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> |
99d263d4 | 37 | #include <linux/hash.h> |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | ||
e81e3f4d | 40 | #include "internal.h" |
c7105365 | 41 | #include "mount.h" |
e81e3f4d | 42 | |
1da177e4 LT |
43 | /* [Feb-1997 T. Schoebel-Theuer] |
44 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
45 | * were necessary because of omirr. The reason is that omirr needs | |
46 | * to know the _real_ pathname, not the user-supplied one, in case | |
47 | * of symlinks (and also when transname replacements occur). | |
48 | * | |
49 | * The new code replaces the old recursive symlink resolution with | |
50 | * an iterative one (in case of non-nested symlink chains). It does | |
51 | * this with calls to <fs>_follow_link(). | |
52 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
53 | * replaced with a single function lookup_dentry() that can handle all | |
54 | * the special cases of the former code. | |
55 | * | |
56 | * With the new dcache, the pathname is stored at each inode, at least as | |
57 | * long as the refcount of the inode is positive. As a side effect, the | |
58 | * size of the dcache depends on the inode cache and thus is dynamic. | |
59 | * | |
60 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
61 | * resolution to correspond with current state of the code. | |
62 | * | |
63 | * Note that the symlink resolution is not *completely* iterative. | |
64 | * There is still a significant amount of tail- and mid- recursion in | |
65 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
66 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
67 | * may return different results than <fs>_follow_link(). Many virtual | |
68 | * filesystems (including /proc) exhibit this behavior. | |
69 | */ | |
70 | ||
71 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
72 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
73 | * and the name already exists in form of a symlink, try to create the new | |
74 | * name indicated by the symlink. The old code always complained that the | |
75 | * name already exists, due to not following the symlink even if its target | |
76 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 77 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
78 | * |
79 | * I don't know which semantics is the right one, since I have no access | |
80 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
81 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
82 | * "old" one. Personally, I think the new semantics is much more logical. | |
83 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
84 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
85 | * and in the old Linux semantics. | |
86 | */ | |
87 | ||
88 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
89 | * semantics. See the comments in "open_namei" and "do_link" below. | |
90 | * | |
91 | * [10-Sep-98 Alan Modra] Another symlink change. | |
92 | */ | |
93 | ||
94 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
95 | * inside the path - always follow. | |
96 | * in the last component in creation/removal/renaming - never follow. | |
97 | * if LOOKUP_FOLLOW passed - follow. | |
98 | * if the pathname has trailing slashes - follow. | |
99 | * otherwise - don't follow. | |
100 | * (applied in that order). | |
101 | * | |
102 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
103 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
104 | * During the 2.4 we need to fix the userland stuff depending on it - | |
105 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
106 | * XEmacs seems to be relying on it... | |
107 | */ | |
108 | /* | |
109 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 110 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
111 | * any extra contention... |
112 | */ | |
113 | ||
114 | /* In order to reduce some races, while at the same time doing additional | |
115 | * checking and hopefully speeding things up, we copy filenames to the | |
