]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namei.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Some corrections by tytso. | |
10 | */ | |
11 | ||
12 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
13 | * lookup logic. | |
14 | */ | |
15 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
16 | */ | |
17 | ||
18 | #include <linux/init.h> | |
630d9c47 | 19 | #include <linux/export.h> |
44696908 | 20 | #include <linux/kernel.h> |
1da177e4 LT |
21 | #include <linux/slab.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/namei.h> | |
1da177e4 | 24 | #include <linux/pagemap.h> |
0eeca283 | 25 | #include <linux/fsnotify.h> |
1da177e4 LT |
26 | #include <linux/personality.h> |
27 | #include <linux/security.h> | |
6146f0d5 | 28 | #include <linux/ima.h> |
1da177e4 LT |
29 | #include <linux/syscalls.h> |
30 | #include <linux/mount.h> | |
31 | #include <linux/audit.h> | |
16f7e0fe | 32 | #include <linux/capability.h> |
834f2a4a | 33 | #include <linux/file.h> |
5590ff0d | 34 | #include <linux/fcntl.h> |
08ce5f16 | 35 | #include <linux/device_cgroup.h> |
5ad4e53b | 36 | #include <linux/fs_struct.h> |
e77819e5 | 37 | #include <linux/posix_acl.h> |
99d263d4 | 38 | #include <linux/hash.h> |
2a18da7a | 39 | #include <linux/bitops.h> |
aeaa4a79 | 40 | #include <linux/init_task.h> |
7c0f6ba6 | 41 | #include <linux/uaccess.h> |
1da177e4 | 42 | |
e81e3f4d | 43 | #include "internal.h" |
c7105365 | 44 | #include "mount.h" |
e81e3f4d | 45 | |
1da177e4 LT |
46 | /* [Feb-1997 T. Schoebel-Theuer] |
47 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
48 | * were necessary because of omirr. The reason is that omirr needs | |
49 | * to know the _real_ pathname, not the user-supplied one, in case | |
50 | * of symlinks (and also when transname replacements occur). | |
51 | * | |
52 | * The new code replaces the old recursive symlink resolution with | |
53 | * an iterative one (in case of non-nested symlink chains). It does | |
54 | * this with calls to <fs>_follow_link(). | |
55 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
56 | * replaced with a single function lookup_dentry() that can handle all | |
57 | * the special cases of the former code. | |
58 | * | |
59 | * With the new dcache, the pathname is stored at each inode, at least as | |
60 | * long as the refcount of the inode is positive. As a side effect, the | |
61 | * size of the dcache depends on the inode cache and thus is dynamic. | |
62 | * | |
63 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
64 | * resolution to correspond with current state of the code. | |
65 | * | |
66 | * Note that the symlink resolution is not *completely* iterative. | |
67 | * There is still a significant amount of tail- and mid- recursion in | |
68 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
69 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
70 | * may return different results than <fs>_follow_link(). Many virtual | |
71 | * filesystems (including /proc) exhibit this behavior. | |
72 | */ | |
73 | ||
74 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
75 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
76 | * and the name already exists in form of a symlink, try to create the new | |
77 | * name indicated by the symlink. The old code always complained that the | |
78 | * name already exists, due to not following the symlink even if its target | |
79 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 80 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
81 | * |
82 | * I don't know which semantics is the right one, since I have no access | |
83 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
84 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
85 | * "old" one. Personally, I think the new semantics is much more logical. | |
86 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
87 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
88 | * and in the old Linux semantics. | |
89 | */ | |
90 | ||
91 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
92 | * semantics. See the comments in "open_namei" and "do_link" below. | |
93 | * | |
94 | * [10-Sep-98 Alan Modra] Another symlink change. | |
95 | */ | |
96 | ||
97 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
98 | * inside the path - always follow. | |
99 | * in the last component in creation/removal/renaming - never follow. | |
100 | * if LOOKUP_FOLLOW passed - follow. | |
101 | * if the pathname has trailing slashes - follow. | |
102 | * otherwise - don't follow. | |
103 | * (applied in that order). | |
104 | * | |
105 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
106 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
107 | * During the 2.4 we need to fix the userland stuff depending on it - | |
108 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
109 | * XEmacs seems to be relying on it... | |
110 | */ | |
111 | /* | |
112 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 113 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
114 | * any extra contention... |
115 | */ | |
116 | ||
117 | /* In order to reduce some races, while at the same time doing additional | |
118 | * checking and hopefully speeding things up, we copy filenames to the | |
119 | * kernel data space before using them.. | |
120 | * | |
121 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
122 | * PATH_MAX includes the nul terminator --RR. | |
123 | */ | |
91a27b2a | 124 | |
fd2f7cb5 | 125 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) |
7950e385 | 126 | |
51f39a1f | 127 | struct filename * |
91a27b2a JL |
128 | getname_flags(const char __user *filename, int flags, int *empty) |
129 | { | |
94b5d262 | 130 | struct filename *result; |
7950e385 | 131 | char *kname; |
94b5d262 | 132 | int len; |
4043cde8 | 133 | |
7ac86265 JL |
134 | result = audit_reusename(filename); |
135 | if (result) | |
136 | return result; | |
137 | ||
7950e385 | 138 | result = __getname(); |
3f9f0aa6 | 139 | if (unlikely(!result)) |
4043cde8 EP |
140 | return ERR_PTR(-ENOMEM); |
141 | ||
7950e385 JL |
142 | /* |
143 | * First, try to embed the struct filename inside the names_cache | |
144 | * allocation | |
145 | */ | |
fd2f7cb5 | 146 | kname = (char *)result->iname; |
91a27b2a | 147 | result->name = kname; |
7950e385 | 148 | |
94b5d262 | 149 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); |
91a27b2a | 150 | if (unlikely(len < 0)) { |
94b5d262 AV |
151 | __putname(result); |
152 | return ERR_PTR(len); | |
91a27b2a | 153 | } |
3f9f0aa6 | 154 | |
7950e385 JL |
155 | /* |
156 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
157 | * separate struct filename so we can dedicate the entire | |
158 | * names_cache allocation for the pathname, and re-do the copy from | |
159 | * userland. | |
160 | */ | |
94b5d262 | 161 | if (unlikely(len == EMBEDDED_NAME_MAX)) { |
fd2f7cb5 | 162 | const size_t size = offsetof(struct filename, iname[1]); |
7950e385 JL |
163 | kname = (char *)result; |
164 | ||
fd2f7cb5 AV |
165 | /* |
166 | * size is chosen that way we to guarantee that | |
167 | * result->iname[0] is within the same object and that | |
168 | * kname can't be equal to result->iname, no matter what. | |
169 | */ | |
170 | result = kzalloc(size, GFP_KERNEL); | |
94b5d262 AV |
171 | if (unlikely(!result)) { |
172 | __putname(kname); | |
173 | return ERR_PTR(-ENOMEM); | |
7950e385 JL |
174 | } |
175 | result->name = kname; | |
94b5d262 AV |
176 | len = strncpy_from_user(kname, filename, PATH_MAX); |
177 | if (unlikely(len < 0)) { | |
178 | __putname(kname); | |
179 | kfree(result); | |
180 | return ERR_PTR(len); | |
181 | } | |
182 | if (unlikely(len == PATH_MAX)) { | |
183 | __putname(kname); | |
184 | kfree(result); | |
185 | return ERR_PTR(-ENAMETOOLONG); | |
186 | } | |
7950e385 JL |
187 | } |
188 | ||
94b5d262 | 189 | result->refcnt = 1; |
3f9f0aa6 LT |
190 | /* The empty path is special. */ |
191 | if (unlikely(!len)) { | |
192 | if (empty) | |
4043cde8 | 193 | *empty = 1; |
94b5d262 AV |
194 | if (!(flags & LOOKUP_EMPTY)) { |
195 | putname(result); | |
196 | return ERR_PTR(-ENOENT); | |
197 | } | |
1da177e4 | 198 | } |
3f9f0aa6 | 199 | |
7950e385 | 200 | result->uptr = filename; |
c4ad8f98 | 201 | result->aname = NULL; |
7950e385 JL |
202 | audit_getname(result); |
203 | return result; | |
1da177e4 LT |
204 | } |
205 | ||
91a27b2a JL |
206 | struct filename * |
207 | getname(const char __user * filename) | |
f52e0c11 | 208 | { |
f7493e5d | 209 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
210 | } |
211 | ||
c4ad8f98 LT |
212 | struct filename * |
213 | getname_kernel(const char * filename) | |
214 | { | |
215 | struct filename *result; | |
08518549 | 216 | int len = strlen(filename) + 1; |
c4ad8f98 LT |
217 | |
218 | result = __getname(); | |
219 | if (unlikely(!result)) | |
220 | return ERR_PTR(-ENOMEM); | |
221 | ||
08518549 | 222 | if (len <= EMBEDDED_NAME_MAX) { |
fd2f7cb5 | 223 | result->name = (char *)result->iname; |
08518549 | 224 | } else if (len <= PATH_MAX) { |
0585bd66 | 225 | const size_t size = offsetof(struct filename, iname[1]); |
08518549 PM |
226 | struct filename *tmp; |
227 | ||
0585bd66 | 228 | tmp = kmalloc(size, GFP_KERNEL); |
08518549 PM |
229 | if (unlikely(!tmp)) { |
230 | __putname(result); | |
231 | return ERR_PTR(-ENOMEM); | |
232 | } | |
233 | tmp->name = (char *)result; | |
08518549 PM |
234 | result = tmp; |
235 | } else { | |
236 | __putname(result); | |
237 | return ERR_PTR(-ENAMETOOLONG); | |
238 | } | |
239 | memcpy((char *)result->name, filename, len); | |
c4ad8f98 LT |
240 | result->uptr = NULL; |
241 | result->aname = NULL; | |
55422d0b | 242 | result->refcnt = 1; |
fd3522fd | 243 | audit_getname(result); |
c4ad8f98 | 244 | |
c4ad8f98 LT |
245 | return result; |
246 | } | |
247 | ||
91a27b2a | 248 | void putname(struct filename *name) |
1da177e4 | 249 | { |
55422d0b PM |
250 | BUG_ON(name->refcnt <= 0); |
251 | ||
252 | if (--name->refcnt > 0) | |
253 | return; | |
254 | ||
fd2f7cb5 | 255 | if (name->name != name->iname) { |
55422d0b PM |
256 | __putname(name->name); |
257 | kfree(name); | |
258 | } else | |
259 | __putname(name); | |
1da177e4 | 260 | } |
1da177e4 | 261 | |
e77819e5 LT |
262 | static int check_acl(struct inode *inode, int mask) |
263 | { | |
84635d68 | 264 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
265 | struct posix_acl *acl; |
266 | ||
e77819e5 | 267 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
268 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
269 | if (!acl) | |
e77819e5 | 270 | return -EAGAIN; |
3567866b | 271 | /* no ->get_acl() calls in RCU mode... */ |
b8a7a3a6 | 272 | if (is_uncached_acl(acl)) |
3567866b | 273 | return -ECHILD; |
206b1d09 | 274 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
275 | } |
276 | ||
2982baa2 CH |
277 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
278 | if (IS_ERR(acl)) | |
279 | return PTR_ERR(acl); | |
e77819e5 LT |
280 | if (acl) { |
281 | int error = posix_acl_permission(inode, acl, mask); | |
282 | posix_acl_release(acl); | |
283 | return error; | |
284 | } | |
84635d68 | 285 | #endif |
e77819e5 LT |
286 | |
287 | return -EAGAIN; | |
288 | } | |
289 | ||
5909ccaa | 290 | /* |
948409c7 | 291 | * This does the basic permission checking |
1da177e4 | 292 | */ |
7e40145e | 293 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 294 | { |
26cf46be | 295 | unsigned int mode = inode->i_mode; |
1da177e4 | 296 | |
8e96e3b7 | 297 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
298 | mode >>= 6; |
299 | else { | |
e77819e5 | 300 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 301 | int error = check_acl(inode, mask); |
b74c79e9 NP |
302 | if (error != -EAGAIN) |
303 | return error; | |
1da177e4 LT |
304 | } |
305 | ||
306 | if (in_group_p(inode->i_gid)) | |
307 | mode >>= 3; | |
308 | } | |
309 | ||
310 | /* | |
311 | * If the DACs are ok we don't need any capability check. | |
312 | */ | |
9c2c7039 | 313 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 314 | return 0; |
5909ccaa LT |
315 | return -EACCES; |
316 | } | |
317 | ||
318 | /** | |
b74c79e9 | 319 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 320 | * @inode: inode to check access rights for |
8fd90c8d | 321 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
322 | * |
323 | * Used to check for read/write/execute permissions on a file. | |
324 | * We use "fsuid" for this, letting us set arbitrary permissions | |
325 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
326 | * are used for other things. |
327 | * | |
328 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
329 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
330 | * It would then be called again in ref-walk mode. | |
5909ccaa | 331 | */ |
2830ba7f | 332 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
333 | { |
334 | int ret; | |
335 | ||
336 | /* | |
948409c7 | 337 | * Do the basic permission checks. |
5909ccaa | 338 | */ |
7e40145e | 339 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
340 | if (ret != -EACCES) |
341 | return ret; | |
1da177e4 | 342 | |
d594e7ec AV |
343 | if (S_ISDIR(inode->i_mode)) { |
344 | /* DACs are overridable for directories */ | |
d594e7ec | 345 | if (!(mask & MAY_WRITE)) |
23adbe12 AL |
346 | if (capable_wrt_inode_uidgid(inode, |
347 | CAP_DAC_READ_SEARCH)) | |
d594e7ec | 348 | return 0; |
23adbe12 | 349 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 | 350 | return 0; |
2a4c2242 SS |
351 | return -EACCES; |
352 | } | |
1da177e4 LT |
353 | |
354 | /* | |
355 | * Searching includes executable on directories, else just read. | |
356 | */ | |
7ea66001 | 357 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 358 | if (mask == MAY_READ) |
23adbe12 | 359 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 | 360 | return 0; |
2a4c2242 SS |
361 | /* |
362 | * Read/write DACs are always overridable. | |
363 | * Executable DACs are overridable when there is | |
364 | * at least one exec bit set. | |
365 | */ | |
366 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
367 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
368 | return 0; | |
1da177e4 LT |
369 | |
370 | return -EACCES; | |
371 | } | |
4d359507 | 372 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 373 | |
3ddcd056 LT |
374 | /* |
375 | * We _really_ want to just do "generic_permission()" without | |
376 | * even looking at the inode->i_op values. So we keep a cache | |
377 | * flag in inode->i_opflags, that says "this has not special | |
378 | * permission function, use the fast case". | |
379 | */ | |
380 | static inline int do_inode_permission(struct inode *inode, int mask) | |
381 | { | |
382 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
383 | if (likely(inode->i_op->permission)) | |
384 | return inode->i_op->permission(inode, mask); | |
385 | ||
386 | /* This gets set once for the inode lifetime */ | |
387 | spin_lock(&inode->i_lock); | |
388 | inode->i_opflags |= IOP_FASTPERM; | |
389 | spin_unlock(&inode->i_lock); | |
390 | } | |
391 | return generic_permission(inode, mask); | |
392 | } | |
393 | ||
cb23beb5 | 394 | /** |
0bdaea90 DH |
395 | * __inode_permission - Check for access rights to a given inode |
396 | * @inode: Inode to check permission on | |
397 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 398 | * |
0bdaea90 | 399 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
400 | * |
401 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
402 | * |
403 | * This does not check for a read-only file system. You probably want | |
404 | * inode_permission(). | |
cb23beb5 | 405 | */ |
0bdaea90 | 406 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 407 | { |
e6305c43 | 408 | int retval; |
1da177e4 | 409 | |
3ddcd056 | 410 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
411 | /* |
412 | * Nobody gets write access to an immutable file. | |
413 | */ | |
414 | if (IS_IMMUTABLE(inode)) | |
337684a1 | 415 | return -EPERM; |
0bd23d09 EB |
416 | |
417 | /* | |
418 | * Updating mtime will likely cause i_uid and i_gid to be | |
419 | * written back improperly if their true value is unknown | |
420 | * to the vfs. | |
421 | */ | |
422 | if (HAS_UNMAPPED_ID(inode)) | |
423 | return -EACCES; | |
1da177e4 LT |
424 | } |
425 | ||
3ddcd056 | 426 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
427 | if (retval) |
428 | return retval; | |
429 | ||
08ce5f16 SH |
430 | retval = devcgroup_inode_permission(inode, mask); |
431 | if (retval) | |
432 | return retval; | |
433 | ||
d09ca739 | 434 | return security_inode_permission(inode, mask); |
1da177e4 | 435 | } |
bd5d0856 | 436 | EXPORT_SYMBOL(__inode_permission); |
1da177e4 | 437 | |
0bdaea90 DH |
438 | /** |
439 | * sb_permission - Check superblock-level permissions | |
440 | * @sb: Superblock of inode to check permission on | |
55852635 | 441 | * @inode: Inode to check permission on |
0bdaea90 DH |
442 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
443 | * | |
444 | * Separate out file-system wide checks from inode-specific permission checks. | |
445 | */ | |
446 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
447 | { | |
448 | if (unlikely(mask & MAY_WRITE)) { | |
449 | umode_t mode = inode->i_mode; | |
450 | ||
451 | /* Nobody gets write access to a read-only fs. */ | |
bc98a42c | 452 | if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
0bdaea90 DH |
453 | return -EROFS; |
454 | } | |
455 | return 0; | |
456 | } | |
457 | ||
458 | /** | |
459 | * inode_permission - Check for access rights to a given inode | |
460 | * @inode: Inode to check permission on | |
461 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
462 | * | |
463 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
464 | * this, letting us set arbitrary permissions for filesystem access without | |
465 | * changing the "normal" UIDs which are used for other things. | |
466 | * | |
467 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
468 | */ | |
469 | int inode_permission(struct inode *inode, int mask) | |
470 | { | |
471 | int retval; | |
472 | ||
473 | retval = sb_permission(inode->i_sb, inode, mask); | |
474 | if (retval) | |
475 | return retval; | |
476 | return __inode_permission(inode, mask); | |
477 | } | |
4d359507 | 478 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 479 | |
5dd784d0 JB |
480 | /** |
481 | * path_get - get a reference to a path | |
482 | * @path: path to get the reference to | |
483 | * | |
484 | * Given a path increment the reference count to the dentry and the vfsmount. | |
485 | */ | |
dcf787f3 | 486 | void path_get(const struct path *path) |
5dd784d0 JB |
487 | { |
488 | mntget(path->mnt); | |
489 | dget(path->dentry); | |
490 | } | |
491 | EXPORT_SYMBOL(path_get); | |
492 | ||
1d957f9b JB |
493 | /** |
494 | * path_put - put a reference to a path | |
495 | * @path: path to put the reference to | |
496 | * | |
497 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
498 | */ | |
dcf787f3 | 499 | void path_put(const struct path *path) |
1da177e4 | 500 | { |
1d957f9b JB |
501 | dput(path->dentry); |
502 | mntput(path->mnt); | |
1da177e4 | 503 | } |
1d957f9b | 504 | EXPORT_SYMBOL(path_put); |
1da177e4 | 505 | |
894bc8c4 | 506 | #define EMBEDDED_LEVELS 2 |
1f55a6ec AV |
507 | struct nameidata { |
508 | struct path path; | |
1cf2665b | 509 | struct qstr last; |
1f55a6ec AV |
510 | struct path root; |
511 | struct inode *inode; /* path.dentry.d_inode */ | |
512 | unsigned int flags; | |
9883d185 | 513 | unsigned seq, m_seq; |
1f55a6ec AV |
514 | int last_type; |
515 | unsigned depth; | |
756daf26 | 516 | int total_link_count; |
697fc6ca AV |
517 | struct saved { |
518 | struct path link; | |
fceef393 | 519 | struct delayed_call done; |
697fc6ca | 520 | const char *name; |
0450b2d1 | 521 | unsigned seq; |
894bc8c4 | 522 | } *stack, internal[EMBEDDED_LEVELS]; |
9883d185 AV |
523 | struct filename *name; |
524 | struct nameidata *saved; | |
fceef393 | 525 | struct inode *link_inode; |
9883d185 AV |
526 | unsigned root_seq; |
527 | int dfd; | |
3859a271 | 528 | } __randomize_layout; |
1f55a6ec | 529 | |
9883d185 | 530 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) |
894bc8c4 | 531 | { |
756daf26 N |
532 | struct nameidata *old = current->nameidata; |
533 | p->stack = p->internal; | |
c8a53ee5 AV |
534 | p->dfd = dfd; |
535 | p->name = name; | |
756daf26 | 536 | p->total_link_count = old ? old->total_link_count : 0; |
9883d185 | 537 | p->saved = old; |
756daf26 | 538 | current->nameidata = p; |
894bc8c4 AV |
539 | } |
540 | ||
9883d185 | 541 | static void restore_nameidata(void) |
894bc8c4 | 542 | { |
9883d185 | 543 | struct nameidata *now = current->nameidata, *old = now->saved; |
756daf26 N |
544 | |
545 | current->nameidata = old; | |
546 | if (old) | |
547 | old->total_link_count = now->total_link_count; | |
e1a63bbc | 548 | if (now->stack != now->internal) |
756daf26 | 549 | kfree(now->stack); |
894bc8c4 AV |
550 | } |
551 | ||
552 | static int __nd_alloc_stack(struct nameidata *nd) | |
553 | { | |
bc40aee0 AV |
554 | struct saved *p; |
555 | ||
556 | if (nd->flags & LOOKUP_RCU) { | |
557 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
558 | GFP_ATOMIC); | |
559 | if (unlikely(!p)) | |
560 | return -ECHILD; | |
561 | } else { | |
562 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
894bc8c4 | 563 | GFP_KERNEL); |
bc40aee0 AV |
564 | if (unlikely(!p)) |
565 | return -ENOMEM; | |
566 | } | |
894bc8c4 AV |
567 | memcpy(p, nd->internal, sizeof(nd->internal)); |
568 | nd->stack = p; | |
569 | return 0; | |
570 | } | |
571 | ||
397d425d EB |
572 | /** |
573 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
574 | * @path: nameidate to verify | |
575 | * | |
576 | * Rename can sometimes move a file or directory outside of a bind | |
577 | * mount, path_connected allows those cases to be detected. | |
578 | */ | |
579 | static bool path_connected(const struct path *path) | |
580 | { | |
581 | struct vfsmount *mnt = path->mnt; | |
4c0ba74e | 582 | struct super_block *sb = mnt->mnt_sb; |
397d425d | 583 | |
4c0ba74e EB |
584 | /* Bind mounts and multi-root filesystems can have disconnected paths */ |
585 | if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) | |
397d425d EB |
586 | return true; |
587 | ||
588 | return is_subdir(path->dentry, mnt->mnt_root); | |
589 | } | |
590 | ||
894bc8c4 AV |
591 | static inline int nd_alloc_stack(struct nameidata *nd) |
592 | { | |
da4e0be0 | 593 | if (likely(nd->depth != EMBEDDED_LEVELS)) |
894bc8c4 AV |
594 | return 0; |
595 | if (likely(nd->stack != nd->internal)) | |
596 | return 0; | |
597 | return __nd_alloc_stack(nd); | |
598 | } | |
599 | ||
7973387a AV |
600 | static void drop_links(struct nameidata *nd) |
601 | { | |
602 | int i = nd->depth; | |
603 | while (i--) { | |
604 | struct saved *last = nd->stack + i; | |
fceef393 AV |
605 | do_delayed_call(&last->done); |
606 | clear_delayed_call(&last->done); | |
7973387a AV |
607 | } |
608 | } | |
609 | ||
610 | static void terminate_walk(struct nameidata *nd) | |
611 | { | |
612 | drop_links(nd); | |
613 | if (!(nd->flags & LOOKUP_RCU)) { | |
614 | int i; | |
