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 */ | |
1a48e2ac | 324 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
325 | return 0; |
326 | if (!(mask & MAY_WRITE)) | |
1a48e2ac | 327 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
d594e7ec AV |
328 | return 0; |
329 | return -EACCES; | |
330 | } | |
1da177e4 LT |
331 | /* |
332 | * Read/write DACs are always overridable. | |
d594e7ec AV |
333 | * Executable DACs are overridable when there is |
334 | * at least one exec bit set. | |
1da177e4 | 335 | */ |
d594e7ec | 336 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
1a48e2ac | 337 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
338 | return 0; |
339 | ||
340 | /* | |
341 | * Searching includes executable on directories, else just read. | |
342 | */ | |
7ea66001 | 343 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 344 | if (mask == MAY_READ) |
1a48e2ac | 345 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
346 | return 0; |
347 | ||
348 | return -EACCES; | |
349 | } | |
350 | ||
3ddcd056 LT |
351 | /* |
352 | * We _really_ want to just do "generic_permission()" without | |
353 | * even looking at the inode->i_op values. So we keep a cache | |
354 | * flag in inode->i_opflags, that says "this has not special | |
355 | * permission function, use the fast case". | |
356 | */ | |
357 | static inline int do_inode_permission(struct inode *inode, int mask) | |
358 | { | |
359 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
360 | if (likely(inode->i_op->permission)) | |
361 | return inode->i_op->permission(inode, mask); | |
362 | ||
363 | /* This gets set once for the inode lifetime */ | |
364 | spin_lock(&inode->i_lock); | |
365 | inode->i_opflags |= IOP_FASTPERM; | |
366 | spin_unlock(&inode->i_lock); | |
367 | } | |
368 | return generic_permission(inode, mask); | |
369 | } | |
370 | ||
cb23beb5 | 371 | /** |
0bdaea90 DH |
372 | * __inode_permission - Check for access rights to a given inode |
373 | * @inode: Inode to check permission on | |
374 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 375 | * |
0bdaea90 | 376 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
377 | * |
378 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
379 | * |
380 | * This does not check for a read-only file system. You probably want | |
381 | * inode_permission(). | |
cb23beb5 | 382 | */ |
0bdaea90 | 383 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 384 | { |
e6305c43 | 385 | int retval; |
1da177e4 | 386 | |
3ddcd056 | 387 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
388 | /* |
389 | * Nobody gets write access to an immutable file. | |
390 | */ | |
391 | if (IS_IMMUTABLE(inode)) | |
392 | return -EACCES; | |
393 | } | |
394 | ||
3ddcd056 | 395 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
396 | if (retval) |
397 | return retval; | |
398 | ||
08ce5f16 SH |
399 | retval = devcgroup_inode_permission(inode, mask); |
400 | if (retval) | |
401 | return retval; | |
402 | ||
d09ca739 | 403 | return security_inode_permission(inode, mask); |
1da177e4 LT |
404 | } |
405 | ||
0bdaea90 DH |
406 | /** |
407 | * sb_permission - Check superblock-level permissions | |
408 | * @sb: Superblock of inode to check permission on | |
55852635 | 409 | * @inode: Inode to check permission on |
0bdaea90 DH |
410 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
411 | * | |
412 | * Separate out file-system wide checks from inode-specific permission checks. | |
413 | */ | |
414 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
415 | { | |
416 | if (unlikely(mask & MAY_WRITE)) { | |
417 | umode_t mode = inode->i_mode; | |
418 | ||
419 | /* Nobody gets write access to a read-only fs. */ | |
420 | if ((sb->s_flags & MS_RDONLY) && | |
421 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
422 | return -EROFS; | |
423 | } | |
424 | return 0; | |
425 | } | |
426 | ||
427 | /** | |
428 | * inode_permission - Check for access rights to a given inode | |
429 | * @inode: Inode to check permission on | |
430 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
431 | * | |
432 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
433 | * this, letting us set arbitrary permissions for filesystem access without | |
434 | * changing the "normal" UIDs which are used for other things. | |
435 | * | |
436 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
437 | */ | |
438 | int inode_permission(struct inode *inode, int mask) | |
439 | { | |
440 | int retval; | |
441 | ||
442 | retval = sb_permission(inode->i_sb, inode, mask); | |
443 | if (retval) | |
444 | return retval; | |
445 | return __inode_permission(inode, mask); | |
446 | } | |
447 | ||
5dd784d0 JB |
448 | /** |
449 | * path_get - get a reference to a path | |
450 | * @path: path to get the reference to | |
451 | * | |
452 | * Given a path increment the reference count to the dentry and the vfsmount. | |
453 | */ | |
dcf787f3 | 454 | void path_get(const struct path *path) |
5dd784d0 JB |
455 | { |
456 | mntget(path->mnt); | |
457 | dget(path->dentry); | |
458 | } | |
459 | EXPORT_SYMBOL(path_get); | |
460 | ||
1d957f9b JB |
461 | /** |
462 | * path_put - put a reference to a path | |
463 | * @path: path to put the reference to | |
464 | * | |
465 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
466 | */ | |
dcf787f3 | 467 | void path_put(const struct path *path) |
1da177e4 | 468 | { |
1d957f9b JB |
469 | dput(path->dentry); |
470 | mntput(path->mnt); | |
1da177e4 | 471 | } |
1d957f9b | 472 | EXPORT_SYMBOL(path_put); |
1da177e4 | 473 | |
19660af7 | 474 | /* |
31e6b01f | 475 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
476 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
477 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
478 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
479 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
480 | * got stuck, so ref-walk may continue from there. If this is not successful | |
481 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
482 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 483 | */ |
31e6b01f | 484 | |
32a7991b AV |
485 | static inline void lock_rcu_walk(void) |
486 | { | |
487 | br_read_lock(&vfsmount_lock); | |
488 | rcu_read_lock(); | |
489 | } | |
490 | ||
491 | static inline void unlock_rcu_walk(void) | |
492 | { | |
493 | rcu_read_unlock(); | |
494 | br_read_unlock(&vfsmount_lock); | |
495 | } | |
496 | ||
31e6b01f | 497 | /** |
19660af7 AV |
498 | * unlazy_walk - try to switch to ref-walk mode. |
499 | * @nd: nameidata pathwalk data | |
500 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 501 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 502 | * |
19660af7 AV |
503 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
504 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
505 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 506 | */ |
19660af7 | 507 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
508 | { |
509 | struct fs_struct *fs = current->fs; | |
510 | struct dentry *parent = nd->path.dentry; | |
511 | ||
512 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
513 | |
514 | /* | |
515 | * Get a reference to the parent first: we're | |
516 | * going to make "path_put(nd->path)" valid in | |
517 | * non-RCU context for "terminate_walk()". | |
518 | * | |
519 | * If this doesn't work, return immediately with | |
520 | * RCU walking still active (and then we will do | |
521 | * the RCU walk cleanup in terminate_walk()). | |
522 | */ | |
523 | if (!lockref_get_not_dead(&parent->d_lockref)) | |
524 | return -ECHILD; | |
525 | ||
526 | /* | |
527 | * After the mntget(), we terminate_walk() will do | |
528 | * the right thing for non-RCU mode, and all our | |
529 | * subsequent exit cases should unlock_rcu_walk() | |
530 | * before returning. | |
531 | */ | |
532 | mntget(nd->path.mnt); | |
533 | nd->flags &= ~LOOKUP_RCU; | |
15570086 LT |
534 | |
535 | /* | |
536 | * For a negative lookup, the lookup sequence point is the parents | |
537 | * sequence point, and it only needs to revalidate the parent dentry. | |
538 | * | |
539 | * For a positive lookup, we need to move both the parent and the | |
540 | * dentry from the RCU domain to be properly refcounted. And the | |
541 | * sequence number in the dentry validates *both* dentry counters, | |
542 | * since we checked the sequence number of the parent after we got | |
543 | * the child sequence number. So we know the parent must still | |
544 | * be valid if the child sequence number is still valid. | |
545 | */ | |
19660af7 | 546 | if (!dentry) { |
e5c832d5 LT |
547 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
548 | goto out; | |
19660af7 AV |
549 | BUG_ON(nd->inode != parent->d_inode); |
550 | } else { | |
e5c832d5 LT |
551 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
552 | goto out; | |
553 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
554 | goto drop_dentry; | |
19660af7 | 555 | } |
e5c832d5 LT |
556 | |
557 | /* | |
558 | * Sequence counts matched. Now make sure that the root is | |
559 | * still valid and get it if required. | |
560 | */ | |
561 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
562 | spin_lock(&fs->lock); | |
563 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
564 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
565 | path_get(&nd->root); |
566 | spin_unlock(&fs->lock); | |
567 | } | |
31e6b01f | 568 | |
32a7991b | 569 | unlock_rcu_walk(); |
31e6b01f | 570 | return 0; |
19660af7 | 571 | |
e5c832d5 LT |
572 | unlock_and_drop_dentry: |
573 | spin_unlock(&fs->lock); | |
574 | drop_dentry: | |
575 | unlock_rcu_walk(); | |
15570086 | 576 | dput(dentry); |
d0d27277 | 577 | goto drop_root_mnt; |
e5c832d5 LT |
578 | out: |
579 | unlock_rcu_walk(); | |
d0d27277 LT |
580 | drop_root_mnt: |
581 | if (!(nd->flags & LOOKUP_ROOT)) | |
582 | nd->root.mnt = NULL; | |
31e6b01f NP |
583 | return -ECHILD; |
584 | } | |
585 | ||
4ce16ef3 | 586 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 587 | { |
4ce16ef3 | 588 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
589 | } |
590 | ||
9f1fafee AV |
591 | /** |
592 | * complete_walk - successful completion of path walk | |
593 | * @nd: pointer nameidata | |
39159de2 | 594 | * |
9f1fafee AV |
595 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
596 | * Revalidate the final result, unless we'd already done that during | |
597 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
598 | * success, -error on failure. In case of failure caller does not | |
599 | * need to drop nd->path. | |
39159de2 | 600 | */ |
9f1fafee | 601 | static int complete_walk(struct nameidata *nd) |
39159de2 | 602 | { |
16c2cd71 | 603 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 604 | int status; |
39159de2 | 605 | |
9f1fafee AV |
606 | if (nd->flags & LOOKUP_RCU) { |
607 | nd->flags &= ~LOOKUP_RCU; | |
608 | if (!(nd->flags & LOOKUP_ROOT)) | |
609 | nd->root.mnt = NULL; | |
15570086 | 610 | |
e5c832d5 LT |
611 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
612 | unlock_rcu_walk(); | |
613 | return -ECHILD; | |
614 | } | |
615 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
32a7991b | 616 | unlock_rcu_walk(); |
e5c832d5 | 617 | dput(dentry); |
9f1fafee AV |
618 | return -ECHILD; |
619 | } | |
9f1fafee | 620 | mntget(nd->path.mnt); |
32a7991b | 621 | unlock_rcu_walk(); |
9f1fafee AV |
622 | } |
623 | ||
16c2cd71 AV |
624 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
625 | return 0; | |
626 | ||
ecf3d1f1 | 627 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
628 | return 0; |
629 | ||
ecf3d1f1 | 630 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
631 | if (status > 0) |
632 | return 0; | |
633 | ||
16c2cd71 | 634 | if (!status) |
39159de2 | 635 | status = -ESTALE; |
16c2cd71 | 636 | |
9f1fafee | 637 | path_put(&nd->path); |
39159de2 JL |
638 | return status; |
639 | } | |
640 | ||
2a737871 AV |
641 | static __always_inline void set_root(struct nameidata *nd) |
642 | { | |
f7ad3c6b MS |
643 | if (!nd->root.mnt) |
644 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
645 | } |
646 | ||
6de88d72 AV |
647 | static int link_path_walk(const char *, struct nameidata *); |
648 | ||
31e6b01f NP |
649 | static __always_inline void set_root_rcu(struct nameidata *nd) |
650 | { | |
651 | if (!nd->root.mnt) { | |
652 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
653 | unsigned seq; |
654 | ||
655 | do { | |
656 | seq = read_seqcount_begin(&fs->seq); | |
657 | nd->root = fs->root; | |
c1530019 | 658 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 659 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
660 | } |
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) { | |
443ed254 AV |
854 | if (unlikely(IS_ERR(s))) { |
855 | path_put(&nd->path); | |
856 | put_link(nd, link, *p); | |
857 | return PTR_ERR(s); | |
858 | } | |
859 | if (*s == '/') { | |
860 | set_root(nd); | |
861 | path_put(&nd->path); | |
862 | nd->path = nd->root; | |
863 | path_get(&nd->root); | |
864 | nd->flags |= LOOKUP_JUMPED; | |
865 | } | |
866 | nd->inode = nd->path.dentry->d_inode; | |
867 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
868 | if (unlikely(error)) |
869 | put_link(nd, link, *p); | |
1da177e4 | 870 | } |
6d7b5aae AV |
871 | |
872 | return error; | |
873 | ||
874 | out_put_nd_path: | |
98f6ef64 | 875 | *p = NULL; |
6d7b5aae | 876 | path_put(&nd->path); |
6d7b5aae | 877 | path_put(link); |
1da177e4 LT |
878 | return error; |
879 | } | |
880 | ||
31e6b01f NP |
881 | static int follow_up_rcu(struct path *path) |
882 | { | |
0714a533 AV |
883 | struct mount *mnt = real_mount(path->mnt); |
884 | struct mount *parent; | |
31e6b01f NP |
885 | struct dentry *mountpoint; |
886 | ||
0714a533 AV |
887 | parent = mnt->mnt_parent; |
888 | if (&parent->mnt == path->mnt) | |
31e6b01f | 889 | return 0; |
a73324da | 890 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 891 | path->dentry = mountpoint; |
0714a533 | 892 | path->mnt = &parent->mnt; |
31e6b01f NP |
893 | return 1; |
894 | } | |
895 | ||
f015f126 DH |
896 | /* |
897 | * follow_up - Find the mountpoint of path's vfsmount | |
898 | * | |
899 | * Given a path, find the mountpoint of its source file system. | |
900 | * Replace @path with the path of the mountpoint in the parent mount. | |
901 | * Up is towards /. | |
902 | * | |
903 | * Return 1 if we went up a level and 0 if we were already at the | |
904 | * root. | |
905 | */ | |
bab77ebf | 906 | int follow_up(struct path *path) |
1da177e4 | 907 | { |
0714a533 AV |
908 | struct mount *mnt = real_mount(path->mnt); |
909 | struct mount *parent; | |
1da177e4 | 910 | struct dentry *mountpoint; |
99b7db7b | 911 | |
962830df | 912 | br_read_lock(&vfsmount_lock); |
0714a533 | 913 | parent = mnt->mnt_parent; |
3c0a6163 | 914 | if (parent == mnt) { |
962830df | 915 | br_read_unlock(&vfsmount_lock); |
1da177e4 LT |
916 | return 0; |
917 | } | |
0714a533 | 918 | mntget(&parent->mnt); |
a73324da | 919 | mountpoint = dget(mnt->mnt_mountpoint); |
962830df | 920 | br_read_unlock(&vfsmount_lock); |
bab77ebf AV |
921 | dput(path->dentry); |
922 | path->dentry = mountpoint; | |
923 | mntput(path->mnt); | |
0714a533 | 924 | path->mnt = &parent->mnt; |
1da177e4 LT |
925 | return 1; |
926 | } | |
927 | ||
b5c84bf6 | 928 | /* |
9875cf80 DH |
929 | * Perform an automount |
930 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
931 | * were called with. | |
1da177e4 | 932 | */ |
9875cf80 DH |
933 | static int follow_automount(struct path *path, unsigned flags, |
934 | bool *need_mntput) | |
31e6b01f | 935 | { |
9875cf80 | 936 | struct vfsmount *mnt; |
ea5b778a | 937 | int err; |
9875cf80 DH |
938 | |
939 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
940 | return -EREMOTE; | |
941 | ||
0ec26fd0 MS |
942 | /* We don't want to mount if someone's just doing a stat - |
943 | * unless they're stat'ing a directory and appended a '/' to | |
944 | * the name. | |
945 | * | |
946 | * We do, however, want to mount if someone wants to open or | |
947 | * create a file of any type under the mountpoint, wants to | |
948 | * traverse through the mountpoint or wants to open the | |
949 | * mounted directory. Also, autofs may mark negative dentries | |
950 | * as being automount points. These will need the attentions | |
951 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 952 | */ |
0ec26fd0 | 953 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 954 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
955 | path->dentry->d_inode) |
956 | return -EISDIR; | |
957 | ||
9875cf80 DH |
958 | current->total_link_count++; |
959 | if (current->total_link_count >= 40) | |
960 | return -ELOOP; | |
961 | ||
962 | mnt = path->dentry->d_op->d_automount(path); | |
963 | if (IS_ERR(mnt)) { | |
964 | /* | |
965 | * The filesystem is allowed to return -EISDIR here to indicate | |
966 | * it doesn't want to automount. For instance, autofs would do | |
967 | * this so that its userspace daemon can mount on this dentry. | |
968 | * | |
969 | * However, we can only permit this if it's a terminal point in | |
970 | * the path being looked up; if it wasn't then the remainder of | |
971 | * the path is inaccessible and we should say so. | |
972 | */ | |
49084c3b | 973 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
974 | return -EREMOTE; |
975 | return PTR_ERR(mnt); | |
31e6b01f | 976 | } |
ea5b778a | 977 | |
9875cf80 DH |
978 | if (!mnt) /* mount collision */ |
979 | return 0; | |
31e6b01f | 980 | |
