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 NP |
484 | |
485 | /** | |
19660af7 AV |
486 | * unlazy_walk - try to switch to ref-walk mode. |
487 | * @nd: nameidata pathwalk data | |
488 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 489 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 490 | * |
19660af7 AV |
491 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
492 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
493 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 494 | */ |
19660af7 | 495 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
496 | { |
497 | struct fs_struct *fs = current->fs; | |
498 | struct dentry *parent = nd->path.dentry; | |
499 | ||
500 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
501 | |
502 | /* | |
48a066e7 AV |
503 | * After legitimizing the bastards, terminate_walk() |
504 | * will do the right thing for non-RCU mode, and all our | |
505 | * subsequent exit cases should rcu_read_unlock() | |
506 | * before returning. Do vfsmount first; if dentry | |
507 | * can't be legitimized, just set nd->path.dentry to NULL | |
508 | * and rely on dput(NULL) being a no-op. | |
e5c832d5 | 509 | */ |
48a066e7 | 510 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) |
e5c832d5 | 511 | return -ECHILD; |
e5c832d5 | 512 | nd->flags &= ~LOOKUP_RCU; |
15570086 | 513 | |
48a066e7 AV |
514 | if (!lockref_get_not_dead(&parent->d_lockref)) { |
515 | nd->path.dentry = NULL; | |
8b61e74f | 516 | rcu_read_unlock(); |
48a066e7 AV |
517 | return -ECHILD; |
518 | } | |
519 | ||
15570086 LT |
520 | /* |
521 | * For a negative lookup, the lookup sequence point is the parents | |
522 | * sequence point, and it only needs to revalidate the parent dentry. | |
523 | * | |
524 | * For a positive lookup, we need to move both the parent and the | |
525 | * dentry from the RCU domain to be properly refcounted. And the | |
526 | * sequence number in the dentry validates *both* dentry counters, | |
527 | * since we checked the sequence number of the parent after we got | |
528 | * the child sequence number. So we know the parent must still | |
529 | * be valid if the child sequence number is still valid. | |
530 | */ | |
19660af7 | 531 | if (!dentry) { |
e5c832d5 LT |
532 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
533 | goto out; | |
19660af7 AV |
534 | BUG_ON(nd->inode != parent->d_inode); |
535 | } else { | |
e5c832d5 LT |
536 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
537 | goto out; | |
538 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
539 | goto drop_dentry; | |
19660af7 | 540 | } |
e5c832d5 LT |
541 | |
542 | /* | |
543 | * Sequence counts matched. Now make sure that the root is | |
544 | * still valid and get it if required. | |
545 | */ | |
546 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
547 | spin_lock(&fs->lock); | |
548 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
549 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
550 | path_get(&nd->root); |
551 | spin_unlock(&fs->lock); | |
552 | } | |
31e6b01f | 553 | |
8b61e74f | 554 | rcu_read_unlock(); |
31e6b01f | 555 | return 0; |
19660af7 | 556 | |
e5c832d5 LT |
557 | unlock_and_drop_dentry: |
558 | spin_unlock(&fs->lock); | |
559 | drop_dentry: | |
8b61e74f | 560 | rcu_read_unlock(); |
15570086 | 561 | dput(dentry); |
d0d27277 | 562 | goto drop_root_mnt; |
e5c832d5 | 563 | out: |
8b61e74f | 564 | rcu_read_unlock(); |
d0d27277 LT |
565 | drop_root_mnt: |
566 | if (!(nd->flags & LOOKUP_ROOT)) | |
567 | nd->root.mnt = NULL; | |
31e6b01f NP |
568 | return -ECHILD; |
569 | } | |
570 | ||
4ce16ef3 | 571 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 572 | { |
4ce16ef3 | 573 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
574 | } |
575 | ||
9f1fafee AV |
576 | /** |
577 | * complete_walk - successful completion of path walk | |
578 | * @nd: pointer nameidata | |
39159de2 | 579 | * |
9f1fafee AV |
580 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
581 | * Revalidate the final result, unless we'd already done that during | |
582 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
583 | * success, -error on failure. In case of failure caller does not | |
584 | * need to drop nd->path. | |
39159de2 | 585 | */ |
9f1fafee | 586 | static int complete_walk(struct nameidata *nd) |
39159de2 | 587 | { |
16c2cd71 | 588 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 589 | int status; |
39159de2 | 590 | |
9f1fafee AV |
591 | if (nd->flags & LOOKUP_RCU) { |
592 | nd->flags &= ~LOOKUP_RCU; | |
593 | if (!(nd->flags & LOOKUP_ROOT)) | |
594 | nd->root.mnt = NULL; | |
15570086 | 595 | |
48a066e7 | 596 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { |
8b61e74f | 597 | rcu_read_unlock(); |
48a066e7 AV |
598 | return -ECHILD; |
599 | } | |
e5c832d5 | 600 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
8b61e74f | 601 | rcu_read_unlock(); |
48a066e7 | 602 | mntput(nd->path.mnt); |
e5c832d5 LT |
603 | return -ECHILD; |
604 | } | |
605 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
8b61e74f | 606 | rcu_read_unlock(); |
e5c832d5 | 607 | dput(dentry); |
48a066e7 | 608 | mntput(nd->path.mnt); |
9f1fafee AV |
609 | return -ECHILD; |
610 | } | |
8b61e74f | 611 | rcu_read_unlock(); |
9f1fafee AV |
612 | } |
613 | ||
16c2cd71 AV |
614 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
615 | return 0; | |
616 | ||
ecf3d1f1 | 617 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
618 | return 0; |
619 | ||
ecf3d1f1 | 620 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
621 | if (status > 0) |
622 | return 0; | |
623 | ||
16c2cd71 | 624 | if (!status) |
39159de2 | 625 | status = -ESTALE; |
16c2cd71 | 626 | |
9f1fafee | 627 | path_put(&nd->path); |
39159de2 JL |
628 | return status; |
629 | } | |
630 | ||
2a737871 AV |
631 | static __always_inline void set_root(struct nameidata *nd) |
632 | { | |
f7ad3c6b MS |
633 | if (!nd->root.mnt) |
634 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
635 | } |
636 | ||
6de88d72 AV |
637 | static int link_path_walk(const char *, struct nameidata *); |
638 | ||
31e6b01f NP |
639 | static __always_inline void set_root_rcu(struct nameidata *nd) |
640 | { | |
641 | if (!nd->root.mnt) { | |
642 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
643 | unsigned seq; |
644 | ||
645 | do { | |
646 | seq = read_seqcount_begin(&fs->seq); | |
647 | nd->root = fs->root; | |
c1530019 | 648 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 649 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
650 | } |
651 | } | |
652 | ||
1d957f9b | 653 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
654 | { |
655 | dput(path->dentry); | |
4ac91378 | 656 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
657 | mntput(path->mnt); |
658 | } | |
659 | ||
7b9337aa NP |
660 | static inline void path_to_nameidata(const struct path *path, |
661 | struct nameidata *nd) | |
051d3812 | 662 | { |
31e6b01f NP |
663 | if (!(nd->flags & LOOKUP_RCU)) { |
664 | dput(nd->path.dentry); | |
665 | if (nd->path.mnt != path->mnt) | |
666 | mntput(nd->path.mnt); | |
9a229683 | 667 | } |
31e6b01f | 668 | nd->path.mnt = path->mnt; |
4ac91378 | 669 | nd->path.dentry = path->dentry; |
051d3812 IK |
670 | } |
671 | ||
b5fb63c1 CH |
672 | /* |
673 | * Helper to directly jump to a known parsed path from ->follow_link, | |
674 | * caller must have taken a reference to path beforehand. | |
675 | */ | |
676 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
677 | { | |
678 | path_put(&nd->path); | |
679 | ||
680 | nd->path = *path; | |
681 | nd->inode = nd->path.dentry->d_inode; | |
682 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
683 | } |
684 | ||
574197e0 AV |
685 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
686 | { | |
687 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 688 | if (inode->i_op->put_link) |
574197e0 AV |
689 | inode->i_op->put_link(link->dentry, nd, cookie); |
690 | path_put(link); | |
691 | } | |
692 | ||
561ec64a LT |
693 | int sysctl_protected_symlinks __read_mostly = 0; |
694 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
695 | |
696 | /** | |
697 | * may_follow_link - Check symlink following for unsafe situations | |
698 | * @link: The path of the symlink | |
55852635 | 699 | * @nd: nameidata pathwalk data |
800179c9 KC |
700 | * |
701 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
702 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
703 | * in a sticky world-writable directory. This is to protect privileged | |
704 | * processes from failing races against path names that may change out | |
705 | * from under them by way of other users creating malicious symlinks. | |
706 | * It will permit symlinks to be followed only when outside a sticky | |
707 | * world-writable directory, or when the uid of the symlink and follower | |
708 | * match, or when the directory owner matches the symlink's owner. | |
709 | * | |
710 | * Returns 0 if following the symlink is allowed, -ve on error. | |
711 | */ | |
712 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
713 | { | |
714 | const struct inode *inode; | |
715 | const struct inode *parent; | |
716 | ||
717 | if (!sysctl_protected_symlinks) | |
718 | return 0; | |
719 | ||
720 | /* Allowed if owner and follower match. */ | |
721 | inode = link->dentry->d_inode; | |
81abe27b | 722 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
723 | return 0; |
724 | ||
725 | /* Allowed if parent directory not sticky and world-writable. */ | |
726 | parent = nd->path.dentry->d_inode; | |
727 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
728 | return 0; | |
729 | ||
730 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 731 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
732 | return 0; |
733 | ||
ffd8d101 | 734 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
735 | path_put_conditional(link, nd); |
736 | path_put(&nd->path); | |
737 | return -EACCES; | |
738 | } | |
739 | ||
740 | /** | |
741 | * safe_hardlink_source - Check for safe hardlink conditions | |
742 | * @inode: the source inode to hardlink from | |
743 | * | |
744 | * Return false if at least one of the following conditions: | |
745 | * - inode is not a regular file | |
746 | * - inode is setuid | |
747 | * - inode is setgid and group-exec | |
748 | * - access failure for read and write | |
749 | * | |
750 | * Otherwise returns true. | |
751 | */ | |
752 | static bool safe_hardlink_source(struct inode *inode) | |
753 | { | |
754 | umode_t mode = inode->i_mode; | |
755 | ||
756 | /* Special files should not get pinned to the filesystem. */ | |
757 | if (!S_ISREG(mode)) | |
758 | return false; | |
759 | ||
760 | /* Setuid files should not get pinned to the filesystem. */ | |
761 | if (mode & S_ISUID) | |
762 | return false; | |
763 | ||
764 | /* Executable setgid files should not get pinned to the filesystem. */ | |
765 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
766 | return false; | |
767 | ||
768 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
769 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
770 | return false; | |
771 | ||
772 | return true; | |
773 | } | |
774 | ||
775 | /** | |
776 | * may_linkat - Check permissions for creating a hardlink | |
777 | * @link: the source to hardlink from | |
778 | * | |
779 | * Block hardlink when all of: | |
780 | * - sysctl_protected_hardlinks enabled | |
781 | * - fsuid does not match inode | |
782 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
783 | * - not CAP_FOWNER | |
784 | * | |
785 | * Returns 0 if successful, -ve on error. | |
786 | */ | |
787 | static int may_linkat(struct path *link) | |
788 | { | |
789 | const struct cred *cred; | |
790 | struct inode *inode; | |
791 | ||
792 | if (!sysctl_protected_hardlinks) | |
793 | return 0; | |
794 | ||
795 | cred = current_cred(); | |
796 | inode = link->dentry->d_inode; | |
797 | ||
798 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
799 | * otherwise, it must be a safe source. | |
800 | */ | |
81abe27b | 801 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
802 | capable(CAP_FOWNER)) |
803 | return 0; | |
804 | ||
a51d9eaa | 805 | audit_log_link_denied("linkat", link); |
800179c9 KC |
806 | return -EPERM; |
807 | } | |
808 | ||
def4af30 | 809 | static __always_inline int |
574197e0 | 810 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 811 | { |
7b9337aa | 812 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
813 | int error; |
814 | char *s; | |
1da177e4 | 815 | |
844a3917 AV |
816 | BUG_ON(nd->flags & LOOKUP_RCU); |
817 | ||
0e794589 AV |
818 | if (link->mnt == nd->path.mnt) |
819 | mntget(link->mnt); | |
820 | ||
6d7b5aae AV |
821 | error = -ELOOP; |
822 | if (unlikely(current->total_link_count >= 40)) | |
823 | goto out_put_nd_path; | |
824 | ||
574197e0 AV |
825 | cond_resched(); |
826 | current->total_link_count++; | |
827 | ||
68ac1234 | 828 | touch_atime(link); |
1da177e4 | 829 | nd_set_link(nd, NULL); |
cd4e91d3 | 830 | |
36f3b4f6 | 831 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
832 | if (error) |
833 | goto out_put_nd_path; | |
36f3b4f6 | 834 | |
86acdca1 | 835 | nd->last_type = LAST_BIND; |
def4af30 AV |
836 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
837 | error = PTR_ERR(*p); | |
6d7b5aae | 838 | if (IS_ERR(*p)) |
408ef013 | 839 | goto out_put_nd_path; |
6d7b5aae AV |
840 | |
841 | error = 0; | |
842 | s = nd_get_link(nd); | |
843 | if (s) { | |
443ed254 AV |
844 | if (unlikely(IS_ERR(s))) { |
845 | path_put(&nd->path); | |
846 | put_link(nd, link, *p); | |
847 | return PTR_ERR(s); | |
848 | } | |
849 | if (*s == '/') { | |
850 | set_root(nd); | |
851 | path_put(&nd->path); | |
852 | nd->path = nd->root; | |
853 | path_get(&nd->root); | |
854 | nd->flags |= LOOKUP_JUMPED; | |
855 | } | |
856 | nd->inode = nd->path.dentry->d_inode; | |
857 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
858 | if (unlikely(error)) |
859 | put_link(nd, link, *p); | |
1da177e4 | 860 | } |
6d7b5aae AV |
861 | |
862 | return error; | |
863 | ||
864 | out_put_nd_path: | |
98f6ef64 | 865 | *p = NULL; |
6d7b5aae | 866 | path_put(&nd->path); |
6d7b5aae | 867 | path_put(link); |
1da177e4 LT |
868 | return error; |
869 | } | |
870 | ||
31e6b01f NP |
871 | static int follow_up_rcu(struct path *path) |
872 | { | |
0714a533 AV |
873 | struct mount *mnt = real_mount(path->mnt); |
874 | struct mount *parent; | |
31e6b01f NP |
875 | struct dentry *mountpoint; |
876 | ||
0714a533 AV |
877 | parent = mnt->mnt_parent; |
878 | if (&parent->mnt == path->mnt) | |
31e6b01f | 879 | return 0; |
a73324da | 880 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 881 | path->dentry = mountpoint; |
0714a533 | 882 | path->mnt = &parent->mnt; |
31e6b01f NP |
883 | return 1; |
884 | } | |
885 | ||
f015f126 DH |
886 | /* |
887 | * follow_up - Find the mountpoint of path's vfsmount | |
888 | * | |
889 | * Given a path, find the mountpoint of its source file system. | |
890 | * Replace @path with the path of the mountpoint in the parent mount. | |
891 | * Up is towards /. | |
892 | * | |
893 | * Return 1 if we went up a level and 0 if we were already at the | |
894 | * root. | |
895 | */ | |
bab77ebf | 896 | int follow_up(struct path *path) |
1da177e4 | 897 | { |
0714a533 AV |
898 | struct mount *mnt = real_mount(path->mnt); |
899 | struct mount *parent; | |
1da177e4 | 900 | struct dentry *mountpoint; |
99b7db7b | 901 | |
48a066e7 | 902 | read_seqlock_excl(&mount_lock); |
0714a533 | 903 | parent = mnt->mnt_parent; |
3c0a6163 | 904 | if (parent == mnt) { |
48a066e7 | 905 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
906 | return 0; |
907 | } | |
0714a533 | 908 | mntget(&parent->mnt); |
a73324da | 909 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 910 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
911 | dput(path->dentry); |
912 | path->dentry = mountpoint; | |
913 | mntput(path->mnt); | |
0714a533 | 914 | path->mnt = &parent->mnt; |
1da177e4 LT |
915 | return 1; |
916 | } | |
917 | ||
b5c84bf6 | 918 | /* |
9875cf80 DH |
919 | * Perform an automount |
920 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
921 | * were called with. | |
1da177e4 | 922 | */ |
9875cf80 DH |
923 | static int follow_automount(struct path *path, unsigned flags, |
924 | bool *need_mntput) | |
31e6b01f | 925 | { |
9875cf80 | 926 | struct vfsmount *mnt; |
ea5b778a | 927 | int err; |
9875cf80 DH |
928 | |
929 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
930 | return -EREMOTE; | |
931 | ||
0ec26fd0 MS |
932 | /* We don't want to mount if someone's just doing a stat - |
933 | * unless they're stat'ing a directory and appended a '/' to | |
934 | * the name. | |
935 | * | |
936 | * We do, however, want to mount if someone wants to open or | |
937 | * create a file of any type under the mountpoint, wants to | |
938 | * traverse through the mountpoint or wants to open the | |
939 | * mounted directory. Also, autofs may mark negative dentries | |
940 | * as being automount points. These will need the attentions | |
941 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 942 | */ |
0ec26fd0 | 943 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 944 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
945 | path->dentry->d_inode) |
946 | return -EISDIR; | |
947 | ||
9875cf80 DH |
948 | current->total_link_count++; |
949 | if (current->total_link_count >= 40) | |
950 | return -ELOOP; | |
951 | ||
952 | mnt = path->dentry->d_op->d_automount(path); | |
953 | if (IS_ERR(mnt)) { | |
954 | /* | |
955 | * The filesystem is allowed to return -EISDIR here to indicate | |