615 | path_put(&nd->path); | |
616 | for (i = 0; i < nd->depth; i++) | |
617 | path_put(&nd->stack[i].link); | |
102b8af2 AV |
618 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
619 | path_put(&nd->root); | |
620 | nd->root.mnt = NULL; | |
621 | } | |
7973387a AV |
622 | } else { |
623 | nd->flags &= ~LOOKUP_RCU; | |
624 | if (!(nd->flags & LOOKUP_ROOT)) | |
625 | nd->root.mnt = NULL; | |
626 | rcu_read_unlock(); | |
627 | } | |
628 | nd->depth = 0; | |
629 | } | |
630 | ||
631 | /* path_put is needed afterwards regardless of success or failure */ | |
632 | static bool legitimize_path(struct nameidata *nd, | |
633 | struct path *path, unsigned seq) | |
634 | { | |
635 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
636 | if (unlikely(res)) { | |
637 | if (res > 0) | |
638 | path->mnt = NULL; | |
639 | path->dentry = NULL; | |
640 | return false; | |
641 | } | |
642 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
643 | path->dentry = NULL; | |
644 | return false; | |
645 | } | |
646 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
647 | } | |
648 | ||
649 | static bool legitimize_links(struct nameidata *nd) | |
650 | { | |
651 | int i; | |
652 | for (i = 0; i < nd->depth; i++) { | |
653 | struct saved *last = nd->stack + i; | |
654 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
655 | drop_links(nd); | |
656 | nd->depth = i + 1; | |
657 | return false; | |
658 | } | |
659 | } | |
660 | return true; | |
661 | } | |
662 | ||
19660af7 | 663 | /* |
31e6b01f | 664 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
665 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
666 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
57e3715c | 667 | * normal reference counts on dentries and vfsmounts to transition to ref-walk |
19660af7 AV |
668 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
669 | * got stuck, so ref-walk may continue from there. If this is not successful | |
670 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
671 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 672 | */ |
31e6b01f NP |
673 | |
674 | /** | |
19660af7 AV |
675 | * unlazy_walk - try to switch to ref-walk mode. |
676 | * @nd: nameidata pathwalk data | |
39191628 | 677 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 678 | * |
4675ac39 AV |
679 | * unlazy_walk attempts to legitimize the current nd->path and nd->root |
680 | * for ref-walk mode. | |
681 | * Must be called from rcu-walk context. | |
7973387a AV |
682 | * Nothing should touch nameidata between unlazy_walk() failure and |
683 | * terminate_walk(). | |
31e6b01f | 684 | */ |
4675ac39 | 685 | static int unlazy_walk(struct nameidata *nd) |
31e6b01f | 686 | { |
31e6b01f NP |
687 | struct dentry *parent = nd->path.dentry; |
688 | ||
689 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 | 690 | |
4675ac39 AV |
691 | nd->flags &= ~LOOKUP_RCU; |
692 | if (unlikely(!legitimize_links(nd))) | |
693 | goto out2; | |
694 | if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) | |
695 | goto out1; | |
696 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
697 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) | |
698 | goto out; | |
699 | } | |
700 | rcu_read_unlock(); | |
701 | BUG_ON(nd->inode != parent->d_inode); | |
702 | return 0; | |
703 | ||
704 | out2: | |
705 | nd->path.mnt = NULL; | |
706 | nd->path.dentry = NULL; | |
707 | out1: | |
708 | if (!(nd->flags & LOOKUP_ROOT)) | |
709 | nd->root.mnt = NULL; | |
710 | out: | |
711 | rcu_read_unlock(); | |
712 | return -ECHILD; | |
713 | } | |
714 | ||
715 | /** | |
716 | * unlazy_child - try to switch to ref-walk mode. | |
717 | * @nd: nameidata pathwalk data | |
718 | * @dentry: child of nd->path.dentry | |
719 | * @seq: seq number to check dentry against | |
720 | * Returns: 0 on success, -ECHILD on failure | |
721 | * | |
722 | * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry | |
723 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
724 | * @nd. Must be called from rcu-walk context. | |
725 | * Nothing should touch nameidata between unlazy_child() failure and | |
726 | * terminate_walk(). | |
727 | */ | |
728 | static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
729 | { | |
730 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
731 | ||
e5c832d5 | 732 | nd->flags &= ~LOOKUP_RCU; |
7973387a AV |
733 | if (unlikely(!legitimize_links(nd))) |
734 | goto out2; | |
735 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
736 | goto out2; | |
4675ac39 | 737 | if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) |
7973387a | 738 | goto out1; |
48a066e7 | 739 | |
15570086 | 740 | /* |
4675ac39 AV |
741 | * We need to move both the parent and the dentry from the RCU domain |
742 | * to be properly refcounted. And the sequence number in the dentry | |
743 | * validates *both* dentry counters, since we checked the sequence | |
744 | * number of the parent after we got the child sequence number. So we | |
745 | * know the parent must still be valid if the child sequence number is | |
15570086 | 746 | */ |
4675ac39 AV |
747 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) |
748 | goto out; | |
749 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) { | |
750 | rcu_read_unlock(); | |
751 | dput(dentry); | |
752 | goto drop_root_mnt; | |
19660af7 | 753 | } |
e5c832d5 LT |
754 | /* |
755 | * Sequence counts matched. Now make sure that the root is | |
756 | * still valid and get it if required. | |
757 | */ | |
758 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
5a8d87e8 AV |
759 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { |
760 | rcu_read_unlock(); | |
761 | dput(dentry); | |
762 | return -ECHILD; | |
7973387a | 763 | } |
31e6b01f | 764 | } |
31e6b01f | 765 | |
8b61e74f | 766 | rcu_read_unlock(); |
31e6b01f | 767 | return 0; |
19660af7 | 768 | |
7973387a AV |
769 | out2: |
770 | nd->path.mnt = NULL; | |
771 | out1: | |
772 | nd->path.dentry = NULL; | |
e5c832d5 | 773 | out: |
8b61e74f | 774 | rcu_read_unlock(); |
d0d27277 LT |
775 | drop_root_mnt: |
776 | if (!(nd->flags & LOOKUP_ROOT)) | |
777 | nd->root.mnt = NULL; | |
31e6b01f NP |
778 | return -ECHILD; |
779 | } | |
780 | ||
4ce16ef3 | 781 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 782 | { |
a89f8337 AV |
783 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) |
784 | return dentry->d_op->d_revalidate(dentry, flags); | |
785 | else | |
786 | return 1; | |
34286d66 NP |
787 | } |
788 | ||
9f1fafee AV |
789 | /** |
790 | * complete_walk - successful completion of path walk | |
791 | * @nd: pointer nameidata | |
39159de2 | 792 | * |
9f1fafee AV |
793 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
794 | * Revalidate the final result, unless we'd already done that during | |
795 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
796 | * success, -error on failure. In case of failure caller does not | |
797 | * need to drop nd->path. | |
39159de2 | 798 | */ |
9f1fafee | 799 | static int complete_walk(struct nameidata *nd) |
39159de2 | 800 | { |
16c2cd71 | 801 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 802 | int status; |
39159de2 | 803 | |
9f1fafee | 804 | if (nd->flags & LOOKUP_RCU) { |
9f1fafee AV |
805 | if (!(nd->flags & LOOKUP_ROOT)) |
806 | nd->root.mnt = NULL; | |
4675ac39 | 807 | if (unlikely(unlazy_walk(nd))) |
9f1fafee | 808 | return -ECHILD; |
9f1fafee AV |
809 | } |
810 | ||
16c2cd71 AV |
811 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
812 | return 0; | |
813 | ||
ecf3d1f1 | 814 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
815 | return 0; |
816 | ||
ecf3d1f1 | 817 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
818 | if (status > 0) |
819 | return 0; | |
820 | ||
16c2cd71 | 821 | if (!status) |
39159de2 | 822 | status = -ESTALE; |
16c2cd71 | 823 | |
39159de2 JL |
824 | return status; |
825 | } | |
826 | ||
18d8c860 | 827 | static void set_root(struct nameidata *nd) |
31e6b01f | 828 | { |
7bd88377 | 829 | struct fs_struct *fs = current->fs; |
c28cc364 | 830 | |
9e6697e2 AV |
831 | if (nd->flags & LOOKUP_RCU) { |
832 | unsigned seq; | |
833 | ||
834 | do { | |
835 | seq = read_seqcount_begin(&fs->seq); | |
836 | nd->root = fs->root; | |
837 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
838 | } while (read_seqcount_retry(&fs->seq, seq)); | |
839 | } else { | |
840 | get_fs_root(fs, &nd->root); | |
841 | } | |
31e6b01f NP |
842 | } |
843 | ||
1d957f9b | 844 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
845 | { |
846 | dput(path->dentry); | |
4ac91378 | 847 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
848 | mntput(path->mnt); |
849 | } | |
850 | ||
7b9337aa NP |
851 | static inline void path_to_nameidata(const struct path *path, |
852 | struct nameidata *nd) | |
051d3812 | 853 | { |
31e6b01f NP |
854 | if (!(nd->flags & LOOKUP_RCU)) { |
855 | dput(nd->path.dentry); | |
856 | if (nd->path.mnt != path->mnt) | |
857 | mntput(nd->path.mnt); | |
9a229683 | 858 | } |
31e6b01f | 859 | nd->path.mnt = path->mnt; |
4ac91378 | 860 | nd->path.dentry = path->dentry; |
051d3812 IK |
861 | } |
862 | ||
248fb5b9 AV |
863 | static int nd_jump_root(struct nameidata *nd) |
864 | { | |
865 | if (nd->flags & LOOKUP_RCU) { | |
866 | struct dentry *d; | |
867 | nd->path = nd->root; | |
868 | d = nd->path.dentry; | |
869 | nd->inode = d->d_inode; | |
870 | nd->seq = nd->root_seq; | |
871 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
872 | return -ECHILD; | |
873 | } else { | |
874 | path_put(&nd->path); | |
875 | nd->path = nd->root; | |
876 | path_get(&nd->path); | |
877 | nd->inode = nd->path.dentry->d_inode; | |
878 | } | |
879 | nd->flags |= LOOKUP_JUMPED; | |
880 | return 0; | |
881 | } | |
882 | ||
b5fb63c1 | 883 | /* |
6b255391 | 884 | * Helper to directly jump to a known parsed path from ->get_link, |
b5fb63c1 CH |
885 | * caller must have taken a reference to path beforehand. |
886 | */ | |
6e77137b | 887 | void nd_jump_link(struct path *path) |
b5fb63c1 | 888 | { |
6e77137b | 889 | struct nameidata *nd = current->nameidata; |
b5fb63c1 CH |
890 | path_put(&nd->path); |
891 | ||
892 | nd->path = *path; | |
893 | nd->inode = nd->path.dentry->d_inode; | |
894 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
895 | } |
896 | ||
b9ff4429 | 897 | static inline void put_link(struct nameidata *nd) |
574197e0 | 898 | { |
21c3003d | 899 | struct saved *last = nd->stack + --nd->depth; |
fceef393 | 900 | do_delayed_call(&last->done); |
6548fae2 AV |
901 | if (!(nd->flags & LOOKUP_RCU)) |
902 | path_put(&last->link); | |
574197e0 AV |
903 | } |
904 | ||
8ab447f5 TG |
905 | int sysctl_protected_symlinks __read_mostly = 1; |
906 | int sysctl_protected_hardlinks __read_mostly = 1; | |
800179c9 KC |
907 | |
908 | /** | |
909 | * may_follow_link - Check symlink following for unsafe situations | |
55852635 | 910 | * @nd: nameidata pathwalk data |
800179c9 KC |
911 | * |
912 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
913 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
914 | * in a sticky world-writable directory. This is to protect privileged | |
915 | * processes from failing races against path names that may change out | |
916 | * from under them by way of other users creating malicious symlinks. | |
917 | * It will permit symlinks to be followed only when outside a sticky | |
918 | * world-writable directory, or when the uid of the symlink and follower | |
919 | * match, or when the directory owner matches the symlink's owner. | |
920 | * | |
921 | * Returns 0 if following the symlink is allowed, -ve on error. | |
922 | */ | |
fec2fa24 | 923 | static inline int may_follow_link(struct nameidata *nd) |
800179c9 KC |
924 | { |
925 | const struct inode *inode; | |
926 | const struct inode *parent; | |
2d7f9e2a | 927 | kuid_t puid; |
800179c9 KC |
928 | |
929 | if (!sysctl_protected_symlinks) | |
930 | return 0; | |
931 | ||
932 | /* Allowed if owner and follower match. */ | |
fceef393 | 933 | inode = nd->link_inode; |
81abe27b | 934 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
935 | return 0; |
936 | ||
937 | /* Allowed if parent directory not sticky and world-writable. */ | |
aa65fa35 | 938 | parent = nd->inode; |
800179c9 KC |
939 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) |
940 | return 0; | |
941 | ||
942 | /* Allowed if parent directory and link owner match. */ | |
2d7f9e2a SF |
943 | puid = parent->i_uid; |
944 | if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) | |
800179c9 KC |
945 | return 0; |
946 | ||
31956502 AV |
947 | if (nd->flags & LOOKUP_RCU) |
948 | return -ECHILD; | |
949 | ||
1cf2665b | 950 | audit_log_link_denied("follow_link", &nd->stack[0].link); |
800179c9 KC |
951 | return -EACCES; |
952 | } | |
953 | ||
954 | /** | |
955 | * safe_hardlink_source - Check for safe hardlink conditions | |
956 | * @inode: the source inode to hardlink from | |
957 | * | |
958 | * Return false if at least one of the following conditions: | |
959 | * - inode is not a regular file | |
960 | * - inode is setuid | |
961 | * - inode is setgid and group-exec | |
962 | * - access failure for read and write | |
963 | * | |
964 | * Otherwise returns true. | |
965 | */ | |
966 | static bool safe_hardlink_source(struct inode *inode) | |
967 | { | |
968 | umode_t mode = inode->i_mode; | |
969 | ||
970 | /* Special files should not get pinned to the filesystem. */ | |
971 | if (!S_ISREG(mode)) | |
972 | return false; | |
973 | ||
974 | /* Setuid files should not get pinned to the filesystem. */ | |
975 | if (mode & S_ISUID) | |
976 | return false; | |
977 | ||
978 | /* Executable setgid files should not get pinned to the filesystem. */ | |
979 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
980 | return false; | |
981 | ||
982 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
983 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
984 | return false; | |
985 | ||
986 | return true; | |
987 | } | |
988 | ||
989 | /** | |
990 | * may_linkat - Check permissions for creating a hardlink | |
991 | * @link: the source to hardlink from | |
992 | * | |
993 | * Block hardlink when all of: | |
994 | * - sysctl_protected_hardlinks enabled | |
995 | * - fsuid does not match inode | |
996 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
f2ca3796 | 997 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped |
800179c9 KC |
998 | * |
999 | * Returns 0 if successful, -ve on error. | |
1000 | */ | |
1001 | static int may_linkat(struct path *link) | |
1002 | { | |
800179c9 KC |
1003 | struct inode *inode; |
1004 | ||
1005 | if (!sysctl_protected_hardlinks) | |
1006 | return 0; | |
1007 | ||
800179c9 KC |
1008 | inode = link->dentry->d_inode; |
1009 | ||
1010 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
1011 | * otherwise, it must be a safe source. | |
1012 | */ | |
cc658db4 | 1013 | if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) |
800179c9 KC |
1014 | return 0; |
1015 | ||
a51d9eaa | 1016 | audit_log_link_denied("linkat", link); |
800179c9 KC |
1017 | return -EPERM; |
1018 | } | |
1019 | ||
3b2e7f75 AV |
1020 | static __always_inline |
1021 | const char *get_link(struct nameidata *nd) | |
1da177e4 | 1022 | { |
ab104923 | 1023 | struct saved *last = nd->stack + nd->depth - 1; |
1cf2665b | 1024 | struct dentry *dentry = last->link.dentry; |
fceef393 | 1025 | struct inode *inode = nd->link_inode; |
6d7b5aae | 1026 | int error; |
0a959df5 | 1027 | const char *res; |
1da177e4 | 1028 | |
8fa9dd24 N |
1029 | if (!(nd->flags & LOOKUP_RCU)) { |
1030 | touch_atime(&last->link); | |
1031 | cond_resched(); | |
598e3c8f | 1032 | } else if (atime_needs_update_rcu(&last->link, inode)) { |
4675ac39 | 1033 | if (unlikely(unlazy_walk(nd))) |
bc40aee0 | 1034 | return ERR_PTR(-ECHILD); |
8fa9dd24 | 1035 | touch_atime(&last->link); |
bc40aee0 | 1036 | } |
574197e0 | 1037 | |
bda0be7a N |
1038 | error = security_inode_follow_link(dentry, inode, |
1039 | nd->flags & LOOKUP_RCU); | |
1040 | if (unlikely(error)) | |
6920a440 | 1041 | return ERR_PTR(error); |
36f3b4f6 | 1042 | |
86acdca1 | 1043 | nd->last_type = LAST_BIND; |
d4dee48b AV |
1044 | res = inode->i_link; |
1045 | if (!res) { | |
fceef393 AV |
1046 | const char * (*get)(struct dentry *, struct inode *, |
1047 | struct delayed_call *); | |
1048 | get = inode->i_op->get_link; | |
8c1b4566 | 1049 | if (nd->flags & LOOKUP_RCU) { |
fceef393 | 1050 | res = get(NULL, inode, &last->done); |
6b255391 | 1051 | if (res == ERR_PTR(-ECHILD)) { |
4675ac39 | 1052 | if (unlikely(unlazy_walk(nd))) |
6b255391 | 1053 | return ERR_PTR(-ECHILD); |
fceef393 | 1054 | res = get(dentry, inode, &last->done); |
6b255391 AV |
1055 | } |
1056 | } else { | |
fceef393 | 1057 | res = get(dentry, inode, &last->done); |
8c1b4566 | 1058 | } |
fceef393 | 1059 | if (IS_ERR_OR_NULL(res)) |
fab51e8a | 1060 | return res; |
fab51e8a AV |
1061 | } |
1062 | if (*res == '/') { | |
9e6697e2 AV |
1063 | if (!nd->root.mnt) |
1064 | set_root(nd); | |
248fb5b9 AV |
1065 | if (unlikely(nd_jump_root(nd))) |
1066 | return ERR_PTR(-ECHILD); | |
fab51e8a AV |
1067 | while (unlikely(*++res == '/')) |
1068 | ; | |
1da177e4 | 1069 | } |
fab51e8a AV |
1070 | if (!*res) |
1071 | res = NULL; | |
0a959df5 AV |
1072 | return res; |
1073 | } | |
6d7b5aae | 1074 | |
f015f126 DH |
1075 | /* |
1076 | * follow_up - Find the mountpoint of path's vfsmount | |
1077 | * | |
1078 | * Given a path, find the mountpoint of its source file system. | |
1079 | * Replace @path with the path of the mountpoint in the parent mount. | |
1080 | * Up is towards /. | |
1081 | * | |
1082 | * Return 1 if we went up a level and 0 if we were already at the | |
1083 | * root. | |
1084 | */ | |
bab77ebf | 1085 | int follow_up(struct path *path) |
1da177e4 | 1086 | { |
0714a533 AV |
1087 | struct mount *mnt = real_mount(path->mnt); |
1088 | struct mount *parent; | |
1da177e4 | 1089 | struct dentry *mountpoint; |
99b7db7b | 1090 | |
48a066e7 | 1091 | read_seqlock_excl(&mount_lock); |
0714a533 | 1092 | parent = mnt->mnt_parent; |
3c0a6163 | 1093 | if (parent == mnt) { |
48a066e7 | 1094 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
1095 | return 0; |
1096 | } | |
0714a533 | 1097 | mntget(&parent->mnt); |
a73324da | 1098 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 1099 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
1100 | dput(path->dentry); |
1101 | path->dentry = mountpoint; | |
1102 | mntput(path->mnt); | |
0714a533 | 1103 | path->mnt = &parent->mnt; |
1da177e4 LT |
1104 | return 1; |
1105 | } | |
4d359507 | 1106 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 1107 | |
b5c84bf6 | 1108 | /* |
9875cf80 DH |
1109 | * Perform an automount |
1110 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1111 | * were called with. | |
1da177e4 | 1112 | */ |
756daf26 | 1113 | static int follow_automount(struct path *path, struct nameidata *nd, |
9875cf80 | 1114 | bool *need_mntput) |
31e6b01f | 1115 | { |
9875cf80 | 1116 | struct vfsmount *mnt; |
ea5b778a | 1117 | int err; |
9875cf80 DH |
1118 | |
1119 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1120 | return -EREMOTE; | |
1121 | ||
0ec26fd0 MS |
1122 | /* We don't want to mount if someone's just doing a stat - |
1123 | * unless they're stat'ing a directory and appended a '/' to | |
1124 | * the name. | |
1125 | * | |
1126 | * We do, however, want to mount if someone wants to open or | |
1127 | * create a file of any type under the mountpoint, wants to | |
1128 | * traverse through the mountpoint or wants to open the | |
1129 | * mounted directory. Also, autofs may mark negative dentries | |
1130 | * as being automount points. These will need the attentions | |
1131 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 1132 | */ |
756daf26 | 1133 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
5d38f049 IK |
1134 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
1135 | path->dentry->d_inode) | |
1136 | return -EISDIR; | |
0ec26fd0 | 1137 | |
756daf26 N |
1138 | nd->total_link_count++; |
1139 | if (nd->total_link_count >= 40) | |
9875cf80 DH |
1140 | return -ELOOP; |
1141 | ||
1142 | mnt = path->dentry->d_op->d_automount(path); | |
1143 | if (IS_ERR(mnt)) { | |
1144 | /* | |
1145 | * The filesystem is allowed to return -EISDIR here to indicate | |
1146 | * it doesn't want to automount. For instance, autofs would do | |
1147 | * this so that its userspace daemon can mount on this dentry. | |
1148 | * | |
1149 | * However, we can only permit this if it's a terminal point in | |
1150 | * the path being looked up; if it wasn't then the remainder of | |
1151 | * the path is inaccessible and we should say so. | |
1152 | */ | |
756daf26 | 1153 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) |
9875cf80 DH |
1154 | return -EREMOTE; |
1155 | return PTR_ERR(mnt); | |
31e6b01f | 1156 | } |
ea5b778a | 1157 | |
9875cf80 DH |
1158 | if (!mnt) /* mount collision */ |
1159 | return 0; | |
31e6b01f | 1160 | |
8aef1884 AV |
1161 | if (!*need_mntput) { |
1162 | /* lock_mount() may release path->mnt on error */ | |
1163 | mntget(path->mnt); | |
1164 | *need_mntput = true; | |
1165 | } | |
19a167af | 1166 | err = finish_automount(mnt, path); |
9875cf80 | 1167 | |
ea5b778a DH |
1168 | switch (err) { |
1169 | case -EBUSY: | |
1170 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1171 | return 0; |
ea5b778a | 1172 | case 0: |
8aef1884 | 1173 | path_put(path); |
ea5b778a DH |
1174 | path->mnt = mnt; |
1175 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1176 | return 0; |
19a167af AV |
1177 | default: |
1178 | return err; | |
ea5b778a | 1179 | } |
19a167af | 1180 | |
463ffb2e AV |
1181 | } |
1182 | ||
9875cf80 DH |
1183 | /* |
1184 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1185 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1186 | * - Flagged as mountpoint |
1187 | * - Flagged as automount point | |
1188 | * | |
1189 | * This may only be called in refwalk mode. | |
1190 | * | |
1191 | * Serialization is taken care of in namespace.c | |
1192 | */ | |
756daf26 | 1193 | static int follow_managed(struct path *path, struct nameidata *nd) |
1da177e4 | 1194 | { |
8aef1884 | 1195 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1196 | unsigned managed; |
1197 | bool need_mntput = false; | |
8aef1884 | 1198 | int ret = 0; |
9875cf80 DH |
1199 | |
1200 | /* Given that we're not holding a lock here, we retain the value in a | |
1201 | * local variable for each dentry as we look at it so that we don't see | |
1202 | * the components of that value change under us */ | |
6aa7de05 | 1203 | while (managed = READ_ONCE(path->dentry->d_flags), |
9875cf80 DH |
1204 | managed &= DCACHE_MANAGED_DENTRY, |
1205 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1206 | /* Allow the filesystem to manage the transit without i_mutex |
1207 | * being held. */ | |
1208 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1209 | BUG_ON(!path->dentry->d_op); | |
1210 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1211 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 | 1212 | if (ret < 0) |
8aef1884 | 1213 | break; |
cc53ce53 DH |
1214 | } |
1215 | ||
9875cf80 DH |