8aef1884 AV |
981 | if (!*need_mntput) { |
982 | /* lock_mount() may release path->mnt on error */ | |
983 | mntget(path->mnt); | |
984 | *need_mntput = true; | |
985 | } | |
19a167af | 986 | err = finish_automount(mnt, path); |
9875cf80 | 987 | |
ea5b778a DH |
988 | switch (err) { |
989 | case -EBUSY: | |
990 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 991 | return 0; |
ea5b778a | 992 | case 0: |
8aef1884 | 993 | path_put(path); |
ea5b778a DH |
994 | path->mnt = mnt; |
995 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 996 | return 0; |
19a167af AV |
997 | default: |
998 | return err; | |
ea5b778a | 999 | } |
19a167af | 1000 | |
463ffb2e AV |
1001 | } |
1002 | ||
9875cf80 DH |
1003 | /* |
1004 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1005 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1006 | * - Flagged as mountpoint |
1007 | * - Flagged as automount point | |
1008 | * | |
1009 | * This may only be called in refwalk mode. | |
1010 | * | |
1011 | * Serialization is taken care of in namespace.c | |
1012 | */ | |
1013 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1014 | { |
8aef1884 | 1015 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1016 | unsigned managed; |
1017 | bool need_mntput = false; | |
8aef1884 | 1018 | int ret = 0; |
9875cf80 DH |
1019 | |
1020 | /* Given that we're not holding a lock here, we retain the value in a | |
1021 | * local variable for each dentry as we look at it so that we don't see | |
1022 | * the components of that value change under us */ | |
1023 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1024 | managed &= DCACHE_MANAGED_DENTRY, | |
1025 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1026 | /* Allow the filesystem to manage the transit without i_mutex |
1027 | * being held. */ | |
1028 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1029 | BUG_ON(!path->dentry->d_op); | |
1030 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1031 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1032 | if (ret < 0) |
8aef1884 | 1033 | break; |
cc53ce53 DH |
1034 | } |
1035 | ||
9875cf80 DH |
1036 | /* Transit to a mounted filesystem. */ |
1037 | if (managed & DCACHE_MOUNTED) { | |
1038 | struct vfsmount *mounted = lookup_mnt(path); | |
1039 | if (mounted) { | |
1040 | dput(path->dentry); | |
1041 | if (need_mntput) | |
1042 | mntput(path->mnt); | |
1043 | path->mnt = mounted; | |
1044 | path->dentry = dget(mounted->mnt_root); | |
1045 | need_mntput = true; | |
1046 | continue; | |
1047 | } | |
1048 | ||
1049 | /* Something is mounted on this dentry in another | |
1050 | * namespace and/or whatever was mounted there in this | |
1051 | * namespace got unmounted before we managed to get the | |
1052 | * vfsmount_lock */ | |
1053 | } | |
1054 | ||
1055 | /* Handle an automount point */ | |
1056 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1057 | ret = follow_automount(path, flags, &need_mntput); | |
1058 | if (ret < 0) | |
8aef1884 | 1059 | break; |
9875cf80 DH |
1060 | continue; |
1061 | } | |
1062 | ||
1063 | /* We didn't change the current path point */ | |
1064 | break; | |
1da177e4 | 1065 | } |
8aef1884 AV |
1066 | |
1067 | if (need_mntput && path->mnt == mnt) | |
1068 | mntput(path->mnt); | |
1069 | if (ret == -EISDIR) | |
1070 | ret = 0; | |
a3fbbde7 | 1071 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1072 | } |
1073 | ||
cc53ce53 | 1074 | int follow_down_one(struct path *path) |
1da177e4 LT |
1075 | { |
1076 | struct vfsmount *mounted; | |
1077 | ||
1c755af4 | 1078 | mounted = lookup_mnt(path); |
1da177e4 | 1079 | if (mounted) { |
9393bd07 AV |
1080 | dput(path->dentry); |
1081 | mntput(path->mnt); | |
1082 | path->mnt = mounted; | |
1083 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1084 | return 1; |
1085 | } | |
1086 | return 0; | |
1087 | } | |
1088 | ||
62a7375e IK |
1089 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
1090 | { | |
1091 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
1092 | dentry->d_op->d_manage(dentry, true) < 0); | |
1093 | } | |
1094 | ||
9875cf80 | 1095 | /* |
287548e4 AV |
1096 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1097 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1098 | */ |
1099 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1100 | struct inode **inode) |
9875cf80 | 1101 | { |
62a7375e | 1102 | for (;;) { |
c7105365 | 1103 | struct mount *mounted; |
62a7375e IK |
1104 | /* |
1105 | * Don't forget we might have a non-mountpoint managed dentry | |
1106 | * that wants to block transit. | |
1107 | */ | |
287548e4 | 1108 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 1109 | return false; |
62a7375e IK |
1110 | |
1111 | if (!d_mountpoint(path->dentry)) | |
1112 | break; | |
1113 | ||
9875cf80 DH |
1114 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1115 | if (!mounted) | |
1116 | break; | |
c7105365 AV |
1117 | path->mnt = &mounted->mnt; |
1118 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1119 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1120 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1121 | /* |
1122 | * Update the inode too. We don't need to re-check the | |
1123 | * dentry sequence number here after this d_inode read, | |
1124 | * because a mount-point is always pinned. | |
1125 | */ | |
1126 | *inode = path->dentry->d_inode; | |
9875cf80 | 1127 | } |
9875cf80 DH |
1128 | return true; |
1129 | } | |
1130 | ||
dea39376 | 1131 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 1132 | { |
dea39376 | 1133 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 1134 | struct mount *mounted; |
dea39376 | 1135 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
287548e4 AV |
1136 | if (!mounted) |
1137 | break; | |
c7105365 AV |
1138 | nd->path.mnt = &mounted->mnt; |
1139 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 1140 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
1141 | } |
1142 | } | |
1143 | ||
31e6b01f NP |
1144 | static int follow_dotdot_rcu(struct nameidata *nd) |
1145 | { | |
31e6b01f NP |
1146 | set_root_rcu(nd); |
1147 | ||
9875cf80 | 1148 | while (1) { |
31e6b01f NP |
1149 | if (nd->path.dentry == nd->root.dentry && |
1150 | nd->path.mnt == nd->root.mnt) { | |
1151 | break; | |
1152 | } | |
1153 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1154 | struct dentry *old = nd->path.dentry; | |
1155 | struct dentry *parent = old->d_parent; | |
1156 | unsigned seq; | |
1157 | ||
1158 | seq = read_seqcount_begin(&parent->d_seq); | |
1159 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1160 | goto failed; |
31e6b01f NP |
1161 | nd->path.dentry = parent; |
1162 | nd->seq = seq; | |
1163 | break; | |
1164 | } | |
1165 | if (!follow_up_rcu(&nd->path)) | |
1166 | break; | |
1167 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1168 | } |
dea39376 AV |
1169 | follow_mount_rcu(nd); |
1170 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 1171 | return 0; |
ef7562d5 AV |
1172 | |
1173 | failed: | |
1174 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1175 | if (!(nd->flags & LOOKUP_ROOT)) |
1176 | nd->root.mnt = NULL; | |
32a7991b | 1177 | unlock_rcu_walk(); |
ef7562d5 | 1178 | return -ECHILD; |
31e6b01f NP |
1179 | } |
1180 | ||
cc53ce53 DH |
1181 | /* |
1182 | * Follow down to the covering mount currently visible to userspace. At each | |
1183 | * point, the filesystem owning that dentry may be queried as to whether the | |
1184 | * caller is permitted to proceed or not. | |
cc53ce53 | 1185 | */ |
7cc90cc3 | 1186 | int follow_down(struct path *path) |
cc53ce53 DH |
1187 | { |
1188 | unsigned managed; | |
1189 | int ret; | |
1190 | ||
1191 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1192 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1193 | /* Allow the filesystem to manage the transit without i_mutex | |
1194 | * being held. | |
1195 | * | |
1196 | * We indicate to the filesystem if someone is trying to mount | |
1197 | * something here. This gives autofs the chance to deny anyone | |
1198 | * other than its daemon the right to mount on its | |
1199 | * superstructure. | |
1200 | * | |
1201 | * The filesystem may sleep at this point. | |
1202 | */ | |
1203 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1204 | BUG_ON(!path->dentry->d_op); | |
1205 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1206 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1207 | path->dentry, false); |
cc53ce53 DH |
1208 | if (ret < 0) |
1209 | return ret == -EISDIR ? 0 : ret; | |
1210 | } | |
1211 | ||
1212 | /* Transit to a mounted filesystem. */ | |
1213 | if (managed & DCACHE_MOUNTED) { | |
1214 | struct vfsmount *mounted = lookup_mnt(path); | |
1215 | if (!mounted) | |
1216 | break; | |
1217 | dput(path->dentry); | |
1218 | mntput(path->mnt); | |
1219 | path->mnt = mounted; | |
1220 | path->dentry = dget(mounted->mnt_root); | |
1221 | continue; | |
1222 | } | |
1223 | ||
1224 | /* Don't handle automount points here */ | |
1225 | break; | |
1226 | } | |
1227 | return 0; | |
1228 | } | |
1229 | ||
9875cf80 DH |
1230 | /* |
1231 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1232 | */ | |
1233 | static void follow_mount(struct path *path) | |
1234 | { | |
1235 | while (d_mountpoint(path->dentry)) { | |
1236 | struct vfsmount *mounted = lookup_mnt(path); | |
1237 | if (!mounted) | |
1238 | break; | |
1239 | dput(path->dentry); | |
1240 | mntput(path->mnt); | |
1241 | path->mnt = mounted; | |
1242 | path->dentry = dget(mounted->mnt_root); | |
1243 | } | |
1244 | } | |
1245 | ||
31e6b01f | 1246 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1247 | { |
2a737871 | 1248 | set_root(nd); |
e518ddb7 | 1249 | |
1da177e4 | 1250 | while(1) { |
4ac91378 | 1251 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1252 | |
2a737871 AV |
1253 | if (nd->path.dentry == nd->root.dentry && |
1254 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1255 | break; |
1256 | } | |
4ac91378 | 1257 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1258 | /* rare case of legitimate dget_parent()... */ |
1259 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1260 | dput(old); |
1261 | break; | |
1262 | } | |
3088dd70 | 1263 | if (!follow_up(&nd->path)) |
1da177e4 | 1264 | break; |
1da177e4 | 1265 | } |
79ed0226 | 1266 | follow_mount(&nd->path); |
31e6b01f | 1267 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1268 | } |
1269 | ||
baa03890 | 1270 | /* |
bad61189 MS |
1271 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1272 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1273 | * returns whether i_op->lookup is necessary. | |
1274 | * | |
1275 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1276 | */ |
bad61189 | 1277 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1278 | unsigned int flags, bool *need_lookup) |
baa03890 | 1279 | { |
baa03890 | 1280 | struct dentry *dentry; |
bad61189 | 1281 | int error; |
baa03890 | 1282 | |
bad61189 MS |
1283 | *need_lookup = false; |
1284 | dentry = d_lookup(dir, name); | |
1285 | if (dentry) { | |
39e3c955 | 1286 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1287 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1288 | if (unlikely(error <= 0)) { |
1289 | if (error < 0) { | |
1290 | dput(dentry); | |
1291 | return ERR_PTR(error); | |
1292 | } else if (!d_invalidate(dentry)) { | |
1293 | dput(dentry); | |
1294 | dentry = NULL; | |
1295 | } | |
1296 | } | |
1297 | } | |
1298 | } | |
baa03890 | 1299 | |
bad61189 MS |
1300 | if (!dentry) { |
1301 | dentry = d_alloc(dir, name); | |
1302 | if (unlikely(!dentry)) | |
1303 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1304 | |
bad61189 | 1305 | *need_lookup = true; |
baa03890 NP |
1306 | } |
1307 | return dentry; | |
1308 | } | |
1309 | ||
44396f4b | 1310 | /* |
bad61189 MS |
1311 | * Call i_op->lookup on the dentry. The dentry must be negative but may be |
1312 | * hashed if it was pouplated with DCACHE_NEED_LOOKUP. | |
1313 | * | |
1314 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1315 | */ |
bad61189 | 1316 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1317 | unsigned int flags) |
44396f4b | 1318 | { |
44396f4b JB |
1319 | struct dentry *old; |
1320 | ||
1321 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1322 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1323 | dput(dentry); |
44396f4b | 1324 | return ERR_PTR(-ENOENT); |
e188dc02 | 1325 | } |
44396f4b | 1326 | |
72bd866a | 1327 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1328 | if (unlikely(old)) { |
1329 | dput(dentry); | |
1330 | dentry = old; | |
1331 | } | |
1332 | return dentry; | |
1333 | } | |
1334 | ||
a3255546 | 1335 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1336 | struct dentry *base, unsigned int flags) |
a3255546 | 1337 | { |
bad61189 | 1338 | bool need_lookup; |
a3255546 AV |
1339 | struct dentry *dentry; |
1340 | ||
72bd866a | 1341 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1342 | if (!need_lookup) |
1343 | return dentry; | |
a3255546 | 1344 | |
72bd866a | 1345 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1346 | } |
1347 | ||
1da177e4 LT |
1348 | /* |
1349 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1350 | * small and for now I'd prefer to have fast path as straight as possible. | |
1351 | * It _is_ time-critical. | |
1352 | */ | |
e97cdc87 | 1353 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1354 | struct path *path, struct inode **inode) |
1da177e4 | 1355 | { |
4ac91378 | 1356 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1357 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1358 | int need_reval = 1; |
1359 | int status = 1; | |
9875cf80 DH |
1360 | int err; |
1361 | ||
b04f784e NP |
1362 | /* |
1363 | * Rename seqlock is not required here because in the off chance | |
1364 | * of a false negative due to a concurrent rename, we're going to | |
1365 | * do the non-racy lookup, below. | |
1366 | */ | |
31e6b01f NP |
1367 | if (nd->flags & LOOKUP_RCU) { |
1368 | unsigned seq; | |
da53be12 | 1369 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1370 | if (!dentry) |
1371 | goto unlazy; | |
1372 | ||
12f8ad4b LT |
1373 | /* |
1374 | * This sequence count validates that the inode matches | |
1375 | * the dentry name information from lookup. | |
1376 | */ | |
1377 | *inode = dentry->d_inode; | |
1378 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1379 | return -ECHILD; | |
1380 | ||
1381 | /* | |
1382 | * This sequence count validates that the parent had no | |
1383 | * changes while we did the lookup of the dentry above. | |
1384 | * | |
1385 | * The memory barrier in read_seqcount_begin of child is | |
1386 | * enough, we can use __read_seqcount_retry here. | |
1387 | */ | |
31e6b01f NP |
1388 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1389 | return -ECHILD; | |
31e6b01f | 1390 | nd->seq = seq; |
5a18fff2 | 1391 | |
24643087 | 1392 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1393 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1394 | if (unlikely(status <= 0)) { |
1395 | if (status != -ECHILD) | |
1396 | need_reval = 0; | |
1397 | goto unlazy; | |
1398 | } | |
24643087 | 1399 | } |
31e6b01f NP |
1400 | path->mnt = mnt; |
1401 | path->dentry = dentry; | |
d6e9bd25 AV |
1402 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1403 | goto unlazy; | |
1404 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1405 | goto unlazy; | |
1406 | return 0; | |
5a18fff2 | 1407 | unlazy: |
19660af7 AV |
1408 | if (unlazy_walk(nd, dentry)) |
1409 | return -ECHILD; | |
5a18fff2 | 1410 | } else { |
e97cdc87 | 1411 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1412 | } |
5a18fff2 | 1413 | |
81e6f520 AV |
1414 | if (unlikely(!dentry)) |
1415 | goto need_lookup; | |
1416 | ||
5a18fff2 | 1417 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1418 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1419 | if (unlikely(status <= 0)) { |
1420 | if (status < 0) { | |
1421 | dput(dentry); | |
1422 | return status; | |
1423 | } | |
1424 | if (!d_invalidate(dentry)) { | |
1425 | dput(dentry); | |
81e6f520 | 1426 | goto need_lookup; |
5a18fff2 | 1427 | } |
24643087 | 1428 | } |
697f514d | 1429 | |
9875cf80 DH |
1430 | path->mnt = mnt; |
1431 | path->dentry = dentry; | |
1432 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1433 | if (unlikely(err < 0)) { |
1434 | path_put_conditional(path, nd); | |
9875cf80 | 1435 | return err; |
89312214 | 1436 | } |
a3fbbde7 AV |
1437 | if (err) |
1438 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1439 | *inode = path->dentry->d_inode; |
1da177e4 | 1440 | return 0; |
81e6f520 AV |
1441 | |
1442 | need_lookup: | |
697f514d MS |
1443 | return 1; |
1444 | } | |
1445 | ||
1446 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1447 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1448 | { |
1449 | struct dentry *dentry, *parent; | |
1450 | int err; | |
1451 | ||
1452 | parent = nd->path.dentry; | |
81e6f520 AV |
1453 | BUG_ON(nd->inode != parent->d_inode); |
1454 | ||
1455 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1456 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1457 | mutex_unlock(&parent->d_inode->i_mutex); |
1458 | if (IS_ERR(dentry)) | |
1459 | return PTR_ERR(dentry); | |
697f514d MS |
1460 | path->mnt = nd->path.mnt; |
1461 | path->dentry = dentry; | |
1462 | err = follow_managed(path, nd->flags); | |
1463 | if (unlikely(err < 0)) { | |
1464 | path_put_conditional(path, nd); | |
1465 | return err; | |
1466 | } | |