116 | * kernel data space before using them.. | |
117 | * | |
118 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
119 | * PATH_MAX includes the nul terminator --RR. | |
120 | */ | |
91a27b2a | 121 | void final_putname(struct filename *name) |
1da177e4 | 122 | { |
7950e385 JL |
123 | if (name->separate) { |
124 | __putname(name->name); | |
125 | kfree(name); | |
126 | } else { | |
127 | __putname(name); | |
128 | } | |
91a27b2a JL |
129 | } |
130 | ||
7950e385 JL |
131 | #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) |
132 | ||
91a27b2a JL |
133 | static struct filename * |
134 | getname_flags(const char __user *filename, int flags, int *empty) | |
135 | { | |
136 | struct filename *result, *err; | |
3f9f0aa6 | 137 | int len; |
7950e385 JL |
138 | long max; |
139 | char *kname; | |
4043cde8 | 140 | |
7ac86265 JL |
141 | result = audit_reusename(filename); |
142 | if (result) | |
143 | return result; | |
144 | ||
7950e385 | 145 | result = __getname(); |
3f9f0aa6 | 146 | if (unlikely(!result)) |
4043cde8 EP |
147 | return ERR_PTR(-ENOMEM); |
148 | ||
7950e385 JL |
149 | /* |
150 | * First, try to embed the struct filename inside the names_cache | |
151 | * allocation | |
152 | */ | |
153 | kname = (char *)result + sizeof(*result); | |
91a27b2a | 154 | result->name = kname; |
7950e385 JL |
155 | result->separate = false; |
156 | max = EMBEDDED_NAME_MAX; | |
157 | ||
158 | recopy: | |
159 | len = strncpy_from_user(kname, filename, max); | |
91a27b2a JL |
160 | if (unlikely(len < 0)) { |
161 | err = ERR_PTR(len); | |
3f9f0aa6 | 162 | goto error; |
91a27b2a | 163 | } |
3f9f0aa6 | 164 | |
7950e385 JL |
165 | /* |
166 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
167 | * separate struct filename so we can dedicate the entire | |
168 | * names_cache allocation for the pathname, and re-do the copy from | |
169 | * userland. | |
170 | */ | |
171 | if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { | |
172 | kname = (char *)result; | |
173 | ||
174 | result = kzalloc(sizeof(*result), GFP_KERNEL); | |
175 | if (!result) { | |
176 | err = ERR_PTR(-ENOMEM); | |
177 | result = (struct filename *)kname; | |
178 | goto error; | |
179 | } | |
180 | result->name = kname; | |
181 | result->separate = true; | |
182 | max = PATH_MAX; | |
183 | goto recopy; | |
184 | } | |
185 | ||
3f9f0aa6 LT |
186 | /* The empty path is special. */ |
187 | if (unlikely(!len)) { | |
188 | if (empty) | |
4043cde8 | 189 | *empty = 1; |
3f9f0aa6 LT |
190 | err = ERR_PTR(-ENOENT); |
191 | if (!(flags & LOOKUP_EMPTY)) | |
192 | goto error; | |
1da177e4 | 193 | } |
3f9f0aa6 LT |
194 | |
195 | err = ERR_PTR(-ENAMETOOLONG); | |
7950e385 JL |
196 | if (unlikely(len >= PATH_MAX)) |
197 | goto error; | |
198 | ||
199 | result->uptr = filename; | |
c4ad8f98 | 200 | result->aname = NULL; |
7950e385 JL |
201 | audit_getname(result); |
202 | return result; | |
3f9f0aa6 LT |
203 | |
204 | error: | |
7950e385 | 205 | final_putname(result); |
3f9f0aa6 | 206 | return err; |
1da177e4 LT |
207 | } |
208 | ||
91a27b2a JL |
209 | struct filename * |
210 | getname(const char __user * filename) | |
f52e0c11 | 211 | { |
f7493e5d | 212 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
213 | } |
214 | ||
c4ad8f98 LT |
215 | /* |
216 | * The "getname_kernel()" interface doesn't do pathnames longer | |
217 | * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user. | |
218 | */ | |
219 | struct filename * | |
220 | getname_kernel(const char * filename) | |
221 | { | |
222 | struct filename *result; | |
223 | char *kname; | |
224 | int len; | |
225 | ||
226 | len = strlen(filename); | |
227 | if (len >= EMBEDDED_NAME_MAX) | |
228 | return ERR_PTR(-ENAMETOOLONG); | |
229 | ||
230 | result = __getname(); | |
231 | if (unlikely(!result)) | |
232 | return ERR_PTR(-ENOMEM); | |
233 | ||
234 | kname = (char *)result + sizeof(*result); | |
235 | result->name = kname; | |
236 | result->uptr = NULL; | |
237 | result->aname = NULL; | |