1216 | /* Transit to a mounted filesystem. */ |
1217 | if (managed & DCACHE_MOUNTED) { | |
1218 | struct vfsmount *mounted = lookup_mnt(path); | |
1219 | if (mounted) { | |
1220 | dput(path->dentry); | |
1221 | if (need_mntput) | |
1222 | mntput(path->mnt); | |
1223 | path->mnt = mounted; | |
1224 | path->dentry = dget(mounted->mnt_root); | |
1225 | need_mntput = true; | |
1226 | continue; | |
1227 | } | |
1228 | ||
1229 | /* Something is mounted on this dentry in another | |
1230 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1231 | * namespace got unmounted before lookup_mnt() could |
1232 | * get it */ | |
9875cf80 DH |
1233 | } |
1234 | ||
1235 | /* Handle an automount point */ | |
1236 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
756daf26 | 1237 | ret = follow_automount(path, nd, &need_mntput); |
9875cf80 | 1238 | if (ret < 0) |
8aef1884 | 1239 | break; |
9875cf80 DH |
1240 | continue; |
1241 | } | |
1242 | ||
1243 | /* We didn't change the current path point */ | |
1244 | break; | |
1da177e4 | 1245 | } |
8aef1884 AV |
1246 | |
1247 | if (need_mntput && path->mnt == mnt) | |
1248 | mntput(path->mnt); | |
e9742b53 AV |
1249 | if (ret == -EISDIR || !ret) |
1250 | ret = 1; | |
8402752e AV |
1251 | if (need_mntput) |
1252 | nd->flags |= LOOKUP_JUMPED; | |
1253 | if (unlikely(ret < 0)) | |
1254 | path_put_conditional(path, nd); | |
1255 | return ret; | |
1da177e4 LT |
1256 | } |
1257 | ||
cc53ce53 | 1258 | int follow_down_one(struct path *path) |
1da177e4 LT |
1259 | { |
1260 | struct vfsmount *mounted; | |
1261 | ||
1c755af4 | 1262 | mounted = lookup_mnt(path); |
1da177e4 | 1263 | if (mounted) { |
9393bd07 AV |
1264 | dput(path->dentry); |
1265 | mntput(path->mnt); | |
1266 | path->mnt = mounted; | |
1267 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1268 | return 1; |
1269 | } | |
1270 | return 0; | |
1271 | } | |
4d359507 | 1272 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1273 | |
fb5f51c7 | 1274 | static inline int managed_dentry_rcu(const struct path *path) |
62a7375e | 1275 | { |
fb5f51c7 IK |
1276 | return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1277 | path->dentry->d_op->d_manage(path, true) : 0; | |
62a7375e IK |
1278 | } |
1279 | ||
9875cf80 | 1280 | /* |
287548e4 AV |
1281 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1282 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1283 | */ |
1284 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
254cf582 | 1285 | struct inode **inode, unsigned *seqp) |
9875cf80 | 1286 | { |
62a7375e | 1287 | for (;;) { |
c7105365 | 1288 | struct mount *mounted; |
62a7375e IK |
1289 | /* |
1290 | * Don't forget we might have a non-mountpoint managed dentry | |
1291 | * that wants to block transit. | |
1292 | */ | |
fb5f51c7 | 1293 | switch (managed_dentry_rcu(path)) { |
b8faf035 N |
1294 | case -ECHILD: |
1295 | default: | |
ab90911f | 1296 | return false; |
b8faf035 N |
1297 | case -EISDIR: |
1298 | return true; | |
1299 | case 0: | |
1300 | break; | |
1301 | } | |
62a7375e IK |
1302 | |
1303 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1304 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1305 | |
474279dc | 1306 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1307 | if (!mounted) |
1308 | break; | |
c7105365 AV |
1309 | path->mnt = &mounted->mnt; |
1310 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1311 | nd->flags |= LOOKUP_JUMPED; |
254cf582 | 1312 | *seqp = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1313 | /* |
1314 | * Update the inode too. We don't need to re-check the | |
1315 | * dentry sequence number here after this d_inode read, | |
1316 | * because a mount-point is always pinned. | |
1317 | */ | |
1318 | *inode = path->dentry->d_inode; | |
9875cf80 | 1319 | } |
f5be3e29 | 1320 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1321 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1322 | } |
1323 | ||
31e6b01f NP |
1324 | static int follow_dotdot_rcu(struct nameidata *nd) |
1325 | { | |
4023bfc9 | 1326 | struct inode *inode = nd->inode; |
31e6b01f | 1327 | |
9875cf80 | 1328 | while (1) { |
aed434ad | 1329 | if (path_equal(&nd->path, &nd->root)) |
31e6b01f | 1330 | break; |
31e6b01f NP |
1331 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1332 | struct dentry *old = nd->path.dentry; | |
1333 | struct dentry *parent = old->d_parent; | |
1334 | unsigned seq; | |
1335 | ||
4023bfc9 | 1336 | inode = parent->d_inode; |
31e6b01f | 1337 | seq = read_seqcount_begin(&parent->d_seq); |
aed434ad AV |
1338 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) |
1339 | return -ECHILD; | |
31e6b01f NP |
1340 | nd->path.dentry = parent; |
1341 | nd->seq = seq; | |
397d425d EB |
1342 | if (unlikely(!path_connected(&nd->path))) |
1343 | return -ENOENT; | |
31e6b01f | 1344 | break; |
aed434ad AV |
1345 | } else { |
1346 | struct mount *mnt = real_mount(nd->path.mnt); | |
1347 | struct mount *mparent = mnt->mnt_parent; | |
1348 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1349 | struct inode *inode2 = mountpoint->d_inode; | |
1350 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1351 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1352 | return -ECHILD; | |
1353 | if (&mparent->mnt == nd->path.mnt) | |
1354 | break; | |
1355 | /* we know that mountpoint was pinned */ | |
1356 | nd->path.dentry = mountpoint; | |
1357 | nd->path.mnt = &mparent->mnt; | |
1358 | inode = inode2; | |
1359 | nd->seq = seq; | |
31e6b01f | 1360 | } |
31e6b01f | 1361 | } |
aed434ad | 1362 | while (unlikely(d_mountpoint(nd->path.dentry))) { |
b37199e6 AV |
1363 | struct mount *mounted; |
1364 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
aed434ad AV |
1365 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) |
1366 | return -ECHILD; | |
b37199e6 AV |
1367 | if (!mounted) |
1368 | break; | |
1369 | nd->path.mnt = &mounted->mnt; | |
1370 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1371 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1372 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
b37199e6 | 1373 | } |
4023bfc9 | 1374 | nd->inode = inode; |
31e6b01f NP |
1375 | return 0; |
1376 | } | |
1377 | ||
cc53ce53 DH |
1378 | /* |
1379 | * Follow down to the covering mount currently visible to userspace. At each | |
1380 | * point, the filesystem owning that dentry may be queried as to whether the | |
1381 | * caller is permitted to proceed or not. | |
cc53ce53 | 1382 | */ |
7cc90cc3 | 1383 | int follow_down(struct path *path) |
cc53ce53 DH |
1384 | { |
1385 | unsigned managed; | |
1386 | int ret; | |
1387 | ||
6aa7de05 | 1388 | while (managed = READ_ONCE(path->dentry->d_flags), |
cc53ce53 DH |
1389 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { |
1390 | /* Allow the filesystem to manage the transit without i_mutex | |
1391 | * being held. | |
1392 | * | |
1393 | * We indicate to the filesystem if someone is trying to mount | |
1394 | * something here. This gives autofs the chance to deny anyone | |
1395 | * other than its daemon the right to mount on its | |
1396 | * superstructure. | |
1397 | * | |
1398 | * The filesystem may sleep at this point. | |
1399 | */ | |
1400 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1401 | BUG_ON(!path->dentry->d_op); | |
1402 | BUG_ON(!path->dentry->d_op->d_manage); | |
fb5f51c7 | 1403 | ret = path->dentry->d_op->d_manage(path, false); |
cc53ce53 DH |
1404 | if (ret < 0) |
1405 | return ret == -EISDIR ? 0 : ret; | |
1406 | } | |
1407 | ||
1408 | /* Transit to a mounted filesystem. */ | |
1409 | if (managed & DCACHE_MOUNTED) { | |
1410 | struct vfsmount *mounted = lookup_mnt(path); | |
1411 | if (!mounted) | |
1412 | break; | |
1413 | dput(path->dentry); | |
1414 | mntput(path->mnt); | |
1415 | path->mnt = mounted; | |
1416 | path->dentry = dget(mounted->mnt_root); | |
1417 | continue; | |
1418 | } | |
1419 | ||
1420 | /* Don't handle automount points here */ | |
1421 | break; | |
1422 | } | |
1423 | return 0; | |
1424 | } | |
4d359507 | 1425 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1426 | |
9875cf80 DH |
1427 | /* |
1428 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1429 | */ | |
1430 | static void follow_mount(struct path *path) | |
1431 | { | |
1432 | while (d_mountpoint(path->dentry)) { | |
1433 | struct vfsmount *mounted = lookup_mnt(path); | |
1434 | if (!mounted) | |
1435 | break; | |
1436 | dput(path->dentry); | |
1437 | mntput(path->mnt); | |
1438 | path->mnt = mounted; | |
1439 | path->dentry = dget(mounted->mnt_root); | |
1440 | } | |
1441 | } | |
1442 | ||
eedf265a EB |
1443 | static int path_parent_directory(struct path *path) |
1444 | { | |
1445 | struct dentry *old = path->dentry; | |
1446 | /* rare case of legitimate dget_parent()... */ | |
1447 | path->dentry = dget_parent(path->dentry); | |
1448 | dput(old); | |
1449 | if (unlikely(!path_connected(path))) | |
1450 | return -ENOENT; | |
1451 | return 0; | |
1452 | } | |
1453 | ||
397d425d | 1454 | static int follow_dotdot(struct nameidata *nd) |
1da177e4 LT |
1455 | { |
1456 | while(1) { | |
2a737871 AV |
1457 | if (nd->path.dentry == nd->root.dentry && |
1458 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1459 | break; |
1460 | } | |
4ac91378 | 1461 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
eedf265a EB |
1462 | int ret = path_parent_directory(&nd->path); |
1463 | if (ret) | |
1464 | return ret; | |
1da177e4 LT |
1465 | break; |
1466 | } | |
3088dd70 | 1467 | if (!follow_up(&nd->path)) |
1da177e4 | 1468 | break; |
1da177e4 | 1469 | } |
79ed0226 | 1470 | follow_mount(&nd->path); |
31e6b01f | 1471 | nd->inode = nd->path.dentry->d_inode; |
397d425d | 1472 | return 0; |
1da177e4 LT |
1473 | } |
1474 | ||
baa03890 | 1475 | /* |
f4fdace9 OD |
1476 | * This looks up the name in dcache and possibly revalidates the found dentry. |
1477 | * NULL is returned if the dentry does not exist in the cache. | |
baa03890 | 1478 | */ |
e3c13928 AV |
1479 | static struct dentry *lookup_dcache(const struct qstr *name, |
1480 | struct dentry *dir, | |
6c51e513 | 1481 | unsigned int flags) |
baa03890 | 1482 | { |
a89f8337 | 1483 | struct dentry *dentry = d_lookup(dir, name); |
bad61189 | 1484 | if (dentry) { |
a89f8337 AV |
1485 | int error = d_revalidate(dentry, flags); |
1486 | if (unlikely(error <= 0)) { | |
1487 | if (!error) | |
1488 | d_invalidate(dentry); | |
1489 | dput(dentry); | |
1490 | return ERR_PTR(error); | |
bad61189 MS |
1491 | } |
1492 | } | |
baa03890 NP |
1493 | return dentry; |
1494 | } | |
1495 | ||
44396f4b | 1496 | /* |
13a2c3be BF |
1497 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1498 | * unhashed. | |
bad61189 MS |
1499 | * |
1500 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1501 | */ |
bad61189 | 1502 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1503 | unsigned int flags) |
44396f4b | 1504 | { |
44396f4b JB |
1505 | struct dentry *old; |
1506 | ||
1507 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1508 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1509 | dput(dentry); |
44396f4b | 1510 | return ERR_PTR(-ENOENT); |
e188dc02 | 1511 | } |
44396f4b | 1512 | |
72bd866a | 1513 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1514 | if (unlikely(old)) { |
1515 | dput(dentry); | |
1516 | dentry = old; | |
1517 | } | |
1518 | return dentry; | |
1519 | } | |
1520 | ||
e3c13928 | 1521 | static struct dentry *__lookup_hash(const struct qstr *name, |
72bd866a | 1522 | struct dentry *base, unsigned int flags) |
a3255546 | 1523 | { |
6c51e513 | 1524 | struct dentry *dentry = lookup_dcache(name, base, flags); |
a3255546 | 1525 | |
6c51e513 | 1526 | if (dentry) |
bad61189 | 1527 | return dentry; |
a3255546 | 1528 | |
6c51e513 AV |
1529 | dentry = d_alloc(base, name); |
1530 | if (unlikely(!dentry)) | |
1531 | return ERR_PTR(-ENOMEM); | |
1532 | ||
72bd866a | 1533 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1534 | } |
1535 | ||
e97cdc87 | 1536 | static int lookup_fast(struct nameidata *nd, |
254cf582 AV |
1537 | struct path *path, struct inode **inode, |
1538 | unsigned *seqp) | |
1da177e4 | 1539 | { |
4ac91378 | 1540 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1541 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 | 1542 | int status = 1; |
9875cf80 DH |
1543 | int err; |
1544 | ||
b04f784e NP |
1545 | /* |
1546 | * Rename seqlock is not required here because in the off chance | |
5d0f49c1 AV |
1547 | * of a false negative due to a concurrent rename, the caller is |
1548 | * going to fall back to non-racy lookup. | |
b04f784e | 1549 | */ |
31e6b01f NP |
1550 | if (nd->flags & LOOKUP_RCU) { |
1551 | unsigned seq; | |
766c4cbf | 1552 | bool negative; |
da53be12 | 1553 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5d0f49c1 | 1554 | if (unlikely(!dentry)) { |
4675ac39 | 1555 | if (unlazy_walk(nd)) |
5d0f49c1 | 1556 | return -ECHILD; |
e9742b53 | 1557 | return 0; |
5d0f49c1 | 1558 | } |
5a18fff2 | 1559 | |
12f8ad4b LT |
1560 | /* |
1561 | * This sequence count validates that the inode matches | |
1562 | * the dentry name information from lookup. | |
1563 | */ | |
63afdfc7 | 1564 | *inode = d_backing_inode(dentry); |
766c4cbf | 1565 | negative = d_is_negative(dentry); |
5d0f49c1 | 1566 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) |
12f8ad4b LT |
1567 | return -ECHILD; |
1568 | ||
1569 | /* | |
1570 | * This sequence count validates that the parent had no | |
1571 | * changes while we did the lookup of the dentry above. | |
1572 | * | |
1573 | * The memory barrier in read_seqcount_begin of child is | |
1574 | * enough, we can use __read_seqcount_retry here. | |
1575 | */ | |
5d0f49c1 | 1576 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) |
31e6b01f | 1577 | return -ECHILD; |
5a18fff2 | 1578 | |
254cf582 | 1579 | *seqp = seq; |
a89f8337 | 1580 | status = d_revalidate(dentry, nd->flags); |
209a7fb2 | 1581 | if (likely(status > 0)) { |
5d0f49c1 AV |
1582 | /* |
1583 | * Note: do negative dentry check after revalidation in | |
1584 | * case that drops it. | |
1585 | */ | |
1586 | if (unlikely(negative)) | |
1587 | return -ENOENT; | |
1588 | path->mnt = mnt; | |
1589 | path->dentry = dentry; | |
1590 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
e9742b53 | 1591 | return 1; |
24643087 | 1592 | } |
4675ac39 | 1593 | if (unlazy_child(nd, dentry, seq)) |
209a7fb2 AV |
1594 | return -ECHILD; |
1595 | if (unlikely(status == -ECHILD)) | |
1596 | /* we'd been told to redo it in non-rcu mode */ | |
1597 | status = d_revalidate(dentry, nd->flags); | |
5a18fff2 | 1598 | } else { |
e97cdc87 | 1599 | dentry = __d_lookup(parent, &nd->last); |
5d0f49c1 | 1600 | if (unlikely(!dentry)) |
e9742b53 | 1601 | return 0; |
a89f8337 | 1602 | status = d_revalidate(dentry, nd->flags); |
9875cf80 | 1603 | } |
5a18fff2 | 1604 | if (unlikely(status <= 0)) { |
e9742b53 | 1605 | if (!status) |
5d0f49c1 | 1606 | d_invalidate(dentry); |
5542aa2f | 1607 | dput(dentry); |
5d0f49c1 | 1608 | return status; |
24643087 | 1609 | } |
766c4cbf AV |
1610 | if (unlikely(d_is_negative(dentry))) { |
1611 | dput(dentry); | |
1612 | return -ENOENT; | |
1613 | } | |
5d0f49c1 | 1614 | |
9875cf80 DH |
1615 | path->mnt = mnt; |
1616 | path->dentry = dentry; | |
756daf26 | 1617 | err = follow_managed(path, nd); |
e9742b53 | 1618 | if (likely(err > 0)) |
63afdfc7 | 1619 | *inode = d_backing_inode(path->dentry); |
8402752e | 1620 | return err; |
697f514d MS |
1621 | } |
1622 | ||
1623 | /* Fast lookup failed, do it the slow way */ | |
e3c13928 AV |
1624 | static struct dentry *lookup_slow(const struct qstr *name, |
1625 | struct dentry *dir, | |
1626 | unsigned int flags) | |
697f514d | 1627 | { |
94bdd655 | 1628 | struct dentry *dentry = ERR_PTR(-ENOENT), *old; |
1936386e | 1629 | struct inode *inode = dir->d_inode; |
d9171b93 | 1630 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
1936386e | 1631 | |
9902af79 | 1632 | inode_lock_shared(inode); |
1936386e | 1633 | /* Don't go there if it's already dead */ |
94bdd655 AV |
1634 | if (unlikely(IS_DEADDIR(inode))) |
1635 | goto out; | |
1636 | again: | |
d9171b93 | 1637 | dentry = d_alloc_parallel(dir, name, &wq); |
94bdd655 AV |
1638 | if (IS_ERR(dentry)) |
1639 | goto out; | |
1640 | if (unlikely(!d_in_lookup(dentry))) { | |
a89f8337 | 1641 | if (!(flags & LOOKUP_NO_REVAL)) { |
949a852e AV |
1642 | int error = d_revalidate(dentry, flags); |
1643 | if (unlikely(error <= 0)) { | |
94bdd655 | 1644 | if (!error) { |
949a852e | 1645 | d_invalidate(dentry); |
94bdd655 AV |
1646 | dput(dentry); |
1647 | goto again; | |
1648 | } | |
949a852e AV |
1649 | dput(dentry); |
1650 | dentry = ERR_PTR(error); | |
1651 | } | |
1652 | } | |
94bdd655 AV |
1653 | } else { |
1654 | old = inode->i_op->lookup(inode, dentry, flags); | |
1655 | d_lookup_done(dentry); | |
1656 | if (unlikely(old)) { | |
1657 | dput(dentry); | |
1658 | dentry = old; | |
949a852e AV |
1659 | } |
1660 | } | |
94bdd655 | 1661 | out: |
9902af79 | 1662 | inode_unlock_shared(inode); |
e3c13928 | 1663 | return dentry; |
1da177e4 LT |
1664 | } |
1665 | ||
52094c8a AV |
1666 | static inline int may_lookup(struct nameidata *nd) |
1667 | { | |
1668 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1669 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1670 | if (err != -ECHILD) |
1671 | return err; | |
4675ac39 | 1672 | if (unlazy_walk(nd)) |
52094c8a AV |
1673 | return -ECHILD; |
1674 | } | |
4ad5abb3 | 1675 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1676 | } |
1677 | ||
9856fa1b AV |
1678 | static inline int handle_dots(struct nameidata *nd, int type) |
1679 | { | |
1680 | if (type == LAST_DOTDOT) { | |
9e6697e2 AV |
1681 | if (!nd->root.mnt) |
1682 | set_root(nd); | |
9856fa1b | 1683 | if (nd->flags & LOOKUP_RCU) { |
70291aec | 1684 | return follow_dotdot_rcu(nd); |
9856fa1b | 1685 | } else |
397d425d | 1686 | return follow_dotdot(nd); |
9856fa1b AV |
1687 | } |
1688 | return 0; | |
1689 | } | |
1690 | ||
181548c0 AV |
1691 | static int pick_link(struct nameidata *nd, struct path *link, |
1692 | struct inode *inode, unsigned seq) | |
d63ff28f | 1693 | { |
626de996 | 1694 | int error; |
1cf2665b | 1695 | struct saved *last; |
756daf26 | 1696 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { |
626de996 AV |
1697 | path_to_nameidata(link, nd); |
1698 | return -ELOOP; | |
1699 | } | |
bc40aee0 | 1700 | if (!(nd->flags & LOOKUP_RCU)) { |
7973387a AV |
1701 | if (link->mnt == nd->path.mnt) |
1702 | mntget(link->mnt); | |
d63ff28f | 1703 | } |
626de996 AV |
1704 | error = nd_alloc_stack(nd); |
1705 | if (unlikely(error)) { | |
bc40aee0 | 1706 | if (error == -ECHILD) { |
ad1633a1 AV |
1707 | if (unlikely(!legitimize_path(nd, link, seq))) { |
1708 | drop_links(nd); | |
1709 | nd->depth = 0; | |
1710 | nd->flags &= ~LOOKUP_RCU; | |
1711 | nd->path.mnt = NULL; | |
1712 | nd->path.dentry = NULL; | |
1713 | if (!(nd->flags & LOOKUP_ROOT)) | |
1714 | nd->root.mnt = NULL; | |
1715 | rcu_read_unlock(); | |
4675ac39 | 1716 | } else if (likely(unlazy_walk(nd)) == 0) |
ad1633a1 | 1717 | error = nd_alloc_stack(nd); |
bc40aee0 AV |
1718 | } |
1719 | if (error) { | |
1720 | path_put(link); | |
1721 | return error; | |
1722 | } | |
626de996 AV |
1723 | } |
1724 | ||
ab104923 | 1725 | last = nd->stack + nd->depth++; |
1cf2665b | 1726 | last->link = *link; |
fceef393 AV |
1727 | clear_delayed_call(&last->done); |
1728 | nd->link_inode = inode; | |
0450b2d1 | 1729 | last->seq = seq; |
d63ff28f AV |
1730 | return 1; |
1731 | } | |
1732 | ||
8f64fb1c | 1733 | enum {WALK_FOLLOW = 1, WALK_MORE = 2}; |
31d66bcd | 1734 | |
3ddcd056 LT |
1735 | /* |
1736 | * Do we need to follow links? We _really_ want to be able | |
1737 | * to do this check without having to look at inode->i_op, | |
1738 | * so we keep a cache of "no, this doesn't need follow_link" | |
1739 | * for the common case. | |
1740 | */ | |
8f64fb1c AV |
1741 | static inline int step_into(struct nameidata *nd, struct path *path, |
1742 | int flags, struct inode *inode, unsigned seq) | |
3ddcd056 | 1743 | { |
31d66bcd AV |
1744 | if (!(flags & WALK_MORE) && nd->depth) |
1745 | put_link(nd); | |
8f64fb1c AV |
1746 | if (likely(!d_is_symlink(path->dentry)) || |
1747 | !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { | |
1748 | /* not a symlink or should not follow */ | |
1749 | path_to_nameidata(path, nd); | |
1750 | nd->inode = inode; | |
1751 | nd->seq = seq; | |
d63ff28f | 1752 | return 0; |
8f64fb1c | 1753 | } |
a7f77542 AV |
1754 | /* make sure that d_is_symlink above matches inode */ |
1755 | if (nd->flags & LOOKUP_RCU) { | |
8f64fb1c | 1756 | if (read_seqcount_retry(&path->dentry->d_seq, seq)) |
a7f77542 AV |
1757 | return -ECHILD; |
1758 | } | |
8f64fb1c | 1759 | return pick_link(nd, path, inode, seq); |
3ddcd056 LT |
1760 | } |
1761 | ||
4693a547 | 1762 | static int walk_component(struct nameidata *nd, int flags) |
ce57dfc1 | 1763 | { |
caa85634 | 1764 | struct path path; |
ce57dfc1 | 1765 | struct inode *inode; |
254cf582 | 1766 | unsigned seq; |
ce57dfc1 AV |
1767 | int err; |
1768 | /* | |
1769 | * "." and ".." are special - ".." especially so because it has | |
1770 | * to be able to know about the current root directory and | |
1771 | * parent relationships. | |
1772 | */ | |
4693a547 AV |
1773 | if (unlikely(nd->last_type != LAST_NORM)) { |
1774 | err = handle_dots(nd, nd->last_type); | |
1c4ff1a8 | 1775 | if (!(flags & WALK_MORE) && nd->depth) |
4693a547 AV |
1776 | put_link(nd); |
1777 | return err; | |
1778 | } | |
254cf582 | 1779 | err = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 1780 | if (unlikely(err <= 0)) { |
697f514d | 1781 | if (err < 0) |
f0a9ba70 | 1782 | return err; |
e3c13928 AV |
1783 | path.dentry = lookup_slow(&nd->last, nd->path.dentry, |
1784 | nd->flags); | |
1785 | if (IS_ERR(path.dentry)) | |
1786 | return PTR_ERR(path.dentry); | |
7500c38a | 1787 | |
e3c13928 AV |
1788 | path.mnt = nd->path.mnt; |
1789 | err = follow_managed(&path, nd); | |
1790 | if (unlikely(err < 0)) | |
f0a9ba70 | 1791 | return err; |
697f514d | 1792 | |
7500c38a AV |
1793 | if (unlikely(d_is_negative(path.dentry))) { |
1794 | path_to_nameidata(&path, nd); | |
1795 | return -ENOENT; | |
1796 | } | |
1797 | ||
254cf582 | 1798 | seq = 0; /* we are already out of RCU mode */ |
d4565649 | 1799 | inode = d_backing_inode(path.dentry); |
ce57dfc1 | 1800 | } |
697f514d | 1801 | |
8f64fb1c | 1802 | return step_into(nd, &path, flags, inode, seq); |
ce57dfc1 AV |
1803 | } |
1804 | ||
bfcfaa77 LT |
1805 | /* |
1806 | * We can do the critical dentry name comparison and hashing | |
1807 | * operations one word at a time, but we are limited to: | |
1808 | * | |
1809 | * - Architectures with fast unaligned word accesses. We could | |
1810 | * do a "get_unaligned()" if this helps and is sufficiently | |
1811 | * fast. | |
1812 | * | |
bfcfaa77 LT |
1813 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1814 | * do not trap on the (extremely unlikely) case of a page | |
1815 | * crossing operation. | |
1816 | * | |
1817 | * - Furthermore, we need an efficient 64-bit compile for the | |
1818 | * 64-bit case in order to generate the "number of bytes in | |