1467 | if (err) | |
1468 | nd->flags |= LOOKUP_JUMPED; | |
1469 | return 0; | |
1da177e4 LT |
1470 | } |
1471 | ||
52094c8a AV |
1472 | static inline int may_lookup(struct nameidata *nd) |
1473 | { | |
1474 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1475 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1476 | if (err != -ECHILD) |
1477 | return err; | |
19660af7 | 1478 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1479 | return -ECHILD; |
1480 | } | |
4ad5abb3 | 1481 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1482 | } |
1483 | ||
9856fa1b AV |
1484 | static inline int handle_dots(struct nameidata *nd, int type) |
1485 | { | |
1486 | if (type == LAST_DOTDOT) { | |
1487 | if (nd->flags & LOOKUP_RCU) { | |
1488 | if (follow_dotdot_rcu(nd)) | |
1489 | return -ECHILD; | |
1490 | } else | |
1491 | follow_dotdot(nd); | |
1492 | } | |
1493 | return 0; | |
1494 | } | |
1495 | ||
951361f9 AV |
1496 | static void terminate_walk(struct nameidata *nd) |
1497 | { | |
1498 | if (!(nd->flags & LOOKUP_RCU)) { | |
1499 | path_put(&nd->path); | |
1500 | } else { | |
1501 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1502 | if (!(nd->flags & LOOKUP_ROOT)) |
1503 | nd->root.mnt = NULL; | |
32a7991b | 1504 | unlock_rcu_walk(); |
951361f9 AV |
1505 | } |
1506 | } | |
1507 | ||
3ddcd056 LT |
1508 | /* |
1509 | * Do we need to follow links? We _really_ want to be able | |
1510 | * to do this check without having to look at inode->i_op, | |
1511 | * so we keep a cache of "no, this doesn't need follow_link" | |
1512 | * for the common case. | |
1513 | */ | |
7813b94a | 1514 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1515 | { |
1516 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1517 | if (likely(inode->i_op->follow_link)) | |
1518 | return follow; | |
1519 | ||
1520 | /* This gets set once for the inode lifetime */ | |
1521 | spin_lock(&inode->i_lock); | |
1522 | inode->i_opflags |= IOP_NOFOLLOW; | |
1523 | spin_unlock(&inode->i_lock); | |
1524 | } | |
1525 | return 0; | |
1526 | } | |
1527 | ||
ce57dfc1 | 1528 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1529 | int follow) |
ce57dfc1 AV |
1530 | { |
1531 | struct inode *inode; | |
1532 | int err; | |
1533 | /* | |
1534 | * "." and ".." are special - ".." especially so because it has | |
1535 | * to be able to know about the current root directory and | |
1536 | * parent relationships. | |
1537 | */ | |
21b9b073 AV |
1538 | if (unlikely(nd->last_type != LAST_NORM)) |
1539 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1540 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1541 | if (unlikely(err)) { |
697f514d MS |
1542 | if (err < 0) |
1543 | goto out_err; | |
1544 | ||
cc2a5271 | 1545 | err = lookup_slow(nd, path); |
697f514d MS |
1546 | if (err < 0) |
1547 | goto out_err; | |
1548 | ||
1549 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1550 | } |
697f514d MS |
1551 | err = -ENOENT; |
1552 | if (!inode) | |
1553 | goto out_path_put; | |
1554 | ||
7813b94a | 1555 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1556 | if (nd->flags & LOOKUP_RCU) { |
1557 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1558 | err = -ECHILD; |
1559 | goto out_err; | |
19660af7 AV |
1560 | } |
1561 | } | |
ce57dfc1 AV |
1562 | BUG_ON(inode != path->dentry->d_inode); |
1563 | return 1; | |
1564 | } | |
1565 | path_to_nameidata(path, nd); | |
1566 | nd->inode = inode; | |
1567 | return 0; | |
697f514d MS |
1568 | |
1569 | out_path_put: | |
1570 | path_to_nameidata(path, nd); | |
1571 | out_err: | |
1572 | terminate_walk(nd); | |
1573 | return err; | |
ce57dfc1 AV |
1574 | } |
1575 | ||
b356379a AV |
1576 | /* |
1577 | * This limits recursive symlink follows to 8, while | |
1578 | * limiting consecutive symlinks to 40. | |
1579 | * | |
1580 | * Without that kind of total limit, nasty chains of consecutive | |
1581 | * symlinks can cause almost arbitrarily long lookups. | |
1582 | */ | |
1583 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1584 | { | |
1585 | int res; | |
1586 | ||
b356379a AV |
1587 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1588 | path_put_conditional(path, nd); | |
1589 | path_put(&nd->path); | |
1590 | return -ELOOP; | |
1591 | } | |
1a4022f8 | 1592 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1593 | |
1594 | nd->depth++; | |
1595 | current->link_count++; | |
1596 | ||
1597 | do { | |
1598 | struct path link = *path; | |
1599 | void *cookie; | |
574197e0 AV |
1600 | |
1601 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1602 | if (res) |
1603 | break; | |
21b9b073 | 1604 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1605 | put_link(nd, &link, cookie); |
b356379a AV |
1606 | } while (res > 0); |
1607 | ||
1608 | current->link_count--; | |
1609 | nd->depth--; | |
1610 | return res; | |
1611 | } | |
1612 | ||
3ddcd056 LT |
1613 | /* |
1614 | * We really don't want to look at inode->i_op->lookup | |
1615 | * when we don't have to. So we keep a cache bit in | |
1616 | * the inode ->i_opflags field that says "yes, we can | |
1617 | * do lookup on this inode". | |
1618 | */ | |
1619 | static inline int can_lookup(struct inode *inode) | |
1620 | { | |
1621 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1622 | return 1; | |
1623 | if (likely(!inode->i_op->lookup)) | |
1624 | return 0; | |
1625 | ||
1626 | /* We do this once for the lifetime of the inode */ | |
1627 | spin_lock(&inode->i_lock); | |
1628 | inode->i_opflags |= IOP_LOOKUP; | |
1629 | spin_unlock(&inode->i_lock); | |
1630 | return 1; | |
1631 | } | |
1632 | ||
bfcfaa77 LT |
1633 | /* |
1634 | * We can do the critical dentry name comparison and hashing | |
1635 | * operations one word at a time, but we are limited to: | |
1636 | * | |
1637 | * - Architectures with fast unaligned word accesses. We could | |
1638 | * do a "get_unaligned()" if this helps and is sufficiently | |
1639 | * fast. | |
1640 | * | |
1641 | * - Little-endian machines (so that we can generate the mask | |
1642 | * of low bytes efficiently). Again, we *could* do a byte | |
1643 | * swapping load on big-endian architectures if that is not | |
1644 | * expensive enough to make the optimization worthless. | |
1645 | * | |
1646 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1647 | * do not trap on the (extremely unlikely) case of a page | |
1648 | * crossing operation. | |
1649 | * | |
1650 | * - Furthermore, we need an efficient 64-bit compile for the | |
1651 | * 64-bit case in order to generate the "number of bytes in | |
1652 | * the final mask". Again, that could be replaced with a | |
1653 | * efficient population count instruction or similar. | |
1654 | */ | |
1655 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1656 | ||
f68e556e | 1657 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1658 | |
f68e556e | 1659 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1660 | |
1661 | static inline unsigned int fold_hash(unsigned long hash) | |
1662 | { | |
1663 | hash += hash >> (8*sizeof(int)); | |
1664 | return hash; | |
1665 | } | |
1666 | ||
1667 | #else /* 32-bit case */ | |
1668 | ||
bfcfaa77 LT |
1669 | #define fold_hash(x) (x) |
1670 | ||
1671 | #endif | |
1672 | ||
1673 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1674 | { | |
1675 | unsigned long a, mask; | |
1676 | unsigned long hash = 0; | |
1677 | ||
1678 | for (;;) { | |
e419b4cc | 1679 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1680 | if (len < sizeof(unsigned long)) |
1681 | break; | |
1682 | hash += a; | |
f132c5be | 1683 | hash *= 9; |
bfcfaa77 LT |
1684 | name += sizeof(unsigned long); |
1685 | len -= sizeof(unsigned long); | |
1686 | if (!len) | |
1687 | goto done; | |
1688 | } | |
1689 | mask = ~(~0ul << len*8); | |
1690 | hash += mask & a; | |
1691 | done: | |
1692 | return fold_hash(hash); | |
1693 | } | |
1694 | EXPORT_SYMBOL(full_name_hash); | |
1695 | ||
bfcfaa77 LT |
1696 | /* |
1697 | * Calculate the length and hash of the path component, and | |
1698 | * return the length of the component; | |
1699 | */ | |
1700 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1701 | { | |
36126f8f LT |
1702 | unsigned long a, b, adata, bdata, mask, hash, len; |
1703 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1704 | |
1705 | hash = a = 0; | |
1706 | len = -sizeof(unsigned long); | |
1707 | do { | |
1708 | hash = (hash + a) * 9; | |
1709 | len += sizeof(unsigned long); | |
e419b4cc | 1710 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1711 | b = a ^ REPEAT_BYTE('/'); |
1712 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1713 | ||
1714 | adata = prep_zero_mask(a, adata, &constants); | |
1715 | bdata = prep_zero_mask(b, bdata, &constants); | |
1716 | ||
1717 | mask = create_zero_mask(adata | bdata); | |
1718 | ||
1719 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1720 | *hashp = fold_hash(hash); |
1721 | ||
36126f8f | 1722 | return len + find_zero(mask); |
bfcfaa77 LT |
1723 | } |
1724 | ||
1725 | #else | |
1726 | ||
0145acc2 LT |
1727 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1728 | { | |
1729 | unsigned long hash = init_name_hash(); | |
1730 | while (len--) | |
1731 | hash = partial_name_hash(*name++, hash); | |
1732 | return end_name_hash(hash); | |
1733 | } | |
ae942ae7 | 1734 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1735 | |
200e9ef7 LT |
1736 | /* |
1737 | * We know there's a real path component here of at least | |
1738 | * one character. | |
1739 | */ | |
1740 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1741 | { | |
1742 | unsigned long hash = init_name_hash(); | |
1743 | unsigned long len = 0, c; | |
1744 | ||
1745 | c = (unsigned char)*name; | |
1746 | do { | |
1747 | len++; | |
1748 | hash = partial_name_hash(c, hash); | |
1749 | c = (unsigned char)name[len]; | |
1750 | } while (c && c != '/'); | |
1751 | *hashp = end_name_hash(hash); | |
1752 | return len; | |
1753 | } | |
1754 | ||
bfcfaa77 LT |
1755 | #endif |
1756 | ||
1da177e4 LT |
1757 | /* |
1758 | * Name resolution. | |
ea3834d9 PM |
1759 | * This is the basic name resolution function, turning a pathname into |
1760 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1761 | * |
ea3834d9 PM |
1762 | * Returns 0 and nd will have valid dentry and mnt on success. |
1763 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1764 | */ |
6de88d72 | 1765 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1766 | { |
1767 | struct path next; | |
1da177e4 | 1768 | int err; |
1da177e4 LT |
1769 | |
1770 | while (*name=='/') | |
1771 | name++; | |
1772 | if (!*name) | |
086e183a | 1773 | return 0; |
1da177e4 | 1774 | |
1da177e4 LT |
1775 | /* At this point we know we have a real path component. */ |
1776 | for(;;) { | |
1da177e4 | 1777 | struct qstr this; |
200e9ef7 | 1778 | long len; |
fe479a58 | 1779 | int type; |
1da177e4 | 1780 | |
52094c8a | 1781 | err = may_lookup(nd); |
1da177e4 LT |
1782 | if (err) |
1783 | break; | |
1784 | ||
200e9ef7 | 1785 | len = hash_name(name, &this.hash); |
1da177e4 | 1786 | this.name = name; |
200e9ef7 | 1787 | this.len = len; |
1da177e4 | 1788 | |
fe479a58 | 1789 | type = LAST_NORM; |
200e9ef7 | 1790 | if (name[0] == '.') switch (len) { |
fe479a58 | 1791 | case 2: |
200e9ef7 | 1792 | if (name[1] == '.') { |
fe479a58 | 1793 | type = LAST_DOTDOT; |
16c2cd71 AV |
1794 | nd->flags |= LOOKUP_JUMPED; |
1795 | } | |
fe479a58 AV |
1796 | break; |
1797 | case 1: | |
1798 | type = LAST_DOT; | |
1799 | } | |
5a202bcd AV |
1800 | if (likely(type == LAST_NORM)) { |
1801 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1802 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1803 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1804 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1805 | if (err < 0) |
1806 | break; | |
1807 | } | |
1808 | } | |
fe479a58 | 1809 | |
5f4a6a69 AV |
1810 | nd->last = this; |
1811 | nd->last_type = type; | |
1812 | ||
200e9ef7 | 1813 | if (!name[len]) |
5f4a6a69 | 1814 | return 0; |
200e9ef7 LT |
1815 | /* |
1816 | * If it wasn't NUL, we know it was '/'. Skip that | |
1817 | * slash, and continue until no more slashes. | |
1818 | */ | |
1819 | do { | |
1820 | len++; | |
1821 | } while (unlikely(name[len] == '/')); | |
1822 | if (!name[len]) | |
5f4a6a69 AV |
1823 | return 0; |
1824 | ||
200e9ef7 | 1825 | name += len; |
1da177e4 | 1826 | |
21b9b073 | 1827 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1828 | if (err < 0) |
1829 | return err; | |
1da177e4 | 1830 | |
ce57dfc1 | 1831 | if (err) { |
b356379a | 1832 | err = nested_symlink(&next, nd); |
1da177e4 | 1833 | if (err) |
a7472bab | 1834 | return err; |
31e6b01f | 1835 | } |
5f4a6a69 AV |
1836 | if (!can_lookup(nd->inode)) { |
1837 | err = -ENOTDIR; | |
1838 | break; | |
1839 | } | |
1da177e4 | 1840 | } |
951361f9 | 1841 | terminate_walk(nd); |
1da177e4 LT |
1842 | return err; |
1843 | } | |
1844 | ||
70e9b357 AV |
1845 | static int path_init(int dfd, const char *name, unsigned int flags, |
1846 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1847 | { |
1848 | int retval = 0; | |
31e6b01f NP |
1849 | |
1850 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1851 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1852 | nd->depth = 0; |
5b6ca027 AV |
1853 | if (flags & LOOKUP_ROOT) { |
1854 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 | 1855 | if (*name) { |
741b7c3f | 1856 | if (!can_lookup(inode)) |
73d049a4 AV |
1857 | return -ENOTDIR; |
1858 | retval = inode_permission(inode, MAY_EXEC); | |
1859 | if (retval) | |
1860 | return retval; | |
1861 | } | |
5b6ca027 AV |
1862 | nd->path = nd->root; |
1863 | nd->inode = inode; | |
1864 | if (flags & LOOKUP_RCU) { | |
32a7991b | 1865 | lock_rcu_walk(); |
5b6ca027 AV |
1866 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1867 | } else { | |
1868 | path_get(&nd->path); | |
1869 | } | |
1870 | return 0; | |
1871 | } | |
1872 | ||
31e6b01f | 1873 | nd->root.mnt = NULL; |
31e6b01f NP |
1874 | |
1875 | if (*name=='/') { | |
e41f7d4e | 1876 | if (flags & LOOKUP_RCU) { |
32a7991b | 1877 | lock_rcu_walk(); |
e41f7d4e AV |
1878 | set_root_rcu(nd); |
1879 | } else { | |
1880 | set_root(nd); | |
1881 | path_get(&nd->root); | |
1882 | } | |
1883 | nd->path = nd->root; | |
31e6b01f | 1884 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1885 | if (flags & LOOKUP_RCU) { |
1886 | struct fs_struct *fs = current->fs; | |
1887 | unsigned seq; | |
31e6b01f | 1888 | |
32a7991b | 1889 | lock_rcu_walk(); |
c28cc364 | 1890 | |
e41f7d4e AV |
1891 | do { |
1892 | seq = read_seqcount_begin(&fs->seq); | |
1893 | nd->path = fs->pwd; | |
1894 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1895 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1896 | } else { | |
1897 | get_fs_pwd(current->fs, &nd->path); | |
1898 | } | |
31e6b01f | 1899 | } else { |
582aa64a | 1900 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1901 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1902 | struct dentry *dentry; |
1903 | ||
2903ff01 AV |
1904 | if (!f.file) |
1905 | return -EBADF; | |
31e6b01f | 1906 | |
2903ff01 | 1907 | dentry = f.file->f_path.dentry; |
31e6b01f | 1908 | |
f52e0c11 | 1909 | if (*name) { |
741b7c3f | 1910 | if (!can_lookup(dentry->d_inode)) { |
2903ff01 AV |
1911 | fdput(f); |
1912 | return -ENOTDIR; | |
1913 | } | |
f52e0c11 | 1914 | } |
31e6b01f | 1915 | |
2903ff01 | 1916 | nd->path = f.file->f_path; |
e41f7d4e | 1917 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1918 | if (f.need_put) |
1919 | *fp = f.file; | |
e41f7d4e | 1920 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
32a7991b | 1921 | lock_rcu_walk(); |
e41f7d4e | 1922 | } else { |
2903ff01 AV |
1923 | path_get(&nd->path); |
1924 | fdput(f); | |
e41f7d4e | 1925 | } |
31e6b01f | 1926 | } |
31e6b01f | 1927 | |
31e6b01f | 1928 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1929 | return 0; |
9b4a9b14 AV |
1930 | } |
1931 | ||
bd92d7fe AV |
1932 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1933 | { | |
1934 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1935 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1936 | ||
1937 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1938 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1939 | } |
1940 | ||
9b4a9b14 | 1941 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1942 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1943 | unsigned int flags, struct nameidata *nd) |
1944 | { | |
70e9b357 | 1945 | struct file *base = NULL; |
bd92d7fe AV |
1946 | struct path path; |
1947 | int err; | |
31e6b01f NP |
1948 | |
1949 | /* | |
1950 | * Path walking is largely split up into 2 different synchronisation | |
1951 | * schemes, rcu-walk and ref-walk (explained in | |
1952 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1953 | * path walk code, but some things particularly setup, cleanup, and | |
1954 | * following mounts are sufficiently divergent that functions are | |
1955 | * duplicated. Typically there is a function foo(), and its RCU | |
1956 | * analogue, foo_rcu(). | |
1957 | * | |
1958 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1959 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1960 | * be handled by restarting a traditional ref-walk (which will always | |
1961 | * be able to complete). | |
1962 | */ | |
bd92d7fe | 1963 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1964 | |
bd92d7fe AV |
1965 | if (unlikely(err)) |
1966 | return err; | |
ee0827cd AV |
1967 | |
1968 | current->total_link_count = 0; | |