238 | result->separate = false; | |
239 | ||
240 | strlcpy(kname, filename, EMBEDDED_NAME_MAX); | |
241 | return result; | |
242 | } | |
243 | ||
1da177e4 | 244 | #ifdef CONFIG_AUDITSYSCALL |
91a27b2a | 245 | void putname(struct filename *name) |
1da177e4 | 246 | { |
5ac3a9c2 | 247 | if (unlikely(!audit_dummy_context())) |
91a27b2a JL |
248 | return audit_putname(name); |
249 | final_putname(name); | |
1da177e4 | 250 | } |
1da177e4 LT |
251 | #endif |
252 | ||
e77819e5 LT |
253 | static int check_acl(struct inode *inode, int mask) |
254 | { | |
84635d68 | 255 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
256 | struct posix_acl *acl; |
257 | ||
e77819e5 | 258 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
259 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
260 | if (!acl) | |
e77819e5 | 261 | return -EAGAIN; |
3567866b AV |
262 | /* no ->get_acl() calls in RCU mode... */ |
263 | if (acl == ACL_NOT_CACHED) | |
264 | return -ECHILD; | |
206b1d09 | 265 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
266 | } |
267 | ||
2982baa2 CH |
268 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
269 | if (IS_ERR(acl)) | |
270 | return PTR_ERR(acl); | |
e77819e5 LT |
271 | if (acl) { |
272 | int error = posix_acl_permission(inode, acl, mask); | |
273 | posix_acl_release(acl); | |
274 | return error; | |
275 | } | |
84635d68 | 276 | #endif |
e77819e5 LT |
277 | |
278 | return -EAGAIN; | |
279 | } | |
280 | ||
5909ccaa | 281 | /* |
948409c7 | 282 | * This does the basic permission checking |
1da177e4 | 283 | */ |
7e40145e | 284 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 285 | { |
26cf46be | 286 | unsigned int mode = inode->i_mode; |
1da177e4 | 287 | |
8e96e3b7 | 288 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
289 | mode >>= 6; |
290 | else { | |
e77819e5 | 291 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 292 | int error = check_acl(inode, mask); |
b74c79e9 NP |
293 | if (error != -EAGAIN) |
294 | return error; | |
1da177e4 LT |
295 | } |
296 | ||
297 | if (in_group_p(inode->i_gid)) | |
298 | mode >>= 3; | |
299 | } | |
300 | ||
301 | /* | |
302 | * If the DACs are ok we don't need any capability check. | |
303 | */ | |
9c2c7039 | 304 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 305 | return 0; |
5909ccaa LT |
306 | return -EACCES; |
307 | } | |
308 | ||
309 | /** | |
b74c79e9 | 310 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 311 | * @inode: inode to check access rights for |
8fd90c8d | 312 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
313 | * |
314 | * Used to check for read/write/execute permissions on a file. | |
315 | * We use "fsuid" for this, letting us set arbitrary permissions | |
316 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
317 | * are used for other things. |
318 | * | |
319 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
320 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
321 | * It would then be called again in ref-walk mode. | |
5909ccaa | 322 | */ |
2830ba7f | 323 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
324 | { |
325 | int ret; | |
326 | ||
327 | /* | |
948409c7 | 328 | * Do the basic permission checks. |
5909ccaa | 329 | */ |
7e40145e | 330 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
331 | if (ret != -EACCES) |
332 | return ret; | |
1da177e4 | 333 | |
d594e7ec AV |
334 | if (S_ISDIR(inode->i_mode)) { |
335 | /* DACs are overridable for directories */ | |
23adbe12 | 336 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
337 | return 0; |
338 | if (!(mask & MAY_WRITE)) | |
23adbe12 AL |
339 | if (capable_wrt_inode_uidgid(inode, |
340 | CAP_DAC_READ_SEARCH)) | |
d594e7ec AV |
341 | return 0; |
342 | return -EACCES; | |
343 | } | |
1da177e4 LT |
344 | /* |
345 | * Read/write DACs are always overridable. | |
d594e7ec AV |
346 | * Executable DACs are overridable when there is |
347 | * at least one exec bit set. | |
1da177e4 | 348 | */ |
d594e7ec | 349 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
