Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | ||
e81e3f4d | 39 | #include "internal.h" |
c7105365 | 40 | #include "mount.h" |
e81e3f4d | 41 | |
1da177e4 LT |
42 | /* [Feb-1997 T. Schoebel-Theuer] |
43 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
44 | * were necessary because of omirr. The reason is that omirr needs | |
45 | * to know the _real_ pathname, not the user-supplied one, in case | |
46 | * of symlinks (and also when transname replacements occur). | |
47 | * | |
48 | * The new code replaces the old recursive symlink resolution with | |
49 | * an iterative one (in case of non-nested symlink chains). It does | |
50 | * this with calls to <fs>_follow_link(). | |
51 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
52 | * replaced with a single function lookup_dentry() that can handle all | |
53 | * the special cases of the former code. | |
54 | * | |
55 | * With the new dcache, the pathname is stored at each inode, at least as | |
56 | * long as the refcount of the inode is positive. As a side effect, the | |
57 | * size of the dcache depends on the inode cache and thus is dynamic. | |
58 | * | |
59 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
60 | * resolution to correspond with current state of the code. | |
61 | * | |
62 | * Note that the symlink resolution is not *completely* iterative. | |
63 | * There is still a significant amount of tail- and mid- recursion in | |
64 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
65 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
66 | * may return different results than <fs>_follow_link(). Many virtual | |
67 | * filesystems (including /proc) exhibit this behavior. | |
68 | */ | |
69 | ||
70 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
71 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
72 | * and the name already exists in form of a symlink, try to create the new | |
73 | * name indicated by the symlink. The old code always complained that the | |
74 | * name already exists, due to not following the symlink even if its target | |
75 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 76 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
77 | * |
78 | * I don't know which semantics is the right one, since I have no access | |
79 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
80 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
81 | * "old" one. Personally, I think the new semantics is much more logical. | |
82 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
83 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
84 | * and in the old Linux semantics. | |
85 | */ | |
86 | ||
87 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
88 | * semantics. See the comments in "open_namei" and "do_link" below. | |
89 | * | |
90 | * [10-Sep-98 Alan Modra] Another symlink change. | |
91 | */ | |
92 | ||
93 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
94 | * inside the path - always follow. | |
95 | * in the last component in creation/removal/renaming - never follow. | |
96 | * if LOOKUP_FOLLOW passed - follow. | |
97 | * if the pathname has trailing slashes - follow. | |
98 | * otherwise - don't follow. | |
99 | * (applied in that order). | |
100 | * | |
101 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
102 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
103 | * During the 2.4 we need to fix the userland stuff depending on it - | |
104 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
105 | * XEmacs seems to be relying on it... | |
106 | */ | |
107 | /* | |
108 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 109 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
110 | * any extra contention... |
111 | */ | |
112 | ||
113 | /* In order to reduce some races, while at the same time doing additional | |
114 | * checking and hopefully speeding things up, we copy filenames to the | |
115 | * kernel data space before using them.. | |
116 | * | |
117 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
118 | * PATH_MAX includes the nul terminator --RR. | |
119 | */ | |
91a27b2a | 120 | void final_putname(struct filename *name) |
1da177e4 | 121 | { |
7950e385 JL |
122 | if (name->separate) { |
123 | __putname(name->name); | |
124 | kfree(name); | |
125 | } else { | |
126 | __putname(name); | |
127 | } | |
91a27b2a JL |
128 | } |
129 | ||
7950e385 JL |
130 | #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) |
131 | ||
91a27b2a JL |
132 | static struct filename * |
133 | getname_flags(const char __user *filename, int flags, int *empty) | |
134 | { | |
135 | struct filename *result, *err; | |
3f9f0aa6 | 136 | int len; |
7950e385 JL |
137 | long max; |
138 | char *kname; | |
4043cde8 | 139 | |
7ac86265 JL |
140 | result = audit_reusename(filename); |
141 | if (result) | |
142 | return result; | |
143 | ||
7950e385 | 144 | result = __getname(); |
3f9f0aa6 | 145 | if (unlikely(!result)) |
4043cde8 EP |
146 | return ERR_PTR(-ENOMEM); |
147 | ||
7950e385 JL |
148 | /* |
149 | * First, try to embed the struct filename inside the names_cache | |
150 | * allocation | |
151 | */ | |
152 | kname = (char *)result + sizeof(*result); | |
91a27b2a | 153 | result->name = kname; |
7950e385 JL |
154 | result->separate = false; |
155 | max = EMBEDDED_NAME_MAX; | |
156 | ||
157 | recopy: | |
158 | len = strncpy_from_user(kname, filename, max); | |
91a27b2a JL |
159 | if (unlikely(len < 0)) { |
160 | err = ERR_PTR(len); | |
3f9f0aa6 | 161 | goto error; |
91a27b2a | 162 | } |
3f9f0aa6 | 163 | |
7950e385 JL |
164 | /* |
165 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
166 | * separate struct filename so we can dedicate the entire | |
167 | * names_cache allocation for the pathname, and re-do the copy from | |
168 | * userland. | |
169 | */ | |
170 | if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { | |
171 | kname = (char *)result; | |
172 | ||
173 | result = kzalloc(sizeof(*result), GFP_KERNEL); | |
174 | if (!result) { | |
175 | err = ERR_PTR(-ENOMEM); | |
176 | result = (struct filename *)kname; | |
177 | goto error; | |
178 | } | |
179 | result->name = kname; | |
180 | result->separate = true; | |
181 | max = PATH_MAX; | |
182 | goto recopy; | |
183 | } | |
184 | ||
3f9f0aa6 LT |
185 | /* The empty path is special. */ |
186 | if (unlikely(!len)) { | |
187 | if (empty) | |
4043cde8 | 188 | *empty = 1; |
3f9f0aa6 LT |
189 | err = ERR_PTR(-ENOENT); |
190 | if (!(flags & LOOKUP_EMPTY)) | |
191 | goto error; | |
1da177e4 | 192 | } |
3f9f0aa6 LT |
193 | |
194 | err = ERR_PTR(-ENAMETOOLONG); | |
7950e385 JL |
195 | if (unlikely(len >= PATH_MAX)) |
196 | goto error; | |
197 | ||
198 | result->uptr = filename; | |
199 | audit_getname(result); | |
200 | return result; | |
3f9f0aa6 LT |
201 | |
202 | error: | |
7950e385 | 203 | final_putname(result); |
3f9f0aa6 | 204 | return err; |
1da177e4 LT |
205 | } |
206 | ||
91a27b2a JL |
207 | struct filename * |
208 | getname(const char __user * filename) | |
f52e0c11 | 209 | { |
f7493e5d | 210 | return getname_flags(filename, 0, NULL); |
f52e0c11 | 211 | } |
91a27b2a | 212 | EXPORT_SYMBOL(getname); |
f52e0c11 | 213 | |
1da177e4 | 214 | #ifdef CONFIG_AUDITSYSCALL |
91a27b2a | 215 | void putname(struct filename *name) |
1da177e4 | 216 | { |
5ac3a9c2 | 217 | if (unlikely(!audit_dummy_context())) |
91a27b2a JL |
218 | return audit_putname(name); |
219 | final_putname(name); | |
1da177e4 | 220 | } |
1da177e4 LT |
221 | #endif |
222 | ||
e77819e5 LT |
223 | static int check_acl(struct inode *inode, int mask) |
224 | { | |
84635d68 | 225 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
226 | struct posix_acl *acl; |
227 | ||
e77819e5 | 228 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
229 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
230 | if (!acl) | |
e77819e5 | 231 | return -EAGAIN; |
3567866b AV |
232 | /* no ->get_acl() calls in RCU mode... */ |
233 | if (acl == ACL_NOT_CACHED) | |
234 | return -ECHILD; | |
206b1d09 | 235 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
236 | } |
237 | ||
238 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); | |
239 | ||
240 | /* | |
4e34e719 CH |
241 | * A filesystem can force a ACL callback by just never filling the |
242 | * ACL cache. But normally you'd fill the cache either at inode | |
243 | * instantiation time, or on the first ->get_acl call. | |
e77819e5 | 244 | * |
4e34e719 CH |
245 | * If the filesystem doesn't have a get_acl() function at all, we'll |
246 | * just create the negative cache entry. | |
e77819e5 LT |
247 | */ |
248 | if (acl == ACL_NOT_CACHED) { | |
4e34e719 CH |
249 | if (inode->i_op->get_acl) { |
250 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); | |
251 | if (IS_ERR(acl)) | |
252 | return PTR_ERR(acl); | |
253 | } else { | |
254 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); | |
255 | return -EAGAIN; | |
256 | } | |
e77819e5 LT |
257 | } |
258 | ||
259 | if (acl) { | |
260 | int error = posix_acl_permission(inode, acl, mask); | |
261 | posix_acl_release(acl); | |
262 | return error; | |
263 | } | |
84635d68 | 264 | #endif |
e77819e5 LT |
265 | |
266 | return -EAGAIN; | |
267 | } | |
268 | ||
5909ccaa | 269 | /* |
948409c7 | 270 | * This does the basic permission checking |
1da177e4 | 271 | */ |
7e40145e | 272 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 273 | { |
26cf46be | 274 | unsigned int mode = inode->i_mode; |
1da177e4 | 275 | |
8e96e3b7 | 276 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
277 | mode >>= 6; |
278 | else { | |
e77819e5 | 279 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 280 | int error = check_acl(inode, mask); |
b74c79e9 NP |
281 | if (error != -EAGAIN) |
282 | return error; | |
1da177e4 LT |
283 | } |
284 | ||
285 | if (in_group_p(inode->i_gid)) | |
286 | mode >>= 3; | |
287 | } | |
288 | ||
289 | /* | |
290 | * If the DACs are ok we don't need any capability check. | |
291 | */ | |
9c2c7039 | 292 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 293 | return 0; |
5909ccaa LT |
294 | return -EACCES; |
295 | } | |
296 | ||
297 | /** | |
b74c79e9 | 298 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 299 | * @inode: inode to check access rights for |
8fd90c8d | 300 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
301 | * |
302 | * Used to check for read/write/execute permissions on a file. | |
303 | * We use "fsuid" for this, letting us set arbitrary permissions | |
304 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
305 | * are used for other things. |
306 | * | |
307 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
308 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
309 | * It would then be called again in ref-walk mode. | |
5909ccaa | 310 | */ |
2830ba7f | 311 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
312 | { |
313 | int ret; | |
314 | ||
315 | /* | |
948409c7 | 316 | * Do the basic permission checks. |
5909ccaa | 317 | */ |
7e40145e | 318 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
319 | if (ret != -EACCES) |
320 | return ret; | |
1da177e4 | 321 | |
d594e7ec AV |
322 | if (S_ISDIR(inode->i_mode)) { |
323 | /* DACs are overridable for directories */ | |
4f80c6c1 | 324 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
325 | return 0; |
326 | if (!(mask & MAY_WRITE)) | |
4f80c6c1 AL |
327 | if (capable_wrt_inode_uidgid(inode, |
328 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
329 | return 0; |
330 | return -EACCES; | |
331 | } | |
1da177e4 LT |
332 | /* |
333 | * Read/write DACs are always overridable. | |
d594e7ec AV |
334 | * Executable DACs are overridable when there is |
335 | * at least one exec bit set. | |
1da177e4 | 336 | */ |
d594e7ec | 337 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
4f80c6c1 | 338 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
339 | return 0; |
340 | ||
341 | /* | |
342 | * Searching includes executable on directories, else just read. | |
343 | */ | |
7ea66001 | 344 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 345 | if (mask == MAY_READ) |
4f80c6c1 | 346 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
347 | return 0; |
348 | ||
349 | return -EACCES; | |
350 | } | |
351 | ||
3ddcd056 LT |
352 | /* |
353 | * We _really_ want to just do "generic_permission()" without | |
354 | * even looking at the inode->i_op values. So we keep a cache | |
355 | * flag in inode->i_opflags, that says "this has not special | |
356 | * permission function, use the fast case". | |
357 | */ | |
358 | static inline int do_inode_permission(struct inode *inode, int mask) | |
359 | { | |
360 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
361 | if (likely(inode->i_op->permission)) | |
362 | return inode->i_op->permission(inode, mask); | |
363 | ||
364 | /* This gets set once for the inode lifetime */ | |
365 | spin_lock(&inode->i_lock); | |
366 | inode->i_opflags |= IOP_FASTPERM; | |
367 | spin_unlock(&inode->i_lock); | |
368 | } | |
369 | return generic_permission(inode, mask); | |
370 | } | |
371 | ||
cb23beb5 | 372 | /** |
0bdaea90 DH |
373 | * __inode_permission - Check for access rights to a given inode |
374 | * @inode: Inode to check permission on | |
375 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 376 | * |
0bdaea90 | 377 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
378 | * |
379 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
380 | * |
381 | * This does not check for a read-only file system. You probably want | |
382 | * inode_permission(). | |
cb23beb5 | 383 | */ |
0bdaea90 | 384 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 385 | { |
e6305c43 | 386 | int retval; |
1da177e4 | 387 | |
3ddcd056 | 388 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
389 | /* |
390 | * Nobody gets write access to an immutable file. | |
391 | */ | |
392 | if (IS_IMMUTABLE(inode)) | |
393 | return -EACCES; | |
394 | } | |
395 | ||
3ddcd056 | 396 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
397 | if (retval) |
398 | return retval; | |
399 | ||
08ce5f16 SH |
400 | retval = devcgroup_inode_permission(inode, mask); |
401 | if (retval) | |
402 | return retval; | |
403 | ||
d09ca739 | 404 | return security_inode_permission(inode, mask); |
1da177e4 LT |
405 | } |
406 | ||
0bdaea90 DH |
407 | /** |
408 | * sb_permission - Check superblock-level permissions | |
409 | * @sb: Superblock of inode to check permission on | |
55852635 | 410 | * @inode: Inode to check permission on |
0bdaea90 DH |
411 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
412 | * | |
413 | * Separate out file-system wide checks from inode-specific permission checks. | |
414 | */ | |
415 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
416 | { | |
417 | if (unlikely(mask & MAY_WRITE)) { | |
418 | umode_t mode = inode->i_mode; | |
419 | ||
420 | /* Nobody gets write access to a read-only fs. */ | |
421 | if ((sb->s_flags & MS_RDONLY) && | |
422 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
423 | return -EROFS; | |
424 | } | |
425 | return 0; | |
426 | } | |
427 | ||
428 | /** | |
429 | * inode_permission - Check for access rights to a given inode | |
430 | * @inode: Inode to check permission on | |
431 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
432 | * | |
433 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
434 | * this, letting us set arbitrary permissions for filesystem access without | |
435 | * changing the "normal" UIDs which are used for other things. | |
436 | * | |
437 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
438 | */ | |
439 | int inode_permission(struct inode *inode, int mask) | |
440 | { | |
441 | int retval; | |
442 | ||
443 | retval = sb_permission(inode->i_sb, inode, mask); | |
444 | if (retval) | |
445 | return retval; | |
446 | return __inode_permission(inode, mask); | |
447 | } | |
448 | ||
5dd784d0 JB |
449 | /** |
450 | * path_get - get a reference to a path | |
451 | * @path: path to get the reference to | |
452 | * | |
453 | * Given a path increment the reference count to the dentry and the vfsmount. | |
454 | */ | |
dcf787f3 | 455 | void path_get(const struct path *path) |
5dd784d0 JB |
456 | { |
457 | mntget(path->mnt); | |
458 | dget(path->dentry); | |
459 | } | |
460 | EXPORT_SYMBOL(path_get); | |
461 | ||
1d957f9b JB |
462 | /** |
463 | * path_put - put a reference to a path | |
464 | * @path: path to put the reference to | |
465 | * | |
466 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
467 | */ | |
dcf787f3 | 468 | void path_put(const struct path *path) |
1da177e4 | 469 | { |
1d957f9b JB |
470 | dput(path->dentry); |
471 | mntput(path->mnt); | |
1da177e4 | 472 | } |
1d957f9b | 473 | EXPORT_SYMBOL(path_put); |
1da177e4 | 474 | |
19660af7 | 475 | /* |
31e6b01f | 476 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
477 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
478 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
479 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
480 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
481 | * got stuck, so ref-walk may continue from there. If this is not successful | |
482 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
483 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 484 | */ |
31e6b01f | 485 | |
32a7991b AV |
486 | static inline void lock_rcu_walk(void) |
487 | { | |
488 | br_read_lock(&vfsmount_lock); | |
489 | rcu_read_lock(); | |
490 | } | |
491 | ||
492 | static inline void unlock_rcu_walk(void) | |
493 | { | |
494 | rcu_read_unlock(); | |
495 | br_read_unlock(&vfsmount_lock); | |
496 | } | |
497 | ||
31e6b01f | 498 | /** |
19660af7 AV |
499 | * unlazy_walk - try to switch to ref-walk mode. |
500 | * @nd: nameidata pathwalk data | |
501 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 502 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 503 | * |
19660af7 AV |
504 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
505 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
506 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 507 | */ |
19660af7 | 508 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
509 | { |
510 | struct fs_struct *fs = current->fs; | |
511 | struct dentry *parent = nd->path.dentry; | |
5b6ca027 | 512 | int want_root = 0; |
31e6b01f NP |
513 | |
514 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
5b6ca027 AV |
515 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
516 | want_root = 1; | |
31e6b01f NP |
517 | spin_lock(&fs->lock); |
518 | if (nd->root.mnt != fs->root.mnt || | |
519 | nd->root.dentry != fs->root.dentry) | |