bd92d7fe AV |
1969 | err = link_path_walk(name, nd); |
1970 | ||
1971 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1972 | err = lookup_last(nd, &path); |
1973 | while (err > 0) { | |
1974 | void *cookie; | |
1975 | struct path link = path; | |
800179c9 KC |
1976 | err = may_follow_link(&link, nd); |
1977 | if (unlikely(err)) | |
1978 | break; | |
bd92d7fe | 1979 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1980 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1981 | if (err) |
1982 | break; | |
1983 | err = lookup_last(nd, &path); | |
574197e0 | 1984 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1985 | } |
1986 | } | |
ee0827cd | 1987 | |
9f1fafee AV |
1988 | if (!err) |
1989 | err = complete_walk(nd); | |
bd92d7fe AV |
1990 | |
1991 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
05252901 | 1992 | if (!can_lookup(nd->inode)) { |
bd92d7fe | 1993 | path_put(&nd->path); |
bd23a539 | 1994 | err = -ENOTDIR; |
bd92d7fe AV |
1995 | } |
1996 | } | |
16c2cd71 | 1997 | |
70e9b357 AV |
1998 | if (base) |
1999 | fput(base); | |
ee0827cd | 2000 | |
5b6ca027 | 2001 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
2002 | path_put(&nd->root); |
2003 | nd->root.mnt = NULL; | |
2004 | } | |
bd92d7fe | 2005 | return err; |
ee0827cd | 2006 | } |
31e6b01f | 2007 | |
873f1eed | 2008 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
2009 | unsigned int flags, struct nameidata *nd) |
2010 | { | |
873f1eed | 2011 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 2012 | if (unlikely(retval == -ECHILD)) |
873f1eed | 2013 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 2014 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
2015 | retval = path_lookupat(dfd, name->name, |
2016 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 2017 | |
f78570dd | 2018 | if (likely(!retval)) |
adb5c247 | 2019 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 2020 | return retval; |
1da177e4 LT |
2021 | } |
2022 | ||
873f1eed JL |
2023 | static int do_path_lookup(int dfd, const char *name, |
2024 | unsigned int flags, struct nameidata *nd) | |
2025 | { | |
2026 | struct filename filename = { .name = name }; | |
2027 | ||
2028 | return filename_lookup(dfd, &filename, flags, nd); | |
2029 | } | |
2030 | ||
79714f72 AV |
2031 | /* does lookup, returns the object with parent locked */ |
2032 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2033 | { |
79714f72 AV |
2034 | struct nameidata nd; |
2035 | struct dentry *d; | |
2036 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2037 | if (err) | |
2038 | return ERR_PTR(err); | |
2039 | if (nd.last_type != LAST_NORM) { | |
2040 | path_put(&nd.path); | |
2041 | return ERR_PTR(-EINVAL); | |
2042 | } | |
2043 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2044 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2045 | if (IS_ERR(d)) { |
2046 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2047 | path_put(&nd.path); | |
2048 | return d; | |
2049 | } | |
2050 | *path = nd.path; | |
2051 | return d; | |
5590ff0d UD |
2052 | } |
2053 | ||
d1811465 AV |
2054 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2055 | { | |
2056 | struct nameidata nd; | |
2057 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2058 | if (!res) | |
2059 | *path = nd.path; | |
2060 | return res; | |
2061 | } | |
2062 | ||
16f18200 JJS |
2063 | /** |
2064 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2065 | * @dentry: pointer to dentry of the base directory | |
2066 | * @mnt: pointer to vfs mount of the base directory | |
2067 | * @name: pointer to file name | |
2068 | * @flags: lookup flags | |
e0a01249 | 2069 | * @path: pointer to struct path to fill |
16f18200 JJS |
2070 | */ |
2071 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2072 | const char *name, unsigned int flags, | |
e0a01249 | 2073 | struct path *path) |
16f18200 | 2074 | { |
e0a01249 AV |
2075 | struct nameidata nd; |
2076 | int err; | |
2077 | nd.root.dentry = dentry; | |
2078 | nd.root.mnt = mnt; | |
2079 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2080 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2081 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2082 | if (!err) | |
2083 | *path = nd.path; | |
2084 | return err; | |
16f18200 JJS |
2085 | } |
2086 | ||
057f6c01 JM |
2087 | /* |
2088 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2089 | * needs parent already locked. Doesn't follow mounts. | |
2090 | * SMP-safe. | |
2091 | */ | |
eead1911 | 2092 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2093 | { |
72bd866a | 2094 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2095 | } |
2096 | ||
eead1911 | 2097 | /** |
a6b91919 | 2098 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2099 | * @name: pathname component to lookup |
2100 | * @base: base directory to lookup from | |
2101 | * @len: maximum length @len should be interpreted to | |
2102 | * | |
a6b91919 RD |
2103 | * Note that this routine is purely a helper for filesystem usage and should |
2104 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2105 | * nameidata argument is passed to the filesystem methods and a filesystem |
2106 | * using this helper needs to be prepared for that. | |
2107 | */ | |
057f6c01 JM |
2108 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2109 | { | |
057f6c01 | 2110 | struct qstr this; |
6a96ba54 | 2111 | unsigned int c; |
cda309de | 2112 | int err; |
057f6c01 | 2113 | |
2f9092e1 DW |
2114 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2115 | ||
6a96ba54 AV |
2116 | this.name = name; |
2117 | this.len = len; | |
0145acc2 | 2118 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2119 | if (!len) |
2120 | return ERR_PTR(-EACCES); | |
2121 | ||
21d8a15a AV |
2122 | if (unlikely(name[0] == '.')) { |
2123 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2124 | return ERR_PTR(-EACCES); | |
2125 | } | |
2126 | ||
6a96ba54 AV |
2127 | while (len--) { |
2128 | c = *(const unsigned char *)name++; | |
2129 | if (c == '/' || c == '\0') | |
2130 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2131 | } |
5a202bcd AV |
2132 | /* |
2133 | * See if the low-level filesystem might want | |
2134 | * to use its own hash.. | |
2135 | */ | |
2136 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2137 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2138 | if (err < 0) |
2139 | return ERR_PTR(err); | |
2140 | } | |
eead1911 | 2141 | |
cda309de MS |
2142 | err = inode_permission(base->d_inode, MAY_EXEC); |
2143 | if (err) | |
2144 | return ERR_PTR(err); | |
2145 | ||
72bd866a | 2146 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2147 | } |
2148 | ||
1fa1e7f6 AW |
2149 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2150 | struct path *path, int *empty) | |
1da177e4 | 2151 | { |
2d8f3038 | 2152 | struct nameidata nd; |
91a27b2a | 2153 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2154 | int err = PTR_ERR(tmp); |
1da177e4 | 2155 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2156 | |
2157 | BUG_ON(flags & LOOKUP_PARENT); | |
2158 | ||
873f1eed | 2159 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2160 | putname(tmp); |
2d8f3038 AV |
2161 | if (!err) |
2162 | *path = nd.path; | |
1da177e4 LT |
2163 | } |
2164 | return err; | |
2165 | } | |
2166 | ||
1fa1e7f6 AW |
2167 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2168 | struct path *path) | |
2169 | { | |
f7493e5d | 2170 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2171 | } |
2172 | ||
873f1eed JL |
2173 | /* |
2174 | * NB: most callers don't do anything directly with the reference to the | |
2175 | * to struct filename, but the nd->last pointer points into the name string | |
2176 | * allocated by getname. So we must hold the reference to it until all | |
2177 | * path-walking is complete. | |
2178 | */ | |
91a27b2a | 2179 | static struct filename * |
9e790bd6 JL |
2180 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2181 | unsigned int flags) | |
2ad94ae6 | 2182 | { |
91a27b2a | 2183 | struct filename *s = getname(path); |
2ad94ae6 AV |
2184 | int error; |
2185 | ||
9e790bd6 JL |
2186 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2187 | flags &= LOOKUP_REVAL; | |
2188 | ||
2ad94ae6 | 2189 | if (IS_ERR(s)) |
91a27b2a | 2190 | return s; |
2ad94ae6 | 2191 | |
9e790bd6 | 2192 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2193 | if (error) { |
2ad94ae6 | 2194 | putname(s); |
91a27b2a JL |
2195 | return ERR_PTR(error); |
2196 | } | |
2ad94ae6 | 2197 | |
91a27b2a | 2198 | return s; |
2ad94ae6 AV |
2199 | } |
2200 | ||
8033426e | 2201 | /** |
197df04c | 2202 | * mountpoint_last - look up last component for umount |
8033426e JL |
2203 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2204 | * @path: pointer to container for result | |
2205 | * | |
2206 | * This is a special lookup_last function just for umount. In this case, we | |
2207 | * need to resolve the path without doing any revalidation. | |
2208 | * | |
2209 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2210 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2211 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2212 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2213 | * bogus and it doesn't exist. | |
2214 | * | |
2215 | * Returns: | |
2216 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2217 | * lookup found a negative dentry. The nd->path reference will also be | |
2218 | * put in this case. | |
2219 | * | |
2220 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2221 | * symlink that needs to be followed. "path" will also be populated. | |
2222 | * The nd->path reference will also be put. | |
2223 | * | |
2224 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2225 | * that needs to be followed. "path" will be populated with the path | |
2226 | * to the link, and nd->path will *not* be put. | |
2227 | */ | |
2228 | static int | |
197df04c | 2229 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2230 | { |
2231 | int error = 0; | |
2232 | struct dentry *dentry; | |
2233 | struct dentry *dir = nd->path.dentry; | |
2234 | ||
35759521 AV |
2235 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2236 | if (nd->flags & LOOKUP_RCU) { | |
2237 | if (unlazy_walk(nd, NULL)) { | |
2238 | error = -ECHILD; | |
2239 | goto out; | |
2240 | } | |
8033426e JL |
2241 | } |
2242 | ||
2243 | nd->flags &= ~LOOKUP_PARENT; | |
2244 | ||
2245 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2246 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2247 | if (error) |
2248 | goto out; | |
2249 | dentry = dget(nd->path.dentry); | |
2250 | goto done; | |
8033426e JL |
2251 | } |
2252 | ||
2253 | mutex_lock(&dir->d_inode->i_mutex); | |
2254 | dentry = d_lookup(dir, &nd->last); | |
2255 | if (!dentry) { | |
2256 | /* | |
2257 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2258 | * so that means that this dentry is probably a symlink or the | |
2259 | * path doesn't actually point to a mounted dentry. | |
2260 | */ | |
2261 | dentry = d_alloc(dir, &nd->last); | |
2262 | if (!dentry) { | |
2263 | error = -ENOMEM; | |
bcceeeba | 2264 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2265 | goto out; |
8033426e | 2266 | } |
35759521 AV |
2267 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2268 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2269 | if (IS_ERR(dentry)) { |
2270 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2271 | goto out; |
bcceeeba | 2272 | } |
8033426e JL |
2273 | } |
2274 | mutex_unlock(&dir->d_inode->i_mutex); | |
2275 | ||
35759521 AV |
2276 | done: |
2277 | if (!dentry->d_inode) { | |
2278 | error = -ENOENT; | |
2279 | dput(dentry); | |
2280 | goto out; | |
8033426e | 2281 | } |
35759521 AV |
2282 | path->dentry = dentry; |
2283 | path->mnt = mntget(nd->path.mnt); | |
2284 | if (should_follow_link(dentry->d_inode, nd->flags & LOOKUP_FOLLOW)) | |
2285 | return 1; | |
2286 | follow_mount(path); | |
2287 | error = 0; | |
2288 | out: | |
8033426e JL |
2289 | terminate_walk(nd); |
2290 | return error; | |
2291 | } | |
2292 | ||
2293 | /** | |
197df04c | 2294 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2295 | * @dfd: directory file descriptor to start walk from |
2296 | * @name: full pathname to walk | |
2297 | * @flags: lookup flags | |
8033426e JL |
2298 | * |
2299 | * Look up the given name, but don't attempt to revalidate the last component. | |
2300 | * Returns 0 and "path" will be valid on success; Retuns error otherwise. | |
2301 | */ | |
2302 | static int | |
197df04c | 2303 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2304 | { |
2305 | struct file *base = NULL; | |
2306 | struct nameidata nd; | |
2307 | int err; | |
2308 | ||
2309 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2310 | if (unlikely(err)) | |
2311 | return err; | |
2312 | ||
2313 | current->total_link_count = 0; | |
2314 | err = link_path_walk(name, &nd); | |
2315 | if (err) | |
2316 | goto out; | |
2317 | ||
197df04c | 2318 | err = mountpoint_last(&nd, path); |
8033426e JL |
2319 | while (err > 0) { |
2320 | void *cookie; | |
2321 | struct path link = *path; | |
2322 | err = may_follow_link(&link, &nd); | |
2323 | if (unlikely(err)) | |
2324 | break; | |
2325 | nd.flags |= LOOKUP_PARENT; | |
2326 | err = follow_link(&link, &nd, &cookie); | |
2327 | if (err) | |
2328 | break; | |
197df04c | 2329 | err = mountpoint_last(&nd, path); |
8033426e JL |
2330 | put_link(&nd, &link, cookie); |
2331 | } | |
2332 | out: | |
2333 | if (base) | |
2334 | fput(base); | |
2335 | ||
2336 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2337 | path_put(&nd.root); | |
2338 | ||
2339 | return err; | |
2340 | } | |
2341 | ||
2d864651 AV |
2342 | static int |
2343 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2344 | unsigned int flags) | |
2345 | { | |
2346 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2347 | if (unlikely(error == -ECHILD)) | |
2348 | error = path_mountpoint(dfd, s->name, path, flags); | |
2349 | if (unlikely(error == -ESTALE)) | |
2350 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2351 | if (likely(!error)) | |
2352 | audit_inode(s, path->dentry, 0); | |
2353 | return error; | |
2354 | } | |
2355 | ||
8033426e | 2356 | /** |
197df04c | 2357 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2358 | * @dfd: directory file descriptor |
2359 | * @name: pathname from userland | |
2360 | * @flags: lookup flags | |
2361 | * @path: pointer to container to hold result | |
2362 | * | |
2363 | * A umount is a special case for path walking. We're not actually interested | |
2364 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2365 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2366 | * and avoid revalidating the last component. | |
2367 | * | |
2368 | * Returns 0 and populates "path" on success. | |
2369 | */ | |
2370 | int | |
197df04c | 2371 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2372 | struct path *path) |
2373 | { | |
2374 | struct filename *s = getname(name); | |
2375 | int error; | |
8033426e JL |
2376 | if (IS_ERR(s)) |
2377 | return PTR_ERR(s); | |
2d864651 | 2378 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2379 | putname(s); |
2380 | return error; | |
2381 | } | |
2382 | ||
2d864651 AV |
2383 | int |
2384 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2385 | unsigned int flags) | |
2386 | { | |
2387 | struct filename s = {.name = name}; | |
2388 | return filename_mountpoint(dfd, &s, path, flags); | |
2389 | } | |
2390 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2391 | ||
1da177e4 LT |
2392 | /* |
2393 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2394 | * minimal. | |
2395 | */ | |
2396 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2397 | { | |
8e96e3b7 | 2398 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2399 | |
1da177e4 LT |
2400 | if (!(dir->i_mode & S_ISVTX)) |
2401 | return 0; | |
8e96e3b7 | 2402 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2403 | return 0; |
8e96e3b7 | 2404 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2405 | return 0; |
1a48e2ac | 2406 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2407 | } |
2408 | ||
2409 | /* | |
2410 | * Check whether we can remove a link victim from directory dir, check | |
2411 | * whether the type of victim is right. | |
2412 | * 1. We can't do it if dir is read-only (done in permission()) | |
2413 | * 2. We should have write and exec permissions on dir | |
2414 | * 3. We can't remove anything from append-only dir | |
2415 | * 4. We can't do anything with immutable dir (done in permission()) | |
2416 | * 5. If the sticky bit on dir is set we should either | |
2417 | * a. be owner of dir, or | |
2418 | * b. be owner of victim, or | |
2419 | * c. have CAP_FOWNER capability | |
2420 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2421 | * links pointing to it. | |
2422 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2423 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2424 | * 9. We can't remove a root or mountpoint. | |
2425 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2426 | * nfs_async_unlink(). | |
2427 | */ | |
858119e1 | 2428 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
2429 | { |
2430 | int error; | |
2431 | ||
2432 | if (!victim->d_inode) | |
2433 | return -ENOENT; | |
2434 | ||
2435 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2436 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2437 | |
f419a2e3 | 2438 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2439 | if (error) |