23adbe12 | 350 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
351 | return 0; |
352 | ||
353 | /* | |
354 | * Searching includes executable on directories, else just read. | |
355 | */ | |
7ea66001 | 356 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 357 | if (mask == MAY_READ) |
23adbe12 | 358 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
359 | return 0; |
360 | ||
361 | return -EACCES; | |
362 | } | |
4d359507 | 363 | EXPORT_SYMBOL(generic_permission); |
1da177e4 | 364 | |
3ddcd056 LT |
365 | /* |
366 | * We _really_ want to just do "generic_permission()" without | |
367 | * even looking at the inode->i_op values. So we keep a cache | |
368 | * flag in inode->i_opflags, that says "this has not special | |
369 | * permission function, use the fast case". | |
370 | */ | |
371 | static inline int do_inode_permission(struct inode *inode, int mask) | |
372 | { | |
373 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
374 | if (likely(inode->i_op->permission)) | |
375 | return inode->i_op->permission(inode, mask); | |
376 | ||
377 | /* This gets set once for the inode lifetime */ | |
378 | spin_lock(&inode->i_lock); | |
379 | inode->i_opflags |= IOP_FASTPERM; | |
380 | spin_unlock(&inode->i_lock); | |
381 | } | |
382 | return generic_permission(inode, mask); | |
383 | } | |
384 | ||
cb23beb5 | 385 | /** |
0bdaea90 DH |
386 | * __inode_permission - Check for access rights to a given inode |
387 | * @inode: Inode to check permission on | |
388 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 389 | * |
0bdaea90 | 390 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
391 | * |
392 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
393 | * |
394 | * This does not check for a read-only file system. You probably want | |
395 | * inode_permission(). | |
cb23beb5 | 396 | */ |
0bdaea90 | 397 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 398 | { |
e6305c43 | 399 | int retval; |
1da177e4 | 400 | |
3ddcd056 | 401 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
402 | /* |
403 | * Nobody gets write access to an immutable file. | |
404 | */ | |
405 | if (IS_IMMUTABLE(inode)) | |
406 | return -EACCES; | |
407 | } | |
408 | ||
3ddcd056 | 409 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
410 | if (retval) |
411 | return retval; | |
412 | ||
08ce5f16 SH |
413 | retval = devcgroup_inode_permission(inode, mask); |
414 | if (retval) | |
415 | return retval; | |
416 | ||
d09ca739 | 417 | return security_inode_permission(inode, mask); |
1da177e4 | 418 | } |
bd5d0856 | 419 | EXPORT_SYMBOL(__inode_permission); |
1da177e4 | 420 | |
0bdaea90 DH |
421 | /** |
422 | * sb_permission - Check superblock-level permissions | |
423 | * @sb: Superblock of inode to check permission on | |
55852635 | 424 | * @inode: Inode to check permission on |
0bdaea90 DH |
425 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
426 | * | |
427 | * Separate out file-system wide checks from inode-specific permission checks. | |
428 | */ | |
429 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
430 | { | |
431 | if (unlikely(mask & MAY_WRITE)) { | |
432 | umode_t mode = inode->i_mode; | |
433 | ||
434 | /* Nobody gets write access to a read-only fs. */ | |
435 | if ((sb->s_flags & MS_RDONLY) && | |
436 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
437 | return -EROFS; | |
438 | } | |
439 | return 0; | |
440 | } | |
441 | ||
442 | /** | |
443 | * inode_permission - Check for access rights to a given inode | |
444 | * @inode: Inode to check permission on | |
445 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
446 | * | |
447 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
448 | * this, letting us set arbitrary permissions for filesystem access without | |
449 | * changing the "normal" UIDs which are used for other things. | |
450 | * | |
451 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
452 | */ | |
453 | int inode_permission(struct inode *inode, int mask) | |
454 | { | |