956 | * it doesn't want to automount. For instance, autofs would do | |
957 | * this so that its userspace daemon can mount on this dentry. | |
958 | * | |
959 | * However, we can only permit this if it's a terminal point in | |
960 | * the path being looked up; if it wasn't then the remainder of | |
961 | * the path is inaccessible and we should say so. | |
962 | */ | |
49084c3b | 963 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
964 | return -EREMOTE; |
965 | return PTR_ERR(mnt); | |
31e6b01f | 966 | } |
ea5b778a | 967 | |
9875cf80 DH |
968 | if (!mnt) /* mount collision */ |
969 | return 0; | |
31e6b01f | 970 | |
8aef1884 AV |
971 | if (!*need_mntput) { |
972 | /* lock_mount() may release path->mnt on error */ | |
973 | mntget(path->mnt); | |
974 | *need_mntput = true; | |
975 | } | |
19a167af | 976 | err = finish_automount(mnt, path); |
9875cf80 | 977 | |
ea5b778a DH |
978 | switch (err) { |
979 | case -EBUSY: | |
980 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 981 | return 0; |
ea5b778a | 982 | case 0: |
8aef1884 | 983 | path_put(path); |
ea5b778a DH |
984 | path->mnt = mnt; |
985 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 986 | return 0; |
19a167af AV |
987 | default: |
988 | return err; | |
ea5b778a | 989 | } |
19a167af | 990 | |
463ffb2e AV |
991 | } |
992 | ||
9875cf80 DH |
993 | /* |
994 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 995 | * - Flagged for transit management (autofs) |
9875cf80 DH |
996 | * - Flagged as mountpoint |
997 | * - Flagged as automount point | |
998 | * | |
999 | * This may only be called in refwalk mode. | |
1000 | * | |
1001 | * Serialization is taken care of in namespace.c | |
1002 | */ | |
1003 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1004 | { |
8aef1884 | 1005 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1006 | unsigned managed; |
1007 | bool need_mntput = false; | |
8aef1884 | 1008 | int ret = 0; |
9875cf80 DH |
1009 | |
1010 | /* Given that we're not holding a lock here, we retain the value in a | |
1011 | * local variable for each dentry as we look at it so that we don't see | |
1012 | * the components of that value change under us */ | |
1013 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1014 | managed &= DCACHE_MANAGED_DENTRY, | |
1015 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1016 | /* Allow the filesystem to manage the transit without i_mutex |
1017 | * being held. */ | |
1018 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1019 | BUG_ON(!path->dentry->d_op); | |
1020 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1021 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1022 | if (ret < 0) |
8aef1884 | 1023 | break; |
cc53ce53 DH |
1024 | } |
1025 | ||
9875cf80 DH |
1026 | /* Transit to a mounted filesystem. */ |
1027 | if (managed & DCACHE_MOUNTED) { | |
1028 | struct vfsmount *mounted = lookup_mnt(path); | |
1029 | if (mounted) { | |
1030 | dput(path->dentry); | |
1031 | if (need_mntput) | |
1032 | mntput(path->mnt); | |
1033 | path->mnt = mounted; | |
1034 | path->dentry = dget(mounted->mnt_root); | |
1035 | need_mntput = true; | |
1036 | continue; | |
1037 | } | |
1038 | ||
1039 | /* Something is mounted on this dentry in another | |
1040 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1041 | * namespace got unmounted before lookup_mnt() could |
1042 | * get it */ | |
9875cf80 DH |
1043 | } |
1044 | ||
1045 | /* Handle an automount point */ | |
1046 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1047 | ret = follow_automount(path, flags, &need_mntput); | |
1048 | if (ret < 0) | |
8aef1884 | 1049 | break; |
9875cf80 DH |
1050 | continue; |
1051 | } | |
1052 | ||
1053 | /* We didn't change the current path point */ | |
1054 | break; | |
1da177e4 | 1055 | } |
8aef1884 AV |
1056 | |
1057 | if (need_mntput && path->mnt == mnt) | |
1058 | mntput(path->mnt); | |
1059 | if (ret == -EISDIR) | |
1060 | ret = 0; | |
a3fbbde7 | 1061 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1062 | } |
1063 | ||
cc53ce53 | 1064 | int follow_down_one(struct path *path) |
1da177e4 LT |
1065 | { |
1066 | struct vfsmount *mounted; | |
1067 | ||
1c755af4 | 1068 | mounted = lookup_mnt(path); |
1da177e4 | 1069 | if (mounted) { |
9393bd07 AV |
1070 | dput(path->dentry); |
1071 | mntput(path->mnt); | |
1072 | path->mnt = mounted; | |
1073 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1074 | return 1; |
1075 | } | |
1076 | return 0; | |
1077 | } | |
1078 | ||
62a7375e IK |
1079 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
1080 | { | |
1081 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
1082 | dentry->d_op->d_manage(dentry, true) < 0); | |
1083 | } | |
1084 | ||
9875cf80 | 1085 | /* |
287548e4 AV |
1086 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1087 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1088 | */ |
1089 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1090 | struct inode **inode) |
9875cf80 | 1091 | { |
62a7375e | 1092 | for (;;) { |
c7105365 | 1093 | struct mount *mounted; |
62a7375e IK |
1094 | /* |
1095 | * Don't forget we might have a non-mountpoint managed dentry | |
1096 | * that wants to block transit. | |
1097 | */ | |
287548e4 | 1098 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 1099 | return false; |
62a7375e IK |
1100 | |
1101 | if (!d_mountpoint(path->dentry)) | |
1102 | break; | |
1103 | ||
474279dc | 1104 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1105 | if (!mounted) |
1106 | break; | |
c7105365 AV |
1107 | path->mnt = &mounted->mnt; |
1108 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1109 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1110 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1111 | /* |
1112 | * Update the inode too. We don't need to re-check the | |
1113 | * dentry sequence number here after this d_inode read, | |
1114 | * because a mount-point is always pinned. | |
1115 | */ | |
1116 | *inode = path->dentry->d_inode; | |
9875cf80 | 1117 | } |
9875cf80 DH |
1118 | return true; |
1119 | } | |
1120 | ||
dea39376 | 1121 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 1122 | { |
dea39376 | 1123 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 1124 | struct mount *mounted; |
474279dc | 1125 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); |
287548e4 AV |
1126 | if (!mounted) |
1127 | break; | |
c7105365 AV |
1128 | nd->path.mnt = &mounted->mnt; |
1129 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 1130 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
1131 | } |
1132 | } | |
1133 | ||
31e6b01f NP |
1134 | static int follow_dotdot_rcu(struct nameidata *nd) |
1135 | { | |
31e6b01f NP |
1136 | set_root_rcu(nd); |
1137 | ||
9875cf80 | 1138 | while (1) { |
31e6b01f NP |
1139 | if (nd->path.dentry == nd->root.dentry && |
1140 | nd->path.mnt == nd->root.mnt) { | |
1141 | break; | |
1142 | } | |
1143 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1144 | struct dentry *old = nd->path.dentry; | |
1145 | struct dentry *parent = old->d_parent; | |
1146 | unsigned seq; | |
1147 | ||
1148 | seq = read_seqcount_begin(&parent->d_seq); | |
1149 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1150 | goto failed; |
31e6b01f NP |
1151 | nd->path.dentry = parent; |
1152 | nd->seq = seq; | |
1153 | break; | |
1154 | } | |
1155 | if (!follow_up_rcu(&nd->path)) | |
1156 | break; | |
1157 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1158 | } |
dea39376 AV |
1159 | follow_mount_rcu(nd); |
1160 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 1161 | return 0; |
ef7562d5 AV |
1162 | |
1163 | failed: | |
1164 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1165 | if (!(nd->flags & LOOKUP_ROOT)) |
1166 | nd->root.mnt = NULL; | |
8b61e74f | 1167 | rcu_read_unlock(); |
ef7562d5 | 1168 | return -ECHILD; |
31e6b01f NP |
1169 | } |
1170 | ||
cc53ce53 DH |
1171 | /* |
1172 | * Follow down to the covering mount currently visible to userspace. At each | |
1173 | * point, the filesystem owning that dentry may be queried as to whether the | |
1174 | * caller is permitted to proceed or not. | |
cc53ce53 | 1175 | */ |
7cc90cc3 | 1176 | int follow_down(struct path *path) |
cc53ce53 DH |
1177 | { |
1178 | unsigned managed; | |
1179 | int ret; | |
1180 | ||
1181 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1182 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1183 | /* Allow the filesystem to manage the transit without i_mutex | |
1184 | * being held. | |
1185 | * | |
1186 | * We indicate to the filesystem if someone is trying to mount | |
1187 | * something here. This gives autofs the chance to deny anyone | |
1188 | * other than its daemon the right to mount on its | |
1189 | * superstructure. | |
1190 | * | |
1191 | * The filesystem may sleep at this point. | |
1192 | */ | |
1193 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1194 | BUG_ON(!path->dentry->d_op); | |
1195 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1196 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1197 | path->dentry, false); |
cc53ce53 DH |
1198 | if (ret < 0) |
1199 | return ret == -EISDIR ? 0 : ret; | |
1200 | } | |
1201 | ||
1202 | /* Transit to a mounted filesystem. */ | |
1203 | if (managed & DCACHE_MOUNTED) { | |
1204 | struct vfsmount *mounted = lookup_mnt(path); | |
1205 | if (!mounted) | |
1206 | break; | |
1207 | dput(path->dentry); | |
1208 | mntput(path->mnt); | |
1209 | path->mnt = mounted; | |
1210 | path->dentry = dget(mounted->mnt_root); | |
1211 | continue; | |
1212 | } | |
1213 | ||
1214 | /* Don't handle automount points here */ | |
1215 | break; | |
1216 | } | |
1217 | return 0; | |
1218 | } | |
1219 | ||
9875cf80 DH |
1220 | /* |
1221 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1222 | */ | |
1223 | static void follow_mount(struct path *path) | |
1224 | { | |
1225 | while (d_mountpoint(path->dentry)) { | |
1226 | struct vfsmount *mounted = lookup_mnt(path); | |
1227 | if (!mounted) | |
1228 | break; | |
1229 | dput(path->dentry); | |
1230 | mntput(path->mnt); | |
1231 | path->mnt = mounted; | |
1232 | path->dentry = dget(mounted->mnt_root); | |
1233 | } | |
1234 | } | |
1235 | ||
31e6b01f | 1236 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1237 | { |
2a737871 | 1238 | set_root(nd); |
e518ddb7 | 1239 | |
1da177e4 | 1240 | while(1) { |
4ac91378 | 1241 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1242 | |
2a737871 AV |
1243 | if (nd->path.dentry == nd->root.dentry && |
1244 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1245 | break; |
1246 | } | |
4ac91378 | 1247 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1248 | /* rare case of legitimate dget_parent()... */ |
1249 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1250 | dput(old); |
1251 | break; | |
1252 | } | |
3088dd70 | 1253 | if (!follow_up(&nd->path)) |
1da177e4 | 1254 | break; |
1da177e4 | 1255 | } |
79ed0226 | 1256 | follow_mount(&nd->path); |
31e6b01f | 1257 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1258 | } |
1259 | ||
baa03890 | 1260 | /* |
bad61189 MS |
1261 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1262 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1263 | * returns whether i_op->lookup is necessary. | |
1264 | * | |
1265 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1266 | */ |
bad61189 | 1267 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1268 | unsigned int flags, bool *need_lookup) |
baa03890 | 1269 | { |
baa03890 | 1270 | struct dentry *dentry; |
bad61189 | 1271 | int error; |
baa03890 | 1272 | |
bad61189 MS |
1273 | *need_lookup = false; |
1274 | dentry = d_lookup(dir, name); | |
1275 | if (dentry) { | |
39e3c955 | 1276 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1277 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1278 | if (unlikely(error <= 0)) { |
1279 | if (error < 0) { | |
1280 | dput(dentry); | |
1281 | return ERR_PTR(error); | |
1282 | } else if (!d_invalidate(dentry)) { | |
1283 | dput(dentry); | |
1284 | dentry = NULL; | |
1285 | } | |
1286 | } | |
1287 | } | |
1288 | } | |
baa03890 | 1289 | |
bad61189 MS |
1290 | if (!dentry) { |
1291 | dentry = d_alloc(dir, name); | |
1292 | if (unlikely(!dentry)) | |
1293 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1294 | |
bad61189 | 1295 | *need_lookup = true; |
baa03890 NP |
1296 | } |
1297 | return dentry; | |
1298 | } | |
1299 | ||
44396f4b | 1300 | /* |
bad61189 MS |
1301 | * Call i_op->lookup on the dentry. The dentry must be negative but may be |
1302 | * hashed if it was pouplated with DCACHE_NEED_LOOKUP. | |
1303 | * | |
1304 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1305 | */ |
bad61189 | 1306 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1307 | unsigned int flags) |
44396f4b | 1308 | { |
44396f4b JB |
1309 | struct dentry *old; |
1310 | ||
1311 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1312 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1313 | dput(dentry); |
44396f4b | 1314 | return ERR_PTR(-ENOENT); |
e188dc02 | 1315 | } |
44396f4b | 1316 | |
72bd866a | 1317 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1318 | if (unlikely(old)) { |
1319 | dput(dentry); | |
1320 | dentry = old; | |
1321 | } | |
1322 | return dentry; | |
1323 | } | |
1324 | ||
a3255546 | 1325 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1326 | struct dentry *base, unsigned int flags) |
a3255546 | 1327 | { |
bad61189 | 1328 | bool need_lookup; |
a3255546 AV |
1329 | struct dentry *dentry; |
1330 | ||
72bd866a | 1331 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1332 | if (!need_lookup) |
1333 | return dentry; | |
a3255546 | 1334 | |
72bd866a | 1335 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1336 | } |
1337 | ||
1da177e4 LT |
1338 | /* |
1339 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1340 | * small and for now I'd prefer to have fast path as straight as possible. | |
1341 | * It _is_ time-critical. | |
1342 | */ | |
e97cdc87 | 1343 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1344 | struct path *path, struct inode **inode) |
1da177e4 | 1345 | { |
4ac91378 | 1346 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1347 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1348 | int need_reval = 1; |
1349 | int status = 1; | |
9875cf80 DH |
1350 | int err; |
1351 | ||
b04f784e NP |
1352 | /* |
1353 | * Rename seqlock is not required here because in the off chance | |
1354 | * of a false negative due to a concurrent rename, we're going to | |
1355 | * do the non-racy lookup, below. | |
1356 | */ | |
31e6b01f NP |
1357 | if (nd->flags & LOOKUP_RCU) { |
1358 | unsigned seq; | |
da53be12 | 1359 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1360 | if (!dentry) |
1361 | goto unlazy; | |
1362 | ||
12f8ad4b LT |
1363 | /* |
1364 | * This sequence count validates that the inode matches | |
1365 | * the dentry name information from lookup. | |
1366 | */ | |
1367 | *inode = dentry->d_inode; | |
1368 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1369 | return -ECHILD; | |
1370 | ||
1371 | /* | |
1372 | * This sequence count validates that the parent had no | |
1373 | * changes while we did the lookup of the dentry above. | |
1374 | * | |
1375 | * The memory barrier in read_seqcount_begin of child is | |
1376 | * enough, we can use __read_seqcount_retry here. | |
1377 | */ | |
31e6b01f NP |
1378 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1379 | return -ECHILD; | |
31e6b01f | 1380 | nd->seq = seq; |
5a18fff2 | 1381 | |
24643087 | 1382 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1383 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1384 | if (unlikely(status <= 0)) { |
1385 | if (status != -ECHILD) | |
1386 | need_reval = 0; | |
1387 | goto unlazy; | |
1388 | } | |
24643087 | 1389 | } |
31e6b01f NP |
1390 | path->mnt = mnt; |
1391 | path->dentry = dentry; | |
d6e9bd25 AV |
1392 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1393 | goto unlazy; | |
1394 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1395 | goto unlazy; | |
1396 | return 0; | |
5a18fff2 | 1397 | unlazy: |
19660af7 AV |
1398 | if (unlazy_walk(nd, dentry)) |
1399 | return -ECHILD; | |
5a18fff2 | 1400 | } else { |
e97cdc87 | 1401 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1402 | } |
5a18fff2 | 1403 | |
81e6f520 AV |
1404 | if (unlikely(!dentry)) |
1405 | goto need_lookup; | |
1406 | ||
5a18fff2 | 1407 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1408 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1409 | if (unlikely(status <= 0)) { |
1410 | if (status < 0) { | |
1411 | dput(dentry); | |
1412 | return status; | |
1413 | } | |
1414 | if (!d_invalidate(dentry)) { | |
1415 | dput(dentry); | |
81e6f520 | 1416 | goto need_lookup; |
5a18fff2 | 1417 | } |
24643087 | 1418 | } |
697f514d | 1419 | |
9875cf80 DH |
1420 | path->mnt = mnt; |
1421 | path->dentry = dentry; | |
1422 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1423 | if (unlikely(err < 0)) { |
1424 | path_put_conditional(path, nd); | |
9875cf80 | 1425 | return err; |
89312214 | 1426 | } |
a3fbbde7 AV |
1427 | if (err) |
1428 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1429 | *inode = path->dentry->d_inode; |
1da177e4 | 1430 | return 0; |
81e6f520 AV |
1431 | |
1432 | need_lookup: | |
697f514d MS |
1433 | return 1; |
1434 | } | |
1435 | ||
1436 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1437 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1438 | { |
1439 | struct dentry *dentry, *parent; | |
1440 | int err; | |
1441 | ||
1442 | parent = nd->path.dentry; | |
81e6f520 AV |
1443 | BUG_ON(nd->inode != parent->d_inode); |
1444 | ||