2440 | return error; | |
2441 | if (IS_APPEND(dir)) | |
2442 | return -EPERM; | |
2443 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2444 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2445 | return -EPERM; |
2446 | if (isdir) { | |
2447 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2448 | return -ENOTDIR; | |
2449 | if (IS_ROOT(victim)) | |
2450 | return -EBUSY; | |
2451 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2452 | return -EISDIR; | |
2453 | if (IS_DEADDIR(dir)) | |
2454 | return -ENOENT; | |
2455 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2456 | return -EBUSY; | |
2457 | return 0; | |
2458 | } | |
2459 | ||
2460 | /* Check whether we can create an object with dentry child in directory | |
2461 | * dir. | |
2462 | * 1. We can't do it if child already exists (open has special treatment for | |
2463 | * this case, but since we are inlined it's OK) | |
2464 | * 2. We can't do it if dir is read-only (done in permission()) | |
2465 | * 3. We should have write and exec permissions on dir | |
2466 | * 4. We can't do it if dir is immutable (done in permission()) | |
2467 | */ | |
a95164d9 | 2468 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2469 | { |
2470 | if (child->d_inode) | |
2471 | return -EEXIST; | |
2472 | if (IS_DEADDIR(dir)) | |
2473 | return -ENOENT; | |
f419a2e3 | 2474 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2475 | } |
2476 | ||
1da177e4 LT |
2477 | /* |
2478 | * p1 and p2 should be directories on the same fs. | |
2479 | */ | |
2480 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2481 | { | |
2482 | struct dentry *p; | |
2483 | ||
2484 | if (p1 == p2) { | |
f2eace23 | 2485 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2486 | return NULL; |
2487 | } | |
2488 | ||
a11f3a05 | 2489 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2490 | |
e2761a11 OH |
2491 | p = d_ancestor(p2, p1); |
2492 | if (p) { | |
2493 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2494 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2495 | return p; | |
1da177e4 LT |
2496 | } |
2497 | ||
e2761a11 OH |
2498 | p = d_ancestor(p1, p2); |
2499 | if (p) { | |
2500 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2501 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2502 | return p; | |
1da177e4 LT |
2503 | } |
2504 | ||
f2eace23 IM |
2505 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2506 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2507 | return NULL; |
2508 | } | |
2509 | ||
2510 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2511 | { | |
1b1dcc1b | 2512 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2513 | if (p1 != p2) { |
1b1dcc1b | 2514 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2515 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2516 | } |
2517 | } | |
2518 | ||
4acdaf27 | 2519 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2520 | bool want_excl) |
1da177e4 | 2521 | { |
a95164d9 | 2522 | int error = may_create(dir, dentry); |
1da177e4 LT |
2523 | if (error) |
2524 | return error; | |
2525 | ||
acfa4380 | 2526 | if (!dir->i_op->create) |
1da177e4 LT |
2527 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2528 | mode &= S_IALLUGO; | |
2529 | mode |= S_IFREG; | |
2530 | error = security_inode_create(dir, dentry, mode); | |
2531 | if (error) | |
2532 | return error; | |
312b63fb | 2533 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2534 | if (!error) |
f38aa942 | 2535 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2536 | return error; |
2537 | } | |
2538 | ||
73d049a4 | 2539 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2540 | { |
3fb64190 | 2541 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2542 | struct inode *inode = dentry->d_inode; |
2543 | int error; | |
2544 | ||
bcda7652 AV |
2545 | /* O_PATH? */ |
2546 | if (!acc_mode) | |
2547 | return 0; | |
2548 | ||
1da177e4 LT |
2549 | if (!inode) |
2550 | return -ENOENT; | |
2551 | ||
c8fe8f30 CH |
2552 | switch (inode->i_mode & S_IFMT) { |
2553 | case S_IFLNK: | |
1da177e4 | 2554 | return -ELOOP; |
c8fe8f30 CH |
2555 | case S_IFDIR: |
2556 | if (acc_mode & MAY_WRITE) | |
2557 | return -EISDIR; | |
2558 | break; | |
2559 | case S_IFBLK: | |
2560 | case S_IFCHR: | |
3fb64190 | 2561 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2562 | return -EACCES; |
c8fe8f30 CH |
2563 | /*FALLTHRU*/ |
2564 | case S_IFIFO: | |
2565 | case S_IFSOCK: | |
1da177e4 | 2566 | flag &= ~O_TRUNC; |
c8fe8f30 | 2567 | break; |
4a3fd211 | 2568 | } |
b41572e9 | 2569 | |
3fb64190 | 2570 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2571 | if (error) |
2572 | return error; | |
6146f0d5 | 2573 | |
1da177e4 LT |
2574 | /* |
2575 | * An append-only file must be opened in append mode for writing. | |
2576 | */ | |
2577 | if (IS_APPEND(inode)) { | |
8737c930 | 2578 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2579 | return -EPERM; |
1da177e4 | 2580 | if (flag & O_TRUNC) |
7715b521 | 2581 | return -EPERM; |
1da177e4 LT |
2582 | } |
2583 | ||
2584 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2585 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2586 | return -EPERM; |
1da177e4 | 2587 | |
f3c7691e | 2588 | return 0; |
7715b521 | 2589 | } |
1da177e4 | 2590 | |
e1181ee6 | 2591 | static int handle_truncate(struct file *filp) |
7715b521 | 2592 | { |
e1181ee6 | 2593 | struct path *path = &filp->f_path; |
7715b521 AV |
2594 | struct inode *inode = path->dentry->d_inode; |
2595 | int error = get_write_access(inode); | |
2596 | if (error) | |
2597 | return error; | |
2598 | /* | |
2599 | * Refuse to truncate files with mandatory locks held on them. | |
2600 | */ | |
2601 | error = locks_verify_locked(inode); | |
2602 | if (!error) | |
ea0d3ab2 | 2603 | error = security_path_truncate(path); |
7715b521 AV |
2604 | if (!error) { |
2605 | error = do_truncate(path->dentry, 0, | |
2606 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2607 | filp); |
7715b521 AV |
2608 | } |
2609 | put_write_access(inode); | |
acd0c935 | 2610 | return error; |
1da177e4 LT |
2611 | } |
2612 | ||
d57999e1 DH |
2613 | static inline int open_to_namei_flags(int flag) |
2614 | { | |
8a5e929d AV |
2615 | if ((flag & O_ACCMODE) == 3) |
2616 | flag--; | |
d57999e1 DH |
2617 | return flag; |
2618 | } | |
2619 | ||
d18e9008 MS |
2620 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2621 | { | |
2622 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2623 | if (error) | |
2624 | return error; | |
2625 | ||
2626 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2627 | if (error) | |
2628 | return error; | |
2629 | ||
2630 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2631 | } | |
2632 | ||
1acf0af9 DH |
2633 | /* |
2634 | * Attempt to atomically look up, create and open a file from a negative | |
2635 | * dentry. | |
2636 | * | |
2637 | * Returns 0 if successful. The file will have been created and attached to | |
2638 | * @file by the filesystem calling finish_open(). | |
2639 | * | |
2640 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2641 | * caller will need to perform the open themselves. @path will have been | |
2642 | * updated to point to the new dentry. This may be negative. | |
2643 | * | |
2644 | * Returns an error code otherwise. | |
2645 | */ | |
2675a4eb AV |
2646 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2647 | struct path *path, struct file *file, | |
2648 | const struct open_flags *op, | |
64894cf8 | 2649 | bool got_write, bool need_lookup, |
2675a4eb | 2650 | int *opened) |
d18e9008 MS |
2651 | { |
2652 | struct inode *dir = nd->path.dentry->d_inode; | |
2653 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2654 | umode_t mode; | |
2655 | int error; | |
2656 | int acc_mode; | |
d18e9008 MS |
2657 | int create_error = 0; |
2658 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2659 | bool excl; |
d18e9008 MS |
2660 | |
2661 | BUG_ON(dentry->d_inode); | |
2662 | ||
2663 | /* Don't create child dentry for a dead directory. */ | |
2664 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2665 | error = -ENOENT; |
d18e9008 MS |
2666 | goto out; |
2667 | } | |
2668 | ||
62b259d8 | 2669 | mode = op->mode; |
d18e9008 MS |
2670 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2671 | mode &= ~current_umask(); | |
2672 | ||
116cc022 MS |
2673 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2674 | if (excl) | |
d18e9008 | 2675 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2676 | |
2677 | /* | |
2678 | * Checking write permission is tricky, bacuse we don't know if we are | |
2679 | * going to actually need it: O_CREAT opens should work as long as the | |
2680 | * file exists. But checking existence breaks atomicity. The trick is | |
2681 | * to check access and if not granted clear O_CREAT from the flags. | |
2682 | * | |
2683 | * Another problem is returing the "right" error value (e.g. for an | |
2684 | * O_EXCL open we want to return EEXIST not EROFS). | |
2685 | */ | |
64894cf8 AV |
2686 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2687 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2688 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2689 | /* |
2690 | * No O_CREATE -> atomicity not a requirement -> fall | |
2691 | * back to lookup + open | |
2692 | */ | |
2693 | goto no_open; | |
2694 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2695 | /* Fall back and fail with the right error */ | |
64894cf8 | 2696 | create_error = -EROFS; |
d18e9008 MS |
2697 | goto no_open; |
2698 | } else { | |
2699 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2700 | create_error = -EROFS; |
d18e9008 MS |
2701 | open_flag &= ~O_CREAT; |
2702 | } | |
2703 | } | |
2704 | ||
2705 | if (open_flag & O_CREAT) { | |
38227f78 | 2706 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2707 | if (error) { |
2708 | create_error = error; | |
2709 | if (open_flag & O_EXCL) | |
2710 | goto no_open; | |
2711 | open_flag &= ~O_CREAT; | |
2712 | } | |
2713 | } | |
2714 | ||
2715 | if (nd->flags & LOOKUP_DIRECTORY) | |
2716 | open_flag |= O_DIRECTORY; | |
2717 | ||
30d90494 AV |
2718 | file->f_path.dentry = DENTRY_NOT_SET; |
2719 | file->f_path.mnt = nd->path.mnt; | |
2720 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2721 | opened); |
d9585277 | 2722 | if (error < 0) { |
d9585277 AV |
2723 | if (create_error && error == -ENOENT) |
2724 | error = create_error; | |
d18e9008 MS |
2725 | goto out; |
2726 | } | |
2727 | ||
d9585277 | 2728 | if (error) { /* returned 1, that is */ |
30d90494 | 2729 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2730 | error = -EIO; |
d18e9008 MS |
2731 | goto out; |
2732 | } | |
30d90494 | 2733 | if (file->f_path.dentry) { |
d18e9008 | 2734 | dput(dentry); |
30d90494 | 2735 | dentry = file->f_path.dentry; |
d18e9008 | 2736 | } |
03da633a AV |
2737 | if (*opened & FILE_CREATED) |
2738 | fsnotify_create(dir, dentry); | |
2739 | if (!dentry->d_inode) { | |
2740 | WARN_ON(*opened & FILE_CREATED); | |
2741 | if (create_error) { | |
2742 | error = create_error; | |
2743 | goto out; | |
2744 | } | |
2745 | } else { | |
2746 | if (excl && !(*opened & FILE_CREATED)) { | |
2747 | error = -EEXIST; | |
2748 | goto out; | |
2749 | } | |
62b2ce96 | 2750 | } |
d18e9008 MS |
2751 | goto looked_up; |
2752 | } | |
2753 | ||
2754 | /* | |
2755 | * We didn't have the inode before the open, so check open permission | |
2756 | * here. | |
2757 | */ | |
03da633a AV |
2758 | acc_mode = op->acc_mode; |
2759 | if (*opened & FILE_CREATED) { | |
2760 | WARN_ON(!(open_flag & O_CREAT)); | |
2761 | fsnotify_create(dir, dentry); | |
2762 | acc_mode = MAY_OPEN; | |
2763 | } | |
2675a4eb AV |
2764 | error = may_open(&file->f_path, acc_mode, open_flag); |
2765 | if (error) | |
2766 | fput(file); | |
d18e9008 MS |
2767 | |
2768 | out: | |
2769 | dput(dentry); | |
2675a4eb | 2770 | return error; |
d18e9008 | 2771 | |
d18e9008 MS |
2772 | no_open: |
2773 | if (need_lookup) { | |
72bd866a | 2774 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2775 | if (IS_ERR(dentry)) |
2675a4eb | 2776 | return PTR_ERR(dentry); |
d18e9008 MS |
2777 | |
2778 | if (create_error) { | |
2779 | int open_flag = op->open_flag; | |
2780 | ||
2675a4eb | 2781 | error = create_error; |
d18e9008 MS |
2782 | if ((open_flag & O_EXCL)) { |
2783 | if (!dentry->d_inode) | |
2784 | goto out; | |
2785 | } else if (!dentry->d_inode) { | |
2786 | goto out; | |
2787 | } else if ((open_flag & O_TRUNC) && | |
2788 | S_ISREG(dentry->d_inode->i_mode)) { | |
2789 | goto out; | |
2790 | } | |
2791 | /* will fail later, go on to get the right error */ | |
2792 | } | |
2793 | } | |
2794 | looked_up: | |
2795 | path->dentry = dentry; | |
2796 | path->mnt = nd->path.mnt; | |
2675a4eb | 2797 | return 1; |
d18e9008 MS |
2798 | } |
2799 | ||
d58ffd35 | 2800 | /* |
1acf0af9 | 2801 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2802 | * |
2803 | * Must be called with i_mutex held on parent. | |
2804 | * | |
1acf0af9 DH |
2805 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2806 | * opened. In this case the file will be returned attached to @file. | |
2807 | * | |
2808 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2809 | * and creations will have been performed and the dentry returned in @path will | |
2810 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2811 | * specified then a negative dentry may be returned. | |
2812 | * | |
2813 | * An error code is returned otherwise. | |
2814 | * | |
2815 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2816 | * cleared otherwise prior to returning. | |
d58ffd35 | 2817 | */ |
2675a4eb AV |
2818 | static int lookup_open(struct nameidata *nd, struct path *path, |
2819 | struct file *file, | |
2820 | const struct open_flags *op, | |
64894cf8 | 2821 | bool got_write, int *opened) |
d58ffd35 MS |
2822 | { |
2823 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2824 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2825 | struct dentry *dentry; |
2826 | int error; | |
54ef4872 | 2827 | bool need_lookup; |
d58ffd35 | 2828 | |
47237687 | 2829 | *opened &= ~FILE_CREATED; |
201f956e | 2830 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2831 | if (IS_ERR(dentry)) |
2675a4eb | 2832 | return PTR_ERR(dentry); |
d58ffd35 | 2833 | |
d18e9008 MS |
2834 | /* Cached positive dentry: will open in f_op->open */ |
2835 | if (!need_lookup && dentry->d_inode) | |
2836 | goto out_no_open; | |
2837 | ||
2838 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2839 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2840 | need_lookup, opened); |
d18e9008 MS |
2841 | } |
2842 | ||
54ef4872 MS |
2843 | if (need_lookup) { |
2844 | BUG_ON(dentry->d_inode); | |
2845 | ||
72bd866a | 2846 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2847 | if (IS_ERR(dentry)) |
2675a4eb | 2848 | return PTR_ERR(dentry); |
54ef4872 MS |
2849 | } |
2850 | ||
d58ffd35 MS |
2851 | /* Negative dentry, just create the file */ |
2852 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2853 | umode_t mode = op->mode; | |
2854 | if (!IS_POSIXACL(dir->d_inode)) | |
2855 | mode &= ~current_umask(); | |
2856 | /* | |
2857 | * This write is needed to ensure that a | |
2858 | * rw->ro transition does not occur between | |
2859 | * the time when the file is created and when | |
2860 | * a permanent write count is taken through | |
015c3bbc | 2861 | * the 'struct file' in finish_open(). |
d58ffd35 | 2862 | */ |
64894cf8 AV |
2863 | if (!got_write) { |
2864 | error = -EROFS; | |
d58ffd35 | 2865 | goto out_dput; |
64894cf8 | 2866 | } |
47237687 | 2867 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2868 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2869 | if (error) | |
2870 | goto out_dput; | |
312b63fb AV |
2871 | error = vfs_create(dir->d_inode, dentry, mode, |
2872 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2873 | if (error) |
2874 | goto out_dput; | |
2875 | } | |
d18e9008 | 2876 | out_no_open: |
d58ffd35 MS |
2877 | path->dentry = dentry; |
2878 | path->mnt = nd->path.mnt; | |
2675a4eb | 2879 | return 1; |
d58ffd35 MS |
2880 | |
2881 | out_dput: | |
2882 | dput(dentry); | |
2675a4eb | 2883 | return error; |
d58ffd35 MS |
2884 | } |
2885 | ||
31e6b01f | 2886 | /* |
fe2d35ff | 2887 | * Handle the last step of open() |
31e6b01f | 2888 | */ |
2675a4eb AV |
2889 | static int do_last(struct nameidata *nd, struct path *path, |
2890 | struct file *file, const struct open_flags *op, | |
669abf4e | 2891 | int *opened, struct filename *name) |
fb1cc555 | 2892 | { |
a1e28038 | 2893 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2894 | int open_flag = op->open_flag; |
77d660a8 | 2895 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2896 | bool got_write = false; |
bcda7652 | 2897 | int acc_mode = op->acc_mode; |
a1eb3315 | 2898 | struct inode *inode; |
77d660a8 | 2899 | bool symlink_ok = false; |
16b1c1cd MS |
2900 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2901 | bool retried = false; | |
16c2cd71 | 2902 | int error; |
1f36f774 | 2903 | |
c3e380b0 AV |
2904 | nd->flags &= ~LOOKUP_PARENT; |
2905 | nd->flags |= op->intent; | |
2906 | ||
bc77daa7 | 2907 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2908 | error = handle_dots(nd, nd->last_type); |
2909 | if (error) | |
2675a4eb | 2910 | return error; |
e83db167 | 2911 | goto finish_open; |
1f36f774 | 2912 | } |
67ee3ad2 | 2913 | |
ca344a89 | 2914 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2915 | if (nd->last.name[nd->last.len]) |