455 | int retval; | |
456 | ||
457 | retval = sb_permission(inode->i_sb, inode, mask); | |
458 | if (retval) | |
459 | return retval; | |
460 | return __inode_permission(inode, mask); | |
461 | } | |
4d359507 | 462 | EXPORT_SYMBOL(inode_permission); |
0bdaea90 | 463 | |
5dd784d0 JB |
464 | /** |
465 | * path_get - get a reference to a path | |
466 | * @path: path to get the reference to | |
467 | * | |
468 | * Given a path increment the reference count to the dentry and the vfsmount. | |
469 | */ | |
dcf787f3 | 470 | void path_get(const struct path *path) |
5dd784d0 JB |
471 | { |
472 | mntget(path->mnt); | |
473 | dget(path->dentry); | |
474 | } | |
475 | EXPORT_SYMBOL(path_get); | |
476 | ||
1d957f9b JB |
477 | /** |
478 | * path_put - put a reference to a path | |
479 | * @path: path to put the reference to | |
480 | * | |
481 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
482 | */ | |
dcf787f3 | 483 | void path_put(const struct path *path) |
1da177e4 | 484 | { |
1d957f9b JB |
485 | dput(path->dentry); |
486 | mntput(path->mnt); | |
1da177e4 | 487 | } |
1d957f9b | 488 | EXPORT_SYMBOL(path_put); |
1da177e4 | 489 | |
1f55a6ec AV |
490 | struct nameidata { |
491 | struct path path; | |
492 | struct qstr last; | |
493 | struct path root; | |
494 | struct inode *inode; /* path.dentry.d_inode */ | |
495 | unsigned int flags; | |
496 | unsigned seq, m_seq; | |
497 | int last_type; | |
498 | unsigned depth; | |
499 | char *saved_names[MAX_NESTED_LINKS + 1]; | |
500 | }; | |
501 | ||
19660af7 | 502 | /* |
31e6b01f | 503 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
504 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
505 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
506 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
507 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
508 | * got stuck, so ref-walk may continue from there. If this is not successful | |
509 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
510 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 511 | */ |
31e6b01f NP |
512 | |
513 | /** | |
19660af7 AV |
514 | * unlazy_walk - try to switch to ref-walk mode. |
515 | * @nd: nameidata pathwalk data | |
516 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 517 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 518 | * |
19660af7 AV |
519 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
520 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
521 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 522 | */ |
19660af7 | 523 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
524 | { |
525 | struct fs_struct *fs = current->fs; | |
526 | struct dentry *parent = nd->path.dentry; | |
527 | ||
528 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
529 | |
530 | /* | |
48a066e7 AV |
531 | * After legitimizing the bastards, terminate_walk() |
532 | * will do the right thing for non-RCU mode, and all our | |
533 | * subsequent exit cases should rcu_read_unlock() | |
534 | * before returning. Do vfsmount first; if dentry | |
535 | * can't be legitimized, just set nd->path.dentry to NULL | |
536 | * and rely on dput(NULL) being a no-op. | |
e5c832d5 | 537 | */ |
48a066e7 | 538 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) |
e5c832d5 | 539 | return -ECHILD; |
e5c832d5 | 540 | nd->flags &= ~LOOKUP_RCU; |
15570086 | 541 | |
48a066e7 AV |
542 | if (!lockref_get_not_dead(&parent->d_lockref)) { |
543 | nd->path.dentry = NULL; | |
d870b4a1 | 544 | goto out; |
48a066e7 AV |
545 | } |
546 | ||
15570086 LT |
547 | /* |
548 | * For a negative lookup, the lookup sequence point is the parents | |
549 | * sequence point, and it only needs to revalidate the parent dentry. | |
550 | * | |
551 | * For a positive lookup, we need to move both the parent and the | |
552 | * dentry from the RCU domain to be properly refcounted. And the | |
553 | * sequence number in the dentry validates *both* dentry counters, | |
554 | * since we checked the sequence number of the parent after we got | |
