]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namei.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Some corrections by tytso. | |
10 | */ | |
11 | ||
12 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
13 | * lookup logic. | |
14 | */ | |
15 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
16 | */ | |
17 | ||
18 | #include <linux/init.h> | |
630d9c47 | 19 | #include <linux/export.h> |
1da177e4 | 20 | #include <linux/slab.h> |
66a5c40f | 21 | #include <linux/wordpart.h> |
1da177e4 | 22 | #include <linux/fs.h> |
5970e15d | 23 | #include <linux/filelock.h> |
1da177e4 | 24 | #include <linux/namei.h> |
1da177e4 | 25 | #include <linux/pagemap.h> |
2d878178 | 26 | #include <linux/sched/mm.h> |
0eeca283 | 27 | #include <linux/fsnotify.h> |
1da177e4 LT |
28 | #include <linux/personality.h> |
29 | #include <linux/security.h> | |
30 | #include <linux/syscalls.h> | |
31 | #include <linux/mount.h> | |
32 | #include <linux/audit.h> | |
16f7e0fe | 33 | #include <linux/capability.h> |
834f2a4a | 34 | #include <linux/file.h> |
5590ff0d | 35 | #include <linux/fcntl.h> |
08ce5f16 | 36 | #include <linux/device_cgroup.h> |
5ad4e53b | 37 | #include <linux/fs_struct.h> |
e77819e5 | 38 | #include <linux/posix_acl.h> |
99d263d4 | 39 | #include <linux/hash.h> |
2a18da7a | 40 | #include <linux/bitops.h> |
aeaa4a79 | 41 | #include <linux/init_task.h> |
7c0f6ba6 | 42 | #include <linux/uaccess.h> |
1da177e4 | 43 | |
e81e3f4d | 44 | #include "internal.h" |
c7105365 | 45 | #include "mount.h" |
e81e3f4d | 46 | |
1da177e4 LT |
47 | /* [Feb-1997 T. Schoebel-Theuer] |
48 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
49 | * were necessary because of omirr. The reason is that omirr needs | |
50 | * to know the _real_ pathname, not the user-supplied one, in case | |
51 | * of symlinks (and also when transname replacements occur). | |
52 | * | |
53 | * The new code replaces the old recursive symlink resolution with | |
54 | * an iterative one (in case of non-nested symlink chains). It does | |
55 | * this with calls to <fs>_follow_link(). | |
56 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
57 | * replaced with a single function lookup_dentry() that can handle all | |
58 | * the special cases of the former code. | |
59 | * | |
60 | * With the new dcache, the pathname is stored at each inode, at least as | |
61 | * long as the refcount of the inode is positive. As a side effect, the | |
62 | * size of the dcache depends on the inode cache and thus is dynamic. | |
63 | * | |
64 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
65 | * resolution to correspond with current state of the code. | |
66 | * | |
67 | * Note that the symlink resolution is not *completely* iterative. | |
68 | * There is still a significant amount of tail- and mid- recursion in | |
69 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
70 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
71 | * may return different results than <fs>_follow_link(). Many virtual | |
72 | * filesystems (including /proc) exhibit this behavior. | |
73 | */ | |
74 | ||
75 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
76 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
77 | * and the name already exists in form of a symlink, try to create the new | |
78 | * name indicated by the symlink. The old code always complained that the | |
79 | * name already exists, due to not following the symlink even if its target | |
80 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 81 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
82 | * |
83 | * I don't know which semantics is the right one, since I have no access | |
84 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
85 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
86 | * "old" one. Personally, I think the new semantics is much more logical. | |
87 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
88 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
89 | * and in the old Linux semantics. | |
90 | */ | |
91 | ||
92 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
93 | * semantics. See the comments in "open_namei" and "do_link" below. | |
94 | * | |
95 | * [10-Sep-98 Alan Modra] Another symlink change. | |
96 | */ | |
97 | ||
98 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
99 | * inside the path - always follow. | |
100 | * in the last component in creation/removal/renaming - never follow. | |
101 | * if LOOKUP_FOLLOW passed - follow. | |
102 | * if the pathname has trailing slashes - follow. | |
103 | * otherwise - don't follow. | |
104 | * (applied in that order). | |
105 | * | |
106 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
107 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
108 | * During the 2.4 we need to fix the userland stuff depending on it - | |
109 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
110 | * XEmacs seems to be relying on it... | |
111 | */ | |
112 | /* | |
113 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 114 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
115 | * any extra contention... |
116 | */ | |
117 | ||
118 | /* In order to reduce some races, while at the same time doing additional | |
119 | * checking and hopefully speeding things up, we copy filenames to the | |
120 | * kernel data space before using them.. | |
121 | * | |
122 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
123 | * PATH_MAX includes the nul terminator --RR. | |
124 | */ | |
91a27b2a | 125 | |
fd2f7cb5 | 126 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 127 | |
51f39a1f | 128 | struct filename * |
91a27b2a JL |
129 | getname_flags(const char __user *filename, int flags, int *empty) |
130 | { | |
94b5d262 | 131 | struct filename *result; |
7950e385 | 132 | char *kname; |
94b5d262 | 133 | int len; |
4043cde8 | 134 | |
7ac86265 JL |
135 | result = audit_reusename(filename); |
136 | if (result) | |
137 | return result; | |
138 | ||
7950e385 | 139 | result = __getname(); |
3f9f0aa6 | 140 | if (unlikely(!result)) |
4043cde8 EP |
141 | return ERR_PTR(-ENOMEM); |
142 | ||
7950e385 JL |
143 | /* |
144 | * First, try to embed the struct filename inside the names_cache | |
145 | * allocation | |
146 | */ | |
fd2f7cb5 | 147 | kname = (char *)result->iname; |
91a27b2a | 148 | result->name = kname; |
7950e385 | 149 | |
94b5d262 | 150 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 151 | if (unlikely(len < 0)) { |
94b5d262 AV |
152 | __putname(result); |
153 | return ERR_PTR(len); | |
91a27b2a | 154 | } |
3f9f0aa6 | 155 | |
7950e385 JL |
156 | /* |
157 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
158 | * separate struct filename so we can dedicate the entire | |
159 | * names_cache allocation for the pathname, and re-do the copy from | |
160 | * userland. | |
161 | */ | |
94b5d262 | 162 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 163 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
164 | kname = (char *)result; |
165 | ||
fd2f7cb5 AV |
166 | /* |
167 | * size is chosen that way we to guarantee that | |
168 | * result->iname[0] is within the same object and that | |
169 | * kname can't be equal to result->iname, no matter what. | |
170 | */ | |
171 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
172 | if (unlikely(!result)) { |
173 | __putname(kname); | |
174 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
175 | } |
176 | result->name = kname; | |
94b5d262 AV |
177 | len = strncpy_from_user(kname, filename, PATH_MAX); |
178 | if (unlikely(len < 0)) { | |
179 | __putname(kname); | |
180 | kfree(result); | |
181 | return ERR_PTR(len); | |
182 | } | |
183 | if (unlikely(len == PATH_MAX)) { | |
184 | __putname(kname); | |
185 | kfree(result); | |
186 | return ERR_PTR(-ENAMETOOLONG); | |
187 | } | |
7950e385 JL |
188 | } |
189 | ||
03adc61e | 190 | atomic_set(&result->refcnt, 1); |
3f9f0aa6 LT |
191 | /* The empty path is special. */ |
192 | if (unlikely(!len)) { | |
193 | if (empty) | |
4043cde8 | 194 | *empty = 1; |
94b5d262 AV |
195 | if (!(flags & LOOKUP_EMPTY)) { |
196 | putname(result); | |
197 | return ERR_PTR(-ENOENT); | |
198 | } | |
1da177e4 | 199 | } |
3f9f0aa6 | 200 | |
7950e385 | 201 | result->uptr = filename; |
c4ad8f98 | 202 | result->aname = NULL; |
7950e385 JL |
203 | audit_getname(result); |
204 | return result; | |
1da177e4 LT |
205 | } |
206 | ||
8228e2c3 DK |
207 | struct filename * |
208 | getname_uflags(const char __user *filename, int uflags) | |
209 | { | |
210 | int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; | |
211 | ||
212 | return getname_flags(filename, flags, NULL); | |
213 | } | |
214 | ||
91a27b2a JL |
215 | struct filename * |
216 | getname(const char __user * filename) | |
f52e0c11 | 217 | { |
f7493e5d | 218 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
219 | } |
220 | ||
c4ad8f98 LT |
221 | struct filename * |
222 | getname_kernel(const char * filename) | |
223 | { | |
224 | struct filename *result; | |
08518549 | 225 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
226 | |
227 | result = __getname(); | |
228 | if (unlikely(!result)) | |
229 | return ERR_PTR(-ENOMEM); | |
230 | ||
08518549 | 231 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 232 | result->name = (char *)result->iname; |
08518549 | 233 | } else if (len <= PATH_MAX) { |
30ce4d19 | 234 | const size_t size = offsetof(struct filename, iname[1]); |
08518549 PM |
235 | struct filename *tmp; |
236 | ||
30ce4d19 | 237 | tmp = kmalloc(size, GFP_KERNEL); |
08518549 PM |
238 | if (unlikely(!tmp)) { |
239 | __putname(result); | |
240 | return ERR_PTR(-ENOMEM); | |
241 | } | |
242 | tmp->name = (char *)result; | |
08518549 PM |
243 | result = tmp; |
244 | } else { | |
245 | __putname(result); | |
246 | return ERR_PTR(-ENAMETOOLONG); | |
247 | } | |
248 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
249 | result->uptr = NULL; |
250 | result->aname = NULL; | |
03adc61e | 251 | atomic_set(&result->refcnt, 1); |
fd3522fd | 252 | audit_getname(result); |
c4ad8f98 | 253 | |
c4ad8f98 LT |
254 | return result; |
255 | } | |
74d7970f | 256 | EXPORT_SYMBOL(getname_kernel); |
c4ad8f98 | 257 | |
91a27b2a | 258 | void putname(struct filename *name) |
1da177e4 | 259 | { |
ea47ab11 | 260 | if (IS_ERR(name)) |
91ef658f DK |
261 | return; |
262 | ||
03adc61e DC |
263 | if (WARN_ON_ONCE(!atomic_read(&name->refcnt))) |
264 | return; | |
55422d0b | 265 | |
03adc61e | 266 | if (!atomic_dec_and_test(&name->refcnt)) |
55422d0b PM |
267 | return; |
268 | ||
fd2f7cb5 | 269 | if (name->name != name->iname) { |
55422d0b PM |
270 | __putname(name->name); |
271 | kfree(name); | |
272 | } else | |
273 | __putname(name); | |
1da177e4 | 274 | } |
74d7970f | 275 | EXPORT_SYMBOL(putname); |
1da177e4 | 276 | |
47291baa CB |
277 | /** |
278 | * check_acl - perform ACL permission checking | |
700b7940 | 279 | * @idmap: idmap of the mount the inode was found from |
47291baa CB |
280 | * @inode: inode to check permissions on |
281 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
282 | * | |
283 | * This function performs the ACL permission checking. Since this function | |
284 | * retrieve POSIX acls it needs to know whether it is called from a blocking or | |
285 | * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. | |
286 | * | |
700b7940 CB |
287 | * If the inode has been found through an idmapped mount the idmap of |
288 | * the vfsmount must be passed through @idmap. This function will then take | |
289 | * care to map the inode according to @idmap before checking permissions. | |
47291baa | 290 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 291 | * raw inode simply pass @nop_mnt_idmap. |
47291baa | 292 | */ |
700b7940 | 293 | static int check_acl(struct mnt_idmap *idmap, |
47291baa | 294 | struct inode *inode, int mask) |
e77819e5 | 295 | { |
84635d68 | 296 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
297 | struct posix_acl *acl; |
298 | ||
e77819e5 | 299 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
300 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
301 | if (!acl) | |
e77819e5 | 302 | return -EAGAIN; |
cac2f8b8 | 303 | /* no ->get_inode_acl() calls in RCU mode... */ |
b8a7a3a6 | 304 | if (is_uncached_acl(acl)) |
3567866b | 305 | return -ECHILD; |
700b7940 | 306 | return posix_acl_permission(idmap, inode, acl, mask); |
e77819e5 LT |
307 | } |
308 | ||
cac2f8b8 | 309 | acl = get_inode_acl(inode, ACL_TYPE_ACCESS); |
2982baa2 CH |
310 | if (IS_ERR(acl)) |
311 | return PTR_ERR(acl); | |
e77819e5 | 312 | if (acl) { |
700b7940 | 313 | int error = posix_acl_permission(idmap, inode, acl, mask); |
e77819e5 LT |
314 | posix_acl_release(acl); |
315 | return error; | |
316 | } | |
84635d68 | 317 | #endif |
e77819e5 LT |
318 | |
319 | return -EAGAIN; | |
320 | } | |
321 | ||
47291baa CB |
322 | /** |
323 | * acl_permission_check - perform basic UNIX permission checking | |
700b7940 | 324 | * @idmap: idmap of the mount the inode was found from |
47291baa CB |
325 | * @inode: inode to check permissions on |
326 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
327 | * | |
328 | * This function performs the basic UNIX permission checking. Since this | |
329 | * function may retrieve POSIX acls it needs to know whether it is called from a | |
330 | * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. | |
5fc475b7 | 331 | * |
700b7940 CB |
332 | * If the inode has been found through an idmapped mount the idmap of |
333 | * the vfsmount must be passed through @idmap. This function will then take | |
334 | * care to map the inode according to @idmap before checking permissions. | |
47291baa | 335 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 336 | * raw inode simply pass @nop_mnt_idmap. |
1da177e4 | 337 | */ |
700b7940 | 338 | static int acl_permission_check(struct mnt_idmap *idmap, |
47291baa | 339 | struct inode *inode, int mask) |
1da177e4 | 340 | { |
26cf46be | 341 | unsigned int mode = inode->i_mode; |
a2bd096f | 342 | vfsuid_t vfsuid; |
1da177e4 | 343 | |
5fc475b7 | 344 | /* Are we the owner? If so, ACL's don't matter */ |
e67fe633 | 345 | vfsuid = i_uid_into_vfsuid(idmap, inode); |
a2bd096f | 346 | if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) { |
5fc475b7 | 347 | mask &= 7; |
1da177e4 | 348 | mode >>= 6; |
5fc475b7 LT |
349 | return (mask & ~mode) ? -EACCES : 0; |
350 | } | |
1da177e4 | 351 | |
5fc475b7 LT |
352 | /* Do we have ACL's? */ |
353 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { | |
700b7940 | 354 | int error = check_acl(idmap, inode, mask); |
5fc475b7 LT |
355 | if (error != -EAGAIN) |
356 | return error; | |
1da177e4 LT |
357 | } |
358 | ||
5fc475b7 LT |
359 | /* Only RWX matters for group/other mode bits */ |
360 | mask &= 7; | |
361 | ||
1da177e4 | 362 | /* |
5fc475b7 LT |
363 | * Are the group permissions different from |
364 | * the other permissions in the bits we care | |
365 | * about? Need to check group ownership if so. | |
1da177e4 | 366 | */ |
5fc475b7 | 367 | if (mask & (mode ^ (mode >> 3))) { |
e67fe633 | 368 | vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); |
a2bd096f | 369 | if (vfsgid_in_group_p(vfsgid)) |
5fc475b7 LT |
370 | mode >>= 3; |
371 | } | |
372 | ||
373 | /* Bits in 'mode' clear that we require? */ | |
374 | return (mask & ~mode) ? -EACCES : 0; | |
5909ccaa LT |
375 | } |
376 | ||
377 | /** | |
b74c79e9 | 378 | * generic_permission - check for access rights on a Posix-like filesystem |
4609e1f1 | 379 | * @idmap: idmap of the mount the inode was found from |
5909ccaa | 380 | * @inode: inode to check access rights for |
5fc475b7 LT |
381 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, |
382 | * %MAY_NOT_BLOCK ...) | |
5909ccaa LT |
383 | * |
384 | * Used to check for read/write/execute permissions on a file. | |
385 | * We use "fsuid" for this, letting us set arbitrary permissions | |
386 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
387 | * are used for other things. |
388 | * | |
389 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
390 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
391 | * It would then be called again in ref-walk mode. | |
47291baa | 392 | * |
4609e1f1 CB |
393 | * If the inode has been found through an idmapped mount the idmap of |
394 | * the vfsmount must be passed through @idmap. This function will then take | |
395 | * care to map the inode according to @idmap before checking permissions. | |
47291baa | 396 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 397 | * raw inode simply pass @nop_mnt_idmap. |
5909ccaa | 398 | */ |
4609e1f1 | 399 | int generic_permission(struct mnt_idmap *idmap, struct inode *inode, |
47291baa | 400 | int mask) |
5909ccaa LT |
401 | { |
402 | int ret; | |
403 | ||
404 | /* | |
948409c7 | 405 | * Do the basic permission checks. |
5909ccaa | 406 | */ |
700b7940 | 407 | ret = acl_permission_check(idmap, inode, mask); |
5909ccaa LT |
408 | if (ret != -EACCES) |
409 | return ret; | |
1da177e4 | 410 | |
d594e7ec AV |
411 | if (S_ISDIR(inode->i_mode)) { |
412 | /* DACs are overridable for directories */ | |
d594e7ec | 413 | if (!(mask & MAY_WRITE)) |
9452e93e | 414 | if (capable_wrt_inode_uidgid(idmap, inode, |
23adbe12 | 415 | CAP_DAC_READ_SEARCH)) |
d594e7ec | 416 | return 0; |
9452e93e | 417 | if (capable_wrt_inode_uidgid(idmap, inode, |
0558c1bf | 418 | CAP_DAC_OVERRIDE)) |
1da177e4 | 419 | return 0; |
2a4c2242 SS |
420 | return -EACCES; |
421 | } | |
1da177e4 LT |
422 | |
423 | /* | |
424 | * Searching includes executable on directories, else just read. | |
425 | */ | |
7ea66001 | 426 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 427 | if (mask == MAY_READ) |
9452e93e | 428 | if (capable_wrt_inode_uidgid(idmap, inode, |
0558c1bf | 429 | CAP_DAC_READ_SEARCH)) |
1da177e4 | 430 | return 0; |
2a4c2242 SS |
431 | /* |
432 | * Read/write DACs are always overridable. | |
433 | * Executable DACs are overridable when there is | |
434 | * at least one exec bit set. | |
435 | */ | |
436 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
9452e93e | 437 | if (capable_wrt_inode_uidgid(idmap, inode, |
0558c1bf | 438 | CAP_DAC_OVERRIDE)) |
2a4c2242 | 439 | return 0; |
1da177e4 LT |
440 | |
441 | return -EACCES; | |
442 | } | |
4d359507 | 443 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 444 | |
47291baa CB |
445 | /** |
446 | * do_inode_permission - UNIX permission checking | |
4609e1f1 | 447 | * @idmap: idmap of the mount the inode was found from |
47291baa CB |
448 | * @inode: inode to check permissions on |
449 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) | |
450 | * | |
3ddcd056 LT |
451 | * We _really_ want to just do "generic_permission()" without |
452 | * even looking at the inode->i_op values. So we keep a cache | |
453 | * flag in inode->i_opflags, that says "this has not special | |
454 | * permission function, use the fast case". | |
455 | */ | |
4609e1f1 | 456 | static inline int do_inode_permission(struct mnt_idmap *idmap, |
47291baa | 457 | struct inode *inode, int mask) |
3ddcd056 LT |
458 | { |
459 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
460 | if (likely(inode->i_op->permission)) | |
4609e1f1 | 461 | return inode->i_op->permission(idmap, inode, mask); |
3ddcd056 LT |
462 | |
463 | /* This gets set once for the inode lifetime */ | |
464 | spin_lock(&inode->i_lock); | |
465 | inode->i_opflags |= IOP_FASTPERM; | |
466 | spin_unlock(&inode->i_lock); | |
467 | } | |
4609e1f1 | 468 | return generic_permission(idmap, inode, mask); |
3ddcd056 LT |
469 | } |
470 | ||
0bdaea90 DH |
471 | /** |
472 | * sb_permission - Check superblock-level permissions | |
473 | * @sb: Superblock of inode to check permission on | |
55852635 | 474 | * @inode: Inode to check permission on |
0bdaea90 DH |
475 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
476 | * | |
477 | * Separate out file-system wide checks from inode-specific permission checks. | |
478 | */ | |
479 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
480 | { | |
481 | if (unlikely(mask & MAY_WRITE)) { | |
482 | umode_t mode = inode->i_mode; | |
483 | ||
484 | /* Nobody gets write access to a read-only fs. */ | |
bc98a42c | 485 | if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
0bdaea90 DH |
486 | return -EROFS; |
487 | } | |
488 | return 0; | |
489 | } | |
490 | ||
491 | /** | |
492 | * inode_permission - Check for access rights to a given inode | |
4609e1f1 | 493 | * @idmap: idmap of the mount the inode was found from |
47291baa CB |
494 | * @inode: Inode to check permission on |
495 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
0bdaea90 DH |
496 | * |
497 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
498 | * this, letting us set arbitrary permissions for filesystem access without | |
499 | * changing the "normal" UIDs which are used for other things. | |
500 | * | |
501 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
502 | */ | |
4609e1f1 | 503 | int inode_permission(struct mnt_idmap *idmap, |
47291baa | 504 | struct inode *inode, int mask) |
0bdaea90 DH |
505 | { |
506 | int retval; | |
507 | ||
508 | retval = sb_permission(inode->i_sb, inode, mask); | |
509 | if (retval) | |
510 | return retval; | |
4bfd054a EB |
511 | |
512 | if (unlikely(mask & MAY_WRITE)) { | |
513 | /* | |
514 | * Nobody gets write access to an immutable file. | |
515 | */ | |
516 | if (IS_IMMUTABLE(inode)) | |
517 | return -EPERM; | |
518 | ||
519 | /* | |
520 | * Updating mtime will likely cause i_uid and i_gid to be | |
521 | * written back improperly if their true value is unknown | |
522 | * to the vfs. | |
523 | */ | |
4609e1f1 | 524 | if (HAS_UNMAPPED_ID(idmap, inode)) |
4bfd054a EB |
525 | return -EACCES; |
526 | } | |
527 | ||
4609e1f1 | 528 | retval = do_inode_permission(idmap, inode, mask); |
4bfd054a EB |
529 | if (retval) |
530 | return retval; | |
531 | ||
532 | retval = devcgroup_inode_permission(inode, mask); | |
533 | if (retval) | |
534 | return retval; | |
535 | ||
536 | return security_inode_permission(inode, mask); | |
0bdaea90 | 537 | } |
4d359507 | 538 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 539 | |
5dd784d0 JB |
540 | /** |
541 | * path_get - get a reference to a path | |
542 | * @path: path to get the reference to | |
543 | * | |
544 | * Given a path increment the reference count to the dentry and the vfsmount. | |
545 | */ | |
dcf787f3 | 546 | void path_get(const struct path *path) |
5dd784d0 JB |
547 | { |
548 | mntget(path->mnt); | |
549 | dget(path->dentry); | |
550 | } | |
551 | EXPORT_SYMBOL(path_get); | |
552 | ||
1d957f9b JB |
553 | /** |
554 | * path_put - put a reference to a path | |
555 | * @path: path to put the reference to | |
556 | * | |
557 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
558 | */ | |
dcf787f3 | 559 | void path_put(const struct path *path) |
1da177e4 | 560 | { |
1d957f9b JB |
561 | dput(path->dentry); |
562 | mntput(path->mnt); | |
1da177e4 | 563 | } |
1d957f9b | 564 | EXPORT_SYMBOL(path_put); |
1da177e4 | 565 | |
894bc8c4 | 566 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
567 | struct nameidata { |
568 | struct path path; | |
1cf2665b | 569 | struct qstr last; |
1f55a6ec AV |
570 | struct path root; |
571 | struct inode *inode; /* path.dentry.d_inode */ | |
bcba1e7d | 572 | unsigned int flags, state; |
03fa86e9 | 573 | unsigned seq, next_seq, m_seq, r_seq; |
1f55a6ec AV |
574 | int last_type; |
575 | unsigned depth; | |
756daf26 | 576 | int total_link_count; |
697fc6ca AV |
577 | struct saved { |
578 | struct path link; | |
fceef393 | 579 | struct delayed_call done; |
697fc6ca | 580 | const char *name; |
0450b2d1 | 581 | unsigned seq; |
894bc8c4 | 582 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
583 | struct filename *name; |
584 | struct nameidata *saved; | |
585 | unsigned root_seq; | |
586 | int dfd; | |
a2bd096f | 587 | vfsuid_t dir_vfsuid; |
0f705953 | 588 | umode_t dir_mode; |
3859a271 | 589 | } __randomize_layout; |
1f55a6ec | 590 | |
bcba1e7d AV |
591 | #define ND_ROOT_PRESET 1 |
592 | #define ND_ROOT_GRABBED 2 | |
593 | #define ND_JUMPED 4 | |
594 | ||
06422964 | 595 | static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 596 | { |
756daf26 N |
597 | struct nameidata *old = current->nameidata; |
598 | p->stack = p->internal; | |
7962c7d1 | 599 | p->depth = 0; |
c8a53ee5 AV |
600 | p->dfd = dfd; |
601 | p->name = name; | |
7d01ef75 AV |
602 | p->path.mnt = NULL; |
603 | p->path.dentry = NULL; | |
756daf26 | 604 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 605 | p->saved = old; |
756daf26 | 606 | current->nameidata = p; |
894bc8c4 AV |
607 | } |
608 | ||
06422964 AV |
609 | static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, |
610 | const struct path *root) | |
611 | { | |
612 | __set_nameidata(p, dfd, name); | |
613 | p->state = 0; | |
614 | if (unlikely(root)) { | |
615 | p->state = ND_ROOT_PRESET; | |
616 | p->root = *root; | |
617 | } | |
618 | } | |
619 | ||
9883d185 | 620 | static void restore_nameidata(void) |
894bc8c4 | 621 | { |
9883d185 | 622 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
623 | |
624 | current->nameidata = old; | |
625 | if (old) | |
626 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 627 | if (now->stack != now->internal) |
756daf26 | 628 | kfree(now->stack); |
894bc8c4 AV |
629 | } |
630 | ||
60ef60c7 | 631 | static bool nd_alloc_stack(struct nameidata *nd) |
894bc8c4 | 632 | { |
bc40aee0 AV |
633 | struct saved *p; |
634 | ||
60ef60c7 AV |
635 | p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), |
636 | nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); | |
637 | if (unlikely(!p)) | |
638 | return false; | |
894bc8c4 AV |
639 | memcpy(p, nd->internal, sizeof(nd->internal)); |
640 | nd->stack = p; | |
60ef60c7 | 641 | return true; |
894bc8c4 AV |
642 | } |
643 | ||
397d425d | 644 | /** |
6b03f7ed | 645 | * path_connected - Verify that a dentry is below mnt.mnt_root |
35931eb3 MWO |
646 | * @mnt: The mountpoint to check. |
647 | * @dentry: The dentry to check. | |
397d425d EB |
648 | * |
649 | * Rename can sometimes move a file or directory outside of a bind | |
650 | * mount, path_connected allows those cases to be detected. | |
651 | */ | |
6b03f7ed | 652 | static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) |
397d425d | 653 | { |
95dd7758 | 654 | struct super_block *sb = mnt->mnt_sb; |
397d425d | 655 | |
402dd2cf CH |
656 | /* Bind mounts can have disconnected paths */ |
657 | if (mnt->mnt_root == sb->s_root) | |
397d425d EB |
658 | return true; |
659 | ||
6b03f7ed | 660 | return is_subdir(dentry, mnt->mnt_root); |