1819 | * the final mask". Again, that could be replaced with a | |
1820 | * efficient population count instruction or similar. | |
1821 | */ | |
1822 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1823 | ||
f68e556e | 1824 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1825 | |
468a9428 | 1826 | #ifdef HASH_MIX |
bfcfaa77 | 1827 | |
468a9428 | 1828 | /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ |
bfcfaa77 | 1829 | |
468a9428 | 1830 | #elif defined(CONFIG_64BIT) |
0fed3ac8 | 1831 | /* |
2a18da7a GS |
1832 | * Register pressure in the mixing function is an issue, particularly |
1833 | * on 32-bit x86, but almost any function requires one state value and | |
1834 | * one temporary. Instead, use a function designed for two state values | |
1835 | * and no temporaries. | |
1836 | * | |
1837 | * This function cannot create a collision in only two iterations, so | |
1838 | * we have two iterations to achieve avalanche. In those two iterations, | |
1839 | * we have six layers of mixing, which is enough to spread one bit's | |
1840 | * influence out to 2^6 = 64 state bits. | |
1841 | * | |
1842 | * Rotate constants are scored by considering either 64 one-bit input | |
1843 | * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the | |
1844 | * probability of that delta causing a change to each of the 128 output | |
1845 | * bits, using a sample of random initial states. | |
1846 | * | |
1847 | * The Shannon entropy of the computed probabilities is then summed | |
1848 | * to produce a score. Ideally, any input change has a 50% chance of | |
1849 | * toggling any given output bit. | |
1850 | * | |
1851 | * Mixing scores (in bits) for (12,45): | |
1852 | * Input delta: 1-bit 2-bit | |
1853 | * 1 round: 713.3 42542.6 | |
1854 | * 2 rounds: 2753.7 140389.8 | |
1855 | * 3 rounds: 5954.1 233458.2 | |
1856 | * 4 rounds: 7862.6 256672.2 | |
1857 | * Perfect: 8192 258048 | |
1858 | * (64*128) (64*63/2 * 128) | |
0fed3ac8 | 1859 | */ |
2a18da7a GS |
1860 | #define HASH_MIX(x, y, a) \ |
1861 | ( x ^= (a), \ | |
1862 | y ^= x, x = rol64(x,12),\ | |
1863 | x += y, y = rol64(y,45),\ | |
1864 | y *= 9 ) | |
bfcfaa77 | 1865 | |
0fed3ac8 | 1866 | /* |
2a18da7a GS |
1867 | * Fold two longs into one 32-bit hash value. This must be fast, but |
1868 | * latency isn't quite as critical, as there is a fair bit of additional | |
1869 | * work done before the hash value is used. | |
0fed3ac8 | 1870 | */ |
2a18da7a | 1871 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1872 | { |
2a18da7a GS |
1873 | y ^= x * GOLDEN_RATIO_64; |
1874 | y *= GOLDEN_RATIO_64; | |
1875 | return y >> 32; | |
0fed3ac8 GS |
1876 | } |
1877 | ||
bfcfaa77 LT |
1878 | #else /* 32-bit case */ |
1879 | ||
2a18da7a GS |
1880 | /* |
1881 | * Mixing scores (in bits) for (7,20): | |
1882 | * Input delta: 1-bit 2-bit | |
1883 | * 1 round: 330.3 9201.6 | |
1884 | * 2 rounds: 1246.4 25475.4 | |
1885 | * 3 rounds: 1907.1 31295.1 | |
1886 | * 4 rounds: 2042.3 31718.6 | |
1887 | * Perfect: 2048 31744 | |
1888 | * (32*64) (32*31/2 * 64) | |
1889 | */ | |
1890 | #define HASH_MIX(x, y, a) \ | |
1891 | ( x ^= (a), \ | |
1892 | y ^= x, x = rol32(x, 7),\ | |
1893 | x += y, y = rol32(y,20),\ | |
1894 | y *= 9 ) | |
bfcfaa77 | 1895 | |
2a18da7a | 1896 | static inline unsigned int fold_hash(unsigned long x, unsigned long y) |
0fed3ac8 | 1897 | { |
2a18da7a GS |
1898 | /* Use arch-optimized multiply if one exists */ |
1899 | return __hash_32(y ^ __hash_32(x)); | |
0fed3ac8 GS |
1900 | } |
1901 | ||
bfcfaa77 LT |
1902 | #endif |
1903 | ||
2a18da7a GS |
1904 | /* |
1905 | * Return the hash of a string of known length. This is carfully | |
1906 | * designed to match hash_name(), which is the more critical function. | |
1907 | * In particular, we must end by hashing a final word containing 0..7 | |
1908 | * payload bytes, to match the way that hash_name() iterates until it | |
1909 | * finds the delimiter after the name. | |
1910 | */ | |
8387ff25 | 1911 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
bfcfaa77 | 1912 | { |
8387ff25 | 1913 | unsigned long a, x = 0, y = (unsigned long)salt; |
bfcfaa77 LT |
1914 | |
1915 | for (;;) { | |
fcfd2fbf GS |
1916 | if (!len) |
1917 | goto done; | |
e419b4cc | 1918 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1919 | if (len < sizeof(unsigned long)) |
1920 | break; | |
2a18da7a | 1921 | HASH_MIX(x, y, a); |
bfcfaa77 LT |
1922 | name += sizeof(unsigned long); |
1923 | len -= sizeof(unsigned long); | |
bfcfaa77 | 1924 | } |
2a18da7a | 1925 | x ^= a & bytemask_from_count(len); |
bfcfaa77 | 1926 | done: |
2a18da7a | 1927 | return fold_hash(x, y); |
bfcfaa77 LT |
1928 | } |
1929 | EXPORT_SYMBOL(full_name_hash); | |
1930 | ||
fcfd2fbf | 1931 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1932 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1933 | { |
8387ff25 LT |
1934 | unsigned long a = 0, x = 0, y = (unsigned long)salt; |
1935 | unsigned long adata, mask, len; | |
fcfd2fbf GS |
1936 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
1937 | ||
8387ff25 LT |
1938 | len = 0; |
1939 | goto inside; | |
1940 | ||
fcfd2fbf | 1941 | do { |
2a18da7a | 1942 | HASH_MIX(x, y, a); |
fcfd2fbf | 1943 | len += sizeof(unsigned long); |
8387ff25 | 1944 | inside: |
fcfd2fbf GS |
1945 | a = load_unaligned_zeropad(name+len); |
1946 | } while (!has_zero(a, &adata, &constants)); | |
1947 | ||
1948 | adata = prep_zero_mask(a, adata, &constants); | |
1949 | mask = create_zero_mask(adata); | |
2a18da7a | 1950 | x ^= a & zero_bytemask(mask); |
fcfd2fbf | 1951 | |
2a18da7a | 1952 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
fcfd2fbf GS |
1953 | } |
1954 | EXPORT_SYMBOL(hashlen_string); | |
1955 | ||
bfcfaa77 LT |
1956 | /* |
1957 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1958 | * return the "hash_len" as the result. |
bfcfaa77 | 1959 | */ |
8387ff25 | 1960 | static inline u64 hash_name(const void *salt, const char *name) |
bfcfaa77 | 1961 | { |
8387ff25 LT |
1962 | unsigned long a = 0, b, x = 0, y = (unsigned long)salt; |
1963 | unsigned long adata, bdata, mask, len; | |
36126f8f | 1964 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; |
bfcfaa77 | 1965 | |
8387ff25 LT |
1966 | len = 0; |
1967 | goto inside; | |
1968 | ||
bfcfaa77 | 1969 | do { |
2a18da7a | 1970 | HASH_MIX(x, y, a); |
bfcfaa77 | 1971 | len += sizeof(unsigned long); |
8387ff25 | 1972 | inside: |
e419b4cc | 1973 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1974 | b = a ^ REPEAT_BYTE('/'); |
1975 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1976 | ||
1977 | adata = prep_zero_mask(a, adata, &constants); | |
1978 | bdata = prep_zero_mask(b, bdata, &constants); | |
36126f8f | 1979 | mask = create_zero_mask(adata | bdata); |
2a18da7a | 1980 | x ^= a & zero_bytemask(mask); |
36126f8f | 1981 | |
2a18da7a | 1982 | return hashlen_create(fold_hash(x, y), len + find_zero(mask)); |
bfcfaa77 LT |
1983 | } |
1984 | ||
2a18da7a | 1985 | #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ |
bfcfaa77 | 1986 | |
fcfd2fbf | 1987 | /* Return the hash of a string of known length */ |
8387ff25 | 1988 | unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) |
0145acc2 | 1989 | { |
8387ff25 | 1990 | unsigned long hash = init_name_hash(salt); |
0145acc2 | 1991 | while (len--) |
fcfd2fbf | 1992 | hash = partial_name_hash((unsigned char)*name++, hash); |
0145acc2 LT |
1993 | return end_name_hash(hash); |
1994 | } | |
ae942ae7 | 1995 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1996 | |
fcfd2fbf | 1997 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 1998 | u64 hashlen_string(const void *salt, const char *name) |
fcfd2fbf | 1999 | { |
8387ff25 | 2000 | unsigned long hash = init_name_hash(salt); |
fcfd2fbf GS |
2001 | unsigned long len = 0, c; |
2002 | ||
2003 | c = (unsigned char)*name; | |
e0ab7af9 | 2004 | while (c) { |
fcfd2fbf GS |
2005 | len++; |
2006 | hash = partial_name_hash(c, hash); | |
2007 | c = (unsigned char)name[len]; | |
e0ab7af9 | 2008 | } |
fcfd2fbf GS |
2009 | return hashlen_create(end_name_hash(hash), len); |
2010 | } | |
f2a031b6 | 2011 | EXPORT_SYMBOL(hashlen_string); |
fcfd2fbf | 2012 | |
200e9ef7 LT |
2013 | /* |
2014 | * We know there's a real path component here of at least | |
2015 | * one character. | |
2016 | */ | |
8387ff25 | 2017 | static inline u64 hash_name(const void *salt, const char *name) |
200e9ef7 | 2018 | { |
8387ff25 | 2019 | unsigned long hash = init_name_hash(salt); |
200e9ef7 LT |
2020 | unsigned long len = 0, c; |
2021 | ||
2022 | c = (unsigned char)*name; | |
2023 | do { | |
2024 | len++; | |
2025 | hash = partial_name_hash(c, hash); | |
2026 | c = (unsigned char)name[len]; | |
2027 | } while (c && c != '/'); | |
d6bb3e90 | 2028 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
2029 | } |
2030 | ||
bfcfaa77 LT |
2031 | #endif |
2032 | ||
1da177e4 LT |
2033 | /* |
2034 | * Name resolution. | |
ea3834d9 PM |
2035 | * This is the basic name resolution function, turning a pathname into |
2036 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 2037 | * |
ea3834d9 PM |
2038 | * Returns 0 and nd will have valid dentry and mnt on success. |
2039 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 2040 | */ |
6de88d72 | 2041 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 | 2042 | { |
1da177e4 | 2043 | int err; |
32cd7468 | 2044 | |
1da177e4 LT |
2045 | while (*name=='/') |
2046 | name++; | |
2047 | if (!*name) | |
9e18f10a | 2048 | return 0; |
1da177e4 | 2049 | |
1da177e4 LT |
2050 | /* At this point we know we have a real path component. */ |
2051 | for(;;) { | |
d6bb3e90 | 2052 | u64 hash_len; |
fe479a58 | 2053 | int type; |
1da177e4 | 2054 | |
52094c8a | 2055 | err = may_lookup(nd); |
2a18da7a | 2056 | if (err) |
3595e234 | 2057 | return err; |
1da177e4 | 2058 | |
8387ff25 | 2059 | hash_len = hash_name(nd->path.dentry, name); |
1da177e4 | 2060 | |
fe479a58 | 2061 | type = LAST_NORM; |
d6bb3e90 | 2062 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 2063 | case 2: |
200e9ef7 | 2064 | if (name[1] == '.') { |
fe479a58 | 2065 | type = LAST_DOTDOT; |
16c2cd71 AV |
2066 | nd->flags |= LOOKUP_JUMPED; |
2067 | } | |
fe479a58 AV |
2068 | break; |
2069 | case 1: | |
2070 | type = LAST_DOT; | |
2071 | } | |
5a202bcd AV |
2072 | if (likely(type == LAST_NORM)) { |
2073 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 2074 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 2075 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 2076 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 2077 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd | 2078 | if (err < 0) |
3595e234 | 2079 | return err; |
d6bb3e90 LT |
2080 | hash_len = this.hash_len; |
2081 | name = this.name; | |
5a202bcd AV |
2082 | } |
2083 | } | |
fe479a58 | 2084 | |
d6bb3e90 LT |
2085 | nd->last.hash_len = hash_len; |
2086 | nd->last.name = name; | |
5f4a6a69 AV |
2087 | nd->last_type = type; |
2088 | ||
d6bb3e90 LT |
2089 | name += hashlen_len(hash_len); |
2090 | if (!*name) | |
bdf6cbf1 | 2091 | goto OK; |
200e9ef7 LT |
2092 | /* |
2093 | * If it wasn't NUL, we know it was '/'. Skip that | |
2094 | * slash, and continue until no more slashes. | |
2095 | */ | |
2096 | do { | |
d6bb3e90 LT |
2097 | name++; |
2098 | } while (unlikely(*name == '/')); | |
8620c238 AV |
2099 | if (unlikely(!*name)) { |
2100 | OK: | |
368ee9ba | 2101 | /* pathname body, done */ |
8620c238 AV |
2102 | if (!nd->depth) |
2103 | return 0; | |
2104 | name = nd->stack[nd->depth - 1].name; | |
368ee9ba | 2105 | /* trailing symlink, done */ |
8620c238 AV |
2106 | if (!name) |
2107 | return 0; | |
2108 | /* last component of nested symlink */ | |
8f64fb1c | 2109 | err = walk_component(nd, WALK_FOLLOW); |
1c4ff1a8 AV |
2110 | } else { |
2111 | /* not the last component */ | |
8f64fb1c | 2112 | err = walk_component(nd, WALK_FOLLOW | WALK_MORE); |
8620c238 | 2113 | } |
ce57dfc1 | 2114 | if (err < 0) |
3595e234 | 2115 | return err; |
1da177e4 | 2116 | |
ce57dfc1 | 2117 | if (err) { |
626de996 | 2118 | const char *s = get_link(nd); |
5a460275 | 2119 | |
a1c83681 | 2120 | if (IS_ERR(s)) |
3595e234 | 2121 | return PTR_ERR(s); |
d40bcc09 AV |
2122 | err = 0; |
2123 | if (unlikely(!s)) { | |
2124 | /* jumped */ | |
b9ff4429 | 2125 | put_link(nd); |
d40bcc09 | 2126 | } else { |
fab51e8a AV |
2127 | nd->stack[nd->depth - 1].name = name; |
2128 | name = s; | |
2129 | continue; | |
d40bcc09 | 2130 | } |
31e6b01f | 2131 | } |
97242f99 AV |
2132 | if (unlikely(!d_can_lookup(nd->path.dentry))) { |
2133 | if (nd->flags & LOOKUP_RCU) { | |
4675ac39 | 2134 | if (unlazy_walk(nd)) |
97242f99 AV |
2135 | return -ECHILD; |
2136 | } | |
3595e234 | 2137 | return -ENOTDIR; |
97242f99 | 2138 | } |
1da177e4 | 2139 | } |
1da177e4 LT |
2140 | } |
2141 | ||
c8a53ee5 | 2142 | static const char *path_init(struct nameidata *nd, unsigned flags) |
31e6b01f | 2143 | { |
c8a53ee5 | 2144 | const char *s = nd->name->name; |
31e6b01f | 2145 | |
c0eb027e LT |
2146 | if (!*s) |
2147 | flags &= ~LOOKUP_RCU; | |
2148 | ||
31e6b01f | 2149 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ |
980f3ea2 | 2150 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; |
31e6b01f | 2151 | nd->depth = 0; |
5b6ca027 | 2152 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
2153 | struct dentry *root = nd->root.dentry; |
2154 | struct inode *inode = root->d_inode; | |
93893862 AV |
2155 | if (*s && unlikely(!d_can_lookup(root))) |
2156 | return ERR_PTR(-ENOTDIR); | |
5b6ca027 AV |
2157 | nd->path = nd->root; |
2158 | nd->inode = inode; | |
2159 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 2160 | rcu_read_lock(); |
5b6ca027 | 2161 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8f47a016 | 2162 | nd->root_seq = nd->seq; |
48a066e7 | 2163 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
2164 | } else { |
2165 | path_get(&nd->path); | |
2166 | } | |
368ee9ba | 2167 | return s; |
5b6ca027 AV |
2168 | } |
2169 | ||
31e6b01f | 2170 | nd->root.mnt = NULL; |
248fb5b9 AV |
2171 | nd->path.mnt = NULL; |
2172 | nd->path.dentry = NULL; | |
31e6b01f | 2173 | |
48a066e7 | 2174 | nd->m_seq = read_seqbegin(&mount_lock); |
fd2f7cb5 | 2175 | if (*s == '/') { |
9e6697e2 | 2176 | if (flags & LOOKUP_RCU) |
8b61e74f | 2177 | rcu_read_lock(); |
9e6697e2 | 2178 | set_root(nd); |
248fb5b9 | 2179 | if (likely(!nd_jump_root(nd))) |
ef55d917 | 2180 | return s; |
248fb5b9 | 2181 | nd->root.mnt = NULL; |
ef55d917 AV |
2182 | rcu_read_unlock(); |
2183 | return ERR_PTR(-ECHILD); | |
c8a53ee5 | 2184 | } else if (nd->dfd == AT_FDCWD) { |
e41f7d4e AV |
2185 | if (flags & LOOKUP_RCU) { |
2186 | struct fs_struct *fs = current->fs; | |
2187 | unsigned seq; | |
31e6b01f | 2188 | |
8b61e74f | 2189 | rcu_read_lock(); |
c28cc364 | 2190 | |
e41f7d4e AV |
2191 | do { |
2192 | seq = read_seqcount_begin(&fs->seq); | |
2193 | nd->path = fs->pwd; | |
ef55d917 | 2194 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e AV |
2195 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
2196 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2197 | } else { | |
2198 | get_fs_pwd(current->fs, &nd->path); | |
ef55d917 | 2199 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2200 | } |
ef55d917 | 2201 | return s; |
31e6b01f | 2202 | } else { |
582aa64a | 2203 | /* Caller must check execute permissions on the starting path component */ |
c8a53ee5 | 2204 | struct fd f = fdget_raw(nd->dfd); |
31e6b01f NP |
2205 | struct dentry *dentry; |
2206 | ||
2903ff01 | 2207 | if (!f.file) |
368ee9ba | 2208 | return ERR_PTR(-EBADF); |
31e6b01f | 2209 | |
2903ff01 | 2210 | dentry = f.file->f_path.dentry; |
31e6b01f | 2211 | |
fd2f7cb5 | 2212 | if (*s) { |
44b1d530 | 2213 | if (!d_can_lookup(dentry)) { |
2903ff01 | 2214 | fdput(f); |
368ee9ba | 2215 | return ERR_PTR(-ENOTDIR); |
2903ff01 | 2216 | } |
f52e0c11 | 2217 | } |
31e6b01f | 2218 | |
2903ff01 | 2219 | nd->path = f.file->f_path; |
e41f7d4e | 2220 | if (flags & LOOKUP_RCU) { |
8b61e74f | 2221 | rcu_read_lock(); |
34a26b99 AV |
2222 | nd->inode = nd->path.dentry->d_inode; |
2223 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
e41f7d4e | 2224 | } else { |
2903ff01 | 2225 | path_get(&nd->path); |
34a26b99 | 2226 | nd->inode = nd->path.dentry->d_inode; |
e41f7d4e | 2227 | } |
34a26b99 | 2228 | fdput(f); |
368ee9ba | 2229 | return s; |
31e6b01f | 2230 | } |
9b4a9b14 AV |
2231 | } |
2232 | ||
3bdba28b | 2233 | static const char *trailing_symlink(struct nameidata *nd) |
95fa25d9 AV |
2234 | { |
2235 | const char *s; | |
fec2fa24 | 2236 | int error = may_follow_link(nd); |
deb106c6 | 2237 | if (unlikely(error)) |
3bdba28b | 2238 | return ERR_PTR(error); |
95fa25d9 | 2239 | nd->flags |= LOOKUP_PARENT; |
fab51e8a | 2240 | nd->stack[0].name = NULL; |
3b2e7f75 | 2241 | s = get_link(nd); |
deb106c6 | 2242 | return s ? s : ""; |
95fa25d9 AV |
2243 | } |
2244 | ||
caa85634 | 2245 | static inline int lookup_last(struct nameidata *nd) |
bd92d7fe AV |
2246 | { |
2247 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2248 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2249 | ||
2250 | nd->flags &= ~LOOKUP_PARENT; | |
1c4ff1a8 | 2251 | return walk_component(nd, 0); |
bd92d7fe AV |
2252 | } |
2253 | ||
4f757f3c AV |
2254 | static int handle_lookup_down(struct nameidata *nd) |
2255 | { | |
2256 | struct path path = nd->path; | |
2257 | struct inode *inode = nd->inode; | |
2258 | unsigned seq = nd->seq; | |
2259 | int err; | |
2260 | ||
2261 | if (nd->flags & LOOKUP_RCU) { | |
2262 | /* | |
2263 | * don't bother with unlazy_walk on failure - we are | |
2264 | * at the very beginning of walk, so we lose nothing | |
2265 | * if we simply redo everything in non-RCU mode | |
2266 | */ | |
2267 | if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) | |
2268 | return -ECHILD; | |
2269 | } else { | |
2270 | dget(path.dentry); | |
2271 | err = follow_managed(&path, nd); | |
2272 | if (unlikely(err < 0)) | |
2273 | return err; | |
2274 | inode = d_backing_inode(path.dentry); | |
2275 | seq = 0; | |
2276 | } | |
2277 | path_to_nameidata(&path, nd); | |
2278 | nd->inode = inode; | |
2279 | nd->seq = seq; | |
2280 | return 0; | |
2281 | } | |
2282 | ||
9b4a9b14 | 2283 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2284 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) |
9b4a9b14 | 2285 | { |
c8a53ee5 | 2286 | const char *s = path_init(nd, flags); |
bd92d7fe | 2287 | int err; |
31e6b01f | 2288 | |
368ee9ba AV |
2289 | if (IS_ERR(s)) |
2290 | return PTR_ERR(s); | |
4f757f3c AV |
2291 | |
2292 | if (unlikely(flags & LOOKUP_DOWN)) { | |
2293 | err = handle_lookup_down(nd); | |
2294 | if (unlikely(err < 0)) { | |
2295 | terminate_walk(nd); | |
2296 | return err; | |
2297 | } | |
2298 | } | |
2299 | ||
3bdba28b AV |
2300 | while (!(err = link_path_walk(s, nd)) |
2301 | && ((err = lookup_last(nd)) > 0)) { | |
2302 | s = trailing_symlink(nd); | |
2303 | if (IS_ERR(s)) { | |
2304 | err = PTR_ERR(s); | |
2305 | break; | |
bd92d7fe AV |
2306 | } |
2307 | } | |
9f1fafee AV |
2308 | if (!err) |
2309 | err = complete_walk(nd); | |
bd92d7fe | 2310 | |
deb106c6 AV |
2311 | if (!err && nd->flags & LOOKUP_DIRECTORY) |
2312 | if (!d_can_lookup(nd->path.dentry)) | |
bd23a539 | 2313 | err = -ENOTDIR; |
625b6d10 AV |
2314 | if (!err) { |
2315 | *path = nd->path; | |
2316 | nd->path.mnt = NULL; | |
2317 | nd->path.dentry = NULL; | |
2318 | } | |
2319 | terminate_walk(nd); | |
bd92d7fe | 2320 | return err; |
ee0827cd | 2321 | } |
31e6b01f | 2322 | |
625b6d10 | 2323 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, |
9ad1aaa6 | 2324 | struct path *path, struct path *root) |
ee0827cd | 2325 | { |
894bc8c4 | 2326 | int retval; |
9883d185 | 2327 | struct nameidata nd; |
abc9f5be AV |
2328 | if (IS_ERR(name)) |
2329 | return PTR_ERR(name); | |
9ad1aaa6 AV |
2330 | if (unlikely(root)) { |
2331 | nd.root = *root; | |
2332 | flags |= LOOKUP_ROOT; | |
2333 | } | |
9883d185 | 2334 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2335 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); |
ee0827cd | 2336 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2337 | retval = path_lookupat(&nd, flags, path); |
ee0827cd | 2338 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2339 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); |
31e6b01f | 2340 | |
f78570dd | 2341 | if (likely(!retval)) |
625b6d10 | 2342 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); |
9883d185 | 2343 | restore_nameidata(); |
e4bd1c1a | 2344 | putname(name); |
170aa3d0 | 2345 | return retval; |
1da177e4 LT |
2346 | } |
2347 | ||
8bcb77fa | 2348 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
c8a53ee5 | 2349 | static int path_parentat(struct nameidata *nd, unsigned flags, |
391172c4 | 2350 | struct path *parent) |
8bcb77fa | 2351 | { |
c8a53ee5 | 2352 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2353 | int err; |
2354 | if (IS_ERR(s)) | |
2355 | return PTR_ERR(s); | |
2356 | err = link_path_walk(s, nd); | |
8bcb77fa AV |
2357 | if (!err) |
2358 | err = complete_walk(nd); | |
391172c4 AV |
2359 | if (!err) { |
2360 | *parent = nd->path; | |
2361 | nd->path.mnt = NULL; | |
2362 | nd->path.dentry = NULL; | |
2363 | } | |
2364 | terminate_walk(nd); | |
8bcb77fa AV |
2365 | return err; |
2366 | } | |
2367 | ||
5c31b6ce | 2368 | static struct filename *filename_parentat(int dfd, struct filename *name, |
391172c4 AV |
2369 | unsigned int flags, struct path *parent, |
2370 | struct qstr *last, int *type) | |
8bcb77fa AV |
2371 | { |
2372 | int retval; | |
9883d185 | 2373 | struct nameidata nd; |
8bcb77fa | 2374 | |
5c31b6ce AV |
2375 | if (IS_ERR(name)) |
2376 | return name; | |
9883d185 | 2377 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2378 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); |
8bcb77fa | 2379 | if (unlikely(retval == -ECHILD)) |
c8a53ee5 | 2380 | retval = path_parentat(&nd, flags, parent); |
8bcb77fa | 2381 | if (unlikely(retval == -ESTALE)) |
c8a53ee5 | 2382 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); |
391172c4 AV |
2383 | if (likely(!retval)) { |
2384 | *last = nd.last; | |
2385 | *type = nd.last_type; | |
2386 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
5c31b6ce AV |
2387 | } else { |
2388 | putname(name); | |
2389 | name = ERR_PTR(retval); | |
391172c4 | 2390 | } |
9883d185 | 2391 | restore_nameidata(); |
5c31b6ce | 2392 | return name; |
8bcb77fa AV |
2393 | } |
2394 | ||
79714f72 AV |
2395 | /* does lookup, returns the object with parent locked */ |
2396 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2397 | { |
5c31b6ce AV |
2398 | struct filename *filename; |
2399 | struct dentry *d; | |
391172c4 AV |
2400 | struct qstr last; |
2401 | int type; | |
51689104 | 2402 | |
5c31b6ce AV |
2403 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, |
2404 | &last, &type); | |
51689104 PM |
2405 | if (IS_ERR(filename)) |
2406 | return ERR_CAST(filename); | |
5c31b6ce | 2407 | if (unlikely(type != LAST_NORM)) { |
391172c4 | 2408 | path_put(path); |
5c31b6ce AV |
2409 | putname(filename); |
2410 | return ERR_PTR(-EINVAL); | |
79714f72 | 2411 | } |