555 | * the child sequence number. So we know the parent must still | |
556 | * be valid if the child sequence number is still valid. | |
557 | */ | |
19660af7 | 558 | if (!dentry) { |
e5c832d5 LT |
559 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
560 | goto out; | |
19660af7 AV |
561 | BUG_ON(nd->inode != parent->d_inode); |
562 | } else { | |
e5c832d5 LT |
563 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
564 | goto out; | |
565 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
566 | goto drop_dentry; | |
19660af7 | 567 | } |
e5c832d5 LT |
568 | |
569 | /* | |
570 | * Sequence counts matched. Now make sure that the root is | |
571 | * still valid and get it if required. | |
572 | */ | |
573 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
574 | spin_lock(&fs->lock); | |
575 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
576 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
577 | path_get(&nd->root); |
578 | spin_unlock(&fs->lock); | |
579 | } | |
31e6b01f | 580 | |
8b61e74f | 581 | rcu_read_unlock(); |
31e6b01f | 582 | return 0; |
19660af7 | 583 | |
e5c832d5 LT |
584 | unlock_and_drop_dentry: |
585 | spin_unlock(&fs->lock); | |
586 | drop_dentry: | |
8b61e74f | 587 | rcu_read_unlock(); |
15570086 | 588 | dput(dentry); |
d0d27277 | 589 | goto drop_root_mnt; |
e5c832d5 | 590 | out: |
8b61e74f | 591 | rcu_read_unlock(); |
d0d27277 LT |
592 | drop_root_mnt: |
593 | if (!(nd->flags & LOOKUP_ROOT)) | |
594 | nd->root.mnt = NULL; | |
31e6b01f NP |
595 | return -ECHILD; |
596 | } | |
597 | ||
4ce16ef3 | 598 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 599 | { |
4ce16ef3 | 600 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
601 | } |
602 | ||
9f1fafee AV |
603 | /** |
604 | * complete_walk - successful completion of path walk | |
605 | * @nd: pointer nameidata | |
39159de2 | 606 | * |
9f1fafee AV |
607 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
608 | * Revalidate the final result, unless we'd already done that during | |
609 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
610 | * success, -error on failure. In case of failure caller does not | |
611 | * need to drop nd->path. | |
39159de2 | 612 | */ |
9f1fafee | 613 | static int complete_walk(struct nameidata *nd) |
39159de2 | 614 | { |
16c2cd71 | 615 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 616 | int status; |
39159de2 | 617 | |
9f1fafee AV |
618 | if (nd->flags & LOOKUP_RCU) { |
619 | nd->flags &= ~LOOKUP_RCU; | |
620 | if (!(nd->flags & LOOKUP_ROOT)) | |
621 | nd->root.mnt = NULL; | |
15570086 | 622 | |
48a066e7 | 623 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { |
8b61e74f | 624 | rcu_read_unlock(); |
48a066e7 AV |
625 | return -ECHILD; |
626 | } | |
e5c832d5 | 627 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
8b61e74f | 628 | rcu_read_unlock(); |
48a066e7 | 629 | mntput(nd->path.mnt); |
e5c832d5 LT |
630 | return -ECHILD; |
631 | } | |
632 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
8b61e74f | 633 | rcu_read_unlock(); |
e5c832d5 | 634 | dput(dentry); |
48a066e7 | 635 | mntput(nd->path.mnt); |
9f1fafee AV |
636 | return -ECHILD; |
637 | } | |
8b61e74f | 638 | rcu_read_unlock(); |
9f1fafee AV |
639 | } |
640 | ||
16c2cd71 AV |
641 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
642 | return 0; | |
643 | ||
ecf3d1f1 | 644 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
645 | return 0; |
646 | ||
ecf3d1f1 | 647 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
648 | if (status > 0) |
649 | return 0; | |
650 | ||
16c2cd71 | 651 | if (!status) |
39159de2 | 652 | status = -ESTALE; |
16c2cd71 | 653 | |
9f1fafee | 654 | path_put(&nd->path); |
39159de2 JL |
655 | return status; |
656 | } | |
657 | ||
2a737871 AV |
658 | static __always_inline void set_root(struct nameidata *nd) |
659 | { | |
7bd88377 | 660 | get_fs_root(current->fs, &nd->root); |
2a737871 AV |
661 | } |
662 | ||
6de88d72 AV |
663 | static int link_path_walk(const char *, struct nameidata *); |