520 | goto err_root; | |
521 | } | |
522 | spin_lock(&parent->d_lock); | |
19660af7 AV |
523 | if (!dentry) { |
524 | if (!__d_rcu_to_refcount(parent, nd->seq)) | |
525 | goto err_parent; | |
526 | BUG_ON(nd->inode != parent->d_inode); | |
527 | } else { | |
94c0d4ec AV |
528 | if (dentry->d_parent != parent) |
529 | goto err_parent; | |
19660af7 AV |
530 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
531 | if (!__d_rcu_to_refcount(dentry, nd->seq)) | |
532 | goto err_child; | |
533 | /* | |
534 | * If the sequence check on the child dentry passed, then | |
535 | * the child has not been removed from its parent. This | |
536 | * means the parent dentry must be valid and able to take | |
537 | * a reference at this point. | |
538 | */ | |
539 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); | |
540 | BUG_ON(!parent->d_count); | |
541 | parent->d_count++; | |
542 | spin_unlock(&dentry->d_lock); | |
543 | } | |
31e6b01f | 544 | spin_unlock(&parent->d_lock); |
5b6ca027 | 545 | if (want_root) { |
31e6b01f NP |
546 | path_get(&nd->root); |
547 | spin_unlock(&fs->lock); | |
548 | } | |
549 | mntget(nd->path.mnt); | |
550 | ||
32a7991b | 551 | unlock_rcu_walk(); |
31e6b01f NP |
552 | nd->flags &= ~LOOKUP_RCU; |
553 | return 0; | |
19660af7 AV |
554 | |
555 | err_child: | |
31e6b01f | 556 | spin_unlock(&dentry->d_lock); |
19660af7 | 557 | err_parent: |
31e6b01f NP |
558 | spin_unlock(&parent->d_lock); |
559 | err_root: | |
5b6ca027 | 560 | if (want_root) |
31e6b01f NP |
561 | spin_unlock(&fs->lock); |
562 | return -ECHILD; | |
563 | } | |
564 | ||
4ce16ef3 | 565 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 566 | { |
4ce16ef3 | 567 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
568 | } |
569 | ||
9f1fafee AV |
570 | /** |
571 | * complete_walk - successful completion of path walk | |
572 | * @nd: pointer nameidata | |
39159de2 | 573 | * |
9f1fafee AV |
574 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
575 | * Revalidate the final result, unless we'd already done that during | |
576 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
577 | * success, -error on failure. In case of failure caller does not | |
578 | * need to drop nd->path. | |
39159de2 | 579 | */ |
9f1fafee | 580 | static int complete_walk(struct nameidata *nd) |
39159de2 | 581 | { |
16c2cd71 | 582 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 583 | int status; |
39159de2 | 584 | |
9f1fafee AV |
585 | if (nd->flags & LOOKUP_RCU) { |
586 | nd->flags &= ~LOOKUP_RCU; | |
587 | if (!(nd->flags & LOOKUP_ROOT)) | |
588 | nd->root.mnt = NULL; | |
589 | spin_lock(&dentry->d_lock); | |
590 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { | |
591 | spin_unlock(&dentry->d_lock); | |
32a7991b | 592 | unlock_rcu_walk(); |
9f1fafee AV |
593 | return -ECHILD; |
594 | } | |
595 | BUG_ON(nd->inode != dentry->d_inode); | |
596 | spin_unlock(&dentry->d_lock); | |
597 | mntget(nd->path.mnt); | |
32a7991b | 598 | unlock_rcu_walk(); |
9f1fafee AV |
599 | } |
600 | ||
16c2cd71 AV |
601 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
602 | return 0; | |
603 | ||
ecf3d1f1 | 604 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
605 | return 0; |
606 | ||
ecf3d1f1 | 607 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
608 | if (status > 0) |
609 | return 0; | |
610 | ||
16c2cd71 | 611 | if (!status) |
39159de2 | 612 | status = -ESTALE; |
16c2cd71 | 613 | |
9f1fafee | 614 | path_put(&nd->path); |
39159de2 JL |
615 | return status; |
616 | } | |
617 | ||
2a737871 AV |
618 | static __always_inline void set_root(struct nameidata *nd) |
619 | { | |
f7ad3c6b MS |
620 | if (!nd->root.mnt) |
621 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
622 | } |
623 | ||
6de88d72 AV |
624 | static int link_path_walk(const char *, struct nameidata *); |
625 | ||
31e6b01f NP |
626 | static __always_inline void set_root_rcu(struct nameidata *nd) |
627 | { | |
628 | if (!nd->root.mnt) { | |
629 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
630 | unsigned seq; |
631 | ||
632 | do { | |
633 | seq = read_seqcount_begin(&fs->seq); | |
634 | nd->root = fs->root; | |
c1530019 | 635 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 636 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
637 | } |
638 | } | |
639 | ||
f1662356 | 640 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
1da177e4 | 641 | { |
31e6b01f NP |
642 | int ret; |
643 | ||
1da177e4 LT |
644 | if (IS_ERR(link)) |
645 | goto fail; | |
646 | ||
647 | if (*link == '/') { | |
2a737871 | 648 | set_root(nd); |
1d957f9b | 649 | path_put(&nd->path); |
2a737871 AV |
650 | nd->path = nd->root; |
651 | path_get(&nd->root); | |
16c2cd71 | 652 | nd->flags |= LOOKUP_JUMPED; |
1da177e4 | 653 | } |
31e6b01f | 654 | nd->inode = nd->path.dentry->d_inode; |
b4091d5f | 655 | |
31e6b01f NP |
656 | ret = link_path_walk(link, nd); |
657 | return ret; | |
1da177e4 | 658 | fail: |
1d957f9b | 659 | path_put(&nd->path); |
1da177e4 LT |
660 | return PTR_ERR(link); |
661 | } | |
662 | ||
1d957f9b | 663 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
664 | { |
665 | dput(path->dentry); | |
4ac91378 | 666 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
667 | mntput(path->mnt); |
668 | } | |
669 | ||
7b9337aa NP |
670 | static inline void path_to_nameidata(const struct path *path, |
671 | struct nameidata *nd) | |
051d3812 | 672 | { |
31e6b01f NP |
673 | if (!(nd->flags & LOOKUP_RCU)) { |
674 | dput(nd->path.dentry); | |
675 | if (nd->path.mnt != path->mnt) | |
676 | mntput(nd->path.mnt); | |
9a229683 | 677 | } |
31e6b01f | 678 | nd->path.mnt = path->mnt; |
4ac91378 | 679 | nd->path.dentry = path->dentry; |
051d3812 IK |
680 | } |
681 | ||
b5fb63c1 CH |
682 | /* |
683 | * Helper to directly jump to a known parsed path from ->follow_link, | |
684 | * caller must have taken a reference to path beforehand. | |
685 | */ | |
686 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
687 | { | |
688 | path_put(&nd->path); | |
689 | ||
690 | nd->path = *path; | |
691 | nd->inode = nd->path.dentry->d_inode; | |
692 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
693 | } |
694 | ||
574197e0 AV |
695 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
696 | { | |
697 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 698 | if (inode->i_op->put_link) |
574197e0 AV |
699 | inode->i_op->put_link(link->dentry, nd, cookie); |
700 | path_put(link); | |
701 | } | |
702 | ||
561ec64a LT |
703 | int sysctl_protected_symlinks __read_mostly = 0; |
704 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
705 | |
706 | /** | |
707 | * may_follow_link - Check symlink following for unsafe situations | |
708 | * @link: The path of the symlink | |
55852635 | 709 | * @nd: nameidata pathwalk data |
800179c9 KC |
710 | * |
711 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
712 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
713 | * in a sticky world-writable directory. This is to protect privileged | |
714 | * processes from failing races against path names that may change out | |
715 | * from under them by way of other users creating malicious symlinks. | |
716 | * It will permit symlinks to be followed only when outside a sticky | |
717 | * world-writable directory, or when the uid of the symlink and follower | |
718 | * match, or when the directory owner matches the symlink's owner. | |
719 | * | |
720 | * Returns 0 if following the symlink is allowed, -ve on error. | |
721 | */ | |
722 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
723 | { | |
724 | const struct inode *inode; | |
725 | const struct inode *parent; | |
726 | ||
727 | if (!sysctl_protected_symlinks) | |
728 | return 0; | |
729 | ||
730 | /* Allowed if owner and follower match. */ | |
731 | inode = link->dentry->d_inode; | |
81abe27b | 732 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
733 | return 0; |
734 | ||
735 | /* Allowed if parent directory not sticky and world-writable. */ | |
736 | parent = nd->path.dentry->d_inode; | |
737 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
738 | return 0; | |
739 | ||
740 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 741 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
742 | return 0; |
743 | ||
ffd8d101 | 744 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
745 | path_put_conditional(link, nd); |
746 | path_put(&nd->path); | |
747 | return -EACCES; | |
748 | } | |
749 | ||
750 | /** | |
751 | * safe_hardlink_source - Check for safe hardlink conditions | |
752 | * @inode: the source inode to hardlink from | |
753 | * | |
754 | * Return false if at least one of the following conditions: | |
755 | * - inode is not a regular file | |
756 | * - inode is setuid | |
757 | * - inode is setgid and group-exec | |
758 | * - access failure for read and write | |
759 | * | |
760 | * Otherwise returns true. | |
761 | */ | |
762 | static bool safe_hardlink_source(struct inode *inode) | |
763 | { | |
764 | umode_t mode = inode->i_mode; | |
765 | ||
766 | /* Special files should not get pinned to the filesystem. */ | |
767 | if (!S_ISREG(mode)) | |
768 | return false; | |
769 | ||
770 | /* Setuid files should not get pinned to the filesystem. */ | |
771 | if (mode & S_ISUID) | |
772 | return false; | |
773 | ||
774 | /* Executable setgid files should not get pinned to the filesystem. */ | |
775 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
776 | return false; | |
777 | ||
778 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
779 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
780 | return false; | |
781 | ||
782 | return true; | |
783 | } | |
784 | ||
785 | /** | |
786 | * may_linkat - Check permissions for creating a hardlink | |
787 | * @link: the source to hardlink from | |
788 | * | |
789 | * Block hardlink when all of: | |
790 | * - sysctl_protected_hardlinks enabled | |
791 | * - fsuid does not match inode | |
792 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
793 | * - not CAP_FOWNER | |
794 | * | |
795 | * Returns 0 if successful, -ve on error. | |
796 | */ | |
797 | static int may_linkat(struct path *link) | |
798 | { | |
799 | const struct cred *cred; | |
800 | struct inode *inode; | |
801 | ||
802 | if (!sysctl_protected_hardlinks) | |
803 | return 0; | |
804 | ||
805 | cred = current_cred(); | |
806 | inode = link->dentry->d_inode; | |
807 | ||
808 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
809 | * otherwise, it must be a safe source. | |
810 | */ | |
81abe27b | 811 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
812 | capable(CAP_FOWNER)) |
813 | return 0; | |
814 | ||
a51d9eaa | 815 | audit_log_link_denied("linkat", link); |
800179c9 KC |
816 | return -EPERM; |
817 | } | |
818 | ||
def4af30 | 819 | static __always_inline int |
574197e0 | 820 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 821 | { |
7b9337aa | 822 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
823 | int error; |
824 | char *s; | |
1da177e4 | 825 | |
844a3917 AV |
826 | BUG_ON(nd->flags & LOOKUP_RCU); |
827 | ||
0e794589 AV |
828 | if (link->mnt == nd->path.mnt) |
829 | mntget(link->mnt); | |
830 | ||
6d7b5aae AV |
831 | error = -ELOOP; |
832 | if (unlikely(current->total_link_count >= 40)) | |
833 | goto out_put_nd_path; | |
834 | ||
574197e0 AV |
835 | cond_resched(); |
836 | current->total_link_count++; | |
837 | ||
68ac1234 | 838 | touch_atime(link); |
1da177e4 | 839 | nd_set_link(nd, NULL); |
cd4e91d3 | 840 | |
36f3b4f6 | 841 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
842 | if (error) |
843 | goto out_put_nd_path; | |
36f3b4f6 | 844 | |
86acdca1 | 845 | nd->last_type = LAST_BIND; |
def4af30 AV |
846 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
847 | error = PTR_ERR(*p); | |
6d7b5aae | 848 | if (IS_ERR(*p)) |
408ef013 | 849 | goto out_put_nd_path; |
6d7b5aae AV |
850 | |
851 | error = 0; | |
852 | s = nd_get_link(nd); | |
853 | if (s) { | |
854 | error = __vfs_follow_link(nd, s); | |
b5fb63c1 CH |
855 | if (unlikely(error)) |
856 | put_link(nd, link, *p); | |
1da177e4 | 857 | } |
6d7b5aae AV |
858 | |
859 | return error; | |
860 | ||
861 | out_put_nd_path: | |
98f6ef64 | 862 | *p = NULL; |
6d7b5aae | 863 | path_put(&nd->path); |
6d7b5aae | 864 | path_put(link); |
1da177e4 LT |
865 | return error; |
866 | } | |
867 | ||
31e6b01f NP |
868 | static int follow_up_rcu(struct path *path) |
869 | { | |
0714a533 AV |
870 | struct mount *mnt = real_mount(path->mnt); |
871 | struct mount *parent; | |
31e6b01f NP |
872 | struct dentry *mountpoint; |
873 | ||
0714a533 AV |
874 | parent = mnt->mnt_parent; |
875 | if (&parent->mnt == path->mnt) | |
31e6b01f | 876 | return 0; |
a73324da | 877 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 878 | path->dentry = mountpoint; |
0714a533 | 879 | path->mnt = &parent->mnt; |
31e6b01f NP |
880 | return 1; |
881 | } | |
882 | ||
f015f126 DH |
883 | /* |
884 | * follow_up - Find the mountpoint of path's vfsmount | |
885 | * | |
886 | * Given a path, find the mountpoint of its source file system. | |
887 | * Replace @path with the path of the mountpoint in the parent mount. | |
888 | * Up is towards /. | |
889 | * | |
890 | * Return 1 if we went up a level and 0 if we were already at the | |
891 | * root. | |
892 | */ | |
bab77ebf | 893 | int follow_up(struct path *path) |
1da177e4 | 894 | { |
0714a533 AV |
895 | struct mount *mnt = real_mount(path->mnt); |
896 | struct mount *parent; | |
1da177e4 | 897 | struct dentry *mountpoint; |
99b7db7b | 898 | |
962830df | 899 | br_read_lock(&vfsmount_lock); |
0714a533 | 900 | parent = mnt->mnt_parent; |
3c0a6163 | 901 | if (parent == mnt) { |
962830df | 902 | br_read_unlock(&vfsmount_lock); |
1da177e4 LT |
903 | return 0; |
904 | } | |
0714a533 | 905 | mntget(&parent->mnt); |
a73324da | 906 | mountpoint = dget(mnt->mnt_mountpoint); |
962830df | 907 | br_read_unlock(&vfsmount_lock); |
bab77ebf AV |
908 | dput(path->dentry); |
909 | path->dentry = mountpoint; | |
910 | mntput(path->mnt); | |
0714a533 | 911 | path->mnt = &parent->mnt; |
1da177e4 LT |
912 | return 1; |
913 | } | |
914 | ||
b5c84bf6 | 915 | /* |
9875cf80 DH |
916 | * Perform an automount |
917 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
918 | * were called with. | |
1da177e4 | 919 | */ |
9875cf80 DH |
920 | static int follow_automount(struct path *path, unsigned flags, |
921 | bool *need_mntput) | |
31e6b01f | 922 | { |
9875cf80 | 923 | struct vfsmount *mnt; |
ea5b778a | 924 | int err; |
9875cf80 DH |
925 | |
926 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
927 | return -EREMOTE; | |
928 | ||
0ec26fd0 MS |
929 | /* We don't want to mount if someone's just doing a stat - |
930 | * unless they're stat'ing a directory and appended a '/' to | |
931 | * the name. | |
932 | * | |
933 | * We do, however, want to mount if someone wants to open or | |
934 | * create a file of any type under the mountpoint, wants to | |
935 | * traverse through the mountpoint or wants to open the | |
936 | * mounted directory. Also, autofs may mark negative dentries | |
937 | * as being automount points. These will need the attentions | |
938 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 939 | */ |
0ec26fd0 | 940 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 941 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
942 | path->dentry->d_inode) |
943 | return -EISDIR; | |
944 | ||
9875cf80 DH |
945 | current->total_link_count++; |
946 | if (current->total_link_count >= 40) | |
947 | return -ELOOP; | |
948 | ||
949 | mnt = path->dentry->d_op->d_automount(path); | |
950 | if (IS_ERR(mnt)) { | |
951 | /* | |
952 | * The filesystem is allowed to return -EISDIR here to indicate | |
953 | * it doesn't want to automount. For instance, autofs would do | |
954 | * this so that its userspace daemon can mount on this dentry. | |
955 | * | |
956 | * However, we can only permit this if it's a terminal point in | |
957 | * the path being looked up; if it wasn't then the remainder of | |
958 | * the path is inaccessible and we should say so. | |
959 | */ | |
49084c3b | 960 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
961 | return -EREMOTE; |
962 | return PTR_ERR(mnt); | |
31e6b01f | 963 | } |
ea5b778a | 964 | |
9875cf80 DH |
965 | if (!mnt) /* mount collision */ |
966 | return 0; | |
31e6b01f | 967 | |
8aef1884 AV |
968 | if (!*need_mntput) { |
969 | /* lock_mount() may release path->mnt on error */ | |
970 | mntget(path->mnt); | |
971 | *need_mntput = true; | |
972 | } | |
19a167af | 973 | err = finish_automount(mnt, path); |
9875cf80 | 974 | |
ea5b778a DH |
975 | switch (err) { |
976 | case -EBUSY: | |
977 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 978 | return 0; |
ea5b778a | 979 | case 0: |
8aef1884 | 980 | path_put(path); |
ea5b778a DH |
981 | path->mnt = mnt; |
982 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 983 | return 0; |
19a167af AV |
984 | default: |
985 | return err; | |
ea5b778a | 986 | } |
19a167af | 987 | |
463ffb2e AV |
988 | } |
989 | ||
9875cf80 DH |
990 | /* |
991 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 992 | * - Flagged for transit management (autofs) |
9875cf80 DH |
993 | * - Flagged as mountpoint |
994 | * - Flagged as automount point | |
995 | * | |
996 | * This may only be called in refwalk mode. | |
997 | * | |
998 | * Serialization is taken care of in namespace.c | |
999 | */ | |
1000 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1001 | { |
8aef1884 | 1002 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1003 | unsigned managed; |
1004 | bool need_mntput = false; | |
8aef1884 | 1005 | int ret = 0; |
9875cf80 DH |
1006 | |
1007 | /* Given that we're not holding a lock here, we retain the value in a | |
1008 | * local variable for each dentry as we look at it so that we don't see | |
1009 | * the components of that value change under us */ | |
1010 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1011 | managed &= DCACHE_MANAGED_DENTRY, | |
1012 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1013 | /* Allow the filesystem to manage the transit without i_mutex |
1014 | * being held. */ | |
1015 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1016 | BUG_ON(!path->dentry->d_op); | |
1017 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1018 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1019 | if (ret < 0) |
8aef1884 | 1020 | break; |
cc53ce53 DH |
1021 | } |
1022 | ||
9875cf80 DH |
1023 | /* Transit to a mounted filesystem. */ |
1024 | if (managed & DCACHE_MOUNTED) { | |
1025 | struct vfsmount *mounted = lookup_mnt(path); | |