5955102c | 2412 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 2413 | d = __lookup_hash(&last, path->dentry, 0); |
79714f72 | 2414 | if (IS_ERR(d)) { |
5955102c | 2415 | inode_unlock(path->dentry->d_inode); |
391172c4 | 2416 | path_put(path); |
79714f72 | 2417 | } |
51689104 | 2418 | putname(filename); |
79714f72 | 2419 | return d; |
5590ff0d UD |
2420 | } |
2421 | ||
d1811465 AV |
2422 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2423 | { | |
abc9f5be AV |
2424 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2425 | flags, path, NULL); | |
d1811465 | 2426 | } |
4d359507 | 2427 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2428 | |
16f18200 JJS |
2429 | /** |
2430 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2431 | * @dentry: pointer to dentry of the base directory | |
2432 | * @mnt: pointer to vfs mount of the base directory | |
2433 | * @name: pointer to file name | |
2434 | * @flags: lookup flags | |
e0a01249 | 2435 | * @path: pointer to struct path to fill |
16f18200 JJS |
2436 | */ |
2437 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2438 | const char *name, unsigned int flags, | |
e0a01249 | 2439 | struct path *path) |
16f18200 | 2440 | { |
9ad1aaa6 | 2441 | struct path root = {.mnt = mnt, .dentry = dentry}; |
9ad1aaa6 | 2442 | /* the first argument of filename_lookup() is ignored with root */ |
abc9f5be AV |
2443 | return filename_lookup(AT_FDCWD, getname_kernel(name), |
2444 | flags , path, &root); | |
16f18200 | 2445 | } |
4d359507 | 2446 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2447 | |
eead1911 | 2448 | /** |
a6b91919 | 2449 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2450 | * @name: pathname component to lookup |
2451 | * @base: base directory to lookup from | |
2452 | * @len: maximum length @len should be interpreted to | |
2453 | * | |
a6b91919 | 2454 | * Note that this routine is purely a helper for filesystem usage and should |
9e7543e9 | 2455 | * not be called by generic code. |
bbddca8e N |
2456 | * |
2457 | * The caller must hold base->i_mutex. | |
eead1911 | 2458 | */ |
057f6c01 JM |
2459 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2460 | { | |
057f6c01 | 2461 | struct qstr this; |
6a96ba54 | 2462 | unsigned int c; |
cda309de | 2463 | int err; |
057f6c01 | 2464 | |
5955102c | 2465 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); |
2f9092e1 | 2466 | |
6a96ba54 AV |
2467 | this.name = name; |
2468 | this.len = len; | |
8387ff25 | 2469 | this.hash = full_name_hash(base, name, len); |
6a96ba54 AV |
2470 | if (!len) |
2471 | return ERR_PTR(-EACCES); | |
2472 | ||
21d8a15a AV |
2473 | if (unlikely(name[0] == '.')) { |
2474 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2475 | return ERR_PTR(-EACCES); | |
2476 | } | |
2477 | ||
6a96ba54 AV |
2478 | while (len--) { |
2479 | c = *(const unsigned char *)name++; | |
2480 | if (c == '/' || c == '\0') | |
2481 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2482 | } |
5a202bcd AV |
2483 | /* |
2484 | * See if the low-level filesystem might want | |
2485 | * to use its own hash.. | |
2486 | */ | |
2487 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2488 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2489 | if (err < 0) |
2490 | return ERR_PTR(err); | |
2491 | } | |
eead1911 | 2492 | |
cda309de MS |
2493 | err = inode_permission(base->d_inode, MAY_EXEC); |
2494 | if (err) | |
2495 | return ERR_PTR(err); | |
2496 | ||
72bd866a | 2497 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2498 | } |
4d359507 | 2499 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2500 | |
bbddca8e N |
2501 | /** |
2502 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2503 | * @name: pathname component to lookup | |
2504 | * @base: base directory to lookup from | |
2505 | * @len: maximum length @len should be interpreted to | |
2506 | * | |
2507 | * Note that this routine is purely a helper for filesystem usage and should | |
2508 | * not be called by generic code. | |
2509 | * | |
2510 | * Unlike lookup_one_len, it should be called without the parent | |
2511 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2512 | */ | |
2513 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2514 | struct dentry *base, int len) | |
2515 | { | |
2516 | struct qstr this; | |
2517 | unsigned int c; | |
2518 | int err; | |
20d00ee8 | 2519 | struct dentry *ret; |
bbddca8e N |
2520 | |
2521 | this.name = name; | |
2522 | this.len = len; | |
8387ff25 | 2523 | this.hash = full_name_hash(base, name, len); |
bbddca8e N |
2524 | if (!len) |
2525 | return ERR_PTR(-EACCES); | |
2526 | ||
2527 | if (unlikely(name[0] == '.')) { | |
2528 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2529 | return ERR_PTR(-EACCES); | |
2530 | } | |
2531 | ||
2532 | while (len--) { | |
2533 | c = *(const unsigned char *)name++; | |
2534 | if (c == '/' || c == '\0') | |
2535 | return ERR_PTR(-EACCES); | |
2536 | } | |
2537 | /* | |
2538 | * See if the low-level filesystem might want | |
2539 | * to use its own hash.. | |
2540 | */ | |
2541 | if (base->d_flags & DCACHE_OP_HASH) { | |
2542 | int err = base->d_op->d_hash(base, &this); | |
2543 | if (err < 0) | |
2544 | return ERR_PTR(err); | |
2545 | } | |
2546 | ||
2547 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2548 | if (err) | |
2549 | return ERR_PTR(err); | |
2550 | ||
20d00ee8 LT |
2551 | ret = lookup_dcache(&this, base, 0); |
2552 | if (!ret) | |
2553 | ret = lookup_slow(&this, base, 0); | |
2554 | return ret; | |
bbddca8e N |
2555 | } |
2556 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2557 | ||
eedf265a EB |
2558 | #ifdef CONFIG_UNIX98_PTYS |
2559 | int path_pts(struct path *path) | |
2560 | { | |
2561 | /* Find something mounted on "pts" in the same directory as | |
2562 | * the input path. | |
2563 | */ | |
2564 | struct dentry *child, *parent; | |
2565 | struct qstr this; | |
2566 | int ret; | |
2567 | ||
2568 | ret = path_parent_directory(path); | |
2569 | if (ret) | |
2570 | return ret; | |
2571 | ||
2572 | parent = path->dentry; | |
2573 | this.name = "pts"; | |
2574 | this.len = 3; | |
2575 | child = d_hash_and_lookup(parent, &this); | |
2576 | if (!child) | |
2577 | return -ENOENT; | |
2578 | ||
2579 | path->dentry = child; | |
2580 | dput(parent); | |
2581 | follow_mount(path); | |
2582 | return 0; | |
2583 | } | |
2584 | #endif | |
2585 | ||
1fa1e7f6 AW |
2586 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2587 | struct path *path, int *empty) | |
1da177e4 | 2588 | { |
abc9f5be AV |
2589 | return filename_lookup(dfd, getname_flags(name, flags, empty), |
2590 | flags, path, NULL); | |
1da177e4 | 2591 | } |
b853a161 | 2592 | EXPORT_SYMBOL(user_path_at_empty); |
1fa1e7f6 | 2593 | |
8033426e | 2594 | /** |
197df04c | 2595 | * mountpoint_last - look up last component for umount |
8033426e | 2596 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
8033426e JL |
2597 | * |
2598 | * This is a special lookup_last function just for umount. In this case, we | |
2599 | * need to resolve the path without doing any revalidation. | |
2600 | * | |
2601 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2602 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2603 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2604 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2605 | * bogus and it doesn't exist. | |
2606 | * | |
2607 | * Returns: | |
2608 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
ba8f4613 | 2609 | * lookup found a negative dentry. |
8033426e | 2610 | * |
ba8f4613 AV |
2611 | * 0: if we successfully resolved nd->last and found it to not to be a |
2612 | * symlink that needs to be followed. | |
8033426e JL |
2613 | * |
2614 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
ba8f4613 | 2615 | * that needs to be followed. |
8033426e JL |
2616 | */ |
2617 | static int | |
ba8f4613 | 2618 | mountpoint_last(struct nameidata *nd) |
8033426e JL |
2619 | { |
2620 | int error = 0; | |
8033426e | 2621 | struct dentry *dir = nd->path.dentry; |
ba8f4613 | 2622 | struct path path; |
8033426e | 2623 | |
35759521 AV |
2624 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2625 | if (nd->flags & LOOKUP_RCU) { | |
4675ac39 | 2626 | if (unlazy_walk(nd)) |
deb106c6 | 2627 | return -ECHILD; |
8033426e JL |
2628 | } |
2629 | ||
2630 | nd->flags &= ~LOOKUP_PARENT; | |
2631 | ||
2632 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2633 | error = handle_dots(nd, nd->last_type); | |
35759521 | 2634 | if (error) |
deb106c6 | 2635 | return error; |
ba8f4613 | 2636 | path.dentry = dget(nd->path.dentry); |
949a852e | 2637 | } else { |
ba8f4613 AV |
2638 | path.dentry = d_lookup(dir, &nd->last); |
2639 | if (!path.dentry) { | |
949a852e AV |
2640 | /* |
2641 | * No cached dentry. Mounted dentries are pinned in the | |
2642 | * cache, so that means that this dentry is probably | |
2643 | * a symlink or the path doesn't actually point | |
2644 | * to a mounted dentry. | |
2645 | */ | |
ba8f4613 | 2646 | path.dentry = lookup_slow(&nd->last, dir, |
949a852e | 2647 | nd->flags | LOOKUP_NO_REVAL); |
ba8f4613 AV |
2648 | if (IS_ERR(path.dentry)) |
2649 | return PTR_ERR(path.dentry); | |
bcceeeba | 2650 | } |
8033426e | 2651 | } |
ba8f4613 AV |
2652 | if (d_is_negative(path.dentry)) { |
2653 | dput(path.dentry); | |
deb106c6 | 2654 | return -ENOENT; |
8033426e | 2655 | } |
ba8f4613 | 2656 | path.mnt = nd->path.mnt; |
8f64fb1c | 2657 | return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); |
8033426e JL |
2658 | } |
2659 | ||
2660 | /** | |
197df04c | 2661 | * path_mountpoint - look up a path to be umounted |
2a78b857 | 2662 | * @nd: lookup context |
8033426e | 2663 | * @flags: lookup flags |
c8a53ee5 | 2664 | * @path: pointer to container for result |
8033426e JL |
2665 | * |
2666 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2667 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2668 | */ |
2669 | static int | |
c8a53ee5 | 2670 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) |
8033426e | 2671 | { |
c8a53ee5 | 2672 | const char *s = path_init(nd, flags); |
368ee9ba AV |
2673 | int err; |
2674 | if (IS_ERR(s)) | |
2675 | return PTR_ERR(s); | |
3bdba28b | 2676 | while (!(err = link_path_walk(s, nd)) && |
ba8f4613 | 2677 | (err = mountpoint_last(nd)) > 0) { |
3bdba28b AV |
2678 | s = trailing_symlink(nd); |
2679 | if (IS_ERR(s)) { | |
2680 | err = PTR_ERR(s); | |
8033426e | 2681 | break; |
3bdba28b | 2682 | } |
8033426e | 2683 | } |
ba8f4613 AV |
2684 | if (!err) { |
2685 | *path = nd->path; | |
2686 | nd->path.mnt = NULL; | |
2687 | nd->path.dentry = NULL; | |
2688 | follow_mount(path); | |
2689 | } | |
deb106c6 | 2690 | terminate_walk(nd); |
8033426e JL |
2691 | return err; |
2692 | } | |
2693 | ||
2d864651 | 2694 | static int |
668696dc | 2695 | filename_mountpoint(int dfd, struct filename *name, struct path *path, |
2d864651 AV |
2696 | unsigned int flags) |
2697 | { | |
9883d185 | 2698 | struct nameidata nd; |
cbaab2db | 2699 | int error; |
668696dc AV |
2700 | if (IS_ERR(name)) |
2701 | return PTR_ERR(name); | |
9883d185 | 2702 | set_nameidata(&nd, dfd, name); |
c8a53ee5 | 2703 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); |
2d864651 | 2704 | if (unlikely(error == -ECHILD)) |
c8a53ee5 | 2705 | error = path_mountpoint(&nd, flags, path); |
2d864651 | 2706 | if (unlikely(error == -ESTALE)) |
c8a53ee5 | 2707 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); |
2d864651 | 2708 | if (likely(!error)) |
668696dc | 2709 | audit_inode(name, path->dentry, 0); |
9883d185 | 2710 | restore_nameidata(); |
668696dc | 2711 | putname(name); |
2d864651 AV |
2712 | return error; |
2713 | } | |
2714 | ||
8033426e | 2715 | /** |
197df04c | 2716 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2717 | * @dfd: directory file descriptor |
2718 | * @name: pathname from userland | |
2719 | * @flags: lookup flags | |
2720 | * @path: pointer to container to hold result | |
2721 | * | |
2722 | * A umount is a special case for path walking. We're not actually interested | |
2723 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2724 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2725 | * and avoid revalidating the last component. | |
2726 | * | |
2727 | * Returns 0 and populates "path" on success. | |
2728 | */ | |
2729 | int | |
197df04c | 2730 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2731 | struct path *path) |
2732 | { | |
cbaab2db | 2733 | return filename_mountpoint(dfd, getname(name), path, flags); |
8033426e JL |
2734 | } |
2735 | ||
2d864651 AV |
2736 | int |
2737 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2738 | unsigned int flags) | |
2739 | { | |
cbaab2db | 2740 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); |
2d864651 AV |
2741 | } |
2742 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2743 | ||
cbdf35bc | 2744 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2745 | { |
8e96e3b7 | 2746 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2747 | |
8e96e3b7 | 2748 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2749 | return 0; |
8e96e3b7 | 2750 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2751 | return 0; |
23adbe12 | 2752 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2753 | } |
cbdf35bc | 2754 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2755 | |
2756 | /* | |
2757 | * Check whether we can remove a link victim from directory dir, check | |
2758 | * whether the type of victim is right. | |
2759 | * 1. We can't do it if dir is read-only (done in permission()) | |
2760 | * 2. We should have write and exec permissions on dir | |
2761 | * 3. We can't remove anything from append-only dir | |
2762 | * 4. We can't do anything with immutable dir (done in permission()) | |
2763 | * 5. If the sticky bit on dir is set we should either | |
2764 | * a. be owner of dir, or | |
2765 | * b. be owner of victim, or | |
2766 | * c. have CAP_FOWNER capability | |
2767 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2768 | * links pointing to it. | |
0bd23d09 EB |
2769 | * 7. If the victim has an unknown uid or gid we can't change the inode. |
2770 | * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2771 | * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2772 | * 10. We can't remove a root or mountpoint. | |
2773 | * 11. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1da177e4 LT |
2774 | * nfs_async_unlink(). |
2775 | */ | |
b18825a7 | 2776 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2777 | { |
63afdfc7 | 2778 | struct inode *inode = d_backing_inode(victim); |
1da177e4 LT |
2779 | int error; |
2780 | ||
b18825a7 | 2781 | if (d_is_negative(victim)) |
1da177e4 | 2782 | return -ENOENT; |
b18825a7 | 2783 | BUG_ON(!inode); |
1da177e4 LT |
2784 | |
2785 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2786 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2787 | |
f419a2e3 | 2788 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2789 | if (error) |
2790 | return error; | |
2791 | if (IS_APPEND(dir)) | |
2792 | return -EPERM; | |
b18825a7 DH |
2793 | |
2794 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
0bd23d09 | 2795 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) |
1da177e4 LT |
2796 | return -EPERM; |
2797 | if (isdir) { | |
44b1d530 | 2798 | if (!d_is_dir(victim)) |
1da177e4 LT |
2799 | return -ENOTDIR; |
2800 | if (IS_ROOT(victim)) | |
2801 | return -EBUSY; | |
44b1d530 | 2802 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2803 | return -EISDIR; |
2804 | if (IS_DEADDIR(dir)) | |
2805 | return -ENOENT; | |
2806 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2807 | return -EBUSY; | |
2808 | return 0; | |
2809 | } | |
2810 | ||
2811 | /* Check whether we can create an object with dentry child in directory | |
2812 | * dir. | |
2813 | * 1. We can't do it if child already exists (open has special treatment for | |
2814 | * this case, but since we are inlined it's OK) | |
2815 | * 2. We can't do it if dir is read-only (done in permission()) | |
036d5236 EB |
2816 | * 3. We can't do it if the fs can't represent the fsuid or fsgid. |
2817 | * 4. We should have write and exec permissions on dir | |
2818 | * 5. We can't do it if dir is immutable (done in permission()) | |
1da177e4 | 2819 | */ |
a95164d9 | 2820 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2821 | { |
036d5236 | 2822 | struct user_namespace *s_user_ns; |
14e972b4 | 2823 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2824 | if (child->d_inode) |
2825 | return -EEXIST; | |
2826 | if (IS_DEADDIR(dir)) | |
2827 | return -ENOENT; | |
036d5236 EB |
2828 | s_user_ns = dir->i_sb->s_user_ns; |
2829 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2830 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2831 | return -EOVERFLOW; | |
f419a2e3 | 2832 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2833 | } |
2834 | ||
1da177e4 LT |
2835 | /* |
2836 | * p1 and p2 should be directories on the same fs. | |
2837 | */ | |
2838 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2839 | { | |
2840 | struct dentry *p; | |
2841 | ||
2842 | if (p1 == p2) { | |
5955102c | 2843 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
1da177e4 LT |
2844 | return NULL; |
2845 | } | |
2846 | ||
fc64005c | 2847 | mutex_lock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 | 2848 | |
e2761a11 OH |
2849 | p = d_ancestor(p2, p1); |
2850 | if (p) { | |
5955102c AV |
2851 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); |
2852 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2853 | return p; |
1da177e4 LT |
2854 | } |
2855 | ||
e2761a11 OH |
2856 | p = d_ancestor(p1, p2); |
2857 | if (p) { | |
5955102c AV |
2858 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2859 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
e2761a11 | 2860 | return p; |
1da177e4 LT |
2861 | } |
2862 | ||
5955102c AV |
2863 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); |
2864 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
1da177e4 LT |
2865 | return NULL; |
2866 | } | |
4d359507 | 2867 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2868 | |
2869 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2870 | { | |
5955102c | 2871 | inode_unlock(p1->d_inode); |
1da177e4 | 2872 | if (p1 != p2) { |
5955102c | 2873 | inode_unlock(p2->d_inode); |
fc64005c | 2874 | mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2875 | } |
2876 | } | |
4d359507 | 2877 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2878 | |
4acdaf27 | 2879 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2880 | bool want_excl) |
1da177e4 | 2881 | { |
a95164d9 | 2882 | int error = may_create(dir, dentry); |
1da177e4 LT |
2883 | if (error) |
2884 | return error; | |
2885 | ||
acfa4380 | 2886 | if (!dir->i_op->create) |
1da177e4 LT |
2887 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2888 | mode &= S_IALLUGO; | |
2889 | mode |= S_IFREG; | |
2890 | error = security_inode_create(dir, dentry, mode); | |
2891 | if (error) | |
2892 | return error; | |
312b63fb | 2893 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2894 | if (!error) |
f38aa942 | 2895 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2896 | return error; |
2897 | } | |
4d359507 | 2898 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2899 | |
a2982cc9 EB |
2900 | bool may_open_dev(const struct path *path) |
2901 | { | |
2902 | return !(path->mnt->mnt_flags & MNT_NODEV) && | |
2903 | !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); | |
2904 | } | |
2905 | ||
f0bb5aaf | 2906 | static int may_open(const struct path *path, int acc_mode, int flag) |
1da177e4 | 2907 | { |
3fb64190 | 2908 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2909 | struct inode *inode = dentry->d_inode; |
2910 | int error; | |
2911 | ||
2912 | if (!inode) | |
2913 | return -ENOENT; | |
2914 | ||
c8fe8f30 CH |
2915 | switch (inode->i_mode & S_IFMT) { |
2916 | case S_IFLNK: | |
1da177e4 | 2917 | return -ELOOP; |
c8fe8f30 CH |
2918 | case S_IFDIR: |
2919 | if (acc_mode & MAY_WRITE) | |
2920 | return -EISDIR; | |
2921 | break; | |
2922 | case S_IFBLK: | |
2923 | case S_IFCHR: | |
a2982cc9 | 2924 | if (!may_open_dev(path)) |
1da177e4 | 2925 | return -EACCES; |
c8fe8f30 CH |
2926 | /*FALLTHRU*/ |
2927 | case S_IFIFO: | |
2928 | case S_IFSOCK: | |
1da177e4 | 2929 | flag &= ~O_TRUNC; |
c8fe8f30 | 2930 | break; |
4a3fd211 | 2931 | } |
b41572e9 | 2932 | |
62fb4a15 | 2933 | error = inode_permission(inode, MAY_OPEN | acc_mode); |
b41572e9 DH |
2934 | if (error) |
2935 | return error; | |
6146f0d5 | 2936 | |
1da177e4 LT |
2937 | /* |
2938 | * An append-only file must be opened in append mode for writing. | |
2939 | */ | |
2940 | if (IS_APPEND(inode)) { | |
8737c930 | 2941 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2942 | return -EPERM; |
1da177e4 | 2943 | if (flag & O_TRUNC) |
7715b521 | 2944 | return -EPERM; |
1da177e4 LT |
2945 | } |
2946 | ||
2947 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2948 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2949 | return -EPERM; |
1da177e4 | 2950 | |
f3c7691e | 2951 | return 0; |
7715b521 | 2952 | } |
1da177e4 | 2953 | |
e1181ee6 | 2954 | static int handle_truncate(struct file *filp) |
7715b521 | 2955 | { |
f0bb5aaf | 2956 | const struct path *path = &filp->f_path; |
7715b521 AV |
2957 | struct inode *inode = path->dentry->d_inode; |
2958 | int error = get_write_access(inode); | |
2959 | if (error) | |
2960 | return error; | |
2961 | /* | |
2962 | * Refuse to truncate files with mandatory locks held on them. | |
2963 | */ | |
d7a06983 | 2964 | error = locks_verify_locked(filp); |
7715b521 | 2965 | if (!error) |
ea0d3ab2 | 2966 | error = security_path_truncate(path); |
7715b521 AV |
2967 | if (!error) { |
2968 | error = do_truncate(path->dentry, 0, | |
2969 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2970 | filp); |
7715b521 AV |
2971 | } |
2972 | put_write_access(inode); | |
acd0c935 | 2973 | return error; |
1da177e4 LT |
2974 | } |
2975 | ||
d57999e1 DH |
2976 | static inline int open_to_namei_flags(int flag) |
2977 | { | |
8a5e929d AV |
2978 | if ((flag & O_ACCMODE) == 3) |
2979 | flag--; | |
d57999e1 DH |
2980 | return flag; |
2981 | } | |
2982 | ||
d3607752 | 2983 | static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) |
d18e9008 | 2984 | { |
1328c727 | 2985 | struct user_namespace *s_user_ns; |
d18e9008 MS |
2986 | int error = security_path_mknod(dir, dentry, mode, 0); |
2987 | if (error) | |
2988 | return error; | |
2989 | ||
1328c727 SF |
2990 | s_user_ns = dir->dentry->d_sb->s_user_ns; |
2991 | if (!kuid_has_mapping(s_user_ns, current_fsuid()) || | |
2992 | !kgid_has_mapping(s_user_ns, current_fsgid())) | |
2993 | return -EOVERFLOW; | |
2994 | ||
d18e9008 MS |
2995 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); |
2996 | if (error) | |
2997 | return error; | |
2998 | ||
2999 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
3000 | } | |
3001 | ||
1acf0af9 DH |
3002 | /* |
3003 | * Attempt to atomically look up, create and open a file from a negative | |
3004 | * dentry. | |
3005 | * | |
3006 | * Returns 0 if successful. The file will have been created and attached to | |