664 | ||
7bd88377 | 665 | static __always_inline unsigned set_root_rcu(struct nameidata *nd) |
31e6b01f | 666 | { |
7bd88377 AV |
667 | struct fs_struct *fs = current->fs; |
668 | unsigned seq, res; | |
c28cc364 | 669 | |
7bd88377 AV |
670 | do { |
671 | seq = read_seqcount_begin(&fs->seq); | |
672 | nd->root = fs->root; | |
673 | res = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
674 | } while (read_seqcount_retry(&fs->seq, seq)); | |
675 | return res; | |
31e6b01f NP |
676 | } |
677 | ||
1d957f9b | 678 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
679 | { |
680 | dput(path->dentry); | |
4ac91378 | 681 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
682 | mntput(path->mnt); |
683 | } | |
684 | ||
7b9337aa NP |
685 | static inline void path_to_nameidata(const struct path *path, |
686 | struct nameidata *nd) | |
051d3812 | 687 | { |
31e6b01f NP |
688 | if (!(nd->flags & LOOKUP_RCU)) { |
689 | dput(nd->path.dentry); | |
690 | if (nd->path.mnt != path->mnt) | |
691 | mntput(nd->path.mnt); | |
9a229683 | 692 | } |
31e6b01f | 693 | nd->path.mnt = path->mnt; |
4ac91378 | 694 | nd->path.dentry = path->dentry; |
051d3812 IK |
695 | } |
696 | ||
b5fb63c1 CH |
697 | /* |
698 | * Helper to directly jump to a known parsed path from ->follow_link, | |
699 | * caller must have taken a reference to path beforehand. | |
700 | */ | |
701 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
702 | { | |
703 | path_put(&nd->path); | |
704 | ||
705 | nd->path = *path; | |
706 | nd->inode = nd->path.dentry->d_inode; | |
707 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
708 | } |
709 | ||
1f55a6ec AV |
710 | void nd_set_link(struct nameidata *nd, char *path) |
711 | { | |
712 | nd->saved_names[nd->depth] = path; | |
713 | } | |
714 | EXPORT_SYMBOL(nd_set_link); | |
715 | ||
716 | char *nd_get_link(struct nameidata *nd) | |
717 | { | |
718 | return nd->saved_names[nd->depth]; | |
719 | } | |
720 | EXPORT_SYMBOL(nd_get_link); | |
721 | ||
574197e0 AV |
722 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
723 | { | |
724 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 725 | if (inode->i_op->put_link) |
574197e0 AV |
726 | inode->i_op->put_link(link->dentry, nd, cookie); |
727 | path_put(link); | |
728 | } | |
729 | ||
561ec64a LT |
730 | int sysctl_protected_symlinks __read_mostly = 0; |
731 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
732 | |
733 | /** | |
734 | * may_follow_link - Check symlink following for unsafe situations | |
735 | * @link: The path of the symlink | |
55852635 | 736 | * @nd: nameidata pathwalk data |
800179c9 KC |
737 | * |
738 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
739 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
740 | * in a sticky world-writable directory. This is to protect privileged | |
741 | * processes from failing races against path names that may change out | |
742 | * from under them by way of other users creating malicious symlinks. | |
743 | * It will permit symlinks to be followed only when outside a sticky | |
744 | * world-writable directory, or when the uid of the symlink and follower | |
745 | * match, or when the directory owner matches the symlink's owner. | |
746 | * | |
747 | * Returns 0 if following the symlink is allowed, -ve on error. | |
748 | */ | |
749 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
750 | { | |
751 | const struct inode *inode; | |
752 | const struct inode *parent; | |
753 | ||
754 | if (!sysctl_protected_symlinks) | |
755 | return 0; | |
756 | ||
757 | /* Allowed if owner and follower match. */ | |
758 | inode = link->dentry->d_inode; | |
81abe27b | 759 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
760 | return 0; |
761 | ||
762 | /* Allowed if parent directory not sticky and world-writable. */ | |
763 | parent = nd->path.dentry->d_inode; | |
764 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
765 | return 0; | |
766 | ||