397d425d EB |
661 | } |
662 | ||
7973387a AV |
663 | static void drop_links(struct nameidata *nd) |
664 | { | |
665 | int i = nd->depth; | |
666 | while (i--) { | |
667 | struct saved *last = nd->stack + i; | |
fceef393 AV |
668 | do_delayed_call(&last->done); |
669 | clear_delayed_call(&last->done); | |
7973387a AV |
670 | } |
671 | } | |
672 | ||
6e180327 AV |
673 | static void leave_rcu(struct nameidata *nd) |
674 | { | |
675 | nd->flags &= ~LOOKUP_RCU; | |
03fa86e9 | 676 | nd->seq = nd->next_seq = 0; |
6e180327 AV |
677 | rcu_read_unlock(); |
678 | } | |
679 | ||
7973387a AV |
680 | static void terminate_walk(struct nameidata *nd) |
681 | { | |
682 | drop_links(nd); | |
683 | if (!(nd->flags & LOOKUP_RCU)) { | |
684 | int i; | |
685 | path_put(&nd->path); | |
686 | for (i = 0; i < nd->depth; i++) | |
687 | path_put(&nd->stack[i].link); | |
bcba1e7d | 688 | if (nd->state & ND_ROOT_GRABBED) { |
102b8af2 | 689 | path_put(&nd->root); |
bcba1e7d | 690 | nd->state &= ~ND_ROOT_GRABBED; |
102b8af2 | 691 | } |
7973387a | 692 | } else { |
6e180327 | 693 | leave_rcu(nd); |
7973387a AV |
694 | } |
695 | nd->depth = 0; | |
7d01ef75 AV |
696 | nd->path.mnt = NULL; |
697 | nd->path.dentry = NULL; | |
7973387a AV |
698 | } |
699 | ||
700 | /* path_put is needed afterwards regardless of success or failure */ | |
2aa38470 | 701 | static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) |
7973387a | 702 | { |
2aa38470 | 703 | int res = __legitimize_mnt(path->mnt, mseq); |
7973387a AV |
704 | if (unlikely(res)) { |
705 | if (res > 0) | |
706 | path->mnt = NULL; | |
707 | path->dentry = NULL; | |
708 | return false; | |
709 | } | |
710 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
711 | path->dentry = NULL; | |
712 | return false; | |
713 | } | |
714 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
715 | } | |
716 | ||
2aa38470 AV |
717 | static inline bool legitimize_path(struct nameidata *nd, |
718 | struct path *path, unsigned seq) | |
719 | { | |
5bd73286 | 720 | return __legitimize_path(path, seq, nd->m_seq); |
2aa38470 AV |
721 | } |
722 | ||
7973387a AV |
723 | static bool legitimize_links(struct nameidata *nd) |
724 | { | |
725 | int i; | |
eacd9aa8 AV |
726 | if (unlikely(nd->flags & LOOKUP_CACHED)) { |
727 | drop_links(nd); | |
728 | nd->depth = 0; | |
729 | return false; | |
730 | } | |
7973387a AV |
731 | for (i = 0; i < nd->depth; i++) { |
732 | struct saved *last = nd->stack + i; | |
733 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
734 | drop_links(nd); | |
735 | nd->depth = i + 1; | |
736 | return false; | |
737 | } | |
738 | } | |
739 | return true; | |
740 | } | |
741 | ||
ee594bff AV |
742 | static bool legitimize_root(struct nameidata *nd) |
743 | { | |
adb21d2b | 744 | /* Nothing to do if nd->root is zero or is managed by the VFS user. */ |
bcba1e7d | 745 | if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) |
ee594bff | 746 | return true; |
bcba1e7d | 747 | nd->state |= ND_ROOT_GRABBED; |
ee594bff AV |
748 | return legitimize_path(nd, &nd->root, nd->root_seq); |
749 | } | |
750 | ||
19660af7 | 751 | /* |
31e6b01f | 752 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
753 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
754 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 755 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
756 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
757 | * got stuck, so ref-walk may continue from there. If this is not successful | |
758 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
759 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 760 | */ |
31e6b01f NP |
761 | |
762 | /** | |
e36cffed | 763 | * try_to_unlazy - try to switch to ref-walk mode. |
19660af7 | 764 | * @nd: nameidata pathwalk data |
e36cffed | 765 | * Returns: true on success, false on failure |
31e6b01f | 766 | * |
e36cffed | 767 | * try_to_unlazy attempts to legitimize the current nd->path and nd->root |
4675ac39 AV |
768 | * for ref-walk mode. |
769 | * Must be called from rcu-walk context. | |
e36cffed | 770 | * Nothing should touch nameidata between try_to_unlazy() failure and |
7973387a | 771 | * terminate_walk(). |
31e6b01f | 772 | */ |
e36cffed | 773 | static bool try_to_unlazy(struct nameidata *nd) |
31e6b01f | 774 | { |
31e6b01f NP |
775 | struct dentry *parent = nd->path.dentry; |
776 | ||
777 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 778 | |
4675ac39 | 779 | if (unlikely(!legitimize_links(nd))) |
4675ac39 | 780 | goto out1; |
84a2bd39 AV |
781 | if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) |
782 | goto out; | |
ee594bff AV |
783 | if (unlikely(!legitimize_root(nd))) |
784 | goto out; | |
6e180327 | 785 | leave_rcu(nd); |
4675ac39 | 786 | BUG_ON(nd->inode != parent->d_inode); |
e36cffed | 787 | return true; |
4675ac39 | 788 | |
84a2bd39 | 789 | out1: |
4675ac39 AV |
790 | nd->path.mnt = NULL; |
791 | nd->path.dentry = NULL; | |
4675ac39 | 792 | out: |
6e180327 | 793 | leave_rcu(nd); |
e36cffed | 794 | return false; |
4675ac39 AV |
795 | } |
796 | ||
797 | /** | |
ae66db45 | 798 | * try_to_unlazy_next - try to switch to ref-walk mode. |
4675ac39 | 799 | * @nd: nameidata pathwalk data |
ae66db45 | 800 | * @dentry: next dentry to step into |
ae66db45 | 801 | * Returns: true on success, false on failure |
4675ac39 | 802 | * |
30476f7e | 803 | * Similar to try_to_unlazy(), but here we have the next dentry already |
ae66db45 AV |
804 | * picked by rcu-walk and want to legitimize that in addition to the current |
805 | * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. | |
806 | * Nothing should touch nameidata between try_to_unlazy_next() failure and | |
4675ac39 AV |
807 | * terminate_walk(). |
808 | */ | |
03fa86e9 | 809 | static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry) |
4675ac39 | 810 | { |
7e4745a0 | 811 | int res; |
4675ac39 AV |
812 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
813 | ||
7973387a AV |
814 | if (unlikely(!legitimize_links(nd))) |
815 | goto out2; | |
7e4745a0 AV |
816 | res = __legitimize_mnt(nd->path.mnt, nd->m_seq); |
817 | if (unlikely(res)) { | |
818 | if (res > 0) | |
819 | goto out2; | |
820 | goto out1; | |
821 | } | |
4675ac39 | 822 | if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) |
7973387a | 823 | goto out1; |
48a066e7 | 824 | |
15570086 | 825 | /* |
4675ac39 AV |
826 | * We need to move both the parent and the dentry from the RCU domain |
827 | * to be properly refcounted. And the sequence number in the dentry | |
828 | * validates *both* dentry counters, since we checked the sequence | |
829 | * number of the parent after we got the child sequence number. So we | |
830 | * know the parent must still be valid if the child sequence number is | |
15570086 | 831 | */ |
4675ac39 AV |
832 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) |
833 | goto out; | |
03fa86e9 | 834 | if (read_seqcount_retry(&dentry->d_seq, nd->next_seq)) |
84a2bd39 | 835 | goto out_dput; |
e5c832d5 LT |
836 | /* |
837 | * Sequence counts matched. Now make sure that the root is | |
838 | * still valid and get it if required. | |
839 | */ | |
84a2bd39 AV |
840 | if (unlikely(!legitimize_root(nd))) |
841 | goto out_dput; | |
6e180327 | 842 | leave_rcu(nd); |
ae66db45 | 843 | return true; |
19660af7 | 844 | |
7973387a AV |
845 | out2: |
846 | nd->path.mnt = NULL; | |
847 | out1: | |
848 | nd->path.dentry = NULL; | |
e5c832d5 | 849 | out: |
6e180327 | 850 | leave_rcu(nd); |
ae66db45 | 851 | return false; |
84a2bd39 | 852 | out_dput: |
6e180327 | 853 | leave_rcu(nd); |
84a2bd39 | 854 | dput(dentry); |
ae66db45 | 855 | return false; |
31e6b01f NP |
856 | } |
857 | ||
4ce16ef3 | 858 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 859 | { |
a89f8337 AV |
860 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
861 | return dentry->d_op->d_revalidate(dentry, flags); | |
862 | else | |
863 | return 1; | |
34286d66 NP |
864 | } |
865 | ||
9f1fafee AV |
866 | /** |
867 | * complete_walk - successful completion of path walk | |
868 | * @nd: pointer nameidata | |
39159de2 | 869 | * |
9f1fafee AV |
870 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
871 | * Revalidate the final result, unless we'd already done that during | |
872 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
873 | * success, -error on failure. In case of failure caller does not | |
874 | * need to drop nd->path. | |
39159de2 | 875 | */ |
9f1fafee | 876 | static int complete_walk(struct nameidata *nd) |
39159de2 | 877 | { |
16c2cd71 | 878 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 879 | int status; |
39159de2 | 880 | |
9f1fafee | 881 | if (nd->flags & LOOKUP_RCU) { |
adb21d2b AS |
882 | /* |
883 | * We don't want to zero nd->root for scoped-lookups or | |
884 | * externally-managed nd->root. | |
885 | */ | |
bcba1e7d AV |
886 | if (!(nd->state & ND_ROOT_PRESET)) |
887 | if (!(nd->flags & LOOKUP_IS_SCOPED)) | |
888 | nd->root.mnt = NULL; | |
6c6ec2b0 | 889 | nd->flags &= ~LOOKUP_CACHED; |
e36cffed | 890 | if (!try_to_unlazy(nd)) |
9f1fafee | 891 | return -ECHILD; |
9f1fafee AV |
892 | } |
893 | ||
adb21d2b AS |
894 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { |
895 | /* | |
896 | * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't | |
897 | * ever step outside the root during lookup" and should already | |
898 | * be guaranteed by the rest of namei, we want to avoid a namei | |
899 | * BUG resulting in userspace being given a path that was not | |
900 | * scoped within the root at some point during the lookup. | |
901 | * | |
902 | * So, do a final sanity-check to make sure that in the | |
903 | * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) | |
904 | * we won't silently return an fd completely outside of the | |
905 | * requested root to userspace. | |
906 | * | |
907 | * Userspace could move the path outside the root after this | |
908 | * check, but as discussed elsewhere this is not a concern (the | |
909 | * resolved file was inside the root at some point). | |
910 | */ | |
911 | if (!path_is_under(&nd->path, &nd->root)) | |
912 | return -EXDEV; | |
913 | } | |
914 | ||
bcba1e7d | 915 | if (likely(!(nd->state & ND_JUMPED))) |
16c2cd71 AV |
916 | return 0; |
917 | ||
ecf3d1f1 | 918 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
919 | return 0; |
920 | ||
ecf3d1f1 | 921 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
922 | if (status > 0) |
923 | return 0; | |
924 | ||
16c2cd71 | 925 | if (!status) |
39159de2 | 926 | status = -ESTALE; |
16c2cd71 | 927 | |
39159de2 JL |
928 | return status; |
929 | } | |
930 | ||
740a1678 | 931 | static int set_root(struct nameidata *nd) |
31e6b01f | 932 | { |
7bd88377 | 933 | struct fs_struct *fs = current->fs; |
c28cc364 | 934 | |
adb21d2b AS |
935 | /* |
936 | * Jumping to the real root in a scoped-lookup is a BUG in namei, but we | |
937 | * still have to ensure it doesn't happen because it will cause a breakout | |
938 | * from the dirfd. | |
939 | */ | |
940 | if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) | |
941 | return -ENOTRECOVERABLE; | |
942 | ||
9e6697e2 AV |
943 | if (nd->flags & LOOKUP_RCU) { |
944 | unsigned seq; | |
945 | ||
946 | do { | |
947 | seq = read_seqcount_begin(&fs->seq); | |
948 | nd->root = fs->root; | |
949 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
950 | } while (read_seqcount_retry(&fs->seq, seq)); | |
951 | } else { | |
952 | get_fs_root(fs, &nd->root); | |
bcba1e7d | 953 | nd->state |= ND_ROOT_GRABBED; |
9e6697e2 | 954 | } |
740a1678 | 955 | return 0; |
31e6b01f NP |
956 | } |
957 | ||
248fb5b9 AV |
958 | static int nd_jump_root(struct nameidata *nd) |
959 | { | |
adb21d2b AS |
960 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
961 | return -EXDEV; | |
72ba2929 AS |
962 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { |
963 | /* Absolute path arguments to path_init() are allowed. */ | |
964 | if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) | |
965 | return -EXDEV; | |
966 | } | |
740a1678 AS |
967 | if (!nd->root.mnt) { |
968 | int error = set_root(nd); | |
969 | if (error) | |
970 | return error; | |
971 | } | |
248fb5b9 AV |
972 | if (nd->flags & LOOKUP_RCU) { |
973 | struct dentry *d; | |
974 | nd->path = nd->root; | |
975 | d = nd->path.dentry; | |
976 | nd->inode = d->d_inode; | |
977 | nd->seq = nd->root_seq; | |
82ef0698 | 978 | if (read_seqcount_retry(&d->d_seq, nd->seq)) |
248fb5b9 AV |
979 | return -ECHILD; |
980 | } else { | |
981 | path_put(&nd->path); | |
982 | nd->path = nd->root; | |
983 | path_get(&nd->path); | |
984 | nd->inode = nd->path.dentry->d_inode; | |
985 | } | |
bcba1e7d | 986 | nd->state |= ND_JUMPED; |
248fb5b9 AV |
987 | return 0; |
988 | } | |
989 | ||
b5fb63c1 | 990 | /* |
6b255391 | 991 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
992 | * caller must have taken a reference to path beforehand. |
993 | */ | |
ea4af4aa | 994 | int nd_jump_link(const struct path *path) |
b5fb63c1 | 995 | { |
4b99d499 | 996 | int error = -ELOOP; |
6e77137b | 997 | struct nameidata *nd = current->nameidata; |
b5fb63c1 | 998 | |
4b99d499 AS |
999 | if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) |
1000 | goto err; | |
1001 | ||
72ba2929 AS |
1002 | error = -EXDEV; |
1003 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { | |
1004 | if (nd->path.mnt != path->mnt) | |
1005 | goto err; | |
1006 | } | |
adb21d2b AS |
1007 | /* Not currently safe for scoped-lookups. */ |
1008 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) | |
1009 | goto err; | |
72ba2929 | 1010 | |
4b99d499 | 1011 | path_put(&nd->path); |
b5fb63c1 CH |
1012 | nd->path = *path; |
1013 | nd->inode = nd->path.dentry->d_inode; | |
bcba1e7d | 1014 | nd->state |= ND_JUMPED; |
1bc82070 | 1015 | return 0; |
4b99d499 AS |
1016 | |
1017 | err: | |
1018 | path_put(path); | |
1019 | return error; | |
b5fb63c1 CH |
1020 | } |
1021 | ||
b9ff4429 | 1022 | static inline void put_link(struct nameidata *nd) |
574197e0 | 1023 | { |
21c3003d | 1024 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 1025 | do_delayed_call(&last->done); |
6548fae2 AV |
1026 | if (!(nd->flags & LOOKUP_RCU)) |
1027 | path_put(&last->link); | |
574197e0 AV |
1028 | } |
1029 | ||
9c011be1 LC |
1030 | static int sysctl_protected_symlinks __read_mostly; |
1031 | static int sysctl_protected_hardlinks __read_mostly; | |
1032 | static int sysctl_protected_fifos __read_mostly; | |
1033 | static int sysctl_protected_regular __read_mostly; | |
1034 | ||
1035 | #ifdef CONFIG_SYSCTL | |
1036 | static struct ctl_table namei_sysctls[] = { | |
1037 | { | |
1038 | .procname = "protected_symlinks", | |
1039 | .data = &sysctl_protected_symlinks, | |
1040 | .maxlen = sizeof(int), | |
c7031c14 | 1041 | .mode = 0644, |
9c011be1 LC |
1042 | .proc_handler = proc_dointvec_minmax, |
1043 | .extra1 = SYSCTL_ZERO, | |
1044 | .extra2 = SYSCTL_ONE, | |
1045 | }, | |
1046 | { | |
1047 | .procname = "protected_hardlinks", | |
1048 | .data = &sysctl_protected_hardlinks, | |
1049 | .maxlen = sizeof(int), | |
c7031c14 | 1050 | .mode = 0644, |
9c011be1 LC |
1051 | .proc_handler = proc_dointvec_minmax, |
1052 | .extra1 = SYSCTL_ZERO, | |
1053 | .extra2 = SYSCTL_ONE, | |
1054 | }, | |
1055 | { | |
1056 | .procname = "protected_fifos", | |
1057 | .data = &sysctl_protected_fifos, | |
1058 | .maxlen = sizeof(int), | |
c7031c14 | 1059 | .mode = 0644, |
9c011be1 LC |
1060 | .proc_handler = proc_dointvec_minmax, |
1061 | .extra1 = SYSCTL_ZERO, | |
1062 | .extra2 = SYSCTL_TWO, | |
1063 | }, | |
1064 | { | |
1065 | .procname = "protected_regular", | |
1066 | .data = &sysctl_protected_regular, | |
1067 | .maxlen = sizeof(int), | |
c7031c14 | 1068 | .mode = 0644, |
9c011be1 LC |
1069 | .proc_handler = proc_dointvec_minmax, |
1070 | .extra1 = SYSCTL_ZERO, | |
1071 | .extra2 = SYSCTL_TWO, | |
1072 | }, | |
9c011be1 LC |
1073 | }; |
1074 | ||
1075 | static int __init init_fs_namei_sysctls(void) | |
1076 | { | |
1077 | register_sysctl_init("fs", namei_sysctls); | |
1078 | return 0; | |
1079 | } | |
1080 | fs_initcall(init_fs_namei_sysctls); | |
1081 | ||
1082 | #endif /* CONFIG_SYSCTL */ | |
800179c9 KC |
1083 | |
1084 | /** | |
1085 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 1086 | * @nd: nameidata pathwalk data |
35931eb3 | 1087 | * @inode: Used for idmapping. |
800179c9 KC |
1088 | * |
1089 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
1090 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
1091 | * in a sticky world-writable directory. This is to protect privileged | |
1092 | * processes from failing races against path names that may change out | |
1093 | * from under them by way of other users creating malicious symlinks. | |
1094 | * It will permit symlinks to be followed only when outside a sticky | |
1095 | * world-writable directory, or when the uid of the symlink and follower | |
1096 | * match, or when the directory owner matches the symlink's owner. | |
1097 | * | |
1098 | * Returns 0 if following the symlink is allowed, -ve on error. | |
1099 | */ | |
ad6cc4c3 | 1100 | static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) |
800179c9 | 1101 | { |
e67fe633 | 1102 | struct mnt_idmap *idmap; |
a2bd096f | 1103 | vfsuid_t vfsuid; |
ba73d987 | 1104 | |
800179c9 KC |
1105 | if (!sysctl_protected_symlinks) |
1106 | return 0; | |
1107 | ||
e67fe633 CB |
1108 | idmap = mnt_idmap(nd->path.mnt); |
1109 | vfsuid = i_uid_into_vfsuid(idmap, inode); | |
800179c9 | 1110 | /* Allowed if owner and follower match. */ |
a2bd096f | 1111 | if (vfsuid_eq_kuid(vfsuid, current_fsuid())) |
800179c9 KC |
1112 | return 0; |
1113 | ||
1114 | /* Allowed if parent directory not sticky and world-writable. */ | |
0f705953 | 1115 | if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
800179c9 KC |
1116 | return 0; |
1117 | ||
1118 | /* Allowed if parent directory and link owner match. */ | |
a2bd096f | 1119 | if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid)) |
800179c9 KC |
1120 | return 0; |
1121 | ||
31956502 AV |
1122 | if (nd->flags & LOOKUP_RCU) |
1123 | return -ECHILD; | |
1124 | ||
ea841baf | 1125 | audit_inode(nd->name, nd->stack[0].link.dentry, 0); |
245d7369 | 1126 | audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); |
800179c9 KC |
1127 | return -EACCES; |
1128 | } | |
1129 | ||
1130 | /** | |
1131 | * safe_hardlink_source - Check for safe hardlink conditions | |
4609e1f1 | 1132 | * @idmap: idmap of the mount the inode was found from |
800179c9 KC |
1133 | * @inode: the source inode to hardlink from |
1134 | * | |
1135 | * Return false if at least one of the following conditions: | |
1136 | * - inode is not a regular file | |
1137 | * - inode is setuid | |
1138 | * - inode is setgid and group-exec | |
1139 | * - access failure for read and write | |
1140 | * | |
1141 | * Otherwise returns true. | |
1142 | */ | |
4609e1f1 | 1143 | static bool safe_hardlink_source(struct mnt_idmap *idmap, |
ba73d987 | 1144 | struct inode *inode) |
800179c9 KC |
1145 | { |
1146 | umode_t mode = inode->i_mode; | |
1147 | ||
1148 | /* Special files should not get pinned to the filesystem. */ | |
1149 | if (!S_ISREG(mode)) | |
1150 | return false; | |
1151 | ||
1152 | /* Setuid files should not get pinned to the filesystem. */ | |
1153 | if (mode & S_ISUID) | |
1154 | return false; | |
1155 | ||
1156 | /* Executable setgid files should not get pinned to the filesystem. */ | |
1157 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
1158 | return false; | |
1159 | ||
1160 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
4609e1f1 | 1161 | if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE)) |
800179c9 KC |
1162 | return false; |
1163 | ||
1164 | return true; | |
1165 | } | |
1166 | ||
1167 | /** | |
1168 | * may_linkat - Check permissions for creating a hardlink | |
4609e1f1 CB |
1169 | * @idmap: idmap of the mount the inode was found from |
1170 | * @link: the source to hardlink from | |
800179c9 KC |
1171 | * |
1172 | * Block hardlink when all of: | |
1173 | * - sysctl_protected_hardlinks enabled | |
1174 | * - fsuid does not match inode | |
1175 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 1176 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 | 1177 | * |
4609e1f1 CB |
1178 | * If the inode has been found through an idmapped mount the idmap of |
1179 | * the vfsmount must be passed through @idmap. This function will then take | |
1180 | * care to map the inode according to @idmap before checking permissions. | |
ba73d987 | 1181 | * On non-idmapped mounts or if permission checking is to be performed on the |
4609e1f1 | 1182 | * raw inode simply pass @nop_mnt_idmap. |
ba73d987 | 1183 | * |
800179c9 KC |
1184 | * Returns 0 if successful, -ve on error. |
1185 | */ | |
4609e1f1 | 1186 | int may_linkat(struct mnt_idmap *idmap, const struct path *link) |
800179c9 | 1187 | { |
593d1ce8 EB |
1188 | struct inode *inode = link->dentry->d_inode; |
1189 | ||
1190 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
e67fe633 CB |
1191 | if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || |
1192 | !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) | |
593d1ce8 | 1193 | return -EOVERFLOW; |
800179c9 KC |
1194 | |
1195 | if (!sysctl_protected_hardlinks) | |
1196 | return 0; | |
1197 | ||
800179c9 KC |
1198 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, |
1199 | * otherwise, it must be a safe source. | |
1200 | */ | |
4609e1f1 | 1201 | if (safe_hardlink_source(idmap, inode) || |
01beba79 | 1202 | inode_owner_or_capable(idmap, inode)) |
800179c9 KC |
1203 | return 0; |
1204 | ||
245d7369 | 1205 | audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); |
800179c9 KC |
1206 | return -EPERM; |
1207 | } | |
1208 | ||
30aba665 SM |
1209 | /** |
1210 | * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory | |
1211 | * should be allowed, or not, on files that already | |
1212 | * exist. | |
e67fe633 | 1213 | * @idmap: idmap of the mount the inode was found from |
2111c3c0 | 1214 | * @nd: nameidata pathwalk data |
30aba665 SM |
1215 | * @inode: the inode of the file to open |
1216 | * | |
1217 | * Block an O_CREAT open of a FIFO (or a regular file) when: | |
1218 | * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled | |
1219 | * - the file already exists | |
1220 | * - we are in a sticky directory | |
1221 | * - we don't own the file | |
1222 | * - the owner of the directory doesn't own the file | |
1223 | * - the directory is world writable | |
1224 | * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 | |
1225 | * the directory doesn't have to be world writable: being group writable will | |
1226 | * be enough. | |
1227 | * | |
e67fe633 CB |
1228 | * If the inode has been found through an idmapped mount the idmap of |
1229 | * the vfsmount must be passed through @idmap. This function will then take | |
1230 | * care to map the inode according to @idmap before checking permissions. | |
ba73d987 | 1231 | * On non-idmapped mounts or if permission checking is to be performed on the |
e67fe633 | 1232 | * raw inode simply pass @nop_mnt_idmap. |
ba73d987 | 1233 | * |
30aba665 SM |
1234 | * Returns 0 if the open is allowed, -ve on error. |
1235 | */ | |
e67fe633 | 1236 | static int may_create_in_sticky(struct mnt_idmap *idmap, |
ba73d987 | 1237 | struct nameidata *nd, struct inode *const inode) |
30aba665 | 1238 | { |
ba73d987 | 1239 | umode_t dir_mode = nd->dir_mode; |
a2bd096f | 1240 | vfsuid_t dir_vfsuid = nd->dir_vfsuid; |
ba73d987 | 1241 | |
30aba665 SM |
1242 | if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || |
1243 | (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || | |
d0cb5018 | 1244 | likely(!(dir_mode & S_ISVTX)) || |
e67fe633 CB |
1245 | vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) || |
1246 | vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid())) | |
30aba665 SM |
1247 | return 0; |
1248 | ||
d0cb5018 AV |
1249 | if (likely(dir_mode & 0002) || |
1250 | (dir_mode & 0020 && | |
30aba665 SM |
1251 | ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || |
1252 | (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { | |
245d7369 KC |
1253 | const char *operation = S_ISFIFO(inode->i_mode) ? |
1254 | "sticky_create_fifo" : | |
1255 | "sticky_create_regular"; | |
1256 | audit_log_path_denied(AUDIT_ANOM_CREAT, operation); | |
30aba665 SM |
1257 | return -EACCES; |
1258 | } | |
1259 | return 0; | |
1260 | } | |
1261 | ||
f015f126 DH |
1262 | /* |
1263 | * follow_up - Find the mountpoint of path's vfsmount | |
1264 | * | |
1265 | * Given a path, find the mountpoint of its source file system. | |
1266 | * Replace @path with the path of the mountpoint in the parent mount. | |
1267 | * Up is towards /. | |
1268 | * | |
1269 | * Return 1 if we went up a level and 0 if we were already at the | |
1270 | * root. | |
1271 | */ | |
bab77ebf | 1272 | int follow_up(struct path *path) |
1da177e4 | 1273 | { |
0714a533 AV |
1274 | struct mount *mnt = real_mount(path->mnt); |
1275 | struct mount *parent; | |
1da177e4 | 1276 | struct dentry *mountpoint; |
99b7db7b | 1277 | |
48a066e7 | 1278 | read_seqlock_excl(&mount_lock); |
0714a533 | 1279 | parent = mnt->mnt_parent; |
3c0a6163 | 1280 | if (parent == mnt) { |
48a066e7 | 1281 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1282 | return 0; |
1283 | } | |
0714a533 | 1284 | mntget(&parent->mnt); |
a73324da | 1285 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1286 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1287 | dput(path->dentry); |
1288 | path->dentry = mountpoint; | |
1289 | mntput(path->mnt); | |
0714a533 | 1290 | path->mnt = &parent->mnt; |
1da177e4 LT |
1291 | return 1; |
1292 | } | |
4d359507 | 1293 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1294 | |
7ef482fa AV |
1295 | static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, |
1296 | struct path *path, unsigned *seqp) | |
1297 | { | |
1298 | while (mnt_has_parent(m)) { | |
1299 | struct dentry *mountpoint = m->mnt_mountpoint; | |
1300 | ||
1301 | m = m->mnt_parent; | |
1302 | if (unlikely(root->dentry == mountpoint && | |
1303 | root->mnt == &m->mnt)) | |
1304 | break; | |
1305 | if (mountpoint != m->mnt.mnt_root) { | |
1306 | path->mnt = &m->mnt; | |
1307 | path->dentry = mountpoint; | |
1308 | *seqp = read_seqcount_begin(&mountpoint->d_seq); | |
1309 | return true; | |
1310 | } | |
1311 | } | |
1312 | return false; | |
1313 | } | |
1314 | ||
2aa38470 AV |
1315 | static bool choose_mountpoint(struct mount *m, const struct path *root, |
1316 | struct path *path) | |
1317 | { | |