1026 | if (mounted) { | |
1027 | dput(path->dentry); | |
1028 | if (need_mntput) | |
1029 | mntput(path->mnt); | |
1030 | path->mnt = mounted; | |
1031 | path->dentry = dget(mounted->mnt_root); | |
1032 | need_mntput = true; | |
1033 | continue; | |
1034 | } | |
1035 | ||
1036 | /* Something is mounted on this dentry in another | |
1037 | * namespace and/or whatever was mounted there in this | |
1038 | * namespace got unmounted before we managed to get the | |
1039 | * vfsmount_lock */ | |
1040 | } | |
1041 | ||
1042 | /* Handle an automount point */ | |
1043 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1044 | ret = follow_automount(path, flags, &need_mntput); | |
1045 | if (ret < 0) | |
8aef1884 | 1046 | break; |
9875cf80 DH |
1047 | continue; |
1048 | } | |
1049 | ||
1050 | /* We didn't change the current path point */ | |
1051 | break; | |
1da177e4 | 1052 | } |
8aef1884 AV |
1053 | |
1054 | if (need_mntput && path->mnt == mnt) | |
1055 | mntput(path->mnt); | |
1056 | if (ret == -EISDIR) | |
1057 | ret = 0; | |
a3fbbde7 | 1058 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1059 | } |
1060 | ||
cc53ce53 | 1061 | int follow_down_one(struct path *path) |
1da177e4 LT |
1062 | { |
1063 | struct vfsmount *mounted; | |
1064 | ||
1c755af4 | 1065 | mounted = lookup_mnt(path); |
1da177e4 | 1066 | if (mounted) { |
9393bd07 AV |
1067 | dput(path->dentry); |
1068 | mntput(path->mnt); | |
1069 | path->mnt = mounted; | |
1070 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1071 | return 1; |
1072 | } | |
1073 | return 0; | |
1074 | } | |
1075 | ||
62a7375e IK |
1076 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
1077 | { | |
1078 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
1079 | dentry->d_op->d_manage(dentry, true) < 0); | |
1080 | } | |
1081 | ||
9875cf80 | 1082 | /* |
287548e4 AV |
1083 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1084 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1085 | */ |
1086 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1087 | struct inode **inode) |
9875cf80 | 1088 | { |
62a7375e | 1089 | for (;;) { |
c7105365 | 1090 | struct mount *mounted; |
62a7375e IK |
1091 | /* |
1092 | * Don't forget we might have a non-mountpoint managed dentry | |
1093 | * that wants to block transit. | |
1094 | */ | |
287548e4 | 1095 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 1096 | return false; |
62a7375e IK |
1097 | |
1098 | if (!d_mountpoint(path->dentry)) | |
1099 | break; | |
1100 | ||
9875cf80 DH |
1101 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1102 | if (!mounted) | |
1103 | break; | |
c7105365 AV |
1104 | path->mnt = &mounted->mnt; |
1105 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1106 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1107 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1108 | /* |
1109 | * Update the inode too. We don't need to re-check the | |
1110 | * dentry sequence number here after this d_inode read, | |
1111 | * because a mount-point is always pinned. | |
1112 | */ | |
1113 | *inode = path->dentry->d_inode; | |
9875cf80 | 1114 | } |
9875cf80 DH |
1115 | return true; |
1116 | } | |
1117 | ||
dea39376 | 1118 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 1119 | { |
dea39376 | 1120 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 1121 | struct mount *mounted; |
dea39376 | 1122 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
1123 | if (!mounted) |
1124 | break; | |
c7105365 AV |
1125 | nd->path.mnt = &mounted->mnt; |
1126 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 1127 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
1128 | } |
1129 | } | |
1130 | ||
31e6b01f NP |
1131 | static int follow_dotdot_rcu(struct nameidata *nd) |
1132 | { | |
31e6b01f NP |
1133 | set_root_rcu(nd); |
1134 | ||
9875cf80 | 1135 | while (1) { |
31e6b01f NP |
1136 | if (nd->path.dentry == nd->root.dentry && |
1137 | nd->path.mnt == nd->root.mnt) { | |
1138 | break; | |
1139 | } | |
1140 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1141 | struct dentry *old = nd->path.dentry; | |
1142 | struct dentry *parent = old->d_parent; | |
1143 | unsigned seq; | |
1144 | ||
1145 | seq = read_seqcount_begin(&parent->d_seq); | |
1146 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1147 | goto failed; |
31e6b01f NP |
1148 | nd->path.dentry = parent; |
1149 | nd->seq = seq; | |
1150 | break; | |
1151 | } | |
1152 | if (!follow_up_rcu(&nd->path)) | |
1153 | break; | |
1154 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1155 | } |
dea39376 AV |
1156 | follow_mount_rcu(nd); |
1157 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 1158 | return 0; |
ef7562d5 AV |
1159 | |
1160 | failed: | |
1161 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1162 | if (!(nd->flags & LOOKUP_ROOT)) |
1163 | nd->root.mnt = NULL; | |
32a7991b | 1164 | unlock_rcu_walk(); |
ef7562d5 | 1165 | return -ECHILD; |
31e6b01f NP |
1166 | } |
1167 | ||
cc53ce53 DH |
1168 | /* |
1169 | * Follow down to the covering mount currently visible to userspace. At each | |
1170 | * point, the filesystem owning that dentry may be queried as to whether the | |
1171 | * caller is permitted to proceed or not. | |
cc53ce53 | 1172 | */ |
7cc90cc3 | 1173 | int follow_down(struct path *path) |
cc53ce53 DH |
1174 | { |
1175 | unsigned managed; | |
1176 | int ret; | |
1177 | ||
1178 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1179 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1180 | /* Allow the filesystem to manage the transit without i_mutex | |
1181 | * being held. | |
1182 | * | |
1183 | * We indicate to the filesystem if someone is trying to mount | |
1184 | * something here. This gives autofs the chance to deny anyone | |
1185 | * other than its daemon the right to mount on its | |
1186 | * superstructure. | |
1187 | * | |
1188 | * The filesystem may sleep at this point. | |
1189 | */ | |
1190 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1191 | BUG_ON(!path->dentry->d_op); | |
1192 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1193 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1194 | path->dentry, false); |
cc53ce53 DH |
1195 | if (ret < 0) |
1196 | return ret == -EISDIR ? 0 : ret; | |
1197 | } | |
1198 | ||
1199 | /* Transit to a mounted filesystem. */ | |
1200 | if (managed & DCACHE_MOUNTED) { | |
1201 | struct vfsmount *mounted = lookup_mnt(path); | |
1202 | if (!mounted) | |
1203 | break; | |
1204 | dput(path->dentry); | |
1205 | mntput(path->mnt); | |
1206 | path->mnt = mounted; | |
1207 | path->dentry = dget(mounted->mnt_root); | |
1208 | continue; | |
1209 | } | |
1210 | ||
1211 | /* Don't handle automount points here */ | |
1212 | break; | |
1213 | } | |
1214 | return 0; | |
1215 | } | |
1216 | ||
9875cf80 DH |
1217 | /* |
1218 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1219 | */ | |
1220 | static void follow_mount(struct path *path) | |
1221 | { | |
1222 | while (d_mountpoint(path->dentry)) { | |
1223 | struct vfsmount *mounted = lookup_mnt(path); | |
1224 | if (!mounted) | |
1225 | break; | |
1226 | dput(path->dentry); | |
1227 | mntput(path->mnt); | |
1228 | path->mnt = mounted; | |
1229 | path->dentry = dget(mounted->mnt_root); | |
1230 | } | |
1231 | } | |
1232 | ||
31e6b01f | 1233 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1234 | { |
2a737871 | 1235 | set_root(nd); |
e518ddb7 | 1236 | |
1da177e4 | 1237 | while(1) { |
4ac91378 | 1238 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1239 | |
2a737871 AV |
1240 | if (nd->path.dentry == nd->root.dentry && |
1241 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1242 | break; |
1243 | } | |
4ac91378 | 1244 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1245 | /* rare case of legitimate dget_parent()... */ |
1246 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1247 | dput(old); |
1248 | break; | |
1249 | } | |
3088dd70 | 1250 | if (!follow_up(&nd->path)) |
1da177e4 | 1251 | break; |
1da177e4 | 1252 | } |
79ed0226 | 1253 | follow_mount(&nd->path); |
31e6b01f | 1254 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1255 | } |
1256 | ||
baa03890 | 1257 | /* |
bad61189 MS |
1258 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1259 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1260 | * returns whether i_op->lookup is necessary. | |
1261 | * | |
1262 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1263 | */ |
bad61189 | 1264 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1265 | unsigned int flags, bool *need_lookup) |
baa03890 | 1266 | { |
baa03890 | 1267 | struct dentry *dentry; |
bad61189 | 1268 | int error; |
baa03890 | 1269 | |
bad61189 MS |
1270 | *need_lookup = false; |
1271 | dentry = d_lookup(dir, name); | |
1272 | if (dentry) { | |
39e3c955 | 1273 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1274 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1275 | if (unlikely(error <= 0)) { |
1276 | if (error < 0) { | |
1277 | dput(dentry); | |
1278 | return ERR_PTR(error); | |
1279 | } else if (!d_invalidate(dentry)) { | |
1280 | dput(dentry); | |
1281 | dentry = NULL; | |
1282 | } | |
1283 | } | |
1284 | } | |
1285 | } | |
baa03890 | 1286 | |
bad61189 MS |
1287 | if (!dentry) { |
1288 | dentry = d_alloc(dir, name); | |
1289 | if (unlikely(!dentry)) | |
1290 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1291 | |
bad61189 | 1292 | *need_lookup = true; |
baa03890 NP |
1293 | } |
1294 | return dentry; | |
1295 | } | |
1296 | ||
44396f4b | 1297 | /* |
bad61189 MS |
1298 | * Call i_op->lookup on the dentry. The dentry must be negative but may be |
1299 | * hashed if it was pouplated with DCACHE_NEED_LOOKUP. | |
1300 | * | |
1301 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1302 | */ |
bad61189 | 1303 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1304 | unsigned int flags) |
44396f4b | 1305 | { |
44396f4b JB |
1306 | struct dentry *old; |
1307 | ||
1308 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1309 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1310 | dput(dentry); |
44396f4b | 1311 | return ERR_PTR(-ENOENT); |
e188dc02 | 1312 | } |
44396f4b | 1313 | |
72bd866a | 1314 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1315 | if (unlikely(old)) { |
1316 | dput(dentry); | |
1317 | dentry = old; | |
1318 | } | |
1319 | return dentry; | |
1320 | } | |
1321 | ||
a3255546 | 1322 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1323 | struct dentry *base, unsigned int flags) |
a3255546 | 1324 | { |
bad61189 | 1325 | bool need_lookup; |
a3255546 AV |
1326 | struct dentry *dentry; |
1327 | ||
72bd866a | 1328 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1329 | if (!need_lookup) |
1330 | return dentry; | |
a3255546 | 1331 | |
72bd866a | 1332 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1333 | } |
1334 | ||
1da177e4 LT |
1335 | /* |
1336 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1337 | * small and for now I'd prefer to have fast path as straight as possible. | |
1338 | * It _is_ time-critical. | |
1339 | */ | |
e97cdc87 | 1340 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1341 | struct path *path, struct inode **inode) |
1da177e4 | 1342 | { |
4ac91378 | 1343 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1344 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1345 | int need_reval = 1; |
1346 | int status = 1; | |
9875cf80 DH |
1347 | int err; |
1348 | ||
b04f784e NP |
1349 | /* |
1350 | * Rename seqlock is not required here because in the off chance | |
1351 | * of a false negative due to a concurrent rename, we're going to | |
1352 | * do the non-racy lookup, below. | |
1353 | */ | |
31e6b01f NP |
1354 | if (nd->flags & LOOKUP_RCU) { |
1355 | unsigned seq; | |
e97cdc87 | 1356 | dentry = __d_lookup_rcu(parent, &nd->last, &seq, nd->inode); |
5a18fff2 AV |
1357 | if (!dentry) |
1358 | goto unlazy; | |
1359 | ||
12f8ad4b LT |
1360 | /* |
1361 | * This sequence count validates that the inode matches | |
1362 | * the dentry name information from lookup. | |
1363 | */ | |
1364 | *inode = dentry->d_inode; | |
1365 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1366 | return -ECHILD; | |
1367 | ||
1368 | /* | |
1369 | * This sequence count validates that the parent had no | |
1370 | * changes while we did the lookup of the dentry above. | |
1371 | * | |
1372 | * The memory barrier in read_seqcount_begin of child is | |
1373 | * enough, we can use __read_seqcount_retry here. | |
1374 | */ | |
31e6b01f NP |
1375 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1376 | return -ECHILD; | |
31e6b01f | 1377 | nd->seq = seq; |
5a18fff2 | 1378 | |
24643087 | 1379 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1380 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1381 | if (unlikely(status <= 0)) { |
1382 | if (status != -ECHILD) | |
1383 | need_reval = 0; | |
1384 | goto unlazy; | |
1385 | } | |
24643087 | 1386 | } |
31e6b01f NP |
1387 | path->mnt = mnt; |
1388 | path->dentry = dentry; | |
d6e9bd25 AV |
1389 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1390 | goto unlazy; | |
1391 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1392 | goto unlazy; | |
1393 | return 0; | |
5a18fff2 | 1394 | unlazy: |
19660af7 AV |
1395 | if (unlazy_walk(nd, dentry)) |
1396 | return -ECHILD; | |
5a18fff2 | 1397 | } else { |
e97cdc87 | 1398 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1399 | } |
5a18fff2 | 1400 | |
81e6f520 AV |
1401 | if (unlikely(!dentry)) |
1402 | goto need_lookup; | |
1403 | ||
5a18fff2 | 1404 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1405 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1406 | if (unlikely(status <= 0)) { |
1407 | if (status < 0) { | |
1408 | dput(dentry); | |
1409 | return status; | |
1410 | } | |
1411 | if (!d_invalidate(dentry)) { | |
1412 | dput(dentry); | |
81e6f520 | 1413 | goto need_lookup; |
5a18fff2 | 1414 | } |
24643087 | 1415 | } |
697f514d | 1416 | |
9875cf80 DH |
1417 | path->mnt = mnt; |
1418 | path->dentry = dentry; | |
1419 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1420 | if (unlikely(err < 0)) { |
1421 | path_put_conditional(path, nd); | |
9875cf80 | 1422 | return err; |
89312214 | 1423 | } |
a3fbbde7 AV |
1424 | if (err) |
1425 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1426 | *inode = path->dentry->d_inode; |
1da177e4 | 1427 | return 0; |
81e6f520 AV |
1428 | |
1429 | need_lookup: | |
697f514d MS |
1430 | return 1; |
1431 | } | |
1432 | ||
1433 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1434 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1435 | { |
1436 | struct dentry *dentry, *parent; | |
1437 | int err; | |
1438 | ||
1439 | parent = nd->path.dentry; | |
81e6f520 AV |
1440 | BUG_ON(nd->inode != parent->d_inode); |
1441 | ||
1442 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1443 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1444 | mutex_unlock(&parent->d_inode->i_mutex); |
1445 | if (IS_ERR(dentry)) | |
1446 | return PTR_ERR(dentry); | |
697f514d MS |
1447 | path->mnt = nd->path.mnt; |
1448 | path->dentry = dentry; | |
1449 | err = follow_managed(path, nd->flags); | |
1450 | if (unlikely(err < 0)) { | |
1451 | path_put_conditional(path, nd); | |
1452 | return err; | |
1453 | } | |
1454 | if (err) | |
1455 | nd->flags |= LOOKUP_JUMPED; | |
1456 | return 0; | |
1da177e4 LT |
1457 | } |
1458 | ||
52094c8a AV |
1459 | static inline int may_lookup(struct nameidata *nd) |
1460 | { | |
1461 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1462 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1463 | if (err != -ECHILD) |
1464 | return err; | |
19660af7 | 1465 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1466 | return -ECHILD; |
1467 | } | |
4ad5abb3 | 1468 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1469 | } |
1470 | ||
9856fa1b AV |
1471 | static inline int handle_dots(struct nameidata *nd, int type) |
1472 | { | |
1473 | if (type == LAST_DOTDOT) { | |
1474 | if (nd->flags & LOOKUP_RCU) { | |
1475 | if (follow_dotdot_rcu(nd)) | |
1476 | return -ECHILD; | |
1477 | } else | |
1478 | follow_dotdot(nd); | |
1479 | } | |
1480 | return 0; | |
1481 | } | |
1482 | ||
951361f9 AV |
1483 | static void terminate_walk(struct nameidata *nd) |
1484 | { | |
1485 | if (!(nd->flags & LOOKUP_RCU)) { | |
1486 | path_put(&nd->path); | |
1487 | } else { | |
1488 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1489 | if (!(nd->flags & LOOKUP_ROOT)) |
1490 | nd->root.mnt = NULL; | |
32a7991b | 1491 | unlock_rcu_walk(); |
951361f9 AV |
1492 | } |
1493 | } | |
1494 | ||
3ddcd056 LT |
1495 | /* |
1496 | * Do we need to follow links? We _really_ want to be able | |
1497 | * to do this check without having to look at inode->i_op, | |
1498 | * so we keep a cache of "no, this doesn't need follow_link" | |
1499 | * for the common case. | |
1500 | */ | |
7813b94a | 1501 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1502 | { |
1503 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1504 | if (likely(inode->i_op->follow_link)) | |
1505 | return follow; | |
1506 | ||
1507 | /* This gets set once for the inode lifetime */ | |
1508 | spin_lock(&inode->i_lock); | |
1509 | inode->i_opflags |= IOP_NOFOLLOW; | |
1510 | spin_unlock(&inode->i_lock); | |
1511 | } | |
1512 | return 0; | |
1513 | } | |
1514 | ||
ce57dfc1 | 1515 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1516 | int follow) |
ce57dfc1 AV |
1517 | { |
1518 | struct inode *inode; | |
1519 | int err; | |
1520 | /* | |
1521 | * "." and ".." are special - ".." especially so because it has | |
1522 | * to be able to know about the current root directory and | |
1523 | * parent relationships. | |
1524 | */ | |
21b9b073 AV |
1525 | if (unlikely(nd->last_type != LAST_NORM)) |
1526 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1527 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1528 | if (unlikely(err)) { |
697f514d MS |
1529 | if (err < 0) |
1530 | goto out_err; | |
1531 | ||
cc2a5271 | 1532 | err = lookup_slow(nd, path); |
697f514d MS |
1533 | if (err < 0) |
1534 | goto out_err; | |
1535 | ||
1536 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1537 | } |
697f514d MS |
1538 | err = -ENOENT; |
1539 | if (!inode) | |
1540 | goto out_path_put; | |
1541 | ||
7813b94a | 1542 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1543 | if (nd->flags & LOOKUP_RCU) { |
1544 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1545 | err = -ECHILD; |
1546 | goto out_err; | |
19660af7 AV |
1547 | } |
1548 | } | |
ce57dfc1 AV |