1445 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1446 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1447 | mutex_unlock(&parent->d_inode->i_mutex); |
1448 | if (IS_ERR(dentry)) | |
1449 | return PTR_ERR(dentry); | |
697f514d MS |
1450 | path->mnt = nd->path.mnt; |
1451 | path->dentry = dentry; | |
1452 | err = follow_managed(path, nd->flags); | |
1453 | if (unlikely(err < 0)) { | |
1454 | path_put_conditional(path, nd); | |
1455 | return err; | |
1456 | } | |
1457 | if (err) | |
1458 | nd->flags |= LOOKUP_JUMPED; | |
1459 | return 0; | |
1da177e4 LT |
1460 | } |
1461 | ||
52094c8a AV |
1462 | static inline int may_lookup(struct nameidata *nd) |
1463 | { | |
1464 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1465 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1466 | if (err != -ECHILD) |
1467 | return err; | |
19660af7 | 1468 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1469 | return -ECHILD; |
1470 | } | |
4ad5abb3 | 1471 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1472 | } |
1473 | ||
9856fa1b AV |
1474 | static inline int handle_dots(struct nameidata *nd, int type) |
1475 | { | |
1476 | if (type == LAST_DOTDOT) { | |
1477 | if (nd->flags & LOOKUP_RCU) { | |
1478 | if (follow_dotdot_rcu(nd)) | |
1479 | return -ECHILD; | |
1480 | } else | |
1481 | follow_dotdot(nd); | |
1482 | } | |
1483 | return 0; | |
1484 | } | |
1485 | ||
951361f9 AV |
1486 | static void terminate_walk(struct nameidata *nd) |
1487 | { | |
1488 | if (!(nd->flags & LOOKUP_RCU)) { | |
1489 | path_put(&nd->path); | |
1490 | } else { | |
1491 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1492 | if (!(nd->flags & LOOKUP_ROOT)) |
1493 | nd->root.mnt = NULL; | |
8b61e74f | 1494 | rcu_read_unlock(); |
951361f9 AV |
1495 | } |
1496 | } | |
1497 | ||
3ddcd056 LT |
1498 | /* |
1499 | * Do we need to follow links? We _really_ want to be able | |
1500 | * to do this check without having to look at inode->i_op, | |
1501 | * so we keep a cache of "no, this doesn't need follow_link" | |
1502 | * for the common case. | |
1503 | */ | |
7813b94a | 1504 | static inline int should_follow_link(struct inode *inode, int follow) |
3ddcd056 LT |
1505 | { |
1506 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1507 | if (likely(inode->i_op->follow_link)) | |
1508 | return follow; | |
1509 | ||
1510 | /* This gets set once for the inode lifetime */ | |
1511 | spin_lock(&inode->i_lock); | |
1512 | inode->i_opflags |= IOP_NOFOLLOW; | |
1513 | spin_unlock(&inode->i_lock); | |
1514 | } | |
1515 | return 0; | |
1516 | } | |
1517 | ||
ce57dfc1 | 1518 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1519 | int follow) |
ce57dfc1 AV |
1520 | { |
1521 | struct inode *inode; | |
1522 | int err; | |
1523 | /* | |
1524 | * "." and ".." are special - ".." especially so because it has | |
1525 | * to be able to know about the current root directory and | |
1526 | * parent relationships. | |
1527 | */ | |
21b9b073 AV |
1528 | if (unlikely(nd->last_type != LAST_NORM)) |
1529 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1530 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1531 | if (unlikely(err)) { |
697f514d MS |
1532 | if (err < 0) |
1533 | goto out_err; | |
1534 | ||
cc2a5271 | 1535 | err = lookup_slow(nd, path); |
697f514d MS |
1536 | if (err < 0) |
1537 | goto out_err; | |
1538 | ||
1539 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1540 | } |
697f514d MS |
1541 | err = -ENOENT; |
1542 | if (!inode) | |
1543 | goto out_path_put; | |
1544 | ||
7813b94a | 1545 | if (should_follow_link(inode, follow)) { |
19660af7 AV |
1546 | if (nd->flags & LOOKUP_RCU) { |
1547 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1548 | err = -ECHILD; |
1549 | goto out_err; | |
19660af7 AV |
1550 | } |
1551 | } | |
ce57dfc1 AV |
1552 | BUG_ON(inode != path->dentry->d_inode); |
1553 | return 1; | |
1554 | } | |
1555 | path_to_nameidata(path, nd); | |
1556 | nd->inode = inode; | |
1557 | return 0; | |
697f514d MS |
1558 | |
1559 | out_path_put: | |
1560 | path_to_nameidata(path, nd); | |
1561 | out_err: | |
1562 | terminate_walk(nd); | |
1563 | return err; | |
ce57dfc1 AV |
1564 | } |
1565 | ||
b356379a AV |
1566 | /* |
1567 | * This limits recursive symlink follows to 8, while | |
1568 | * limiting consecutive symlinks to 40. | |
1569 | * | |
1570 | * Without that kind of total limit, nasty chains of consecutive | |
1571 | * symlinks can cause almost arbitrarily long lookups. | |
1572 | */ | |
1573 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1574 | { | |
1575 | int res; | |
1576 | ||
b356379a AV |
1577 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1578 | path_put_conditional(path, nd); | |
1579 | path_put(&nd->path); | |
1580 | return -ELOOP; | |
1581 | } | |
1a4022f8 | 1582 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1583 | |
1584 | nd->depth++; | |
1585 | current->link_count++; | |
1586 | ||
1587 | do { | |
1588 | struct path link = *path; | |
1589 | void *cookie; | |
574197e0 AV |
1590 | |
1591 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1592 | if (res) |
1593 | break; | |
21b9b073 | 1594 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1595 | put_link(nd, &link, cookie); |
b356379a AV |
1596 | } while (res > 0); |
1597 | ||
1598 | current->link_count--; | |
1599 | nd->depth--; | |
1600 | return res; | |
1601 | } | |
1602 | ||
3ddcd056 LT |
1603 | /* |
1604 | * We really don't want to look at inode->i_op->lookup | |
1605 | * when we don't have to. So we keep a cache bit in | |
1606 | * the inode ->i_opflags field that says "yes, we can | |
1607 | * do lookup on this inode". | |
1608 | */ | |
1609 | static inline int can_lookup(struct inode *inode) | |
1610 | { | |
1611 | if (likely(inode->i_opflags & IOP_LOOKUP)) | |
1612 | return 1; | |
1613 | if (likely(!inode->i_op->lookup)) | |
1614 | return 0; | |
1615 | ||
1616 | /* We do this once for the lifetime of the inode */ | |
1617 | spin_lock(&inode->i_lock); | |
1618 | inode->i_opflags |= IOP_LOOKUP; | |
1619 | spin_unlock(&inode->i_lock); | |
1620 | return 1; | |
1621 | } | |
1622 | ||
bfcfaa77 LT |
1623 | /* |
1624 | * We can do the critical dentry name comparison and hashing | |
1625 | * operations one word at a time, but we are limited to: | |
1626 | * | |
1627 | * - Architectures with fast unaligned word accesses. We could | |
1628 | * do a "get_unaligned()" if this helps and is sufficiently | |
1629 | * fast. | |
1630 | * | |
1631 | * - Little-endian machines (so that we can generate the mask | |
1632 | * of low bytes efficiently). Again, we *could* do a byte | |
1633 | * swapping load on big-endian architectures if that is not | |
1634 | * expensive enough to make the optimization worthless. | |
1635 | * | |
1636 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1637 | * do not trap on the (extremely unlikely) case of a page | |
1638 | * crossing operation. | |
1639 | * | |
1640 | * - Furthermore, we need an efficient 64-bit compile for the | |
1641 | * 64-bit case in order to generate the "number of bytes in | |
1642 | * the final mask". Again, that could be replaced with a | |
1643 | * efficient population count instruction or similar. | |
1644 | */ | |
1645 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1646 | ||
f68e556e | 1647 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1648 | |
f68e556e | 1649 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1650 | |
1651 | static inline unsigned int fold_hash(unsigned long hash) | |
1652 | { | |
1653 | hash += hash >> (8*sizeof(int)); | |
1654 | return hash; | |
1655 | } | |
1656 | ||
1657 | #else /* 32-bit case */ | |
1658 | ||
bfcfaa77 LT |
1659 | #define fold_hash(x) (x) |
1660 | ||
1661 | #endif | |
1662 | ||
1663 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1664 | { | |
1665 | unsigned long a, mask; | |
1666 | unsigned long hash = 0; | |
1667 | ||
1668 | for (;;) { | |
e419b4cc | 1669 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1670 | if (len < sizeof(unsigned long)) |
1671 | break; | |
1672 | hash += a; | |
f132c5be | 1673 | hash *= 9; |
bfcfaa77 LT |
1674 | name += sizeof(unsigned long); |
1675 | len -= sizeof(unsigned long); | |
1676 | if (!len) | |
1677 | goto done; | |
1678 | } | |
1679 | mask = ~(~0ul << len*8); | |
1680 | hash += mask & a; | |
1681 | done: | |
1682 | return fold_hash(hash); | |
1683 | } | |
1684 | EXPORT_SYMBOL(full_name_hash); | |
1685 | ||
bfcfaa77 LT |
1686 | /* |
1687 | * Calculate the length and hash of the path component, and | |
1688 | * return the length of the component; | |
1689 | */ | |
1690 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1691 | { | |
36126f8f LT |
1692 | unsigned long a, b, adata, bdata, mask, hash, len; |
1693 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1694 | |
1695 | hash = a = 0; | |
1696 | len = -sizeof(unsigned long); | |
1697 | do { | |
1698 | hash = (hash + a) * 9; | |
1699 | len += sizeof(unsigned long); | |
e419b4cc | 1700 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1701 | b = a ^ REPEAT_BYTE('/'); |
1702 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1703 | ||
1704 | adata = prep_zero_mask(a, adata, &constants); | |
1705 | bdata = prep_zero_mask(b, bdata, &constants); | |
1706 | ||
1707 | mask = create_zero_mask(adata | bdata); | |
1708 | ||
1709 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1710 | *hashp = fold_hash(hash); |
1711 | ||
36126f8f | 1712 | return len + find_zero(mask); |
bfcfaa77 LT |
1713 | } |
1714 | ||
1715 | #else | |
1716 | ||
0145acc2 LT |
1717 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1718 | { | |
1719 | unsigned long hash = init_name_hash(); | |
1720 | while (len--) | |
1721 | hash = partial_name_hash(*name++, hash); | |
1722 | return end_name_hash(hash); | |
1723 | } | |
ae942ae7 | 1724 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1725 | |
200e9ef7 LT |
1726 | /* |
1727 | * We know there's a real path component here of at least | |
1728 | * one character. | |
1729 | */ | |
1730 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1731 | { | |
1732 | unsigned long hash = init_name_hash(); | |
1733 | unsigned long len = 0, c; | |
1734 | ||
1735 | c = (unsigned char)*name; | |
1736 | do { | |
1737 | len++; | |
1738 | hash = partial_name_hash(c, hash); | |
1739 | c = (unsigned char)name[len]; | |
1740 | } while (c && c != '/'); | |
1741 | *hashp = end_name_hash(hash); | |
1742 | return len; | |
1743 | } | |
1744 | ||
bfcfaa77 LT |
1745 | #endif |
1746 | ||
1da177e4 LT |
1747 | /* |
1748 | * Name resolution. | |
ea3834d9 PM |
1749 | * This is the basic name resolution function, turning a pathname into |
1750 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1751 | * |
ea3834d9 PM |
1752 | * Returns 0 and nd will have valid dentry and mnt on success. |
1753 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1754 | */ |
6de88d72 | 1755 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1756 | { |
1757 | struct path next; | |
1da177e4 | 1758 | int err; |
1da177e4 LT |
1759 | |
1760 | while (*name=='/') | |
1761 | name++; | |
1762 | if (!*name) | |
086e183a | 1763 | return 0; |
1da177e4 | 1764 | |
1da177e4 LT |
1765 | /* At this point we know we have a real path component. */ |
1766 | for(;;) { | |
1da177e4 | 1767 | struct qstr this; |
200e9ef7 | 1768 | long len; |
fe479a58 | 1769 | int type; |
1da177e4 | 1770 | |
52094c8a | 1771 | err = may_lookup(nd); |
1da177e4 LT |
1772 | if (err) |
1773 | break; | |
1774 | ||
200e9ef7 | 1775 | len = hash_name(name, &this.hash); |
1da177e4 | 1776 | this.name = name; |
200e9ef7 | 1777 | this.len = len; |
1da177e4 | 1778 | |
fe479a58 | 1779 | type = LAST_NORM; |
200e9ef7 | 1780 | if (name[0] == '.') switch (len) { |
fe479a58 | 1781 | case 2: |
200e9ef7 | 1782 | if (name[1] == '.') { |
fe479a58 | 1783 | type = LAST_DOTDOT; |
16c2cd71 AV |
1784 | nd->flags |= LOOKUP_JUMPED; |
1785 | } | |
fe479a58 AV |
1786 | break; |
1787 | case 1: | |
1788 | type = LAST_DOT; | |
1789 | } | |
5a202bcd AV |
1790 | if (likely(type == LAST_NORM)) { |
1791 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1792 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1793 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1794 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1795 | if (err < 0) |
1796 | break; | |
1797 | } | |
1798 | } | |
fe479a58 | 1799 | |
5f4a6a69 AV |
1800 | nd->last = this; |
1801 | nd->last_type = type; | |
1802 | ||
200e9ef7 | 1803 | if (!name[len]) |
5f4a6a69 | 1804 | return 0; |
200e9ef7 LT |
1805 | /* |
1806 | * If it wasn't NUL, we know it was '/'. Skip that | |
1807 | * slash, and continue until no more slashes. | |
1808 | */ | |
1809 | do { | |
1810 | len++; | |
1811 | } while (unlikely(name[len] == '/')); | |
1812 | if (!name[len]) | |
5f4a6a69 AV |
1813 | return 0; |
1814 | ||
200e9ef7 | 1815 | name += len; |
1da177e4 | 1816 | |
21b9b073 | 1817 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1818 | if (err < 0) |
1819 | return err; | |
1da177e4 | 1820 | |
ce57dfc1 | 1821 | if (err) { |
b356379a | 1822 | err = nested_symlink(&next, nd); |
1da177e4 | 1823 | if (err) |
a7472bab | 1824 | return err; |
31e6b01f | 1825 | } |
5f4a6a69 AV |
1826 | if (!can_lookup(nd->inode)) { |
1827 | err = -ENOTDIR; | |
1828 | break; | |
1829 | } | |
1da177e4 | 1830 | } |
951361f9 | 1831 | terminate_walk(nd); |
1da177e4 LT |
1832 | return err; |
1833 | } | |
1834 | ||
70e9b357 AV |
1835 | static int path_init(int dfd, const char *name, unsigned int flags, |
1836 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1837 | { |
1838 | int retval = 0; | |
31e6b01f NP |
1839 | |
1840 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1841 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1842 | nd->depth = 0; |
5b6ca027 AV |
1843 | if (flags & LOOKUP_ROOT) { |
1844 | struct inode *inode = nd->root.dentry->d_inode; | |
73d049a4 | 1845 | if (*name) { |
741b7c3f | 1846 | if (!can_lookup(inode)) |
73d049a4 AV |
1847 | return -ENOTDIR; |
1848 | retval = inode_permission(inode, MAY_EXEC); | |
1849 | if (retval) | |
1850 | return retval; | |
1851 | } | |
5b6ca027 AV |
1852 | nd->path = nd->root; |
1853 | nd->inode = inode; | |
1854 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1855 | rcu_read_lock(); |
5b6ca027 | 1856 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1857 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1858 | } else { |
1859 | path_get(&nd->path); | |
1860 | } | |
1861 | return 0; | |
1862 | } | |
1863 | ||
31e6b01f | 1864 | nd->root.mnt = NULL; |
31e6b01f | 1865 | |
48a066e7 | 1866 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1867 | if (*name=='/') { |
e41f7d4e | 1868 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1869 | rcu_read_lock(); |
e41f7d4e AV |
1870 | set_root_rcu(nd); |
1871 | } else { | |
1872 | set_root(nd); | |
1873 | path_get(&nd->root); | |
1874 | } | |
1875 | nd->path = nd->root; | |
31e6b01f | 1876 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1877 | if (flags & LOOKUP_RCU) { |
1878 | struct fs_struct *fs = current->fs; | |
1879 | unsigned seq; | |
31e6b01f | 1880 | |
8b61e74f | 1881 | rcu_read_lock(); |
c28cc364 | 1882 | |
e41f7d4e AV |
1883 | do { |
1884 | seq = read_seqcount_begin(&fs->seq); | |
1885 | nd->path = fs->pwd; | |
1886 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1887 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1888 | } else { | |
1889 | get_fs_pwd(current->fs, &nd->path); | |
1890 | } | |
31e6b01f | 1891 | } else { |
582aa64a | 1892 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1893 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1894 | struct dentry *dentry; |
1895 | ||
2903ff01 AV |
1896 | if (!f.file) |
1897 | return -EBADF; | |
31e6b01f | 1898 | |
2903ff01 | 1899 | dentry = f.file->f_path.dentry; |
31e6b01f | 1900 | |
f52e0c11 | 1901 | if (*name) { |
741b7c3f | 1902 | if (!can_lookup(dentry->d_inode)) { |
2903ff01 AV |
1903 | fdput(f); |
1904 | return -ENOTDIR; | |
1905 | } | |
f52e0c11 | 1906 | } |
31e6b01f | 1907 | |
2903ff01 | 1908 | nd->path = f.file->f_path; |
e41f7d4e | 1909 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1910 | if (f.need_put) |
1911 | *fp = f.file; | |
e41f7d4e | 1912 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1913 | rcu_read_lock(); |
e41f7d4e | 1914 | } else { |
2903ff01 AV |
1915 | path_get(&nd->path); |
1916 | fdput(f); | |
e41f7d4e | 1917 | } |
31e6b01f | 1918 | } |
31e6b01f | 1919 | |
31e6b01f | 1920 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1921 | return 0; |
9b4a9b14 AV |
1922 | } |
1923 | ||
bd92d7fe AV |
1924 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1925 | { | |
1926 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1927 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1928 | ||
1929 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1930 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1931 | } |
1932 | ||
9b4a9b14 | 1933 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1934 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1935 | unsigned int flags, struct nameidata *nd) |
1936 | { | |
70e9b357 | 1937 | struct file *base = NULL; |
bd92d7fe AV |
1938 | struct path path; |
1939 | int err; | |
31e6b01f NP |
1940 | |
1941 | /* | |
1942 | * Path walking is largely split up into 2 different synchronisation | |
1943 | * schemes, rcu-walk and ref-walk (explained in | |
1944 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1945 | * path walk code, but some things particularly setup, cleanup, and | |
1946 | * following mounts are sufficiently divergent that functions are | |
1947 | * duplicated. Typically there is a function foo(), and its RCU | |