1318 | bool found; | |
1319 | ||
1320 | rcu_read_lock(); | |
1321 | while (1) { | |
1322 | unsigned seq, mseq = read_seqbegin(&mount_lock); | |
1323 | ||
1324 | found = choose_mountpoint_rcu(m, root, path, &seq); | |
1325 | if (unlikely(!found)) { | |
1326 | if (!read_seqretry(&mount_lock, mseq)) | |
1327 | break; | |
1328 | } else { | |
1329 | if (likely(__legitimize_path(path, seq, mseq))) | |
1330 | break; | |
1331 | rcu_read_unlock(); | |
1332 | path_put(path); | |
1333 | rcu_read_lock(); | |
1334 | } | |
1335 | } | |
1336 | rcu_read_unlock(); | |
1337 | return found; | |
1338 | } | |
1339 | ||
b5c84bf6 | 1340 | /* |
9875cf80 DH |
1341 | * Perform an automount |
1342 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1343 | * were called with. | |
1da177e4 | 1344 | */ |
1c9f5e06 | 1345 | static int follow_automount(struct path *path, int *count, unsigned lookup_flags) |
31e6b01f | 1346 | { |
25e195aa | 1347 | struct dentry *dentry = path->dentry; |
9875cf80 | 1348 | |
0ec26fd0 MS |
1349 | /* We don't want to mount if someone's just doing a stat - |
1350 | * unless they're stat'ing a directory and appended a '/' to | |
1351 | * the name. | |
1352 | * | |
1353 | * We do, however, want to mount if someone wants to open or | |
1354 | * create a file of any type under the mountpoint, wants to | |
1355 | * traverse through the mountpoint or wants to open the | |
1356 | * mounted directory. Also, autofs may mark negative dentries | |
1357 | * as being automount points. These will need the attentions | |
1358 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1359 | */ |
1c9f5e06 | 1360 | if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
5d38f049 | 1361 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
25e195aa | 1362 | dentry->d_inode) |
5d38f049 | 1363 | return -EISDIR; |
0ec26fd0 | 1364 | |
1c9f5e06 | 1365 | if (count && (*count)++ >= MAXSYMLINKS) |
9875cf80 DH |
1366 | return -ELOOP; |
1367 | ||
25e195aa | 1368 | return finish_automount(dentry->d_op->d_automount(path), path); |
463ffb2e AV |
1369 | } |
1370 | ||
9875cf80 | 1371 | /* |
9deed3eb AV |
1372 | * mount traversal - out-of-line part. One note on ->d_flags accesses - |
1373 | * dentries are pinned but not locked here, so negative dentry can go | |
1374 | * positive right under us. Use of smp_load_acquire() provides a barrier | |
1375 | * sufficient for ->d_inode and ->d_flags consistency. | |
9875cf80 | 1376 | */ |
9deed3eb AV |
1377 | static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, |
1378 | int *count, unsigned lookup_flags) | |
1da177e4 | 1379 | { |
9deed3eb | 1380 | struct vfsmount *mnt = path->mnt; |
9875cf80 | 1381 | bool need_mntput = false; |
8aef1884 | 1382 | int ret = 0; |
9875cf80 | 1383 | |
9deed3eb | 1384 | while (flags & DCACHE_MANAGED_DENTRY) { |
cc53ce53 DH |
1385 | /* Allow the filesystem to manage the transit without i_mutex |
1386 | * being held. */ | |
d41efb52 | 1387 | if (flags & DCACHE_MANAGE_TRANSIT) { |
fb5f51c7 | 1388 | ret = path->dentry->d_op->d_manage(path, false); |
508c8772 | 1389 | flags = smp_load_acquire(&path->dentry->d_flags); |
cc53ce53 | 1390 | if (ret < 0) |
8aef1884 | 1391 | break; |
cc53ce53 DH |
1392 | } |
1393 | ||
9deed3eb | 1394 | if (flags & DCACHE_MOUNTED) { // something's mounted on it.. |
9875cf80 | 1395 | struct vfsmount *mounted = lookup_mnt(path); |
9deed3eb | 1396 | if (mounted) { // ... in our namespace |
9875cf80 DH |
1397 | dput(path->dentry); |
1398 | if (need_mntput) | |
1399 | mntput(path->mnt); | |
1400 | path->mnt = mounted; | |
1401 | path->dentry = dget(mounted->mnt_root); | |
9deed3eb AV |
1402 | // here we know it's positive |
1403 | flags = path->dentry->d_flags; | |
9875cf80 DH |
1404 | need_mntput = true; |
1405 | continue; | |
1406 | } | |
9875cf80 DH |
1407 | } |
1408 | ||
9deed3eb AV |
1409 | if (!(flags & DCACHE_NEED_AUTOMOUNT)) |
1410 | break; | |
9875cf80 | 1411 | |
9deed3eb AV |
1412 | // uncovered automount point |
1413 | ret = follow_automount(path, count, lookup_flags); | |
1414 | flags = smp_load_acquire(&path->dentry->d_flags); | |
1415 | if (ret < 0) | |
1416 | break; | |
1da177e4 | 1417 | } |
8aef1884 | 1418 | |
9deed3eb AV |
1419 | if (ret == -EISDIR) |
1420 | ret = 0; | |
1421 | // possible if you race with several mount --move | |
1422 | if (need_mntput && path->mnt == mnt) | |
1423 | mntput(path->mnt); | |
1424 | if (!ret && unlikely(d_flags_negative(flags))) | |
d41efb52 | 1425 | ret = -ENOENT; |
9deed3eb | 1426 | *jumped = need_mntput; |
8402752e | 1427 | return ret; |
1da177e4 LT |
1428 | } |
1429 | ||
9deed3eb AV |
1430 | static inline int traverse_mounts(struct path *path, bool *jumped, |
1431 | int *count, unsigned lookup_flags) | |
1432 | { | |
1433 | unsigned flags = smp_load_acquire(&path->dentry->d_flags); | |
1434 | ||
1435 | /* fastpath */ | |
1436 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { | |
1437 | *jumped = false; | |
1438 | if (unlikely(d_flags_negative(flags))) | |
1439 | return -ENOENT; | |
1440 | return 0; | |
1441 | } | |
1442 | return __traverse_mounts(path, flags, jumped, count, lookup_flags); | |
1443 | } | |
1444 | ||
cc53ce53 | 1445 | int follow_down_one(struct path *path) |
1da177e4 LT |
1446 | { |
1447 | struct vfsmount *mounted; | |
1448 | ||
1c755af4 | 1449 | mounted = lookup_mnt(path); |
1da177e4 | 1450 | if (mounted) { |
9393bd07 AV |
1451 | dput(path->dentry); |
1452 | mntput(path->mnt); | |
1453 | path->mnt = mounted; | |
1454 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1455 | return 1; |
1456 | } | |
1457 | return 0; | |
1458 | } | |
4d359507 | 1459 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1460 | |
9deed3eb AV |
1461 | /* |
1462 | * Follow down to the covering mount currently visible to userspace. At each | |
1463 | * point, the filesystem owning that dentry may be queried as to whether the | |
1464 | * caller is permitted to proceed or not. | |
1465 | */ | |
e1f19857 | 1466 | int follow_down(struct path *path, unsigned int flags) |
9deed3eb AV |
1467 | { |
1468 | struct vfsmount *mnt = path->mnt; | |
1469 | bool jumped; | |
e1f19857 | 1470 | int ret = traverse_mounts(path, &jumped, NULL, flags); |
9deed3eb AV |
1471 | |
1472 | if (path->mnt != mnt) | |
1473 | mntput(mnt); | |
1474 | return ret; | |
1475 | } | |
1476 | EXPORT_SYMBOL(follow_down); | |
1477 | ||
9875cf80 | 1478 | /* |
287548e4 AV |
1479 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1480 | * we meet a managed dentry that would need blocking. | |
9875cf80 | 1481 | */ |
3bd8bc89 | 1482 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path) |
9875cf80 | 1483 | { |
ea936aeb AV |
1484 | struct dentry *dentry = path->dentry; |
1485 | unsigned int flags = dentry->d_flags; | |
1486 | ||
1487 | if (likely(!(flags & DCACHE_MANAGED_DENTRY))) | |
1488 | return true; | |
1489 | ||
1490 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1491 | return false; | |
1492 | ||
62a7375e | 1493 | for (;;) { |
62a7375e IK |
1494 | /* |
1495 | * Don't forget we might have a non-mountpoint managed dentry | |
1496 | * that wants to block transit. | |
1497 | */ | |
ea936aeb AV |
1498 | if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { |
1499 | int res = dentry->d_op->d_manage(path, true); | |
1500 | if (res) | |
1501 | return res == -EISDIR; | |
1502 | flags = dentry->d_flags; | |
b8faf035 | 1503 | } |
62a7375e | 1504 | |
ea936aeb AV |
1505 | if (flags & DCACHE_MOUNTED) { |
1506 | struct mount *mounted = __lookup_mnt(path->mnt, dentry); | |
1507 | if (mounted) { | |
1508 | path->mnt = &mounted->mnt; | |
1509 | dentry = path->dentry = mounted->mnt.mnt_root; | |
bcba1e7d | 1510 | nd->state |= ND_JUMPED; |
03fa86e9 | 1511 | nd->next_seq = read_seqcount_begin(&dentry->d_seq); |
ea936aeb | 1512 | flags = dentry->d_flags; |
03fa86e9 AV |
1513 | // makes sure that non-RCU pathwalk could reach |
1514 | // this state. | |
20aac6c6 AV |
1515 | if (read_seqretry(&mount_lock, nd->m_seq)) |
1516 | return false; | |
ea936aeb AV |
1517 | continue; |
1518 | } | |
1519 | if (read_seqretry(&mount_lock, nd->m_seq)) | |
1520 | return false; | |
1521 | } | |
1522 | return !(flags & DCACHE_NEED_AUTOMOUNT); | |
9875cf80 | 1523 | } |
287548e4 AV |
1524 | } |
1525 | ||
db3c9ade | 1526 | static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, |
3bd8bc89 | 1527 | struct path *path) |
bd7c4b50 | 1528 | { |
9deed3eb | 1529 | bool jumped; |
db3c9ade | 1530 | int ret; |
bd7c4b50 | 1531 | |
db3c9ade AV |
1532 | path->mnt = nd->path.mnt; |
1533 | path->dentry = dentry; | |
c153007b | 1534 | if (nd->flags & LOOKUP_RCU) { |
03fa86e9 | 1535 | unsigned int seq = nd->next_seq; |
3bd8bc89 | 1536 | if (likely(__follow_mount_rcu(nd, path))) |
9deed3eb | 1537 | return 0; |
03fa86e9 | 1538 | // *path and nd->next_seq might've been clobbered |
c153007b AV |
1539 | path->mnt = nd->path.mnt; |
1540 | path->dentry = dentry; | |
03fa86e9 AV |
1541 | nd->next_seq = seq; |
1542 | if (!try_to_unlazy_next(nd, dentry)) | |
1543 | return -ECHILD; | |
c153007b | 1544 | } |
9deed3eb AV |
1545 | ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); |
1546 | if (jumped) { | |
1547 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) | |
1548 | ret = -EXDEV; | |
1549 | else | |
bcba1e7d | 1550 | nd->state |= ND_JUMPED; |
9deed3eb AV |
1551 | } |
1552 | if (unlikely(ret)) { | |
1553 | dput(path->dentry); | |
1554 | if (path->mnt != nd->path.mnt) | |
1555 | mntput(path->mnt); | |
bd7c4b50 AV |
1556 | } |
1557 | return ret; | |
1558 | } | |
1559 | ||
baa03890 | 1560 | /* |
f4fdace9 OD |
1561 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1562 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1563 | */ |
e3c13928 AV |
1564 | static struct dentry *lookup_dcache(const struct qstr *name, |
1565 | struct dentry *dir, | |
6c51e513 | 1566 | unsigned int flags) |
baa03890 | 1567 | { |
a89f8337 | 1568 | struct dentry *dentry = d_lookup(dir, name); |
bad61189 | 1569 | if (dentry) { |
a89f8337 AV |
1570 | int error = d_revalidate(dentry, flags); |
1571 | if (unlikely(error <= 0)) { | |
1572 | if (!error) | |
1573 | d_invalidate(dentry); | |
1574 | dput(dentry); | |
1575 | return ERR_PTR(error); | |
bad61189 MS |
1576 | } |
1577 | } | |
baa03890 NP |
1578 | return dentry; |
1579 | } | |
1580 | ||
44396f4b | 1581 | /* |
a03ece5f AV |
1582 | * Parent directory has inode locked exclusive. This is one |
1583 | * and only case when ->lookup() gets called on non in-lookup | |
1584 | * dentries - as the matter of fact, this only gets called | |
1585 | * when directory is guaranteed to have no in-lookup children | |
1586 | * at all. | |
44396f4b | 1587 | */ |
74d7970f NJ |
1588 | struct dentry *lookup_one_qstr_excl(const struct qstr *name, |
1589 | struct dentry *base, | |
1590 | unsigned int flags) | |
a3255546 | 1591 | { |
6c51e513 | 1592 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a03ece5f AV |
1593 | struct dentry *old; |
1594 | struct inode *dir = base->d_inode; | |
a3255546 | 1595 | |
6c51e513 | 1596 | if (dentry) |
bad61189 | 1597 | return dentry; |
a3255546 | 1598 | |
a03ece5f AV |
1599 | /* Don't create child dentry for a dead directory. */ |
1600 | if (unlikely(IS_DEADDIR(dir))) | |
1601 | return ERR_PTR(-ENOENT); | |
1602 | ||
6c51e513 AV |
1603 | dentry = d_alloc(base, name); |
1604 | if (unlikely(!dentry)) | |
1605 | return ERR_PTR(-ENOMEM); | |
1606 | ||
a03ece5f AV |
1607 | old = dir->i_op->lookup(dir, dentry, flags); |
1608 | if (unlikely(old)) { | |
1609 | dput(dentry); | |
1610 | dentry = old; | |
1611 | } | |
1612 | return dentry; | |
a3255546 | 1613 | } |
74d7970f | 1614 | EXPORT_SYMBOL(lookup_one_qstr_excl); |
a3255546 | 1615 | |
4cb64024 | 1616 | static struct dentry *lookup_fast(struct nameidata *nd) |
1da177e4 | 1617 | { |
31e6b01f | 1618 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1619 | int status = 1; |
9875cf80 | 1620 | |
b04f784e NP |
1621 | /* |
1622 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1623 | * of a false negative due to a concurrent rename, the caller is |
1624 | * going to fall back to non-racy lookup. | |
b04f784e | 1625 | */ |
31e6b01f | 1626 | if (nd->flags & LOOKUP_RCU) { |
03fa86e9 | 1627 | dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq); |
5d0f49c1 | 1628 | if (unlikely(!dentry)) { |
e36cffed | 1629 | if (!try_to_unlazy(nd)) |
20e34357 AV |
1630 | return ERR_PTR(-ECHILD); |
1631 | return NULL; | |
5d0f49c1 | 1632 | } |
5a18fff2 | 1633 | |
12f8ad4b LT |
1634 | /* |
1635 | * This sequence count validates that the parent had no | |
1636 | * changes while we did the lookup of the dentry above. | |
12f8ad4b | 1637 | */ |
4cb64024 | 1638 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
20e34357 | 1639 | return ERR_PTR(-ECHILD); |
5a18fff2 | 1640 | |
a89f8337 | 1641 | status = d_revalidate(dentry, nd->flags); |
c153007b | 1642 | if (likely(status > 0)) |
20e34357 | 1643 | return dentry; |
03fa86e9 | 1644 | if (!try_to_unlazy_next(nd, dentry)) |
20e34357 | 1645 | return ERR_PTR(-ECHILD); |
26ddb45e | 1646 | if (status == -ECHILD) |
209a7fb2 AV |
1647 | /* we'd been told to redo it in non-rcu mode */ |
1648 | status = d_revalidate(dentry, nd->flags); | |
5a18fff2 | 1649 | } else { |
e97cdc87 | 1650 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1651 | if (unlikely(!dentry)) |
20e34357 | 1652 | return NULL; |
a89f8337 | 1653 | status = d_revalidate(dentry, nd->flags); |
9875cf80 | 1654 | } |
5a18fff2 | 1655 | if (unlikely(status <= 0)) { |
e9742b53 | 1656 | if (!status) |
5d0f49c1 | 1657 | d_invalidate(dentry); |
5542aa2f | 1658 | dput(dentry); |
20e34357 | 1659 | return ERR_PTR(status); |
24643087 | 1660 | } |
20e34357 | 1661 | return dentry; |
697f514d MS |
1662 | } |
1663 | ||
1664 | /* Fast lookup failed, do it the slow way */ | |
88d8331a AV |
1665 | static struct dentry *__lookup_slow(const struct qstr *name, |
1666 | struct dentry *dir, | |
1667 | unsigned int flags) | |
697f514d | 1668 | { |
88d8331a | 1669 | struct dentry *dentry, *old; |
1936386e | 1670 | struct inode *inode = dir->d_inode; |
d9171b93 | 1671 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1672 | |
1936386e | 1673 | /* Don't go there if it's already dead */ |
94bdd655 | 1674 | if (unlikely(IS_DEADDIR(inode))) |
88d8331a | 1675 | return ERR_PTR(-ENOENT); |
94bdd655 | 1676 | again: |
d9171b93 | 1677 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 | 1678 | if (IS_ERR(dentry)) |
88d8331a | 1679 | return dentry; |
94bdd655 | 1680 | if (unlikely(!d_in_lookup(dentry))) { |
c64cd6e3 AV |
1681 | int error = d_revalidate(dentry, flags); |
1682 | if (unlikely(error <= 0)) { | |
1683 | if (!error) { | |
1684 | d_invalidate(dentry); | |
949a852e | 1685 | dput(dentry); |
c64cd6e3 | 1686 | goto again; |
949a852e | 1687 | } |
c64cd6e3 AV |
1688 | dput(dentry); |
1689 | dentry = ERR_PTR(error); | |
949a852e | 1690 | } |
94bdd655 AV |
1691 | } else { |
1692 | old = inode->i_op->lookup(inode, dentry, flags); | |
1693 | d_lookup_done(dentry); | |
1694 | if (unlikely(old)) { | |
1695 | dput(dentry); | |
1696 | dentry = old; | |
949a852e AV |
1697 | } |
1698 | } | |
e3c13928 | 1699 | return dentry; |
1da177e4 LT |
1700 | } |
1701 | ||
88d8331a AV |
1702 | static struct dentry *lookup_slow(const struct qstr *name, |
1703 | struct dentry *dir, | |
1704 | unsigned int flags) | |
1705 | { | |
1706 | struct inode *inode = dir->d_inode; | |
1707 | struct dentry *res; | |
1708 | inode_lock_shared(inode); | |
1709 | res = __lookup_slow(name, dir, flags); | |
1710 | inode_unlock_shared(inode); | |
1711 | return res; | |
1712 | } | |
1713 | ||
4609e1f1 | 1714 | static inline int may_lookup(struct mnt_idmap *idmap, |
ba73d987 | 1715 | struct nameidata *nd) |
52094c8a AV |
1716 | { |
1717 | if (nd->flags & LOOKUP_RCU) { | |
4609e1f1 | 1718 | int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
cdb67fde AV |
1719 | if (!err) // success, keep going |
1720 | return 0; | |
1721 | if (!try_to_unlazy(nd)) | |
1722 | return -ECHILD; // redo it all non-lazy | |
1723 | if (err != -ECHILD) // hard error | |
52094c8a | 1724 | return err; |
52094c8a | 1725 | } |
4609e1f1 | 1726 | return inode_permission(idmap, nd->inode, MAY_EXEC); |
52094c8a AV |
1727 | } |
1728 | ||
03fa86e9 | 1729 | static int reserve_stack(struct nameidata *nd, struct path *link) |
49055906 | 1730 | { |
49055906 AV |
1731 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) |
1732 | return -ELOOP; | |
4542576b AV |
1733 | |
1734 | if (likely(nd->depth != EMBEDDED_LEVELS)) | |
1735 | return 0; | |
1736 | if (likely(nd->stack != nd->internal)) | |
1737 | return 0; | |
60ef60c7 | 1738 | if (likely(nd_alloc_stack(nd))) |
49055906 | 1739 | return 0; |
60ef60c7 AV |
1740 | |
1741 | if (nd->flags & LOOKUP_RCU) { | |
1742 | // we need to grab link before we do unlazy. And we can't skip | |
1743 | // unlazy even if we fail to grab the link - cleanup needs it | |
03fa86e9 | 1744 | bool grabbed_link = legitimize_path(nd, link, nd->next_seq); |
60ef60c7 | 1745 | |
e5ca024e | 1746 | if (!try_to_unlazy(nd) || !grabbed_link) |
60ef60c7 AV |
1747 | return -ECHILD; |
1748 | ||
1749 | if (nd_alloc_stack(nd)) | |
1750 | return 0; | |
49055906 | 1751 | } |
60ef60c7 | 1752 | return -ENOMEM; |
49055906 AV |
1753 | } |
1754 | ||
b1a81972 AV |
1755 | enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; |
1756 | ||
06708adb | 1757 | static const char *pick_link(struct nameidata *nd, struct path *link, |
03fa86e9 | 1758 | struct inode *inode, int flags) |
d63ff28f | 1759 | { |
1cf2665b | 1760 | struct saved *last; |
ad6cc4c3 | 1761 | const char *res; |
03fa86e9 | 1762 | int error = reserve_stack(nd, link); |
ad6cc4c3 | 1763 | |
626de996 | 1764 | if (unlikely(error)) { |
49055906 | 1765 | if (!(nd->flags & LOOKUP_RCU)) |
bc40aee0 | 1766 | path_put(link); |
49055906 | 1767 | return ERR_PTR(error); |
626de996 | 1768 | } |
ab104923 | 1769 | last = nd->stack + nd->depth++; |
1cf2665b | 1770 | last->link = *link; |
fceef393 | 1771 | clear_delayed_call(&last->done); |
03fa86e9 | 1772 | last->seq = nd->next_seq; |
ad6cc4c3 | 1773 | |
b1a81972 | 1774 | if (flags & WALK_TRAILING) { |
ad6cc4c3 AV |
1775 | error = may_follow_link(nd, inode); |
1776 | if (unlikely(error)) | |
1777 | return ERR_PTR(error); | |
1778 | } | |
1779 | ||
dab741e0 MN |
1780 | if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || |
1781 | unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) | |
ad6cc4c3 AV |
1782 | return ERR_PTR(-ELOOP); |
1783 | ||
1784 | if (!(nd->flags & LOOKUP_RCU)) { | |
1785 | touch_atime(&last->link); | |
1786 | cond_resched(); | |
1787 | } else if (atime_needs_update(&last->link, inode)) { | |
e36cffed | 1788 | if (!try_to_unlazy(nd)) |
ad6cc4c3 AV |
1789 | return ERR_PTR(-ECHILD); |
1790 | touch_atime(&last->link); | |
1791 | } | |
1792 | ||
1793 | error = security_inode_follow_link(link->dentry, inode, | |
1794 | nd->flags & LOOKUP_RCU); | |
1795 | if (unlikely(error)) | |
1796 | return ERR_PTR(error); | |
1797 | ||
ad6cc4c3 AV |
1798 | res = READ_ONCE(inode->i_link); |
1799 | if (!res) { | |
1800 | const char * (*get)(struct dentry *, struct inode *, | |
1801 | struct delayed_call *); | |
1802 | get = inode->i_op->get_link; | |
1803 | if (nd->flags & LOOKUP_RCU) { | |
1804 | res = get(NULL, inode, &last->done); | |
e36cffed | 1805 | if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) |
ad6cc4c3 | 1806 | res = get(link->dentry, inode, &last->done); |
ad6cc4c3 AV |
1807 | } else { |
1808 | res = get(link->dentry, inode, &last->done); | |
1809 | } | |
1810 | if (!res) | |
1811 | goto all_done; | |
1812 | if (IS_ERR(res)) | |
1813 | return res; | |
1814 | } | |
1815 | if (*res == '/') { | |
1816 | error = nd_jump_root(nd); | |
1817 | if (unlikely(error)) | |
1818 | return ERR_PTR(error); | |
1819 | while (unlikely(*++res == '/')) | |
1820 | ; | |
1821 | } | |
1822 | if (*res) | |
1823 | return res; | |
1824 | all_done: // pure jump | |
1825 | put_link(nd); | |
1826 | return NULL; | |
d63ff28f AV |
1827 | } |
1828 | ||
3ddcd056 LT |
1829 | /* |
1830 | * Do we need to follow links? We _really_ want to be able | |
1831 | * to do this check without having to look at inode->i_op, | |
1832 | * so we keep a cache of "no, this doesn't need follow_link" | |
1833 | * for the common case. | |
03fa86e9 AV |
1834 | * |
1835 | * NOTE: dentry must be what nd->next_seq had been sampled from. | |
3ddcd056 | 1836 | */ |
b0417d2c | 1837 | static const char *step_into(struct nameidata *nd, int flags, |
a4f5b521 | 1838 | struct dentry *dentry) |
3ddcd056 | 1839 | { |
cbae4d12 | 1840 | struct path path; |
a4f5b521 | 1841 | struct inode *inode; |
3bd8bc89 | 1842 | int err = handle_mounts(nd, dentry, &path); |
cbae4d12 AV |
1843 | |
1844 | if (err < 0) | |
b0417d2c | 1845 | return ERR_PTR(err); |
3bd8bc89 | 1846 | inode = path.dentry->d_inode; |
cbae4d12 | 1847 | if (likely(!d_is_symlink(path.dentry)) || |
8c4efe22 | 1848 | ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || |
aca2903e | 1849 | (flags & WALK_NOFOLLOW)) { |
8f64fb1c | 1850 | /* not a symlink or should not follow */ |
3bd8bc89 AV |
1851 | if (nd->flags & LOOKUP_RCU) { |
1852 | if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) | |
1853 | return ERR_PTR(-ECHILD); | |
1854 | if (unlikely(!inode)) | |
1855 | return ERR_PTR(-ENOENT); | |
1856 | } else { | |
c99687a0 AV |
1857 | dput(nd->path.dentry); |
1858 | if (nd->path.mnt != path.mnt) | |
1859 | mntput(nd->path.mnt); | |
1860 | } | |
1861 | nd->path = path; | |
8f64fb1c | 1862 | nd->inode = inode; |
03fa86e9 | 1863 | nd->seq = nd->next_seq; |
b0417d2c | 1864 | return NULL; |
8f64fb1c | 1865 | } |
a7f77542 | 1866 | if (nd->flags & LOOKUP_RCU) { |
84f0cd9e | 1867 | /* make sure that d_is_symlink above matches inode */ |
03fa86e9 | 1868 | if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) |
b0417d2c | 1869 | return ERR_PTR(-ECHILD); |
84f0cd9e AV |
1870 | } else { |
1871 | if (path.mnt == nd->path.mnt) | |
1872 | mntget(path.mnt); | |
a7f77542 | 1873 | } |
03fa86e9 | 1874 | return pick_link(nd, &path, inode, flags); |
3ddcd056 LT |
1875 | } |
1876 | ||
b16c001d | 1877 | static struct dentry *follow_dotdot_rcu(struct nameidata *nd) |
957dd41d | 1878 | { |
12487f30 | 1879 | struct dentry *parent, *old; |
957dd41d | 1880 | |
12487f30 AV |
1881 | if (path_equal(&nd->path, &nd->root)) |
1882 | goto in_root; | |
1883 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
7ef482fa | 1884 | struct path path; |
efe772d6 | 1885 | unsigned seq; |
7ef482fa AV |
1886 | if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), |
1887 | &nd->root, &path, &seq)) | |
1888 | goto in_root; | |
efe772d6 AV |
1889 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1890 | return ERR_PTR(-ECHILD); | |
1891 | nd->path = path; | |
1892 | nd->inode = path.dentry->d_inode; | |
1893 | nd->seq = seq; | |
03fa86e9 | 1894 | // makes sure that non-RCU pathwalk could reach this state |
82ef0698 | 1895 | if (read_seqretry(&mount_lock, nd->m_seq)) |
efe772d6 AV |
1896 | return ERR_PTR(-ECHILD); |
1897 | /* we know that mountpoint was pinned */ | |
957dd41d | 1898 | } |
12487f30 AV |
1899 | old = nd->path.dentry; |
1900 | parent = old->d_parent; | |
03fa86e9 AV |
1901 | nd->next_seq = read_seqcount_begin(&parent->d_seq); |
1902 | // makes sure that non-RCU pathwalk could reach this state | |
82ef0698 | 1903 | if (read_seqcount_retry(&old->d_seq, nd->seq)) |
12487f30 AV |
1904 | return ERR_PTR(-ECHILD); |
1905 | if (unlikely(!path_connected(nd->path.mnt, parent))) | |
1906 | return ERR_PTR(-ECHILD); | |
1907 | return parent; | |
1908 | in_root: | |
82ef0698 | 1909 | if (read_seqretry(&mount_lock, nd->m_seq)) |
efe772d6 | 1910 | return ERR_PTR(-ECHILD); |
c2df1968 AV |
1911 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1912 | return ERR_PTR(-ECHILD); | |
03fa86e9 | 1913 | nd->next_seq = nd->seq; |
51c6546c | 1914 | return nd->path.dentry; |
957dd41d AV |
1915 | } |
1916 | ||
b16c001d | 1917 | static struct dentry *follow_dotdot(struct nameidata *nd) |
957dd41d | 1918 | { |
12487f30 AV |
1919 | struct dentry *parent; |
1920 | ||
1921 | if (path_equal(&nd->path, &nd->root)) | |
1922 | goto in_root; | |
1923 | if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { | |
2aa38470 AV |
1924 | struct path path; |
1925 | ||
1926 | if (!choose_mountpoint(real_mount(nd->path.mnt), | |
1927 | &nd->root, &path)) | |
1928 | goto in_root; | |
165200d6 AV |
1929 | path_put(&nd->path); |
1930 | nd->path = path; | |
2aa38470 | 1931 | nd->inode = path.dentry->d_inode; |
165200d6 AV |
1932 | if (unlikely(nd->flags & LOOKUP_NO_XDEV)) |
1933 | return ERR_PTR(-EXDEV); | |
957dd41d | 1934 | } |
12487f30 AV |
1935 | /* rare case of legitimate dget_parent()... */ |
1936 | parent = dget_parent(nd->path.dentry); | |
1937 | if (unlikely(!path_connected(nd->path.mnt, parent))) { | |
1938 | dput(parent); | |
1939 | return ERR_PTR(-ENOENT); | |
1940 | } | |
12487f30 AV |
1941 | return parent; |
1942 | ||
1943 | in_root: | |
c2df1968 AV |
1944 | if (unlikely(nd->flags & LOOKUP_BENEATH)) |
1945 | return ERR_PTR(-EXDEV); | |
51c6546c | 1946 | return dget(nd->path.dentry); |
957dd41d AV |
1947 | } |
1948 | ||
7521f22b | 1949 | static const char *handle_dots(struct nameidata *nd, int type) |
957dd41d AV |
1950 | { |
1951 | if (type == LAST_DOTDOT) { | |
7521f22b | 1952 | const char *error = NULL; |
c2df1968 | 1953 | struct dentry *parent; |
957dd41d AV |
1954 | |
1955 | if (!nd->root.mnt) { | |
7521f22b | 1956 | error = ERR_PTR(set_root(nd)); |
957dd41d AV |
1957 | if (error) |
1958 | return error; | |
1959 | } | |
1960 | if (nd->flags & LOOKUP_RCU) | |
b16c001d | 1961 | parent = follow_dotdot_rcu(nd); |
957dd41d | 1962 | else |
b16c001d | 1963 | parent = follow_dotdot(nd); |
c2df1968 AV |
1964 | if (IS_ERR(parent)) |
1965 | return ERR_CAST(parent); | |
a4f5b521 | 1966 | error = step_into(nd, WALK_NOFOLLOW, parent); |
c2df1968 | 1967 | if (unlikely(error)) |
957dd41d AV |
1968 | return error; |
1969 | ||
1970 | if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { | |
1971 | /* | |
1972 | * If there was a racing rename or mount along our | |
1973 | * path, then we can't be sure that ".." hasn't jumped | |
1974 | * above nd->root (and so userspace should retry or use | |
1975 | * some fallback). | |
1976 | */ | |
1977 | smp_rmb(); | |
82ef0698 | 1978 | if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) |
7521f22b | 1979 | return ERR_PTR(-EAGAIN); |
82ef0698 | 1980 | if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) |
7521f22b | 1981 | return ERR_PTR(-EAGAIN); |
957dd41d AV |
1982 | } |
1983 | } | |
7521f22b | 1984 | return NULL; |