1549 | BUG_ON(inode != path->dentry->d_inode); |
1550 | return 1; | |
1551 | } | |
1552 | path_to_nameidata(path, nd); | |
1553 | nd->inode = inode; | |
1554 | return 0; | |
697f514d MS |
1555 | |
1556 | out_path_put: | |
1557 | path_to_nameidata(path, nd); | |
1558 | out_err: | |
1559 | terminate_walk(nd); | |
1560 | return err; | |
ce57dfc1 AV |
1561 | } |
1562 | ||
b356379a AV |
1563 | /* |
1564 | * This limits recursive symlink follows to 8, while | |
1565 | * limiting consecutive symlinks to 40. | |
1566 | * | |
1567 | * Without that kind of total limit, nasty chains of consecutive | |
1568 | * symlinks can cause almost arbitrarily long lookups. | |
1569 | */ | |
1570 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1571 | { | |
1572 | int res; | |
1573 | ||
b356379a AV |
1574 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1575 | path_put_conditional(path, nd); | |
1576 | path_put(&nd->path); | |
1577 | return -ELOOP; | |
1578 | } | |
1a4022f8 | 1579 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1580 | |
1581 | nd->depth++; | |
1582 | current->link_count++; | |
1583 | ||
1584 | do { | |
1585 | struct path link = *path; | |
1586 | void *cookie; | |
574197e0 AV |
1587 | |
1588 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1589 | if (res) |
1590 | break; | |
21b9b073 | 1591 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1592 | put_link(nd, &link, cookie); |
b356379a AV |
1593 | } while (res > 0); |
1594 | ||
1595 | current->link_count--; | |
1596 | nd->depth--; | |
1597 | return res; | |
1598 | } | |
1599 | ||
3ddcd056 LT |
1600 | /* |
1601 | * We really don't want to look at inode->i_op->lookup | |
1602 | * when we don't have to. So we keep a cache bit in | |
1603 | * the inode ->i_opflags field that says "yes, we can | |
1604 | * do lookup on this inode". | |
1605 | */ | |
1606 | static inline int can_lookup(struct inode *inode) | |
1607 | { | |
1608 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1609 | return 1; | |
1610 | if (likely(!inode->i_op->lookup)) | |
1611 | return 0; | |
1612 | ||
1613 | /* We do this once for the lifetime of the inode */ | |
1614 | spin_lock(&inode->i_lock); | |
1615 | inode->i_opflags |= IOP_LOOKUP; | |
1616 | spin_unlock(&inode->i_lock); | |
1617 | return 1; | |
1618 | } | |
1619 | ||
bfcfaa77 LT |
1620 | /* |
1621 | * We can do the critical dentry name comparison and hashing | |
1622 | * operations one word at a time, but we are limited to: | |
1623 | * | |
1624 | * - Architectures with fast unaligned word accesses. We could | |
1625 | * do a "get_unaligned()" if this helps and is sufficiently | |
1626 | * fast. | |
1627 | * | |
1628 | * - Little-endian machines (so that we can generate the mask | |
1629 | * of low bytes efficiently). Again, we *could* do a byte | |
1630 | * swapping load on big-endian architectures if that is not | |
1631 | * expensive enough to make the optimization worthless. | |
1632 | * | |
1633 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1634 | * do not trap on the (extremely unlikely) case of a page | |
1635 | * crossing operation. | |
1636 | * | |
1637 | * - Furthermore, we need an efficient 64-bit compile for the | |
1638 | * 64-bit case in order to generate the "number of bytes in | |
1639 | * the final mask". Again, that could be replaced with a | |
1640 | * efficient population count instruction or similar. | |
1641 | */ | |
1642 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1643 | ||
f68e556e | 1644 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1645 | |
f68e556e | 1646 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1647 | |
1648 | static inline unsigned int fold_hash(unsigned long hash) | |
1649 | { | |
1650 | hash += hash >> (8*sizeof(int)); | |
1651 | return hash; | |
1652 | } | |
1653 | ||
1654 | #else /* 32-bit case */ | |
1655 | ||
bfcfaa77 LT |
1656 | #define fold_hash(x) (x) |
1657 | ||
1658 | #endif | |
1659 | ||
1660 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1661 | { | |
1662 | unsigned long a, mask; | |
1663 | unsigned long hash = 0; | |
1664 | ||
1665 | for (;;) { | |
e419b4cc | 1666 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1667 | if (len < sizeof(unsigned long)) |
1668 | break; | |
1669 | hash += a; | |
f132c5be | 1670 | hash *= 9; |
bfcfaa77 LT |
1671 | name += sizeof(unsigned long); |
1672 | len -= sizeof(unsigned long); | |
1673 | if (!len) | |
1674 | goto done; | |
1675 | } | |
1676 | mask = ~(~0ul << len*8); | |
1677 | hash += mask & a; | |
1678 | done: | |
1679 | return fold_hash(hash); | |
1680 | } | |
1681 | EXPORT_SYMBOL(full_name_hash); | |
1682 | ||
bfcfaa77 LT |
1683 | /* |
1684 | * Calculate the length and hash of the path component, and | |
1685 | * return the length of the component; | |
1686 | */ | |
1687 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1688 | { | |
36126f8f LT |
1689 | unsigned long a, b, adata, bdata, mask, hash, len; |
1690 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1691 | |
1692 | hash = a = 0; | |
1693 | len = -sizeof(unsigned long); | |
1694 | do { | |
1695 | hash = (hash + a) * 9; | |
1696 | len += sizeof(unsigned long); | |
e419b4cc | 1697 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1698 | b = a ^ REPEAT_BYTE('/'); |
1699 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1700 | ||
1701 | adata = prep_zero_mask(a, adata, &constants); | |
1702 | bdata = prep_zero_mask(b, bdata, &constants); | |
1703 | ||
1704 | mask = create_zero_mask(adata | bdata); | |
1705 | ||
1706 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1707 | *hashp = fold_hash(hash); |
1708 | ||
36126f8f | 1709 | return len + find_zero(mask); |
bfcfaa77 LT |
1710 | } |
1711 | ||
1712 | #else | |
1713 | ||
0145acc2 LT |
1714 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1715 | { | |
1716 | unsigned long hash = init_name_hash(); | |
1717 | while (len--) | |
1718 | hash = partial_name_hash(*name++, hash); | |
1719 | return end_name_hash(hash); | |
1720 | } | |
ae942ae7 | 1721 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1722 | |
200e9ef7 LT |
1723 | /* |
1724 | * We know there's a real path component here of at least | |
1725 | * one character. | |
1726 | */ | |
1727 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1728 | { | |
1729 | unsigned long hash = init_name_hash(); | |
1730 | unsigned long len = 0, c; | |
1731 | ||
1732 | c = (unsigned char)*name; | |
1733 | do { | |
1734 | len++; | |
1735 | hash = partial_name_hash(c, hash); | |
1736 | c = (unsigned char)name[len]; | |
1737 | } while (c && c != '/'); | |
1738 | *hashp = end_name_hash(hash); | |
1739 | return len; | |
1740 | } | |
1741 | ||
bfcfaa77 LT |
1742 | #endif |
1743 | ||
1da177e4 LT |
1744 | /* |
1745 | * Name resolution. | |
ea3834d9 PM |
1746 | * This is the basic name resolution function, turning a pathname into |
1747 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1748 | * |
ea3834d9 PM |
1749 | * Returns 0 and nd will have valid dentry and mnt on success. |
1750 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1751 | */ |
6de88d72 | 1752 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1753 | { |
1754 | struct path next; | |
1da177e4 | 1755 | int err; |
1da177e4 LT |
1756 | |
1757 | while (*name=='/') | |
1758 | name++; | |
1759 | if (!*name) | |
086e183a | 1760 | return 0; |
1da177e4 | 1761 | |
1da177e4 LT |
1762 | /* At this point we know we have a real path component. */ |
1763 | for(;;) { | |
1da177e4 | 1764 | struct qstr this; |
200e9ef7 | 1765 | long len; |
fe479a58 | 1766 | int type; |
1da177e4 | 1767 | |
52094c8a | 1768 | err = may_lookup(nd); |
1da177e4 LT |
1769 | if (err) |
1770 | break; | |
1771 | ||
200e9ef7 | 1772 | len = hash_name(name, &this.hash); |
1da177e4 | 1773 | this.name = name; |
200e9ef7 | 1774 | this.len = len; |
1da177e4 | 1775 | |
fe479a58 | 1776 | type = LAST_NORM; |
200e9ef7 | 1777 | if (name[0] == '.') switch (len) { |
fe479a58 | 1778 | case 2: |
200e9ef7 | 1779 | if (name[1] == '.') { |
fe479a58 | 1780 | type = LAST_DOTDOT; |
16c2cd71 AV |
1781 | nd->flags |= LOOKUP_JUMPED; |
1782 | } | |
fe479a58 AV |
1783 | break; |
1784 | case 1: | |
1785 | type = LAST_DOT; | |
1786 | } | |
5a202bcd AV |
1787 | if (likely(type == LAST_NORM)) { |
1788 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1789 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd AV |
1790 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1791 | err = parent->d_op->d_hash(parent, nd->inode, | |
1792 | &this); | |
1793 | if (err < 0) | |
1794 | break; | |
1795 | } | |
1796 | } | |
fe479a58 | 1797 | |
5f4a6a69 AV |
1798 | nd->last = this; |
1799 | nd->last_type = type; | |
1800 | ||
200e9ef7 | 1801 | if (!name[len]) |
5f4a6a69 | 1802 | return 0; |
200e9ef7 LT |
1803 | /* |
1804 | * If it wasn't NUL, we know it was '/'. Skip that | |
1805 | * slash, and continue until no more slashes. | |
1806 | */ | |
1807 | do { | |
1808 | len++; | |
1809 | } while (unlikely(name[len] == '/')); | |
1810 | if (!name[len]) | |
5f4a6a69 AV |
1811 | return 0; |
1812 | ||
200e9ef7 | 1813 | name += len; |
1da177e4 | 1814 | |
21b9b073 | 1815 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1816 | if (err < 0) |
1817 | return err; | |
1da177e4 | 1818 | |
ce57dfc1 | 1819 | if (err) { |
b356379a | 1820 | err = nested_symlink(&next, nd); |
1da177e4 | 1821 | if (err) |
a7472bab | 1822 | return err; |
31e6b01f | 1823 | } |
5f4a6a69 AV |
1824 | if (!can_lookup(nd->inode)) { |
1825 | err = -ENOTDIR; | |
1826 | break; | |
1827 | } | |
1da177e4 | 1828 | } |
951361f9 | 1829 | terminate_walk(nd); |
1da177e4 LT |
1830 | return err; |
1831 | } | |
1832 | ||
70e9b357 AV |
1833 | static int path_init(int dfd, const char *name, unsigned int flags, |
1834 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1835 | { |
1836 | int retval = 0; | |
31e6b01f NP |
1837 | |
1838 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1839 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1840 | nd->depth = 0; |
5b6ca027 AV |
1841 | if (flags & LOOKUP_ROOT) { |
1842 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 | 1843 | if (*name) { |
741b7c3f | 1844 | if (!can_lookup(inode)) |
73d049a4 AV |
1845 | return -ENOTDIR; |
1846 | retval = inode_permission(inode, MAY_EXEC); | |
1847 | if (retval) | |
1848 | return retval; | |
1849 | } | |
5b6ca027 AV |
1850 | nd->path = nd->root; |
1851 | nd->inode = inode; | |
1852 | if (flags & LOOKUP_RCU) { | |
32a7991b | 1853 | lock_rcu_walk(); |
5b6ca027 AV |
1854 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1855 | } else { | |
1856 | path_get(&nd->path); | |
1857 | } | |
1858 | return 0; | |
1859 | } | |
1860 | ||
31e6b01f | 1861 | nd->root.mnt = NULL; |
31e6b01f NP |
1862 | |
1863 | if (*name=='/') { | |
e41f7d4e | 1864 | if (flags & LOOKUP_RCU) { |
32a7991b | 1865 | lock_rcu_walk(); |
e41f7d4e AV |
1866 | set_root_rcu(nd); |
1867 | } else { | |
1868 | set_root(nd); | |
1869 | path_get(&nd->root); | |
1870 | } | |
1871 | nd->path = nd->root; | |
31e6b01f | 1872 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1873 | if (flags & LOOKUP_RCU) { |
1874 | struct fs_struct *fs = current->fs; | |
1875 | unsigned seq; | |
31e6b01f | 1876 | |
32a7991b | 1877 | lock_rcu_walk(); |
c28cc364 | 1878 | |
e41f7d4e AV |
1879 | do { |
1880 | seq = read_seqcount_begin(&fs->seq); | |
1881 | nd->path = fs->pwd; | |
1882 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1883 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1884 | } else { | |
1885 | get_fs_pwd(current->fs, &nd->path); | |
1886 | } | |
31e6b01f | 1887 | } else { |
582aa64a | 1888 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1889 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1890 | struct dentry *dentry; |
1891 | ||
2903ff01 AV |
1892 | if (!f.file) |
1893 | return -EBADF; | |
31e6b01f | 1894 | |
2903ff01 | 1895 | dentry = f.file->f_path.dentry; |
31e6b01f | 1896 | |
f52e0c11 | 1897 | if (*name) { |
741b7c3f | 1898 | if (!can_lookup(dentry->d_inode)) { |
2903ff01 AV |
1899 | fdput(f); |
1900 | return -ENOTDIR; | |
1901 | } | |
f52e0c11 | 1902 | } |
31e6b01f | 1903 | |
2903ff01 | 1904 | nd->path = f.file->f_path; |
e41f7d4e | 1905 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1906 | if (f.need_put) |
1907 | *fp = f.file; | |
e41f7d4e | 1908 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
32a7991b | 1909 | lock_rcu_walk(); |
e41f7d4e | 1910 | } else { |
2903ff01 AV |
1911 | path_get(&nd->path); |
1912 | fdput(f); | |
e41f7d4e | 1913 | } |
31e6b01f | 1914 | } |
31e6b01f | 1915 | |
31e6b01f | 1916 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1917 | return 0; |
9b4a9b14 AV |
1918 | } |
1919 | ||
bd92d7fe AV |
1920 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1921 | { | |
1922 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1923 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1924 | ||
1925 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1926 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1927 | } |
1928 | ||
9b4a9b14 | 1929 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1930 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1931 | unsigned int flags, struct nameidata *nd) |
1932 | { | |
70e9b357 | 1933 | struct file *base = NULL; |
bd92d7fe AV |
1934 | struct path path; |
1935 | int err; | |
31e6b01f NP |
1936 | |
1937 | /* | |
1938 | * Path walking is largely split up into 2 different synchronisation | |
1939 | * schemes, rcu-walk and ref-walk (explained in | |
1940 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1941 | * path walk code, but some things particularly setup, cleanup, and | |
1942 | * following mounts are sufficiently divergent that functions are | |
1943 | * duplicated. Typically there is a function foo(), and its RCU | |
1944 | * analogue, foo_rcu(). | |
1945 | * | |
1946 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1947 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1948 | * be handled by restarting a traditional ref-walk (which will always | |
1949 | * be able to complete). | |
1950 | */ | |
bd92d7fe | 1951 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1952 | |
bd92d7fe AV |
1953 | if (unlikely(err)) |
1954 | return err; | |
ee0827cd AV |
1955 | |
1956 | current->total_link_count = 0; | |
bd92d7fe AV |
1957 | err = link_path_walk(name, nd); |
1958 | ||
1959 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1960 | err = lookup_last(nd, &path); |
1961 | while (err > 0) { | |
1962 | void *cookie; | |
1963 | struct path link = path; | |
800179c9 KC |
1964 | err = may_follow_link(&link, nd); |
1965 | if (unlikely(err)) | |
1966 | break; | |
bd92d7fe | 1967 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1968 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1969 | if (err) |
1970 | break; | |
1971 | err = lookup_last(nd, &path); | |
574197e0 | 1972 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1973 | } |
1974 | } | |
ee0827cd | 1975 | |
9f1fafee AV |
1976 | if (!err) |
1977 | err = complete_walk(nd); | |
bd92d7fe AV |
1978 | |
1979 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
05252901 | 1980 | if (!can_lookup(nd->inode)) { |
bd92d7fe | 1981 | path_put(&nd->path); |
bd23a539 | 1982 | err = -ENOTDIR; |
bd92d7fe AV |
1983 | } |
1984 | } | |
16c2cd71 | 1985 | |
70e9b357 AV |
1986 | if (base) |
1987 | fput(base); | |
ee0827cd | 1988 | |
5b6ca027 | 1989 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1990 | path_put(&nd->root); |
1991 | nd->root.mnt = NULL; | |
1992 | } | |
bd92d7fe | 1993 | return err; |
ee0827cd | 1994 | } |
31e6b01f | 1995 | |
873f1eed | 1996 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
1997 | unsigned int flags, struct nameidata *nd) |
1998 | { | |
873f1eed | 1999 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 2000 | if (unlikely(retval == -ECHILD)) |
873f1eed | 2001 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 2002 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
2003 | retval = path_lookupat(dfd, name->name, |
2004 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 2005 | |
f78570dd | 2006 | if (likely(!retval)) |
adb5c247 | 2007 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 2008 | return retval; |
1da177e4 LT |
2009 | } |
2010 | ||
873f1eed JL |
2011 | static int do_path_lookup(int dfd, const char *name, |
2012 | unsigned int flags, struct nameidata *nd) | |
2013 | { | |
2014 | struct filename filename = { .name = name }; | |
2015 | ||
2016 | return filename_lookup(dfd, &filename, flags, nd); | |
2017 | } | |
2018 | ||
79714f72 AV |
2019 | /* does lookup, returns the object with parent locked */ |
2020 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2021 | { |
79714f72 AV |
2022 | struct nameidata nd; |
2023 | struct dentry *d; | |
2024 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2025 | if (err) | |
2026 | return ERR_PTR(err); | |
2027 | if (nd.last_type != LAST_NORM) { | |
2028 | path_put(&nd.path); | |
2029 | return ERR_PTR(-EINVAL); | |
2030 | } | |
2031 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2032 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2033 | if (IS_ERR(d)) { |
2034 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2035 | path_put(&nd.path); | |
2036 | return d; | |
2037 | } | |
2038 | *path = nd.path; | |
2039 | return d; | |
5590ff0d UD |
2040 | } |
2041 | ||
d1811465 AV |
2042 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2043 | { | |
2044 | struct nameidata nd; | |
2045 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2046 | if (!res) | |
2047 | *path = nd.path; | |
2048 | return res; | |
2049 | } | |
2050 | ||
16f18200 JJS |
2051 | /** |
2052 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2053 | * @dentry: pointer to dentry of the base directory | |
2054 | * @mnt: pointer to vfs mount of the base directory | |
2055 | * @name: pointer to file name | |
2056 | * @flags: lookup flags | |
e0a01249 | 2057 | * @path: pointer to struct path to fill |
16f18200 JJS |
2058 | */ |
2059 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2060 | const char *name, unsigned int flags, | |
e0a01249 | 2061 | struct path *path) |
16f18200 | 2062 | { |
e0a01249 AV |
2063 | struct nameidata nd; |
2064 | int err; | |
2065 | nd.root.dentry = dentry; | |
2066 | nd.root.mnt = mnt; | |
2067 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2068 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2069 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2070 | if (!err) | |
2071 | *path = nd.path; | |
2072 | return err; | |
16f18200 JJS |
2073 | } |
2074 | ||
057f6c01 JM |