2916 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2917 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2918 | symlink_ok = true; |
fe2d35ff | 2919 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2920 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2921 | if (likely(!error)) |
2922 | goto finish_lookup; | |
2923 | ||
2924 | if (error < 0) | |
2675a4eb | 2925 | goto out; |
71574865 MS |
2926 | |
2927 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2928 | } else { |
2929 | /* create side of things */ | |
2930 | /* | |
2931 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2932 | * has been cleared when we got to the last component we are | |
2933 | * about to look up | |
2934 | */ | |
2935 | error = complete_walk(nd); | |
2936 | if (error) | |
2675a4eb | 2937 | return error; |
fe2d35ff | 2938 | |
33e2208a | 2939 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2940 | error = -EISDIR; |
2941 | /* trailing slashes? */ | |
2942 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2943 | goto out; |
b6183df7 | 2944 | } |
a2c36b45 | 2945 | |
16b1c1cd | 2946 | retry_lookup: |
64894cf8 AV |
2947 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2948 | error = mnt_want_write(nd->path.mnt); | |
2949 | if (!error) | |
2950 | got_write = true; | |
2951 | /* | |
2952 | * do _not_ fail yet - we might not need that or fail with | |
2953 | * a different error; let lookup_open() decide; we'll be | |
2954 | * dropping this one anyway. | |
2955 | */ | |
2956 | } | |
a1e28038 | 2957 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2958 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2959 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2960 | |
2675a4eb AV |
2961 | if (error <= 0) { |
2962 | if (error) | |
d18e9008 MS |
2963 | goto out; |
2964 | ||
47237687 | 2965 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2966 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2967 | will_truncate = false; |
d18e9008 | 2968 | |
adb5c247 | 2969 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2970 | goto opened; |
2971 | } | |
fb1cc555 | 2972 | |
47237687 | 2973 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2974 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2975 | open_flag &= ~O_TRUNC; |
77d660a8 | 2976 | will_truncate = false; |
bcda7652 | 2977 | acc_mode = MAY_OPEN; |
d58ffd35 | 2978 | path_to_nameidata(path, nd); |
e83db167 | 2979 | goto finish_open_created; |
fb1cc555 AV |
2980 | } |
2981 | ||
2982 | /* | |
3134f37e | 2983 | * create/update audit record if it already exists. |
fb1cc555 | 2984 | */ |
3134f37e | 2985 | if (path->dentry->d_inode) |
adb5c247 | 2986 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2987 | |
d18e9008 MS |
2988 | /* |
2989 | * If atomic_open() acquired write access it is dropped now due to | |
2990 | * possible mount and symlink following (this might be optimized away if | |
2991 | * necessary...) | |
2992 | */ | |
64894cf8 | 2993 | if (got_write) { |
d18e9008 | 2994 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2995 | got_write = false; |
d18e9008 MS |
2996 | } |
2997 | ||
fb1cc555 | 2998 | error = -EEXIST; |
f8310c59 | 2999 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
3000 | goto exit_dput; |
3001 | ||
9875cf80 DH |
3002 | error = follow_managed(path, nd->flags); |
3003 | if (error < 0) | |
3004 | goto exit_dput; | |
fb1cc555 | 3005 | |
a3fbbde7 AV |
3006 | if (error) |
3007 | nd->flags |= LOOKUP_JUMPED; | |
3008 | ||
decf3400 MS |
3009 | BUG_ON(nd->flags & LOOKUP_RCU); |
3010 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
3011 | finish_lookup: |
3012 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 3013 | error = -ENOENT; |
54c33e7f MS |
3014 | if (!inode) { |
3015 | path_to_nameidata(path, nd); | |
2675a4eb | 3016 | goto out; |
54c33e7f | 3017 | } |
9e67f361 | 3018 | |
d45ea867 MS |
3019 | if (should_follow_link(inode, !symlink_ok)) { |
3020 | if (nd->flags & LOOKUP_RCU) { | |
3021 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
3022 | error = -ECHILD; | |
2675a4eb | 3023 | goto out; |
d45ea867 MS |
3024 | } |
3025 | } | |
3026 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 3027 | return 1; |
d45ea867 | 3028 | } |
fb1cc555 | 3029 | |
16b1c1cd MS |
3030 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
3031 | path_to_nameidata(path, nd); | |
3032 | } else { | |
3033 | save_parent.dentry = nd->path.dentry; | |
3034 | save_parent.mnt = mntget(path->mnt); | |
3035 | nd->path.dentry = path->dentry; | |
3036 | ||
3037 | } | |
decf3400 | 3038 | nd->inode = inode; |
a3fbbde7 | 3039 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3040 | finish_open: |
a3fbbde7 | 3041 | error = complete_walk(nd); |
16b1c1cd MS |
3042 | if (error) { |
3043 | path_put(&save_parent); | |
2675a4eb | 3044 | return error; |
16b1c1cd | 3045 | } |
bc77daa7 | 3046 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3047 | error = -EISDIR; |
050ac841 | 3048 | if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) |
2675a4eb | 3049 | goto out; |
af2f5542 | 3050 | error = -ENOTDIR; |
05252901 | 3051 | if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode)) |
2675a4eb | 3052 | goto out; |
6c0d46c4 | 3053 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3054 | will_truncate = false; |
6c0d46c4 | 3055 | |
0f9d1a10 AV |
3056 | if (will_truncate) { |
3057 | error = mnt_want_write(nd->path.mnt); | |
3058 | if (error) | |
2675a4eb | 3059 | goto out; |
64894cf8 | 3060 | got_write = true; |
0f9d1a10 | 3061 | } |
e83db167 | 3062 | finish_open_created: |
bcda7652 | 3063 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3064 | if (error) |
2675a4eb | 3065 | goto out; |
30d90494 AV |
3066 | file->f_path.mnt = nd->path.mnt; |
3067 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3068 | if (error) { | |
30d90494 | 3069 | if (error == -EOPENSTALE) |
f60dc3db | 3070 | goto stale_open; |
015c3bbc | 3071 | goto out; |
f60dc3db | 3072 | } |
a8277b9b | 3073 | opened: |
2675a4eb | 3074 | error = open_check_o_direct(file); |
015c3bbc MS |
3075 | if (error) |
3076 | goto exit_fput; | |
2675a4eb | 3077 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3078 | if (error) |
3079 | goto exit_fput; | |
3080 | ||
3081 | if (will_truncate) { | |
2675a4eb | 3082 | error = handle_truncate(file); |
aa4caadb MS |
3083 | if (error) |
3084 | goto exit_fput; | |
0f9d1a10 | 3085 | } |
ca344a89 | 3086 | out: |
64894cf8 | 3087 | if (got_write) |
0f9d1a10 | 3088 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3089 | path_put(&save_parent); |
e276ae67 | 3090 | terminate_walk(nd); |
2675a4eb | 3091 | return error; |
fb1cc555 | 3092 | |
fb1cc555 AV |
3093 | exit_dput: |
3094 | path_put_conditional(path, nd); | |
ca344a89 | 3095 | goto out; |
015c3bbc | 3096 | exit_fput: |
2675a4eb AV |
3097 | fput(file); |
3098 | goto out; | |
015c3bbc | 3099 | |
f60dc3db MS |
3100 | stale_open: |
3101 | /* If no saved parent or already retried then can't retry */ | |
3102 | if (!save_parent.dentry || retried) | |
3103 | goto out; | |
3104 | ||
3105 | BUG_ON(save_parent.dentry != dir); | |
3106 | path_put(&nd->path); | |
3107 | nd->path = save_parent; | |
3108 | nd->inode = dir->d_inode; | |
3109 | save_parent.mnt = NULL; | |
3110 | save_parent.dentry = NULL; | |
64894cf8 | 3111 | if (got_write) { |
f60dc3db | 3112 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3113 | got_write = false; |
f60dc3db MS |
3114 | } |
3115 | retried = true; | |
3116 | goto retry_lookup; | |
fb1cc555 AV |
3117 | } |
3118 | ||
60545d0d AV |
3119 | static int do_tmpfile(int dfd, struct filename *pathname, |
3120 | struct nameidata *nd, int flags, | |
3121 | const struct open_flags *op, | |
3122 | struct file *file, int *opened) | |
3123 | { | |
3124 | static const struct qstr name = QSTR_INIT("/", 1); | |
3125 | struct dentry *dentry, *child; | |
3126 | struct inode *dir; | |
3127 | int error = path_lookupat(dfd, pathname->name, | |
3128 | flags | LOOKUP_DIRECTORY, nd); | |
3129 | if (unlikely(error)) | |
3130 | return error; | |
3131 | error = mnt_want_write(nd->path.mnt); | |
3132 | if (unlikely(error)) | |
3133 | goto out; | |
3134 | /* we want directory to be writable */ | |
3135 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3136 | if (error) | |
3137 | goto out2; | |
3138 | dentry = nd->path.dentry; | |
3139 | dir = dentry->d_inode; | |
3140 | if (!dir->i_op->tmpfile) { | |
3141 | error = -EOPNOTSUPP; | |
3142 | goto out2; | |
3143 | } | |
3144 | child = d_alloc(dentry, &name); | |
3145 | if (unlikely(!child)) { | |
3146 | error = -ENOMEM; | |
3147 | goto out2; | |
3148 | } | |
3149 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3150 | nd->flags |= op->intent; | |
3151 | dput(nd->path.dentry); | |
3152 | nd->path.dentry = child; | |
3153 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3154 | if (error) | |
3155 | goto out2; | |
3156 | audit_inode(pathname, nd->path.dentry, 0); | |
3157 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3158 | if (error) | |
3159 | goto out2; | |
3160 | file->f_path.mnt = nd->path.mnt; | |
3161 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3162 | if (error) | |
3163 | goto out2; | |
3164 | error = open_check_o_direct(file); | |
f4e0c30c | 3165 | if (error) { |
60545d0d | 3166 | fput(file); |
f4e0c30c AV |
3167 | } else if (!(op->open_flag & O_EXCL)) { |
3168 | struct inode *inode = file_inode(file); | |
3169 | spin_lock(&inode->i_lock); | |
3170 | inode->i_state |= I_LINKABLE; | |
3171 | spin_unlock(&inode->i_lock); | |
3172 | } | |
60545d0d AV |
3173 | out2: |
3174 | mnt_drop_write(nd->path.mnt); | |
3175 | out: | |
3176 | path_put(&nd->path); | |
3177 | return error; | |
3178 | } | |
3179 | ||
669abf4e | 3180 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3181 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3182 | { |
fe2d35ff | 3183 | struct file *base = NULL; |
30d90494 | 3184 | struct file *file; |
9850c056 | 3185 | struct path path; |
47237687 | 3186 | int opened = 0; |
13aab428 | 3187 | int error; |
31e6b01f | 3188 | |
30d90494 | 3189 | file = get_empty_filp(); |
1afc99be AV |
3190 | if (IS_ERR(file)) |
3191 | return file; | |
31e6b01f | 3192 | |
30d90494 | 3193 | file->f_flags = op->open_flag; |
31e6b01f | 3194 | |
bb458c64 | 3195 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3196 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3197 | goto out; | |
3198 | } | |
3199 | ||
669abf4e | 3200 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3201 | if (unlikely(error)) |
2675a4eb | 3202 | goto out; |
31e6b01f | 3203 | |
fe2d35ff | 3204 | current->total_link_count = 0; |
669abf4e | 3205 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3206 | if (unlikely(error)) |
2675a4eb | 3207 | goto out; |
1da177e4 | 3208 | |
2675a4eb AV |
3209 | error = do_last(nd, &path, file, op, &opened, pathname); |
3210 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3211 | struct path link = path; |
def4af30 | 3212 | void *cookie; |
574197e0 | 3213 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3214 | path_put_conditional(&path, nd); |
3215 | path_put(&nd->path); | |
2675a4eb | 3216 | error = -ELOOP; |
40b39136 AV |
3217 | break; |
3218 | } | |
800179c9 KC |
3219 | error = may_follow_link(&link, nd); |
3220 | if (unlikely(error)) | |
3221 | break; | |
73d049a4 AV |
3222 | nd->flags |= LOOKUP_PARENT; |
3223 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3224 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3225 | if (unlikely(error)) |
2675a4eb AV |
3226 | break; |
3227 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3228 | put_link(nd, &link, cookie); |
806b681c | 3229 | } |
10fa8e62 | 3230 | out: |
73d049a4 AV |
3231 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3232 | path_put(&nd->root); | |
fe2d35ff AV |
3233 | if (base) |
3234 | fput(base); | |
2675a4eb AV |
3235 | if (!(opened & FILE_OPENED)) { |
3236 | BUG_ON(!error); | |
30d90494 | 3237 | put_filp(file); |
16b1c1cd | 3238 | } |
2675a4eb AV |
3239 | if (unlikely(error)) { |
3240 | if (error == -EOPENSTALE) { | |
3241 | if (flags & LOOKUP_RCU) | |
3242 | error = -ECHILD; | |
3243 | else | |
3244 | error = -ESTALE; | |
3245 | } | |
3246 | file = ERR_PTR(error); | |
3247 | } | |
3248 | return file; | |
1da177e4 LT |
3249 | } |
3250 | ||
669abf4e | 3251 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3252 | const struct open_flags *op) |
13aab428 | 3253 | { |
73d049a4 | 3254 | struct nameidata nd; |
f9652e10 | 3255 | int flags = op->lookup_flags; |
13aab428 AV |
3256 | struct file *filp; |
3257 | ||
73d049a4 | 3258 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3259 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3260 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3261 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3262 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3263 | return filp; |
3264 | } | |
3265 | ||
73d049a4 | 3266 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3267 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3268 | { |
3269 | struct nameidata nd; | |
3270 | struct file *file; | |
669abf4e | 3271 | struct filename filename = { .name = name }; |
f9652e10 | 3272 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3273 | |
3274 | nd.root.mnt = mnt; | |
3275 | nd.root.dentry = dentry; | |
3276 | ||
bcda7652 | 3277 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3278 | return ERR_PTR(-ELOOP); |
3279 | ||
669abf4e | 3280 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3281 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3282 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3283 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3284 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3285 | return file; |
3286 | } | |
3287 | ||
1ac12b4b JL |
3288 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3289 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3290 | { |
c663e5d8 | 3291 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3292 | struct nameidata nd; |
c30dabfe | 3293 | int err2; |
1ac12b4b JL |
3294 | int error; |
3295 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3296 | ||
3297 | /* | |
3298 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3299 | * other flags passed in are ignored! | |
3300 | */ | |
3301 | lookup_flags &= LOOKUP_REVAL; | |
3302 | ||
3303 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3304 | if (error) |
3305 | return ERR_PTR(error); | |
1da177e4 | 3306 | |
c663e5d8 CH |
3307 | /* |
3308 | * Yucky last component or no last component at all? | |
3309 | * (foo/., foo/.., /////) | |
3310 | */ | |
ed75e95d AV |
3311 | if (nd.last_type != LAST_NORM) |
3312 | goto out; | |
3313 | nd.flags &= ~LOOKUP_PARENT; | |
3314 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3315 | |
c30dabfe JK |
3316 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3317 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3318 | /* |
3319 | * Do the final lookup. | |
3320 | */ | |
ed75e95d AV |
3321 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3322 | dentry = lookup_hash(&nd); | |
1da177e4 | 3323 | if (IS_ERR(dentry)) |
a8104a9f | 3324 | goto unlock; |
c663e5d8 | 3325 | |
a8104a9f | 3326 | error = -EEXIST; |
e9baf6e5 | 3327 | if (dentry->d_inode) |
a8104a9f | 3328 | goto fail; |
c663e5d8 CH |
3329 | /* |
3330 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3331 | * From the vfs_mknod() POV we just have a negative dentry - | |
3332 | * all is fine. Let's be bastards - you had / on the end, you've | |
3333 | * been asking for (non-existent) directory. -ENOENT for you. | |
3334 | */ | |
ed75e95d | 3335 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3336 | error = -ENOENT; |
ed75e95d | 3337 | goto fail; |
e9baf6e5 | 3338 | } |
c30dabfe JK |
3339 | if (unlikely(err2)) { |
3340 | error = err2; | |
a8104a9f | 3341 | goto fail; |
c30dabfe | 3342 | } |
ed75e95d | 3343 | *path = nd.path; |
1da177e4 | 3344 | return dentry; |
1da177e4 | 3345 | fail: |
a8104a9f AV |
3346 | dput(dentry); |
3347 | dentry = ERR_PTR(error); | |
3348 | unlock: | |
ed75e95d | 3349 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3350 | if (!err2) |
3351 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3352 | out: |
3353 | path_put(&nd.path); | |
1da177e4 LT |
3354 | return dentry; |
3355 | } | |
dae6ad8f AV |
3356 | EXPORT_SYMBOL(kern_path_create); |
3357 | ||
921a1650 AV |
3358 | void done_path_create(struct path *path, struct dentry *dentry) |
3359 | { | |
3360 | dput(dentry); | |
3361 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3362 | mnt_drop_write(path->mnt); |
921a1650 AV |
3363 | path_put(path); |
3364 | } | |
3365 | EXPORT_SYMBOL(done_path_create); | |
3366 | ||
1ac12b4b JL |
3367 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3368 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3369 | { |
91a27b2a | 3370 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3371 | struct dentry *res; |
3372 | if (IS_ERR(tmp)) | |
3373 | return ERR_CAST(tmp); | |
1ac12b4b | 3374 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3375 | putname(tmp); |
3376 | return res; | |
3377 | } | |
3378 | EXPORT_SYMBOL(user_path_create); | |
3379 | ||
1a67aafb | 3380 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3381 | { |
a95164d9 | 3382 | int error = may_create(dir, dentry); |
1da177e4 LT |
3383 | |
3384 | if (error) | |
3385 | return error; | |
3386 | ||
975d6b39 | 3387 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3388 | return -EPERM; |