957dd41d AV |
1985 | } |
1986 | ||
92d27016 | 1987 | static const char *walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1988 | { |
db3c9ade | 1989 | struct dentry *dentry; |
ce57dfc1 AV |
1990 | /* |
1991 | * "." and ".." are special - ".." especially so because it has | |
1992 | * to be able to know about the current root directory and | |
1993 | * parent relationships. | |
1994 | */ | |
4693a547 | 1995 | if (unlikely(nd->last_type != LAST_NORM)) { |
1c4ff1a8 | 1996 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 | 1997 | put_link(nd); |
7521f22b | 1998 | return handle_dots(nd, nd->last_type); |
4693a547 | 1999 | } |
4cb64024 | 2000 | dentry = lookup_fast(nd); |
20e34357 | 2001 | if (IS_ERR(dentry)) |
92d27016 | 2002 | return ERR_CAST(dentry); |
20e34357 | 2003 | if (unlikely(!dentry)) { |
db3c9ade AV |
2004 | dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); |
2005 | if (IS_ERR(dentry)) | |
92d27016 | 2006 | return ERR_CAST(dentry); |
ce57dfc1 | 2007 | } |
56676ec3 AV |
2008 | if (!(flags & WALK_MORE) && nd->depth) |
2009 | put_link(nd); | |
a4f5b521 | 2010 | return step_into(nd, flags, dentry); |
ce57dfc1 AV |
2011 | } |
2012 | ||
bfcfaa77 LT |
2013 | /* |
2014 | * We can do the critical dentry name comparison and hashing | |
2015 | * operations one word at a time, but we are limited to: | |
2016 | * | |
2017 | * - Architectures with fast unaligned word accesses. We could | |
2018 | * do a "get_unaligned()" if this helps and is sufficiently | |
2019 | * fast. | |
2020 | * | |
bfcfaa77 LT |
2021 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
2022 | * do not trap on the (extremely unlikely) case of a page | |
2023 | * crossing operation. | |
2024 | * | |
2025 | * - Furthermore, we need an efficient 64-bit compile for the | |
2026 | * 64-bit case in order to generate the "number of bytes in | |
2027 | * the final mask". Again, that could be replaced with a | |
2028 | * efficient population count instruction or similar. | |
2029 | */ | |
2030 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
2031 | ||
f68e556e | 2032 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 2033 | |
468a9428 | 2034 | #ifdef HASH_MIX |
bfcfaa77 | 2035 | |
468a9428 | 2036 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 2037 | |
468a9428 | 2038 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 2039 | /* |
2a18da7a GS |
2040 | * Register pressure in the mixing function is an issue, particularly |
2041 | * on 32-bit x86, but almost any function requires one state value and | |
2042 | * one temporary. Instead, use a function designed for two state values | |
2043 | * and no temporaries. | |
2044 | * | |
2045 | * This function cannot create a collision in only two iterations, so | |
2046 | * we have two iterations to achieve avalanche. In those two iterations, | |
2047 | * we have six layers of mixing, which is enough to spread one bit's | |
2048 | * influence out to 2^6 = 64 state bits. | |
2049 | * | |
2050 | * Rotate constants are scored by considering either 64 one-bit input | |
2051 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
2052 | * probability of that delta causing a change to each of the 128 output | |
2053 | * bits, using a sample of random initial states. | |
2054 | * | |
2055 | * The Shannon entropy of the computed probabilities is then summed | |
2056 | * to produce a score. Ideally, any input change has a 50% chance of | |
2057 | * toggling any given output bit. | |
2058 | * | |
2059 | * Mixing scores (in bits) for (12,45): | |
2060 | * Input delta: 1-bit 2-bit | |
2061 | * 1 round: 713.3 42542.6 | |
2062 | * 2 rounds: 2753.7 140389.8 | |
2063 | * 3 rounds: 5954.1 233458.2 | |
2064 | * 4 rounds: 7862.6 256672.2 | |
2065 | * Perfect: 8192 258048 | |
2066 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 2067 | */ |
2a18da7a GS |
2068 | #define HASH_MIX(x, y, a) \ |
2069 | ( x ^= (a), \ | |
2070 | y ^= x, x = rol64(x,12),\ | |
2071 | x += y, y = rol64(y,45),\ | |
2072 | y *= 9 ) | |
bfcfaa77 | 2073 | |
0fed3ac8 | 2074 | /* |
2a18da7a GS |
2075 | * Fold two longs into one 32-bit hash value. This must be fast, but |
2076 | * latency isn't quite as critical, as there is a fair bit of additional | |
2077 | * work done before the hash value is used. | |
0fed3ac8 | 2078 | */ |
2a18da7a | 2079 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 2080 | { |
2a18da7a GS |
2081 | y ^= x * GOLDEN_RATIO_64; |
2082 | y *= GOLDEN_RATIO_64; | |
2083 | return y >> 32; | |
0fed3ac8 GS |
2084 | } |
2085 | ||
bfcfaa77 LT |
2086 | #else /* 32-bit case */ |
2087 | ||
2a18da7a GS |
2088 | /* |
2089 | * Mixing scores (in bits) for (7,20): | |
2090 | * Input delta: 1-bit 2-bit | |
2091 | * 1 round: 330.3 9201.6 | |
2092 | * 2 rounds: 1246.4 25475.4 | |
2093 | * 3 rounds: 1907.1 31295.1 | |
2094 | * 4 rounds: 2042.3 31718.6 | |
2095 | * Perfect: 2048 31744 | |
2096 | * (32*64) (32*31/2 * 64) | |
2097 | */ | |
2098 | #define HASH_MIX(x, y, a) \ | |
2099 | ( x ^= (a), \ | |
2100 | y ^= x, x = rol32(x, 7),\ | |
2101 | x += y, y = rol32(y,20),\ | |
2102 | y *= 9 ) | |
bfcfaa77 | 2103 | |
2a18da7a | 2104 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 2105 | { |
2a18da7a GS |
2106 | /* Use arch-optimized multiply if one exists */ |
2107 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
2108 | } |
2109 | ||
bfcfaa77 LT |
2110 | #endif |
2111 | ||
2a18da7a GS |
2112 | /* |
2113 | * Return the hash of a string of known length. This is carfully | |
2114 | * designed to match hash_name(), which is the more critical function. | |
2115 | * In particular, we must end by hashing a final word containing 0..7 | |
2116 | * payload bytes, to match the way that hash_name() iterates until it | |
2117 | * finds the delimiter after the name. | |
2118 | */ | |
8387ff25 | 2119 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 2120 | { |
8387ff25 | 2121 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
2122 | |
2123 | for (;;) { | |
fcfd2fbf GS |
2124 | if (!len) |
2125 | goto done; | |
e419b4cc | 2126 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
2127 | if (len < sizeof(unsigned long)) |
2128 | break; | |
2a18da7a | 2129 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
2130 | name += sizeof(unsigned long); |
2131 | len -= sizeof(unsigned long); | |
bfcfaa77 | 2132 | } |
2a18da7a | 2133 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 2134 | done: |
2a18da7a | 2135 | return fold_hash(x, y); |
bfcfaa77 LT |
2136 | } |
2137 | EXPORT_SYMBOL(full_name_hash); | |
2138 | ||
fcfd2fbf | 2139 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2140 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2141 | { |
8387ff25 LT |
2142 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
2143 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
2144 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
2145 | ||
8387ff25 LT |
2146 | len = 0; |
2147 | goto inside; | |
2148 | ||
fcfd2fbf | 2149 | do { |
2a18da7a | 2150 | HASH_MIX(x, y, a); |
fcfd2fbf | 2151 | len += sizeof(unsigned long); |
8387ff25 | 2152 | inside: |
fcfd2fbf GS |
2153 | a = load_unaligned_zeropad(name+len); |
2154 | } while (!has_zero(a, &adata, &constants)); | |
2155 | ||
2156 | adata = prep_zero_mask(a, adata, &constants); | |
2157 | mask = create_zero_mask(adata); | |
2a18da7a | 2158 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 2159 | |
2a18da7a | 2160 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
2161 | } |
2162 | EXPORT_SYMBOL(hashlen_string); | |
2163 | ||
bfcfaa77 LT |
2164 | /* |
2165 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 2166 | * return the "hash_len" as the result. |
bfcfaa77 | 2167 | */ |
8387ff25 | 2168 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 2169 | { |
8387ff25 LT |
2170 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
2171 | unsigned long adata, bdata, mask, len; | |
36126f8f | 2172 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 2173 | |
8387ff25 LT |
2174 | len = 0; |
2175 | goto inside; | |
2176 | ||
bfcfaa77 | 2177 | do { |
2a18da7a | 2178 | HASH_MIX(x, y, a); |
bfcfaa77 | 2179 | len += sizeof(unsigned long); |
8387ff25 | 2180 | inside: |
e419b4cc | 2181 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
2182 | b = a ^ REPEAT_BYTE('/'); |
2183 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
2184 | ||
2185 | adata = prep_zero_mask(a, adata, &constants); | |
2186 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 2187 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 2188 | x ^= a & zero_bytemask(mask); |
36126f8f | 2189 | |
2a18da7a | 2190 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
2191 | } |
2192 | ||
2a18da7a | 2193 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 2194 | |
fcfd2fbf | 2195 | /* Return the hash of a string of known length */ |
8387ff25 | 2196 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 2197 | { |
8387ff25 | 2198 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 2199 | while (len--) |
fcfd2fbf | 2200 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
2201 | return end_name_hash(hash); |
2202 | } | |
ae942ae7 | 2203 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 2204 | |
fcfd2fbf | 2205 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 2206 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 2207 | { |
8387ff25 | 2208 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
2209 | unsigned long len = 0, c; |
2210 | ||
2211 | c = (unsigned char)*name; | |
e0ab7af9 | 2212 | while (c) { |
fcfd2fbf GS |
2213 | len++; |
2214 | hash = partial_name_hash(c, hash); | |
2215 | c = (unsigned char)name[len]; | |
e0ab7af9 | 2216 | } |
fcfd2fbf GS |
2217 | return hashlen_create(end_name_hash(hash), len); |
2218 | } | |
f2a031b6 | 2219 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 2220 | |
200e9ef7 LT |
2221 | /* |
2222 | * We know there's a real path component here of at least | |
2223 | * one character. | |
2224 | */ | |
8387ff25 | 2225 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2226 | { |
8387ff25 | 2227 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2228 | unsigned long len = 0, c; |
2229 | ||
2230 | c = (unsigned char)*name; | |
2231 | do { | |
2232 | len++; | |
2233 | hash = partial_name_hash(c, hash); | |
2234 | c = (unsigned char)name[len]; | |
2235 | } while (c && c != '/'); | |
d6bb3e90 | 2236 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2237 | } |
2238 | ||
bfcfaa77 LT |
2239 | #endif |
2240 | ||
1da177e4 LT |
2241 | /* |
2242 | * Name resolution. | |
ea3834d9 PM |
2243 | * This is the basic name resolution function, turning a pathname into |
2244 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2245 | * |
ea3834d9 PM |
2246 | * Returns 0 and nd will have valid dentry and mnt on success. |
2247 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2248 | */ |
6de88d72 | 2249 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2250 | { |
d8d4611a | 2251 | int depth = 0; // depth <= nd->depth |
1da177e4 | 2252 | int err; |
32cd7468 | 2253 | |
b4c03536 | 2254 | nd->last_type = LAST_ROOT; |
c108837e | 2255 | nd->flags |= LOOKUP_PARENT; |
9b5858e9 AV |
2256 | if (IS_ERR(name)) |
2257 | return PTR_ERR(name); | |
1da177e4 LT |
2258 | while (*name=='/') |
2259 | name++; | |
1a97d899 AV |
2260 | if (!*name) { |
2261 | nd->dir_mode = 0; // short-circuit the 'hardening' idiocy | |
9e18f10a | 2262 | return 0; |
1a97d899 | 2263 | } |
1da177e4 | 2264 | |
1da177e4 LT |
2265 | /* At this point we know we have a real path component. */ |
2266 | for(;;) { | |
4609e1f1 | 2267 | struct mnt_idmap *idmap; |
92d27016 | 2268 | const char *link; |
d6bb3e90 | 2269 | u64 hash_len; |
fe479a58 | 2270 | int type; |
1da177e4 | 2271 | |
4609e1f1 | 2272 | idmap = mnt_idmap(nd->path.mnt); |
4609e1f1 | 2273 | err = may_lookup(idmap, nd); |
2a18da7a | 2274 | if (err) |
3595e234 | 2275 | return err; |
1da177e4 | 2276 | |
8387ff25 | 2277 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2278 | |
fe479a58 | 2279 | type = LAST_NORM; |
d6bb3e90 | 2280 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2281 | case 2: |
200e9ef7 | 2282 | if (name[1] == '.') { |
fe479a58 | 2283 | type = LAST_DOTDOT; |
bcba1e7d | 2284 | nd->state |= ND_JUMPED; |
16c2cd71 | 2285 | } |
fe479a58 AV |
2286 | break; |
2287 | case 1: | |
2288 | type = LAST_DOT; | |
2289 | } | |
5a202bcd AV |
2290 | if (likely(type == LAST_NORM)) { |
2291 | struct dentry *parent = nd->path.dentry; | |
bcba1e7d | 2292 | nd->state &= ~ND_JUMPED; |
5a202bcd | 2293 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2294 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2295 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2296 | if (err < 0) |
3595e234 | 2297 | return err; |
d6bb3e90 LT |
2298 | hash_len = this.hash_len; |
2299 | name = this.name; | |
5a202bcd AV |
2300 | } |
2301 | } | |
fe479a58 | 2302 | |
d6bb3e90 LT |
2303 | nd->last.hash_len = hash_len; |
2304 | nd->last.name = name; | |
5f4a6a69 AV |
2305 | nd->last_type = type; |
2306 | ||
d6bb3e90 LT |
2307 | name += hashlen_len(hash_len); |
2308 | if (!*name) | |
bdf6cbf1 | 2309 | goto OK; |
200e9ef7 LT |
2310 | /* |
2311 | * If it wasn't NUL, we know it was '/'. Skip that | |
2312 | * slash, and continue until no more slashes. | |
2313 | */ | |
2314 | do { | |
d6bb3e90 LT |
2315 | name++; |
2316 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2317 | if (unlikely(!*name)) { |
2318 | OK: | |
d8d4611a | 2319 | /* pathname or trailing symlink, done */ |
c108837e | 2320 | if (!depth) { |
e67fe633 | 2321 | nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode); |
0f705953 | 2322 | nd->dir_mode = nd->inode->i_mode; |
c108837e | 2323 | nd->flags &= ~LOOKUP_PARENT; |
8620c238 | 2324 | return 0; |
c108837e | 2325 | } |
8620c238 | 2326 | /* last component of nested symlink */ |
d8d4611a | 2327 | name = nd->stack[--depth].name; |
8c4efe22 | 2328 | link = walk_component(nd, 0); |
1c4ff1a8 AV |
2329 | } else { |
2330 | /* not the last component */ | |
8c4efe22 | 2331 | link = walk_component(nd, WALK_MORE); |
8620c238 | 2332 | } |
92d27016 AV |
2333 | if (unlikely(link)) { |
2334 | if (IS_ERR(link)) | |
2335 | return PTR_ERR(link); | |
2336 | /* a symlink to follow */ | |
d8d4611a | 2337 | nd->stack[depth++].name = name; |
92d27016 AV |
2338 | name = link; |
2339 | continue; | |
31e6b01f | 2340 | } |
97242f99 AV |
2341 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2342 | if (nd->flags & LOOKUP_RCU) { | |
e36cffed | 2343 | if (!try_to_unlazy(nd)) |
97242f99 AV |
2344 | return -ECHILD; |
2345 | } | |
3595e234 | 2346 | return -ENOTDIR; |
97242f99 | 2347 | } |
1da177e4 | 2348 | } |
1da177e4 LT |
2349 | } |
2350 | ||
edc2b1da | 2351 | /* must be paired with terminate_walk() */ |
c8a53ee5 | 2352 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f | 2353 | { |
740a1678 | 2354 | int error; |
c8a53ee5 | 2355 | const char *s = nd->name->name; |
31e6b01f | 2356 | |
6c6ec2b0 JA |
2357 | /* LOOKUP_CACHED requires RCU, ask caller to retry */ |
2358 | if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) | |
2359 | return ERR_PTR(-EAGAIN); | |
2360 | ||
c0eb027e LT |
2361 | if (!*s) |
2362 | flags &= ~LOOKUP_RCU; | |
edc2b1da AV |
2363 | if (flags & LOOKUP_RCU) |
2364 | rcu_read_lock(); | |
03fa86e9 AV |
2365 | else |
2366 | nd->seq = nd->next_seq = 0; | |
c0eb027e | 2367 | |
bcba1e7d AV |
2368 | nd->flags = flags; |
2369 | nd->state |= ND_JUMPED; | |
ab87f9a5 AS |
2370 | |
2371 | nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); | |
2372 | nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); | |
2373 | smp_rmb(); | |
2374 | ||
bcba1e7d | 2375 | if (nd->state & ND_ROOT_PRESET) { |
b18825a7 DH |
2376 | struct dentry *root = nd->root.dentry; |
2377 | struct inode *inode = root->d_inode; | |
93893862 AV |
2378 | if (*s && unlikely(!d_can_lookup(root))) |
2379 | return ERR_PTR(-ENOTDIR); | |
5b6ca027 AV |
2380 | nd->path = nd->root; |
2381 | nd->inode = inode; | |
2382 | if (flags & LOOKUP_RCU) { | |
ab87f9a5 | 2383 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2384 | nd->root_seq = nd->seq; |
5b6ca027 AV |
2385 | } else { |
2386 | path_get(&nd->path); | |
2387 | } | |
368ee9ba | 2388 | return s; |
5b6ca027 AV |
2389 | } |
2390 | ||
31e6b01f | 2391 | nd->root.mnt = NULL; |
31e6b01f | 2392 | |
8db52c7e AS |
2393 | /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ |
2394 | if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { | |
740a1678 AS |
2395 | error = nd_jump_root(nd); |
2396 | if (unlikely(error)) | |
2397 | return ERR_PTR(error); | |
2398 | return s; | |
8db52c7e AS |
2399 | } |
2400 | ||
2401 | /* Relative pathname -- get the starting-point it is relative to. */ | |
2402 | if (nd->dfd == AT_FDCWD) { | |
e41f7d4e AV |
2403 | if (flags & LOOKUP_RCU) { |
2404 | struct fs_struct *fs = current->fs; | |
2405 | unsigned seq; | |
31e6b01f | 2406 | |
e41f7d4e AV |
2407 | do { |
2408 | seq = read_seqcount_begin(&fs->seq); | |
2409 | nd->path = fs->pwd; | |
ef55d917 | 2410 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2411 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2412 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2413 | } else { | |
2414 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2415 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2416 | } |
31e6b01f | 2417 | } else { |
582aa64a | 2418 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2419 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2420 | struct dentry *dentry; |
2421 | ||
2903ff01 | 2422 | if (!f.file) |
368ee9ba | 2423 | return ERR_PTR(-EBADF); |
31e6b01f | 2424 | |
2903ff01 | 2425 | dentry = f.file->f_path.dentry; |
31e6b01f | 2426 | |
edc2b1da AV |
2427 | if (*s && unlikely(!d_can_lookup(dentry))) { |
2428 | fdput(f); | |
2429 | return ERR_PTR(-ENOTDIR); | |
f52e0c11 | 2430 | } |
31e6b01f | 2431 | |
2903ff01 | 2432 | nd->path = f.file->f_path; |
e41f7d4e | 2433 | if (flags & LOOKUP_RCU) { |
34a26b99 AV |
2434 | nd->inode = nd->path.dentry->d_inode; |
2435 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2436 | } else { |
2903ff01 | 2437 | path_get(&nd->path); |
34a26b99 | 2438 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2439 | } |
34a26b99 | 2440 | fdput(f); |
31e6b01f | 2441 | } |
8db52c7e | 2442 | |
adb21d2b AS |
2443 | /* For scoped-lookups we need to set the root to the dirfd as well. */ |
2444 | if (flags & LOOKUP_IS_SCOPED) { | |
2445 | nd->root = nd->path; | |
2446 | if (flags & LOOKUP_RCU) { | |
2447 | nd->root_seq = nd->seq; | |
2448 | } else { | |
2449 | path_get(&nd->root); | |
bcba1e7d | 2450 | nd->state |= ND_ROOT_GRABBED; |
adb21d2b AS |
2451 | } |
2452 | } | |
2453 | return s; | |
9b4a9b14 AV |
2454 | } |
2455 | ||
1ccac622 | 2456 | static inline const char *lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2457 | { |
2458 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2459 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2460 | ||
c108837e | 2461 | return walk_component(nd, WALK_TRAILING); |
bd92d7fe AV |
2462 | } |
2463 | ||
4f757f3c AV |
2464 | static int handle_lookup_down(struct nameidata *nd) |
2465 | { | |
c153007b | 2466 | if (!(nd->flags & LOOKUP_RCU)) |
db3c9ade | 2467 | dget(nd->path.dentry); |
03fa86e9 | 2468 | nd->next_seq = nd->seq; |
a4f5b521 | 2469 | return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry)); |
4f757f3c AV |
2470 | } |
2471 | ||
6f672f7b | 2472 | /* Returns 0 and nd will be valid on success; Returns error, otherwise. */ |
c8a53ee5 | 2473 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2474 | { |
c8a53ee5 | 2475 | const char *s = path_init(nd, flags); |
bd92d7fe | 2476 | int err; |
31e6b01f | 2477 | |
9b5858e9 | 2478 | if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { |
4f757f3c | 2479 | err = handle_lookup_down(nd); |
5f336e72 AV |
2480 | if (unlikely(err < 0)) |
2481 | s = ERR_PTR(err); | |
4f757f3c AV |
2482 | } |
2483 | ||
1ccac622 AV |
2484 | while (!(err = link_path_walk(s, nd)) && |
2485 | (s = lookup_last(nd)) != NULL) | |
2486 | ; | |
4f0ed93f AV |
2487 | if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { |
2488 | err = handle_lookup_down(nd); | |
bcba1e7d | 2489 | nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... |
4f0ed93f | 2490 | } |
9f1fafee AV |
2491 | if (!err) |
2492 | err = complete_walk(nd); | |
bd92d7fe | 2493 | |
deb106c6 AV |
2494 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2495 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2496 | err = -ENOTDIR; |
625b6d10 AV |
2497 | if (!err) { |
2498 | *path = nd->path; | |
2499 | nd->path.mnt = NULL; | |
2500 | nd->path.dentry = NULL; | |
2501 | } | |
2502 | terminate_walk(nd); | |
bd92d7fe | 2503 | return err; |
ee0827cd | 2504 | } |
31e6b01f | 2505 | |
794ebcea | 2506 | int filename_lookup(int dfd, struct filename *name, unsigned flags, |
31d921c7 | 2507 | struct path *path, struct path *root) |
ee0827cd | 2508 | { |
894bc8c4 | 2509 | int retval; |
9883d185 | 2510 | struct nameidata nd; |
abc9f5be AV |
2511 | if (IS_ERR(name)) |
2512 | return PTR_ERR(name); | |
06422964 | 2513 | set_nameidata(&nd, dfd, name, root); |
c8a53ee5 | 2514 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2515 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2516 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2517 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2518 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2519 | |
f78570dd | 2520 | if (likely(!retval)) |
161aff1d AV |
2521 | audit_inode(name, path->dentry, |
2522 | flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); | |
9883d185 | 2523 | restore_nameidata(); |
020250f3 DK |
2524 | return retval; |
2525 | } | |
2526 | ||
6f672f7b | 2527 | /* Returns 0 and nd will be valid on success; Returns error, otherwise. */ |
c8a53ee5 | 2528 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2529 | struct path *parent) |
8bcb77fa | 2530 | { |
c8a53ee5 | 2531 | const char *s = path_init(nd, flags); |
9b5858e9 | 2532 | int err = link_path_walk(s, nd); |
8bcb77fa AV |
2533 | if (!err) |
2534 | err = complete_walk(nd); | |
391172c4 AV |
2535 | if (!err) { |
2536 | *parent = nd->path; | |
2537 | nd->path.mnt = NULL; | |
2538 | nd->path.dentry = NULL; | |
2539 | } | |
2540 | terminate_walk(nd); | |
8bcb77fa AV |
2541 | return err; |
2542 | } | |
2543 | ||
0766ec82 | 2544 | /* Note: this does not consume "name" */ |
74d7970f NJ |
2545 | static int __filename_parentat(int dfd, struct filename *name, |
2546 | unsigned int flags, struct path *parent, | |
2547 | struct qstr *last, int *type, | |
2548 | const struct path *root) | |
8bcb77fa AV |
2549 | { |
2550 | int retval; | |
9883d185 | 2551 | struct nameidata nd; |
8bcb77fa | 2552 | |
5c31b6ce | 2553 | if (IS_ERR(name)) |
0ee50b47 | 2554 | return PTR_ERR(name); |
74d7970f | 2555 | set_nameidata(&nd, dfd, name, root); |
c8a53ee5 | 2556 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2557 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2558 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2559 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2560 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2561 | if (likely(!retval)) { |
2562 | *last = nd.last; | |
2563 | *type = nd.last_type; | |
c9b07eab | 2564 | audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); |
391172c4 | 2565 | } |
9883d185 | 2566 | restore_nameidata(); |
0ee50b47 DK |
2567 | return retval; |
2568 | } | |
2569 | ||
74d7970f NJ |
2570 | static int filename_parentat(int dfd, struct filename *name, |
2571 | unsigned int flags, struct path *parent, | |
2572 | struct qstr *last, int *type) | |
2573 | { | |
2574 | return __filename_parentat(dfd, name, flags, parent, last, type, NULL); | |
2575 | } | |
2576 | ||
79714f72 | 2577 | /* does lookup, returns the object with parent locked */ |
74d016ec | 2578 | static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path) |
5590ff0d | 2579 | { |
5c31b6ce | 2580 | struct dentry *d; |
391172c4 | 2581 | struct qstr last; |
0ee50b47 | 2582 | int type, error; |
51689104 | 2583 | |
74d016ec | 2584 | error = filename_parentat(dfd, name, 0, path, &last, &type); |
0ee50b47 DK |
2585 | if (error) |
2586 | return ERR_PTR(error); | |
5c31b6ce | 2587 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2588 | path_put(path); |
5c31b6ce | 2589 | return ERR_PTR(-EINVAL); |
79714f72 | 2590 | } |
5955102c | 2591 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
74d7970f | 2592 | d = lookup_one_qstr_excl(&last, path->dentry, 0); |
79714f72 | 2593 | if (IS_ERR(d)) { |
5955102c | 2594 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2595 | path_put(path); |
79714f72 | 2596 | } |
79714f72 | 2597 | return d; |
5590ff0d UD |
2598 | } |
2599 | ||
0766ec82 SB |
2600 | struct dentry *kern_path_locked(const char *name, struct path *path) |
2601 | { | |
2602 | struct filename *filename = getname_kernel(name); | |
74d016ec | 2603 | struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path); |
0766ec82 SB |
2604 | |
2605 | putname(filename); | |
2606 | return res; | |
2607 | } | |
2608 | ||
74d016ec AV |
2609 | struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path) |
2610 | { | |
2611 | struct filename *filename = getname(name); | |
2612 | struct dentry *res = __kern_path_locked(dfd, filename, path); | |
2613 | ||
2614 | putname(filename); | |
2615 | return res; | |
2616 | } | |
2617 | EXPORT_SYMBOL(user_path_locked_at); | |
2618 | ||
d1811465 AV |
2619 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2620 | { | |
794ebcea SB |
2621 | struct filename *filename = getname_kernel(name); |
2622 | int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); | |
2623 | ||
2624 | putname(filename); | |
2625 | return ret; | |
2626 | ||
d1811465 | 2627 | } |
4d359507 | 2628 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2629 | |
74d7970f NJ |
2630 | /** |
2631 | * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair | |
2632 | * @filename: filename structure | |
2633 | * @flags: lookup flags | |
2634 | * @parent: pointer to struct path to fill | |
2635 | * @last: last component | |
2636 | * @type: type of the last component | |
2637 | * @root: pointer to struct path of the base directory | |
2638 | */ | |
2639 | int vfs_path_parent_lookup(struct filename *filename, unsigned int flags, | |
2640 | struct path *parent, struct qstr *last, int *type, | |
2641 | const struct path *root) | |
2642 | { | |
2643 | return __filename_parentat(AT_FDCWD, filename, flags, parent, last, | |
2644 | type, root); | |
2645 | } | |