3007 | * @file by the filesystem calling finish_open(). | |
3008 | * | |
3009 | * Returns 1 if the file was looked up only or didn't need creating. The | |
3010 | * caller will need to perform the open themselves. @path will have been | |
3011 | * updated to point to the new dentry. This may be negative. | |
3012 | * | |
3013 | * Returns an error code otherwise. | |
3014 | */ | |
2675a4eb AV |
3015 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
3016 | struct path *path, struct file *file, | |
3017 | const struct open_flags *op, | |
1643b43f | 3018 | int open_flag, umode_t mode, |
2675a4eb | 3019 | int *opened) |
d18e9008 | 3020 | { |
384f26e2 | 3021 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; |
d18e9008 | 3022 | struct inode *dir = nd->path.dentry->d_inode; |
d18e9008 | 3023 | int error; |
d18e9008 | 3024 | |
384f26e2 | 3025 | if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ |
d18e9008 | 3026 | open_flag &= ~O_TRUNC; |
d18e9008 | 3027 | |
d18e9008 MS |
3028 | if (nd->flags & LOOKUP_DIRECTORY) |
3029 | open_flag |= O_DIRECTORY; | |
3030 | ||
30d90494 AV |
3031 | file->f_path.dentry = DENTRY_NOT_SET; |
3032 | file->f_path.mnt = nd->path.mnt; | |
0fb1ea09 AV |
3033 | error = dir->i_op->atomic_open(dir, dentry, file, |
3034 | open_to_namei_flags(open_flag), | |
3035 | mode, opened); | |
6fbd0714 | 3036 | d_lookup_done(dentry); |
384f26e2 AV |
3037 | if (!error) { |
3038 | /* | |
3039 | * We didn't have the inode before the open, so check open | |
3040 | * permission here. | |
3041 | */ | |
3042 | int acc_mode = op->acc_mode; | |
3043 | if (*opened & FILE_CREATED) { | |
3044 | WARN_ON(!(open_flag & O_CREAT)); | |
3045 | fsnotify_create(dir, dentry); | |
3046 | acc_mode = 0; | |
3047 | } | |
3048 | error = may_open(&file->f_path, acc_mode, open_flag); | |
3049 | if (WARN_ON(error > 0)) | |
3050 | error = -EINVAL; | |
3051 | } else if (error > 0) { | |
30d90494 | 3052 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 3053 | error = -EIO; |
03da633a | 3054 | } else { |
384f26e2 AV |
3055 | if (file->f_path.dentry) { |
3056 | dput(dentry); | |
3057 | dentry = file->f_path.dentry; | |
03da633a | 3058 | } |
384f26e2 AV |
3059 | if (*opened & FILE_CREATED) |
3060 | fsnotify_create(dir, dentry); | |
a01e718f AV |
3061 | if (unlikely(d_is_negative(dentry))) { |
3062 | error = -ENOENT; | |
3063 | } else { | |
3064 | path->dentry = dentry; | |
3065 | path->mnt = nd->path.mnt; | |
3066 | return 1; | |
3067 | } | |
62b2ce96 | 3068 | } |
d18e9008 | 3069 | } |
d18e9008 | 3070 | dput(dentry); |
2675a4eb | 3071 | return error; |
d18e9008 MS |
3072 | } |
3073 | ||
d58ffd35 | 3074 | /* |
1acf0af9 | 3075 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
3076 | * |
3077 | * Must be called with i_mutex held on parent. | |
3078 | * | |
1acf0af9 DH |
3079 | * Returns 0 if the file was successfully atomically created (if necessary) and |
3080 | * opened. In this case the file will be returned attached to @file. | |
3081 | * | |
3082 | * Returns 1 if the file was not completely opened at this time, though lookups | |
3083 | * and creations will have been performed and the dentry returned in @path will | |
3084 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
3085 | * specified then a negative dentry may be returned. | |
3086 | * | |
3087 | * An error code is returned otherwise. | |
3088 | * | |
3089 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
3090 | * cleared otherwise prior to returning. | |
d58ffd35 | 3091 | */ |
2675a4eb AV |
3092 | static int lookup_open(struct nameidata *nd, struct path *path, |
3093 | struct file *file, | |
3094 | const struct open_flags *op, | |
64894cf8 | 3095 | bool got_write, int *opened) |
d58ffd35 MS |
3096 | { |
3097 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 3098 | struct inode *dir_inode = dir->d_inode; |
1643b43f | 3099 | int open_flag = op->open_flag; |
d58ffd35 | 3100 | struct dentry *dentry; |
1643b43f | 3101 | int error, create_error = 0; |
1643b43f | 3102 | umode_t mode = op->mode; |
6fbd0714 | 3103 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
d58ffd35 | 3104 | |
ce8644fc AV |
3105 | if (unlikely(IS_DEADDIR(dir_inode))) |
3106 | return -ENOENT; | |
d58ffd35 | 3107 | |
47237687 | 3108 | *opened &= ~FILE_CREATED; |
6fbd0714 AV |
3109 | dentry = d_lookup(dir, &nd->last); |
3110 | for (;;) { | |
3111 | if (!dentry) { | |
3112 | dentry = d_alloc_parallel(dir, &nd->last, &wq); | |
3113 | if (IS_ERR(dentry)) | |
3114 | return PTR_ERR(dentry); | |
3115 | } | |
3116 | if (d_in_lookup(dentry)) | |
3117 | break; | |
d58ffd35 | 3118 | |
6fbd0714 AV |
3119 | error = d_revalidate(dentry, nd->flags); |
3120 | if (likely(error > 0)) | |
3121 | break; | |
3122 | if (error) | |
3123 | goto out_dput; | |
3124 | d_invalidate(dentry); | |
3125 | dput(dentry); | |
3126 | dentry = NULL; | |
3127 | } | |
3128 | if (dentry->d_inode) { | |
6c51e513 | 3129 | /* Cached positive dentry: will open in f_op->open */ |
d18e9008 | 3130 | goto out_no_open; |
6c51e513 | 3131 | } |
d18e9008 | 3132 | |
1643b43f AV |
3133 | /* |
3134 | * Checking write permission is tricky, bacuse we don't know if we are | |
3135 | * going to actually need it: O_CREAT opens should work as long as the | |
3136 | * file exists. But checking existence breaks atomicity. The trick is | |
3137 | * to check access and if not granted clear O_CREAT from the flags. | |
3138 | * | |
3139 | * Another problem is returing the "right" error value (e.g. for an | |
3140 | * O_EXCL open we want to return EEXIST not EROFS). | |
3141 | */ | |
3142 | if (open_flag & O_CREAT) { | |
3143 | if (!IS_POSIXACL(dir->d_inode)) | |
3144 | mode &= ~current_umask(); | |
3145 | if (unlikely(!got_write)) { | |
3146 | create_error = -EROFS; | |
3147 | open_flag &= ~O_CREAT; | |
3148 | if (open_flag & (O_EXCL | O_TRUNC)) | |
3149 | goto no_open; | |
3150 | /* No side effects, safe to clear O_CREAT */ | |
3151 | } else { | |
3152 | create_error = may_o_create(&nd->path, dentry, mode); | |
3153 | if (create_error) { | |
3154 | open_flag &= ~O_CREAT; | |
3155 | if (open_flag & O_EXCL) | |
3156 | goto no_open; | |
3157 | } | |
3158 | } | |
3159 | } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && | |
3160 | unlikely(!got_write)) { | |
3161 | /* | |
3162 | * No O_CREATE -> atomicity not a requirement -> fall | |
3163 | * back to lookup + open | |
3164 | */ | |
3165 | goto no_open; | |
d18e9008 MS |
3166 | } |
3167 | ||
6ac08709 | 3168 | if (dir_inode->i_op->atomic_open) { |
1643b43f AV |
3169 | error = atomic_open(nd, dentry, path, file, op, open_flag, |
3170 | mode, opened); | |
3171 | if (unlikely(error == -ENOENT) && create_error) | |
3172 | error = create_error; | |
3173 | return error; | |
d18e9008 | 3174 | } |
54ef4872 | 3175 | |
1643b43f | 3176 | no_open: |
6fbd0714 | 3177 | if (d_in_lookup(dentry)) { |
12fa5e24 AV |
3178 | struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, |
3179 | nd->flags); | |
6fbd0714 | 3180 | d_lookup_done(dentry); |
12fa5e24 AV |
3181 | if (unlikely(res)) { |
3182 | if (IS_ERR(res)) { | |
3183 | error = PTR_ERR(res); | |
3184 | goto out_dput; | |
3185 | } | |
3186 | dput(dentry); | |
3187 | dentry = res; | |
3188 | } | |
54ef4872 MS |
3189 | } |
3190 | ||
d58ffd35 | 3191 | /* Negative dentry, just create the file */ |
1643b43f | 3192 | if (!dentry->d_inode && (open_flag & O_CREAT)) { |
47237687 | 3193 | *opened |= FILE_CREATED; |
ce8644fc | 3194 | audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); |
ce8644fc AV |
3195 | if (!dir_inode->i_op->create) { |
3196 | error = -EACCES; | |
d58ffd35 | 3197 | goto out_dput; |
ce8644fc AV |
3198 | } |
3199 | error = dir_inode->i_op->create(dir_inode, dentry, mode, | |
1643b43f | 3200 | open_flag & O_EXCL); |
d58ffd35 MS |
3201 | if (error) |
3202 | goto out_dput; | |
ce8644fc | 3203 | fsnotify_create(dir_inode, dentry); |
d58ffd35 | 3204 | } |
1643b43f AV |
3205 | if (unlikely(create_error) && !dentry->d_inode) { |
3206 | error = create_error; | |
3207 | goto out_dput; | |
d58ffd35 | 3208 | } |
d18e9008 | 3209 | out_no_open: |
d58ffd35 MS |
3210 | path->dentry = dentry; |
3211 | path->mnt = nd->path.mnt; | |
2675a4eb | 3212 | return 1; |
d58ffd35 MS |
3213 | |
3214 | out_dput: | |
3215 | dput(dentry); | |
2675a4eb | 3216 | return error; |
d58ffd35 MS |
3217 | } |
3218 | ||
31e6b01f | 3219 | /* |
fe2d35ff | 3220 | * Handle the last step of open() |
31e6b01f | 3221 | */ |
896475d5 | 3222 | static int do_last(struct nameidata *nd, |
2675a4eb | 3223 | struct file *file, const struct open_flags *op, |
76ae2a5a | 3224 | int *opened) |
fb1cc555 | 3225 | { |
a1e28038 | 3226 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 3227 | int open_flag = op->open_flag; |
77d660a8 | 3228 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 3229 | bool got_write = false; |
bcda7652 | 3230 | int acc_mode = op->acc_mode; |
254cf582 | 3231 | unsigned seq; |
a1eb3315 | 3232 | struct inode *inode; |
896475d5 | 3233 | struct path path; |
16c2cd71 | 3234 | int error; |
1f36f774 | 3235 | |
c3e380b0 AV |
3236 | nd->flags &= ~LOOKUP_PARENT; |
3237 | nd->flags |= op->intent; | |
3238 | ||
bc77daa7 | 3239 | if (nd->last_type != LAST_NORM) { |
fe2d35ff | 3240 | error = handle_dots(nd, nd->last_type); |
deb106c6 | 3241 | if (unlikely(error)) |
2675a4eb | 3242 | return error; |
e83db167 | 3243 | goto finish_open; |
1f36f774 | 3244 | } |
67ee3ad2 | 3245 | |
ca344a89 | 3246 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
3247 | if (nd->last.name[nd->last.len]) |
3248 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3249 | /* we _can_ be in RCU mode here */ | |
254cf582 | 3250 | error = lookup_fast(nd, &path, &inode, &seq); |
e9742b53 | 3251 | if (likely(error > 0)) |
71574865 MS |
3252 | goto finish_lookup; |
3253 | ||
3254 | if (error < 0) | |
deb106c6 | 3255 | return error; |
71574865 MS |
3256 | |
3257 | BUG_ON(nd->inode != dir->d_inode); | |
6583fe22 | 3258 | BUG_ON(nd->flags & LOOKUP_RCU); |
b6183df7 MS |
3259 | } else { |
3260 | /* create side of things */ | |
3261 | /* | |
3262 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3263 | * has been cleared when we got to the last component we are | |
3264 | * about to look up | |
3265 | */ | |
3266 | error = complete_walk(nd); | |
e8bb73df | 3267 | if (error) |
2675a4eb | 3268 | return error; |
fe2d35ff | 3269 | |
76ae2a5a | 3270 | audit_inode(nd->name, dir, LOOKUP_PARENT); |
b6183df7 | 3271 | /* trailing slashes? */ |
deb106c6 AV |
3272 | if (unlikely(nd->last.name[nd->last.len])) |
3273 | return -EISDIR; | |
b6183df7 | 3274 | } |
a2c36b45 | 3275 | |
9cf843e3 | 3276 | if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
64894cf8 AV |
3277 | error = mnt_want_write(nd->path.mnt); |
3278 | if (!error) | |
3279 | got_write = true; | |
3280 | /* | |
3281 | * do _not_ fail yet - we might not need that or fail with | |
3282 | * a different error; let lookup_open() decide; we'll be | |
3283 | * dropping this one anyway. | |
3284 | */ | |
3285 | } | |
9cf843e3 AV |
3286 | if (open_flag & O_CREAT) |
3287 | inode_lock(dir->d_inode); | |
3288 | else | |
3289 | inode_lock_shared(dir->d_inode); | |
896475d5 | 3290 | error = lookup_open(nd, &path, file, op, got_write, opened); |
9cf843e3 AV |
3291 | if (open_flag & O_CREAT) |
3292 | inode_unlock(dir->d_inode); | |
3293 | else | |
3294 | inode_unlock_shared(dir->d_inode); | |
a1e28038 | 3295 | |
2675a4eb AV |
3296 | if (error <= 0) { |
3297 | if (error) | |
d18e9008 MS |
3298 | goto out; |
3299 | ||
47237687 | 3300 | if ((*opened & FILE_CREATED) || |
496ad9aa | 3301 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 3302 | will_truncate = false; |
d18e9008 | 3303 | |
76ae2a5a | 3304 | audit_inode(nd->name, file->f_path.dentry, 0); |
d18e9008 MS |
3305 | goto opened; |
3306 | } | |
fb1cc555 | 3307 | |
47237687 | 3308 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 3309 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 3310 | open_flag &= ~O_TRUNC; |
77d660a8 | 3311 | will_truncate = false; |
62fb4a15 | 3312 | acc_mode = 0; |
896475d5 | 3313 | path_to_nameidata(&path, nd); |
e83db167 | 3314 | goto finish_open_created; |
fb1cc555 AV |
3315 | } |
3316 | ||
d18e9008 MS |
3317 | /* |
3318 | * If atomic_open() acquired write access it is dropped now due to | |
3319 | * possible mount and symlink following (this might be optimized away if | |
3320 | * necessary...) | |
3321 | */ | |
64894cf8 | 3322 | if (got_write) { |
d18e9008 | 3323 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3324 | got_write = false; |
d18e9008 MS |
3325 | } |
3326 | ||
e6ec03a2 AV |
3327 | error = follow_managed(&path, nd); |
3328 | if (unlikely(error < 0)) | |
3329 | return error; | |
3330 | ||
6583fe22 AV |
3331 | if (unlikely(d_is_negative(path.dentry))) { |
3332 | path_to_nameidata(&path, nd); | |
3333 | return -ENOENT; | |
3334 | } | |
3335 | ||
3336 | /* | |
3337 | * create/update audit record if it already exists. | |
3338 | */ | |
3339 | audit_inode(nd->name, path.dentry, 0); | |
3340 | ||
deb106c6 AV |
3341 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { |
3342 | path_to_nameidata(&path, nd); | |
3343 | return -EEXIST; | |
3344 | } | |
fb1cc555 | 3345 | |
254cf582 | 3346 | seq = 0; /* out of RCU mode, so the value doesn't matter */ |
d4565649 | 3347 | inode = d_backing_inode(path.dentry); |
766c4cbf | 3348 | finish_lookup: |
8f64fb1c | 3349 | error = step_into(nd, &path, 0, inode, seq); |
deb106c6 | 3350 | if (unlikely(error)) |
d63ff28f | 3351 | return error; |
bc77daa7 | 3352 | finish_open: |
8f64fb1c | 3353 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
a3fbbde7 | 3354 | error = complete_walk(nd); |
fac7d191 | 3355 | if (error) |
2675a4eb | 3356 | return error; |
76ae2a5a | 3357 | audit_inode(nd->name, nd->path.dentry, 0); |
fb1cc555 | 3358 | error = -EISDIR; |
44b1d530 | 3359 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3360 | goto out; |
af2f5542 | 3361 | error = -ENOTDIR; |
44b1d530 | 3362 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3363 | goto out; |
4bbcbd3b | 3364 | if (!d_is_reg(nd->path.dentry)) |
77d660a8 | 3365 | will_truncate = false; |
6c0d46c4 | 3366 | |
0f9d1a10 AV |
3367 | if (will_truncate) { |
3368 | error = mnt_want_write(nd->path.mnt); | |
3369 | if (error) | |
2675a4eb | 3370 | goto out; |
64894cf8 | 3371 | got_write = true; |
0f9d1a10 | 3372 | } |
e83db167 | 3373 | finish_open_created: |
6ac08709 AV |
3374 | error = may_open(&nd->path, acc_mode, open_flag); |
3375 | if (error) | |
3376 | goto out; | |
4aa7c634 MS |
3377 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ |
3378 | error = vfs_open(&nd->path, file, current_cred()); | |
fac7d191 | 3379 | if (error) |
015c3bbc | 3380 | goto out; |
fac7d191 | 3381 | *opened |= FILE_OPENED; |
a8277b9b | 3382 | opened: |
2675a4eb | 3383 | error = open_check_o_direct(file); |
fe9ec829 AV |
3384 | if (!error) |
3385 | error = ima_file_check(file, op->acc_mode, *opened); | |
3386 | if (!error && will_truncate) | |
2675a4eb | 3387 | error = handle_truncate(file); |
ca344a89 | 3388 | out: |
fe9ec829 AV |
3389 | if (unlikely(error) && (*opened & FILE_OPENED)) |
3390 | fput(file); | |
c80567c8 AV |
3391 | if (unlikely(error > 0)) { |
3392 | WARN_ON(1); | |
3393 | error = -EINVAL; | |
3394 | } | |
64894cf8 | 3395 | if (got_write) |
0f9d1a10 | 3396 | mnt_drop_write(nd->path.mnt); |
2675a4eb | 3397 | return error; |
fb1cc555 AV |
3398 | } |
3399 | ||
af7bd4dc AG |
3400 | struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) |
3401 | { | |
af7bd4dc AG |
3402 | struct dentry *child = NULL; |
3403 | struct inode *dir = dentry->d_inode; | |
3404 | struct inode *inode; | |
3405 | int error; | |
3406 | ||
3407 | /* we want directory to be writable */ | |
3408 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
3409 | if (error) | |
3410 | goto out_err; | |
3411 | error = -EOPNOTSUPP; | |
3412 | if (!dir->i_op->tmpfile) | |
3413 | goto out_err; | |
3414 | error = -ENOMEM; | |
cdf01226 | 3415 | child = d_alloc(dentry, &slash_name); |
af7bd4dc AG |
3416 | if (unlikely(!child)) |
3417 | goto out_err; | |
3418 | error = dir->i_op->tmpfile(dir, child, mode); | |
3419 | if (error) | |
3420 | goto out_err; | |
3421 | error = -ENOENT; | |
3422 | inode = child->d_inode; | |
3423 | if (unlikely(!inode)) | |
3424 | goto out_err; | |
3425 | if (!(open_flag & O_EXCL)) { | |
3426 | spin_lock(&inode->i_lock); | |
3427 | inode->i_state |= I_LINKABLE; | |
3428 | spin_unlock(&inode->i_lock); | |
3429 | } | |
3430 | return child; | |
3431 | ||
3432 | out_err: | |
3433 | dput(child); | |
3434 | return ERR_PTR(error); | |
3435 | } | |
3436 | EXPORT_SYMBOL(vfs_tmpfile); | |
3437 | ||
c8a53ee5 | 3438 | static int do_tmpfile(struct nameidata *nd, unsigned flags, |
60545d0d AV |
3439 | const struct open_flags *op, |
3440 | struct file *file, int *opened) | |
3441 | { | |
625b6d10 | 3442 | struct dentry *child; |
625b6d10 | 3443 | struct path path; |
c8a53ee5 | 3444 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); |
60545d0d AV |
3445 | if (unlikely(error)) |
3446 | return error; | |
625b6d10 | 3447 | error = mnt_want_write(path.mnt); |
60545d0d AV |
3448 | if (unlikely(error)) |
3449 | goto out; | |
af7bd4dc AG |
3450 | child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); |
3451 | error = PTR_ERR(child); | |
684e73be | 3452 | if (IS_ERR(child)) |
60545d0d | 3453 | goto out2; |
625b6d10 AV |
3454 | dput(path.dentry); |
3455 | path.dentry = child; | |
c8a53ee5 | 3456 | audit_inode(nd->name, child, 0); |
69a91c23 | 3457 | /* Don't check for other permissions, the inode was just created */ |
62fb4a15 | 3458 | error = may_open(&path, 0, op->open_flag); |
60545d0d AV |
3459 | if (error) |
3460 | goto out2; | |
625b6d10 AV |
3461 | file->f_path.mnt = path.mnt; |
3462 | error = finish_open(file, child, NULL, opened); | |
60545d0d AV |
3463 | if (error) |
3464 | goto out2; | |
3465 | error = open_check_o_direct(file); | |
af7bd4dc | 3466 | if (error) |
60545d0d AV |
3467 | fput(file); |
3468 | out2: | |
625b6d10 | 3469 | mnt_drop_write(path.mnt); |
60545d0d | 3470 | out: |
625b6d10 | 3471 | path_put(&path); |
60545d0d AV |
3472 | return error; |
3473 | } | |
3474 | ||
6ac08709 AV |
3475 | static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) |
3476 | { | |
3477 | struct path path; | |
3478 | int error = path_lookupat(nd, flags, &path); | |
3479 | if (!error) { | |
3480 | audit_inode(nd->name, path.dentry, 0); | |
3481 | error = vfs_open(&path, file, current_cred()); | |
3482 | path_put(&path); | |
3483 | } | |
3484 | return error; | |
3485 | } | |
3486 | ||
c8a53ee5 AV |
3487 | static struct file *path_openat(struct nameidata *nd, |
3488 | const struct open_flags *op, unsigned flags) | |
1da177e4 | 3489 | { |
368ee9ba | 3490 | const char *s; |
30d90494 | 3491 | struct file *file; |
47237687 | 3492 | int opened = 0; |
13aab428 | 3493 | int error; |
31e6b01f | 3494 | |
30d90494 | 3495 | file = get_empty_filp(); |
1afc99be AV |
3496 | if (IS_ERR(file)) |
3497 | return file; | |
31e6b01f | 3498 | |
30d90494 | 3499 | file->f_flags = op->open_flag; |
31e6b01f | 3500 | |
bb458c64 | 3501 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
c8a53ee5 | 3502 | error = do_tmpfile(nd, flags, op, file, &opened); |
f15133df | 3503 | goto out2; |
60545d0d AV |
3504 | } |
3505 | ||
6ac08709 AV |
3506 | if (unlikely(file->f_flags & O_PATH)) { |
3507 | error = do_o_path(nd, flags, file); | |
3508 | if (!error) | |
3509 | opened |= FILE_OPENED; | |
3510 | goto out2; | |
3511 | } | |
3512 | ||
c8a53ee5 | 3513 | s = path_init(nd, flags); |
368ee9ba AV |
3514 | if (IS_ERR(s)) { |
3515 | put_filp(file); | |
3516 | return ERR_CAST(s); | |
3517 | } | |
3bdba28b | 3518 | while (!(error = link_path_walk(s, nd)) && |
76ae2a5a | 3519 | (error = do_last(nd, file, op, &opened)) > 0) { |
73d049a4 | 3520 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
3bdba28b AV |
3521 | s = trailing_symlink(nd); |
3522 | if (IS_ERR(s)) { | |
3523 | error = PTR_ERR(s); | |
2675a4eb | 3524 | break; |
3bdba28b | 3525 | } |
806b681c | 3526 | } |
deb106c6 | 3527 | terminate_walk(nd); |
f15133df | 3528 | out2: |
2675a4eb AV |
3529 | if (!(opened & FILE_OPENED)) { |
3530 | BUG_ON(!error); | |
30d90494 | 3531 | put_filp(file); |
16b1c1cd | 3532 | } |
2675a4eb AV |
3533 | if (unlikely(error)) { |
3534 | if (error == -EOPENSTALE) { | |
3535 | if (flags & LOOKUP_RCU) | |
3536 | error = -ECHILD; | |
3537 | else | |
3538 | error = -ESTALE; | |
3539 | } | |
3540 | file = ERR_PTR(error); | |
3541 | } | |
3542 | return file; | |
1da177e4 LT |
3543 | } |
3544 | ||
669abf4e | 3545 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3546 | const struct open_flags *op) |
13aab428 | 3547 | { |
9883d185 | 3548 | struct nameidata nd; |
f9652e10 | 3549 | int flags = op->lookup_flags; |
13aab428 AV |
3550 | struct file *filp; |
3551 | ||
9883d185 | 3552 | set_nameidata(&nd, dfd, pathname); |
c8a53ee5 | 3553 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); |
13aab428 | 3554 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3555 | filp = path_openat(&nd, op, flags); |
13aab428 | 3556 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3557 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3558 | restore_nameidata(); |
13aab428 AV |
3559 | return filp; |
3560 | } | |
3561 | ||
73d049a4 | 3562 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3563 | const char *name, const struct open_flags *op) |
73d049a4 | 3564 | { |
9883d185 | 3565 | struct nameidata nd; |
73d049a4 | 3566 | struct file *file; |
51689104 | 3567 | struct filename *filename; |
f9652e10 | 3568 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3569 | |
3570 | nd.root.mnt = mnt; | |
3571 | nd.root.dentry = dentry; | |
3572 | ||
b18825a7 | 3573 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3574 | return ERR_PTR(-ELOOP); |
3575 | ||
51689104 | 3576 | filename = getname_kernel(name); |
a1c83681 | 3577 | if (IS_ERR(filename)) |
51689104 PM |
3578 | return ERR_CAST(filename); |
3579 | ||
9883d185 | 3580 | set_nameidata(&nd, -1, filename); |
c8a53ee5 | 3581 | file = path_openat(&nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3582 | if (unlikely(file == ERR_PTR(-ECHILD))) |
c8a53ee5 | 3583 | file = path_openat(&nd, op, flags); |
73d049a4 | 3584 | if (unlikely(file == ERR_PTR(-ESTALE))) |
c8a53ee5 | 3585 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); |
9883d185 | 3586 | restore_nameidata(); |
51689104 | 3587 | putname(filename); |
73d049a4 AV |
3588 | return file; |
3589 | } | |
3590 | ||
fa14a0b8 | 3591 | static struct dentry *filename_create(int dfd, struct filename *name, |
1ac12b4b | 3592 | struct path *path, unsigned int lookup_flags) |
1da177e4 | 3593 | { |
c663e5d8 | 3594 | struct dentry *dentry = ERR_PTR(-EEXIST); |
391172c4 AV |
3595 | struct qstr last; |
3596 | int type; | |
c30dabfe | 3597 | int err2; |
1ac12b4b JL |
3598 | int error; |
3599 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3600 | ||
3601 | /* | |
3602 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3603 | * other flags passed in are ignored! | |