2075 | /* |
2076 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2077 | * needs parent already locked. Doesn't follow mounts. | |
2078 | * SMP-safe. | |
2079 | */ | |
eead1911 | 2080 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2081 | { |
72bd866a | 2082 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2083 | } |
2084 | ||
eead1911 | 2085 | /** |
a6b91919 | 2086 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2087 | * @name: pathname component to lookup |
2088 | * @base: base directory to lookup from | |
2089 | * @len: maximum length @len should be interpreted to | |
2090 | * | |
a6b91919 RD |
2091 | * Note that this routine is purely a helper for filesystem usage and should |
2092 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2093 | * nameidata argument is passed to the filesystem methods and a filesystem |
2094 | * using this helper needs to be prepared for that. | |
2095 | */ | |
057f6c01 JM |
2096 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2097 | { | |
057f6c01 | 2098 | struct qstr this; |
6a96ba54 | 2099 | unsigned int c; |
cda309de | 2100 | int err; |
057f6c01 | 2101 | |
2f9092e1 DW |
2102 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2103 | ||
6a96ba54 AV |
2104 | this.name = name; |
2105 | this.len = len; | |
0145acc2 | 2106 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2107 | if (!len) |
2108 | return ERR_PTR(-EACCES); | |
2109 | ||
21d8a15a AV |
2110 | if (unlikely(name[0] == '.')) { |
2111 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2112 | return ERR_PTR(-EACCES); | |
2113 | } | |
2114 | ||
6a96ba54 AV |
2115 | while (len--) { |
2116 | c = *(const unsigned char *)name++; | |
2117 | if (c == '/' || c == '\0') | |
2118 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2119 | } |
5a202bcd AV |
2120 | /* |
2121 | * See if the low-level filesystem might want | |
2122 | * to use its own hash.. | |
2123 | */ | |
2124 | if (base->d_flags & DCACHE_OP_HASH) { | |
2125 | int err = base->d_op->d_hash(base, base->d_inode, &this); | |
2126 | if (err < 0) | |
2127 | return ERR_PTR(err); | |
2128 | } | |
eead1911 | 2129 | |
cda309de MS |
2130 | err = inode_permission(base->d_inode, MAY_EXEC); |
2131 | if (err) | |
2132 | return ERR_PTR(err); | |
2133 | ||
72bd866a | 2134 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2135 | } |
2136 | ||
1fa1e7f6 AW |
2137 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2138 | struct path *path, int *empty) | |
1da177e4 | 2139 | { |
2d8f3038 | 2140 | struct nameidata nd; |
91a27b2a | 2141 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2142 | int err = PTR_ERR(tmp); |
1da177e4 | 2143 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2144 | |
2145 | BUG_ON(flags & LOOKUP_PARENT); | |
2146 | ||
873f1eed | 2147 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2148 | putname(tmp); |
2d8f3038 AV |
2149 | if (!err) |
2150 | *path = nd.path; | |
1da177e4 LT |
2151 | } |
2152 | return err; | |
2153 | } | |
2154 | ||
1fa1e7f6 AW |
2155 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2156 | struct path *path) | |
2157 | { | |
f7493e5d | 2158 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2159 | } |
2160 | ||
873f1eed JL |
2161 | /* |
2162 | * NB: most callers don't do anything directly with the reference to the | |
2163 | * to struct filename, but the nd->last pointer points into the name string | |
2164 | * allocated by getname. So we must hold the reference to it until all | |
2165 | * path-walking is complete. | |
2166 | */ | |
91a27b2a | 2167 | static struct filename * |
9e790bd6 JL |
2168 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2169 | unsigned int flags) | |
2ad94ae6 | 2170 | { |
91a27b2a | 2171 | struct filename *s = getname(path); |
2ad94ae6 AV |
2172 | int error; |
2173 | ||
9e790bd6 JL |
2174 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2175 | flags &= LOOKUP_REVAL; | |
2176 | ||
2ad94ae6 | 2177 | if (IS_ERR(s)) |
91a27b2a | 2178 | return s; |
2ad94ae6 | 2179 | |
9e790bd6 | 2180 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2181 | if (error) { |
2ad94ae6 | 2182 | putname(s); |
91a27b2a JL |
2183 | return ERR_PTR(error); |
2184 | } | |
2ad94ae6 | 2185 | |
91a27b2a | 2186 | return s; |
2ad94ae6 AV |
2187 | } |
2188 | ||
1da177e4 LT |
2189 | /* |
2190 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2191 | * minimal. | |
2192 | */ | |
2193 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2194 | { | |
8e96e3b7 | 2195 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2196 | |
1da177e4 LT |
2197 | if (!(dir->i_mode & S_ISVTX)) |
2198 | return 0; | |
8e96e3b7 | 2199 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2200 | return 0; |
8e96e3b7 | 2201 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2202 | return 0; |
4f80c6c1 | 2203 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 LT |
2204 | } |
2205 | ||
2206 | /* | |
2207 | * Check whether we can remove a link victim from directory dir, check | |
2208 | * whether the type of victim is right. | |
2209 | * 1. We can't do it if dir is read-only (done in permission()) | |
2210 | * 2. We should have write and exec permissions on dir | |
2211 | * 3. We can't remove anything from append-only dir | |
2212 | * 4. We can't do anything with immutable dir (done in permission()) | |
2213 | * 5. If the sticky bit on dir is set we should either | |
2214 | * a. be owner of dir, or | |
2215 | * b. be owner of victim, or | |
2216 | * c. have CAP_FOWNER capability | |
2217 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2218 | * links pointing to it. | |
2219 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2220 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2221 | * 9. We can't remove a root or mountpoint. | |
2222 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2223 | * nfs_async_unlink(). | |
2224 | */ | |
858119e1 | 2225 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
2226 | { |
2227 | int error; | |
2228 | ||
2229 | if (!victim->d_inode) | |
2230 | return -ENOENT; | |
2231 | ||
2232 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2233 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2234 | |
f419a2e3 | 2235 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2236 | if (error) |
2237 | return error; | |
2238 | if (IS_APPEND(dir)) | |
2239 | return -EPERM; | |
2240 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2241 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2242 | return -EPERM; |
2243 | if (isdir) { | |
2244 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2245 | return -ENOTDIR; | |
2246 | if (IS_ROOT(victim)) | |
2247 | return -EBUSY; | |
2248 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2249 | return -EISDIR; | |
2250 | if (IS_DEADDIR(dir)) | |
2251 | return -ENOENT; | |
2252 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2253 | return -EBUSY; | |
2254 | return 0; | |
2255 | } | |
2256 | ||
2257 | /* Check whether we can create an object with dentry child in directory | |
2258 | * dir. | |
2259 | * 1. We can't do it if child already exists (open has special treatment for | |
2260 | * this case, but since we are inlined it's OK) | |
2261 | * 2. We can't do it if dir is read-only (done in permission()) | |
2262 | * 3. We should have write and exec permissions on dir | |
2263 | * 4. We can't do it if dir is immutable (done in permission()) | |
2264 | */ | |
a95164d9 | 2265 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2266 | { |
1ec12a81 | 2267 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2268 | if (child->d_inode) |
2269 | return -EEXIST; | |
2270 | if (IS_DEADDIR(dir)) | |
2271 | return -ENOENT; | |
f419a2e3 | 2272 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2273 | } |
2274 | ||
1da177e4 LT |
2275 | /* |
2276 | * p1 and p2 should be directories on the same fs. | |
2277 | */ | |
2278 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2279 | { | |
2280 | struct dentry *p; | |
2281 | ||
2282 | if (p1 == p2) { | |
f2eace23 | 2283 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2284 | return NULL; |
2285 | } | |
2286 | ||
a11f3a05 | 2287 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2288 | |
e2761a11 OH |
2289 | p = d_ancestor(p2, p1); |
2290 | if (p) { | |
2291 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2292 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2293 | return p; | |
1da177e4 LT |
2294 | } |
2295 | ||
e2761a11 OH |
2296 | p = d_ancestor(p1, p2); |
2297 | if (p) { | |
2298 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2299 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2300 | return p; | |
1da177e4 LT |
2301 | } |
2302 | ||
f2eace23 IM |
2303 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2304 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2305 | return NULL; |
2306 | } | |
2307 | ||
2308 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2309 | { | |
1b1dcc1b | 2310 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2311 | if (p1 != p2) { |
1b1dcc1b | 2312 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2313 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2314 | } |
2315 | } | |
2316 | ||
4acdaf27 | 2317 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2318 | bool want_excl) |
1da177e4 | 2319 | { |
a95164d9 | 2320 | int error = may_create(dir, dentry); |
1da177e4 LT |
2321 | if (error) |
2322 | return error; | |
2323 | ||
acfa4380 | 2324 | if (!dir->i_op->create) |
1da177e4 LT |
2325 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2326 | mode &= S_IALLUGO; | |
2327 | mode |= S_IFREG; | |
2328 | error = security_inode_create(dir, dentry, mode); | |
2329 | if (error) | |
2330 | return error; | |
312b63fb | 2331 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2332 | if (!error) |
f38aa942 | 2333 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2334 | return error; |
2335 | } | |
2336 | ||
73d049a4 | 2337 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2338 | { |
3fb64190 | 2339 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2340 | struct inode *inode = dentry->d_inode; |
2341 | int error; | |
2342 | ||
bcda7652 AV |
2343 | /* O_PATH? */ |
2344 | if (!acc_mode) | |
2345 | return 0; | |
2346 | ||
1da177e4 LT |
2347 | if (!inode) |
2348 | return -ENOENT; | |
2349 | ||
c8fe8f30 CH |
2350 | switch (inode->i_mode & S_IFMT) { |
2351 | case S_IFLNK: | |
1da177e4 | 2352 | return -ELOOP; |
c8fe8f30 CH |
2353 | case S_IFDIR: |
2354 | if (acc_mode & MAY_WRITE) | |
2355 | return -EISDIR; | |
2356 | break; | |
2357 | case S_IFBLK: | |
2358 | case S_IFCHR: | |
3fb64190 | 2359 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2360 | return -EACCES; |
c8fe8f30 CH |
2361 | /*FALLTHRU*/ |
2362 | case S_IFIFO: | |
2363 | case S_IFSOCK: | |
1da177e4 | 2364 | flag &= ~O_TRUNC; |
c8fe8f30 | 2365 | break; |
4a3fd211 | 2366 | } |
b41572e9 | 2367 | |
3fb64190 | 2368 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2369 | if (error) |
2370 | return error; | |
6146f0d5 | 2371 | |
1da177e4 LT |
2372 | /* |
2373 | * An append-only file must be opened in append mode for writing. | |
2374 | */ | |
2375 | if (IS_APPEND(inode)) { | |
8737c930 | 2376 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2377 | return -EPERM; |
1da177e4 | 2378 | if (flag & O_TRUNC) |
7715b521 | 2379 | return -EPERM; |
1da177e4 LT |
2380 | } |
2381 | ||
2382 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2383 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2384 | return -EPERM; |
1da177e4 | 2385 | |
f3c7691e | 2386 | return 0; |
7715b521 | 2387 | } |
1da177e4 | 2388 | |
e1181ee6 | 2389 | static int handle_truncate(struct file *filp) |
7715b521 | 2390 | { |
e1181ee6 | 2391 | struct path *path = &filp->f_path; |
7715b521 AV |
2392 | struct inode *inode = path->dentry->d_inode; |
2393 | int error = get_write_access(inode); | |
2394 | if (error) | |
2395 | return error; | |
2396 | /* | |
2397 | * Refuse to truncate files with mandatory locks held on them. | |
2398 | */ | |
2399 | error = locks_verify_locked(inode); | |
2400 | if (!error) | |
ea0d3ab2 | 2401 | error = security_path_truncate(path); |
7715b521 AV |
2402 | if (!error) { |
2403 | error = do_truncate(path->dentry, 0, | |
2404 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2405 | filp); |
7715b521 AV |
2406 | } |
2407 | put_write_access(inode); | |
acd0c935 | 2408 | return error; |
1da177e4 LT |
2409 | } |
2410 | ||
d57999e1 DH |
2411 | static inline int open_to_namei_flags(int flag) |
2412 | { | |
8a5e929d AV |
2413 | if ((flag & O_ACCMODE) == 3) |
2414 | flag--; | |
d57999e1 DH |
2415 | return flag; |
2416 | } | |
2417 | ||
d18e9008 MS |
2418 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2419 | { | |
2420 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2421 | if (error) | |
2422 | return error; | |
2423 | ||
2424 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2425 | if (error) | |
2426 | return error; | |
2427 | ||
2428 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2429 | } | |
2430 | ||
1acf0af9 DH |
2431 | /* |
2432 | * Attempt to atomically look up, create and open a file from a negative | |
2433 | * dentry. | |
2434 | * | |
2435 | * Returns 0 if successful. The file will have been created and attached to | |
2436 | * @file by the filesystem calling finish_open(). | |
2437 | * | |
2438 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2439 | * caller will need to perform the open themselves. @path will have been | |
2440 | * updated to point to the new dentry. This may be negative. | |
2441 | * | |
2442 | * Returns an error code otherwise. | |
2443 | */ | |
2675a4eb AV |
2444 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2445 | struct path *path, struct file *file, | |
2446 | const struct open_flags *op, | |
64894cf8 | 2447 | bool got_write, bool need_lookup, |
2675a4eb | 2448 | int *opened) |
d18e9008 MS |
2449 | { |
2450 | struct inode *dir = nd->path.dentry->d_inode; | |
2451 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2452 | umode_t mode; | |
2453 | int error; | |
2454 | int acc_mode; | |
d18e9008 MS |
2455 | int create_error = 0; |
2456 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
2457 | ||
2458 | BUG_ON(dentry->d_inode); | |
2459 | ||
2460 | /* Don't create child dentry for a dead directory. */ | |
2461 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2462 | error = -ENOENT; |
d18e9008 MS |
2463 | goto out; |
2464 | } | |
2465 | ||
62b259d8 | 2466 | mode = op->mode; |
d18e9008 MS |
2467 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2468 | mode &= ~current_umask(); | |
2469 | ||
f8310c59 | 2470 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) { |
d18e9008 | 2471 | open_flag &= ~O_TRUNC; |
47237687 | 2472 | *opened |= FILE_CREATED; |
d18e9008 MS |
2473 | } |
2474 | ||
2475 | /* | |
2476 | * Checking write permission is tricky, bacuse we don't know if we are | |
2477 | * going to actually need it: O_CREAT opens should work as long as the | |
2478 | * file exists. But checking existence breaks atomicity. The trick is | |
2479 | * to check access and if not granted clear O_CREAT from the flags. | |
2480 | * | |
2481 | * Another problem is returing the "right" error value (e.g. for an | |
2482 | * O_EXCL open we want to return EEXIST not EROFS). | |
2483 | */ | |
64894cf8 AV |
2484 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2485 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2486 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2487 | /* |
2488 | * No O_CREATE -> atomicity not a requirement -> fall | |
2489 | * back to lookup + open | |
2490 | */ | |
2491 | goto no_open; | |
2492 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2493 | /* Fall back and fail with the right error */ | |
64894cf8 | 2494 | create_error = -EROFS; |
d18e9008 MS |
2495 | goto no_open; |
2496 | } else { | |
2497 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2498 | create_error = -EROFS; |
d18e9008 MS |
2499 | open_flag &= ~O_CREAT; |
2500 | } | |
2501 | } | |
2502 | ||
2503 | if (open_flag & O_CREAT) { | |
38227f78 | 2504 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2505 | if (error) { |
2506 | create_error = error; | |
2507 | if (open_flag & O_EXCL) | |
2508 | goto no_open; | |
2509 | open_flag &= ~O_CREAT; | |
2510 | } | |
2511 | } | |
2512 | ||
2513 | if (nd->flags & LOOKUP_DIRECTORY) | |
2514 | open_flag |= O_DIRECTORY; | |
2515 | ||
30d90494 AV |
2516 | file->f_path.dentry = DENTRY_NOT_SET; |
2517 | file->f_path.mnt = nd->path.mnt; | |
2518 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2519 | opened); |
d9585277 | 2520 | if (error < 0) { |
d9585277 AV |
2521 | if (create_error && error == -ENOENT) |
2522 | error = create_error; | |
d18e9008 MS |
2523 | goto out; |
2524 | } | |
2525 | ||
2526 | acc_mode = op->acc_mode; | |
47237687 | 2527 | if (*opened & FILE_CREATED) { |
d18e9008 MS |
2528 | fsnotify_create(dir, dentry); |
2529 | acc_mode = MAY_OPEN; | |
2530 | } | |
2531 | ||
d9585277 | 2532 | if (error) { /* returned 1, that is */ |
30d90494 | 2533 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2534 | error = -EIO; |
d18e9008 MS |
2535 | goto out; |
2536 | } | |
30d90494 | 2537 | if (file->f_path.dentry) { |
d18e9008 | 2538 | dput(dentry); |
30d90494 | 2539 | dentry = file->f_path.dentry; |
d18e9008 | 2540 | } |
62b2ce96 SW |
2541 | if (create_error && dentry->d_inode == NULL) { |
2542 | error = create_error; | |
2543 | goto out; | |
2544 | } | |
d18e9008 MS |
2545 | goto looked_up; |
2546 | } | |
2547 | ||
2548 | /* | |
2549 | * We didn't have the inode before the open, so check open permission | |
2550 | * here. | |
2551 | */ | |
2675a4eb AV |
2552 | error = may_open(&file->f_path, acc_mode, open_flag); |
2553 | if (error) | |
2554 | fput(file); | |
d18e9008 MS |
2555 | |
2556 | out: | |
2557 | dput(dentry); | |
2675a4eb | 2558 | return error; |
d18e9008 | 2559 | |
d18e9008 MS |
2560 | no_open: |
2561 | if (need_lookup) { | |
72bd866a | 2562 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2563 | if (IS_ERR(dentry)) |
2675a4eb | 2564 | return PTR_ERR(dentry); |
d18e9008 MS |
2565 | |
2566 | if (create_error) { | |
2567 | int open_flag = op->open_flag; | |
2568 | ||
2675a4eb | 2569 | error = create_error; |
d18e9008 MS |
2570 | if ((open_flag & O_EXCL)) { |
2571 | if (!dentry->d_inode) | |
2572 | goto out; | |
2573 | } else if (!dentry->d_inode) { | |
2574 | goto out; | |
2575 | } else if ((open_flag & O_TRUNC) && | |
2576 | S_ISREG(dentry->d_inode->i_mode)) { | |
2577 | goto out; | |
2578 | } | |