3389 | ||
acfa4380 | 3390 | if (!dir->i_op->mknod) |
1da177e4 LT |
3391 | return -EPERM; |
3392 | ||
08ce5f16 SH |
3393 | error = devcgroup_inode_mknod(mode, dev); |
3394 | if (error) | |
3395 | return error; | |
3396 | ||
1da177e4 LT |
3397 | error = security_inode_mknod(dir, dentry, mode, dev); |
3398 | if (error) | |
3399 | return error; | |
3400 | ||
1da177e4 | 3401 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3402 | if (!error) |
f38aa942 | 3403 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3404 | return error; |
3405 | } | |
3406 | ||
f69aac00 | 3407 | static int may_mknod(umode_t mode) |
463c3197 DH |
3408 | { |
3409 | switch (mode & S_IFMT) { | |
3410 | case S_IFREG: | |
3411 | case S_IFCHR: | |
3412 | case S_IFBLK: | |
3413 | case S_IFIFO: | |
3414 | case S_IFSOCK: | |
3415 | case 0: /* zero mode translates to S_IFREG */ | |
3416 | return 0; | |
3417 | case S_IFDIR: | |
3418 | return -EPERM; | |
3419 | default: | |
3420 | return -EINVAL; | |
3421 | } | |
3422 | } | |
3423 | ||
8208a22b | 3424 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3425 | unsigned, dev) |
1da177e4 | 3426 | { |
2ad94ae6 | 3427 | struct dentry *dentry; |
dae6ad8f AV |
3428 | struct path path; |
3429 | int error; | |
972567f1 | 3430 | unsigned int lookup_flags = 0; |
1da177e4 | 3431 | |
8e4bfca1 AV |
3432 | error = may_mknod(mode); |
3433 | if (error) | |
3434 | return error; | |
972567f1 JL |
3435 | retry: |
3436 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3437 | if (IS_ERR(dentry)) |
3438 | return PTR_ERR(dentry); | |
2ad94ae6 | 3439 | |
dae6ad8f | 3440 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3441 | mode &= ~current_umask(); |
dae6ad8f | 3442 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3443 | if (error) |
a8104a9f | 3444 | goto out; |
463c3197 | 3445 | switch (mode & S_IFMT) { |
1da177e4 | 3446 | case 0: case S_IFREG: |
312b63fb | 3447 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3448 | break; |
3449 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3450 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3451 | new_decode_dev(dev)); |
3452 | break; | |
3453 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3454 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3455 | break; |
1da177e4 | 3456 | } |
a8104a9f | 3457 | out: |
921a1650 | 3458 | done_path_create(&path, dentry); |
972567f1 JL |
3459 | if (retry_estale(error, lookup_flags)) { |
3460 | lookup_flags |= LOOKUP_REVAL; | |
3461 | goto retry; | |
3462 | } | |
1da177e4 LT |
3463 | return error; |
3464 | } | |
3465 | ||
8208a22b | 3466 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3467 | { |
3468 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3469 | } | |
3470 | ||
18bb1db3 | 3471 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3472 | { |
a95164d9 | 3473 | int error = may_create(dir, dentry); |
8de52778 | 3474 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3475 | |
3476 | if (error) | |
3477 | return error; | |
3478 | ||
acfa4380 | 3479 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3480 | return -EPERM; |
3481 | ||
3482 | mode &= (S_IRWXUGO|S_ISVTX); | |
3483 | error = security_inode_mkdir(dir, dentry, mode); | |
3484 | if (error) | |
3485 | return error; | |
3486 | ||
8de52778 AV |
3487 | if (max_links && dir->i_nlink >= max_links) |
3488 | return -EMLINK; | |
3489 | ||
1da177e4 | 3490 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3491 | if (!error) |
f38aa942 | 3492 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3493 | return error; |
3494 | } | |
3495 | ||
a218d0fd | 3496 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3497 | { |
6902d925 | 3498 | struct dentry *dentry; |
dae6ad8f AV |
3499 | struct path path; |
3500 | int error; | |
b76d8b82 | 3501 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3502 | |
b76d8b82 JL |
3503 | retry: |
3504 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3505 | if (IS_ERR(dentry)) |
dae6ad8f | 3506 | return PTR_ERR(dentry); |
1da177e4 | 3507 | |
dae6ad8f | 3508 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3509 | mode &= ~current_umask(); |
dae6ad8f | 3510 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3511 | if (!error) |
3512 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3513 | done_path_create(&path, dentry); |
b76d8b82 JL |
3514 | if (retry_estale(error, lookup_flags)) { |
3515 | lookup_flags |= LOOKUP_REVAL; | |
3516 | goto retry; | |
3517 | } | |
1da177e4 LT |
3518 | return error; |
3519 | } | |
3520 | ||
a218d0fd | 3521 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3522 | { |
3523 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3524 | } | |
3525 | ||
1da177e4 | 3526 | /* |
a71905f0 | 3527 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3528 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3529 | * dentry, and if that is true (possibly after pruning the dcache), |
3530 | * then we drop the dentry now. | |
1da177e4 LT |
3531 | * |
3532 | * A low-level filesystem can, if it choses, legally | |
3533 | * do a | |
3534 | * | |
3535 | * if (!d_unhashed(dentry)) | |
3536 | * return -EBUSY; | |
3537 | * | |
3538 | * if it cannot handle the case of removing a directory | |
3539 | * that is still in use by something else.. | |
3540 | */ | |
3541 | void dentry_unhash(struct dentry *dentry) | |
3542 | { | |
dc168427 | 3543 | shrink_dcache_parent(dentry); |
1da177e4 | 3544 | spin_lock(&dentry->d_lock); |
98474236 | 3545 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3546 | __d_drop(dentry); |
3547 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3548 | } |
3549 | ||
3550 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3551 | { | |
3552 | int error = may_delete(dir, dentry, 1); | |
3553 | ||
3554 | if (error) | |
3555 | return error; | |
3556 | ||
acfa4380 | 3557 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3558 | return -EPERM; |
3559 | ||
1d2ef590 | 3560 | dget(dentry); |
1b1dcc1b | 3561 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3562 | |
3563 | error = -EBUSY; | |
1da177e4 | 3564 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3565 | goto out; |
3566 | ||
3567 | error = security_inode_rmdir(dir, dentry); | |
3568 | if (error) | |
3569 | goto out; | |
3570 | ||
3cebde24 | 3571 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3572 | error = dir->i_op->rmdir(dir, dentry); |
3573 | if (error) | |
3574 | goto out; | |
3575 | ||
3576 | dentry->d_inode->i_flags |= S_DEAD; | |
3577 | dont_mount(dentry); | |
3578 | ||
3579 | out: | |
1b1dcc1b | 3580 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3581 | dput(dentry); |
912dbc15 | 3582 | if (!error) |
1da177e4 | 3583 | d_delete(dentry); |
1da177e4 LT |
3584 | return error; |
3585 | } | |
3586 | ||
5590ff0d | 3587 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3588 | { |
3589 | int error = 0; | |
91a27b2a | 3590 | struct filename *name; |
1da177e4 LT |
3591 | struct dentry *dentry; |
3592 | struct nameidata nd; | |
c6ee9206 JL |
3593 | unsigned int lookup_flags = 0; |
3594 | retry: | |
3595 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3596 | if (IS_ERR(name)) |
3597 | return PTR_ERR(name); | |
1da177e4 LT |
3598 | |
3599 | switch(nd.last_type) { | |
0612d9fb OH |
3600 | case LAST_DOTDOT: |
3601 | error = -ENOTEMPTY; | |
3602 | goto exit1; | |
3603 | case LAST_DOT: | |
3604 | error = -EINVAL; | |
3605 | goto exit1; | |
3606 | case LAST_ROOT: | |
3607 | error = -EBUSY; | |
3608 | goto exit1; | |
1da177e4 | 3609 | } |
0612d9fb OH |
3610 | |
3611 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3612 | error = mnt_want_write(nd.path.mnt); |
3613 | if (error) | |
3614 | goto exit1; | |
0612d9fb | 3615 | |
4ac91378 | 3616 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3617 | dentry = lookup_hash(&nd); |
1da177e4 | 3618 | error = PTR_ERR(dentry); |
6902d925 DH |
3619 | if (IS_ERR(dentry)) |
3620 | goto exit2; | |
e6bc45d6 TT |
3621 | if (!dentry->d_inode) { |
3622 | error = -ENOENT; | |
3623 | goto exit3; | |
3624 | } | |
be6d3e56 KT |
3625 | error = security_path_rmdir(&nd.path, dentry); |
3626 | if (error) | |
c30dabfe | 3627 | goto exit3; |
4ac91378 | 3628 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3629 | exit3: |
6902d925 DH |
3630 | dput(dentry); |
3631 | exit2: | |
4ac91378 | 3632 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3633 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3634 | exit1: |
1d957f9b | 3635 | path_put(&nd.path); |
1da177e4 | 3636 | putname(name); |
c6ee9206 JL |
3637 | if (retry_estale(error, lookup_flags)) { |
3638 | lookup_flags |= LOOKUP_REVAL; | |
3639 | goto retry; | |
3640 | } | |
1da177e4 LT |
3641 | return error; |
3642 | } | |
3643 | ||
3cdad428 | 3644 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3645 | { |
3646 | return do_rmdir(AT_FDCWD, pathname); | |
3647 | } | |
3648 | ||
1da177e4 LT |
3649 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
3650 | { | |
3651 | int error = may_delete(dir, dentry, 0); | |
3652 | ||
3653 | if (error) | |
3654 | return error; | |
3655 | ||
acfa4380 | 3656 | if (!dir->i_op->unlink) |
1da177e4 LT |
3657 | return -EPERM; |
3658 | ||
1b1dcc1b | 3659 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3660 | if (d_mountpoint(dentry)) |
3661 | error = -EBUSY; | |
3662 | else { | |
3663 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3664 | if (!error) { |
1da177e4 | 3665 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3666 | if (!error) |
d83c49f3 | 3667 | dont_mount(dentry); |
bec1052e | 3668 | } |
1da177e4 | 3669 | } |
1b1dcc1b | 3670 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3671 | |
3672 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3673 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 3674 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 3675 | d_delete(dentry); |
1da177e4 | 3676 | } |
0eeca283 | 3677 | |
1da177e4 LT |
3678 | return error; |
3679 | } | |
3680 | ||
3681 | /* | |
3682 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3683 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3684 | * writeout happening, and we don't want to prevent access to the directory |
3685 | * while waiting on the I/O. | |
3686 | */ | |
5590ff0d | 3687 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3688 | { |
2ad94ae6 | 3689 | int error; |
91a27b2a | 3690 | struct filename *name; |
1da177e4 LT |
3691 | struct dentry *dentry; |
3692 | struct nameidata nd; | |
3693 | struct inode *inode = NULL; | |
5d18f813 JL |
3694 | unsigned int lookup_flags = 0; |
3695 | retry: | |
3696 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3697 | if (IS_ERR(name)) |
3698 | return PTR_ERR(name); | |
2ad94ae6 | 3699 | |
1da177e4 LT |
3700 | error = -EISDIR; |
3701 | if (nd.last_type != LAST_NORM) | |
3702 | goto exit1; | |
0612d9fb OH |
3703 | |
3704 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3705 | error = mnt_want_write(nd.path.mnt); |
3706 | if (error) | |
3707 | goto exit1; | |
0612d9fb | 3708 | |
4ac91378 | 3709 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3710 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3711 | error = PTR_ERR(dentry); |
3712 | if (!IS_ERR(dentry)) { | |
3713 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3714 | if (nd.last.name[nd.last.len]) |
3715 | goto slashes; | |
1da177e4 | 3716 | inode = dentry->d_inode; |
50338b88 | 3717 | if (!inode) |
e6bc45d6 TT |
3718 | goto slashes; |
3719 | ihold(inode); | |
be6d3e56 KT |
3720 | error = security_path_unlink(&nd.path, dentry); |
3721 | if (error) | |
c30dabfe | 3722 | goto exit2; |
4ac91378 | 3723 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
c30dabfe | 3724 | exit2: |
1da177e4 LT |
3725 | dput(dentry); |
3726 | } | |
4ac91378 | 3727 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3728 | if (inode) |
3729 | iput(inode); /* truncate the inode here */ | |
c30dabfe | 3730 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3731 | exit1: |
1d957f9b | 3732 | path_put(&nd.path); |
1da177e4 | 3733 | putname(name); |
5d18f813 JL |
3734 | if (retry_estale(error, lookup_flags)) { |
3735 | lookup_flags |= LOOKUP_REVAL; | |
3736 | inode = NULL; | |
3737 | goto retry; | |
3738 | } | |
1da177e4 LT |
3739 | return error; |
3740 | ||
3741 | slashes: | |
3742 | error = !dentry->d_inode ? -ENOENT : | |
3743 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
3744 | goto exit2; | |
3745 | } | |
3746 | ||
2e4d0924 | 3747 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3748 | { |
3749 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3750 | return -EINVAL; | |
3751 | ||
3752 | if (flag & AT_REMOVEDIR) | |
3753 | return do_rmdir(dfd, pathname); | |
3754 | ||
3755 | return do_unlinkat(dfd, pathname); | |
3756 | } | |
3757 | ||
3480b257 | 3758 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3759 | { |
3760 | return do_unlinkat(AT_FDCWD, pathname); | |
3761 | } | |
3762 | ||
db2e747b | 3763 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3764 | { |
a95164d9 | 3765 | int error = may_create(dir, dentry); |
1da177e4 LT |
3766 | |
3767 | if (error) | |
3768 | return error; | |
3769 | ||
acfa4380 | 3770 | if (!dir->i_op->symlink) |
1da177e4 LT |
3771 | return -EPERM; |
3772 | ||
3773 | error = security_inode_symlink(dir, dentry, oldname); | |
3774 | if (error) | |
3775 | return error; | |
3776 | ||
1da177e4 | 3777 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3778 | if (!error) |
f38aa942 | 3779 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3780 | return error; |
3781 | } | |
3782 | ||
2e4d0924 HC |
3783 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3784 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3785 | { |
2ad94ae6 | 3786 | int error; |
91a27b2a | 3787 | struct filename *from; |
6902d925 | 3788 | struct dentry *dentry; |
dae6ad8f | 3789 | struct path path; |
f46d3567 | 3790 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3791 | |
3792 | from = getname(oldname); | |
2ad94ae6 | 3793 | if (IS_ERR(from)) |
1da177e4 | 3794 | return PTR_ERR(from); |
f46d3567 JL |
3795 | retry: |
3796 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3797 | error = PTR_ERR(dentry); |
3798 | if (IS_ERR(dentry)) | |
dae6ad8f | 3799 | goto out_putname; |
6902d925 | 3800 | |
91a27b2a | 3801 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3802 | if (!error) |
91a27b2a | 3803 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3804 | done_path_create(&path, dentry); |
f46d3567 JL |
3805 | if (retry_estale(error, lookup_flags)) { |
3806 | lookup_flags |= LOOKUP_REVAL; | |
3807 | goto retry; | |
3808 | } | |
6902d925 | 3809 | out_putname: |
1da177e4 LT |
3810 | putname(from); |
3811 | return error; | |
3812 | } | |
3813 | ||
3480b257 | 3814 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3815 | { |
3816 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3817 | } | |
3818 | ||
1da177e4 LT |
3819 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3820 | { | |
3821 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3822 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3823 | int error; |
3824 | ||
3825 | if (!inode) | |
3826 | return -ENOENT; | |
3827 | ||
a95164d9 | 3828 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3829 | if (error) |
3830 | return error; | |
3831 | ||
3832 | if (dir->i_sb != inode->i_sb) | |
3833 | return -EXDEV; | |
3834 | ||
3835 | /* | |
3836 | * A link to an append-only or immutable file cannot be created. | |
3837 | */ | |
3838 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3839 | return -EPERM; | |
acfa4380 | 3840 | if (!dir->i_op->link) |
1da177e4 | 3841 | return -EPERM; |
7e79eedb | 3842 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3843 | return -EPERM; |
3844 | ||
3845 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3846 | if (error) | |
3847 | return error; | |
3848 | ||
7e79eedb | 3849 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3850 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3851 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3852 | error = -ENOENT; |
8de52778 AV |
3853 | else if (max_links && inode->i_nlink >= max_links) |
3854 | error = -EMLINK; | |
aae8a97d AK |
3855 | else |
3856 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
f4e0c30c AV |
3857 | |
3858 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3859 | spin_lock(&inode->i_lock); | |
3860 | inode->i_state &= ~I_LINKABLE; | |
3861 | spin_unlock(&inode->i_lock); | |
3862 | } | |
7e79eedb | 3863 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3864 | if (!error) |
7e79eedb | 3865 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3866 | return error; |
3867 | } | |
3868 | ||
3869 | /* | |
3870 | * Hardlinks are often used in delicate situations. We avoid | |
3871 | * security-related surprises by not following symlinks on the | |
3872 | * newname. --KAB | |
3873 | * | |
3874 | * We don't follow them on the oldname either to be compatible | |
3875 | * with linux 2.0, and to avoid hard-linking to directories | |
3876 | * and other special files. --ADM | |
3877 | */ | |
2e4d0924 HC |
3878 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3879 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3880 | { |
3881 | struct dentry *new_dentry; | |
dae6ad8f | 3882 | struct path old_path, new_path; |
11a7b371 | 3883 | int how = 0; |
1da177e4 | 3884 | int error; |
1da177e4 | 3885 | |
11a7b371 | 3886 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3887 | return -EINVAL; |
11a7b371 | 3888 | /* |