2646 | EXPORT_SYMBOL(vfs_path_parent_lookup); | |
2647 | ||
16f18200 JJS |
2648 | /** |
2649 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2650 | * @dentry: pointer to dentry of the base directory | |
2651 | * @mnt: pointer to vfs mount of the base directory | |
2652 | * @name: pointer to file name | |
2653 | * @flags: lookup flags | |
e0a01249 | 2654 | * @path: pointer to struct path to fill |
16f18200 JJS |
2655 | */ |
2656 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2657 | const char *name, unsigned int flags, | |
e0a01249 | 2658 | struct path *path) |
16f18200 | 2659 | { |
794ebcea | 2660 | struct filename *filename; |
9ad1aaa6 | 2661 | struct path root = {.mnt = mnt, .dentry = dentry}; |
794ebcea SB |
2662 | int ret; |
2663 | ||
2664 | filename = getname_kernel(name); | |
9ad1aaa6 | 2665 | /* the first argument of filename_lookup() is ignored with root */ |
794ebcea SB |
2666 | ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); |
2667 | putname(filename); | |
2668 | return ret; | |
16f18200 | 2669 | } |
4d359507 | 2670 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2671 | |
4609e1f1 | 2672 | static int lookup_one_common(struct mnt_idmap *idmap, |
c2fd68b6 CB |
2673 | const char *name, struct dentry *base, int len, |
2674 | struct qstr *this) | |
057f6c01 | 2675 | { |
3c95f0dc AV |
2676 | this->name = name; |
2677 | this->len = len; | |
2678 | this->hash = full_name_hash(base, name, len); | |
6a96ba54 | 2679 | if (!len) |
3c95f0dc | 2680 | return -EACCES; |
6a96ba54 | 2681 | |
42c3732f CL |
2682 | if (is_dot_dotdot(name, len)) |
2683 | return -EACCES; | |
21d8a15a | 2684 | |
6a96ba54 | 2685 | while (len--) { |
3c95f0dc | 2686 | unsigned int c = *(const unsigned char *)name++; |
6a96ba54 | 2687 | if (c == '/' || c == '\0') |
3c95f0dc | 2688 | return -EACCES; |
6a96ba54 | 2689 | } |
5a202bcd AV |
2690 | /* |
2691 | * See if the low-level filesystem might want | |
2692 | * to use its own hash.. | |
2693 | */ | |
2694 | if (base->d_flags & DCACHE_OP_HASH) { | |
3c95f0dc | 2695 | int err = base->d_op->d_hash(base, this); |
5a202bcd | 2696 | if (err < 0) |
3c95f0dc | 2697 | return err; |
5a202bcd | 2698 | } |
eead1911 | 2699 | |
4609e1f1 | 2700 | return inode_permission(idmap, base->d_inode, MAY_EXEC); |
3c95f0dc AV |
2701 | } |
2702 | ||
0da0b7fd DH |
2703 | /** |
2704 | * try_lookup_one_len - filesystem helper to lookup single pathname component | |
2705 | * @name: pathname component to lookup | |
2706 | * @base: base directory to lookup from | |
2707 | * @len: maximum length @len should be interpreted to | |
2708 | * | |
2709 | * Look up a dentry by name in the dcache, returning NULL if it does not | |
2710 | * currently exist. The function does not try to create a dentry. | |
2711 | * | |
2712 | * Note that this routine is purely a helper for filesystem usage and should | |
2713 | * not be called by generic code. | |
2714 | * | |
2715 | * The caller must hold base->i_mutex. | |
2716 | */ | |
2717 | struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) | |
2718 | { | |
2719 | struct qstr this; | |
2720 | int err; | |
2721 | ||
2722 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2723 | ||
4609e1f1 | 2724 | err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); |
0da0b7fd DH |
2725 | if (err) |
2726 | return ERR_PTR(err); | |
2727 | ||
2728 | return lookup_dcache(&this, base, 0); | |
2729 | } | |
2730 | EXPORT_SYMBOL(try_lookup_one_len); | |
2731 | ||
3c95f0dc AV |
2732 | /** |
2733 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2734 | * @name: pathname component to lookup | |
2735 | * @base: base directory to lookup from | |
2736 | * @len: maximum length @len should be interpreted to | |
2737 | * | |
2738 | * Note that this routine is purely a helper for filesystem usage and should | |
2739 | * not be called by generic code. | |
2740 | * | |
2741 | * The caller must hold base->i_mutex. | |
2742 | */ | |
2743 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2744 | { | |
8613a209 | 2745 | struct dentry *dentry; |
3c95f0dc AV |
2746 | struct qstr this; |
2747 | int err; | |
2748 | ||
2749 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2750 | ||
4609e1f1 | 2751 | err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); |
cda309de MS |
2752 | if (err) |
2753 | return ERR_PTR(err); | |
2754 | ||
8613a209 AV |
2755 | dentry = lookup_dcache(&this, base, 0); |
2756 | return dentry ? dentry : __lookup_slow(&this, base, 0); | |
057f6c01 | 2757 | } |
4d359507 | 2758 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2759 | |
c2fd68b6 CB |
2760 | /** |
2761 | * lookup_one - filesystem helper to lookup single pathname component | |
4609e1f1 | 2762 | * @idmap: idmap of the mount the lookup is performed from |
c2fd68b6 CB |
2763 | * @name: pathname component to lookup |
2764 | * @base: base directory to lookup from | |
2765 | * @len: maximum length @len should be interpreted to | |
2766 | * | |
2767 | * Note that this routine is purely a helper for filesystem usage and should | |
2768 | * not be called by generic code. | |
2769 | * | |
2770 | * The caller must hold base->i_mutex. | |
2771 | */ | |
4609e1f1 | 2772 | struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name, |
c2fd68b6 CB |
2773 | struct dentry *base, int len) |
2774 | { | |
2775 | struct dentry *dentry; | |
2776 | struct qstr this; | |
2777 | int err; | |
2778 | ||
2779 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2780 | ||
4609e1f1 | 2781 | err = lookup_one_common(idmap, name, base, len, &this); |
c2fd68b6 CB |
2782 | if (err) |
2783 | return ERR_PTR(err); | |
2784 | ||
2785 | dentry = lookup_dcache(&this, base, 0); | |
2786 | return dentry ? dentry : __lookup_slow(&this, base, 0); | |
2787 | } | |
2788 | EXPORT_SYMBOL(lookup_one); | |
2789 | ||
bbddca8e | 2790 | /** |
00675017 | 2791 | * lookup_one_unlocked - filesystem helper to lookup single pathname component |
4609e1f1 | 2792 | * @idmap: idmap of the mount the lookup is performed from |
bbddca8e N |
2793 | * @name: pathname component to lookup |
2794 | * @base: base directory to lookup from | |
2795 | * @len: maximum length @len should be interpreted to | |
2796 | * | |
2797 | * Note that this routine is purely a helper for filesystem usage and should | |
2798 | * not be called by generic code. | |
2799 | * | |
2800 | * Unlike lookup_one_len, it should be called without the parent | |
2801 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2802 | */ | |
4609e1f1 | 2803 | struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap, |
00675017 CB |
2804 | const char *name, struct dentry *base, |
2805 | int len) | |
bbddca8e N |
2806 | { |
2807 | struct qstr this; | |
bbddca8e | 2808 | int err; |
20d00ee8 | 2809 | struct dentry *ret; |
bbddca8e | 2810 | |
4609e1f1 | 2811 | err = lookup_one_common(idmap, name, base, len, &this); |
bbddca8e N |
2812 | if (err) |
2813 | return ERR_PTR(err); | |
2814 | ||
20d00ee8 LT |
2815 | ret = lookup_dcache(&this, base, 0); |
2816 | if (!ret) | |
2817 | ret = lookup_slow(&this, base, 0); | |
2818 | return ret; | |
bbddca8e | 2819 | } |
00675017 CB |
2820 | EXPORT_SYMBOL(lookup_one_unlocked); |
2821 | ||
2822 | /** | |
2823 | * lookup_one_positive_unlocked - filesystem helper to lookup single | |
2824 | * pathname component | |
4609e1f1 | 2825 | * @idmap: idmap of the mount the lookup is performed from |
00675017 CB |
2826 | * @name: pathname component to lookup |
2827 | * @base: base directory to lookup from | |
2828 | * @len: maximum length @len should be interpreted to | |
2829 | * | |
2830 | * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns | |
2831 | * known positive or ERR_PTR(). This is what most of the users want. | |
2832 | * | |
2833 | * Note that pinned negative with unlocked parent _can_ become positive at any | |
2834 | * time, so callers of lookup_one_unlocked() need to be very careful; pinned | |
2835 | * positives have >d_inode stable, so this one avoids such problems. | |
2836 | * | |
2837 | * Note that this routine is purely a helper for filesystem usage and should | |
2838 | * not be called by generic code. | |
2839 | * | |
2840 | * The helper should be called without i_mutex held. | |
2841 | */ | |
4609e1f1 | 2842 | struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap, |
00675017 CB |
2843 | const char *name, |
2844 | struct dentry *base, int len) | |
2845 | { | |
4609e1f1 | 2846 | struct dentry *ret = lookup_one_unlocked(idmap, name, base, len); |
00675017 CB |
2847 | |
2848 | if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { | |
2849 | dput(ret); | |
2850 | ret = ERR_PTR(-ENOENT); | |
2851 | } | |
2852 | return ret; | |
2853 | } | |
2854 | EXPORT_SYMBOL(lookup_one_positive_unlocked); | |
2855 | ||
2856 | /** | |
2857 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2858 | * @name: pathname component to lookup | |
2859 | * @base: base directory to lookup from | |
2860 | * @len: maximum length @len should be interpreted to | |
2861 | * | |
2862 | * Note that this routine is purely a helper for filesystem usage and should | |
2863 | * not be called by generic code. | |
2864 | * | |
2865 | * Unlike lookup_one_len, it should be called without the parent | |
2866 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2867 | */ | |
2868 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2869 | struct dentry *base, int len) | |
2870 | { | |
4609e1f1 | 2871 | return lookup_one_unlocked(&nop_mnt_idmap, name, base, len); |
00675017 | 2872 | } |
bbddca8e N |
2873 | EXPORT_SYMBOL(lookup_one_len_unlocked); |
2874 | ||
6c2d4798 AV |
2875 | /* |
2876 | * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) | |
2877 | * on negatives. Returns known positive or ERR_PTR(); that's what | |
2878 | * most of the users want. Note that pinned negative with unlocked parent | |
2879 | * _can_ become positive at any time, so callers of lookup_one_len_unlocked() | |
2880 | * need to be very careful; pinned positives have ->d_inode stable, so | |
2881 | * this one avoids such problems. | |
2882 | */ | |
2883 | struct dentry *lookup_positive_unlocked(const char *name, | |
2884 | struct dentry *base, int len) | |
2885 | { | |
4609e1f1 | 2886 | return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len); |
6c2d4798 AV |
2887 | } |
2888 | EXPORT_SYMBOL(lookup_positive_unlocked); | |
2889 | ||
eedf265a EB |
2890 | #ifdef CONFIG_UNIX98_PTYS |
2891 | int path_pts(struct path *path) | |
2892 | { | |
2893 | /* Find something mounted on "pts" in the same directory as | |
2894 | * the input path. | |
2895 | */ | |
a6a7eb76 AV |
2896 | struct dentry *parent = dget_parent(path->dentry); |
2897 | struct dentry *child; | |
19f6028a | 2898 | struct qstr this = QSTR_INIT("pts", 3); |
eedf265a | 2899 | |
a6a7eb76 AV |
2900 | if (unlikely(!path_connected(path->mnt, parent))) { |
2901 | dput(parent); | |
63b27720 | 2902 | return -ENOENT; |
a6a7eb76 | 2903 | } |
63b27720 AV |
2904 | dput(path->dentry); |
2905 | path->dentry = parent; | |
eedf265a | 2906 | child = d_hash_and_lookup(parent, &this); |
0d5a4f8f | 2907 | if (IS_ERR_OR_NULL(child)) |
eedf265a EB |
2908 | return -ENOENT; |
2909 | ||
2910 | path->dentry = child; | |
2911 | dput(parent); | |
e1f19857 | 2912 | follow_down(path, 0); |
eedf265a EB |
2913 | return 0; |
2914 | } | |
2915 | #endif | |
2916 | ||
1fa1e7f6 AW |
2917 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2918 | struct path *path, int *empty) | |
1da177e4 | 2919 | { |
794ebcea SB |
2920 | struct filename *filename = getname_flags(name, flags, empty); |
2921 | int ret = filename_lookup(dfd, filename, flags, path, NULL); | |
2922 | ||
2923 | putname(filename); | |
2924 | return ret; | |
1da177e4 | 2925 | } |
b853a161 | 2926 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2927 | |
9452e93e | 2928 | int __check_sticky(struct mnt_idmap *idmap, struct inode *dir, |
ba73d987 | 2929 | struct inode *inode) |
1da177e4 | 2930 | { |
8e96e3b7 | 2931 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2932 | |
e67fe633 | 2933 | if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid)) |
1da177e4 | 2934 | return 0; |
e67fe633 | 2935 | if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid)) |
1da177e4 | 2936 | return 0; |
9452e93e | 2937 | return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER); |
1da177e4 | 2938 | } |
cbdf35bc | 2939 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2940 | |
2941 | /* | |
2942 | * Check whether we can remove a link victim from directory dir, check | |
2943 | * whether the type of victim is right. | |
2944 | * 1. We can't do it if dir is read-only (done in permission()) | |
2945 | * 2. We should have write and exec permissions on dir | |
2946 | * 3. We can't remove anything from append-only dir | |
2947 | * 4. We can't do anything with immutable dir (done in permission()) | |
2948 | * 5. If the sticky bit on dir is set we should either | |
2949 | * a. be owner of dir, or | |
2950 | * b. be owner of victim, or | |
2951 | * c. have CAP_FOWNER capability | |
2952 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2953 | * links pointing to it. | |
0bd23d09 EB |
2954 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2955 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2956 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2957 | * 10. We can't remove a root or mountpoint. | |
2958 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2959 | * nfs_async_unlink(). |
2960 | */ | |
4609e1f1 | 2961 | static int may_delete(struct mnt_idmap *idmap, struct inode *dir, |
ba73d987 | 2962 | struct dentry *victim, bool isdir) |
1da177e4 | 2963 | { |
63afdfc7 | 2964 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2965 | int error; |
2966 | ||
b18825a7 | 2967 | if (d_is_negative(victim)) |
1da177e4 | 2968 | return -ENOENT; |
b18825a7 | 2969 | BUG_ON(!inode); |
1da177e4 LT |
2970 | |
2971 | BUG_ON(victim->d_parent->d_inode != dir); | |
593d1ce8 EB |
2972 | |
2973 | /* Inode writeback is not safe when the uid or gid are invalid. */ | |
e67fe633 CB |
2974 | if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || |
2975 | !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) | |
593d1ce8 EB |
2976 | return -EOVERFLOW; |
2977 | ||
4fa6b5ec | 2978 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2979 | |
4609e1f1 | 2980 | error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2981 | if (error) |
2982 | return error; | |
2983 | if (IS_APPEND(dir)) | |
2984 | return -EPERM; | |
b18825a7 | 2985 | |
9452e93e | 2986 | if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) || |
ba73d987 | 2987 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || |
4609e1f1 | 2988 | HAS_UNMAPPED_ID(idmap, inode)) |
1da177e4 LT |
2989 | return -EPERM; |
2990 | if (isdir) { | |
44b1d530 | 2991 | if (!d_is_dir(victim)) |
1da177e4 LT |
2992 | return -ENOTDIR; |
2993 | if (IS_ROOT(victim)) | |
2994 | return -EBUSY; | |
44b1d530 | 2995 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2996 | return -EISDIR; |
2997 | if (IS_DEADDIR(dir)) | |
2998 | return -ENOENT; | |
2999 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
3000 | return -EBUSY; | |
3001 | return 0; | |
3002 | } | |
3003 | ||
3004 | /* Check whether we can create an object with dentry child in directory | |
3005 | * dir. | |
3006 | * 1. We can't do it if child already exists (open has special treatment for | |
3007 | * this case, but since we are inlined it's OK) | |
3008 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
3009 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
3010 | * 4. We should have write and exec permissions on dir | |
3011 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 3012 | */ |
4609e1f1 | 3013 | static inline int may_create(struct mnt_idmap *idmap, |
ba73d987 | 3014 | struct inode *dir, struct dentry *child) |
1da177e4 | 3015 | { |
14e972b4 | 3016 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
3017 | if (child->d_inode) |
3018 | return -EEXIST; | |
3019 | if (IS_DEADDIR(dir)) | |
3020 | return -ENOENT; | |
4609e1f1 | 3021 | if (!fsuidgid_has_mapping(dir->i_sb, idmap)) |
036d5236 | 3022 | return -EOVERFLOW; |
8e538913 | 3023 | |
4609e1f1 | 3024 | return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
3025 | } |
3026 | ||
a8b00268 | 3027 | // p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held |
9bc37e04 | 3028 | static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2) |
1da177e4 | 3029 | { |
a8b00268 | 3030 | struct dentry *p = p1, *q = p2, *r; |
1da177e4 | 3031 | |
a8b00268 AV |
3032 | while ((r = p->d_parent) != p2 && r != p) |
3033 | p = r; | |
3034 | if (r == p2) { | |
3035 | // p is a child of p2 and an ancestor of p1 or p1 itself | |
5955102c | 3036 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
22e111ed | 3037 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2); |
e2761a11 | 3038 | return p; |
1da177e4 | 3039 | } |
a8b00268 AV |
3040 | // p is the root of connected component that contains p1 |
3041 | // p2 does not occur on the path from p to p1 | |
3042 | while ((r = q->d_parent) != p1 && r != p && r != q) | |
3043 | q = r; | |
3044 | if (r == p1) { | |
3045 | // q is a child of p1 and an ancestor of p2 or p2 itself | |
5955102c | 3046 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
a8b00268 AV |
3047 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); |
3048 | return q; | |
3049 | } else if (likely(r == p)) { | |
3050 | // both p2 and p1 are descendents of p | |
3051 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
3052 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
3053 | return NULL; | |
3054 | } else { // no common ancestor at the time we'd been called | |
3055 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); | |
3056 | return ERR_PTR(-EXDEV); | |
1da177e4 | 3057 | } |
1da177e4 | 3058 | } |
9bc37e04 AV |
3059 | |
3060 | /* | |
3061 | * p1 and p2 should be directories on the same fs. | |
3062 | */ | |
3063 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
3064 | { | |
3065 | if (p1 == p2) { | |
3066 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
3067 | return NULL; | |
3068 | } | |
3069 | ||
3070 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); | |
3071 | return lock_two_directories(p1, p2); | |
3072 | } | |
4d359507 | 3073 | EXPORT_SYMBOL(lock_rename); |
1da177e4 | 3074 | |
9bc37e04 AV |
3075 | /* |
3076 | * c1 and p2 should be on the same fs. | |
3077 | */ | |
3078 | struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2) | |
3079 | { | |
3080 | if (READ_ONCE(c1->d_parent) == p2) { | |
3081 | /* | |
3082 | * hopefully won't need to touch ->s_vfs_rename_mutex at all. | |
3083 | */ | |
3084 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); | |
3085 | /* | |
3086 | * now that p2 is locked, nobody can move in or out of it, | |
3087 | * so the test below is safe. | |
3088 | */ | |
3089 | if (likely(c1->d_parent == p2)) | |
3090 | return NULL; | |
3091 | ||
3092 | /* | |
3093 | * c1 got moved out of p2 while we'd been taking locks; | |
3094 | * unlock and fall back to slow case. | |
3095 | */ | |
3096 | inode_unlock(p2->d_inode); | |
3097 | } | |
3098 | ||
3099 | mutex_lock(&c1->d_sb->s_vfs_rename_mutex); | |
3100 | /* | |
3101 | * nobody can move out of any directories on this fs. | |
3102 | */ | |
3103 | if (likely(c1->d_parent != p2)) | |
3104 | return lock_two_directories(c1->d_parent, p2); | |
3105 | ||
3106 | /* | |
3107 | * c1 got moved into p2 while we were taking locks; | |
3108 | * we need p2 locked and ->s_vfs_rename_mutex unlocked, | |
3109 | * for consistency with lock_rename(). | |
3110 | */ | |
3111 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); | |
3112 | mutex_unlock(&c1->d_sb->s_vfs_rename_mutex); | |
3113 | return NULL; | |
3114 | } | |
3115 | EXPORT_SYMBOL(lock_rename_child); | |
3116 | ||
1da177e4 LT |
3117 | void unlock_rename(struct dentry *p1, struct dentry *p2) |
3118 | { | |
5955102c | 3119 | inode_unlock(p1->d_inode); |
1da177e4 | 3120 | if (p1 != p2) { |
5955102c | 3121 | inode_unlock(p2->d_inode); |
fc64005c | 3122 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
3123 | } |
3124 | } | |
4d359507 | 3125 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 3126 | |
1639a49c YX |
3127 | /** |
3128 | * vfs_prepare_mode - prepare the mode to be used for a new inode | |
9452e93e | 3129 | * @idmap: idmap of the mount the inode was found from |
1639a49c YX |
3130 | * @dir: parent directory of the new inode |
3131 | * @mode: mode of the new inode | |
3132 | * @mask_perms: allowed permission by the vfs | |
3133 | * @type: type of file to be created | |
3134 | * | |
3135 | * This helper consolidates and enforces vfs restrictions on the @mode of a new | |
3136 | * object to be created. | |
3137 | * | |
3138 | * Umask stripping depends on whether the filesystem supports POSIX ACLs (see | |
3139 | * the kernel documentation for mode_strip_umask()). Moving umask stripping | |
3140 | * after setgid stripping allows the same ordering for both non-POSIX ACL and | |
3141 | * POSIX ACL supporting filesystems. | |
3142 | * | |
3143 | * Note that it's currently valid for @type to be 0 if a directory is created. | |
3144 | * Filesystems raise that flag individually and we need to check whether each | |
3145 | * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a | |
3146 | * non-zero type. | |
3147 | * | |
3148 | * Returns: mode to be passed to the filesystem | |
3149 | */ | |
9452e93e | 3150 | static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap, |
1639a49c YX |
3151 | const struct inode *dir, umode_t mode, |
3152 | umode_t mask_perms, umode_t type) | |
3153 | { | |
9452e93e | 3154 | mode = mode_strip_sgid(idmap, dir, mode); |
1639a49c YX |
3155 | mode = mode_strip_umask(dir, mode); |
3156 | ||
3157 | /* | |
3158 | * Apply the vfs mandated allowed permission mask and set the type of | |
3159 | * file to be created before we call into the filesystem. | |
3160 | */ | |
3161 | mode &= (mask_perms & ~S_IFMT); | |
3162 | mode |= (type & S_IFMT); | |
3163 | ||
3164 | return mode; | |
3165 | } | |
3166 | ||
6521f891 CB |
3167 | /** |
3168 | * vfs_create - create new file | |
abf08576 | 3169 | * @idmap: idmap of the mount the inode was found from |
6521f891 CB |
3170 | * @dir: inode of @dentry |
3171 | * @dentry: pointer to dentry of the base directory | |
3172 | * @mode: mode of the new file | |
3173 | * @want_excl: whether the file must not yet exist | |
3174 | * | |
3175 | * Create a new file. | |
3176 | * | |
abf08576 CB |
3177 | * If the inode has been found through an idmapped mount the idmap of |
3178 | * the vfsmount must be passed through @idmap. This function will then take | |
3179 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 3180 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 3181 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 3182 | */ |
abf08576 | 3183 | int vfs_create(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 3184 | struct dentry *dentry, umode_t mode, bool want_excl) |
1da177e4 | 3185 | { |
abf08576 CB |
3186 | int error; |
3187 | ||
4609e1f1 | 3188 | error = may_create(idmap, dir, dentry); |
1da177e4 LT |
3189 | if (error) |
3190 | return error; | |
3191 | ||
acfa4380 | 3192 | if (!dir->i_op->create) |
1da177e4 | 3193 | return -EACCES; /* shouldn't it be ENOSYS? */ |
1639a49c | 3194 | |
9452e93e | 3195 | mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG); |
1da177e4 LT |
3196 | error = security_inode_create(dir, dentry, mode); |
3197 | if (error) | |
3198 | return error; | |
6c960e68 | 3199 | error = dir->i_op->create(idmap, dir, dentry, mode, want_excl); |
a74574aa | 3200 | if (!error) |
f38aa942 | 3201 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3202 | return error; |
3203 | } | |
4d359507 | 3204 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 3205 | |
8e6c848e AV |
3206 | int vfs_mkobj(struct dentry *dentry, umode_t mode, |
3207 | int (*f)(struct dentry *, umode_t, void *), | |
3208 | void *arg) | |
3209 | { | |
3210 | struct inode *dir = dentry->d_parent->d_inode; | |
4609e1f1 | 3211 | int error = may_create(&nop_mnt_idmap, dir, dentry); |
8e6c848e AV |
3212 | if (error) |
3213 | return error; | |
3214 | ||
3215 | mode &= S_IALLUGO; | |
3216 | mode |= S_IFREG; | |
3217 | error = security_inode_create(dir, dentry, mode); | |
3218 | if (error) | |
3219 | return error; | |
3220 | error = f(dentry, mode, arg); | |
3221 | if (!error) | |
3222 | fsnotify_create(dir, dentry); | |
3223 | return error; | |
3224 | } | |
3225 | EXPORT_SYMBOL(vfs_mkobj); | |
3226 | ||
a2982cc9 EB |
3227 | bool may_open_dev(const struct path *path) |
3228 | { | |
3229 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
3230 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
3231 | } | |
3232 | ||
4609e1f1 | 3233 | static int may_open(struct mnt_idmap *idmap, const struct path *path, |
ba73d987 | 3234 | int acc_mode, int flag) |
1da177e4 | 3235 | { |
3fb64190 | 3236 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
3237 | struct inode *inode = dentry->d_inode; |
3238 | int error; | |
3239 | ||
3240 | if (!inode) | |
3241 | return -ENOENT; | |
3242 | ||
c8fe8f30 CH |
3243 | switch (inode->i_mode & S_IFMT) { |
3244 | case S_IFLNK: | |
1da177e4 | 3245 | return -ELOOP; |
c8fe8f30 | 3246 | case S_IFDIR: |
fc4177be | 3247 | if (acc_mode & MAY_WRITE) |
c8fe8f30 | 3248 | return -EISDIR; |
fc4177be KC |
3249 | if (acc_mode & MAY_EXEC) |
3250 | return -EACCES; | |
c8fe8f30 CH |
3251 | break; |
3252 | case S_IFBLK: | |
3253 | case S_IFCHR: | |
a2982cc9 | 3254 | if (!may_open_dev(path)) |
1da177e4 | 3255 | return -EACCES; |
633fb6ac | 3256 | fallthrough; |
c8fe8f30 CH |
3257 | case S_IFIFO: |
3258 | case S_IFSOCK: | |
633fb6ac KC |
3259 | if (acc_mode & MAY_EXEC) |
3260 | return -EACCES; | |
1da177e4 | 3261 | flag &= ~O_TRUNC; |
c8fe8f30 | 3262 | break; |
0fd338b2 KC |
3263 | case S_IFREG: |
3264 | if ((acc_mode & MAY_EXEC) && path_noexec(path)) | |
3265 | return -EACCES; | |
3266 | break; | |
4a3fd211 | 3267 | } |
b41572e9 | 3268 | |
4609e1f1 | 3269 | error = inode_permission(idmap, inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
3270 | if (error) |
3271 | return error; | |
6146f0d5 | 3272 | |
1da177e4 LT |
3273 | /* |
3274 | * An append-only file must be opened in append mode for writing. | |
3275 | */ | |
3276 | if (IS_APPEND(inode)) { | |
8737c930 | 3277 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 3278 | return -EPERM; |
1da177e4 | 3279 | if (flag & O_TRUNC) |
7715b521 | 3280 | return -EPERM; |
1da177e4 LT |
3281 | } |
3282 | ||