2579 | /* will fail later, go on to get the right error */ | |
2580 | } | |
2581 | } | |
2582 | looked_up: | |
2583 | path->dentry = dentry; | |
2584 | path->mnt = nd->path.mnt; | |
2675a4eb | 2585 | return 1; |
d18e9008 MS |
2586 | } |
2587 | ||
d58ffd35 | 2588 | /* |
1acf0af9 | 2589 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2590 | * |
2591 | * Must be called with i_mutex held on parent. | |
2592 | * | |
1acf0af9 DH |
2593 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2594 | * opened. In this case the file will be returned attached to @file. | |
2595 | * | |
2596 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2597 | * and creations will have been performed and the dentry returned in @path will | |
2598 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2599 | * specified then a negative dentry may be returned. | |
2600 | * | |
2601 | * An error code is returned otherwise. | |
2602 | * | |
2603 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2604 | * cleared otherwise prior to returning. | |
d58ffd35 | 2605 | */ |
2675a4eb AV |
2606 | static int lookup_open(struct nameidata *nd, struct path *path, |
2607 | struct file *file, | |
2608 | const struct open_flags *op, | |
64894cf8 | 2609 | bool got_write, int *opened) |
d58ffd35 MS |
2610 | { |
2611 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2612 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2613 | struct dentry *dentry; |
2614 | int error; | |
54ef4872 | 2615 | bool need_lookup; |
d58ffd35 | 2616 | |
47237687 | 2617 | *opened &= ~FILE_CREATED; |
201f956e | 2618 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2619 | if (IS_ERR(dentry)) |
2675a4eb | 2620 | return PTR_ERR(dentry); |
d58ffd35 | 2621 | |
d18e9008 MS |
2622 | /* Cached positive dentry: will open in f_op->open */ |
2623 | if (!need_lookup && dentry->d_inode) | |
2624 | goto out_no_open; | |
2625 | ||
2626 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2627 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2628 | need_lookup, opened); |
d18e9008 MS |
2629 | } |
2630 | ||
54ef4872 MS |
2631 | if (need_lookup) { |
2632 | BUG_ON(dentry->d_inode); | |
2633 | ||
72bd866a | 2634 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2635 | if (IS_ERR(dentry)) |
2675a4eb | 2636 | return PTR_ERR(dentry); |
54ef4872 MS |
2637 | } |
2638 | ||
d58ffd35 MS |
2639 | /* Negative dentry, just create the file */ |
2640 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2641 | umode_t mode = op->mode; | |
2642 | if (!IS_POSIXACL(dir->d_inode)) | |
2643 | mode &= ~current_umask(); | |
2644 | /* | |
2645 | * This write is needed to ensure that a | |
2646 | * rw->ro transition does not occur between | |
2647 | * the time when the file is created and when | |
2648 | * a permanent write count is taken through | |
015c3bbc | 2649 | * the 'struct file' in finish_open(). |
d58ffd35 | 2650 | */ |
64894cf8 AV |
2651 | if (!got_write) { |
2652 | error = -EROFS; | |
d58ffd35 | 2653 | goto out_dput; |
64894cf8 | 2654 | } |
47237687 | 2655 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2656 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2657 | if (error) | |
2658 | goto out_dput; | |
312b63fb AV |
2659 | error = vfs_create(dir->d_inode, dentry, mode, |
2660 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2661 | if (error) |
2662 | goto out_dput; | |
2663 | } | |
d18e9008 | 2664 | out_no_open: |
d58ffd35 MS |
2665 | path->dentry = dentry; |
2666 | path->mnt = nd->path.mnt; | |
2675a4eb | 2667 | return 1; |
d58ffd35 MS |
2668 | |
2669 | out_dput: | |
2670 | dput(dentry); | |
2675a4eb | 2671 | return error; |
d58ffd35 MS |
2672 | } |
2673 | ||
31e6b01f | 2674 | /* |
fe2d35ff | 2675 | * Handle the last step of open() |
31e6b01f | 2676 | */ |
2675a4eb AV |
2677 | static int do_last(struct nameidata *nd, struct path *path, |
2678 | struct file *file, const struct open_flags *op, | |
669abf4e | 2679 | int *opened, struct filename *name) |
fb1cc555 | 2680 | { |
a1e28038 | 2681 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2682 | int open_flag = op->open_flag; |
77d660a8 | 2683 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2684 | bool got_write = false; |
bcda7652 | 2685 | int acc_mode = op->acc_mode; |
a1eb3315 | 2686 | struct inode *inode; |
77d660a8 | 2687 | bool symlink_ok = false; |
16b1c1cd MS |
2688 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2689 | bool retried = false; | |
16c2cd71 | 2690 | int error; |
1f36f774 | 2691 | |
c3e380b0 AV |
2692 | nd->flags &= ~LOOKUP_PARENT; |
2693 | nd->flags |= op->intent; | |
2694 | ||
1f36f774 AV |
2695 | switch (nd->last_type) { |
2696 | case LAST_DOTDOT: | |
176306f5 | 2697 | case LAST_DOT: |
fe2d35ff AV |
2698 | error = handle_dots(nd, nd->last_type); |
2699 | if (error) | |
2675a4eb | 2700 | return error; |
1f36f774 | 2701 | /* fallthrough */ |
1f36f774 | 2702 | case LAST_ROOT: |
9f1fafee | 2703 | error = complete_walk(nd); |
16c2cd71 | 2704 | if (error) |
2675a4eb | 2705 | return error; |
adb5c247 | 2706 | audit_inode(name, nd->path.dentry, 0); |
ca344a89 | 2707 | if (open_flag & O_CREAT) { |
fe2d35ff | 2708 | error = -EISDIR; |
2675a4eb | 2709 | goto out; |
fe2d35ff | 2710 | } |
e83db167 | 2711 | goto finish_open; |
1f36f774 | 2712 | case LAST_BIND: |
9f1fafee | 2713 | error = complete_walk(nd); |
16c2cd71 | 2714 | if (error) |
2675a4eb | 2715 | return error; |
adb5c247 | 2716 | audit_inode(name, dir, 0); |
e83db167 | 2717 | goto finish_open; |
1f36f774 | 2718 | } |
67ee3ad2 | 2719 | |
ca344a89 | 2720 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2721 | if (nd->last.name[nd->last.len]) |
2722 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2723 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2724 | symlink_ok = true; |
fe2d35ff | 2725 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2726 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2727 | if (likely(!error)) |
2728 | goto finish_lookup; | |
2729 | ||
2730 | if (error < 0) | |
2675a4eb | 2731 | goto out; |
71574865 MS |
2732 | |
2733 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2734 | } else { |
2735 | /* create side of things */ | |
2736 | /* | |
2737 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2738 | * has been cleared when we got to the last component we are | |
2739 | * about to look up | |
2740 | */ | |
2741 | error = complete_walk(nd); | |
2742 | if (error) | |
2675a4eb | 2743 | return error; |
fe2d35ff | 2744 | |
33e2208a | 2745 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2746 | error = -EISDIR; |
2747 | /* trailing slashes? */ | |
2748 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2749 | goto out; |
b6183df7 | 2750 | } |
a2c36b45 | 2751 | |
16b1c1cd | 2752 | retry_lookup: |
64894cf8 AV |
2753 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2754 | error = mnt_want_write(nd->path.mnt); | |
2755 | if (!error) | |
2756 | got_write = true; | |
2757 | /* | |
2758 | * do _not_ fail yet - we might not need that or fail with | |
2759 | * a different error; let lookup_open() decide; we'll be | |
2760 | * dropping this one anyway. | |
2761 | */ | |
2762 | } | |
a1e28038 | 2763 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2764 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2765 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2766 | |
2675a4eb AV |
2767 | if (error <= 0) { |
2768 | if (error) | |
d18e9008 MS |
2769 | goto out; |
2770 | ||
47237687 | 2771 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2772 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2773 | will_truncate = false; |
d18e9008 | 2774 | |
adb5c247 | 2775 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2776 | goto opened; |
2777 | } | |
fb1cc555 | 2778 | |
47237687 | 2779 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2780 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2781 | open_flag &= ~O_TRUNC; |
77d660a8 | 2782 | will_truncate = false; |
bcda7652 | 2783 | acc_mode = MAY_OPEN; |
d58ffd35 | 2784 | path_to_nameidata(path, nd); |
e83db167 | 2785 | goto finish_open_created; |
fb1cc555 AV |
2786 | } |
2787 | ||
2788 | /* | |
3134f37e | 2789 | * create/update audit record if it already exists. |
fb1cc555 | 2790 | */ |
3134f37e | 2791 | if (path->dentry->d_inode) |
adb5c247 | 2792 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2793 | |
d18e9008 MS |
2794 | /* |
2795 | * If atomic_open() acquired write access it is dropped now due to | |
2796 | * possible mount and symlink following (this might be optimized away if | |
2797 | * necessary...) | |
2798 | */ | |
64894cf8 | 2799 | if (got_write) { |
d18e9008 | 2800 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2801 | got_write = false; |
d18e9008 MS |
2802 | } |
2803 | ||
fb1cc555 | 2804 | error = -EEXIST; |
f8310c59 | 2805 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2806 | goto exit_dput; |
2807 | ||
9875cf80 DH |
2808 | error = follow_managed(path, nd->flags); |
2809 | if (error < 0) | |
2810 | goto exit_dput; | |
fb1cc555 | 2811 | |
a3fbbde7 AV |
2812 | if (error) |
2813 | nd->flags |= LOOKUP_JUMPED; | |
2814 | ||
decf3400 MS |
2815 | BUG_ON(nd->flags & LOOKUP_RCU); |
2816 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2817 | finish_lookup: |
2818 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2819 | error = -ENOENT; |
54c33e7f MS |
2820 | if (!inode) { |
2821 | path_to_nameidata(path, nd); | |
2675a4eb | 2822 | goto out; |
54c33e7f | 2823 | } |
9e67f361 | 2824 | |
d45ea867 MS |
2825 | if (should_follow_link(inode, !symlink_ok)) { |
2826 | if (nd->flags & LOOKUP_RCU) { | |
2827 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2828 | error = -ECHILD; | |
2675a4eb | 2829 | goto out; |
d45ea867 MS |
2830 | } |
2831 | } | |
2832 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 2833 | return 1; |
d45ea867 | 2834 | } |
fb1cc555 | 2835 | |
16b1c1cd MS |
2836 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
2837 | path_to_nameidata(path, nd); | |
2838 | } else { | |
2839 | save_parent.dentry = nd->path.dentry; | |
2840 | save_parent.mnt = mntget(path->mnt); | |
2841 | nd->path.dentry = path->dentry; | |
2842 | ||
2843 | } | |
decf3400 | 2844 | nd->inode = inode; |
a3fbbde7 AV |
2845 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2846 | error = complete_walk(nd); | |
16b1c1cd MS |
2847 | if (error) { |
2848 | path_put(&save_parent); | |
2675a4eb | 2849 | return error; |
16b1c1cd | 2850 | } |
fb1cc555 | 2851 | error = -EISDIR; |
050ac841 | 2852 | if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) |
2675a4eb | 2853 | goto out; |
af2f5542 | 2854 | error = -ENOTDIR; |
05252901 | 2855 | if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode)) |
2675a4eb | 2856 | goto out; |
adb5c247 | 2857 | audit_inode(name, nd->path.dentry, 0); |
e83db167 | 2858 | finish_open: |
6c0d46c4 | 2859 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 2860 | will_truncate = false; |
6c0d46c4 | 2861 | |
0f9d1a10 AV |
2862 | if (will_truncate) { |
2863 | error = mnt_want_write(nd->path.mnt); | |
2864 | if (error) | |
2675a4eb | 2865 | goto out; |
64894cf8 | 2866 | got_write = true; |
0f9d1a10 | 2867 | } |
e83db167 | 2868 | finish_open_created: |
bcda7652 | 2869 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 2870 | if (error) |
2675a4eb | 2871 | goto out; |
30d90494 AV |
2872 | file->f_path.mnt = nd->path.mnt; |
2873 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
2874 | if (error) { | |
30d90494 | 2875 | if (error == -EOPENSTALE) |
f60dc3db | 2876 | goto stale_open; |
015c3bbc | 2877 | goto out; |
f60dc3db | 2878 | } |
a8277b9b | 2879 | opened: |
2675a4eb | 2880 | error = open_check_o_direct(file); |
015c3bbc MS |
2881 | if (error) |
2882 | goto exit_fput; | |
2675a4eb | 2883 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
2884 | if (error) |
2885 | goto exit_fput; | |
2886 | ||
2887 | if (will_truncate) { | |
2675a4eb | 2888 | error = handle_truncate(file); |
aa4caadb MS |
2889 | if (error) |
2890 | goto exit_fput; | |
0f9d1a10 | 2891 | } |
ca344a89 | 2892 | out: |
64894cf8 | 2893 | if (got_write) |
0f9d1a10 | 2894 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 2895 | path_put(&save_parent); |
e276ae67 | 2896 | terminate_walk(nd); |
2675a4eb | 2897 | return error; |
fb1cc555 | 2898 | |
fb1cc555 AV |
2899 | exit_dput: |
2900 | path_put_conditional(path, nd); | |
ca344a89 | 2901 | goto out; |
015c3bbc | 2902 | exit_fput: |
2675a4eb AV |
2903 | fput(file); |
2904 | goto out; | |
015c3bbc | 2905 | |
f60dc3db MS |
2906 | stale_open: |
2907 | /* If no saved parent or already retried then can't retry */ | |
2908 | if (!save_parent.dentry || retried) | |
2909 | goto out; | |
2910 | ||
2911 | BUG_ON(save_parent.dentry != dir); | |
2912 | path_put(&nd->path); | |
2913 | nd->path = save_parent; | |
2914 | nd->inode = dir->d_inode; | |
2915 | save_parent.mnt = NULL; | |
2916 | save_parent.dentry = NULL; | |
64894cf8 | 2917 | if (got_write) { |
f60dc3db | 2918 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2919 | got_write = false; |
f60dc3db MS |
2920 | } |
2921 | retried = true; | |
2922 | goto retry_lookup; | |
fb1cc555 AV |
2923 | } |
2924 | ||
669abf4e | 2925 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 2926 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 2927 | { |
fe2d35ff | 2928 | struct file *base = NULL; |
30d90494 | 2929 | struct file *file; |
9850c056 | 2930 | struct path path; |
47237687 | 2931 | int opened = 0; |
13aab428 | 2932 | int error; |
31e6b01f | 2933 | |
30d90494 | 2934 | file = get_empty_filp(); |
1afc99be AV |
2935 | if (IS_ERR(file)) |
2936 | return file; | |
31e6b01f | 2937 | |
30d90494 | 2938 | file->f_flags = op->open_flag; |
31e6b01f | 2939 | |
669abf4e | 2940 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 2941 | if (unlikely(error)) |
2675a4eb | 2942 | goto out; |
31e6b01f | 2943 | |
fe2d35ff | 2944 | current->total_link_count = 0; |
669abf4e | 2945 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 2946 | if (unlikely(error)) |
2675a4eb | 2947 | goto out; |
1da177e4 | 2948 | |
2675a4eb AV |
2949 | error = do_last(nd, &path, file, op, &opened, pathname); |
2950 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 2951 | struct path link = path; |
def4af30 | 2952 | void *cookie; |
574197e0 | 2953 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
2954 | path_put_conditional(&path, nd); |
2955 | path_put(&nd->path); | |
2675a4eb | 2956 | error = -ELOOP; |
40b39136 AV |
2957 | break; |
2958 | } | |
800179c9 KC |
2959 | error = may_follow_link(&link, nd); |
2960 | if (unlikely(error)) | |
2961 | break; | |
73d049a4 AV |
2962 | nd->flags |= LOOKUP_PARENT; |
2963 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 2964 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 2965 | if (unlikely(error)) |
2675a4eb AV |
2966 | break; |
2967 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 2968 | put_link(nd, &link, cookie); |
806b681c | 2969 | } |
10fa8e62 | 2970 | out: |
73d049a4 AV |
2971 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2972 | path_put(&nd->root); | |
fe2d35ff AV |
2973 | if (base) |
2974 | fput(base); | |
2675a4eb AV |
2975 | if (!(opened & FILE_OPENED)) { |
2976 | BUG_ON(!error); | |
30d90494 | 2977 | put_filp(file); |
16b1c1cd | 2978 | } |
2675a4eb AV |
2979 | if (unlikely(error)) { |
2980 | if (error == -EOPENSTALE) { | |
2981 | if (flags & LOOKUP_RCU) | |
2982 | error = -ECHILD; | |
2983 | else | |
2984 | error = -ESTALE; | |
2985 | } | |
2986 | file = ERR_PTR(error); | |
2987 | } | |
2988 | return file; | |
1da177e4 LT |
2989 | } |
2990 | ||
669abf4e | 2991 | struct file *do_filp_open(int dfd, struct filename *pathname, |
13aab428 AV |
2992 | const struct open_flags *op, int flags) |
2993 | { | |
73d049a4 | 2994 | struct nameidata nd; |
13aab428 AV |
2995 | struct file *filp; |
2996 | ||
73d049a4 | 2997 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 2998 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 2999 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3000 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3001 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3002 | return filp; |
3003 | } | |
3004 | ||
73d049a4 AV |
3005 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
3006 | const char *name, const struct open_flags *op, int flags) | |
3007 | { | |
3008 | struct nameidata nd; | |
3009 | struct file *file; | |
669abf4e | 3010 | struct filename filename = { .name = name }; |
73d049a4 AV |
3011 | |
3012 | nd.root.mnt = mnt; | |
3013 | nd.root.dentry = dentry; | |
3014 | ||
3015 | flags |= LOOKUP_ROOT; | |
3016 | ||
bcda7652 | 3017 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3018 | return ERR_PTR(-ELOOP); |
3019 | ||
669abf4e | 3020 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3021 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3022 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3023 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3024 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3025 | return file; |
3026 | } | |
3027 | ||
1ac12b4b JL |
3028 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3029 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3030 | { |
c663e5d8 | 3031 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3032 | struct nameidata nd; |
c30dabfe | 3033 | int err2; |
1ac12b4b JL |
3034 | int error; |
3035 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3036 | ||
3037 | /* | |
3038 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3039 | * other flags passed in are ignored! | |
3040 | */ | |
3041 | lookup_flags &= LOOKUP_REVAL; | |
3042 | ||
3043 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3044 | if (error) |
3045 | return ERR_PTR(error); | |
1da177e4 | 3046 | |
c663e5d8 CH |
3047 | /* |
3048 | * Yucky last component or no last component at all? | |
3049 | * (foo/., foo/.., /////) | |
3050 | */ | |
ed75e95d AV |
3051 | if (nd.last_type != LAST_NORM) |
3052 | goto out; | |
3053 | nd.flags &= ~LOOKUP_PARENT; | |
3054 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3055 | |
c30dabfe JK |
3056 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3057 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3058 | /* |
3059 | * Do the final lookup. | |
3060 | */ | |
ed75e95d AV |
3061 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3062 | dentry = lookup_hash(&nd); | |
1da177e4 | 3063 | if (IS_ERR(dentry)) |
a8104a9f | 3064 | goto unlock; |
c663e5d8 | 3065 | |
a8104a9f | 3066 | error = -EEXIST; |
e9baf6e5 | 3067 | if (dentry->d_inode) |
a8104a9f | 3068 | goto fail; |