f0cc6ffb LT |
3889 | * To use null names we require CAP_DAC_READ_SEARCH |
3890 | * This ensures that not everyone will be able to create | |
3891 | * handlink using the passed filedescriptor. | |
11a7b371 | 3892 | */ |
f0cc6ffb LT |
3893 | if (flags & AT_EMPTY_PATH) { |
3894 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3895 | return -ENOENT; | |
11a7b371 | 3896 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3897 | } |
11a7b371 AK |
3898 | |
3899 | if (flags & AT_SYMLINK_FOLLOW) | |
3900 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3901 | retry: |
11a7b371 | 3902 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3903 | if (error) |
2ad94ae6 AV |
3904 | return error; |
3905 | ||
442e31ca JL |
3906 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3907 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3908 | error = PTR_ERR(new_dentry); |
6902d925 | 3909 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3910 | goto out; |
3911 | ||
3912 | error = -EXDEV; | |
3913 | if (old_path.mnt != new_path.mnt) | |
3914 | goto out_dput; | |
800179c9 KC |
3915 | error = may_linkat(&old_path); |
3916 | if (unlikely(error)) | |
3917 | goto out_dput; | |
dae6ad8f | 3918 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3919 | if (error) |
a8104a9f | 3920 | goto out_dput; |
dae6ad8f | 3921 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
75c3f29d | 3922 | out_dput: |
921a1650 | 3923 | done_path_create(&new_path, new_dentry); |
442e31ca JL |
3924 | if (retry_estale(error, how)) { |
3925 | how |= LOOKUP_REVAL; | |
3926 | goto retry; | |
3927 | } | |
1da177e4 | 3928 | out: |
2d8f3038 | 3929 | path_put(&old_path); |
1da177e4 LT |
3930 | |
3931 | return error; | |
3932 | } | |
3933 | ||
3480b257 | 3934 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3935 | { |
c04030e1 | 3936 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3937 | } |
3938 | ||
1da177e4 LT |
3939 | /* |
3940 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3941 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3942 | * Problems: | |
3943 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3944 | * b) race potential - two innocent renames can create a loop together. | |
3945 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3946 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3947 | * story. |
3948 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3949 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3950 | * whether the target exists). Solution: try to be smart with locking |
3951 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3952 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3953 | * move will be locked. Thus we can rank directories by the tree |
3954 | * (ancestors first) and rank all non-directories after them. | |
3955 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3956 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3957 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3958 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3959 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3960 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3961 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3962 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3963 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3964 | * locking]. |
3965 | */ | |
75c96f85 AB |
3966 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3967 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3968 | { |
3969 | int error = 0; | |
9055cba7 | 3970 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3971 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3972 | |
3973 | /* | |
3974 | * If we are going to change the parent - check write permissions, | |
3975 | * we'll need to flip '..'. | |
3976 | */ | |
3977 | if (new_dir != old_dir) { | |
f419a2e3 | 3978 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3979 | if (error) |
3980 | return error; | |
3981 | } | |
3982 | ||
3983 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3984 | if (error) | |
3985 | return error; | |
3986 | ||
1d2ef590 | 3987 | dget(new_dentry); |
d83c49f3 | 3988 | if (target) |
1b1dcc1b | 3989 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3990 | |
3991 | error = -EBUSY; | |
3992 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3993 | goto out; | |
3994 | ||
8de52778 AV |
3995 | error = -EMLINK; |
3996 | if (max_links && !target && new_dir != old_dir && | |
3997 | new_dir->i_nlink >= max_links) | |
3998 | goto out; | |
3999 | ||
3cebde24 SW |
4000 | if (target) |
4001 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
4002 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4003 | if (error) | |
4004 | goto out; | |
4005 | ||
1da177e4 | 4006 | if (target) { |
9055cba7 SW |
4007 | target->i_flags |= S_DEAD; |
4008 | dont_mount(new_dentry); | |
1da177e4 | 4009 | } |
9055cba7 SW |
4010 | out: |
4011 | if (target) | |
4012 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 4013 | dput(new_dentry); |
e31e14ec | 4014 | if (!error) |
349457cc MF |
4015 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
4016 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
4017 | return error; |
4018 | } | |
4019 | ||
75c96f85 AB |
4020 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
4021 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 4022 | { |
51892bbb | 4023 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
4024 | int error; |
4025 | ||
4026 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4027 | if (error) | |
4028 | return error; | |
4029 | ||
4030 | dget(new_dentry); | |
1da177e4 | 4031 | if (target) |
1b1dcc1b | 4032 | mutex_lock(&target->i_mutex); |
51892bbb SW |
4033 | |
4034 | error = -EBUSY; | |
1da177e4 | 4035 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
4036 | goto out; |
4037 | ||
4038 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
4039 | if (error) | |
4040 | goto out; | |
4041 | ||
4042 | if (target) | |
4043 | dont_mount(new_dentry); | |
4044 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
4045 | d_move(old_dentry, new_dentry); | |
4046 | out: | |
1da177e4 | 4047 | if (target) |
1b1dcc1b | 4048 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
4049 | dput(new_dentry); |
4050 | return error; | |
4051 | } | |
4052 | ||
4053 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
4054 | struct inode *new_dir, struct dentry *new_dentry) | |
4055 | { | |
4056 | int error; | |
4057 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 4058 | const unsigned char *old_name; |
1da177e4 LT |
4059 | |
4060 | if (old_dentry->d_inode == new_dentry->d_inode) | |
4061 | return 0; | |
4062 | ||
4063 | error = may_delete(old_dir, old_dentry, is_dir); | |
4064 | if (error) | |
4065 | return error; | |
4066 | ||
4067 | if (!new_dentry->d_inode) | |
a95164d9 | 4068 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
4069 | else |
4070 | error = may_delete(new_dir, new_dentry, is_dir); | |
4071 | if (error) | |
4072 | return error; | |
4073 | ||
acfa4380 | 4074 | if (!old_dir->i_op->rename) |
1da177e4 LT |
4075 | return -EPERM; |
4076 | ||
0eeca283 RL |
4077 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
4078 | ||
1da177e4 LT |
4079 | if (is_dir) |
4080 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
4081 | else | |
4082 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
4083 | if (!error) |
4084 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 4085 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
4086 | fsnotify_oldname_free(old_name); |
4087 | ||
1da177e4 LT |
4088 | return error; |
4089 | } | |
4090 | ||
2e4d0924 HC |
4091 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4092 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4093 | { |
2ad94ae6 AV |
4094 | struct dentry *old_dir, *new_dir; |
4095 | struct dentry *old_dentry, *new_dentry; | |
4096 | struct dentry *trap; | |
1da177e4 | 4097 | struct nameidata oldnd, newnd; |
91a27b2a JL |
4098 | struct filename *from; |
4099 | struct filename *to; | |
c6a94284 JL |
4100 | unsigned int lookup_flags = 0; |
4101 | bool should_retry = false; | |
2ad94ae6 | 4102 | int error; |
c6a94284 JL |
4103 | retry: |
4104 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4105 | if (IS_ERR(from)) { |
4106 | error = PTR_ERR(from); | |
1da177e4 | 4107 | goto exit; |
91a27b2a | 4108 | } |
1da177e4 | 4109 | |
c6a94284 | 4110 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4111 | if (IS_ERR(to)) { |
4112 | error = PTR_ERR(to); | |
1da177e4 | 4113 | goto exit1; |
91a27b2a | 4114 | } |
1da177e4 LT |
4115 | |
4116 | error = -EXDEV; | |
4ac91378 | 4117 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4118 | goto exit2; |
4119 | ||
4ac91378 | 4120 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4121 | error = -EBUSY; |
4122 | if (oldnd.last_type != LAST_NORM) | |
4123 | goto exit2; | |
4124 | ||
4ac91378 | 4125 | new_dir = newnd.path.dentry; |
1da177e4 LT |
4126 | if (newnd.last_type != LAST_NORM) |
4127 | goto exit2; | |
4128 | ||
c30dabfe JK |
4129 | error = mnt_want_write(oldnd.path.mnt); |
4130 | if (error) | |
4131 | goto exit2; | |
4132 | ||
0612d9fb OH |
4133 | oldnd.flags &= ~LOOKUP_PARENT; |
4134 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 4135 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 4136 | |
1da177e4 LT |
4137 | trap = lock_rename(new_dir, old_dir); |
4138 | ||
49705b77 | 4139 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4140 | error = PTR_ERR(old_dentry); |
4141 | if (IS_ERR(old_dentry)) | |
4142 | goto exit3; | |
4143 | /* source must exist */ | |
4144 | error = -ENOENT; | |
4145 | if (!old_dentry->d_inode) | |
4146 | goto exit4; | |
4147 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
4148 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
4149 | error = -ENOTDIR; | |
4150 | if (oldnd.last.name[oldnd.last.len]) | |
4151 | goto exit4; | |
4152 | if (newnd.last.name[newnd.last.len]) | |
4153 | goto exit4; | |
4154 | } | |
4155 | /* source should not be ancestor of target */ | |
4156 | error = -EINVAL; | |
4157 | if (old_dentry == trap) | |
4158 | goto exit4; | |
49705b77 | 4159 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
4160 | error = PTR_ERR(new_dentry); |
4161 | if (IS_ERR(new_dentry)) | |
4162 | goto exit4; | |
4163 | /* target should not be an ancestor of source */ | |
4164 | error = -ENOTEMPTY; | |
4165 | if (new_dentry == trap) | |
4166 | goto exit5; | |
4167 | ||
be6d3e56 KT |
4168 | error = security_path_rename(&oldnd.path, old_dentry, |
4169 | &newnd.path, new_dentry); | |
4170 | if (error) | |
c30dabfe | 4171 | goto exit5; |
1da177e4 LT |
4172 | error = vfs_rename(old_dir->d_inode, old_dentry, |
4173 | new_dir->d_inode, new_dentry); | |
4174 | exit5: | |
4175 | dput(new_dentry); | |
4176 | exit4: | |
4177 | dput(old_dentry); | |
4178 | exit3: | |
4179 | unlock_rename(new_dir, old_dir); | |
c30dabfe | 4180 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4181 | exit2: |
c6a94284 JL |
4182 | if (retry_estale(error, lookup_flags)) |
4183 | should_retry = true; | |
1d957f9b | 4184 | path_put(&newnd.path); |
2ad94ae6 | 4185 | putname(to); |
1da177e4 | 4186 | exit1: |
1d957f9b | 4187 | path_put(&oldnd.path); |
1da177e4 | 4188 | putname(from); |
c6a94284 JL |
4189 | if (should_retry) { |
4190 | should_retry = false; | |
4191 | lookup_flags |= LOOKUP_REVAL; | |
4192 | goto retry; | |
4193 | } | |
2ad94ae6 | 4194 | exit: |
1da177e4 LT |
4195 | return error; |
4196 | } | |
4197 | ||
a26eab24 | 4198 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4199 | { |
4200 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
4201 | } | |
4202 | ||
1da177e4 LT |
4203 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
4204 | { | |
4205 | int len; | |
4206 | ||
4207 | len = PTR_ERR(link); | |
4208 | if (IS_ERR(link)) | |
4209 | goto out; | |
4210 | ||
4211 | len = strlen(link); | |
4212 | if (len > (unsigned) buflen) | |
4213 | len = buflen; | |
4214 | if (copy_to_user(buffer, link, len)) | |
4215 | len = -EFAULT; | |
4216 | out: | |
4217 | return len; | |
4218 | } | |
4219 | ||
4220 | /* | |
4221 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4222 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4223 | * using) it for any given inode is up to filesystem. | |
4224 | */ | |
4225 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4226 | { | |
4227 | struct nameidata nd; | |
cc314eef | 4228 | void *cookie; |
694a1764 | 4229 | int res; |
cc314eef | 4230 | |
1da177e4 | 4231 | nd.depth = 0; |
cc314eef | 4232 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4233 | if (IS_ERR(cookie)) |
4234 | return PTR_ERR(cookie); | |
4235 | ||
4236 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
4237 | if (dentry->d_inode->i_op->put_link) | |
4238 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4239 | return res; | |
1da177e4 LT |
4240 | } |
4241 | ||
1da177e4 LT |
4242 | /* get the link contents into pagecache */ |
4243 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4244 | { | |
ebd09abb DG |
4245 | char *kaddr; |
4246 | struct page *page; | |
1da177e4 | 4247 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4248 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4249 | if (IS_ERR(page)) |
6fe6900e | 4250 | return (char*)page; |
1da177e4 | 4251 | *ppage = page; |
ebd09abb DG |
4252 | kaddr = kmap(page); |
4253 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4254 | return kaddr; | |
1da177e4 LT |
4255 | } |
4256 | ||
4257 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4258 | { | |
4259 | struct page *page = NULL; | |
4260 | char *s = page_getlink(dentry, &page); | |
4261 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4262 | if (page) { | |
4263 | kunmap(page); | |
4264 | page_cache_release(page); | |
4265 | } | |
4266 | return res; | |
4267 | } | |
4268 | ||
cc314eef | 4269 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4270 | { |
cc314eef | 4271 | struct page *page = NULL; |
1da177e4 | 4272 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4273 | return page; |
1da177e4 LT |
4274 | } |
4275 | ||
cc314eef | 4276 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4277 | { |
cc314eef LT |
4278 | struct page *page = cookie; |
4279 | ||
4280 | if (page) { | |
1da177e4 LT |
4281 | kunmap(page); |
4282 | page_cache_release(page); | |
1da177e4 LT |
4283 | } |
4284 | } | |
4285 | ||
54566b2c NP |
4286 | /* |
4287 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4288 | */ | |
4289 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4290 | { |
4291 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4292 | struct page *page; |
afddba49 | 4293 | void *fsdata; |
beb497ab | 4294 | int err; |
1da177e4 | 4295 | char *kaddr; |
54566b2c NP |
4296 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4297 | if (nofs) | |
4298 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4299 | |
7e53cac4 | 4300 | retry: |
afddba49 | 4301 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4302 | flags, &page, &fsdata); |
1da177e4 | 4303 | if (err) |
afddba49 NP |
4304 | goto fail; |
4305 | ||
e8e3c3d6 | 4306 | kaddr = kmap_atomic(page); |
1da177e4 | 4307 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4308 | kunmap_atomic(kaddr); |
afddba49 NP |
4309 | |
4310 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4311 | page, fsdata); | |
1da177e4 LT |
4312 | if (err < 0) |
4313 | goto fail; | |
afddba49 NP |
4314 | if (err < len-1) |
4315 | goto retry; | |
4316 | ||
1da177e4 LT |
4317 | mark_inode_dirty(inode); |
4318 | return 0; | |
1da177e4 LT |
4319 | fail: |
4320 | return err; | |
4321 | } | |
4322 | ||
0adb25d2 KK |
4323 | int page_symlink(struct inode *inode, const char *symname, int len) |
4324 | { | |
4325 | return __page_symlink(inode, symname, len, | |
54566b2c | 4326 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4327 | } |
4328 | ||
92e1d5be | 4329 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4330 | .readlink = generic_readlink, |
4331 | .follow_link = page_follow_link_light, | |
4332 | .put_link = page_put_link, | |
4333 | }; | |
4334 | ||
2d8f3038 | 4335 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4336 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4337 | EXPORT_SYMBOL(follow_down); |
4338 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4339 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4340 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4341 | EXPORT_SYMBOL(lookup_one_len); |
4342 | EXPORT_SYMBOL(page_follow_link_light); | |
4343 | EXPORT_SYMBOL(page_put_link); | |
4344 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4345 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4346 | EXPORT_SYMBOL(page_symlink); |
4347 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4348 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4349 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4350 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4351 | EXPORT_SYMBOL(unlock_rename); |
4352 | EXPORT_SYMBOL(vfs_create); | |
1da177e4 LT |
4353 | EXPORT_SYMBOL(vfs_link); |
4354 | EXPORT_SYMBOL(vfs_mkdir); | |
4355 | EXPORT_SYMBOL(vfs_mknod); | |
4356 | EXPORT_SYMBOL(generic_permission); | |
4357 | EXPORT_SYMBOL(vfs_readlink); | |
4358 | EXPORT_SYMBOL(vfs_rename); | |
4359 | EXPORT_SYMBOL(vfs_rmdir); | |
4360 | EXPORT_SYMBOL(vfs_symlink); | |
4361 | EXPORT_SYMBOL(vfs_unlink); | |
4362 | EXPORT_SYMBOL(dentry_unhash); | |
4363 | EXPORT_SYMBOL(generic_readlink); |