3283 | /* O_NOATIME can only be set by the owner or superuser */ | |
01beba79 | 3284 | if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode)) |
7715b521 | 3285 | return -EPERM; |
1da177e4 | 3286 | |
f3c7691e | 3287 | return 0; |
7715b521 | 3288 | } |
1da177e4 | 3289 | |
abf08576 | 3290 | static int handle_truncate(struct mnt_idmap *idmap, struct file *filp) |
7715b521 | 3291 | { |
f0bb5aaf | 3292 | const struct path *path = &filp->f_path; |
7715b521 AV |
3293 | struct inode *inode = path->dentry->d_inode; |
3294 | int error = get_write_access(inode); | |
3295 | if (error) | |
3296 | return error; | |
482e0007 | 3297 | |
3350607d | 3298 | error = security_file_truncate(filp); |
7715b521 | 3299 | if (!error) { |
abf08576 | 3300 | error = do_truncate(idmap, path->dentry, 0, |
7715b521 | 3301 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, |
e1181ee6 | 3302 | filp); |
7715b521 AV |
3303 | } |
3304 | put_write_access(inode); | |
acd0c935 | 3305 | return error; |
1da177e4 LT |
3306 | } |
3307 | ||
d57999e1 DH |
3308 | static inline int open_to_namei_flags(int flag) |
3309 | { | |
8a5e929d AV |
3310 | if ((flag & O_ACCMODE) == 3) |
3311 | flag--; | |
d57999e1 DH |
3312 | return flag; |
3313 | } | |
3314 | ||
4609e1f1 | 3315 | static int may_o_create(struct mnt_idmap *idmap, |
ba73d987 CB |
3316 | const struct path *dir, struct dentry *dentry, |
3317 | umode_t mode) | |
d18e9008 MS |
3318 | { |
3319 | int error = security_path_mknod(dir, dentry, mode, 0); | |
3320 | if (error) | |
3321 | return error; | |
3322 | ||
4609e1f1 | 3323 | if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap)) |
1328c727 SF |
3324 | return -EOVERFLOW; |
3325 | ||
4609e1f1 | 3326 | error = inode_permission(idmap, dir->dentry->d_inode, |
47291baa | 3327 | MAY_WRITE | MAY_EXEC); |
d18e9008 MS |
3328 | if (error) |
3329 | return error; | |
3330 | ||
3331 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
3332 | } | |
3333 | ||
1acf0af9 DH |
3334 | /* |
3335 | * Attempt to atomically look up, create and open a file from a negative | |
3336 | * dentry. | |
3337 | * | |
3338 | * Returns 0 if successful. The file will have been created and attached to | |
3339 | * @file by the filesystem calling finish_open(). | |
3340 | * | |
00a07c15 AV |
3341 | * If the file was looked up only or didn't need creating, FMODE_OPENED won't |
3342 | * be set. The caller will need to perform the open themselves. @path will | |
3343 | * have been updated to point to the new dentry. This may be negative. | |
1acf0af9 DH |
3344 | * |
3345 | * Returns an error code otherwise. | |
3346 | */ | |
239eb983 AV |
3347 | static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, |
3348 | struct file *file, | |
239eb983 | 3349 | int open_flag, umode_t mode) |
d18e9008 | 3350 | { |
384f26e2 | 3351 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 3352 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 3353 | int error; |
d18e9008 | 3354 | |
d18e9008 MS |
3355 | if (nd->flags & LOOKUP_DIRECTORY) |
3356 | open_flag |= O_DIRECTORY; | |
3357 | ||
30d90494 AV |
3358 | file->f_path.dentry = DENTRY_NOT_SET; |
3359 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 | 3360 | error = dir->i_op->atomic_open(dir, dentry, file, |
44907d79 | 3361 | open_to_namei_flags(open_flag), mode); |
6fbd0714 | 3362 | d_lookup_done(dentry); |
384f26e2 | 3363 | if (!error) { |
64e1ac4d | 3364 | if (file->f_mode & FMODE_OPENED) { |
6fb968cd AV |
3365 | if (unlikely(dentry != file->f_path.dentry)) { |
3366 | dput(dentry); | |
3367 | dentry = dget(file->f_path.dentry); | |
3368 | } | |
64e1ac4d | 3369 | } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 3370 | error = -EIO; |
03da633a | 3371 | } else { |
384f26e2 AV |
3372 | if (file->f_path.dentry) { |
3373 | dput(dentry); | |
3374 | dentry = file->f_path.dentry; | |
03da633a | 3375 | } |
239eb983 | 3376 | if (unlikely(d_is_negative(dentry))) |
a01e718f | 3377 | error = -ENOENT; |
62b2ce96 | 3378 | } |
d18e9008 | 3379 | } |
239eb983 AV |
3380 | if (error) { |
3381 | dput(dentry); | |
3382 | dentry = ERR_PTR(error); | |
3383 | } | |
3384 | return dentry; | |
d18e9008 MS |
3385 | } |
3386 | ||
d58ffd35 | 3387 | /* |
1acf0af9 | 3388 | * Look up and maybe create and open the last component. |
d58ffd35 | 3389 | * |
00a07c15 | 3390 | * Must be called with parent locked (exclusive in O_CREAT case). |
1acf0af9 | 3391 | * |
00a07c15 AV |
3392 | * Returns 0 on success, that is, if |
3393 | * the file was successfully atomically created (if necessary) and opened, or | |
3394 | * the file was not completely opened at this time, though lookups and | |
3395 | * creations were performed. | |
3396 | * These case are distinguished by presence of FMODE_OPENED on file->f_mode. | |
3397 | * In the latter case dentry returned in @path might be negative if O_CREAT | |
3398 | * hadn't been specified. | |
1acf0af9 | 3399 | * |
00a07c15 | 3400 | * An error code is returned on failure. |
d58ffd35 | 3401 | */ |
da5ebf5a AV |
3402 | static struct dentry *lookup_open(struct nameidata *nd, struct file *file, |
3403 | const struct open_flags *op, | |
3404 | bool got_write) | |
d58ffd35 | 3405 | { |
6c960e68 | 3406 | struct mnt_idmap *idmap; |
d58ffd35 | 3407 | struct dentry *dir = nd->path.dentry; |
54ef4872 | 3408 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3409 | int open_flag = op->open_flag; |
d58ffd35 | 3410 | struct dentry *dentry; |
1643b43f | 3411 | int error, create_error = 0; |
1643b43f | 3412 | umode_t mode = op->mode; |
6fbd0714 | 3413 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3414 | |
ce8644fc | 3415 | if (unlikely(IS_DEADDIR(dir_inode))) |
da5ebf5a | 3416 | return ERR_PTR(-ENOENT); |
d58ffd35 | 3417 | |
73a09dd9 | 3418 | file->f_mode &= ~FMODE_CREATED; |
6fbd0714 AV |
3419 | dentry = d_lookup(dir, &nd->last); |
3420 | for (;;) { | |
3421 | if (!dentry) { | |
3422 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3423 | if (IS_ERR(dentry)) | |
da5ebf5a | 3424 | return dentry; |
6fbd0714 AV |
3425 | } |
3426 | if (d_in_lookup(dentry)) | |
3427 | break; | |
d58ffd35 | 3428 | |
6fbd0714 AV |
3429 | error = d_revalidate(dentry, nd->flags); |
3430 | if (likely(error > 0)) | |
3431 | break; | |
3432 | if (error) | |
3433 | goto out_dput; | |
3434 | d_invalidate(dentry); | |
3435 | dput(dentry); | |
3436 | dentry = NULL; | |
3437 | } | |
3438 | if (dentry->d_inode) { | |
6c51e513 | 3439 | /* Cached positive dentry: will open in f_op->open */ |
da5ebf5a | 3440 | return dentry; |
6c51e513 | 3441 | } |
d18e9008 | 3442 | |
1643b43f AV |
3443 | /* |
3444 | * Checking write permission is tricky, bacuse we don't know if we are | |
3445 | * going to actually need it: O_CREAT opens should work as long as the | |
3446 | * file exists. But checking existence breaks atomicity. The trick is | |
3447 | * to check access and if not granted clear O_CREAT from the flags. | |
3448 | * | |
3449 | * Another problem is returing the "right" error value (e.g. for an | |
3450 | * O_EXCL open we want to return EEXIST not EROFS). | |
3451 | */ | |
99a4a90c AV |
3452 | if (unlikely(!got_write)) |
3453 | open_flag &= ~O_TRUNC; | |
6c960e68 | 3454 | idmap = mnt_idmap(nd->path.mnt); |
1643b43f | 3455 | if (open_flag & O_CREAT) { |
99a4a90c AV |
3456 | if (open_flag & O_EXCL) |
3457 | open_flag &= ~O_TRUNC; | |
9452e93e | 3458 | mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode); |
99a4a90c | 3459 | if (likely(got_write)) |
4609e1f1 | 3460 | create_error = may_o_create(idmap, &nd->path, |
ba73d987 | 3461 | dentry, mode); |
99a4a90c AV |
3462 | else |
3463 | create_error = -EROFS; | |
d18e9008 | 3464 | } |
99a4a90c AV |
3465 | if (create_error) |
3466 | open_flag &= ~O_CREAT; | |
6ac08709 | 3467 | if (dir_inode->i_op->atomic_open) { |
d489cf9a | 3468 | dentry = atomic_open(nd, dentry, file, open_flag, mode); |
da5ebf5a AV |
3469 | if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) |
3470 | dentry = ERR_PTR(create_error); | |
3471 | return dentry; | |
d18e9008 | 3472 | } |
54ef4872 | 3473 | |
6fbd0714 | 3474 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3475 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3476 | nd->flags); | |
6fbd0714 | 3477 | d_lookup_done(dentry); |
12fa5e24 AV |
3478 | if (unlikely(res)) { |
3479 | if (IS_ERR(res)) { | |
3480 | error = PTR_ERR(res); | |
3481 | goto out_dput; | |
3482 | } | |
3483 | dput(dentry); | |
3484 | dentry = res; | |
3485 | } | |
54ef4872 MS |
3486 | } |
3487 | ||
d58ffd35 | 3488 | /* Negative dentry, just create the file */ |
1643b43f | 3489 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
73a09dd9 | 3490 | file->f_mode |= FMODE_CREATED; |
ce8644fc | 3491 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3492 | if (!dir_inode->i_op->create) { |
3493 | error = -EACCES; | |
d58ffd35 | 3494 | goto out_dput; |
ce8644fc | 3495 | } |
549c7297 | 3496 | |
6c960e68 | 3497 | error = dir_inode->i_op->create(idmap, dir_inode, dentry, |
549c7297 | 3498 | mode, open_flag & O_EXCL); |
d58ffd35 MS |
3499 | if (error) |
3500 | goto out_dput; | |
3501 | } | |
1643b43f AV |
3502 | if (unlikely(create_error) && !dentry->d_inode) { |
3503 | error = create_error; | |
3504 | goto out_dput; | |
d58ffd35 | 3505 | } |
da5ebf5a | 3506 | return dentry; |
d58ffd35 MS |
3507 | |
3508 | out_dput: | |
3509 | dput(dentry); | |
da5ebf5a | 3510 | return ERR_PTR(error); |
d58ffd35 MS |
3511 | } |
3512 | ||
c981a482 | 3513 | static const char *open_last_lookups(struct nameidata *nd, |
3ec2eef1 | 3514 | struct file *file, const struct open_flags *op) |
fb1cc555 | 3515 | { |
a1e28038 | 3516 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3517 | int open_flag = op->open_flag; |
64894cf8 | 3518 | bool got_write = false; |
da5ebf5a | 3519 | struct dentry *dentry; |
b0417d2c | 3520 | const char *res; |
1f36f774 | 3521 | |
c3e380b0 AV |
3522 | nd->flags |= op->intent; |
3523 | ||
bc77daa7 | 3524 | if (nd->last_type != LAST_NORM) { |
56676ec3 AV |
3525 | if (nd->depth) |
3526 | put_link(nd); | |
ff326a32 | 3527 | return handle_dots(nd, nd->last_type); |
1f36f774 | 3528 | } |
67ee3ad2 | 3529 | |
ca344a89 | 3530 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3531 | if (nd->last.name[nd->last.len]) |
3532 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3533 | /* we _can_ be in RCU mode here */ | |
4cb64024 | 3534 | dentry = lookup_fast(nd); |
20e34357 | 3535 | if (IS_ERR(dentry)) |
1ccac622 | 3536 | return ERR_CAST(dentry); |
20e34357 | 3537 | if (likely(dentry)) |
71574865 MS |
3538 | goto finish_lookup; |
3539 | ||
c04d905f BS |
3540 | if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU)) |
3541 | return ERR_PTR(-ECHILD); | |
b6183df7 MS |
3542 | } else { |
3543 | /* create side of things */ | |
72287417 | 3544 | if (nd->flags & LOOKUP_RCU) { |
e36cffed JA |
3545 | if (!try_to_unlazy(nd)) |
3546 | return ERR_PTR(-ECHILD); | |
72287417 | 3547 | } |
c9b07eab | 3548 | audit_inode(nd->name, dir, AUDIT_INODE_PARENT); |
b6183df7 | 3549 | /* trailing slashes? */ |
deb106c6 | 3550 | if (unlikely(nd->last.name[nd->last.len])) |
1ccac622 | 3551 | return ERR_PTR(-EISDIR); |
b6183df7 | 3552 | } |
a2c36b45 | 3553 | |
9cf843e3 | 3554 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
e36cffed | 3555 | got_write = !mnt_want_write(nd->path.mnt); |
64894cf8 AV |
3556 | /* |
3557 | * do _not_ fail yet - we might not need that or fail with | |
3558 | * a different error; let lookup_open() decide; we'll be | |
3559 | * dropping this one anyway. | |
3560 | */ | |
3561 | } | |
9cf843e3 AV |
3562 | if (open_flag & O_CREAT) |
3563 | inode_lock(dir->d_inode); | |
3564 | else | |
3565 | inode_lock_shared(dir->d_inode); | |
da5ebf5a | 3566 | dentry = lookup_open(nd, file, op, got_write); |
f7bb959d AV |
3567 | if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) |
3568 | fsnotify_create(dir->d_inode, dentry); | |
9cf843e3 AV |
3569 | if (open_flag & O_CREAT) |
3570 | inode_unlock(dir->d_inode); | |
3571 | else | |
3572 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3573 | |
c981a482 | 3574 | if (got_write) |
59e96e65 | 3575 | mnt_drop_write(nd->path.mnt); |
d18e9008 | 3576 | |
59e96e65 AV |
3577 | if (IS_ERR(dentry)) |
3578 | return ERR_CAST(dentry); | |
3579 | ||
973d4b73 | 3580 | if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { |
e73cabff AV |
3581 | dput(nd->path.dentry); |
3582 | nd->path.dentry = dentry; | |
c981a482 | 3583 | return NULL; |
fb1cc555 AV |
3584 | } |
3585 | ||
20e34357 | 3586 | finish_lookup: |
56676ec3 AV |
3587 | if (nd->depth) |
3588 | put_link(nd); | |
a4f5b521 | 3589 | res = step_into(nd, WALK_TRAILING, dentry); |
ff326a32 | 3590 | if (unlikely(res)) |
b0417d2c | 3591 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
ff326a32 | 3592 | return res; |
c981a482 AV |
3593 | } |
3594 | ||
3595 | /* | |
3596 | * Handle the last step of open() | |
3597 | */ | |
c5971b8c | 3598 | static int do_open(struct nameidata *nd, |
c981a482 AV |
3599 | struct file *file, const struct open_flags *op) |
3600 | { | |
abf08576 | 3601 | struct mnt_idmap *idmap; |
c981a482 AV |
3602 | int open_flag = op->open_flag; |
3603 | bool do_truncate; | |
3604 | int acc_mode; | |
c981a482 AV |
3605 | int error; |
3606 | ||
ff326a32 AV |
3607 | if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { |
3608 | error = complete_walk(nd); | |
3609 | if (error) | |
3610 | return error; | |
3611 | } | |
973d4b73 AV |
3612 | if (!(file->f_mode & FMODE_CREATED)) |
3613 | audit_inode(nd->name, nd->path.dentry, 0); | |
abf08576 | 3614 | idmap = mnt_idmap(nd->path.mnt); |
30aba665 | 3615 | if (open_flag & O_CREAT) { |
b94e0b32 AV |
3616 | if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) |
3617 | return -EEXIST; | |
30aba665 | 3618 | if (d_is_dir(nd->path.dentry)) |
c5971b8c | 3619 | return -EISDIR; |
e67fe633 | 3620 | error = may_create_in_sticky(idmap, nd, |
30aba665 SM |
3621 | d_backing_inode(nd->path.dentry)); |
3622 | if (unlikely(error)) | |
c5971b8c | 3623 | return error; |
30aba665 | 3624 | } |
44b1d530 | 3625 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
c5971b8c | 3626 | return -ENOTDIR; |
6c0d46c4 | 3627 | |
8795e7d4 AV |
3628 | do_truncate = false; |
3629 | acc_mode = op->acc_mode; | |
5a2d3edd AV |
3630 | if (file->f_mode & FMODE_CREATED) { |
3631 | /* Don't check for write permission, don't truncate */ | |
3632 | open_flag &= ~O_TRUNC; | |
5a2d3edd | 3633 | acc_mode = 0; |
8795e7d4 | 3634 | } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { |
0f9d1a10 AV |
3635 | error = mnt_want_write(nd->path.mnt); |
3636 | if (error) | |
c5971b8c | 3637 | return error; |
8795e7d4 | 3638 | do_truncate = true; |
0f9d1a10 | 3639 | } |
4609e1f1 | 3640 | error = may_open(idmap, &nd->path, acc_mode, open_flag); |
8795e7d4 | 3641 | if (!error && !(file->f_mode & FMODE_OPENED)) |
3ad5615a | 3642 | error = vfs_open(&nd->path, file); |
8795e7d4 | 3643 | if (!error) |
8f46ff57 | 3644 | error = security_file_post_open(file, op->acc_mode); |
8795e7d4 | 3645 | if (!error && do_truncate) |
abf08576 | 3646 | error = handle_truncate(idmap, file); |
c80567c8 AV |
3647 | if (unlikely(error > 0)) { |
3648 | WARN_ON(1); | |
3649 | error = -EINVAL; | |
3650 | } | |
8795e7d4 | 3651 | if (do_truncate) |
0f9d1a10 | 3652 | mnt_drop_write(nd->path.mnt); |
c5971b8c | 3653 | return error; |
fb1cc555 AV |
3654 | } |
3655 | ||
6521f891 CB |
3656 | /** |
3657 | * vfs_tmpfile - create tmpfile | |
abf08576 | 3658 | * @idmap: idmap of the mount the inode was found from |
73bb5a90 RS |
3659 | * @parentpath: pointer to the path of the base directory |
3660 | * @file: file descriptor of the new tmpfile | |
6521f891 | 3661 | * @mode: mode of the new tmpfile |
6521f891 CB |
3662 | * |
3663 | * Create a temporary file. | |
3664 | * | |
abf08576 CB |
3665 | * If the inode has been found through an idmapped mount the idmap of |
3666 | * the vfsmount must be passed through @idmap. This function will then take | |
3667 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 3668 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 3669 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 3670 | */ |
abf08576 | 3671 | static int vfs_tmpfile(struct mnt_idmap *idmap, |
9751b338 MS |
3672 | const struct path *parentpath, |
3673 | struct file *file, umode_t mode) | |
af7bd4dc | 3674 | { |
9751b338 MS |
3675 | struct dentry *child; |
3676 | struct inode *dir = d_inode(parentpath->dentry); | |
af7bd4dc AG |
3677 | struct inode *inode; |
3678 | int error; | |
406c706c | 3679 | int open_flag = file->f_flags; |
af7bd4dc AG |
3680 | |
3681 | /* we want directory to be writable */ | |
4609e1f1 | 3682 | error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); |
af7bd4dc | 3683 | if (error) |
9751b338 | 3684 | return error; |
af7bd4dc | 3685 | if (!dir->i_op->tmpfile) |
9751b338 MS |
3686 | return -EOPNOTSUPP; |
3687 | child = d_alloc(parentpath->dentry, &slash_name); | |
af7bd4dc | 3688 | if (unlikely(!child)) |
9751b338 MS |
3689 | return -ENOMEM; |
3690 | file->f_path.mnt = parentpath->mnt; | |
3691 | file->f_path.dentry = child; | |
9452e93e | 3692 | mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); |
011e2b71 | 3693 | error = dir->i_op->tmpfile(idmap, dir, file, mode); |
9751b338 | 3694 | dput(child); |
af7bd4dc | 3695 | if (error) |
9751b338 MS |
3696 | return error; |
3697 | /* Don't check for other permissions, the inode was just created */ | |
4609e1f1 | 3698 | error = may_open(idmap, &file->f_path, 0, file->f_flags); |
af7bd4dc | 3699 | if (error) |
9751b338 MS |
3700 | return error; |
3701 | inode = file_inode(file); | |
406c706c | 3702 | if (!(open_flag & O_EXCL)) { |
af7bd4dc AG |
3703 | spin_lock(&inode->i_lock); |
3704 | inode->i_state |= I_LINKABLE; | |
3705 | spin_unlock(&inode->i_lock); | |
3706 | } | |
a7811e34 | 3707 | security_inode_post_create_tmpfile(idmap, inode); |
9751b338 | 3708 | return 0; |
af7bd4dc | 3709 | } |
af7bd4dc | 3710 | |
22873dea | 3711 | /** |
d56e0ddb | 3712 | * kernel_tmpfile_open - open a tmpfile for kernel internal use |
abf08576 | 3713 | * @idmap: idmap of the mount the inode was found from |
22873dea MS |
3714 | * @parentpath: path of the base directory |
3715 | * @mode: mode of the new tmpfile | |
3716 | * @open_flag: flags | |
3717 | * @cred: credentials for open | |
3718 | * | |
3719 | * Create and open a temporary file. The file is not accounted in nr_files, | |
3720 | * hence this is only for kernel internal use, and must not be installed into | |
3721 | * file tables or such. | |
3722 | */ | |
d56e0ddb AG |
3723 | struct file *kernel_tmpfile_open(struct mnt_idmap *idmap, |
3724 | const struct path *parentpath, | |
3725 | umode_t mode, int open_flag, | |
3726 | const struct cred *cred) | |
22873dea MS |
3727 | { |
3728 | struct file *file; | |
3729 | int error; | |
22873dea | 3730 | |
9751b338 | 3731 | file = alloc_empty_file_noaccount(open_flag, cred); |
d56e0ddb AG |
3732 | if (IS_ERR(file)) |
3733 | return file; | |
3734 | ||
3735 | error = vfs_tmpfile(idmap, parentpath, file, mode); | |
3736 | if (error) { | |
3737 | fput(file); | |
3738 | file = ERR_PTR(error); | |
9751b338 | 3739 | } |
22873dea | 3740 | return file; |
af7bd4dc | 3741 | } |
d56e0ddb | 3742 | EXPORT_SYMBOL(kernel_tmpfile_open); |
af7bd4dc | 3743 | |
c8a53ee5 | 3744 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d | 3745 | const struct open_flags *op, |
3ec2eef1 | 3746 | struct file *file) |
60545d0d | 3747 | { |
625b6d10 | 3748 | struct path path; |
c8a53ee5 | 3749 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
9751b338 | 3750 | |
60545d0d AV |
3751 | if (unlikely(error)) |
3752 | return error; | |
625b6d10 | 3753 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3754 | if (unlikely(error)) |
3755 | goto out; | |
abf08576 | 3756 | error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode); |
9751b338 | 3757 | if (error) |
60545d0d | 3758 | goto out2; |
9751b338 | 3759 | audit_inode(nd->name, file->f_path.dentry, 0); |
60545d0d | 3760 | out2: |
625b6d10 | 3761 | mnt_drop_write(path.mnt); |
60545d0d | 3762 | out: |
625b6d10 | 3763 | path_put(&path); |
60545d0d AV |
3764 | return error; |
3765 | } | |
3766 | ||
6ac08709 AV |
3767 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3768 | { | |
3769 | struct path path; | |
3770 | int error = path_lookupat(nd, flags, &path); | |
3771 | if (!error) { | |
3772 | audit_inode(nd->name, path.dentry, 0); | |
ae2bb293 | 3773 | error = vfs_open(&path, file); |
6ac08709 AV |
3774 | path_put(&path); |
3775 | } | |
3776 | return error; | |
3777 | } | |
3778 | ||
c8a53ee5 AV |
3779 | static struct file *path_openat(struct nameidata *nd, |
3780 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3781 | { |
30d90494 | 3782 | struct file *file; |
13aab428 | 3783 | int error; |
31e6b01f | 3784 | |
ea73ea72 | 3785 | file = alloc_empty_file(op->open_flag, current_cred()); |
1afc99be AV |
3786 | if (IS_ERR(file)) |
3787 | return file; | |
31e6b01f | 3788 | |
bb458c64 | 3789 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
3ec2eef1 | 3790 | error = do_tmpfile(nd, flags, op, file); |
5f336e72 | 3791 | } else if (unlikely(file->f_flags & O_PATH)) { |
6ac08709 | 3792 | error = do_o_path(nd, flags, file); |
5f336e72 AV |
3793 | } else { |
3794 | const char *s = path_init(nd, flags); | |
3795 | while (!(error = link_path_walk(s, nd)) && | |
c5971b8c | 3796 | (s = open_last_lookups(nd, file, op)) != NULL) |
1ccac622 | 3797 | ; |
c5971b8c AV |
3798 | if (!error) |
3799 | error = do_open(nd, file, op); | |
5f336e72 | 3800 | terminate_walk(nd); |
806b681c | 3801 | } |
7c1c01ec | 3802 | if (likely(!error)) { |
aad888f8 | 3803 | if (likely(file->f_mode & FMODE_OPENED)) |
7c1c01ec AV |
3804 | return file; |
3805 | WARN_ON(1); | |
3806 | error = -EINVAL; | |
16b1c1cd | 3807 | } |
7cb537b6 | 3808 | fput(file); |
7c1c01ec AV |
3809 | if (error == -EOPENSTALE) { |
3810 | if (flags & LOOKUP_RCU) | |
3811 | error = -ECHILD; | |
3812 | else | |
3813 | error = -ESTALE; | |
2675a4eb | 3814 | } |
7c1c01ec | 3815 | return ERR_PTR(error); |
1da177e4 LT |
3816 | } |
3817 | ||
669abf4e | 3818 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3819 | const struct open_flags *op) |
13aab428 | 3820 | { |
9883d185 | 3821 | struct nameidata nd; |
f9652e10 | 3822 | int flags = op->lookup_flags; |
13aab428 AV |
3823 | struct file *filp; |
3824 | ||
06422964 | 3825 | set_nameidata(&nd, dfd, pathname, NULL); |
c8a53ee5 | 3826 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3827 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3828 | filp = path_openat(&nd, op, flags); |
13aab428 | 3829 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3830 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3831 | restore_nameidata(); |
13aab428 AV |
3832 | return filp; |
3833 | } | |
3834 | ||
ffb37ca3 | 3835 | struct file *do_file_open_root(const struct path *root, |
f9652e10 | 3836 | const char *name, const struct open_flags *op) |
73d049a4 | 3837 | { |
9883d185 | 3838 | struct nameidata nd; |
73d049a4 | 3839 | struct file *file; |
51689104 | 3840 | struct filename *filename; |
bcba1e7d | 3841 | int flags = op->lookup_flags; |
73d049a4 | 3842 | |
ffb37ca3 | 3843 | if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3844 | return ERR_PTR(-ELOOP); |
3845 | ||
51689104 | 3846 | filename = getname_kernel(name); |
a1c83681 | 3847 | if (IS_ERR(filename)) |
51689104 PM |
3848 | return ERR_CAST(filename); |
3849 | ||
06422964 | 3850 | set_nameidata(&nd, -1, filename, root); |
c8a53ee5 | 3851 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3852 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3853 | file = path_openat(&nd, op, flags); |
73d049a4 | 3854 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3855 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3856 | restore_nameidata(); |
51689104 | 3857 | putname(filename); |
73d049a4 AV |
3858 | return file; |
3859 | } | |
3860 | ||
b4a4f213 SB |
3861 | static struct dentry *filename_create(int dfd, struct filename *name, |
3862 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3863 | { |
c663e5d8 | 3864 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 | 3865 | struct qstr last; |
b3d4650d N |
3866 | bool want_dir = lookup_flags & LOOKUP_DIRECTORY; |
3867 | unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; | |
3868 | unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; | |
391172c4 | 3869 | int type; |
c30dabfe | 3870 | int err2; |
1ac12b4b | 3871 | int error; |
1ac12b4b | 3872 | |
b3d4650d | 3873 | error = filename_parentat(dfd, name, reval_flag, path, &last, &type); |
0ee50b47 DK |
3874 | if (error) |
3875 | return ERR_PTR(error); | |
1da177e4 | 3876 | |
c663e5d8 CH |
3877 | /* |
3878 | * Yucky last component or no last component at all? | |
3879 | * (foo/., foo/.., /////) | |
3880 | */ | |
5c31b6ce | 3881 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3882 | goto out; |
c663e5d8 | 3883 | |
c30dabfe | 3884 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3885 | err2 = mnt_want_write(path->mnt); |
c663e5d8 | 3886 | /* |
b3d4650d N |
3887 | * Do the final lookup. Suppress 'create' if there is a trailing |
3888 | * '/', and a directory wasn't requested. | |
c663e5d8 | 3889 | */ |
b3d4650d N |
3890 | if (last.name[last.len] && !want_dir) |
3891 | create_flags = 0; | |
5955102c | 3892 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
74d7970f NJ |
3893 | dentry = lookup_one_qstr_excl(&last, path->dentry, |
3894 | reval_flag | create_flags); | |
1da177e4 | 3895 | if (IS_ERR(dentry)) |
a8104a9f | 3896 | goto unlock; |
c663e5d8 | 3897 | |
a8104a9f | 3898 | error = -EEXIST; |
b18825a7 | 3899 | if (d_is_positive(dentry)) |
a8104a9f | 3900 | goto fail; |
b18825a7 | 3901 | |
c663e5d8 CH |
3902 | /* |
3903 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3904 | * From the vfs_mknod() POV we just have a negative dentry - | |
3905 | * all is fine. Let's be bastards - you had / on the end, you've | |
3906 | * been asking for (non-existent) directory. -ENOENT for you. | |
3907 | */ | |
b3d4650d | 3908 | if (unlikely(!create_flags)) { |
a8104a9f | 3909 | error = -ENOENT; |
ed75e95d | 3910 | goto fail; |
e9baf6e5 | 3911 | } |
c30dabfe JK |
3912 | if (unlikely(err2)) { |
3913 | error = err2; | |
a8104a9f | 3914 | goto fail; |
c30dabfe | 3915 | } |
1da177e4 | 3916 | return dentry; |
1da177e4 | 3917 | fail: |
a8104a9f AV |
3918 | dput(dentry); |
3919 | dentry = ERR_PTR(error); | |
3920 | unlock: | |
5955102c | 3921 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3922 | if (!err2) |
391172c4 | 3923 | mnt_drop_write(path->mnt); |