c663e5d8 CH |
3069 | /* |
3070 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3071 | * From the vfs_mknod() POV we just have a negative dentry - | |
3072 | * all is fine. Let's be bastards - you had / on the end, you've | |
3073 | * been asking for (non-existent) directory. -ENOENT for you. | |
3074 | */ | |
ed75e95d | 3075 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3076 | error = -ENOENT; |
ed75e95d | 3077 | goto fail; |
e9baf6e5 | 3078 | } |
c30dabfe JK |
3079 | if (unlikely(err2)) { |
3080 | error = err2; | |
a8104a9f | 3081 | goto fail; |
c30dabfe | 3082 | } |
ed75e95d | 3083 | *path = nd.path; |
1da177e4 | 3084 | return dentry; |
1da177e4 | 3085 | fail: |
a8104a9f AV |
3086 | dput(dentry); |
3087 | dentry = ERR_PTR(error); | |
3088 | unlock: | |
ed75e95d | 3089 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3090 | if (!err2) |
3091 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3092 | out: |
3093 | path_put(&nd.path); | |
1da177e4 LT |
3094 | return dentry; |
3095 | } | |
dae6ad8f AV |
3096 | EXPORT_SYMBOL(kern_path_create); |
3097 | ||
921a1650 AV |
3098 | void done_path_create(struct path *path, struct dentry *dentry) |
3099 | { | |
3100 | dput(dentry); | |
3101 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3102 | mnt_drop_write(path->mnt); |
921a1650 AV |
3103 | path_put(path); |
3104 | } | |
3105 | EXPORT_SYMBOL(done_path_create); | |
3106 | ||
1ac12b4b JL |
3107 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3108 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3109 | { |
91a27b2a | 3110 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3111 | struct dentry *res; |
3112 | if (IS_ERR(tmp)) | |
3113 | return ERR_CAST(tmp); | |
1ac12b4b | 3114 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3115 | putname(tmp); |
3116 | return res; | |
3117 | } | |
3118 | EXPORT_SYMBOL(user_path_create); | |
3119 | ||
1a67aafb | 3120 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3121 | { |
a95164d9 | 3122 | int error = may_create(dir, dentry); |
1da177e4 LT |
3123 | |
3124 | if (error) | |
3125 | return error; | |
3126 | ||
975d6b39 | 3127 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3128 | return -EPERM; |
3129 | ||
acfa4380 | 3130 | if (!dir->i_op->mknod) |
1da177e4 LT |
3131 | return -EPERM; |
3132 | ||
08ce5f16 SH |
3133 | error = devcgroup_inode_mknod(mode, dev); |
3134 | if (error) | |
3135 | return error; | |
3136 | ||
1da177e4 LT |
3137 | error = security_inode_mknod(dir, dentry, mode, dev); |
3138 | if (error) | |
3139 | return error; | |
3140 | ||
1da177e4 | 3141 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3142 | if (!error) |
f38aa942 | 3143 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3144 | return error; |
3145 | } | |
3146 | ||
f69aac00 | 3147 | static int may_mknod(umode_t mode) |
463c3197 DH |
3148 | { |
3149 | switch (mode & S_IFMT) { | |
3150 | case S_IFREG: | |
3151 | case S_IFCHR: | |
3152 | case S_IFBLK: | |
3153 | case S_IFIFO: | |
3154 | case S_IFSOCK: | |
3155 | case 0: /* zero mode translates to S_IFREG */ | |
3156 | return 0; | |
3157 | case S_IFDIR: | |
3158 | return -EPERM; | |
3159 | default: | |
3160 | return -EINVAL; | |
3161 | } | |
3162 | } | |
3163 | ||
8208a22b | 3164 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3165 | unsigned, dev) |
1da177e4 | 3166 | { |
2ad94ae6 | 3167 | struct dentry *dentry; |
dae6ad8f AV |
3168 | struct path path; |
3169 | int error; | |
972567f1 | 3170 | unsigned int lookup_flags = 0; |
1da177e4 | 3171 | |
8e4bfca1 AV |
3172 | error = may_mknod(mode); |
3173 | if (error) | |
3174 | return error; | |
972567f1 JL |
3175 | retry: |
3176 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3177 | if (IS_ERR(dentry)) |
3178 | return PTR_ERR(dentry); | |
2ad94ae6 | 3179 | |
dae6ad8f | 3180 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3181 | mode &= ~current_umask(); |
dae6ad8f | 3182 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3183 | if (error) |
a8104a9f | 3184 | goto out; |
463c3197 | 3185 | switch (mode & S_IFMT) { |
1da177e4 | 3186 | case 0: case S_IFREG: |
312b63fb | 3187 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3188 | break; |
3189 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3190 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3191 | new_decode_dev(dev)); |
3192 | break; | |
3193 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3194 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3195 | break; |
1da177e4 | 3196 | } |
a8104a9f | 3197 | out: |
921a1650 | 3198 | done_path_create(&path, dentry); |
972567f1 JL |
3199 | if (retry_estale(error, lookup_flags)) { |
3200 | lookup_flags |= LOOKUP_REVAL; | |
3201 | goto retry; | |
3202 | } | |
1da177e4 LT |
3203 | return error; |
3204 | } | |
3205 | ||
8208a22b | 3206 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3207 | { |
3208 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3209 | } | |
3210 | ||
18bb1db3 | 3211 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3212 | { |
a95164d9 | 3213 | int error = may_create(dir, dentry); |
8de52778 | 3214 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3215 | |
3216 | if (error) | |
3217 | return error; | |
3218 | ||
acfa4380 | 3219 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3220 | return -EPERM; |
3221 | ||
3222 | mode &= (S_IRWXUGO|S_ISVTX); | |
3223 | error = security_inode_mkdir(dir, dentry, mode); | |
3224 | if (error) | |
3225 | return error; | |
3226 | ||
8de52778 AV |
3227 | if (max_links && dir->i_nlink >= max_links) |
3228 | return -EMLINK; | |
3229 | ||
1da177e4 | 3230 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3231 | if (!error) |
f38aa942 | 3232 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3233 | return error; |
3234 | } | |
3235 | ||
a218d0fd | 3236 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3237 | { |
6902d925 | 3238 | struct dentry *dentry; |
dae6ad8f AV |
3239 | struct path path; |
3240 | int error; | |
b76d8b82 | 3241 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3242 | |
b76d8b82 JL |
3243 | retry: |
3244 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3245 | if (IS_ERR(dentry)) |
dae6ad8f | 3246 | return PTR_ERR(dentry); |
1da177e4 | 3247 | |
dae6ad8f | 3248 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3249 | mode &= ~current_umask(); |
dae6ad8f | 3250 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3251 | if (!error) |
3252 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3253 | done_path_create(&path, dentry); |
b76d8b82 JL |
3254 | if (retry_estale(error, lookup_flags)) { |
3255 | lookup_flags |= LOOKUP_REVAL; | |
3256 | goto retry; | |
3257 | } | |
1da177e4 LT |
3258 | return error; |
3259 | } | |
3260 | ||
a218d0fd | 3261 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3262 | { |
3263 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3264 | } | |
3265 | ||
1da177e4 | 3266 | /* |
a71905f0 | 3267 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3268 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3269 | * dentry, and if that is true (possibly after pruning the dcache), |
3270 | * then we drop the dentry now. | |
1da177e4 LT |
3271 | * |
3272 | * A low-level filesystem can, if it choses, legally | |
3273 | * do a | |
3274 | * | |
3275 | * if (!d_unhashed(dentry)) | |
3276 | * return -EBUSY; | |
3277 | * | |
3278 | * if it cannot handle the case of removing a directory | |
3279 | * that is still in use by something else.. | |
3280 | */ | |
3281 | void dentry_unhash(struct dentry *dentry) | |
3282 | { | |
dc168427 | 3283 | shrink_dcache_parent(dentry); |
1da177e4 | 3284 | spin_lock(&dentry->d_lock); |
64252c75 | 3285 | if (dentry->d_count == 1) |
1da177e4 LT |
3286 | __d_drop(dentry); |
3287 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3288 | } |
3289 | ||
3290 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3291 | { | |
3292 | int error = may_delete(dir, dentry, 1); | |
3293 | ||
3294 | if (error) | |
3295 | return error; | |
3296 | ||
acfa4380 | 3297 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3298 | return -EPERM; |
3299 | ||
1d2ef590 | 3300 | dget(dentry); |
1b1dcc1b | 3301 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3302 | |
3303 | error = -EBUSY; | |
1da177e4 | 3304 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3305 | goto out; |
3306 | ||
3307 | error = security_inode_rmdir(dir, dentry); | |
3308 | if (error) | |
3309 | goto out; | |
3310 | ||
3cebde24 | 3311 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3312 | error = dir->i_op->rmdir(dir, dentry); |
3313 | if (error) | |
3314 | goto out; | |
3315 | ||
3316 | dentry->d_inode->i_flags |= S_DEAD; | |
3317 | dont_mount(dentry); | |
3318 | ||
3319 | out: | |
1b1dcc1b | 3320 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3321 | dput(dentry); |
912dbc15 | 3322 | if (!error) |
1da177e4 | 3323 | d_delete(dentry); |
1da177e4 LT |
3324 | return error; |
3325 | } | |
3326 | ||
5590ff0d | 3327 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3328 | { |
3329 | int error = 0; | |
91a27b2a | 3330 | struct filename *name; |
1da177e4 LT |
3331 | struct dentry *dentry; |
3332 | struct nameidata nd; | |
c6ee9206 JL |
3333 | unsigned int lookup_flags = 0; |
3334 | retry: | |
3335 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3336 | if (IS_ERR(name)) |
3337 | return PTR_ERR(name); | |
1da177e4 LT |
3338 | |
3339 | switch(nd.last_type) { | |
0612d9fb OH |
3340 | case LAST_DOTDOT: |
3341 | error = -ENOTEMPTY; | |
3342 | goto exit1; | |
3343 | case LAST_DOT: | |
3344 | error = -EINVAL; | |
3345 | goto exit1; | |
3346 | case LAST_ROOT: | |
3347 | error = -EBUSY; | |
3348 | goto exit1; | |
1da177e4 | 3349 | } |
0612d9fb OH |
3350 | |
3351 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3352 | error = mnt_want_write(nd.path.mnt); |
3353 | if (error) | |
3354 | goto exit1; | |
0612d9fb | 3355 | |
4ac91378 | 3356 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3357 | dentry = lookup_hash(&nd); |
1da177e4 | 3358 | error = PTR_ERR(dentry); |
6902d925 DH |
3359 | if (IS_ERR(dentry)) |
3360 | goto exit2; | |
e6bc45d6 TT |
3361 | if (!dentry->d_inode) { |
3362 | error = -ENOENT; | |
3363 | goto exit3; | |
3364 | } | |
be6d3e56 KT |
3365 | error = security_path_rmdir(&nd.path, dentry); |
3366 | if (error) | |
c30dabfe | 3367 | goto exit3; |
4ac91378 | 3368 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3369 | exit3: |
6902d925 DH |
3370 | dput(dentry); |
3371 | exit2: | |
4ac91378 | 3372 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3373 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3374 | exit1: |
1d957f9b | 3375 | path_put(&nd.path); |
1da177e4 | 3376 | putname(name); |
c6ee9206 JL |
3377 | if (retry_estale(error, lookup_flags)) { |
3378 | lookup_flags |= LOOKUP_REVAL; | |
3379 | goto retry; | |
3380 | } | |
1da177e4 LT |
3381 | return error; |
3382 | } | |
3383 | ||
3cdad428 | 3384 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3385 | { |
3386 | return do_rmdir(AT_FDCWD, pathname); | |
3387 | } | |
3388 | ||
1da177e4 LT |
3389 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
3390 | { | |
3391 | int error = may_delete(dir, dentry, 0); | |
3392 | ||
3393 | if (error) | |
3394 | return error; | |
3395 | ||
acfa4380 | 3396 | if (!dir->i_op->unlink) |
1da177e4 LT |
3397 | return -EPERM; |
3398 | ||
1b1dcc1b | 3399 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3400 | if (d_mountpoint(dentry)) |
3401 | error = -EBUSY; | |
3402 | else { | |
3403 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3404 | if (!error) { |
1da177e4 | 3405 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3406 | if (!error) |
d83c49f3 | 3407 | dont_mount(dentry); |
bec1052e | 3408 | } |
1da177e4 | 3409 | } |
1b1dcc1b | 3410 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3411 | |
3412 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3413 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 3414 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 3415 | d_delete(dentry); |
1da177e4 | 3416 | } |
0eeca283 | 3417 | |
1da177e4 LT |
3418 | return error; |
3419 | } | |
3420 | ||
3421 | /* | |
3422 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3423 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3424 | * writeout happening, and we don't want to prevent access to the directory |
3425 | * while waiting on the I/O. | |
3426 | */ | |
5590ff0d | 3427 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3428 | { |
2ad94ae6 | 3429 | int error; |
91a27b2a | 3430 | struct filename *name; |
1da177e4 LT |
3431 | struct dentry *dentry; |
3432 | struct nameidata nd; | |
3433 | struct inode *inode = NULL; | |
5d18f813 JL |
3434 | unsigned int lookup_flags = 0; |
3435 | retry: | |
3436 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3437 | if (IS_ERR(name)) |
3438 | return PTR_ERR(name); | |
2ad94ae6 | 3439 | |
1da177e4 LT |
3440 | error = -EISDIR; |
3441 | if (nd.last_type != LAST_NORM) | |
3442 | goto exit1; | |
0612d9fb OH |
3443 | |
3444 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3445 | error = mnt_want_write(nd.path.mnt); |
3446 | if (error) | |
3447 | goto exit1; | |
0612d9fb | 3448 | |
4ac91378 | 3449 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3450 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3451 | error = PTR_ERR(dentry); |
3452 | if (!IS_ERR(dentry)) { | |
3453 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3454 | if (nd.last.name[nd.last.len]) |
3455 | goto slashes; | |
1da177e4 | 3456 | inode = dentry->d_inode; |
50338b88 | 3457 | if (!inode) |
e6bc45d6 TT |
3458 | goto slashes; |
3459 | ihold(inode); | |
be6d3e56 KT |
3460 | error = security_path_unlink(&nd.path, dentry); |
3461 | if (error) | |
c30dabfe | 3462 | goto exit2; |
4ac91378 | 3463 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
c30dabfe | 3464 | exit2: |
1da177e4 LT |
3465 | dput(dentry); |
3466 | } | |
4ac91378 | 3467 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3468 | if (inode) |
3469 | iput(inode); /* truncate the inode here */ | |
c30dabfe | 3470 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3471 | exit1: |
1d957f9b | 3472 | path_put(&nd.path); |
1da177e4 | 3473 | putname(name); |
5d18f813 JL |
3474 | if (retry_estale(error, lookup_flags)) { |
3475 | lookup_flags |= LOOKUP_REVAL; | |
3476 | inode = NULL; | |
3477 | goto retry; | |
3478 | } | |
1da177e4 LT |
3479 | return error; |
3480 | ||
3481 | slashes: | |
3482 | error = !dentry->d_inode ? -ENOENT : | |
3483 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
3484 | goto exit2; | |
3485 | } | |
3486 | ||
2e4d0924 | 3487 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3488 | { |
3489 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3490 | return -EINVAL; | |
3491 | ||
3492 | if (flag & AT_REMOVEDIR) | |
3493 | return do_rmdir(dfd, pathname); | |
3494 | ||
3495 | return do_unlinkat(dfd, pathname); | |
3496 | } | |
3497 | ||
3480b257 | 3498 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3499 | { |
3500 | return do_unlinkat(AT_FDCWD, pathname); | |
3501 | } | |
3502 | ||
db2e747b | 3503 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3504 | { |
a95164d9 | 3505 | int error = may_create(dir, dentry); |
1da177e4 LT |
3506 | |
3507 | if (error) | |
3508 | return error; | |
3509 | ||
acfa4380 | 3510 | if (!dir->i_op->symlink) |
1da177e4 LT |
3511 | return -EPERM; |
3512 | ||
3513 | error = security_inode_symlink(dir, dentry, oldname); | |
3514 | if (error) | |
3515 | return error; | |
3516 | ||
1da177e4 | 3517 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3518 | if (!error) |
f38aa942 | 3519 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3520 | return error; |
3521 | } | |
3522 | ||
2e4d0924 HC |
3523 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3524 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3525 | { |
2ad94ae6 | 3526 | int error; |
91a27b2a | 3527 | struct filename *from; |
6902d925 | 3528 | struct dentry *dentry; |
dae6ad8f | 3529 | struct path path; |
f46d3567 | 3530 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3531 | |
3532 | from = getname(oldname); | |
2ad94ae6 | 3533 | if (IS_ERR(from)) |
1da177e4 | 3534 | return PTR_ERR(from); |
f46d3567 JL |
3535 | retry: |
3536 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3537 | error = PTR_ERR(dentry); |
3538 | if (IS_ERR(dentry)) | |
dae6ad8f | 3539 | goto out_putname; |
6902d925 | 3540 | |
91a27b2a | 3541 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3542 | if (!error) |
91a27b2a | 3543 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3544 | done_path_create(&path, dentry); |
f46d3567 JL |
3545 | if (retry_estale(error, lookup_flags)) { |
3546 | lookup_flags |= LOOKUP_REVAL; | |
3547 | goto retry; | |
3548 | } | |
6902d925 | 3549 | out_putname: |
1da177e4 LT |
3550 | putname(from); |
3551 | return error; | |
3552 | } | |
3553 | ||
3480b257 | 3554 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3555 | { |
3556 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3557 | } | |
3558 | ||
1da177e4 LT |
3559 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3560 | { | |
3561 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3562 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3563 | int error; |
3564 | ||
3565 | if (!inode) | |
3566 | return -ENOENT; | |
3567 | ||
a95164d9 | 3568 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3569 | if (error) |
3570 | return error; | |
3571 | ||
3572 | if (dir->i_sb != inode->i_sb) | |
3573 | return -EXDEV; | |
3574 | ||
3575 | /* | |
3576 | * A link to an append-only or immutable file cannot be created. | |
3577 | */ | |
3578 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3579 | return -EPERM; | |
acfa4380 | 3580 | if (!dir->i_op->link) |
1da177e4 | 3581 | return -EPERM; |
7e79eedb | 3582 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3583 | return -EPERM; |
3584 | ||
3585 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3586 | if (error) | |
3587 | return error; | |
3588 | ||
7e79eedb | 3589 | mutex_lock(&inode->i_mutex); |
aae8a97d AK |
3590 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
3591 | if (inode->i_nlink == 0) | |
3592 | error = -ENOENT; | |
8de52778 AV |
3593 | else if (max_links && inode->i_nlink >= max_links) |
3594 | error = -EMLINK; | |
aae8a97d AK |
3595 | else |
3596 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
7e79eedb | 3597 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3598 | if (!error) |