3604 | */ | |
3605 | lookup_flags &= LOOKUP_REVAL; | |
3606 | ||
5c31b6ce AV |
3607 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); |
3608 | if (IS_ERR(name)) | |
3609 | return ERR_CAST(name); | |
1da177e4 | 3610 | |
c663e5d8 CH |
3611 | /* |
3612 | * Yucky last component or no last component at all? | |
3613 | * (foo/., foo/.., /////) | |
3614 | */ | |
5c31b6ce | 3615 | if (unlikely(type != LAST_NORM)) |
ed75e95d | 3616 | goto out; |
c663e5d8 | 3617 | |
c30dabfe | 3618 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
391172c4 | 3619 | err2 = mnt_want_write(path->mnt); |
c663e5d8 CH |
3620 | /* |
3621 | * Do the final lookup. | |
3622 | */ | |
391172c4 | 3623 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
5955102c | 3624 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); |
391172c4 | 3625 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); |
1da177e4 | 3626 | if (IS_ERR(dentry)) |
a8104a9f | 3627 | goto unlock; |
c663e5d8 | 3628 | |
a8104a9f | 3629 | error = -EEXIST; |
b18825a7 | 3630 | if (d_is_positive(dentry)) |
a8104a9f | 3631 | goto fail; |
b18825a7 | 3632 | |
c663e5d8 CH |
3633 | /* |
3634 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3635 | * From the vfs_mknod() POV we just have a negative dentry - | |
3636 | * all is fine. Let's be bastards - you had / on the end, you've | |
3637 | * been asking for (non-existent) directory. -ENOENT for you. | |
3638 | */ | |
391172c4 | 3639 | if (unlikely(!is_dir && last.name[last.len])) { |
a8104a9f | 3640 | error = -ENOENT; |
ed75e95d | 3641 | goto fail; |
e9baf6e5 | 3642 | } |
c30dabfe JK |
3643 | if (unlikely(err2)) { |
3644 | error = err2; | |
a8104a9f | 3645 | goto fail; |
c30dabfe | 3646 | } |
181c37b6 | 3647 | putname(name); |
1da177e4 | 3648 | return dentry; |
1da177e4 | 3649 | fail: |
a8104a9f AV |
3650 | dput(dentry); |
3651 | dentry = ERR_PTR(error); | |
3652 | unlock: | |
5955102c | 3653 | inode_unlock(path->dentry->d_inode); |
c30dabfe | 3654 | if (!err2) |
391172c4 | 3655 | mnt_drop_write(path->mnt); |
ed75e95d | 3656 | out: |
391172c4 | 3657 | path_put(path); |
181c37b6 | 3658 | putname(name); |
1da177e4 LT |
3659 | return dentry; |
3660 | } | |
fa14a0b8 AV |
3661 | |
3662 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3663 | struct path *path, unsigned int lookup_flags) | |
3664 | { | |
181c37b6 AV |
3665 | return filename_create(dfd, getname_kernel(pathname), |
3666 | path, lookup_flags); | |
fa14a0b8 | 3667 | } |
dae6ad8f AV |
3668 | EXPORT_SYMBOL(kern_path_create); |
3669 | ||
921a1650 AV |
3670 | void done_path_create(struct path *path, struct dentry *dentry) |
3671 | { | |
3672 | dput(dentry); | |
5955102c | 3673 | inode_unlock(path->dentry->d_inode); |
a8104a9f | 3674 | mnt_drop_write(path->mnt); |
921a1650 AV |
3675 | path_put(path); |
3676 | } | |
3677 | EXPORT_SYMBOL(done_path_create); | |
3678 | ||
520ae687 | 3679 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, |
1ac12b4b | 3680 | struct path *path, unsigned int lookup_flags) |
dae6ad8f | 3681 | { |
181c37b6 | 3682 | return filename_create(dfd, getname(pathname), path, lookup_flags); |
dae6ad8f AV |
3683 | } |
3684 | EXPORT_SYMBOL(user_path_create); | |
3685 | ||
1a67aafb | 3686 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3687 | { |
a95164d9 | 3688 | int error = may_create(dir, dentry); |
1da177e4 LT |
3689 | |
3690 | if (error) | |
3691 | return error; | |
3692 | ||
975d6b39 | 3693 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3694 | return -EPERM; |
3695 | ||
acfa4380 | 3696 | if (!dir->i_op->mknod) |
1da177e4 LT |
3697 | return -EPERM; |
3698 | ||
08ce5f16 SH |
3699 | error = devcgroup_inode_mknod(mode, dev); |
3700 | if (error) | |
3701 | return error; | |
3702 | ||
1da177e4 LT |
3703 | error = security_inode_mknod(dir, dentry, mode, dev); |
3704 | if (error) | |
3705 | return error; | |
3706 | ||
1da177e4 | 3707 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3708 | if (!error) |
f38aa942 | 3709 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3710 | return error; |
3711 | } | |
4d359507 | 3712 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3713 | |
f69aac00 | 3714 | static int may_mknod(umode_t mode) |
463c3197 DH |
3715 | { |
3716 | switch (mode & S_IFMT) { | |
3717 | case S_IFREG: | |
3718 | case S_IFCHR: | |
3719 | case S_IFBLK: | |
3720 | case S_IFIFO: | |
3721 | case S_IFSOCK: | |
3722 | case 0: /* zero mode translates to S_IFREG */ | |
3723 | return 0; | |
3724 | case S_IFDIR: | |
3725 | return -EPERM; | |
3726 | default: | |
3727 | return -EINVAL; | |
3728 | } | |
3729 | } | |
3730 | ||
8208a22b | 3731 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3732 | unsigned, dev) |
1da177e4 | 3733 | { |
2ad94ae6 | 3734 | struct dentry *dentry; |
dae6ad8f AV |
3735 | struct path path; |
3736 | int error; | |
972567f1 | 3737 | unsigned int lookup_flags = 0; |
1da177e4 | 3738 | |
8e4bfca1 AV |
3739 | error = may_mknod(mode); |
3740 | if (error) | |
3741 | return error; | |
972567f1 JL |
3742 | retry: |
3743 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3744 | if (IS_ERR(dentry)) |
3745 | return PTR_ERR(dentry); | |
2ad94ae6 | 3746 | |
dae6ad8f | 3747 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3748 | mode &= ~current_umask(); |
dae6ad8f | 3749 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3750 | if (error) |
a8104a9f | 3751 | goto out; |
463c3197 | 3752 | switch (mode & S_IFMT) { |
1da177e4 | 3753 | case 0: case S_IFREG: |
312b63fb | 3754 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
05d1a717 MZ |
3755 | if (!error) |
3756 | ima_post_path_mknod(dentry); | |
1da177e4 LT |
3757 | break; |
3758 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3759 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3760 | new_decode_dev(dev)); |
3761 | break; | |
3762 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3763 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3764 | break; |
1da177e4 | 3765 | } |
a8104a9f | 3766 | out: |
921a1650 | 3767 | done_path_create(&path, dentry); |
972567f1 JL |
3768 | if (retry_estale(error, lookup_flags)) { |
3769 | lookup_flags |= LOOKUP_REVAL; | |
3770 | goto retry; | |
3771 | } | |
1da177e4 LT |
3772 | return error; |
3773 | } | |
3774 | ||
8208a22b | 3775 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3776 | { |
3777 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3778 | } | |
3779 | ||
18bb1db3 | 3780 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3781 | { |
a95164d9 | 3782 | int error = may_create(dir, dentry); |
8de52778 | 3783 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3784 | |
3785 | if (error) | |
3786 | return error; | |
3787 | ||
acfa4380 | 3788 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3789 | return -EPERM; |
3790 | ||
3791 | mode &= (S_IRWXUGO|S_ISVTX); | |
3792 | error = security_inode_mkdir(dir, dentry, mode); | |
3793 | if (error) | |
3794 | return error; | |
3795 | ||
8de52778 AV |
3796 | if (max_links && dir->i_nlink >= max_links) |
3797 | return -EMLINK; | |
3798 | ||
1da177e4 | 3799 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3800 | if (!error) |
f38aa942 | 3801 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3802 | return error; |
3803 | } | |
4d359507 | 3804 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3805 | |
a218d0fd | 3806 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3807 | { |
6902d925 | 3808 | struct dentry *dentry; |
dae6ad8f AV |
3809 | struct path path; |
3810 | int error; | |
b76d8b82 | 3811 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3812 | |
b76d8b82 JL |
3813 | retry: |
3814 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3815 | if (IS_ERR(dentry)) |
dae6ad8f | 3816 | return PTR_ERR(dentry); |
1da177e4 | 3817 | |
dae6ad8f | 3818 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3819 | mode &= ~current_umask(); |
dae6ad8f | 3820 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3821 | if (!error) |
3822 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3823 | done_path_create(&path, dentry); |
b76d8b82 JL |
3824 | if (retry_estale(error, lookup_flags)) { |
3825 | lookup_flags |= LOOKUP_REVAL; | |
3826 | goto retry; | |
3827 | } | |
1da177e4 LT |
3828 | return error; |
3829 | } | |
3830 | ||
a218d0fd | 3831 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3832 | { |
3833 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3834 | } | |
3835 | ||
1da177e4 LT |
3836 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
3837 | { | |
3838 | int error = may_delete(dir, dentry, 1); | |
3839 | ||
3840 | if (error) | |
3841 | return error; | |
3842 | ||
acfa4380 | 3843 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3844 | return -EPERM; |
3845 | ||
1d2ef590 | 3846 | dget(dentry); |
5955102c | 3847 | inode_lock(dentry->d_inode); |
912dbc15 SW |
3848 | |
3849 | error = -EBUSY; | |
7af1364f | 3850 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3851 | goto out; |
3852 | ||
3853 | error = security_inode_rmdir(dir, dentry); | |
3854 | if (error) | |
3855 | goto out; | |
3856 | ||
3cebde24 | 3857 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3858 | error = dir->i_op->rmdir(dir, dentry); |
3859 | if (error) | |
3860 | goto out; | |
3861 | ||
3862 | dentry->d_inode->i_flags |= S_DEAD; | |
3863 | dont_mount(dentry); | |
8ed936b5 | 3864 | detach_mounts(dentry); |
912dbc15 SW |
3865 | |
3866 | out: | |
5955102c | 3867 | inode_unlock(dentry->d_inode); |
1d2ef590 | 3868 | dput(dentry); |
912dbc15 | 3869 | if (!error) |
1da177e4 | 3870 | d_delete(dentry); |
1da177e4 LT |
3871 | return error; |
3872 | } | |
4d359507 | 3873 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3874 | |
5590ff0d | 3875 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3876 | { |
3877 | int error = 0; | |
91a27b2a | 3878 | struct filename *name; |
1da177e4 | 3879 | struct dentry *dentry; |
f5beed75 AV |
3880 | struct path path; |
3881 | struct qstr last; | |
3882 | int type; | |
c6ee9206 JL |
3883 | unsigned int lookup_flags = 0; |
3884 | retry: | |
c1d4dd27 AV |
3885 | name = filename_parentat(dfd, getname(pathname), lookup_flags, |
3886 | &path, &last, &type); | |
91a27b2a JL |
3887 | if (IS_ERR(name)) |
3888 | return PTR_ERR(name); | |
1da177e4 | 3889 | |
f5beed75 | 3890 | switch (type) { |
0612d9fb OH |
3891 | case LAST_DOTDOT: |
3892 | error = -ENOTEMPTY; | |
3893 | goto exit1; | |
3894 | case LAST_DOT: | |
3895 | error = -EINVAL; | |
3896 | goto exit1; | |
3897 | case LAST_ROOT: | |
3898 | error = -EBUSY; | |
3899 | goto exit1; | |
1da177e4 | 3900 | } |
0612d9fb | 3901 | |
f5beed75 | 3902 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
3903 | if (error) |
3904 | goto exit1; | |
0612d9fb | 3905 | |
5955102c | 3906 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 3907 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 | 3908 | error = PTR_ERR(dentry); |
6902d925 DH |
3909 | if (IS_ERR(dentry)) |
3910 | goto exit2; | |
e6bc45d6 TT |
3911 | if (!dentry->d_inode) { |
3912 | error = -ENOENT; | |
3913 | goto exit3; | |
3914 | } | |
f5beed75 | 3915 | error = security_path_rmdir(&path, dentry); |
be6d3e56 | 3916 | if (error) |
c30dabfe | 3917 | goto exit3; |
f5beed75 | 3918 | error = vfs_rmdir(path.dentry->d_inode, dentry); |
0622753b | 3919 | exit3: |
6902d925 DH |
3920 | dput(dentry); |
3921 | exit2: | |
5955102c | 3922 | inode_unlock(path.dentry->d_inode); |
f5beed75 | 3923 | mnt_drop_write(path.mnt); |
1da177e4 | 3924 | exit1: |
f5beed75 | 3925 | path_put(&path); |
1da177e4 | 3926 | putname(name); |
c6ee9206 JL |
3927 | if (retry_estale(error, lookup_flags)) { |
3928 | lookup_flags |= LOOKUP_REVAL; | |
3929 | goto retry; | |
3930 | } | |
1da177e4 LT |
3931 | return error; |
3932 | } | |
3933 | ||
3cdad428 | 3934 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3935 | { |
3936 | return do_rmdir(AT_FDCWD, pathname); | |
3937 | } | |
3938 | ||
b21996e3 BF |
3939 | /** |
3940 | * vfs_unlink - unlink a filesystem object | |
3941 | * @dir: parent directory | |
3942 | * @dentry: victim | |
3943 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3944 | * | |
3945 | * The caller must hold dir->i_mutex. | |
3946 | * | |
3947 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3948 | * return a reference to the inode in delegated_inode. The caller | |
3949 | * should then break the delegation on that inode and retry. Because | |
3950 | * breaking a delegation may take a long time, the caller should drop | |
3951 | * dir->i_mutex before doing so. | |
3952 | * | |
3953 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3954 | * be appropriate for callers that expect the underlying filesystem not | |
3955 | * to be NFS exported. | |
3956 | */ | |
3957 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3958 | { |
9accbb97 | 3959 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3960 | int error = may_delete(dir, dentry, 0); |
3961 | ||
3962 | if (error) | |
3963 | return error; | |
3964 | ||
acfa4380 | 3965 | if (!dir->i_op->unlink) |
1da177e4 LT |
3966 | return -EPERM; |
3967 | ||
5955102c | 3968 | inode_lock(target); |
8ed936b5 | 3969 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3970 | error = -EBUSY; |
3971 | else { | |
3972 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3973 | if (!error) { |
5a14696c BF |
3974 | error = try_break_deleg(target, delegated_inode); |
3975 | if (error) | |
b21996e3 | 3976 | goto out; |
1da177e4 | 3977 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3978 | if (!error) { |
d83c49f3 | 3979 | dont_mount(dentry); |
8ed936b5 EB |
3980 | detach_mounts(dentry); |
3981 | } | |
bec1052e | 3982 | } |
1da177e4 | 3983 | } |
b21996e3 | 3984 | out: |
5955102c | 3985 | inode_unlock(target); |
1da177e4 LT |
3986 | |
3987 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3988 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3989 | fsnotify_link_count(target); |
e234f35c | 3990 | d_delete(dentry); |
1da177e4 | 3991 | } |
0eeca283 | 3992 | |
1da177e4 LT |
3993 | return error; |
3994 | } | |
4d359507 | 3995 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3996 | |
3997 | /* | |
3998 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3999 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
4000 | * writeout happening, and we don't want to prevent access to the directory |
4001 | * while waiting on the I/O. | |
4002 | */ | |
da2f1362 | 4003 | long do_unlinkat(int dfd, struct filename *name) |
1da177e4 | 4004 | { |
2ad94ae6 | 4005 | int error; |
1da177e4 | 4006 | struct dentry *dentry; |
f5beed75 AV |
4007 | struct path path; |
4008 | struct qstr last; | |
4009 | int type; | |
1da177e4 | 4010 | struct inode *inode = NULL; |
b21996e3 | 4011 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
4012 | unsigned int lookup_flags = 0; |
4013 | retry: | |
da2f1362 | 4014 | name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); |
91a27b2a JL |
4015 | if (IS_ERR(name)) |
4016 | return PTR_ERR(name); | |
2ad94ae6 | 4017 | |
1da177e4 | 4018 | error = -EISDIR; |
f5beed75 | 4019 | if (type != LAST_NORM) |
1da177e4 | 4020 | goto exit1; |
0612d9fb | 4021 | |
f5beed75 | 4022 | error = mnt_want_write(path.mnt); |
c30dabfe JK |
4023 | if (error) |
4024 | goto exit1; | |
b21996e3 | 4025 | retry_deleg: |
5955102c | 4026 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); |
f5beed75 | 4027 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); |
1da177e4 LT |
4028 | error = PTR_ERR(dentry); |
4029 | if (!IS_ERR(dentry)) { | |
4030 | /* Why not before? Because we want correct error value */ | |
f5beed75 | 4031 | if (last.name[last.len]) |
50338b88 | 4032 | goto slashes; |
1da177e4 | 4033 | inode = dentry->d_inode; |
b18825a7 | 4034 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
4035 | goto slashes; |
4036 | ihold(inode); | |
f5beed75 | 4037 | error = security_path_unlink(&path, dentry); |
be6d3e56 | 4038 | if (error) |
c30dabfe | 4039 | goto exit2; |
f5beed75 | 4040 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 4041 | exit2: |
1da177e4 LT |
4042 | dput(dentry); |
4043 | } | |
5955102c | 4044 | inode_unlock(path.dentry->d_inode); |
1da177e4 LT |
4045 | if (inode) |
4046 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
4047 | inode = NULL; |
4048 | if (delegated_inode) { | |
5a14696c | 4049 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
4050 | if (!error) |
4051 | goto retry_deleg; | |
4052 | } | |
f5beed75 | 4053 | mnt_drop_write(path.mnt); |
1da177e4 | 4054 | exit1: |
f5beed75 | 4055 | path_put(&path); |
5d18f813 JL |
4056 | if (retry_estale(error, lookup_flags)) { |
4057 | lookup_flags |= LOOKUP_REVAL; | |
4058 | inode = NULL; | |
4059 | goto retry; | |
4060 | } | |
da2f1362 | 4061 | putname(name); |
1da177e4 LT |
4062 | return error; |
4063 | ||
4064 | slashes: | |
b18825a7 DH |
4065 | if (d_is_negative(dentry)) |
4066 | error = -ENOENT; | |
44b1d530 | 4067 | else if (d_is_dir(dentry)) |
b18825a7 DH |
4068 | error = -EISDIR; |
4069 | else | |
4070 | error = -ENOTDIR; | |
1da177e4 LT |
4071 | goto exit2; |
4072 | } | |
4073 | ||
2e4d0924 | 4074 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
4075 | { |
4076 | if ((flag & ~AT_REMOVEDIR) != 0) | |
4077 | return -EINVAL; | |
4078 | ||
4079 | if (flag & AT_REMOVEDIR) | |
4080 | return do_rmdir(dfd, pathname); | |
4081 | ||
da2f1362 | 4082 | return do_unlinkat(dfd, getname(pathname)); |
5590ff0d UD |
4083 | } |
4084 | ||
3480b257 | 4085 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d | 4086 | { |
da2f1362 | 4087 | return do_unlinkat(AT_FDCWD, getname(pathname)); |
5590ff0d UD |
4088 | } |
4089 | ||
db2e747b | 4090 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 4091 | { |
a95164d9 | 4092 | int error = may_create(dir, dentry); |
1da177e4 LT |
4093 | |
4094 | if (error) | |
4095 | return error; | |
4096 | ||
acfa4380 | 4097 | if (!dir->i_op->symlink) |
1da177e4 LT |
4098 | return -EPERM; |
4099 | ||
4100 | error = security_inode_symlink(dir, dentry, oldname); | |
4101 | if (error) | |
4102 | return error; | |
4103 | ||
1da177e4 | 4104 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 4105 | if (!error) |
f38aa942 | 4106 | fsnotify_create(dir, dentry); |
1da177e4 LT |
4107 | return error; |
4108 | } | |
4d359507 | 4109 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 4110 | |
2e4d0924 HC |
4111 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
4112 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4113 | { |
2ad94ae6 | 4114 | int error; |
91a27b2a | 4115 | struct filename *from; |
6902d925 | 4116 | struct dentry *dentry; |
dae6ad8f | 4117 | struct path path; |
f46d3567 | 4118 | unsigned int lookup_flags = 0; |
1da177e4 LT |
4119 | |
4120 | from = getname(oldname); | |
2ad94ae6 | 4121 | if (IS_ERR(from)) |
1da177e4 | 4122 | return PTR_ERR(from); |
f46d3567 JL |
4123 | retry: |
4124 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
4125 | error = PTR_ERR(dentry); |
4126 | if (IS_ERR(dentry)) | |
dae6ad8f | 4127 | goto out_putname; |
6902d925 | 4128 | |
91a27b2a | 4129 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 4130 | if (!error) |
91a27b2a | 4131 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 4132 | done_path_create(&path, dentry); |
f46d3567 JL |
4133 | if (retry_estale(error, lookup_flags)) { |
4134 | lookup_flags |= LOOKUP_REVAL; | |
4135 | goto retry; | |
4136 | } | |
6902d925 | 4137 | out_putname: |
1da177e4 LT |
4138 | putname(from); |
4139 | return error; | |
4140 | } | |
4141 | ||
3480b257 | 4142 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4143 | { |
4144 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4145 | } | |
4146 | ||
146a8595 BF |
4147 | /** |
4148 | * vfs_link - create a new link | |
4149 | * @old_dentry: object to be linked | |
4150 | * @dir: new parent | |
4151 | * @new_dentry: where to create the new link | |
4152 | * @delegated_inode: returns inode needing a delegation break | |
4153 | * | |
4154 | * The caller must hold dir->i_mutex | |
4155 | * | |
4156 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4157 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4158 | * inode in delegated_inode. The caller should then break the delegation | |
4159 | * and retry. Because breaking a delegation may take a long time, the | |
4160 | * caller should drop the i_mutex before doing so. | |
4161 | * | |
4162 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4163 | * be appropriate for callers that expect the underlying filesystem not | |
4164 | * to be NFS exported. | |
4165 | */ | |
4166 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
4167 | { |
4168 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 4169 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
4170 | int error; |
4171 | ||
4172 | if (!inode) | |
4173 | return -ENOENT; | |
4174 | ||
a95164d9 | 4175 | error = may_create(dir, new_dentry); |
1da177e4 LT |
4176 | if (error) |
4177 | return error; | |
4178 | ||
4179 | if (dir->i_sb != inode->i_sb) | |
4180 | return -EXDEV; | |
4181 | ||
4182 | /* | |
4183 | * A link to an append-only or immutable file cannot be created. | |
4184 | */ | |
4185 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4186 | return -EPERM; | |
0bd23d09 EB |
4187 | /* |
4188 | * Updating the link count will likely cause i_uid and i_gid to | |
4189 | * be writen back improperly if their true value is unknown to | |
4190 | * the vfs. | |
4191 | */ | |
4192 | if (HAS_UNMAPPED_ID(inode)) | |
4193 | return -EPERM; | |
acfa4380 | 4194 | if (!dir->i_op->link) |
1da177e4 | 4195 | return -EPERM; |
7e79eedb | 4196 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
4197 | return -EPERM; |
4198 | ||
4199 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4200 | if (error) | |
4201 | return error; | |
4202 | ||
5955102c | 4203 | inode_lock(inode); |
aae8a97d | 4204 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 4205 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 4206 | error = -ENOENT; |
8de52778 AV |
4207 | else if (max_links && inode->i_nlink >= max_links) |
4208 | error = -EMLINK; | |
146a8595 BF |
4209 | else { |
4210 | error = try_break_deleg(inode, delegated_inode); | |
4211 | if (!error) | |
4212 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4213 | } | |
f4e0c30c AV |
4214 | |
4215 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4216 | spin_lock(&inode->i_lock); | |
4217 | inode->i_state &= ~I_LINKABLE; | |
4218 | spin_unlock(&inode->i_lock); | |
4219 | } | |
5955102c | 4220 | inode_unlock(inode); |
e31e14ec | 4221 | if (!error) |
7e79eedb | 4222 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
4223 | return error; |
4224 | } | |
4d359507 | 4225 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
4226 | |
4227 | /* | |
4228 | * Hardlinks are often used in delicate situations. We avoid | |
4229 | * security-related surprises by not following symlinks on the | |
4230 | * newname. --KAB | |
4231 | * | |
4232 | * We don't follow them on the oldname either to be compatible | |
4233 | * with linux 2.0, and to avoid hard-linking to directories | |
4234 | * and other special files. --ADM | |
4235 | */ | |
2e4d0924 HC |