1948 | * analogue, foo_rcu(). | |
1949 | * | |
1950 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1951 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1952 | * be handled by restarting a traditional ref-walk (which will always | |
1953 | * be able to complete). | |
1954 | */ | |
bd92d7fe | 1955 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1956 | |
bd92d7fe AV |
1957 | if (unlikely(err)) |
1958 | return err; | |
ee0827cd AV |
1959 | |
1960 | current->total_link_count = 0; | |
bd92d7fe AV |
1961 | err = link_path_walk(name, nd); |
1962 | ||
1963 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1964 | err = lookup_last(nd, &path); |
1965 | while (err > 0) { | |
1966 | void *cookie; | |
1967 | struct path link = path; | |
800179c9 KC |
1968 | err = may_follow_link(&link, nd); |
1969 | if (unlikely(err)) | |
1970 | break; | |
bd92d7fe | 1971 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1972 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1973 | if (err) |
1974 | break; | |
1975 | err = lookup_last(nd, &path); | |
574197e0 | 1976 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1977 | } |
1978 | } | |
ee0827cd | 1979 | |
9f1fafee AV |
1980 | if (!err) |
1981 | err = complete_walk(nd); | |
bd92d7fe AV |
1982 | |
1983 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
05252901 | 1984 | if (!can_lookup(nd->inode)) { |
bd92d7fe | 1985 | path_put(&nd->path); |
bd23a539 | 1986 | err = -ENOTDIR; |
bd92d7fe AV |
1987 | } |
1988 | } | |
16c2cd71 | 1989 | |
70e9b357 AV |
1990 | if (base) |
1991 | fput(base); | |
ee0827cd | 1992 | |
5b6ca027 | 1993 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1994 | path_put(&nd->root); |
1995 | nd->root.mnt = NULL; | |
1996 | } | |
bd92d7fe | 1997 | return err; |
ee0827cd | 1998 | } |
31e6b01f | 1999 | |
873f1eed | 2000 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
2001 | unsigned int flags, struct nameidata *nd) |
2002 | { | |
873f1eed | 2003 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 2004 | if (unlikely(retval == -ECHILD)) |
873f1eed | 2005 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 2006 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
2007 | retval = path_lookupat(dfd, name->name, |
2008 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 2009 | |
f78570dd | 2010 | if (likely(!retval)) |
adb5c247 | 2011 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 2012 | return retval; |
1da177e4 LT |
2013 | } |
2014 | ||
873f1eed JL |
2015 | static int do_path_lookup(int dfd, const char *name, |
2016 | unsigned int flags, struct nameidata *nd) | |
2017 | { | |
2018 | struct filename filename = { .name = name }; | |
2019 | ||
2020 | return filename_lookup(dfd, &filename, flags, nd); | |
2021 | } | |
2022 | ||
79714f72 AV |
2023 | /* does lookup, returns the object with parent locked */ |
2024 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2025 | { |
79714f72 AV |
2026 | struct nameidata nd; |
2027 | struct dentry *d; | |
2028 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2029 | if (err) | |
2030 | return ERR_PTR(err); | |
2031 | if (nd.last_type != LAST_NORM) { | |
2032 | path_put(&nd.path); | |
2033 | return ERR_PTR(-EINVAL); | |
2034 | } | |
2035 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2036 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2037 | if (IS_ERR(d)) { |
2038 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2039 | path_put(&nd.path); | |
2040 | return d; | |
2041 | } | |
2042 | *path = nd.path; | |
2043 | return d; | |
5590ff0d UD |
2044 | } |
2045 | ||
d1811465 AV |
2046 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2047 | { | |
2048 | struct nameidata nd; | |
2049 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2050 | if (!res) | |
2051 | *path = nd.path; | |
2052 | return res; | |
2053 | } | |
2054 | ||
16f18200 JJS |
2055 | /** |
2056 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2057 | * @dentry: pointer to dentry of the base directory | |
2058 | * @mnt: pointer to vfs mount of the base directory | |
2059 | * @name: pointer to file name | |
2060 | * @flags: lookup flags | |
e0a01249 | 2061 | * @path: pointer to struct path to fill |
16f18200 JJS |
2062 | */ |
2063 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2064 | const char *name, unsigned int flags, | |
e0a01249 | 2065 | struct path *path) |
16f18200 | 2066 | { |
e0a01249 AV |
2067 | struct nameidata nd; |
2068 | int err; | |
2069 | nd.root.dentry = dentry; | |
2070 | nd.root.mnt = mnt; | |
2071 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2072 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2073 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2074 | if (!err) | |
2075 | *path = nd.path; | |
2076 | return err; | |
16f18200 JJS |
2077 | } |
2078 | ||
057f6c01 JM |
2079 | /* |
2080 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2081 | * needs parent already locked. Doesn't follow mounts. | |
2082 | * SMP-safe. | |
2083 | */ | |
eead1911 | 2084 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2085 | { |
72bd866a | 2086 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2087 | } |
2088 | ||
eead1911 | 2089 | /** |
a6b91919 | 2090 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2091 | * @name: pathname component to lookup |
2092 | * @base: base directory to lookup from | |
2093 | * @len: maximum length @len should be interpreted to | |
2094 | * | |
a6b91919 RD |
2095 | * Note that this routine is purely a helper for filesystem usage and should |
2096 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2097 | * nameidata argument is passed to the filesystem methods and a filesystem |
2098 | * using this helper needs to be prepared for that. | |
2099 | */ | |
057f6c01 JM |
2100 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2101 | { | |
057f6c01 | 2102 | struct qstr this; |
6a96ba54 | 2103 | unsigned int c; |
cda309de | 2104 | int err; |
057f6c01 | 2105 | |
2f9092e1 DW |
2106 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2107 | ||
6a96ba54 AV |
2108 | this.name = name; |
2109 | this.len = len; | |
0145acc2 | 2110 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2111 | if (!len) |
2112 | return ERR_PTR(-EACCES); | |
2113 | ||
21d8a15a AV |
2114 | if (unlikely(name[0] == '.')) { |
2115 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2116 | return ERR_PTR(-EACCES); | |
2117 | } | |
2118 | ||
6a96ba54 AV |
2119 | while (len--) { |
2120 | c = *(const unsigned char *)name++; | |
2121 | if (c == '/' || c == '\0') | |
2122 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2123 | } |
5a202bcd AV |
2124 | /* |
2125 | * See if the low-level filesystem might want | |
2126 | * to use its own hash.. | |
2127 | */ | |
2128 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2129 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2130 | if (err < 0) |
2131 | return ERR_PTR(err); | |
2132 | } | |
eead1911 | 2133 | |
cda309de MS |
2134 | err = inode_permission(base->d_inode, MAY_EXEC); |
2135 | if (err) | |
2136 | return ERR_PTR(err); | |
2137 | ||
72bd866a | 2138 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2139 | } |
2140 | ||
1fa1e7f6 AW |
2141 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2142 | struct path *path, int *empty) | |
1da177e4 | 2143 | { |
2d8f3038 | 2144 | struct nameidata nd; |
91a27b2a | 2145 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2146 | int err = PTR_ERR(tmp); |
1da177e4 | 2147 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2148 | |
2149 | BUG_ON(flags & LOOKUP_PARENT); | |
2150 | ||
873f1eed | 2151 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2152 | putname(tmp); |
2d8f3038 AV |
2153 | if (!err) |
2154 | *path = nd.path; | |
1da177e4 LT |
2155 | } |
2156 | return err; | |
2157 | } | |
2158 | ||
1fa1e7f6 AW |
2159 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2160 | struct path *path) | |
2161 | { | |
f7493e5d | 2162 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2163 | } |
2164 | ||
873f1eed JL |
2165 | /* |
2166 | * NB: most callers don't do anything directly with the reference to the | |
2167 | * to struct filename, but the nd->last pointer points into the name string | |
2168 | * allocated by getname. So we must hold the reference to it until all | |
2169 | * path-walking is complete. | |
2170 | */ | |
91a27b2a | 2171 | static struct filename * |
9e790bd6 JL |
2172 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2173 | unsigned int flags) | |
2ad94ae6 | 2174 | { |
91a27b2a | 2175 | struct filename *s = getname(path); |
2ad94ae6 AV |
2176 | int error; |
2177 | ||
9e790bd6 JL |
2178 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2179 | flags &= LOOKUP_REVAL; | |
2180 | ||
2ad94ae6 | 2181 | if (IS_ERR(s)) |
91a27b2a | 2182 | return s; |
2ad94ae6 | 2183 | |
9e790bd6 | 2184 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2185 | if (error) { |
2ad94ae6 | 2186 | putname(s); |
91a27b2a JL |
2187 | return ERR_PTR(error); |
2188 | } | |
2ad94ae6 | 2189 | |
91a27b2a | 2190 | return s; |
2ad94ae6 AV |
2191 | } |
2192 | ||
8033426e | 2193 | /** |
197df04c | 2194 | * mountpoint_last - look up last component for umount |
8033426e JL |
2195 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2196 | * @path: pointer to container for result | |
2197 | * | |
2198 | * This is a special lookup_last function just for umount. In this case, we | |
2199 | * need to resolve the path without doing any revalidation. | |
2200 | * | |
2201 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2202 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2203 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2204 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2205 | * bogus and it doesn't exist. | |
2206 | * | |
2207 | * Returns: | |
2208 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2209 | * lookup found a negative dentry. The nd->path reference will also be | |
2210 | * put in this case. | |
2211 | * | |
2212 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2213 | * symlink that needs to be followed. "path" will also be populated. | |
2214 | * The nd->path reference will also be put. | |
2215 | * | |
2216 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2217 | * that needs to be followed. "path" will be populated with the path | |
2218 | * to the link, and nd->path will *not* be put. | |
2219 | */ | |
2220 | static int | |
197df04c | 2221 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2222 | { |
2223 | int error = 0; | |
2224 | struct dentry *dentry; | |
2225 | struct dentry *dir = nd->path.dentry; | |
2226 | ||
35759521 AV |
2227 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2228 | if (nd->flags & LOOKUP_RCU) { | |
2229 | if (unlazy_walk(nd, NULL)) { | |
2230 | error = -ECHILD; | |
2231 | goto out; | |
2232 | } | |
8033426e JL |
2233 | } |
2234 | ||
2235 | nd->flags &= ~LOOKUP_PARENT; | |
2236 | ||
2237 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2238 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2239 | if (error) |
2240 | goto out; | |
2241 | dentry = dget(nd->path.dentry); | |
2242 | goto done; | |
8033426e JL |
2243 | } |
2244 | ||
2245 | mutex_lock(&dir->d_inode->i_mutex); | |
2246 | dentry = d_lookup(dir, &nd->last); | |
2247 | if (!dentry) { | |
2248 | /* | |
2249 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2250 | * so that means that this dentry is probably a symlink or the | |
2251 | * path doesn't actually point to a mounted dentry. | |
2252 | */ | |
2253 | dentry = d_alloc(dir, &nd->last); | |
2254 | if (!dentry) { | |
2255 | error = -ENOMEM; | |
bcceeeba | 2256 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2257 | goto out; |
8033426e | 2258 | } |
35759521 AV |
2259 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2260 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2261 | if (IS_ERR(dentry)) { |
2262 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2263 | goto out; |
bcceeeba | 2264 | } |
8033426e JL |
2265 | } |
2266 | mutex_unlock(&dir->d_inode->i_mutex); | |
2267 | ||
35759521 AV |
2268 | done: |
2269 | if (!dentry->d_inode) { | |
2270 | error = -ENOENT; | |
2271 | dput(dentry); | |
2272 | goto out; | |
8033426e | 2273 | } |
35759521 AV |
2274 | path->dentry = dentry; |
2275 | path->mnt = mntget(nd->path.mnt); | |
2276 | if (should_follow_link(dentry->d_inode, nd->flags & LOOKUP_FOLLOW)) | |
2277 | return 1; | |
2278 | follow_mount(path); | |
2279 | error = 0; | |
2280 | out: | |
8033426e JL |
2281 | terminate_walk(nd); |
2282 | return error; | |
2283 | } | |
2284 | ||
2285 | /** | |
197df04c | 2286 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2287 | * @dfd: directory file descriptor to start walk from |
2288 | * @name: full pathname to walk | |
2289 | * @flags: lookup flags | |
8033426e JL |
2290 | * |
2291 | * Look up the given name, but don't attempt to revalidate the last component. | |
2292 | * Returns 0 and "path" will be valid on success; Retuns error otherwise. | |
2293 | */ | |
2294 | static int | |
197df04c | 2295 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2296 | { |
2297 | struct file *base = NULL; | |
2298 | struct nameidata nd; | |
2299 | int err; | |
2300 | ||
2301 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2302 | if (unlikely(err)) | |
2303 | return err; | |
2304 | ||
2305 | current->total_link_count = 0; | |
2306 | err = link_path_walk(name, &nd); | |
2307 | if (err) | |
2308 | goto out; | |
2309 | ||
197df04c | 2310 | err = mountpoint_last(&nd, path); |
8033426e JL |
2311 | while (err > 0) { |
2312 | void *cookie; | |
2313 | struct path link = *path; | |
2314 | err = may_follow_link(&link, &nd); | |
2315 | if (unlikely(err)) | |
2316 | break; | |
2317 | nd.flags |= LOOKUP_PARENT; | |
2318 | err = follow_link(&link, &nd, &cookie); | |
2319 | if (err) | |
2320 | break; | |
197df04c | 2321 | err = mountpoint_last(&nd, path); |
8033426e JL |
2322 | put_link(&nd, &link, cookie); |
2323 | } | |
2324 | out: | |
2325 | if (base) | |
2326 | fput(base); | |
2327 | ||
2328 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2329 | path_put(&nd.root); | |
2330 | ||
2331 | return err; | |
2332 | } | |
2333 | ||
2d864651 AV |
2334 | static int |
2335 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2336 | unsigned int flags) | |
2337 | { | |
2338 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2339 | if (unlikely(error == -ECHILD)) | |
2340 | error = path_mountpoint(dfd, s->name, path, flags); | |
2341 | if (unlikely(error == -ESTALE)) | |
2342 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2343 | if (likely(!error)) | |
2344 | audit_inode(s, path->dentry, 0); | |
2345 | return error; | |
2346 | } | |
2347 | ||
8033426e | 2348 | /** |
197df04c | 2349 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2350 | * @dfd: directory file descriptor |
2351 | * @name: pathname from userland | |
2352 | * @flags: lookup flags | |
2353 | * @path: pointer to container to hold result | |
2354 | * | |
2355 | * A umount is a special case for path walking. We're not actually interested | |
2356 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2357 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2358 | * and avoid revalidating the last component. | |
2359 | * | |
2360 | * Returns 0 and populates "path" on success. | |
2361 | */ | |
2362 | int | |
197df04c | 2363 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2364 | struct path *path) |
2365 | { | |
2366 | struct filename *s = getname(name); | |
2367 | int error; | |
8033426e JL |
2368 | if (IS_ERR(s)) |
2369 | return PTR_ERR(s); | |
2d864651 | 2370 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2371 | putname(s); |
2372 | return error; | |
2373 | } | |
2374 | ||
2d864651 AV |
2375 | int |
2376 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2377 | unsigned int flags) | |
2378 | { | |
2379 | struct filename s = {.name = name}; | |
2380 | return filename_mountpoint(dfd, &s, path, flags); | |
2381 | } | |
2382 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2383 | ||
1da177e4 LT |
2384 | /* |
2385 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2386 | * minimal. | |
2387 | */ | |
2388 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2389 | { | |
8e96e3b7 | 2390 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2391 | |
1da177e4 LT |
2392 | if (!(dir->i_mode & S_ISVTX)) |
2393 | return 0; | |
8e96e3b7 | 2394 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2395 | return 0; |
8e96e3b7 | 2396 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2397 | return 0; |
1a48e2ac | 2398 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2399 | } |
2400 | ||
2401 | /* | |
2402 | * Check whether we can remove a link victim from directory dir, check | |
2403 | * whether the type of victim is right. | |
2404 | * 1. We can't do it if dir is read-only (done in permission()) | |
2405 | * 2. We should have write and exec permissions on dir | |
2406 | * 3. We can't remove anything from append-only dir | |
2407 | * 4. We can't do anything with immutable dir (done in permission()) | |
2408 | * 5. If the sticky bit on dir is set we should either | |
2409 | * a. be owner of dir, or | |
2410 | * b. be owner of victim, or | |
2411 | * c. have CAP_FOWNER capability | |
2412 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2413 | * links pointing to it. | |
2414 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2415 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2416 | * 9. We can't remove a root or mountpoint. | |
2417 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2418 | * nfs_async_unlink(). | |