7e79eedb | 3599 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3600 | return error; |
3601 | } | |
3602 | ||
3603 | /* | |
3604 | * Hardlinks are often used in delicate situations. We avoid | |
3605 | * security-related surprises by not following symlinks on the | |
3606 | * newname. --KAB | |
3607 | * | |
3608 | * We don't follow them on the oldname either to be compatible | |
3609 | * with linux 2.0, and to avoid hard-linking to directories | |
3610 | * and other special files. --ADM | |
3611 | */ | |
2e4d0924 HC |
3612 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3613 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3614 | { |
3615 | struct dentry *new_dentry; | |
dae6ad8f | 3616 | struct path old_path, new_path; |
11a7b371 | 3617 | int how = 0; |
1da177e4 | 3618 | int error; |
1da177e4 | 3619 | |
11a7b371 | 3620 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3621 | return -EINVAL; |
11a7b371 AK |
3622 | /* |
3623 | * To use null names we require CAP_DAC_READ_SEARCH | |
3624 | * This ensures that not everyone will be able to create | |
3625 | * handlink using the passed filedescriptor. | |
3626 | */ | |
3627 | if (flags & AT_EMPTY_PATH) { | |
3628 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3629 | return -ENOENT; | |
3630 | how = LOOKUP_EMPTY; | |
3631 | } | |
3632 | ||
3633 | if (flags & AT_SYMLINK_FOLLOW) | |
3634 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3635 | retry: |
11a7b371 | 3636 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3637 | if (error) |
2ad94ae6 AV |
3638 | return error; |
3639 | ||
442e31ca JL |
3640 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3641 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3642 | error = PTR_ERR(new_dentry); |
6902d925 | 3643 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3644 | goto out; |
3645 | ||
3646 | error = -EXDEV; | |
3647 | if (old_path.mnt != new_path.mnt) | |
3648 | goto out_dput; | |
800179c9 KC |
3649 | error = may_linkat(&old_path); |
3650 | if (unlikely(error)) | |
3651 | goto out_dput; | |
dae6ad8f | 3652 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3653 | if (error) |
a8104a9f | 3654 | goto out_dput; |
dae6ad8f | 3655 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
75c3f29d | 3656 | out_dput: |
921a1650 | 3657 | done_path_create(&new_path, new_dentry); |
442e31ca | 3658 | if (retry_estale(error, how)) { |
f1352fb2 | 3659 | path_put(&old_path); |
442e31ca JL |
3660 | how |= LOOKUP_REVAL; |
3661 | goto retry; | |
3662 | } | |
1da177e4 | 3663 | out: |
2d8f3038 | 3664 | path_put(&old_path); |
1da177e4 LT |
3665 | |
3666 | return error; | |
3667 | } | |
3668 | ||
3480b257 | 3669 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3670 | { |
c04030e1 | 3671 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3672 | } |
3673 | ||
1da177e4 LT |
3674 | /* |
3675 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3676 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3677 | * Problems: | |
3678 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3679 | * b) race potential - two innocent renames can create a loop together. | |
3680 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3681 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3682 | * story. |
3683 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3684 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3685 | * whether the target exists). Solution: try to be smart with locking |
3686 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3687 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3688 | * move will be locked. Thus we can rank directories by the tree |
3689 | * (ancestors first) and rank all non-directories after them. | |
3690 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3691 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3692 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3693 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3694 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3695 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3696 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3697 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3698 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3699 | * locking]. |
3700 | */ | |
75c96f85 AB |
3701 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3702 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3703 | { |
3704 | int error = 0; | |
9055cba7 | 3705 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3706 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3707 | |
3708 | /* | |
3709 | * If we are going to change the parent - check write permissions, | |
3710 | * we'll need to flip '..'. | |
3711 | */ | |
3712 | if (new_dir != old_dir) { | |
f419a2e3 | 3713 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3714 | if (error) |
3715 | return error; | |
3716 | } | |
3717 | ||
3718 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3719 | if (error) | |
3720 | return error; | |
3721 | ||
1d2ef590 | 3722 | dget(new_dentry); |
d83c49f3 | 3723 | if (target) |
1b1dcc1b | 3724 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3725 | |
3726 | error = -EBUSY; | |
3727 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3728 | goto out; | |
3729 | ||
8de52778 AV |
3730 | error = -EMLINK; |
3731 | if (max_links && !target && new_dir != old_dir && | |
3732 | new_dir->i_nlink >= max_links) | |
3733 | goto out; | |
3734 | ||
3cebde24 SW |
3735 | if (target) |
3736 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3737 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3738 | if (error) | |
3739 | goto out; | |
3740 | ||
1da177e4 | 3741 | if (target) { |
9055cba7 SW |
3742 | target->i_flags |= S_DEAD; |
3743 | dont_mount(new_dentry); | |
1da177e4 | 3744 | } |
9055cba7 SW |
3745 | out: |
3746 | if (target) | |
3747 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 3748 | dput(new_dentry); |
e31e14ec | 3749 | if (!error) |
349457cc MF |
3750 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3751 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
3752 | return error; |
3753 | } | |
3754 | ||
75c96f85 AB |
3755 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3756 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 3757 | { |
51892bbb | 3758 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
3759 | int error; |
3760 | ||
3761 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3762 | if (error) | |
3763 | return error; | |
3764 | ||
3765 | dget(new_dentry); | |
1da177e4 | 3766 | if (target) |
1b1dcc1b | 3767 | mutex_lock(&target->i_mutex); |
51892bbb SW |
3768 | |
3769 | error = -EBUSY; | |
1da177e4 | 3770 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
3771 | goto out; |
3772 | ||
3773 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
3774 | if (error) | |
3775 | goto out; | |
3776 | ||
3777 | if (target) | |
3778 | dont_mount(new_dentry); | |
3779 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
3780 | d_move(old_dentry, new_dentry); | |
3781 | out: | |
1da177e4 | 3782 | if (target) |
1b1dcc1b | 3783 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3784 | dput(new_dentry); |
3785 | return error; | |
3786 | } | |
3787 | ||
3788 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
3789 | struct inode *new_dir, struct dentry *new_dentry) | |
3790 | { | |
3791 | int error; | |
3792 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 3793 | const unsigned char *old_name; |
1da177e4 LT |
3794 | |
3795 | if (old_dentry->d_inode == new_dentry->d_inode) | |
3796 | return 0; | |
3797 | ||
3798 | error = may_delete(old_dir, old_dentry, is_dir); | |
3799 | if (error) | |
3800 | return error; | |
3801 | ||
3802 | if (!new_dentry->d_inode) | |
a95164d9 | 3803 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
3804 | else |
3805 | error = may_delete(new_dir, new_dentry, is_dir); | |
3806 | if (error) | |
3807 | return error; | |
3808 | ||
acfa4380 | 3809 | if (!old_dir->i_op->rename) |
1da177e4 LT |
3810 | return -EPERM; |
3811 | ||
0eeca283 RL |
3812 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3813 | ||
1da177e4 LT |
3814 | if (is_dir) |
3815 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
3816 | else | |
3817 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
3818 | if (!error) |
3819 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 3820 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
3821 | fsnotify_oldname_free(old_name); |
3822 | ||
1da177e4 LT |
3823 | return error; |
3824 | } | |
3825 | ||
2e4d0924 HC |
3826 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3827 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3828 | { |
2ad94ae6 AV |
3829 | struct dentry *old_dir, *new_dir; |
3830 | struct dentry *old_dentry, *new_dentry; | |
3831 | struct dentry *trap; | |
1da177e4 | 3832 | struct nameidata oldnd, newnd; |
91a27b2a JL |
3833 | struct filename *from; |
3834 | struct filename *to; | |
c6a94284 JL |
3835 | unsigned int lookup_flags = 0; |
3836 | bool should_retry = false; | |
2ad94ae6 | 3837 | int error; |
c6a94284 JL |
3838 | retry: |
3839 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
3840 | if (IS_ERR(from)) { |
3841 | error = PTR_ERR(from); | |
1da177e4 | 3842 | goto exit; |
91a27b2a | 3843 | } |
1da177e4 | 3844 | |
c6a94284 | 3845 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
3846 | if (IS_ERR(to)) { |
3847 | error = PTR_ERR(to); | |
1da177e4 | 3848 | goto exit1; |
91a27b2a | 3849 | } |
1da177e4 LT |
3850 | |
3851 | error = -EXDEV; | |
4ac91378 | 3852 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
3853 | goto exit2; |
3854 | ||
4ac91378 | 3855 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
3856 | error = -EBUSY; |
3857 | if (oldnd.last_type != LAST_NORM) | |
3858 | goto exit2; | |
3859 | ||
4ac91378 | 3860 | new_dir = newnd.path.dentry; |
1da177e4 LT |
3861 | if (newnd.last_type != LAST_NORM) |
3862 | goto exit2; | |
3863 | ||
c30dabfe JK |
3864 | error = mnt_want_write(oldnd.path.mnt); |
3865 | if (error) | |
3866 | goto exit2; | |
3867 | ||
0612d9fb OH |
3868 | oldnd.flags &= ~LOOKUP_PARENT; |
3869 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 3870 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 3871 | |
1da177e4 LT |
3872 | trap = lock_rename(new_dir, old_dir); |
3873 | ||
49705b77 | 3874 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
3875 | error = PTR_ERR(old_dentry); |
3876 | if (IS_ERR(old_dentry)) | |
3877 | goto exit3; | |
3878 | /* source must exist */ | |
3879 | error = -ENOENT; | |
3880 | if (!old_dentry->d_inode) | |
3881 | goto exit4; | |
3882 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
3883 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
3884 | error = -ENOTDIR; | |
3885 | if (oldnd.last.name[oldnd.last.len]) | |
3886 | goto exit4; | |
3887 | if (newnd.last.name[newnd.last.len]) | |
3888 | goto exit4; | |
3889 | } | |
3890 | /* source should not be ancestor of target */ | |
3891 | error = -EINVAL; | |
3892 | if (old_dentry == trap) | |
3893 | goto exit4; | |
49705b77 | 3894 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
3895 | error = PTR_ERR(new_dentry); |
3896 | if (IS_ERR(new_dentry)) | |
3897 | goto exit4; | |
3898 | /* target should not be an ancestor of source */ | |
3899 | error = -ENOTEMPTY; | |
3900 | if (new_dentry == trap) | |
3901 | goto exit5; | |
3902 | ||
be6d3e56 KT |
3903 | error = security_path_rename(&oldnd.path, old_dentry, |
3904 | &newnd.path, new_dentry); | |
3905 | if (error) | |
c30dabfe | 3906 | goto exit5; |
1da177e4 LT |
3907 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3908 | new_dir->d_inode, new_dentry); | |
3909 | exit5: | |
3910 | dput(new_dentry); | |
3911 | exit4: | |
3912 | dput(old_dentry); | |
3913 | exit3: | |
3914 | unlock_rename(new_dir, old_dir); | |
c30dabfe | 3915 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 3916 | exit2: |
c6a94284 JL |
3917 | if (retry_estale(error, lookup_flags)) |
3918 | should_retry = true; | |
1d957f9b | 3919 | path_put(&newnd.path); |
2ad94ae6 | 3920 | putname(to); |
1da177e4 | 3921 | exit1: |
1d957f9b | 3922 | path_put(&oldnd.path); |
1da177e4 | 3923 | putname(from); |
c6a94284 JL |
3924 | if (should_retry) { |
3925 | should_retry = false; | |
3926 | lookup_flags |= LOOKUP_REVAL; | |
3927 | goto retry; | |
3928 | } | |
2ad94ae6 | 3929 | exit: |
1da177e4 LT |
3930 | return error; |
3931 | } | |
3932 | ||
a26eab24 | 3933 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3934 | { |
3935 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
3936 | } | |
3937 | ||
1da177e4 LT |
3938 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3939 | { | |
3940 | int len; | |
3941 | ||
3942 | len = PTR_ERR(link); | |
3943 | if (IS_ERR(link)) | |
3944 | goto out; | |
3945 | ||
3946 | len = strlen(link); | |
3947 | if (len > (unsigned) buflen) | |
3948 | len = buflen; | |
3949 | if (copy_to_user(buffer, link, len)) | |
3950 | len = -EFAULT; | |
3951 | out: | |
3952 | return len; | |
3953 | } | |
3954 | ||
3955 | /* | |
3956 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
3957 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
3958 | * using) it for any given inode is up to filesystem. | |
3959 | */ | |
3960 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3961 | { | |
3962 | struct nameidata nd; | |
cc314eef | 3963 | void *cookie; |
694a1764 | 3964 | int res; |
cc314eef | 3965 | |
1da177e4 | 3966 | nd.depth = 0; |
cc314eef | 3967 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
3968 | if (IS_ERR(cookie)) |
3969 | return PTR_ERR(cookie); | |
3970 | ||
3971 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
3972 | if (dentry->d_inode->i_op->put_link) | |
3973 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
3974 | return res; | |
1da177e4 LT |
3975 | } |
3976 | ||
3977 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
3978 | { | |
3979 | return __vfs_follow_link(nd, link); | |
3980 | } | |
3981 | ||
3982 | /* get the link contents into pagecache */ | |
3983 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
3984 | { | |
ebd09abb DG |
3985 | char *kaddr; |
3986 | struct page *page; | |
1da177e4 | 3987 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 3988 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 3989 | if (IS_ERR(page)) |
6fe6900e | 3990 | return (char*)page; |
1da177e4 | 3991 | *ppage = page; |
ebd09abb DG |
3992 | kaddr = kmap(page); |
3993 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
3994 | return kaddr; | |
1da177e4 LT |
3995 | } |
3996 | ||
3997 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
3998 | { | |
3999 | struct page *page = NULL; | |
4000 | char *s = page_getlink(dentry, &page); | |
4001 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4002 | if (page) { | |
4003 | kunmap(page); | |
4004 | page_cache_release(page); | |
4005 | } | |
4006 | return res; | |
4007 | } | |
4008 | ||
cc314eef | 4009 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4010 | { |
cc314eef | 4011 | struct page *page = NULL; |
1da177e4 | 4012 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4013 | return page; |
1da177e4 LT |
4014 | } |
4015 | ||
cc314eef | 4016 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4017 | { |
cc314eef LT |
4018 | struct page *page = cookie; |
4019 | ||
4020 | if (page) { | |
1da177e4 LT |
4021 | kunmap(page); |
4022 | page_cache_release(page); | |
1da177e4 LT |
4023 | } |
4024 | } | |
4025 | ||
54566b2c NP |
4026 | /* |
4027 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4028 | */ | |
4029 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4030 | { |
4031 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4032 | struct page *page; |
afddba49 | 4033 | void *fsdata; |
beb497ab | 4034 | int err; |
1da177e4 | 4035 | char *kaddr; |
54566b2c NP |
4036 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4037 | if (nofs) | |
4038 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4039 | |
7e53cac4 | 4040 | retry: |
afddba49 | 4041 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4042 | flags, &page, &fsdata); |
1da177e4 | 4043 | if (err) |
afddba49 NP |
4044 | goto fail; |
4045 | ||
e8e3c3d6 | 4046 | kaddr = kmap_atomic(page); |
1da177e4 | 4047 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4048 | kunmap_atomic(kaddr); |
afddba49 NP |
4049 | |
4050 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4051 | page, fsdata); | |
1da177e4 LT |
4052 | if (err < 0) |
4053 | goto fail; | |
afddba49 NP |
4054 | if (err < len-1) |
4055 | goto retry; | |
4056 | ||
1da177e4 LT |
4057 | mark_inode_dirty(inode); |
4058 | return 0; | |
1da177e4 LT |
4059 | fail: |
4060 | return err; | |
4061 | } | |
4062 | ||
0adb25d2 KK |
4063 | int page_symlink(struct inode *inode, const char *symname, int len) |
4064 | { | |
4065 | return __page_symlink(inode, symname, len, | |
54566b2c | 4066 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4067 | } |
4068 | ||
92e1d5be | 4069 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4070 | .readlink = generic_readlink, |
4071 | .follow_link = page_follow_link_light, | |
4072 | .put_link = page_put_link, | |
4073 | }; | |
4074 | ||
2d8f3038 | 4075 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4076 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4077 | EXPORT_SYMBOL(follow_down); |
4078 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4079 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4080 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4081 | EXPORT_SYMBOL(lookup_one_len); |
4082 | EXPORT_SYMBOL(page_follow_link_light); | |
4083 | EXPORT_SYMBOL(page_put_link); | |
4084 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4085 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4086 | EXPORT_SYMBOL(page_symlink); |
4087 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4088 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4089 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4090 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4091 | EXPORT_SYMBOL(unlock_rename); |
4092 | EXPORT_SYMBOL(vfs_create); | |
4093 | EXPORT_SYMBOL(vfs_follow_link); | |
4094 | EXPORT_SYMBOL(vfs_link); | |
4095 | EXPORT_SYMBOL(vfs_mkdir); | |
4096 | EXPORT_SYMBOL(vfs_mknod); | |
4097 | EXPORT_SYMBOL(generic_permission); | |
4098 | EXPORT_SYMBOL(vfs_readlink); | |
4099 | EXPORT_SYMBOL(vfs_rename); | |
4100 | EXPORT_SYMBOL(vfs_rmdir); | |
4101 | EXPORT_SYMBOL(vfs_symlink); | |
4102 | EXPORT_SYMBOL(vfs_unlink); | |
4103 | EXPORT_SYMBOL(dentry_unhash); | |
4104 | EXPORT_SYMBOL(generic_readlink); |