4236 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
4237 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
4238 | { |
4239 | struct dentry *new_dentry; | |
dae6ad8f | 4240 | struct path old_path, new_path; |
146a8595 | 4241 | struct inode *delegated_inode = NULL; |
11a7b371 | 4242 | int how = 0; |
1da177e4 | 4243 | int error; |
1da177e4 | 4244 | |
11a7b371 | 4245 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 4246 | return -EINVAL; |
11a7b371 | 4247 | /* |
f0cc6ffb LT |
4248 | * To use null names we require CAP_DAC_READ_SEARCH |
4249 | * This ensures that not everyone will be able to create | |
4250 | * handlink using the passed filedescriptor. | |
11a7b371 | 4251 | */ |
f0cc6ffb LT |
4252 | if (flags & AT_EMPTY_PATH) { |
4253 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4254 | return -ENOENT; | |
11a7b371 | 4255 | how = LOOKUP_EMPTY; |
f0cc6ffb | 4256 | } |
11a7b371 AK |
4257 | |
4258 | if (flags & AT_SYMLINK_FOLLOW) | |
4259 | how |= LOOKUP_FOLLOW; | |
442e31ca | 4260 | retry: |
11a7b371 | 4261 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 4262 | if (error) |
2ad94ae6 AV |
4263 | return error; |
4264 | ||
442e31ca JL |
4265 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4266 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4267 | error = PTR_ERR(new_dentry); |
6902d925 | 4268 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4269 | goto out; |
4270 | ||
4271 | error = -EXDEV; | |
4272 | if (old_path.mnt != new_path.mnt) | |
4273 | goto out_dput; | |
800179c9 KC |
4274 | error = may_linkat(&old_path); |
4275 | if (unlikely(error)) | |
4276 | goto out_dput; | |
dae6ad8f | 4277 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4278 | if (error) |
a8104a9f | 4279 | goto out_dput; |
146a8595 | 4280 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4281 | out_dput: |
921a1650 | 4282 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4283 | if (delegated_inode) { |
4284 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4285 | if (!error) { |
4286 | path_put(&old_path); | |
146a8595 | 4287 | goto retry; |
d22e6338 | 4288 | } |
146a8595 | 4289 | } |
442e31ca | 4290 | if (retry_estale(error, how)) { |
d22e6338 | 4291 | path_put(&old_path); |
442e31ca JL |
4292 | how |= LOOKUP_REVAL; |
4293 | goto retry; | |
4294 | } | |
1da177e4 | 4295 | out: |
2d8f3038 | 4296 | path_put(&old_path); |
1da177e4 LT |
4297 | |
4298 | return error; | |
4299 | } | |
4300 | ||
3480b257 | 4301 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4302 | { |
c04030e1 | 4303 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4304 | } |
4305 | ||
bc27027a MS |
4306 | /** |
4307 | * vfs_rename - rename a filesystem object | |
4308 | * @old_dir: parent of source | |
4309 | * @old_dentry: source | |
4310 | * @new_dir: parent of destination | |
4311 | * @new_dentry: destination | |
4312 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4313 | * @flags: rename flags |
bc27027a MS |
4314 | * |
4315 | * The caller must hold multiple mutexes--see lock_rename()). | |
4316 | * | |
4317 | * If vfs_rename discovers a delegation in need of breaking at either | |
4318 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4319 | * reference to the inode in delegated_inode. The caller should then | |
4320 | * break the delegation and retry. Because breaking a delegation may | |
4321 | * take a long time, the caller should drop all locks before doing | |
4322 | * so. | |
4323 | * | |
4324 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4325 | * be appropriate for callers that expect the underlying filesystem not | |
4326 | * to be NFS exported. | |
4327 | * | |
1da177e4 LT |
4328 | * The worst of all namespace operations - renaming directory. "Perverted" |
4329 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4330 | * Problems: | |
0117d427 | 4331 | * |
d03b29a2 | 4332 | * a) we can get into loop creation. |
1da177e4 LT |
4333 | * b) race potential - two innocent renames can create a loop together. |
4334 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4335 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4336 | * story. |
6cedba89 BF |
4337 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4338 | * and source (if it is not a directory). | |
1b1dcc1b | 4339 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4340 | * whether the target exists). Solution: try to be smart with locking |
4341 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4342 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4343 | * move will be locked. Thus we can rank directories by the tree |
4344 | * (ancestors first) and rank all non-directories after them. | |
4345 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4346 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4347 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4348 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4349 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4350 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4351 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4352 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4353 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4354 | * locking]. |
4355 | */ | |
bc27027a MS |
4356 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4357 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4358 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4359 | { |
bc27027a MS |
4360 | int error; |
4361 | bool is_dir = d_is_dir(old_dentry); | |
bc27027a | 4362 | struct inode *source = old_dentry->d_inode; |
9055cba7 | 4363 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4364 | bool new_is_dir = false; |
4365 | unsigned max_links = new_dir->i_sb->s_max_links; | |
49d31c2f | 4366 | struct name_snapshot old_name; |
bc27027a | 4367 | |
8d3e2936 | 4368 | if (source == target) |
bc27027a MS |
4369 | return 0; |
4370 | ||
4371 | error = may_delete(old_dir, old_dentry, is_dir); | |
4372 | if (error) | |
4373 | return error; | |
4374 | ||
da1ce067 | 4375 | if (!target) { |
bc27027a | 4376 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4377 | } else { |
4378 | new_is_dir = d_is_dir(new_dentry); | |
4379 | ||
4380 | if (!(flags & RENAME_EXCHANGE)) | |
4381 | error = may_delete(new_dir, new_dentry, is_dir); | |
4382 | else | |
4383 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4384 | } | |
bc27027a MS |
4385 | if (error) |
4386 | return error; | |
4387 | ||
2773bf00 | 4388 | if (!old_dir->i_op->rename) |
bc27027a | 4389 | return -EPERM; |
1da177e4 LT |
4390 | |
4391 | /* | |
4392 | * If we are going to change the parent - check write permissions, | |
4393 | * we'll need to flip '..'. | |
4394 | */ | |
da1ce067 MS |
4395 | if (new_dir != old_dir) { |
4396 | if (is_dir) { | |
4397 | error = inode_permission(source, MAY_WRITE); | |
4398 | if (error) | |
4399 | return error; | |
4400 | } | |
4401 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4402 | error = inode_permission(target, MAY_WRITE); | |
4403 | if (error) | |
4404 | return error; | |
4405 | } | |
1da177e4 LT |
4406 | } |
4407 | ||
0b3974eb MS |
4408 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4409 | flags); | |
1da177e4 LT |
4410 | if (error) |
4411 | return error; | |
4412 | ||
49d31c2f | 4413 | take_dentry_name_snapshot(&old_name, old_dentry); |
1d2ef590 | 4414 | dget(new_dentry); |
da1ce067 | 4415 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4416 | lock_two_nondirectories(source, target); |
4417 | else if (target) | |
5955102c | 4418 | inode_lock(target); |
9055cba7 SW |
4419 | |
4420 | error = -EBUSY; | |
7af1364f | 4421 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4422 | goto out; |
4423 | ||
da1ce067 | 4424 | if (max_links && new_dir != old_dir) { |
bc27027a | 4425 | error = -EMLINK; |
da1ce067 | 4426 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4427 | goto out; |
da1ce067 MS |
4428 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4429 | old_dir->i_nlink >= max_links) | |
4430 | goto out; | |
4431 | } | |
4432 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4433 | shrink_dcache_parent(new_dentry); | |
4434 | if (!is_dir) { | |
bc27027a | 4435 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4436 | if (error) |
4437 | goto out; | |
da1ce067 MS |
4438 | } |
4439 | if (target && !new_is_dir) { | |
4440 | error = try_break_deleg(target, delegated_inode); | |
4441 | if (error) | |
4442 | goto out; | |
8e6d782c | 4443 | } |
2773bf00 | 4444 | error = old_dir->i_op->rename(old_dir, old_dentry, |
18fc84da | 4445 | new_dir, new_dentry, flags); |
51892bbb SW |
4446 | if (error) |
4447 | goto out; | |
4448 | ||
da1ce067 | 4449 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4450 | if (is_dir) |
4451 | target->i_flags |= S_DEAD; | |
51892bbb | 4452 | dont_mount(new_dentry); |
8ed936b5 | 4453 | detach_mounts(new_dentry); |
bc27027a | 4454 | } |
da1ce067 MS |
4455 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4456 | if (!(flags & RENAME_EXCHANGE)) | |
4457 | d_move(old_dentry, new_dentry); | |
4458 | else | |
4459 | d_exchange(old_dentry, new_dentry); | |
4460 | } | |
51892bbb | 4461 | out: |
da1ce067 | 4462 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4463 | unlock_two_nondirectories(source, target); |
4464 | else if (target) | |
5955102c | 4465 | inode_unlock(target); |
1da177e4 | 4466 | dput(new_dentry); |
da1ce067 | 4467 | if (!error) { |
49d31c2f | 4468 | fsnotify_move(old_dir, new_dir, old_name.name, is_dir, |
da1ce067 MS |
4469 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4470 | if (flags & RENAME_EXCHANGE) { | |
4471 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4472 | new_is_dir, NULL, new_dentry); | |
4473 | } | |
4474 | } | |
49d31c2f | 4475 | release_dentry_name_snapshot(&old_name); |
0eeca283 | 4476 | |
1da177e4 LT |
4477 | return error; |
4478 | } | |
4d359507 | 4479 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4480 | |
520c8b16 MS |
4481 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4482 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4483 | { |
2ad94ae6 AV |
4484 | struct dentry *old_dentry, *new_dentry; |
4485 | struct dentry *trap; | |
f5beed75 AV |
4486 | struct path old_path, new_path; |
4487 | struct qstr old_last, new_last; | |
4488 | int old_type, new_type; | |
8e6d782c | 4489 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4490 | struct filename *from; |
4491 | struct filename *to; | |
f5beed75 | 4492 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; |
c6a94284 | 4493 | bool should_retry = false; |
2ad94ae6 | 4494 | int error; |
520c8b16 | 4495 | |
0d7a8555 | 4496 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4497 | return -EINVAL; |
4498 | ||
0d7a8555 MS |
4499 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4500 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4501 | return -EINVAL; |
4502 | ||
0d7a8555 MS |
4503 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4504 | return -EPERM; | |
4505 | ||
f5beed75 AV |
4506 | if (flags & RENAME_EXCHANGE) |
4507 | target_flags = 0; | |
4508 | ||
c6a94284 | 4509 | retry: |
c1d4dd27 AV |
4510 | from = filename_parentat(olddfd, getname(oldname), lookup_flags, |
4511 | &old_path, &old_last, &old_type); | |
91a27b2a JL |
4512 | if (IS_ERR(from)) { |
4513 | error = PTR_ERR(from); | |
1da177e4 | 4514 | goto exit; |
91a27b2a | 4515 | } |
1da177e4 | 4516 | |
c1d4dd27 AV |
4517 | to = filename_parentat(newdfd, getname(newname), lookup_flags, |
4518 | &new_path, &new_last, &new_type); | |
91a27b2a JL |
4519 | if (IS_ERR(to)) { |
4520 | error = PTR_ERR(to); | |
1da177e4 | 4521 | goto exit1; |
91a27b2a | 4522 | } |
1da177e4 LT |
4523 | |
4524 | error = -EXDEV; | |
f5beed75 | 4525 | if (old_path.mnt != new_path.mnt) |
1da177e4 LT |
4526 | goto exit2; |
4527 | ||
1da177e4 | 4528 | error = -EBUSY; |
f5beed75 | 4529 | if (old_type != LAST_NORM) |
1da177e4 LT |
4530 | goto exit2; |
4531 | ||
0a7c3937 MS |
4532 | if (flags & RENAME_NOREPLACE) |
4533 | error = -EEXIST; | |
f5beed75 | 4534 | if (new_type != LAST_NORM) |
1da177e4 LT |
4535 | goto exit2; |
4536 | ||
f5beed75 | 4537 | error = mnt_want_write(old_path.mnt); |
c30dabfe JK |
4538 | if (error) |
4539 | goto exit2; | |
4540 | ||
8e6d782c | 4541 | retry_deleg: |
f5beed75 | 4542 | trap = lock_rename(new_path.dentry, old_path.dentry); |
1da177e4 | 4543 | |
f5beed75 | 4544 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); |
1da177e4 LT |
4545 | error = PTR_ERR(old_dentry); |
4546 | if (IS_ERR(old_dentry)) | |
4547 | goto exit3; | |
4548 | /* source must exist */ | |
4549 | error = -ENOENT; | |
b18825a7 | 4550 | if (d_is_negative(old_dentry)) |
1da177e4 | 4551 | goto exit4; |
f5beed75 | 4552 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); |
0a7c3937 MS |
4553 | error = PTR_ERR(new_dentry); |
4554 | if (IS_ERR(new_dentry)) | |
4555 | goto exit4; | |
4556 | error = -EEXIST; | |
4557 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4558 | goto exit5; | |
da1ce067 MS |
4559 | if (flags & RENAME_EXCHANGE) { |
4560 | error = -ENOENT; | |
4561 | if (d_is_negative(new_dentry)) | |
4562 | goto exit5; | |
4563 | ||
4564 | if (!d_is_dir(new_dentry)) { | |
4565 | error = -ENOTDIR; | |
f5beed75 | 4566 | if (new_last.name[new_last.len]) |
da1ce067 MS |
4567 | goto exit5; |
4568 | } | |
4569 | } | |
1da177e4 | 4570 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4571 | if (!d_is_dir(old_dentry)) { |
1da177e4 | 4572 | error = -ENOTDIR; |
f5beed75 | 4573 | if (old_last.name[old_last.len]) |
0a7c3937 | 4574 | goto exit5; |
f5beed75 | 4575 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) |
0a7c3937 | 4576 | goto exit5; |
1da177e4 LT |
4577 | } |
4578 | /* source should not be ancestor of target */ | |
4579 | error = -EINVAL; | |
4580 | if (old_dentry == trap) | |
0a7c3937 | 4581 | goto exit5; |
1da177e4 | 4582 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4583 | if (!(flags & RENAME_EXCHANGE)) |
4584 | error = -ENOTEMPTY; | |
1da177e4 LT |
4585 | if (new_dentry == trap) |
4586 | goto exit5; | |
4587 | ||
f5beed75 AV |
4588 | error = security_path_rename(&old_path, old_dentry, |
4589 | &new_path, new_dentry, flags); | |
be6d3e56 | 4590 | if (error) |
c30dabfe | 4591 | goto exit5; |
f5beed75 AV |
4592 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, |
4593 | new_path.dentry->d_inode, new_dentry, | |
520c8b16 | 4594 | &delegated_inode, flags); |
1da177e4 LT |
4595 | exit5: |
4596 | dput(new_dentry); | |
4597 | exit4: | |
4598 | dput(old_dentry); | |
4599 | exit3: | |
f5beed75 | 4600 | unlock_rename(new_path.dentry, old_path.dentry); |
8e6d782c BF |
4601 | if (delegated_inode) { |
4602 | error = break_deleg_wait(&delegated_inode); | |
4603 | if (!error) | |
4604 | goto retry_deleg; | |
4605 | } | |
f5beed75 | 4606 | mnt_drop_write(old_path.mnt); |
1da177e4 | 4607 | exit2: |
c6a94284 JL |
4608 | if (retry_estale(error, lookup_flags)) |
4609 | should_retry = true; | |
f5beed75 | 4610 | path_put(&new_path); |
2ad94ae6 | 4611 | putname(to); |
1da177e4 | 4612 | exit1: |
f5beed75 | 4613 | path_put(&old_path); |
1da177e4 | 4614 | putname(from); |
c6a94284 JL |
4615 | if (should_retry) { |
4616 | should_retry = false; | |
4617 | lookup_flags |= LOOKUP_REVAL; | |
4618 | goto retry; | |
4619 | } | |
2ad94ae6 | 4620 | exit: |
1da177e4 LT |
4621 | return error; |
4622 | } | |
4623 | ||
520c8b16 MS |
4624 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4625 | int, newdfd, const char __user *, newname) | |
4626 | { | |
4627 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4628 | } | |
4629 | ||
a26eab24 | 4630 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4631 | { |
520c8b16 | 4632 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4633 | } |
4634 | ||
787fb6bc MS |
4635 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4636 | { | |
4637 | int error = may_create(dir, dentry); | |
4638 | if (error) | |
4639 | return error; | |
4640 | ||
4641 | if (!dir->i_op->mknod) | |
4642 | return -EPERM; | |
4643 | ||
4644 | return dir->i_op->mknod(dir, dentry, | |
4645 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4646 | } | |
4647 | EXPORT_SYMBOL(vfs_whiteout); | |
4648 | ||
5d826c84 | 4649 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4650 | { |
5d826c84 | 4651 | int len = PTR_ERR(link); |
1da177e4 LT |
4652 | if (IS_ERR(link)) |
4653 | goto out; | |
4654 | ||
4655 | len = strlen(link); | |
4656 | if (len > (unsigned) buflen) | |
4657 | len = buflen; | |
4658 | if (copy_to_user(buffer, link, len)) | |
4659 | len = -EFAULT; | |
4660 | out: | |
4661 | return len; | |
4662 | } | |
4663 | ||
4664 | /* | |
4665 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
6b255391 AV |
4666 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it |
4667 | * for any given inode is up to filesystem. | |
1da177e4 | 4668 | */ |
d16744ec MS |
4669 | static int generic_readlink(struct dentry *dentry, char __user *buffer, |
4670 | int buflen) | |
1da177e4 | 4671 | { |
fceef393 | 4672 | DEFINE_DELAYED_CALL(done); |
5f2c4179 AV |
4673 | struct inode *inode = d_inode(dentry); |
4674 | const char *link = inode->i_link; | |
694a1764 | 4675 | int res; |
cc314eef | 4676 | |
d4dee48b | 4677 | if (!link) { |
fceef393 | 4678 | link = inode->i_op->get_link(dentry, inode, &done); |
d4dee48b AV |
4679 | if (IS_ERR(link)) |
4680 | return PTR_ERR(link); | |
4681 | } | |
680baacb | 4682 | res = readlink_copy(buffer, buflen, link); |
fceef393 | 4683 | do_delayed_call(&done); |
694a1764 | 4684 | return res; |
1da177e4 LT |
4685 | } |
4686 | ||
fd4a0edf MS |
4687 | /** |
4688 | * vfs_readlink - copy symlink body into userspace buffer | |
4689 | * @dentry: dentry on which to get symbolic link | |
4690 | * @buffer: user memory pointer | |
4691 | * @buflen: size of buffer | |
4692 | * | |
4693 | * Does not touch atime. That's up to the caller if necessary | |
4694 | * | |
4695 | * Does not call security hook. | |
4696 | */ | |
4697 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4698 | { | |
4699 | struct inode *inode = d_inode(dentry); | |
4700 | ||
76fca90e MS |
4701 | if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { |
4702 | if (unlikely(inode->i_op->readlink)) | |
4703 | return inode->i_op->readlink(dentry, buffer, buflen); | |
4704 | ||
4705 | if (!d_is_symlink(dentry)) | |
4706 | return -EINVAL; | |
4707 | ||
4708 | spin_lock(&inode->i_lock); | |
4709 | inode->i_opflags |= IOP_DEFAULT_READLINK; | |
4710 | spin_unlock(&inode->i_lock); | |
4711 | } | |
fd4a0edf | 4712 | |
76fca90e | 4713 | return generic_readlink(dentry, buffer, buflen); |
fd4a0edf MS |
4714 | } |
4715 | EXPORT_SYMBOL(vfs_readlink); | |
1da177e4 | 4716 | |
d60874cd MS |
4717 | /** |
4718 | * vfs_get_link - get symlink body | |
4719 | * @dentry: dentry on which to get symbolic link | |
4720 | * @done: caller needs to free returned data with this | |
4721 | * | |
4722 | * Calls security hook and i_op->get_link() on the supplied inode. | |
4723 | * | |
4724 | * It does not touch atime. That's up to the caller if necessary. | |
4725 | * | |
4726 | * Does not work on "special" symlinks like /proc/$$/fd/N | |
4727 | */ | |
4728 | const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) | |
4729 | { | |
4730 | const char *res = ERR_PTR(-EINVAL); | |
4731 | struct inode *inode = d_inode(dentry); | |
4732 | ||
4733 | if (d_is_symlink(dentry)) { | |
4734 | res = ERR_PTR(security_inode_readlink(dentry)); | |
4735 | if (!res) | |
4736 | res = inode->i_op->get_link(dentry, inode, done); | |
4737 | } | |
4738 | return res; | |
4739 | } | |
4740 | EXPORT_SYMBOL(vfs_get_link); | |
4741 | ||
1da177e4 | 4742 | /* get the link contents into pagecache */ |
6b255391 | 4743 | const char *page_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 4744 | struct delayed_call *callback) |
1da177e4 | 4745 | { |
ebd09abb DG |
4746 | char *kaddr; |
4747 | struct page *page; | |
6b255391 AV |
4748 | struct address_space *mapping = inode->i_mapping; |
4749 | ||
d3883d4f AV |
4750 | if (!dentry) { |
4751 | page = find_get_page(mapping, 0); | |
4752 | if (!page) | |
4753 | return ERR_PTR(-ECHILD); | |
4754 | if (!PageUptodate(page)) { | |
4755 | put_page(page); | |
4756 | return ERR_PTR(-ECHILD); | |
4757 | } | |
4758 | } else { | |
4759 | page = read_mapping_page(mapping, 0, NULL); | |
4760 | if (IS_ERR(page)) | |
4761 | return (char*)page; | |
4762 | } | |
fceef393 | 4763 | set_delayed_call(callback, page_put_link, page); |
21fc61c7 AV |
4764 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); |
4765 | kaddr = page_address(page); | |
6b255391 | 4766 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); |
ebd09abb | 4767 | return kaddr; |
1da177e4 LT |
4768 | } |
4769 | ||
6b255391 | 4770 | EXPORT_SYMBOL(page_get_link); |
1da177e4 | 4771 | |
fceef393 | 4772 | void page_put_link(void *arg) |
1da177e4 | 4773 | { |
fceef393 | 4774 | put_page(arg); |
1da177e4 | 4775 | } |
4d359507 | 4776 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4777 | |
aa80deab AV |
4778 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
4779 | { | |
fceef393 | 4780 | DEFINE_DELAYED_CALL(done); |
6b255391 AV |
4781 | int res = readlink_copy(buffer, buflen, |
4782 | page_get_link(dentry, d_inode(dentry), | |
fceef393 AV |
4783 | &done)); |
4784 | do_delayed_call(&done); | |
aa80deab AV |
4785 | return res; |
4786 | } | |
4787 | EXPORT_SYMBOL(page_readlink); | |
4788 | ||
54566b2c NP |
4789 | /* |
4790 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4791 | */ | |
4792 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4793 | { |
4794 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4795 | struct page *page; |
afddba49 | 4796 | void *fsdata; |
beb497ab | 4797 | int err; |
c718a975 | 4798 | unsigned int flags = 0; |
54566b2c NP |
4799 | if (nofs) |
4800 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4801 | |
7e53cac4 | 4802 | retry: |
afddba49 | 4803 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4804 | flags, &page, &fsdata); |
1da177e4 | 4805 | if (err) |
afddba49 NP |
4806 | goto fail; |
4807 | ||
21fc61c7 | 4808 | memcpy(page_address(page), symname, len-1); |
afddba49 NP |
4809 | |
4810 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4811 | page, fsdata); | |
1da177e4 LT |
4812 | if (err < 0) |
4813 | goto fail; | |
afddba49 NP |
4814 | if (err < len-1) |
4815 | goto retry; | |
4816 | ||
1da177e4 LT |
4817 | mark_inode_dirty(inode); |
4818 | return 0; | |
1da177e4 LT |
4819 | fail: |
4820 | return err; | |
4821 | } | |
4d359507 | 4822 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4823 | |
0adb25d2 KK |
4824 | int page_symlink(struct inode *inode, const char *symname, int len) |
4825 | { | |
4826 | return __page_symlink(inode, symname, len, | |
c62d2555 | 4827 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); |
0adb25d2 | 4828 | } |
4d359507 | 4829 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4830 | |
92e1d5be | 4831 | const struct inode_operations page_symlink_inode_operations = { |
6b255391 | 4832 | .get_link = page_get_link, |
1da177e4 | 4833 | }; |
1da177e4 | 4834 | EXPORT_SYMBOL(page_symlink_inode_operations); |