2419 | */ | |
858119e1 | 2420 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1da177e4 LT |
2421 | { |
2422 | int error; | |
2423 | ||
2424 | if (!victim->d_inode) | |
2425 | return -ENOENT; | |
2426 | ||
2427 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2428 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2429 | |
f419a2e3 | 2430 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2431 | if (error) |
2432 | return error; | |
2433 | if (IS_APPEND(dir)) | |
2434 | return -EPERM; | |
2435 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
f9454548 | 2436 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1da177e4 LT |
2437 | return -EPERM; |
2438 | if (isdir) { | |
2439 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
2440 | return -ENOTDIR; | |
2441 | if (IS_ROOT(victim)) | |
2442 | return -EBUSY; | |
2443 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
2444 | return -EISDIR; | |
2445 | if (IS_DEADDIR(dir)) | |
2446 | return -ENOENT; | |
2447 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2448 | return -EBUSY; | |
2449 | return 0; | |
2450 | } | |
2451 | ||
2452 | /* Check whether we can create an object with dentry child in directory | |
2453 | * dir. | |
2454 | * 1. We can't do it if child already exists (open has special treatment for | |
2455 | * this case, but since we are inlined it's OK) | |
2456 | * 2. We can't do it if dir is read-only (done in permission()) | |
2457 | * 3. We should have write and exec permissions on dir | |
2458 | * 4. We can't do it if dir is immutable (done in permission()) | |
2459 | */ | |
a95164d9 | 2460 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 LT |
2461 | { |
2462 | if (child->d_inode) | |
2463 | return -EEXIST; | |
2464 | if (IS_DEADDIR(dir)) | |
2465 | return -ENOENT; | |
f419a2e3 | 2466 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2467 | } |
2468 | ||
1da177e4 LT |
2469 | /* |
2470 | * p1 and p2 should be directories on the same fs. | |
2471 | */ | |
2472 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2473 | { | |
2474 | struct dentry *p; | |
2475 | ||
2476 | if (p1 == p2) { | |
f2eace23 | 2477 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2478 | return NULL; |
2479 | } | |
2480 | ||
a11f3a05 | 2481 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2482 | |
e2761a11 OH |
2483 | p = d_ancestor(p2, p1); |
2484 | if (p) { | |
2485 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2486 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2487 | return p; | |
1da177e4 LT |
2488 | } |
2489 | ||
e2761a11 OH |
2490 | p = d_ancestor(p1, p2); |
2491 | if (p) { | |
2492 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2493 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2494 | return p; | |
1da177e4 LT |
2495 | } |
2496 | ||
f2eace23 IM |
2497 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2498 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2499 | return NULL; |
2500 | } | |
2501 | ||
2502 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2503 | { | |
1b1dcc1b | 2504 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2505 | if (p1 != p2) { |
1b1dcc1b | 2506 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2507 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2508 | } |
2509 | } | |
2510 | ||
4acdaf27 | 2511 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2512 | bool want_excl) |
1da177e4 | 2513 | { |
a95164d9 | 2514 | int error = may_create(dir, dentry); |
1da177e4 LT |
2515 | if (error) |
2516 | return error; | |
2517 | ||
acfa4380 | 2518 | if (!dir->i_op->create) |
1da177e4 LT |
2519 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2520 | mode &= S_IALLUGO; | |
2521 | mode |= S_IFREG; | |
2522 | error = security_inode_create(dir, dentry, mode); | |
2523 | if (error) | |
2524 | return error; | |
312b63fb | 2525 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2526 | if (!error) |
f38aa942 | 2527 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2528 | return error; |
2529 | } | |
2530 | ||
73d049a4 | 2531 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2532 | { |
3fb64190 | 2533 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2534 | struct inode *inode = dentry->d_inode; |
2535 | int error; | |
2536 | ||
bcda7652 AV |
2537 | /* O_PATH? */ |
2538 | if (!acc_mode) | |
2539 | return 0; | |
2540 | ||
1da177e4 LT |
2541 | if (!inode) |
2542 | return -ENOENT; | |
2543 | ||
c8fe8f30 CH |
2544 | switch (inode->i_mode & S_IFMT) { |
2545 | case S_IFLNK: | |
1da177e4 | 2546 | return -ELOOP; |
c8fe8f30 CH |
2547 | case S_IFDIR: |
2548 | if (acc_mode & MAY_WRITE) | |
2549 | return -EISDIR; | |
2550 | break; | |
2551 | case S_IFBLK: | |
2552 | case S_IFCHR: | |
3fb64190 | 2553 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2554 | return -EACCES; |
c8fe8f30 CH |
2555 | /*FALLTHRU*/ |
2556 | case S_IFIFO: | |
2557 | case S_IFSOCK: | |
1da177e4 | 2558 | flag &= ~O_TRUNC; |
c8fe8f30 | 2559 | break; |
4a3fd211 | 2560 | } |
b41572e9 | 2561 | |
3fb64190 | 2562 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2563 | if (error) |
2564 | return error; | |
6146f0d5 | 2565 | |
1da177e4 LT |
2566 | /* |
2567 | * An append-only file must be opened in append mode for writing. | |
2568 | */ | |
2569 | if (IS_APPEND(inode)) { | |
8737c930 | 2570 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2571 | return -EPERM; |
1da177e4 | 2572 | if (flag & O_TRUNC) |
7715b521 | 2573 | return -EPERM; |
1da177e4 LT |
2574 | } |
2575 | ||
2576 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2577 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2578 | return -EPERM; |
1da177e4 | 2579 | |
f3c7691e | 2580 | return 0; |
7715b521 | 2581 | } |
1da177e4 | 2582 | |
e1181ee6 | 2583 | static int handle_truncate(struct file *filp) |
7715b521 | 2584 | { |
e1181ee6 | 2585 | struct path *path = &filp->f_path; |
7715b521 AV |
2586 | struct inode *inode = path->dentry->d_inode; |
2587 | int error = get_write_access(inode); | |
2588 | if (error) | |
2589 | return error; | |
2590 | /* | |
2591 | * Refuse to truncate files with mandatory locks held on them. | |
2592 | */ | |
2593 | error = locks_verify_locked(inode); | |
2594 | if (!error) | |
ea0d3ab2 | 2595 | error = security_path_truncate(path); |
7715b521 AV |
2596 | if (!error) { |
2597 | error = do_truncate(path->dentry, 0, | |
2598 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2599 | filp); |
7715b521 AV |
2600 | } |
2601 | put_write_access(inode); | |
acd0c935 | 2602 | return error; |
1da177e4 LT |
2603 | } |
2604 | ||
d57999e1 DH |
2605 | static inline int open_to_namei_flags(int flag) |
2606 | { | |
8a5e929d AV |
2607 | if ((flag & O_ACCMODE) == 3) |
2608 | flag--; | |
d57999e1 DH |
2609 | return flag; |
2610 | } | |
2611 | ||
d18e9008 MS |
2612 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2613 | { | |
2614 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2615 | if (error) | |
2616 | return error; | |
2617 | ||
2618 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2619 | if (error) | |
2620 | return error; | |
2621 | ||
2622 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2623 | } | |
2624 | ||
1acf0af9 DH |
2625 | /* |
2626 | * Attempt to atomically look up, create and open a file from a negative | |
2627 | * dentry. | |
2628 | * | |
2629 | * Returns 0 if successful. The file will have been created and attached to | |
2630 | * @file by the filesystem calling finish_open(). | |
2631 | * | |
2632 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2633 | * caller will need to perform the open themselves. @path will have been | |
2634 | * updated to point to the new dentry. This may be negative. | |
2635 | * | |
2636 | * Returns an error code otherwise. | |
2637 | */ | |
2675a4eb AV |
2638 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2639 | struct path *path, struct file *file, | |
2640 | const struct open_flags *op, | |
64894cf8 | 2641 | bool got_write, bool need_lookup, |
2675a4eb | 2642 | int *opened) |
d18e9008 MS |
2643 | { |
2644 | struct inode *dir = nd->path.dentry->d_inode; | |
2645 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2646 | umode_t mode; | |
2647 | int error; | |
2648 | int acc_mode; | |
d18e9008 MS |
2649 | int create_error = 0; |
2650 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2651 | bool excl; |
d18e9008 MS |
2652 | |
2653 | BUG_ON(dentry->d_inode); | |
2654 | ||
2655 | /* Don't create child dentry for a dead directory. */ | |
2656 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2657 | error = -ENOENT; |
d18e9008 MS |
2658 | goto out; |
2659 | } | |
2660 | ||
62b259d8 | 2661 | mode = op->mode; |
d18e9008 MS |
2662 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2663 | mode &= ~current_umask(); | |
2664 | ||
116cc022 MS |
2665 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2666 | if (excl) | |
d18e9008 | 2667 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2668 | |
2669 | /* | |
2670 | * Checking write permission is tricky, bacuse we don't know if we are | |
2671 | * going to actually need it: O_CREAT opens should work as long as the | |
2672 | * file exists. But checking existence breaks atomicity. The trick is | |
2673 | * to check access and if not granted clear O_CREAT from the flags. | |
2674 | * | |
2675 | * Another problem is returing the "right" error value (e.g. for an | |
2676 | * O_EXCL open we want to return EEXIST not EROFS). | |
2677 | */ | |
64894cf8 AV |
2678 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2679 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2680 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2681 | /* |
2682 | * No O_CREATE -> atomicity not a requirement -> fall | |
2683 | * back to lookup + open | |
2684 | */ | |
2685 | goto no_open; | |
2686 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2687 | /* Fall back and fail with the right error */ | |
64894cf8 | 2688 | create_error = -EROFS; |
d18e9008 MS |
2689 | goto no_open; |
2690 | } else { | |
2691 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2692 | create_error = -EROFS; |
d18e9008 MS |
2693 | open_flag &= ~O_CREAT; |
2694 | } | |
2695 | } | |
2696 | ||
2697 | if (open_flag & O_CREAT) { | |
38227f78 | 2698 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2699 | if (error) { |
2700 | create_error = error; | |
2701 | if (open_flag & O_EXCL) | |
2702 | goto no_open; | |
2703 | open_flag &= ~O_CREAT; | |
2704 | } | |
2705 | } | |
2706 | ||
2707 | if (nd->flags & LOOKUP_DIRECTORY) | |
2708 | open_flag |= O_DIRECTORY; | |
2709 | ||
30d90494 AV |
2710 | file->f_path.dentry = DENTRY_NOT_SET; |
2711 | file->f_path.mnt = nd->path.mnt; | |
2712 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2713 | opened); |
d9585277 | 2714 | if (error < 0) { |
d9585277 AV |
2715 | if (create_error && error == -ENOENT) |
2716 | error = create_error; | |
d18e9008 MS |
2717 | goto out; |
2718 | } | |
2719 | ||
d9585277 | 2720 | if (error) { /* returned 1, that is */ |
30d90494 | 2721 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2722 | error = -EIO; |
d18e9008 MS |
2723 | goto out; |
2724 | } | |
30d90494 | 2725 | if (file->f_path.dentry) { |
d18e9008 | 2726 | dput(dentry); |
30d90494 | 2727 | dentry = file->f_path.dentry; |
d18e9008 | 2728 | } |
03da633a AV |
2729 | if (*opened & FILE_CREATED) |
2730 | fsnotify_create(dir, dentry); | |
2731 | if (!dentry->d_inode) { | |
2732 | WARN_ON(*opened & FILE_CREATED); | |
2733 | if (create_error) { | |
2734 | error = create_error; | |
2735 | goto out; | |
2736 | } | |
2737 | } else { | |
2738 | if (excl && !(*opened & FILE_CREATED)) { | |
2739 | error = -EEXIST; | |
2740 | goto out; | |
2741 | } | |
62b2ce96 | 2742 | } |
d18e9008 MS |
2743 | goto looked_up; |
2744 | } | |
2745 | ||
2746 | /* | |
2747 | * We didn't have the inode before the open, so check open permission | |
2748 | * here. | |
2749 | */ | |
03da633a AV |
2750 | acc_mode = op->acc_mode; |
2751 | if (*opened & FILE_CREATED) { | |
2752 | WARN_ON(!(open_flag & O_CREAT)); | |
2753 | fsnotify_create(dir, dentry); | |
2754 | acc_mode = MAY_OPEN; | |
2755 | } | |
2675a4eb AV |
2756 | error = may_open(&file->f_path, acc_mode, open_flag); |
2757 | if (error) | |
2758 | fput(file); | |
d18e9008 MS |
2759 | |
2760 | out: | |
2761 | dput(dentry); | |
2675a4eb | 2762 | return error; |
d18e9008 | 2763 | |
d18e9008 MS |
2764 | no_open: |
2765 | if (need_lookup) { | |
72bd866a | 2766 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2767 | if (IS_ERR(dentry)) |
2675a4eb | 2768 | return PTR_ERR(dentry); |
d18e9008 MS |
2769 | |
2770 | if (create_error) { | |
2771 | int open_flag = op->open_flag; | |
2772 | ||
2675a4eb | 2773 | error = create_error; |
d18e9008 MS |
2774 | if ((open_flag & O_EXCL)) { |
2775 | if (!dentry->d_inode) | |
2776 | goto out; | |
2777 | } else if (!dentry->d_inode) { | |
2778 | goto out; | |
2779 | } else if ((open_flag & O_TRUNC) && | |
2780 | S_ISREG(dentry->d_inode->i_mode)) { | |
2781 | goto out; | |
2782 | } | |
2783 | /* will fail later, go on to get the right error */ | |
2784 | } | |
2785 | } | |
2786 | looked_up: | |
2787 | path->dentry = dentry; | |
2788 | path->mnt = nd->path.mnt; | |
2675a4eb | 2789 | return 1; |
d18e9008 MS |
2790 | } |
2791 | ||
d58ffd35 | 2792 | /* |
1acf0af9 | 2793 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2794 | * |
2795 | * Must be called with i_mutex held on parent. | |
2796 | * | |
1acf0af9 DH |
2797 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2798 | * opened. In this case the file will be returned attached to @file. | |
2799 | * | |
2800 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2801 | * and creations will have been performed and the dentry returned in @path will | |
2802 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2803 | * specified then a negative dentry may be returned. | |
2804 | * | |
2805 | * An error code is returned otherwise. | |
2806 | * | |
2807 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2808 | * cleared otherwise prior to returning. | |
d58ffd35 | 2809 | */ |
2675a4eb AV |
2810 | static int lookup_open(struct nameidata *nd, struct path *path, |
2811 | struct file *file, | |
2812 | const struct open_flags *op, | |
64894cf8 | 2813 | bool got_write, int *opened) |
d58ffd35 MS |
2814 | { |
2815 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2816 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2817 | struct dentry *dentry; |
2818 | int error; | |
54ef4872 | 2819 | bool need_lookup; |
d58ffd35 | 2820 | |
47237687 | 2821 | *opened &= ~FILE_CREATED; |
201f956e | 2822 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2823 | if (IS_ERR(dentry)) |
2675a4eb | 2824 | return PTR_ERR(dentry); |
d58ffd35 | 2825 | |
d18e9008 MS |
2826 | /* Cached positive dentry: will open in f_op->open */ |
2827 | if (!need_lookup && dentry->d_inode) | |
2828 | goto out_no_open; | |
2829 | ||
2830 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2831 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2832 | need_lookup, opened); |
d18e9008 MS |
2833 | } |
2834 | ||
54ef4872 MS |
2835 | if (need_lookup) { |
2836 | BUG_ON(dentry->d_inode); | |
2837 | ||
72bd866a | 2838 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2839 | if (IS_ERR(dentry)) |
2675a4eb | 2840 | return PTR_ERR(dentry); |
54ef4872 MS |
2841 | } |
2842 | ||
d58ffd35 MS |
2843 | /* Negative dentry, just create the file */ |
2844 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2845 | umode_t mode = op->mode; | |
2846 | if (!IS_POSIXACL(dir->d_inode)) | |
2847 | mode &= ~current_umask(); | |
2848 | /* | |
2849 | * This write is needed to ensure that a | |
2850 | * rw->ro transition does not occur between | |
2851 | * the time when the file is created and when | |
2852 | * a permanent write count is taken through | |
015c3bbc | 2853 | * the 'struct file' in finish_open(). |
d58ffd35 | 2854 | */ |
64894cf8 AV |
2855 | if (!got_write) { |
2856 | error = -EROFS; | |
d58ffd35 | 2857 | goto out_dput; |
64894cf8 | 2858 | } |
47237687 | 2859 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2860 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2861 | if (error) | |
2862 | goto out_dput; | |
312b63fb AV |
2863 | error = vfs_create(dir->d_inode, dentry, mode, |
2864 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2865 | if (error) |
2866 | goto out_dput; | |
2867 | } | |
d18e9008 | 2868 | out_no_open: |
d58ffd35 MS |
2869 | path->dentry = dentry; |
2870 | path->mnt = nd->path.mnt; | |
2675a4eb | 2871 | return 1; |
d58ffd35 MS |
2872 | |
2873 | out_dput: | |
2874 | dput(dentry); | |
2675a4eb | 2875 | return error; |
d58ffd35 MS |
2876 | } |
2877 | ||
31e6b01f | 2878 | /* |
fe2d35ff | 2879 | * Handle the last step of open() |
31e6b01f | 2880 | */ |
2675a4eb AV |
2881 | static int do_last(struct nameidata *nd, struct path *path, |
2882 | struct file *file, const struct open_flags *op, | |
669abf4e | 2883 | int *opened, struct filename *name) |
fb1cc555 | 2884 | { |
a1e28038 | 2885 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2886 | int open_flag = op->open_flag; |
77d660a8 | 2887 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2888 | bool got_write = false; |
bcda7652 | 2889 | int acc_mode = op->acc_mode; |
a1eb3315 | 2890 | struct inode *inode; |
77d660a8 | 2891 | bool symlink_ok = false; |
16b1c1cd MS |
2892 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2893 | bool retried = false; | |
16c2cd71 | 2894 | int error; |
1f36f774 | 2895 | |
c3e380b0 AV |
2896 | nd->flags &= ~LOOKUP_PARENT; |
2897 | nd->flags |= op->intent; | |
2898 | ||