ed75e95d | 3924 | out: |
391172c4 | 3925 | path_put(path); |
1da177e4 LT |
3926 | return dentry; |
3927 | } | |
fa14a0b8 | 3928 | |
b4a4f213 | 3929 | struct dentry *kern_path_create(int dfd, const char *pathname, |
584d3226 DK |
3930 | struct path *path, unsigned int lookup_flags) |
3931 | { | |
b4a4f213 SB |
3932 | struct filename *filename = getname_kernel(pathname); |
3933 | struct dentry *res = filename_create(dfd, filename, path, lookup_flags); | |
584d3226 | 3934 | |
b4a4f213 | 3935 | putname(filename); |
584d3226 DK |
3936 | return res; |
3937 | } | |
dae6ad8f AV |
3938 | EXPORT_SYMBOL(kern_path_create); |
3939 | ||
921a1650 AV |
3940 | void done_path_create(struct path *path, struct dentry *dentry) |
3941 | { | |
3942 | dput(dentry); | |
5955102c | 3943 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3944 | mnt_drop_write(path->mnt); |
921a1650 AV |
3945 | path_put(path); |
3946 | } | |
3947 | EXPORT_SYMBOL(done_path_create); | |
3948 | ||
520ae687 | 3949 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3950 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3951 | { |
b4a4f213 SB |
3952 | struct filename *filename = getname(pathname); |
3953 | struct dentry *res = filename_create(dfd, filename, path, lookup_flags); | |
3954 | ||
3955 | putname(filename); | |
3956 | return res; | |
dae6ad8f AV |
3957 | } |
3958 | EXPORT_SYMBOL(user_path_create); | |
3959 | ||
6521f891 CB |
3960 | /** |
3961 | * vfs_mknod - create device node or file | |
abf08576 | 3962 | * @idmap: idmap of the mount the inode was found from |
6521f891 CB |
3963 | * @dir: inode of @dentry |
3964 | * @dentry: pointer to dentry of the base directory | |
3965 | * @mode: mode of the new device node or file | |
3966 | * @dev: device number of device to create | |
3967 | * | |
3968 | * Create a device node or file. | |
3969 | * | |
abf08576 CB |
3970 | * If the inode has been found through an idmapped mount the idmap of |
3971 | * the vfsmount must be passed through @idmap. This function will then take | |
3972 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 3973 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 3974 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 3975 | */ |
abf08576 | 3976 | int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 3977 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3978 | { |
a3c751a5 | 3979 | bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; |
4609e1f1 | 3980 | int error = may_create(idmap, dir, dentry); |
1da177e4 LT |
3981 | |
3982 | if (error) | |
3983 | return error; | |
3984 | ||
a3c751a5 MS |
3985 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && |
3986 | !capable(CAP_MKNOD)) | |
1da177e4 LT |
3987 | return -EPERM; |
3988 | ||
acfa4380 | 3989 | if (!dir->i_op->mknod) |
1da177e4 LT |
3990 | return -EPERM; |
3991 | ||
9452e93e | 3992 | mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); |
08ce5f16 SH |
3993 | error = devcgroup_inode_mknod(mode, dev); |
3994 | if (error) | |
3995 | return error; | |
3996 | ||
1da177e4 LT |
3997 | error = security_inode_mknod(dir, dentry, mode, dev); |
3998 | if (error) | |
3999 | return error; | |
4000 | ||
5ebb29be | 4001 | error = dir->i_op->mknod(idmap, dir, dentry, mode, dev); |
a74574aa | 4002 | if (!error) |
f38aa942 | 4003 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4004 | return error; |
4005 | } | |
4d359507 | 4006 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 4007 | |
f69aac00 | 4008 | static int may_mknod(umode_t mode) |
463c3197 DH |
4009 | { |
4010 | switch (mode & S_IFMT) { | |
4011 | case S_IFREG: | |
4012 | case S_IFCHR: | |
4013 | case S_IFBLK: | |
4014 | case S_IFIFO: | |
4015 | case S_IFSOCK: | |
4016 | case 0: /* zero mode translates to S_IFREG */ | |
4017 | return 0; | |
4018 | case S_IFDIR: | |
4019 | return -EPERM; | |
4020 | default: | |
4021 | return -EINVAL; | |
4022 | } | |
4023 | } | |
4024 | ||
45f30dab | 4025 | static int do_mknodat(int dfd, struct filename *name, umode_t mode, |
87c4e192 | 4026 | unsigned int dev) |
1da177e4 | 4027 | { |
abf08576 | 4028 | struct mnt_idmap *idmap; |
2ad94ae6 | 4029 | struct dentry *dentry; |
dae6ad8f AV |
4030 | struct path path; |
4031 | int error; | |
972567f1 | 4032 | unsigned int lookup_flags = 0; |
1da177e4 | 4033 | |
8e4bfca1 AV |
4034 | error = may_mknod(mode); |
4035 | if (error) | |
7797251b | 4036 | goto out1; |
972567f1 | 4037 | retry: |
b4a4f213 | 4038 | dentry = filename_create(dfd, name, &path, lookup_flags); |
7797251b | 4039 | error = PTR_ERR(dentry); |
dae6ad8f | 4040 | if (IS_ERR(dentry)) |
7797251b | 4041 | goto out1; |
2ad94ae6 | 4042 | |
1639a49c YX |
4043 | error = security_path_mknod(&path, dentry, |
4044 | mode_strip_umask(path.dentry->d_inode, mode), dev); | |
be6d3e56 | 4045 | if (error) |
7797251b | 4046 | goto out2; |
6521f891 | 4047 | |
abf08576 | 4048 | idmap = mnt_idmap(path.mnt); |
463c3197 | 4049 | switch (mode & S_IFMT) { |
1da177e4 | 4050 | case 0: case S_IFREG: |
abf08576 | 4051 | error = vfs_create(idmap, path.dentry->d_inode, |
6521f891 | 4052 | dentry, mode, true); |
701b3899 RS |
4053 | if (!error) |
4054 | security_path_post_mknod(idmap, dentry); | |
1da177e4 LT |
4055 | break; |
4056 | case S_IFCHR: case S_IFBLK: | |
abf08576 | 4057 | error = vfs_mknod(idmap, path.dentry->d_inode, |
6521f891 | 4058 | dentry, mode, new_decode_dev(dev)); |
1da177e4 LT |
4059 | break; |
4060 | case S_IFIFO: case S_IFSOCK: | |
abf08576 | 4061 | error = vfs_mknod(idmap, path.dentry->d_inode, |
6521f891 | 4062 | dentry, mode, 0); |
1da177e4 | 4063 | break; |
1da177e4 | 4064 | } |
7797251b | 4065 | out2: |
921a1650 | 4066 | done_path_create(&path, dentry); |
972567f1 JL |
4067 | if (retry_estale(error, lookup_flags)) { |
4068 | lookup_flags |= LOOKUP_REVAL; | |
4069 | goto retry; | |
4070 | } | |
7797251b DK |
4071 | out1: |
4072 | putname(name); | |
1da177e4 LT |
4073 | return error; |
4074 | } | |
4075 | ||
87c4e192 DB |
4076 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
4077 | unsigned int, dev) | |
4078 | { | |
7797251b | 4079 | return do_mknodat(dfd, getname(filename), mode, dev); |
87c4e192 DB |
4080 | } |
4081 | ||
8208a22b | 4082 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d | 4083 | { |
7797251b | 4084 | return do_mknodat(AT_FDCWD, getname(filename), mode, dev); |
5590ff0d UD |
4085 | } |
4086 | ||
6521f891 CB |
4087 | /** |
4088 | * vfs_mkdir - create directory | |
abf08576 | 4089 | * @idmap: idmap of the mount the inode was found from |
6521f891 CB |
4090 | * @dir: inode of @dentry |
4091 | * @dentry: pointer to dentry of the base directory | |
4092 | * @mode: mode of the new directory | |
4093 | * | |
4094 | * Create a directory. | |
4095 | * | |
abf08576 CB |
4096 | * If the inode has been found through an idmapped mount the idmap of |
4097 | * the vfsmount must be passed through @idmap. This function will then take | |
4098 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 4099 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 4100 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 4101 | */ |
abf08576 | 4102 | int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 4103 | struct dentry *dentry, umode_t mode) |
1da177e4 | 4104 | { |
abf08576 | 4105 | int error; |
8de52778 | 4106 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 | 4107 | |
4609e1f1 | 4108 | error = may_create(idmap, dir, dentry); |
1da177e4 LT |
4109 | if (error) |
4110 | return error; | |
4111 | ||
acfa4380 | 4112 | if (!dir->i_op->mkdir) |
1da177e4 LT |
4113 | return -EPERM; |
4114 | ||
9452e93e | 4115 | mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0); |
1da177e4 LT |
4116 | error = security_inode_mkdir(dir, dentry, mode); |
4117 | if (error) | |
4118 | return error; | |
4119 | ||
8de52778 AV |
4120 | if (max_links && dir->i_nlink >= max_links) |
4121 | return -EMLINK; | |
4122 | ||
c54bd91e | 4123 | error = dir->i_op->mkdir(idmap, dir, dentry, mode); |
a74574aa | 4124 | if (!error) |
f38aa942 | 4125 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
4126 | return error; |
4127 | } | |
4d359507 | 4128 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 4129 | |
45f30dab | 4130 | int do_mkdirat(int dfd, struct filename *name, umode_t mode) |
1da177e4 | 4131 | { |
6902d925 | 4132 | struct dentry *dentry; |
dae6ad8f AV |
4133 | struct path path; |
4134 | int error; | |
b76d8b82 | 4135 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 4136 | |
b76d8b82 | 4137 | retry: |
b4a4f213 | 4138 | dentry = filename_create(dfd, name, &path, lookup_flags); |
584d3226 | 4139 | error = PTR_ERR(dentry); |
6902d925 | 4140 | if (IS_ERR(dentry)) |
584d3226 | 4141 | goto out_putname; |
1da177e4 | 4142 | |
1639a49c YX |
4143 | error = security_path_mkdir(&path, dentry, |
4144 | mode_strip_umask(path.dentry->d_inode, mode)); | |
6521f891 | 4145 | if (!error) { |
abf08576 CB |
4146 | error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode, |
4147 | dentry, mode); | |
6521f891 | 4148 | } |
921a1650 | 4149 | done_path_create(&path, dentry); |
b76d8b82 JL |
4150 | if (retry_estale(error, lookup_flags)) { |
4151 | lookup_flags |= LOOKUP_REVAL; | |
4152 | goto retry; | |
4153 | } | |
584d3226 DK |
4154 | out_putname: |
4155 | putname(name); | |
1da177e4 LT |
4156 | return error; |
4157 | } | |
4158 | ||
0101db7a DB |
4159 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
4160 | { | |
584d3226 | 4161 | return do_mkdirat(dfd, getname(pathname), mode); |
0101db7a DB |
4162 | } |
4163 | ||
a218d0fd | 4164 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d | 4165 | { |
584d3226 | 4166 | return do_mkdirat(AT_FDCWD, getname(pathname), mode); |
5590ff0d UD |
4167 | } |
4168 | ||
6521f891 CB |
4169 | /** |
4170 | * vfs_rmdir - remove directory | |
abf08576 | 4171 | * @idmap: idmap of the mount the inode was found from |
6521f891 CB |
4172 | * @dir: inode of @dentry |
4173 | * @dentry: pointer to dentry of the base directory | |
4174 | * | |
4175 | * Remove a directory. | |
4176 | * | |
abf08576 CB |
4177 | * If the inode has been found through an idmapped mount the idmap of |
4178 | * the vfsmount must be passed through @idmap. This function will then take | |
4179 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 4180 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 4181 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 4182 | */ |
abf08576 | 4183 | int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 4184 | struct dentry *dentry) |
1da177e4 | 4185 | { |
4609e1f1 | 4186 | int error = may_delete(idmap, dir, dentry, 1); |
1da177e4 LT |
4187 | |
4188 | if (error) | |
4189 | return error; | |
4190 | ||
acfa4380 | 4191 | if (!dir->i_op->rmdir) |
1da177e4 LT |
4192 | return -EPERM; |
4193 | ||
1d2ef590 | 4194 | dget(dentry); |
5955102c | 4195 | inode_lock(dentry->d_inode); |
912dbc15 SW |
4196 | |
4197 | error = -EBUSY; | |
1bd9c4e4 DH |
4198 | if (is_local_mountpoint(dentry) || |
4199 | (dentry->d_inode->i_flags & S_KERNEL_FILE)) | |
912dbc15 SW |
4200 | goto out; |
4201 | ||
4202 | error = security_inode_rmdir(dir, dentry); | |
4203 | if (error) | |
4204 | goto out; | |
4205 | ||
4206 | error = dir->i_op->rmdir(dir, dentry); | |
4207 | if (error) | |
4208 | goto out; | |
4209 | ||
8767712f | 4210 | shrink_dcache_parent(dentry); |
912dbc15 SW |
4211 | dentry->d_inode->i_flags |= S_DEAD; |
4212 | dont_mount(dentry); | |
8ed936b5 | 4213 | detach_mounts(dentry); |
912dbc15 SW |
4214 | |
4215 | out: | |
5955102c | 4216 | inode_unlock(dentry->d_inode); |
1d2ef590 | 4217 | dput(dentry); |
912dbc15 | 4218 | if (!error) |
a37d9a17 | 4219 | d_delete_notify(dir, dentry); |
1da177e4 LT |
4220 | return error; |
4221 | } | |
4d359507 | 4222 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 4223 | |
45f30dab | 4224 | int do_rmdir(int dfd, struct filename *name) |
1da177e4 | 4225 | { |
0ee50b47 | 4226 | int error; |
1da177e4 | 4227 | struct dentry *dentry; |
f5beed75 AV |
4228 | struct path path; |
4229 | struct qstr last; | |
4230 | int type; | |
c6ee9206 JL |
4231 | unsigned int lookup_flags = 0; |
4232 | retry: | |
c5f563f9 | 4233 | error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); |
0ee50b47 DK |
4234 | if (error) |
4235 | goto exit1; | |
1da177e4 | 4236 | |
f5beed75 | 4237 | switch (type) { |
0612d9fb OH |
4238 | case LAST_DOTDOT: |
4239 | error = -ENOTEMPTY; | |
0ee50b47 | 4240 | goto exit2; |
0612d9fb OH |
4241 | case LAST_DOT: |
4242 | error = -EINVAL; | |
0ee50b47 | 4243 | goto exit2; |
0612d9fb OH |
4244 | case LAST_ROOT: |
4245 | error = -EBUSY; | |
0ee50b47 | 4246 | goto exit2; |
1da177e4 | 4247 | } |
0612d9fb | 4248 | |
f5beed75 | 4249 | error = mnt_want_write(path.mnt); |
c30dabfe | 4250 | if (error) |
0ee50b47 | 4251 | goto exit2; |
0612d9fb | 4252 | |
5955102c | 4253 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
74d7970f | 4254 | dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); |
1da177e4 | 4255 | error = PTR_ERR(dentry); |
6902d925 | 4256 | if (IS_ERR(dentry)) |
0ee50b47 | 4257 | goto exit3; |
e6bc45d6 TT |
4258 | if (!dentry->d_inode) { |
4259 | error = -ENOENT; | |
0ee50b47 | 4260 | goto exit4; |
e6bc45d6 | 4261 | } |
f5beed75 | 4262 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 4263 | if (error) |
0ee50b47 | 4264 | goto exit4; |
abf08576 | 4265 | error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry); |
0ee50b47 | 4266 | exit4: |
6902d925 | 4267 | dput(dentry); |
0ee50b47 | 4268 | exit3: |
5955102c | 4269 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 4270 | mnt_drop_write(path.mnt); |
0ee50b47 | 4271 | exit2: |
f5beed75 | 4272 | path_put(&path); |
c6ee9206 JL |
4273 | if (retry_estale(error, lookup_flags)) { |
4274 | lookup_flags |= LOOKUP_REVAL; | |
4275 | goto retry; | |
4276 | } | |
0ee50b47 | 4277 | exit1: |
24fb33d4 | 4278 | putname(name); |
1da177e4 LT |
4279 | return error; |
4280 | } | |
4281 | ||
3cdad428 | 4282 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d | 4283 | { |
e24ab0ef | 4284 | return do_rmdir(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
4285 | } |
4286 | ||
b21996e3 BF |
4287 | /** |
4288 | * vfs_unlink - unlink a filesystem object | |
abf08576 | 4289 | * @idmap: idmap of the mount the inode was found from |
b21996e3 BF |
4290 | * @dir: parent directory |
4291 | * @dentry: victim | |
4292 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
4293 | * | |
4294 | * The caller must hold dir->i_mutex. | |
4295 | * | |
4296 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
4297 | * return a reference to the inode in delegated_inode. The caller | |
4298 | * should then break the delegation on that inode and retry. Because | |
4299 | * breaking a delegation may take a long time, the caller should drop | |
4300 | * dir->i_mutex before doing so. | |
4301 | * | |
4302 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4303 | * be appropriate for callers that expect the underlying filesystem not | |
4304 | * to be NFS exported. | |
6521f891 | 4305 | * |
abf08576 CB |
4306 | * If the inode has been found through an idmapped mount the idmap of |
4307 | * the vfsmount must be passed through @idmap. This function will then take | |
4308 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 4309 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 4310 | * raw inode simply pass @nop_mnt_idmap. |
b21996e3 | 4311 | */ |
abf08576 | 4312 | int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 4313 | struct dentry *dentry, struct inode **delegated_inode) |
1da177e4 | 4314 | { |
9accbb97 | 4315 | struct inode *target = dentry->d_inode; |
4609e1f1 | 4316 | int error = may_delete(idmap, dir, dentry, 0); |
1da177e4 LT |
4317 | |
4318 | if (error) | |
4319 | return error; | |
4320 | ||
acfa4380 | 4321 | if (!dir->i_op->unlink) |
1da177e4 LT |
4322 | return -EPERM; |
4323 | ||
5955102c | 4324 | inode_lock(target); |
51cc3a66 HD |
4325 | if (IS_SWAPFILE(target)) |
4326 | error = -EPERM; | |
4327 | else if (is_local_mountpoint(dentry)) | |
1da177e4 LT |
4328 | error = -EBUSY; |
4329 | else { | |
4330 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 4331 | if (!error) { |
5a14696c BF |
4332 | error = try_break_deleg(target, delegated_inode); |
4333 | if (error) | |
b21996e3 | 4334 | goto out; |
1da177e4 | 4335 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 4336 | if (!error) { |
d83c49f3 | 4337 | dont_mount(dentry); |
8ed936b5 EB |
4338 | detach_mounts(dentry); |
4339 | } | |
bec1052e | 4340 | } |
1da177e4 | 4341 | } |
b21996e3 | 4342 | out: |
5955102c | 4343 | inode_unlock(target); |
1da177e4 LT |
4344 | |
4345 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
a37d9a17 AG |
4346 | if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
4347 | fsnotify_unlink(dir, dentry); | |
4348 | } else if (!error) { | |
9accbb97 | 4349 | fsnotify_link_count(target); |
a37d9a17 | 4350 | d_delete_notify(dir, dentry); |
1da177e4 | 4351 | } |
0eeca283 | 4352 | |
1da177e4 LT |
4353 | return error; |
4354 | } | |
4d359507 | 4355 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
4356 | |
4357 | /* | |
4358 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 4359 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
4360 | * writeout happening, and we don't want to prevent access to the directory |
4361 | * while waiting on the I/O. | |
4362 | */ | |
45f30dab | 4363 | int do_unlinkat(int dfd, struct filename *name) |
1da177e4 | 4364 | { |
2ad94ae6 | 4365 | int error; |
1da177e4 | 4366 | struct dentry *dentry; |
f5beed75 AV |
4367 | struct path path; |
4368 | struct qstr last; | |
4369 | int type; | |
1da177e4 | 4370 | struct inode *inode = NULL; |
b21996e3 | 4371 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
4372 | unsigned int lookup_flags = 0; |
4373 | retry: | |
c5f563f9 | 4374 | error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); |
0ee50b47 DK |
4375 | if (error) |
4376 | goto exit1; | |
2ad94ae6 | 4377 | |
1da177e4 | 4378 | error = -EISDIR; |
f5beed75 | 4379 | if (type != LAST_NORM) |
0ee50b47 | 4380 | goto exit2; |
0612d9fb | 4381 | |
f5beed75 | 4382 | error = mnt_want_write(path.mnt); |
c30dabfe | 4383 | if (error) |
0ee50b47 | 4384 | goto exit2; |
b21996e3 | 4385 | retry_deleg: |
5955102c | 4386 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
74d7970f | 4387 | dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); |
1da177e4 LT |
4388 | error = PTR_ERR(dentry); |
4389 | if (!IS_ERR(dentry)) { | |
6521f891 | 4390 | |
1da177e4 | 4391 | /* Why not before? Because we want correct error value */ |
6036c5f1 | 4392 | if (last.name[last.len] || d_is_negative(dentry)) |
50338b88 | 4393 | goto slashes; |
1da177e4 | 4394 | inode = dentry->d_inode; |
e6bc45d6 | 4395 | ihold(inode); |
f5beed75 | 4396 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 4397 | if (error) |
0ee50b47 | 4398 | goto exit3; |
abf08576 CB |
4399 | error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode, |
4400 | dentry, &delegated_inode); | |
0ee50b47 | 4401 | exit3: |
1da177e4 LT |
4402 | dput(dentry); |
4403 | } | |
5955102c | 4404 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
4405 | if (inode) |
4406 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
4407 | inode = NULL; |
4408 | if (delegated_inode) { | |
5a14696c | 4409 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
4410 | if (!error) |
4411 | goto retry_deleg; | |
4412 | } | |
f5beed75 | 4413 | mnt_drop_write(path.mnt); |
0ee50b47 | 4414 | exit2: |
f5beed75 | 4415 | path_put(&path); |
5d18f813 JL |
4416 | if (retry_estale(error, lookup_flags)) { |
4417 | lookup_flags |= LOOKUP_REVAL; | |
4418 | inode = NULL; | |
4419 | goto retry; | |
4420 | } | |
0ee50b47 | 4421 | exit1: |
da2f1362 | 4422 | putname(name); |
1da177e4 LT |
4423 | return error; |
4424 | ||
4425 | slashes: | |
b18825a7 DH |
4426 | if (d_is_negative(dentry)) |
4427 | error = -ENOENT; | |
44b1d530 | 4428 | else if (d_is_dir(dentry)) |
b18825a7 DH |
4429 | error = -EISDIR; |
4430 | else | |
4431 | error = -ENOTDIR; | |
0ee50b47 | 4432 | goto exit3; |
1da177e4 LT |
4433 | } |
4434 | ||
2e4d0924 | 4435 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4436 | { |
4437 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4438 | return -EINVAL; | |
4439 | ||
4440 | if (flag & AT_REMOVEDIR) | |
e24ab0ef | 4441 | return do_rmdir(dfd, getname(pathname)); |
da2f1362 | 4442 | return do_unlinkat(dfd, getname(pathname)); |
5590ff0d UD |
4443 | } |
4444 | ||
3480b257 | 4445 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d | 4446 | { |
da2f1362 | 4447 | return do_unlinkat(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
4448 | } |
4449 | ||
6521f891 CB |
4450 | /** |
4451 | * vfs_symlink - create symlink | |
abf08576 | 4452 | * @idmap: idmap of the mount the inode was found from |
6521f891 CB |
4453 | * @dir: inode of @dentry |
4454 | * @dentry: pointer to dentry of the base directory | |
4455 | * @oldname: name of the file to link to | |
4456 | * | |
4457 | * Create a symlink. | |
4458 | * | |
abf08576 CB |
4459 | * If the inode has been found through an idmapped mount the idmap of |
4460 | * the vfsmount must be passed through @idmap. This function will then take | |
4461 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 4462 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 4463 | * raw inode simply pass @nop_mnt_idmap. |
6521f891 | 4464 | */ |
abf08576 | 4465 | int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir, |
6521f891 | 4466 | struct dentry *dentry, const char *oldname) |
1da177e4 | 4467 | { |
7a77db95 | 4468 | int error; |
1da177e4 | 4469 | |
4609e1f1 | 4470 | error = may_create(idmap, dir, dentry); |
1da177e4 LT |
4471 | if (error) |
4472 | return error; | |
4473 | ||
acfa4380 | 4474 | if (!dir->i_op->symlink) |
1da177e4 LT |
4475 | return -EPERM; |
4476 | ||
4477 | error = security_inode_symlink(dir, dentry, oldname); | |
4478 | if (error) | |
4479 | return error; | |
4480 | ||
7a77db95 | 4481 | error = dir->i_op->symlink(idmap, dir, dentry, oldname); |
a74574aa | 4482 | if (!error) |
f38aa942 | 4483 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4484 | return error; |
4485 | } | |
4d359507 | 4486 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4487 | |
7a8721f8 | 4488 | int do_symlinkat(struct filename *from, int newdfd, struct filename *to) |
1da177e4 | 4489 | { |
2ad94ae6 | 4490 | int error; |
6902d925 | 4491 | struct dentry *dentry; |
dae6ad8f | 4492 | struct path path; |
f46d3567 | 4493 | unsigned int lookup_flags = 0; |
1da177e4 | 4494 | |
da2d0ced DK |
4495 | if (IS_ERR(from)) { |
4496 | error = PTR_ERR(from); | |
4497 | goto out_putnames; | |
4498 | } | |
f46d3567 | 4499 | retry: |
b4a4f213 | 4500 | dentry = filename_create(newdfd, to, &path, lookup_flags); |
6902d925 DH |
4501 | error = PTR_ERR(dentry); |
4502 | if (IS_ERR(dentry)) | |
da2d0ced | 4503 | goto out_putnames; |
6902d925 | 4504 | |
91a27b2a | 4505 | error = security_path_symlink(&path, dentry, from->name); |
abf08576 CB |
4506 | if (!error) |
4507 | error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode, | |
4508 | dentry, from->name); | |
921a1650 | 4509 | done_path_create(&path, dentry); |
f46d3567 JL |
4510 | if (retry_estale(error, lookup_flags)) { |
4511 | lookup_flags |= LOOKUP_REVAL; | |
4512 | goto retry; | |
4513 | } | |
da2d0ced DK |
4514 | out_putnames: |
4515 | putname(to); | |
1da177e4 LT |
4516 | putname(from); |
4517 | return error; | |
4518 | } | |
4519 | ||
b724e846 DB |
4520 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4521 | int, newdfd, const char __user *, newname) | |
4522 | { | |
da2d0ced | 4523 | return do_symlinkat(getname(oldname), newdfd, getname(newname)); |
b724e846 DB |
4524 | } |
4525 | ||
3480b257 | 4526 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4527 | { |
da2d0ced | 4528 | return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); |
5590ff0d UD |
4529 | } |
4530 | ||
146a8595 BF |
4531 | /** |
4532 | * vfs_link - create a new link | |
4533 | * @old_dentry: object to be linked | |
abf08576 | 4534 | * @idmap: idmap of the mount |
146a8595 BF |
4535 | * @dir: new parent |
4536 | * @new_dentry: where to create the new link | |
4537 | * @delegated_inode: returns inode needing a delegation break | |
4538 | * | |
4539 | * The caller must hold dir->i_mutex | |
4540 | * | |
4541 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4542 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4543 | * inode in delegated_inode. The caller should then break the delegation | |
4544 | * and retry. Because breaking a delegation may take a long time, the | |
4545 | * caller should drop the i_mutex before doing so. | |
4546 | * | |
4547 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4548 | * be appropriate for callers that expect the underlying filesystem not | |
4549 | * to be NFS exported. | |
6521f891 | 4550 | * |
abf08576 CB |
4551 | * If the inode has been found through an idmapped mount the idmap of |
4552 | * the vfsmount must be passed through @idmap. This function will then take | |
4553 | * care to map the inode according to @idmap before checking permissions. | |
6521f891 | 4554 | * On non-idmapped mounts or if permission checking is to be performed on the |
376870aa | 4555 | * raw inode simply pass @nop_mnt_idmap. |
146a8595 | 4556 | */ |
abf08576 | 4557 | int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap, |
6521f891 CB |
4558 | struct inode *dir, struct dentry *new_dentry, |
4559 | struct inode **delegated_inode) | |
1da177e4 LT |
4560 | { |
4561 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4562 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4563 | int error; |
4564 | ||
4565 | if (!inode) | |
4566 | return -ENOENT; | |
4567 | ||
4609e1f1 | 4568 | error = may_create(idmap, dir, new_dentry); |
1da177e4 LT |
4569 | if (error) |
4570 | return error; | |
4571 | ||
4572 | if (dir->i_sb != inode->i_sb) | |
4573 | return -EXDEV; | |
4574 | ||
4575 | /* | |
4576 | * A link to an append-only or immutable file cannot be created. | |
4577 | */ | |
4578 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4579 | return -EPERM; | |
0bd23d09 EB |
4580 | /* |
4581 | * Updating the link count will likely cause i_uid and i_gid to | |
4582 | * be writen back improperly if their true value is unknown to | |
4583 | * the vfs. | |
4584 | */ | |
4609e1f1 | 4585 | if (HAS_UNMAPPED_ID(idmap, inode)) |
0bd23d09 | 4586 | return -EPERM; |
acfa4380 | 4587 | if (!dir->i_op->link) |
1da177e4 | 4588 | return -EPERM; |
7e79eedb | 4589 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4590 | return -EPERM; |