bc77daa7 | 2899 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2900 | error = handle_dots(nd, nd->last_type); |
2901 | if (error) | |
2675a4eb | 2902 | return error; |
e83db167 | 2903 | goto finish_open; |
1f36f774 | 2904 | } |
67ee3ad2 | 2905 | |
ca344a89 | 2906 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2907 | if (nd->last.name[nd->last.len]) |
2908 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2909 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2910 | symlink_ok = true; |
fe2d35ff | 2911 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2912 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2913 | if (likely(!error)) |
2914 | goto finish_lookup; | |
2915 | ||
2916 | if (error < 0) | |
2675a4eb | 2917 | goto out; |
71574865 MS |
2918 | |
2919 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2920 | } else { |
2921 | /* create side of things */ | |
2922 | /* | |
2923 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2924 | * has been cleared when we got to the last component we are | |
2925 | * about to look up | |
2926 | */ | |
2927 | error = complete_walk(nd); | |
2928 | if (error) | |
2675a4eb | 2929 | return error; |
fe2d35ff | 2930 | |
33e2208a | 2931 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2932 | error = -EISDIR; |
2933 | /* trailing slashes? */ | |
2934 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2935 | goto out; |
b6183df7 | 2936 | } |
a2c36b45 | 2937 | |
16b1c1cd | 2938 | retry_lookup: |
64894cf8 AV |
2939 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2940 | error = mnt_want_write(nd->path.mnt); | |
2941 | if (!error) | |
2942 | got_write = true; | |
2943 | /* | |
2944 | * do _not_ fail yet - we might not need that or fail with | |
2945 | * a different error; let lookup_open() decide; we'll be | |
2946 | * dropping this one anyway. | |
2947 | */ | |
2948 | } | |
a1e28038 | 2949 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2950 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2951 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2952 | |
2675a4eb AV |
2953 | if (error <= 0) { |
2954 | if (error) | |
d18e9008 MS |
2955 | goto out; |
2956 | ||
47237687 | 2957 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2958 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2959 | will_truncate = false; |
d18e9008 | 2960 | |
adb5c247 | 2961 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2962 | goto opened; |
2963 | } | |
fb1cc555 | 2964 | |
47237687 | 2965 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2966 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2967 | open_flag &= ~O_TRUNC; |
77d660a8 | 2968 | will_truncate = false; |
bcda7652 | 2969 | acc_mode = MAY_OPEN; |
d58ffd35 | 2970 | path_to_nameidata(path, nd); |
e83db167 | 2971 | goto finish_open_created; |
fb1cc555 AV |
2972 | } |
2973 | ||
2974 | /* | |
3134f37e | 2975 | * create/update audit record if it already exists. |
fb1cc555 | 2976 | */ |
3134f37e | 2977 | if (path->dentry->d_inode) |
adb5c247 | 2978 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2979 | |
d18e9008 MS |
2980 | /* |
2981 | * If atomic_open() acquired write access it is dropped now due to | |
2982 | * possible mount and symlink following (this might be optimized away if | |
2983 | * necessary...) | |
2984 | */ | |
64894cf8 | 2985 | if (got_write) { |
d18e9008 | 2986 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2987 | got_write = false; |
d18e9008 MS |
2988 | } |
2989 | ||
fb1cc555 | 2990 | error = -EEXIST; |
f8310c59 | 2991 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2992 | goto exit_dput; |
2993 | ||
9875cf80 DH |
2994 | error = follow_managed(path, nd->flags); |
2995 | if (error < 0) | |
2996 | goto exit_dput; | |
fb1cc555 | 2997 | |
a3fbbde7 AV |
2998 | if (error) |
2999 | nd->flags |= LOOKUP_JUMPED; | |
3000 | ||
decf3400 MS |
3001 | BUG_ON(nd->flags & LOOKUP_RCU); |
3002 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
3003 | finish_lookup: |
3004 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 3005 | error = -ENOENT; |
54c33e7f MS |
3006 | if (!inode) { |
3007 | path_to_nameidata(path, nd); | |
2675a4eb | 3008 | goto out; |
54c33e7f | 3009 | } |
9e67f361 | 3010 | |
d45ea867 MS |
3011 | if (should_follow_link(inode, !symlink_ok)) { |
3012 | if (nd->flags & LOOKUP_RCU) { | |
3013 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
3014 | error = -ECHILD; | |
2675a4eb | 3015 | goto out; |
d45ea867 MS |
3016 | } |
3017 | } | |
3018 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 3019 | return 1; |
d45ea867 | 3020 | } |
fb1cc555 | 3021 | |
16b1c1cd MS |
3022 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
3023 | path_to_nameidata(path, nd); | |
3024 | } else { | |
3025 | save_parent.dentry = nd->path.dentry; | |
3026 | save_parent.mnt = mntget(path->mnt); | |
3027 | nd->path.dentry = path->dentry; | |
3028 | ||
3029 | } | |
decf3400 | 3030 | nd->inode = inode; |
a3fbbde7 | 3031 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3032 | finish_open: |
a3fbbde7 | 3033 | error = complete_walk(nd); |
16b1c1cd MS |
3034 | if (error) { |
3035 | path_put(&save_parent); | |
2675a4eb | 3036 | return error; |
16b1c1cd | 3037 | } |
bc77daa7 | 3038 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3039 | error = -EISDIR; |
050ac841 | 3040 | if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) |
2675a4eb | 3041 | goto out; |
af2f5542 | 3042 | error = -ENOTDIR; |
05252901 | 3043 | if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode)) |
2675a4eb | 3044 | goto out; |
6c0d46c4 | 3045 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3046 | will_truncate = false; |
6c0d46c4 | 3047 | |
0f9d1a10 AV |
3048 | if (will_truncate) { |
3049 | error = mnt_want_write(nd->path.mnt); | |
3050 | if (error) | |
2675a4eb | 3051 | goto out; |
64894cf8 | 3052 | got_write = true; |
0f9d1a10 | 3053 | } |
e83db167 | 3054 | finish_open_created: |
bcda7652 | 3055 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3056 | if (error) |
2675a4eb | 3057 | goto out; |
30d90494 AV |
3058 | file->f_path.mnt = nd->path.mnt; |
3059 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3060 | if (error) { | |
30d90494 | 3061 | if (error == -EOPENSTALE) |
f60dc3db | 3062 | goto stale_open; |
015c3bbc | 3063 | goto out; |
f60dc3db | 3064 | } |
a8277b9b | 3065 | opened: |
2675a4eb | 3066 | error = open_check_o_direct(file); |
015c3bbc MS |
3067 | if (error) |
3068 | goto exit_fput; | |
2675a4eb | 3069 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3070 | if (error) |
3071 | goto exit_fput; | |
3072 | ||
3073 | if (will_truncate) { | |
2675a4eb | 3074 | error = handle_truncate(file); |
aa4caadb MS |
3075 | if (error) |
3076 | goto exit_fput; | |
0f9d1a10 | 3077 | } |
ca344a89 | 3078 | out: |
64894cf8 | 3079 | if (got_write) |
0f9d1a10 | 3080 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3081 | path_put(&save_parent); |
e276ae67 | 3082 | terminate_walk(nd); |
2675a4eb | 3083 | return error; |
fb1cc555 | 3084 | |
fb1cc555 AV |
3085 | exit_dput: |
3086 | path_put_conditional(path, nd); | |
ca344a89 | 3087 | goto out; |
015c3bbc | 3088 | exit_fput: |
2675a4eb AV |
3089 | fput(file); |
3090 | goto out; | |
015c3bbc | 3091 | |
f60dc3db MS |
3092 | stale_open: |
3093 | /* If no saved parent or already retried then can't retry */ | |
3094 | if (!save_parent.dentry || retried) | |
3095 | goto out; | |
3096 | ||
3097 | BUG_ON(save_parent.dentry != dir); | |
3098 | path_put(&nd->path); | |
3099 | nd->path = save_parent; | |
3100 | nd->inode = dir->d_inode; | |
3101 | save_parent.mnt = NULL; | |
3102 | save_parent.dentry = NULL; | |
64894cf8 | 3103 | if (got_write) { |
f60dc3db | 3104 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3105 | got_write = false; |
f60dc3db MS |
3106 | } |
3107 | retried = true; | |
3108 | goto retry_lookup; | |
fb1cc555 AV |
3109 | } |
3110 | ||
60545d0d AV |
3111 | static int do_tmpfile(int dfd, struct filename *pathname, |
3112 | struct nameidata *nd, int flags, | |
3113 | const struct open_flags *op, | |
3114 | struct file *file, int *opened) | |
3115 | { | |
3116 | static const struct qstr name = QSTR_INIT("/", 1); | |
3117 | struct dentry *dentry, *child; | |
3118 | struct inode *dir; | |
3119 | int error = path_lookupat(dfd, pathname->name, | |
3120 | flags | LOOKUP_DIRECTORY, nd); | |
3121 | if (unlikely(error)) | |
3122 | return error; | |
3123 | error = mnt_want_write(nd->path.mnt); | |
3124 | if (unlikely(error)) | |
3125 | goto out; | |
3126 | /* we want directory to be writable */ | |
3127 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3128 | if (error) | |
3129 | goto out2; | |
3130 | dentry = nd->path.dentry; | |
3131 | dir = dentry->d_inode; | |
3132 | if (!dir->i_op->tmpfile) { | |
3133 | error = -EOPNOTSUPP; | |
3134 | goto out2; | |
3135 | } | |
3136 | child = d_alloc(dentry, &name); | |
3137 | if (unlikely(!child)) { | |
3138 | error = -ENOMEM; | |
3139 | goto out2; | |
3140 | } | |
3141 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3142 | nd->flags |= op->intent; | |
3143 | dput(nd->path.dentry); | |
3144 | nd->path.dentry = child; | |
3145 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3146 | if (error) | |
3147 | goto out2; | |
3148 | audit_inode(pathname, nd->path.dentry, 0); | |
3149 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3150 | if (error) | |
3151 | goto out2; | |
3152 | file->f_path.mnt = nd->path.mnt; | |
3153 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3154 | if (error) | |
3155 | goto out2; | |
3156 | error = open_check_o_direct(file); | |
f4e0c30c | 3157 | if (error) { |
60545d0d | 3158 | fput(file); |
f4e0c30c AV |
3159 | } else if (!(op->open_flag & O_EXCL)) { |
3160 | struct inode *inode = file_inode(file); | |
3161 | spin_lock(&inode->i_lock); | |
3162 | inode->i_state |= I_LINKABLE; | |
3163 | spin_unlock(&inode->i_lock); | |
3164 | } | |
60545d0d AV |
3165 | out2: |
3166 | mnt_drop_write(nd->path.mnt); | |
3167 | out: | |
3168 | path_put(&nd->path); | |
3169 | return error; | |
3170 | } | |
3171 | ||
669abf4e | 3172 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3173 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3174 | { |
fe2d35ff | 3175 | struct file *base = NULL; |
30d90494 | 3176 | struct file *file; |
9850c056 | 3177 | struct path path; |
47237687 | 3178 | int opened = 0; |
13aab428 | 3179 | int error; |
31e6b01f | 3180 | |
30d90494 | 3181 | file = get_empty_filp(); |
1afc99be AV |
3182 | if (IS_ERR(file)) |
3183 | return file; | |
31e6b01f | 3184 | |
30d90494 | 3185 | file->f_flags = op->open_flag; |
31e6b01f | 3186 | |
bb458c64 | 3187 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3188 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3189 | goto out; | |
3190 | } | |
3191 | ||
669abf4e | 3192 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3193 | if (unlikely(error)) |
2675a4eb | 3194 | goto out; |
31e6b01f | 3195 | |
fe2d35ff | 3196 | current->total_link_count = 0; |
669abf4e | 3197 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3198 | if (unlikely(error)) |
2675a4eb | 3199 | goto out; |
1da177e4 | 3200 | |
2675a4eb AV |
3201 | error = do_last(nd, &path, file, op, &opened, pathname); |
3202 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3203 | struct path link = path; |
def4af30 | 3204 | void *cookie; |
574197e0 | 3205 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3206 | path_put_conditional(&path, nd); |
3207 | path_put(&nd->path); | |
2675a4eb | 3208 | error = -ELOOP; |
40b39136 AV |
3209 | break; |
3210 | } | |
800179c9 KC |
3211 | error = may_follow_link(&link, nd); |
3212 | if (unlikely(error)) | |
3213 | break; | |
73d049a4 AV |
3214 | nd->flags |= LOOKUP_PARENT; |
3215 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3216 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3217 | if (unlikely(error)) |
2675a4eb AV |
3218 | break; |
3219 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3220 | put_link(nd, &link, cookie); |
806b681c | 3221 | } |
10fa8e62 | 3222 | out: |
73d049a4 AV |
3223 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3224 | path_put(&nd->root); | |
fe2d35ff AV |
3225 | if (base) |
3226 | fput(base); | |
2675a4eb AV |
3227 | if (!(opened & FILE_OPENED)) { |
3228 | BUG_ON(!error); | |
30d90494 | 3229 | put_filp(file); |
16b1c1cd | 3230 | } |
2675a4eb AV |
3231 | if (unlikely(error)) { |
3232 | if (error == -EOPENSTALE) { | |
3233 | if (flags & LOOKUP_RCU) | |
3234 | error = -ECHILD; | |
3235 | else | |
3236 | error = -ESTALE; | |
3237 | } | |
3238 | file = ERR_PTR(error); | |
3239 | } | |
3240 | return file; | |
1da177e4 LT |
3241 | } |
3242 | ||
669abf4e | 3243 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3244 | const struct open_flags *op) |
13aab428 | 3245 | { |
73d049a4 | 3246 | struct nameidata nd; |
f9652e10 | 3247 | int flags = op->lookup_flags; |
13aab428 AV |
3248 | struct file *filp; |
3249 | ||
73d049a4 | 3250 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3251 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3252 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3253 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3254 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3255 | return filp; |
3256 | } | |
3257 | ||
73d049a4 | 3258 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3259 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3260 | { |
3261 | struct nameidata nd; | |
3262 | struct file *file; | |
669abf4e | 3263 | struct filename filename = { .name = name }; |
f9652e10 | 3264 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3265 | |
3266 | nd.root.mnt = mnt; | |
3267 | nd.root.dentry = dentry; | |
3268 | ||
bcda7652 | 3269 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3270 | return ERR_PTR(-ELOOP); |
3271 | ||
669abf4e | 3272 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3273 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3274 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3275 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3276 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3277 | return file; |
3278 | } | |
3279 | ||
1ac12b4b JL |
3280 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3281 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3282 | { |
c663e5d8 | 3283 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3284 | struct nameidata nd; |
c30dabfe | 3285 | int err2; |
1ac12b4b JL |
3286 | int error; |
3287 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3288 | ||
3289 | /* | |
3290 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3291 | * other flags passed in are ignored! | |
3292 | */ | |
3293 | lookup_flags &= LOOKUP_REVAL; | |
3294 | ||
3295 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3296 | if (error) |
3297 | return ERR_PTR(error); | |
1da177e4 | 3298 | |
c663e5d8 CH |
3299 | /* |
3300 | * Yucky last component or no last component at all? | |
3301 | * (foo/., foo/.., /////) | |
3302 | */ | |
ed75e95d AV |
3303 | if (nd.last_type != LAST_NORM) |
3304 | goto out; | |
3305 | nd.flags &= ~LOOKUP_PARENT; | |
3306 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3307 | |
c30dabfe JK |
3308 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3309 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3310 | /* |
3311 | * Do the final lookup. | |
3312 | */ | |
ed75e95d AV |
3313 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3314 | dentry = lookup_hash(&nd); | |
1da177e4 | 3315 | if (IS_ERR(dentry)) |
a8104a9f | 3316 | goto unlock; |
c663e5d8 | 3317 | |
a8104a9f | 3318 | error = -EEXIST; |
e9baf6e5 | 3319 | if (dentry->d_inode) |
a8104a9f | 3320 | goto fail; |
c663e5d8 CH |
3321 | /* |
3322 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3323 | * From the vfs_mknod() POV we just have a negative dentry - | |
3324 | * all is fine. Let's be bastards - you had / on the end, you've | |
3325 | * been asking for (non-existent) directory. -ENOENT for you. | |
3326 | */ | |
ed75e95d | 3327 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3328 | error = -ENOENT; |
ed75e95d | 3329 | goto fail; |
e9baf6e5 | 3330 | } |
c30dabfe JK |
3331 | if (unlikely(err2)) { |
3332 | error = err2; | |
a8104a9f | 3333 | goto fail; |
c30dabfe | 3334 | } |
ed75e95d | 3335 | *path = nd.path; |
1da177e4 | 3336 | return dentry; |
1da177e4 | 3337 | fail: |
a8104a9f AV |
3338 | dput(dentry); |
3339 | dentry = ERR_PTR(error); | |
3340 | unlock: | |
ed75e95d | 3341 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3342 | if (!err2) |
3343 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3344 | out: |
3345 | path_put(&nd.path); | |
1da177e4 LT |
3346 | return dentry; |
3347 | } | |
dae6ad8f AV |
3348 | EXPORT_SYMBOL(kern_path_create); |
3349 | ||
921a1650 AV |
3350 | void done_path_create(struct path *path, struct dentry *dentry) |
3351 | { | |
3352 | dput(dentry); | |
3353 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3354 | mnt_drop_write(path->mnt); |
921a1650 AV |
3355 | path_put(path); |
3356 | } | |
3357 | EXPORT_SYMBOL(done_path_create); | |
3358 | ||
1ac12b4b JL |
3359 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3360 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3361 | { |
91a27b2a | 3362 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3363 | struct dentry *res; |
3364 | if (IS_ERR(tmp)) | |
3365 | return ERR_CAST(tmp); | |
1ac12b4b | 3366 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3367 | putname(tmp); |
3368 | return res; | |
3369 | } | |
3370 | EXPORT_SYMBOL(user_path_create); | |
3371 | ||
1a67aafb | 3372 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3373 | { |
a95164d9 | 3374 | int error = may_create(dir, dentry); |
1da177e4 LT |
3375 | |
3376 | if (error) | |
3377 | return error; | |
3378 | ||
975d6b39 | 3379 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3380 | return -EPERM; |
3381 | ||
acfa4380 | 3382 | if (!dir->i_op->mknod) |
1da177e4 LT |
3383 | return -EPERM; |
3384 | ||
08ce5f16 SH |
3385 | error = devcgroup_inode_mknod(mode, dev); |
3386 | if (error) | |
3387 | return error; | |
3388 | ||
1da177e4 LT |
3389 | error = security_inode_mknod(dir, dentry, mode, dev); |
3390 | if (error) | |
3391 | return error; | |
3392 | ||
1da177e4 | 3393 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3394 | if (!error) |
f38aa942 | 3395 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3396 | return error; |
3397 | } | |
3398 | ||
f69aac00 | 3399 | static int may_mknod(umode_t mode) |
463c3197 DH |
3400 | { |
3401 | switch (mode & S_IFMT) { | |
3402 | case S_IFREG: | |
3403 | case S_IFCHR: | |
3404 | case S_IFBLK: | |
3405 | case S_IFIFO: | |
3406 | case S_IFSOCK: | |
3407 | case 0: /* zero mode translates to S_IFREG */ | |
3408 | return 0; | |
3409 | case S_IFDIR: | |
3410 | return -EPERM; | |
3411 | default: | |
3412 | return -EINVAL; | |
3413 | } | |
3414 | } | |
3415 | ||
8208a22b | 3416 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3417 | unsigned, dev) |
1da177e4 | 3418 | { |
2ad94ae6 | 3419 | struct dentry *dentry; |
dae6ad8f AV |
3420 | struct path path; |
3421 | int error; | |
972567f1 | 3422 | unsigned int lookup_flags = 0; |
1da177e4 | 3423 | |
8e4bfca1 AV |
3424 | error = may_mknod(mode); |
3425 | if (error) | |
3426 | return error; | |
972567f1 JL |
3427 | retry: |
3428 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3429 | if (IS_ERR(dentry)) |
3430 | return PTR_ERR(dentry); | |
2ad94ae6 | 3431 | |
dae6ad8f | 3432 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3433 | mode &= ~current_umask(); |
dae6ad8f | 3434 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3435 | if (error) |
a8104a9f | 3436 | goto out; |
463c3197 | 3437 | switch (mode & S_IFMT) { |
1da177e4 | 3438 | case 0: case S_IFREG: |
312b63fb | 3439 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3440 | break; |
3441 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3442 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3443 | new_decode_dev(dev)); |
3444 | break; | |
3445 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3446 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3447 | break; |
1da177e4 | 3448 | } |
a8104a9f | 3449 | out: |
921a1650 | 3450 | done_path_create(&path, dentry); |
972567f1 JL |
3451 | if (retry_estale(error, lookup_flags)) { |
3452 | lookup_flags |= LOOKUP_REVAL; | |
3453 | goto retry; | |
3454 | } | |
1da177e4 LT |
3455 | return error; |
3456 | } | |
3457 | ||
8208a22b | 3458 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3459 | { |
3460 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3461 | } | |
3462 | ||
18bb1db3 | 3463 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3464 | { |
a95164d9 | 3465 | int error = may_create(dir, dentry); |
8de52778 | 3466 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3467 | |
3468 | if (error) | |
3469 | return error; | |
3470 | ||
acfa4380 | 3471 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3472 | return -EPERM; |
3473 | ||
3474 | mode &= (S_IRWXUGO|S_ISVTX); | |
3475 | error = security_inode_mkdir(dir, dentry, mode); | |
3476 | if (error) | |
3477 | return error; | |
3478 | ||
8de52778 AV |
3479 | if (max_links && dir->i_nlink >= max_links) |
3480 | return -EMLINK; | |
3481 | ||
1da177e4 | 3482 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3483 | if (!error) |
f38aa942 | 3484 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3485 | return error; |
3486 | } | |
3487 | ||
a218d0fd | 3488 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3489 | { |
6902d925 | 3490 | struct dentry *dentry; |
dae6ad8f AV |
3491 | struct path path; |
3492 | int error; | |
b76d8b82 | 3493 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3494 | |
b76d8b82 JL |
3495 | retry: |
3496 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3497 | if (IS_ERR(dentry)) |
dae6ad8f | 3498 | return PTR_ERR(dentry); |
1da177e4 | 3499 | |
dae6ad8f | 3500 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3501 | mode &= ~current_umask(); |
dae6ad8f | 3502 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3503 | if (!error) |
3504 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3505 | done_path_create(&path, dentry); |
b76d8b82 JL |
3506 | if (retry_estale(error, lookup_flags)) { |
3507 | lookup_flags |= LOOKUP_REVAL; | |
3508 | goto retry; | |
3509 | } | |
1da177e4 LT |
3510 | return error; |
3511 | } | |
3512 | ||
a218d0fd | 3513 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3514 | { |
3515 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3516 | } | |
3517 | ||
1da177e4 | 3518 | /* |
a71905f0 | 3519 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3520 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3521 | * dentry, and if that is true (possibly after pruning the dcache), |
3522 | * then we drop the dentry now. | |
1da177e4 LT |
3523 | * |
3524 | * A low-level filesystem can, if it choses, legally | |
3525 | * do a | |
3526 | * | |
3527 | * if (!d_unhashed(dentry)) | |
3528 | * return -EBUSY; | |
3529 | * | |
3530 | * if it cannot handle the case of removing a directory | |
3531 | * that is still in use by something else.. | |
3532 | */ | |
3533 | void dentry_unhash(struct dentry *dentry) | |
3534 | { | |
dc168427 | 3535 | shrink_dcache_parent(dentry); |
1da177e4 | 3536 | spin_lock(&dentry->d_lock); |
98474236 | 3537 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3538 | __d_drop(dentry); |
3539 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3540 | } |
3541 | ||
3542 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3543 | { | |
3544 | int error = may_delete(dir, dentry, 1); | |
3545 | ||
3546 | if (error) | |
3547 | return error; | |
3548 | ||
acfa4380 | 3549 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3550 | return -EPERM; |
3551 | ||
1d2ef590 | 3552 | dget(dentry); |
1b1dcc1b | 3553 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3554 | |
3555 | error = -EBUSY; | |
1da177e4 | 3556 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3557 | goto out; |
3558 | ||
3559 | error = security_inode_rmdir(dir, dentry); | |
3560 | if (error) | |
3561 | goto out; | |
3562 | ||
3cebde24 | 3563 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3564 | error = dir->i_op->rmdir(dir, dentry); |
3565 | if (error) | |
3566 | goto out; | |
3567 | ||
3568 | dentry->d_inode->i_flags |= S_DEAD; | |
3569 | dont_mount(dentry); | |
3570 | ||
3571 | out: | |
1b1dcc1b | 3572 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3573 | dput(dentry); |
912dbc15 | 3574 | if (!error) |
1da177e4 | 3575 | d_delete(dentry); |
1da177e4 LT |
3576 | return error; |
3577 | } | |
3578 | ||
5590ff0d | 3579 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3580 | { |
3581 | int error = 0; | |
91a27b2a | 3582 | struct filename *name; |
1da177e4 LT |
3583 | struct dentry *dentry; |
3584 | struct nameidata nd; | |
c6ee9206 JL |
3585 | unsigned int lookup_flags = 0; |
3586 | retry: | |
3587 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3588 | if (IS_ERR(name)) |
3589 | return PTR_ERR(name); | |
1da177e4 LT |
3590 | |
3591 | switch(nd.last_type) { | |
0612d9fb OH |
3592 | case LAST_DOTDOT: |
3593 | error = -ENOTEMPTY; | |
3594 | goto exit1; | |
3595 | case LAST_DOT: | |
3596 | error = -EINVAL; | |
3597 | goto exit1; | |
3598 | case LAST_ROOT: | |
3599 | error = -EBUSY; | |
3600 | goto exit1; | |
1da177e4 | 3601 | } |
0612d9fb OH |
3602 | |
3603 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3604 | error = mnt_want_write(nd.path.mnt); |
3605 | if (error) | |
3606 | goto exit1; | |
0612d9fb | 3607 | |
4ac91378 | 3608 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3609 | dentry = lookup_hash(&nd); |
1da177e4 | 3610 | error = PTR_ERR(dentry); |
6902d925 DH |
3611 | if (IS_ERR(dentry)) |
3612 | goto exit2; | |
e6bc45d6 TT |
3613 | if (!dentry->d_inode) { |
3614 | error = -ENOENT; | |
3615 | goto exit3; | |
3616 | } | |
be6d3e56 KT |
3617 | error = security_path_rmdir(&nd.path, dentry); |
3618 | if (error) | |
c30dabfe | 3619 | goto exit3; |
4ac91378 | 3620 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3621 | exit3: |
6902d925 DH |
3622 | dput(dentry); |
3623 | exit2: | |
4ac91378 | 3624 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3625 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3626 | exit1: |
1d957f9b | 3627 | path_put(&nd.path); |
1da177e4 | 3628 | putname(name); |
c6ee9206 JL |
3629 | if (retry_estale(error, lookup_flags)) { |
3630 | lookup_flags |= LOOKUP_REVAL; | |
3631 | goto retry; | |
3632 | } | |
1da177e4 LT |
3633 | return error; |
3634 | } | |
3635 | ||
3cdad428 | 3636 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3637 | { |
3638 | return do_rmdir(AT_FDCWD, pathname); | |
3639 | } | |
3640 | ||
1da177e4 LT |
3641 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
3642 | { | |
3643 | int error = may_delete(dir, dentry, 0); | |
3644 | ||
3645 | if (error) | |
3646 | return error; | |
3647 | ||
acfa4380 | 3648 | if (!dir->i_op->unlink) |
1da177e4 LT |
3649 | return -EPERM; |
3650 | ||
1b1dcc1b | 3651 | mutex_lock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3652 | if (d_mountpoint(dentry)) |
3653 | error = -EBUSY; | |
3654 | else { | |
3655 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3656 | if (!error) { |
1da177e4 | 3657 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3658 | if (!error) |
d83c49f3 | 3659 | dont_mount(dentry); |
bec1052e | 3660 | } |
1da177e4 | 3661 | } |
1b1dcc1b | 3662 | mutex_unlock(&dentry->d_inode->i_mutex); |
1da177e4 LT |
3663 | |
3664 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3665 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
ece95912 | 3666 | fsnotify_link_count(dentry->d_inode); |
e234f35c | 3667 | d_delete(dentry); |
1da177e4 | 3668 | } |
0eeca283 | 3669 | |
1da177e4 LT |
3670 | return error; |
3671 | } | |
3672 | ||
3673 | /* | |
3674 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3675 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3676 | * writeout happening, and we don't want to prevent access to the directory |
3677 | * while waiting on the I/O. | |
3678 | */ | |
5590ff0d | 3679 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3680 | { |
2ad94ae6 | 3681 | int error; |
91a27b2a | 3682 | struct filename *name; |
1da177e4 LT |
3683 | struct dentry *dentry; |
3684 | struct nameidata nd; | |
3685 | struct inode *inode = NULL; | |
5d18f813 JL |
3686 | unsigned int lookup_flags = 0; |
3687 | retry: | |
3688 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3689 | if (IS_ERR(name)) |
3690 | return PTR_ERR(name); | |
2ad94ae6 | 3691 | |
1da177e4 LT |
3692 | error = -EISDIR; |
3693 | if (nd.last_type != LAST_NORM) | |
3694 | goto exit1; | |
0612d9fb OH |
3695 | |
3696 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3697 | error = mnt_want_write(nd.path.mnt); |
3698 | if (error) | |
3699 | goto exit1; | |
0612d9fb | 3700 | |
4ac91378 | 3701 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3702 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3703 | error = PTR_ERR(dentry); |
3704 | if (!IS_ERR(dentry)) { | |
3705 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3706 | if (nd.last.name[nd.last.len]) |
3707 | goto slashes; | |
1da177e4 | 3708 | inode = dentry->d_inode; |
50338b88 | 3709 | if (!inode) |
e6bc45d6 TT |
3710 | goto slashes; |
3711 | ihold(inode); | |
be6d3e56 KT |
3712 | error = security_path_unlink(&nd.path, dentry); |
3713 | if (error) | |
c30dabfe | 3714 | goto exit2; |
4ac91378 | 3715 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
c30dabfe | 3716 | exit2: |
1da177e4 LT |
3717 | dput(dentry); |
3718 | } | |
4ac91378 | 3719 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3720 | if (inode) |
3721 | iput(inode); /* truncate the inode here */ | |
c30dabfe | 3722 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3723 | exit1: |
1d957f9b | 3724 | path_put(&nd.path); |
1da177e4 | 3725 | putname(name); |
5d18f813 JL |
3726 | if (retry_estale(error, lookup_flags)) { |
3727 | lookup_flags |= LOOKUP_REVAL; | |
3728 | inode = NULL; | |
3729 | goto retry; | |
3730 | } | |
1da177e4 LT |
3731 | return error; |
3732 | ||
3733 | slashes: | |
3734 | error = !dentry->d_inode ? -ENOENT : | |
3735 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
3736 | goto exit2; | |
3737 | } | |
3738 | ||
2e4d0924 | 3739 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3740 | { |
3741 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3742 | return -EINVAL; | |
3743 | ||
3744 | if (flag & AT_REMOVEDIR) | |
3745 | return do_rmdir(dfd, pathname); | |
3746 | ||
3747 | return do_unlinkat(dfd, pathname); | |
3748 | } | |
3749 | ||
3480b257 | 3750 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3751 | { |
3752 | return do_unlinkat(AT_FDCWD, pathname); | |
3753 | } | |
3754 | ||
db2e747b | 3755 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3756 | { |
a95164d9 | 3757 | int error = may_create(dir, dentry); |
1da177e4 LT |
3758 | |
3759 | if (error) | |
3760 | return error; | |
3761 | ||
acfa4380 | 3762 | if (!dir->i_op->symlink) |
1da177e4 LT |
3763 | return -EPERM; |
3764 | ||
3765 | error = security_inode_symlink(dir, dentry, oldname); | |
3766 | if (error) | |
3767 | return error; | |
3768 | ||
1da177e4 | 3769 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3770 | if (!error) |
f38aa942 | 3771 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3772 | return error; |
3773 | } | |
3774 | ||
2e4d0924 HC |
3775 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3776 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3777 | { |
2ad94ae6 | 3778 | int error; |
91a27b2a | 3779 | struct filename *from; |
6902d925 | 3780 | struct dentry *dentry; |
dae6ad8f | 3781 | struct path path; |
f46d3567 | 3782 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3783 | |
3784 | from = getname(oldname); | |
2ad94ae6 | 3785 | if (IS_ERR(from)) |
1da177e4 | 3786 | return PTR_ERR(from); |
f46d3567 JL |
3787 | retry: |
3788 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3789 | error = PTR_ERR(dentry); |
3790 | if (IS_ERR(dentry)) | |
dae6ad8f | 3791 | goto out_putname; |
6902d925 | 3792 | |
91a27b2a | 3793 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3794 | if (!error) |
91a27b2a | 3795 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3796 | done_path_create(&path, dentry); |
f46d3567 JL |
3797 | if (retry_estale(error, lookup_flags)) { |
3798 | lookup_flags |= LOOKUP_REVAL; | |
3799 | goto retry; | |
3800 | } | |
6902d925 | 3801 | out_putname: |
1da177e4 LT |
3802 | putname(from); |
3803 | return error; | |
3804 | } | |
3805 | ||
3480b257 | 3806 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3807 | { |
3808 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3809 | } | |
3810 | ||
1da177e4 LT |
3811 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3812 | { | |
3813 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3814 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3815 | int error; |
3816 | ||
3817 | if (!inode) | |
3818 | return -ENOENT; | |
3819 | ||
a95164d9 | 3820 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3821 | if (error) |
3822 | return error; | |
3823 | ||
3824 | if (dir->i_sb != inode->i_sb) | |
3825 | return -EXDEV; | |
3826 | ||
3827 | /* | |
3828 | * A link to an append-only or immutable file cannot be created. | |
3829 | */ | |
3830 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3831 | return -EPERM; | |
acfa4380 | 3832 | if (!dir->i_op->link) |
1da177e4 | 3833 | return -EPERM; |
7e79eedb | 3834 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3835 | return -EPERM; |
3836 | ||
3837 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3838 | if (error) | |
3839 | return error; | |
3840 | ||
7e79eedb | 3841 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3842 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3843 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3844 | error = -ENOENT; |
8de52778 AV |
3845 | else if (max_links && inode->i_nlink >= max_links) |
3846 | error = -EMLINK; | |
aae8a97d AK |
3847 | else |
3848 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
f4e0c30c AV |
3849 | |
3850 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3851 | spin_lock(&inode->i_lock); | |
3852 | inode->i_state &= ~I_LINKABLE; | |
3853 | spin_unlock(&inode->i_lock); | |
3854 | } | |
7e79eedb | 3855 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3856 | if (!error) |
7e79eedb | 3857 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3858 | return error; |
3859 | } | |
3860 | ||
3861 | /* | |
3862 | * Hardlinks are often used in delicate situations. We avoid | |
3863 | * security-related surprises by not following symlinks on the | |
3864 | * newname. --KAB | |
3865 | * | |
3866 | * We don't follow them on the oldname either to be compatible | |
3867 | * with linux 2.0, and to avoid hard-linking to directories | |
3868 | * and other special files. --ADM | |
3869 | */ | |
2e4d0924 HC |
3870 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3871 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3872 | { |
3873 | struct dentry *new_dentry; | |
dae6ad8f | 3874 | struct path old_path, new_path; |
11a7b371 | 3875 | int how = 0; |
1da177e4 | 3876 | int error; |
1da177e4 | 3877 | |