4591 | ||
4592 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4593 | if (error) | |
4594 | return error; | |
4595 | ||
5955102c | 4596 | inode_lock(inode); |
aae8a97d | 4597 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4598 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4599 | error = -ENOENT; |
8de52778 AV |
4600 | else if (max_links && inode->i_nlink >= max_links) |
4601 | error = -EMLINK; | |
146a8595 BF |
4602 | else { |
4603 | error = try_break_deleg(inode, delegated_inode); | |
4604 | if (!error) | |
4605 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4606 | } | |
f4e0c30c AV |
4607 | |
4608 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4609 | spin_lock(&inode->i_lock); | |
4610 | inode->i_state &= ~I_LINKABLE; | |
4611 | spin_unlock(&inode->i_lock); | |
4612 | } | |
5955102c | 4613 | inode_unlock(inode); |
e31e14ec | 4614 | if (!error) |
7e79eedb | 4615 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4616 | return error; |
4617 | } | |
4d359507 | 4618 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4619 | |
4620 | /* | |
4621 | * Hardlinks are often used in delicate situations. We avoid | |
4622 | * security-related surprises by not following symlinks on the | |
4623 | * newname. --KAB | |
4624 | * | |
4625 | * We don't follow them on the oldname either to be compatible | |
4626 | * with linux 2.0, and to avoid hard-linking to directories | |
4627 | * and other special files. --ADM | |
4628 | */ | |
cf30da90 | 4629 | int do_linkat(int olddfd, struct filename *old, int newdfd, |
020250f3 | 4630 | struct filename *new, int flags) |
1da177e4 | 4631 | { |
abf08576 | 4632 | struct mnt_idmap *idmap; |
1da177e4 | 4633 | struct dentry *new_dentry; |
dae6ad8f | 4634 | struct path old_path, new_path; |
146a8595 | 4635 | struct inode *delegated_inode = NULL; |
11a7b371 | 4636 | int how = 0; |
1da177e4 | 4637 | int error; |
1da177e4 | 4638 | |
020250f3 DK |
4639 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { |
4640 | error = -EINVAL; | |
4641 | goto out_putnames; | |
4642 | } | |
11a7b371 | 4643 | /* |
f0cc6ffb LT |
4644 | * To use null names we require CAP_DAC_READ_SEARCH |
4645 | * This ensures that not everyone will be able to create | |
4646 | * handlink using the passed filedescriptor. | |
11a7b371 | 4647 | */ |
020250f3 DK |
4648 | if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { |
4649 | error = -ENOENT; | |
4650 | goto out_putnames; | |
f0cc6ffb | 4651 | } |
11a7b371 AK |
4652 | |
4653 | if (flags & AT_SYMLINK_FOLLOW) | |
4654 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4655 | retry: |
794ebcea | 4656 | error = filename_lookup(olddfd, old, how, &old_path, NULL); |
1da177e4 | 4657 | if (error) |
020250f3 | 4658 | goto out_putnames; |
2ad94ae6 | 4659 | |
b4a4f213 | 4660 | new_dentry = filename_create(newdfd, new, &new_path, |
442e31ca | 4661 | (how & LOOKUP_REVAL)); |
1da177e4 | 4662 | error = PTR_ERR(new_dentry); |
6902d925 | 4663 | if (IS_ERR(new_dentry)) |
020250f3 | 4664 | goto out_putpath; |
dae6ad8f AV |
4665 | |
4666 | error = -EXDEV; | |
4667 | if (old_path.mnt != new_path.mnt) | |
4668 | goto out_dput; | |
abf08576 | 4669 | idmap = mnt_idmap(new_path.mnt); |
4609e1f1 | 4670 | error = may_linkat(idmap, &old_path); |
800179c9 KC |
4671 | if (unlikely(error)) |
4672 | goto out_dput; | |
dae6ad8f | 4673 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4674 | if (error) |
a8104a9f | 4675 | goto out_dput; |
abf08576 | 4676 | error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode, |
6521f891 | 4677 | new_dentry, &delegated_inode); |
75c3f29d | 4678 | out_dput: |
921a1650 | 4679 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4680 | if (delegated_inode) { |
4681 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4682 | if (!error) { |
4683 | path_put(&old_path); | |
146a8595 | 4684 | goto retry; |
d22e6338 | 4685 | } |
146a8595 | 4686 | } |
442e31ca | 4687 | if (retry_estale(error, how)) { |
d22e6338 | 4688 | path_put(&old_path); |
442e31ca JL |
4689 | how |= LOOKUP_REVAL; |
4690 | goto retry; | |
4691 | } | |
020250f3 | 4692 | out_putpath: |
2d8f3038 | 4693 | path_put(&old_path); |
020250f3 DK |
4694 | out_putnames: |
4695 | putname(old); | |
4696 | putname(new); | |
1da177e4 LT |
4697 | |
4698 | return error; | |
4699 | } | |
4700 | ||
46ea89eb DB |
4701 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4702 | int, newdfd, const char __user *, newname, int, flags) | |
4703 | { | |
020250f3 DK |
4704 | return do_linkat(olddfd, getname_uflags(oldname, flags), |
4705 | newdfd, getname(newname), flags); | |
46ea89eb DB |
4706 | } |
4707 | ||
3480b257 | 4708 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4709 | { |
020250f3 | 4710 | return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); |
5590ff0d UD |
4711 | } |
4712 | ||
bc27027a MS |
4713 | /** |
4714 | * vfs_rename - rename a filesystem object | |
2111c3c0 | 4715 | * @rd: pointer to &struct renamedata info |
bc27027a MS |
4716 | * |
4717 | * The caller must hold multiple mutexes--see lock_rename()). | |
4718 | * | |
4719 | * If vfs_rename discovers a delegation in need of breaking at either | |
4720 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4721 | * reference to the inode in delegated_inode. The caller should then | |
4722 | * break the delegation and retry. Because breaking a delegation may | |
4723 | * take a long time, the caller should drop all locks before doing | |
4724 | * so. | |
4725 | * | |
4726 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4727 | * be appropriate for callers that expect the underlying filesystem not | |
4728 | * to be NFS exported. | |
4729 | * | |
1da177e4 LT |
4730 | * The worst of all namespace operations - renaming directory. "Perverted" |
4731 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4732 | * Problems: | |
0117d427 | 4733 | * |
d03b29a2 | 4734 | * a) we can get into loop creation. |
1da177e4 | 4735 | * b) race potential - two innocent renames can create a loop together. |
22e111ed | 4736 | * That's where 4.4BSD screws up. Current fix: serialization on |
a11f3a05 | 4737 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4738 | * story. |
22e111ed AV |
4739 | * c) we may have to lock up to _four_ objects - parents and victim (if it exists), |
4740 | * and source (if it's a non-directory or a subdirectory that moves to | |
4741 | * different parent). | |
1b1dcc1b | 4742 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4743 | * whether the target exists). Solution: try to be smart with locking |
4744 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4745 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4746 | * move will be locked. Thus we can rank directories by the tree |
4747 | * (ancestors first) and rank all non-directories after them. | |
4748 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4749 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4750 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4751 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4752 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4753 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4754 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4755 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4756 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4757 | * locking]. |
4758 | */ | |
9fe61450 | 4759 | int vfs_rename(struct renamedata *rd) |
1da177e4 | 4760 | { |
bc27027a | 4761 | int error; |
9fe61450 CB |
4762 | struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; |
4763 | struct dentry *old_dentry = rd->old_dentry; | |
4764 | struct dentry *new_dentry = rd->new_dentry; | |
4765 | struct inode **delegated_inode = rd->delegated_inode; | |
4766 | unsigned int flags = rd->flags; | |
bc27027a | 4767 | bool is_dir = d_is_dir(old_dentry); |
bc27027a | 4768 | struct inode *source = old_dentry->d_inode; |
9055cba7 | 4769 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4770 | bool new_is_dir = false; |
4771 | unsigned max_links = new_dir->i_sb->s_max_links; | |
49d31c2f | 4772 | struct name_snapshot old_name; |
22e111ed | 4773 | bool lock_old_subdir, lock_new_subdir; |
bc27027a | 4774 | |
8d3e2936 | 4775 | if (source == target) |
bc27027a MS |
4776 | return 0; |
4777 | ||
4609e1f1 | 4778 | error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir); |
bc27027a MS |
4779 | if (error) |
4780 | return error; | |
4781 | ||
da1ce067 | 4782 | if (!target) { |
4609e1f1 | 4783 | error = may_create(rd->new_mnt_idmap, new_dir, new_dentry); |
da1ce067 MS |
4784 | } else { |
4785 | new_is_dir = d_is_dir(new_dentry); | |
4786 | ||
4787 | if (!(flags & RENAME_EXCHANGE)) | |
4609e1f1 | 4788 | error = may_delete(rd->new_mnt_idmap, new_dir, |
6521f891 | 4789 | new_dentry, is_dir); |
da1ce067 | 4790 | else |
4609e1f1 | 4791 | error = may_delete(rd->new_mnt_idmap, new_dir, |
6521f891 | 4792 | new_dentry, new_is_dir); |
da1ce067 | 4793 | } |
bc27027a MS |
4794 | if (error) |
4795 | return error; | |
4796 | ||
2773bf00 | 4797 | if (!old_dir->i_op->rename) |
bc27027a | 4798 | return -EPERM; |
1da177e4 LT |
4799 | |
4800 | /* | |
4801 | * If we are going to change the parent - check write permissions, | |
4802 | * we'll need to flip '..'. | |
4803 | */ | |
da1ce067 MS |
4804 | if (new_dir != old_dir) { |
4805 | if (is_dir) { | |
4609e1f1 | 4806 | error = inode_permission(rd->old_mnt_idmap, source, |
47291baa | 4807 | MAY_WRITE); |
da1ce067 MS |
4808 | if (error) |
4809 | return error; | |
4810 | } | |
4811 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4609e1f1 | 4812 | error = inode_permission(rd->new_mnt_idmap, target, |
47291baa | 4813 | MAY_WRITE); |
da1ce067 MS |
4814 | if (error) |
4815 | return error; | |
4816 | } | |
1da177e4 LT |
4817 | } |
4818 | ||
0b3974eb MS |
4819 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4820 | flags); | |
1da177e4 LT |
4821 | if (error) |
4822 | return error; | |
4823 | ||
49d31c2f | 4824 | take_dentry_name_snapshot(&old_name, old_dentry); |
1d2ef590 | 4825 | dget(new_dentry); |
28eceeda | 4826 | /* |
22e111ed AV |
4827 | * Lock children. |
4828 | * The source subdirectory needs to be locked on cross-directory | |
4829 | * rename or cross-directory exchange since its parent changes. | |
4830 | * The target subdirectory needs to be locked on cross-directory | |
4831 | * exchange due to parent change and on any rename due to becoming | |
4832 | * a victim. | |
4833 | * Non-directories need locking in all cases (for NFS reasons); | |
4834 | * they get locked after any subdirectories (in inode address order). | |
4835 | * | |
4836 | * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE. | |
4837 | * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex. | |
28eceeda | 4838 | */ |
22e111ed AV |
4839 | lock_old_subdir = new_dir != old_dir; |
4840 | lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE); | |
4841 | if (is_dir) { | |
4842 | if (lock_old_subdir) | |
4843 | inode_lock_nested(source, I_MUTEX_CHILD); | |
4844 | if (target && (!new_is_dir || lock_new_subdir)) | |
4845 | inode_lock(target); | |
4846 | } else if (new_is_dir) { | |
4847 | if (lock_new_subdir) | |
4848 | inode_lock_nested(target, I_MUTEX_CHILD); | |
4849 | inode_lock(source); | |
4850 | } else { | |
4851 | lock_two_nondirectories(source, target); | |
4852 | } | |
9055cba7 | 4853 | |
51cc3a66 HD |
4854 | error = -EPERM; |
4855 | if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) | |
4856 | goto out; | |
4857 | ||
9055cba7 | 4858 | error = -EBUSY; |
7af1364f | 4859 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4860 | goto out; |
4861 | ||
da1ce067 | 4862 | if (max_links && new_dir != old_dir) { |
bc27027a | 4863 | error = -EMLINK; |
da1ce067 | 4864 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4865 | goto out; |
da1ce067 MS |
4866 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4867 | old_dir->i_nlink >= max_links) | |
4868 | goto out; | |
4869 | } | |
da1ce067 | 4870 | if (!is_dir) { |
bc27027a | 4871 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4872 | if (error) |
4873 | goto out; | |
da1ce067 MS |
4874 | } |
4875 | if (target && !new_is_dir) { | |
4876 | error = try_break_deleg(target, delegated_inode); | |
4877 | if (error) | |
4878 | goto out; | |
8e6d782c | 4879 | } |
e18275ae | 4880 | error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry, |
549c7297 | 4881 | new_dir, new_dentry, flags); |
51892bbb SW |
4882 | if (error) |
4883 | goto out; | |
4884 | ||
da1ce067 | 4885 | if (!(flags & RENAME_EXCHANGE) && target) { |
8767712f AV |
4886 | if (is_dir) { |
4887 | shrink_dcache_parent(new_dentry); | |
bc27027a | 4888 | target->i_flags |= S_DEAD; |
8767712f | 4889 | } |
51892bbb | 4890 | dont_mount(new_dentry); |
8ed936b5 | 4891 | detach_mounts(new_dentry); |
bc27027a | 4892 | } |
da1ce067 MS |
4893 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4894 | if (!(flags & RENAME_EXCHANGE)) | |
4895 | d_move(old_dentry, new_dentry); | |
4896 | else | |
4897 | d_exchange(old_dentry, new_dentry); | |
4898 | } | |
51892bbb | 4899 | out: |
22e111ed AV |
4900 | if (!is_dir || lock_old_subdir) |
4901 | inode_unlock(source); | |
4902 | if (target && (!new_is_dir || lock_new_subdir)) | |
5955102c | 4903 | inode_unlock(target); |
1da177e4 | 4904 | dput(new_dentry); |
da1ce067 | 4905 | if (!error) { |
f4ec3a3d | 4906 | fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, |
da1ce067 MS |
4907 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4908 | if (flags & RENAME_EXCHANGE) { | |
f4ec3a3d | 4909 | fsnotify_move(new_dir, old_dir, &old_dentry->d_name, |
da1ce067 MS |
4910 | new_is_dir, NULL, new_dentry); |
4911 | } | |
4912 | } | |
49d31c2f | 4913 | release_dentry_name_snapshot(&old_name); |
0eeca283 | 4914 | |
1da177e4 LT |
4915 | return error; |
4916 | } | |
4d359507 | 4917 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4918 | |
e886663c JA |
4919 | int do_renameat2(int olddfd, struct filename *from, int newdfd, |
4920 | struct filename *to, unsigned int flags) | |
1da177e4 | 4921 | { |
9fe61450 | 4922 | struct renamedata rd; |
2ad94ae6 AV |
4923 | struct dentry *old_dentry, *new_dentry; |
4924 | struct dentry *trap; | |
f5beed75 AV |
4925 | struct path old_path, new_path; |
4926 | struct qstr old_last, new_last; | |
4927 | int old_type, new_type; | |
8e6d782c | 4928 | struct inode *delegated_inode = NULL; |
f5beed75 | 4929 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4930 | bool should_retry = false; |
e886663c | 4931 | int error = -EINVAL; |
520c8b16 | 4932 | |
0d7a8555 | 4933 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
0ee50b47 | 4934 | goto put_names; |
da1ce067 | 4935 | |
0d7a8555 MS |
4936 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4937 | (flags & RENAME_EXCHANGE)) | |
0ee50b47 | 4938 | goto put_names; |
520c8b16 | 4939 | |
f5beed75 AV |
4940 | if (flags & RENAME_EXCHANGE) |
4941 | target_flags = 0; | |
4942 | ||
c6a94284 | 4943 | retry: |
c5f563f9 AV |
4944 | error = filename_parentat(olddfd, from, lookup_flags, &old_path, |
4945 | &old_last, &old_type); | |
0ee50b47 DK |
4946 | if (error) |
4947 | goto put_names; | |
1da177e4 | 4948 | |
c5f563f9 AV |
4949 | error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, |
4950 | &new_type); | |
0ee50b47 | 4951 | if (error) |
1da177e4 LT |
4952 | goto exit1; |
4953 | ||
4954 | error = -EXDEV; | |
f5beed75 | 4955 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4956 | goto exit2; |
4957 | ||
1da177e4 | 4958 | error = -EBUSY; |
f5beed75 | 4959 | if (old_type != LAST_NORM) |
1da177e4 LT |
4960 | goto exit2; |
4961 | ||
0a7c3937 MS |
4962 | if (flags & RENAME_NOREPLACE) |
4963 | error = -EEXIST; | |
f5beed75 | 4964 | if (new_type != LAST_NORM) |
1da177e4 LT |
4965 | goto exit2; |
4966 | ||
f5beed75 | 4967 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4968 | if (error) |
4969 | goto exit2; | |
4970 | ||
8e6d782c | 4971 | retry_deleg: |
f5beed75 | 4972 | trap = lock_rename(new_path.dentry, old_path.dentry); |
a8b00268 AV |
4973 | if (IS_ERR(trap)) { |
4974 | error = PTR_ERR(trap); | |
4975 | goto exit_lock_rename; | |
4976 | } | |
1da177e4 | 4977 | |
74d7970f NJ |
4978 | old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry, |
4979 | lookup_flags); | |
1da177e4 LT |
4980 | error = PTR_ERR(old_dentry); |
4981 | if (IS_ERR(old_dentry)) | |
4982 | goto exit3; | |
4983 | /* source must exist */ | |
4984 | error = -ENOENT; | |
b18825a7 | 4985 | if (d_is_negative(old_dentry)) |
1da177e4 | 4986 | goto exit4; |
74d7970f NJ |
4987 | new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry, |
4988 | lookup_flags | target_flags); | |
0a7c3937 MS |
4989 | error = PTR_ERR(new_dentry); |
4990 | if (IS_ERR(new_dentry)) | |
4991 | goto exit4; | |
4992 | error = -EEXIST; | |
4993 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4994 | goto exit5; | |
da1ce067 MS |
4995 | if (flags & RENAME_EXCHANGE) { |
4996 | error = -ENOENT; | |
4997 | if (d_is_negative(new_dentry)) | |
4998 | goto exit5; | |
4999 | ||
5000 | if (!d_is_dir(new_dentry)) { | |
5001 | error = -ENOTDIR; | |
f5beed75 | 5002 | if (new_last.name[new_last.len]) |
da1ce067 MS |
5003 | goto exit5; |
5004 | } | |
5005 | } | |
1da177e4 | 5006 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 5007 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 5008 | error = -ENOTDIR; |
f5beed75 | 5009 | if (old_last.name[old_last.len]) |
0a7c3937 | 5010 | goto exit5; |
f5beed75 | 5011 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 5012 | goto exit5; |
1da177e4 LT |
5013 | } |
5014 | /* source should not be ancestor of target */ | |
5015 | error = -EINVAL; | |
5016 | if (old_dentry == trap) | |
0a7c3937 | 5017 | goto exit5; |
1da177e4 | 5018 | /* target should not be an ancestor of source */ |
da1ce067 MS |
5019 | if (!(flags & RENAME_EXCHANGE)) |
5020 | error = -ENOTEMPTY; | |
1da177e4 LT |
5021 | if (new_dentry == trap) |
5022 | goto exit5; | |
5023 | ||
f5beed75 AV |
5024 | error = security_path_rename(&old_path, old_dentry, |
5025 | &new_path, new_dentry, flags); | |
be6d3e56 | 5026 | if (error) |
c30dabfe | 5027 | goto exit5; |
9fe61450 CB |
5028 | |
5029 | rd.old_dir = old_path.dentry->d_inode; | |
5030 | rd.old_dentry = old_dentry; | |
abf08576 | 5031 | rd.old_mnt_idmap = mnt_idmap(old_path.mnt); |
9fe61450 CB |
5032 | rd.new_dir = new_path.dentry->d_inode; |
5033 | rd.new_dentry = new_dentry; | |
abf08576 | 5034 | rd.new_mnt_idmap = mnt_idmap(new_path.mnt); |
9fe61450 CB |
5035 | rd.delegated_inode = &delegated_inode; |
5036 | rd.flags = flags; | |
5037 | error = vfs_rename(&rd); | |
1da177e4 LT |
5038 | exit5: |
5039 | dput(new_dentry); | |
5040 | exit4: | |
5041 | dput(old_dentry); | |
5042 | exit3: | |
f5beed75 | 5043 | unlock_rename(new_path.dentry, old_path.dentry); |
a8b00268 | 5044 | exit_lock_rename: |
8e6d782c BF |
5045 | if (delegated_inode) { |
5046 | error = break_deleg_wait(&delegated_inode); | |
5047 | if (!error) | |
5048 | goto retry_deleg; | |
5049 | } | |
f5beed75 | 5050 | mnt_drop_write(old_path.mnt); |
1da177e4 | 5051 | exit2: |
c6a94284 JL |
5052 | if (retry_estale(error, lookup_flags)) |
5053 | should_retry = true; | |
f5beed75 | 5054 | path_put(&new_path); |
1da177e4 | 5055 | exit1: |
f5beed75 | 5056 | path_put(&old_path); |
c6a94284 JL |
5057 | if (should_retry) { |
5058 | should_retry = false; | |
5059 | lookup_flags |= LOOKUP_REVAL; | |
5060 | goto retry; | |
5061 | } | |
0ee50b47 | 5062 | put_names: |
91ef658f | 5063 | putname(from); |
91ef658f | 5064 | putname(to); |
1da177e4 LT |
5065 | return error; |
5066 | } | |
5067 | ||
ee81feb6 DB |
5068 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
5069 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
5070 | { | |
e886663c JA |
5071 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), |
5072 | flags); | |
ee81feb6 DB |
5073 | } |
5074 | ||
520c8b16 MS |
5075 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
5076 | int, newdfd, const char __user *, newname) | |
5077 | { | |
e886663c JA |
5078 | return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), |
5079 | 0); | |
520c8b16 MS |
5080 | } |
5081 | ||
a26eab24 | 5082 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 5083 | { |
e886663c JA |
5084 | return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, |
5085 | getname(newname), 0); | |
5590ff0d UD |
5086 | } |
5087 | ||
5d826c84 | 5088 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 5089 | { |
5d826c84 | 5090 | int len = PTR_ERR(link); |
1da177e4 LT |
5091 | if (IS_ERR(link)) |
5092 | goto out; | |
5093 | ||
5094 | len = strlen(link); | |
5095 | if (len > (unsigned) buflen) | |
5096 | len = buflen; | |
5097 | if (copy_to_user(buffer, link, len)) | |
5098 | len = -EFAULT; | |
5099 | out: | |
5100 | return len; | |
5101 | } | |
5102 | ||
fd4a0edf MS |
5103 | /** |
5104 | * vfs_readlink - copy symlink body into userspace buffer | |
5105 | * @dentry: dentry on which to get symbolic link | |
5106 | * @buffer: user memory pointer | |
5107 | * @buflen: size of buffer | |
5108 | * | |
5109 | * Does not touch atime. That's up to the caller if necessary | |
5110 | * | |
5111 | * Does not call security hook. | |
5112 | */ | |
5113 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
5114 | { | |
5115 | struct inode *inode = d_inode(dentry); | |
f2df5da6 AV |
5116 | DEFINE_DELAYED_CALL(done); |
5117 | const char *link; | |
5118 | int res; | |
fd4a0edf | 5119 | |
76fca90e MS |
5120 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
5121 | if (unlikely(inode->i_op->readlink)) | |
5122 | return inode->i_op->readlink(dentry, buffer, buflen); | |
5123 | ||
5124 | if (!d_is_symlink(dentry)) | |
5125 | return -EINVAL; | |
5126 | ||
5127 | spin_lock(&inode->i_lock); | |
5128 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
5129 | spin_unlock(&inode->i_lock); | |
5130 | } | |
fd4a0edf | 5131 | |
4c4f7c19 | 5132 | link = READ_ONCE(inode->i_link); |
f2df5da6 AV |
5133 | if (!link) { |
5134 | link = inode->i_op->get_link(dentry, inode, &done); | |
5135 | if (IS_ERR(link)) | |
5136 | return PTR_ERR(link); | |
5137 | } | |
5138 | res = readlink_copy(buffer, buflen, link); | |
5139 | do_delayed_call(&done); | |
5140 | return res; | |
fd4a0edf MS |
5141 | } |
5142 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 5143 | |
d60874cd MS |
5144 | /** |
5145 | * vfs_get_link - get symlink body | |
5146 | * @dentry: dentry on which to get symbolic link | |
5147 | * @done: caller needs to free returned data with this | |
5148 | * | |
5149 | * Calls security hook and i_op->get_link() on the supplied inode. | |
5150 | * | |
5151 | * It does not touch atime. That's up to the caller if necessary. | |
5152 | * | |
5153 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
5154 | */ | |
5155 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
5156 | { | |
5157 | const char *res = ERR_PTR(-EINVAL); | |
5158 | struct inode *inode = d_inode(dentry); | |
5159 | ||
5160 | if (d_is_symlink(dentry)) { | |
5161 | res = ERR_PTR(security_inode_readlink(dentry)); | |
5162 | if (!res) | |
5163 | res = inode->i_op->get_link(dentry, inode, done); | |
5164 | } | |
5165 | return res; | |
5166 | } | |
5167 | EXPORT_SYMBOL(vfs_get_link); | |
5168 | ||
1da177e4 | 5169 | /* get the link contents into pagecache */ |
6b255391 | 5170 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 5171 | struct delayed_call *callback) |
1da177e4 | 5172 | { |
ebd09abb DG |
5173 | char *kaddr; |
5174 | struct page *page; | |
6b255391 AV |
5175 | struct address_space *mapping = inode->i_mapping; |
5176 | ||
d3883d4f AV |
5177 | if (!dentry) { |
5178 | page = find_get_page(mapping, 0); | |
5179 | if (!page) | |
5180 | return ERR_PTR(-ECHILD); | |
5181 | if (!PageUptodate(page)) { | |
5182 | put_page(page); | |
5183 | return ERR_PTR(-ECHILD); | |
5184 | } | |
5185 | } else { | |
5186 | page = read_mapping_page(mapping, 0, NULL); | |
5187 | if (IS_ERR(page)) | |
5188 | return (char*)page; | |
5189 | } | |
fceef393 | 5190 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
5191 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
5192 | kaddr = page_address(page); | |
6b255391 | 5193 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 5194 | return kaddr; |
1da177e4 LT |
5195 | } |
5196 | ||
6b255391 | 5197 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 5198 | |
fceef393 | 5199 | void page_put_link(void *arg) |
1da177e4 | 5200 | { |
fceef393 | 5201 | put_page(arg); |
1da177e4 | 5202 | } |
4d359507 | 5203 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 5204 | |
aa80deab AV |
5205 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
5206 | { | |
fceef393 | 5207 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
5208 | int res = readlink_copy(buffer, buflen, |
5209 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
5210 | &done)); |
5211 | do_delayed_call(&done); | |
aa80deab AV |
5212 | return res; |
5213 | } | |
5214 | EXPORT_SYMBOL(page_readlink); | |
5215 | ||
56f5746c | 5216 | int page_symlink(struct inode *inode, const char *symname, int len) |
1da177e4 LT |
5217 | { |
5218 | struct address_space *mapping = inode->i_mapping; | |
27a77913 | 5219 | const struct address_space_operations *aops = mapping->a_ops; |
56f5746c | 5220 | bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); |
0adb25d2 | 5221 | struct page *page; |
1468c6f4 | 5222 | void *fsdata = NULL; |
beb497ab | 5223 | int err; |
2d878178 | 5224 | unsigned int flags; |
1da177e4 | 5225 | |
7e53cac4 | 5226 | retry: |
2d878178 MWO |
5227 | if (nofs) |
5228 | flags = memalloc_nofs_save(); | |
27a77913 | 5229 | err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); |
2d878178 MWO |
5230 | if (nofs) |
5231 | memalloc_nofs_restore(flags); | |
1da177e4 | 5232 | if (err) |
afddba49 NP |
5233 | goto fail; |
5234 | ||
21fc61c7 | 5235 | memcpy(page_address(page), symname, len-1); |
afddba49 | 5236 | |
27a77913 | 5237 | err = aops->write_end(NULL, mapping, 0, len-1, len-1, |
afddba49 | 5238 | page, fsdata); |
1da177e4 LT |
5239 | if (err < 0) |
5240 | goto fail; | |
afddba49 NP |
5241 | if (err < len-1) |
5242 | goto retry; | |
5243 | ||
1da177e4 LT |
5244 | mark_inode_dirty(inode); |
5245 | return 0; | |
1da177e4 LT |
5246 | fail: |
5247 | return err; | |
5248 | } | |
4d359507 | 5249 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 5250 | |
92e1d5be | 5251 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 5252 | .get_link = page_get_link, |
1da177e4 | 5253 | }; |
1da177e4 | 5254 | EXPORT_SYMBOL(page_symlink_inode_operations); |