11a7b371 | 3878 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3879 | return -EINVAL; |
11a7b371 | 3880 | /* |
f0cc6ffb LT |
3881 | * To use null names we require CAP_DAC_READ_SEARCH |
3882 | * This ensures that not everyone will be able to create | |
3883 | * handlink using the passed filedescriptor. | |
11a7b371 | 3884 | */ |
f0cc6ffb LT |
3885 | if (flags & AT_EMPTY_PATH) { |
3886 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3887 | return -ENOENT; | |
11a7b371 | 3888 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3889 | } |
11a7b371 AK |
3890 | |
3891 | if (flags & AT_SYMLINK_FOLLOW) | |
3892 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3893 | retry: |
11a7b371 | 3894 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3895 | if (error) |
2ad94ae6 AV |
3896 | return error; |
3897 | ||
442e31ca JL |
3898 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3899 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3900 | error = PTR_ERR(new_dentry); |
6902d925 | 3901 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3902 | goto out; |
3903 | ||
3904 | error = -EXDEV; | |
3905 | if (old_path.mnt != new_path.mnt) | |
3906 | goto out_dput; | |
800179c9 KC |
3907 | error = may_linkat(&old_path); |
3908 | if (unlikely(error)) | |
3909 | goto out_dput; | |
dae6ad8f | 3910 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3911 | if (error) |
a8104a9f | 3912 | goto out_dput; |
dae6ad8f | 3913 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
75c3f29d | 3914 | out_dput: |
921a1650 | 3915 | done_path_create(&new_path, new_dentry); |
442e31ca JL |
3916 | if (retry_estale(error, how)) { |
3917 | how |= LOOKUP_REVAL; | |
3918 | goto retry; | |
3919 | } | |
1da177e4 | 3920 | out: |
2d8f3038 | 3921 | path_put(&old_path); |
1da177e4 LT |
3922 | |
3923 | return error; | |
3924 | } | |
3925 | ||
3480b257 | 3926 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3927 | { |
c04030e1 | 3928 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3929 | } |
3930 | ||
1da177e4 LT |
3931 | /* |
3932 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3933 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3934 | * Problems: | |
3935 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3936 | * b) race potential - two innocent renames can create a loop together. | |
3937 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3938 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 LT |
3939 | * story. |
3940 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
1b1dcc1b | 3941 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3942 | * whether the target exists). Solution: try to be smart with locking |
3943 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3944 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3945 | * move will be locked. Thus we can rank directories by the tree |
3946 | * (ancestors first) and rank all non-directories after them. | |
3947 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3948 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3949 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3950 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3951 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3952 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3953 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3954 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3955 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3956 | * locking]. |
3957 | */ | |
75c96f85 AB |
3958 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3959 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3960 | { |
3961 | int error = 0; | |
9055cba7 | 3962 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3963 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3964 | |
3965 | /* | |
3966 | * If we are going to change the parent - check write permissions, | |
3967 | * we'll need to flip '..'. | |
3968 | */ | |
3969 | if (new_dir != old_dir) { | |
f419a2e3 | 3970 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3971 | if (error) |
3972 | return error; | |
3973 | } | |
3974 | ||
3975 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3976 | if (error) | |
3977 | return error; | |
3978 | ||
1d2ef590 | 3979 | dget(new_dentry); |
d83c49f3 | 3980 | if (target) |
1b1dcc1b | 3981 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
3982 | |
3983 | error = -EBUSY; | |
3984 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
3985 | goto out; | |
3986 | ||
8de52778 AV |
3987 | error = -EMLINK; |
3988 | if (max_links && !target && new_dir != old_dir && | |
3989 | new_dir->i_nlink >= max_links) | |
3990 | goto out; | |
3991 | ||
3cebde24 SW |
3992 | if (target) |
3993 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
3994 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3995 | if (error) | |
3996 | goto out; | |
3997 | ||
1da177e4 | 3998 | if (target) { |
9055cba7 SW |
3999 | target->i_flags |= S_DEAD; |
4000 | dont_mount(new_dentry); | |
1da177e4 | 4001 | } |
9055cba7 SW |
4002 | out: |
4003 | if (target) | |
4004 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 4005 | dput(new_dentry); |
e31e14ec | 4006 | if (!error) |
349457cc MF |
4007 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
4008 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
4009 | return error; |
4010 | } | |
4011 | ||
75c96f85 AB |
4012 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
4013 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 | 4014 | { |
51892bbb | 4015 | struct inode *target = new_dentry->d_inode; |
1da177e4 LT |
4016 | int error; |
4017 | ||
4018 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4019 | if (error) | |
4020 | return error; | |
4021 | ||
4022 | dget(new_dentry); | |
1da177e4 | 4023 | if (target) |
1b1dcc1b | 4024 | mutex_lock(&target->i_mutex); |
51892bbb SW |
4025 | |
4026 | error = -EBUSY; | |
1da177e4 | 4027 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
4028 | goto out; |
4029 | ||
4030 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
4031 | if (error) | |
4032 | goto out; | |
4033 | ||
4034 | if (target) | |
4035 | dont_mount(new_dentry); | |
4036 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
4037 | d_move(old_dentry, new_dentry); | |
4038 | out: | |
1da177e4 | 4039 | if (target) |
1b1dcc1b | 4040 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
4041 | dput(new_dentry); |
4042 | return error; | |
4043 | } | |
4044 | ||
4045 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
4046 | struct inode *new_dir, struct dentry *new_dentry) | |
4047 | { | |
4048 | int error; | |
4049 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
59b0df21 | 4050 | const unsigned char *old_name; |
1da177e4 LT |
4051 | |
4052 | if (old_dentry->d_inode == new_dentry->d_inode) | |
4053 | return 0; | |
4054 | ||
4055 | error = may_delete(old_dir, old_dentry, is_dir); | |
4056 | if (error) | |
4057 | return error; | |
4058 | ||
4059 | if (!new_dentry->d_inode) | |
a95164d9 | 4060 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
4061 | else |
4062 | error = may_delete(new_dir, new_dentry, is_dir); | |
4063 | if (error) | |
4064 | return error; | |
4065 | ||
acfa4380 | 4066 | if (!old_dir->i_op->rename) |
1da177e4 LT |
4067 | return -EPERM; |
4068 | ||
0eeca283 RL |
4069 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
4070 | ||
1da177e4 LT |
4071 | if (is_dir) |
4072 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
4073 | else | |
4074 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
123df294 AV |
4075 | if (!error) |
4076 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 4077 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
4078 | fsnotify_oldname_free(old_name); |
4079 | ||
1da177e4 LT |
4080 | return error; |
4081 | } | |
4082 | ||
2e4d0924 HC |
4083 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4084 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4085 | { |
2ad94ae6 AV |
4086 | struct dentry *old_dir, *new_dir; |
4087 | struct dentry *old_dentry, *new_dentry; | |
4088 | struct dentry *trap; | |
1da177e4 | 4089 | struct nameidata oldnd, newnd; |
91a27b2a JL |
4090 | struct filename *from; |
4091 | struct filename *to; | |
c6a94284 JL |
4092 | unsigned int lookup_flags = 0; |
4093 | bool should_retry = false; | |
2ad94ae6 | 4094 | int error; |
c6a94284 JL |
4095 | retry: |
4096 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4097 | if (IS_ERR(from)) { |
4098 | error = PTR_ERR(from); | |
1da177e4 | 4099 | goto exit; |
91a27b2a | 4100 | } |
1da177e4 | 4101 | |
c6a94284 | 4102 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4103 | if (IS_ERR(to)) { |
4104 | error = PTR_ERR(to); | |
1da177e4 | 4105 | goto exit1; |
91a27b2a | 4106 | } |
1da177e4 LT |
4107 | |
4108 | error = -EXDEV; | |
4ac91378 | 4109 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4110 | goto exit2; |
4111 | ||
4ac91378 | 4112 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4113 | error = -EBUSY; |
4114 | if (oldnd.last_type != LAST_NORM) | |
4115 | goto exit2; | |
4116 | ||
4ac91378 | 4117 | new_dir = newnd.path.dentry; |
1da177e4 LT |
4118 | if (newnd.last_type != LAST_NORM) |
4119 | goto exit2; | |
4120 | ||
c30dabfe JK |
4121 | error = mnt_want_write(oldnd.path.mnt); |
4122 | if (error) | |
4123 | goto exit2; | |
4124 | ||
0612d9fb OH |
4125 | oldnd.flags &= ~LOOKUP_PARENT; |
4126 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 4127 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 4128 | |
1da177e4 LT |
4129 | trap = lock_rename(new_dir, old_dir); |
4130 | ||
49705b77 | 4131 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4132 | error = PTR_ERR(old_dentry); |
4133 | if (IS_ERR(old_dentry)) | |
4134 | goto exit3; | |
4135 | /* source must exist */ | |
4136 | error = -ENOENT; | |
4137 | if (!old_dentry->d_inode) | |
4138 | goto exit4; | |
4139 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
4140 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
4141 | error = -ENOTDIR; | |
4142 | if (oldnd.last.name[oldnd.last.len]) | |
4143 | goto exit4; | |
4144 | if (newnd.last.name[newnd.last.len]) | |
4145 | goto exit4; | |
4146 | } | |
4147 | /* source should not be ancestor of target */ | |
4148 | error = -EINVAL; | |
4149 | if (old_dentry == trap) | |
4150 | goto exit4; | |
49705b77 | 4151 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
4152 | error = PTR_ERR(new_dentry); |
4153 | if (IS_ERR(new_dentry)) | |
4154 | goto exit4; | |
4155 | /* target should not be an ancestor of source */ | |
4156 | error = -ENOTEMPTY; | |
4157 | if (new_dentry == trap) | |
4158 | goto exit5; | |
4159 | ||
be6d3e56 KT |
4160 | error = security_path_rename(&oldnd.path, old_dentry, |
4161 | &newnd.path, new_dentry); | |
4162 | if (error) | |
c30dabfe | 4163 | goto exit5; |
1da177e4 LT |
4164 | error = vfs_rename(old_dir->d_inode, old_dentry, |
4165 | new_dir->d_inode, new_dentry); | |
4166 | exit5: | |
4167 | dput(new_dentry); | |
4168 | exit4: | |
4169 | dput(old_dentry); | |
4170 | exit3: | |
4171 | unlock_rename(new_dir, old_dir); | |
c30dabfe | 4172 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4173 | exit2: |
c6a94284 JL |
4174 | if (retry_estale(error, lookup_flags)) |
4175 | should_retry = true; | |
1d957f9b | 4176 | path_put(&newnd.path); |
2ad94ae6 | 4177 | putname(to); |
1da177e4 | 4178 | exit1: |
1d957f9b | 4179 | path_put(&oldnd.path); |
1da177e4 | 4180 | putname(from); |
c6a94284 JL |
4181 | if (should_retry) { |
4182 | should_retry = false; | |
4183 | lookup_flags |= LOOKUP_REVAL; | |
4184 | goto retry; | |
4185 | } | |
2ad94ae6 | 4186 | exit: |
1da177e4 LT |
4187 | return error; |
4188 | } | |
4189 | ||
a26eab24 | 4190 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4191 | { |
4192 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
4193 | } | |
4194 | ||
1da177e4 LT |
4195 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
4196 | { | |
4197 | int len; | |
4198 | ||
4199 | len = PTR_ERR(link); | |
4200 | if (IS_ERR(link)) | |
4201 | goto out; | |
4202 | ||
4203 | len = strlen(link); | |
4204 | if (len > (unsigned) buflen) | |
4205 | len = buflen; | |
4206 | if (copy_to_user(buffer, link, len)) | |
4207 | len = -EFAULT; | |
4208 | out: | |
4209 | return len; | |
4210 | } | |
4211 | ||
4212 | /* | |
4213 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4214 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4215 | * using) it for any given inode is up to filesystem. | |
4216 | */ | |
4217 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4218 | { | |
4219 | struct nameidata nd; | |
cc314eef | 4220 | void *cookie; |
694a1764 | 4221 | int res; |
cc314eef | 4222 | |
1da177e4 | 4223 | nd.depth = 0; |
cc314eef | 4224 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4225 | if (IS_ERR(cookie)) |
4226 | return PTR_ERR(cookie); | |
4227 | ||
4228 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
4229 | if (dentry->d_inode->i_op->put_link) | |
4230 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4231 | return res; | |
1da177e4 LT |
4232 | } |
4233 | ||
1da177e4 LT |
4234 | /* get the link contents into pagecache */ |
4235 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4236 | { | |
ebd09abb DG |
4237 | char *kaddr; |
4238 | struct page *page; | |
1da177e4 | 4239 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4240 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4241 | if (IS_ERR(page)) |
6fe6900e | 4242 | return (char*)page; |
1da177e4 | 4243 | *ppage = page; |
ebd09abb DG |
4244 | kaddr = kmap(page); |
4245 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4246 | return kaddr; | |
1da177e4 LT |
4247 | } |
4248 | ||
4249 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4250 | { | |
4251 | struct page *page = NULL; | |
4252 | char *s = page_getlink(dentry, &page); | |
4253 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4254 | if (page) { | |
4255 | kunmap(page); | |
4256 | page_cache_release(page); | |
4257 | } | |
4258 | return res; | |
4259 | } | |
4260 | ||
cc314eef | 4261 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4262 | { |
cc314eef | 4263 | struct page *page = NULL; |
1da177e4 | 4264 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4265 | return page; |
1da177e4 LT |
4266 | } |
4267 | ||
cc314eef | 4268 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4269 | { |
cc314eef LT |
4270 | struct page *page = cookie; |
4271 | ||
4272 | if (page) { | |
1da177e4 LT |
4273 | kunmap(page); |
4274 | page_cache_release(page); | |
1da177e4 LT |
4275 | } |
4276 | } | |
4277 | ||
54566b2c NP |
4278 | /* |
4279 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4280 | */ | |
4281 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4282 | { |
4283 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4284 | struct page *page; |
afddba49 | 4285 | void *fsdata; |
beb497ab | 4286 | int err; |
1da177e4 | 4287 | char *kaddr; |
54566b2c NP |
4288 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4289 | if (nofs) | |
4290 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4291 | |
7e53cac4 | 4292 | retry: |
afddba49 | 4293 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4294 | flags, &page, &fsdata); |
1da177e4 | 4295 | if (err) |
afddba49 NP |
4296 | goto fail; |
4297 | ||
e8e3c3d6 | 4298 | kaddr = kmap_atomic(page); |
1da177e4 | 4299 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4300 | kunmap_atomic(kaddr); |
afddba49 NP |
4301 | |
4302 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4303 | page, fsdata); | |
1da177e4 LT |
4304 | if (err < 0) |
4305 | goto fail; | |
afddba49 NP |
4306 | if (err < len-1) |
4307 | goto retry; | |
4308 | ||
1da177e4 LT |
4309 | mark_inode_dirty(inode); |
4310 | return 0; | |
1da177e4 LT |
4311 | fail: |
4312 | return err; | |
4313 | } | |
4314 | ||
0adb25d2 KK |
4315 | int page_symlink(struct inode *inode, const char *symname, int len) |
4316 | { | |
4317 | return __page_symlink(inode, symname, len, | |
54566b2c | 4318 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4319 | } |
4320 | ||
92e1d5be | 4321 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4322 | .readlink = generic_readlink, |
4323 | .follow_link = page_follow_link_light, | |
4324 | .put_link = page_put_link, | |
4325 | }; | |
4326 | ||
2d8f3038 | 4327 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4328 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4329 | EXPORT_SYMBOL(follow_down); |
4330 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4331 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4332 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4333 | EXPORT_SYMBOL(lookup_one_len); |
4334 | EXPORT_SYMBOL(page_follow_link_light); | |
4335 | EXPORT_SYMBOL(page_put_link); | |
4336 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4337 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4338 | EXPORT_SYMBOL(page_symlink); |
4339 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4340 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4341 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4342 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4343 | EXPORT_SYMBOL(unlock_rename); |
4344 | EXPORT_SYMBOL(vfs_create); | |
1da177e4 LT |
4345 | EXPORT_SYMBOL(vfs_link); |
4346 | EXPORT_SYMBOL(vfs_mkdir); | |
4347 | EXPORT_SYMBOL(vfs_mknod); | |
4348 | EXPORT_SYMBOL(generic_permission); | |
4349 | EXPORT_SYMBOL(vfs_readlink); | |
4350 | EXPORT_SYMBOL(vfs_rename); | |
4351 | EXPORT_SYMBOL(vfs_rmdir); | |
4352 | EXPORT_SYMBOL(vfs_symlink); | |
4353 | EXPORT_SYMBOL(vfs_unlink); | |
4354 | EXPORT_SYMBOL(dentry_unhash); | |
4355 | EXPORT_SYMBOL(generic_readlink); |