767 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 768 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
769 | return 0; |
770 | ||
ffd8d101 | 771 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
772 | path_put_conditional(link, nd); |
773 | path_put(&nd->path); | |
774 | return -EACCES; | |
775 | } | |
776 | ||
777 | /** | |
778 | * safe_hardlink_source - Check for safe hardlink conditions | |
779 | * @inode: the source inode to hardlink from | |
780 | * | |
781 | * Return false if at least one of the following conditions: | |
782 | * - inode is not a regular file | |
783 | * - inode is setuid | |
784 | * - inode is setgid and group-exec | |
785 | * - access failure for read and write | |
786 | * | |
787 | * Otherwise returns true. | |
788 | */ | |
789 | static bool safe_hardlink_source(struct inode *inode) | |
790 | { | |
791 | umode_t mode = inode->i_mode; | |
792 | ||
793 | /* Special files should not get pinned to the filesystem. */ | |
794 | if (!S_ISREG(mode)) | |
795 | return false; | |
796 | ||
797 | /* Setuid files should not get pinned to the filesystem. */ | |
798 | if (mode & S_ISUID) | |
799 | return false; | |
800 | ||
801 | /* Executable setgid files should not get pinned to the filesystem. */ | |
802 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
803 | return false; | |
804 | ||
805 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
806 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
807 | return false; | |
808 | ||
809 | return true; | |
810 | } | |
811 | ||
812 | /** | |
813 | * may_linkat - Check permissions for creating a hardlink | |
814 | * @link: the source to hardlink from | |
815 | * | |
816 | * Block hardlink when all of: | |
817 | * - sysctl_protected_hardlinks enabled | |
818 | * - fsuid does not match inode | |
819 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
820 | * - not CAP_FOWNER | |
821 | * | |
822 | * Returns 0 if successful, -ve on error. | |
823 | */ | |
824 | static int may_linkat(struct path *link) | |
825 | { | |
826 | const struct cred *cred; | |
827 | struct inode *inode; | |
828 | ||
829 | if (!sysctl_protected_hardlinks) | |
830 | return 0; | |
831 | ||
832 | cred = current_cred(); | |
833 | inode = link->dentry->d_inode; | |
834 | ||
835 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
836 | * otherwise, it must be a safe source. | |
837 | */ | |
81abe27b | 838 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
839 | capable(CAP_FOWNER)) |
840 | return 0; | |
841 | ||
a51d9eaa | 842 | audit_log_link_denied("linkat", link); |
800179c9 KC |
843 | return -EPERM; |
844 | } | |
845 | ||
def4af30 | 846 | static __always_inline int |
574197e0 | 847 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 848 | { |
7b9337aa | 849 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
850 | int error; |
851 | char *s; | |
1da177e4 | 852 | |
844a3917 AV |
853 | BUG_ON(nd->flags & LOOKUP_RCU); |
854 | ||
0e794589 AV |
855 | if (link->mnt == nd->path.mnt) |
856 | mntget(link->mnt); | |
857 | ||
6d7b5aae AV |
858 | error = -ELOOP; |
859 | if (unlikely(current->total_link_count >= 40)) | |
860 | goto out_put_nd_path; | |
861 | ||
574197e0 AV |
862 | cond_resched(); |
863 | current->total_link_count++; | |
864 | ||
68ac1234 | 865 | touch_atime(link); |
1da177e4 | 866 | nd_set_link(nd, NULL); |
cd4e91d3 | 867 | |
36f3b4f6 | 868 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
869 | if (error) |
870 | goto out_put_nd_path; | |
36f3b4f6 | 871 | |
86acdca1 | 872 | nd->last_type = LAST_BIND; |
def4af30 AV |
873 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
874 | error = PTR_ERR(*p); | |
6d7b5aae | 875 | if (IS_ERR(*p)) |
408ef013 | 876 | goto out_put_nd_path; |
6d7b5aae AV |
877 | |
878 | error = 0; | |
879 | s = nd_get_link(nd); | |
880 | if (s) { | |
443ed254 AV |
881 | if (unlikely(IS_ERR(s))) { |
882 | path_put(&nd->path); | |
883 | put_link(nd, link, *p); | |
884 | return PTR_ERR(s); | |
885 | } | |
886 | if (*s == '/') { | |
7bd88377 AV |
887 | if (!nd->root.mnt) |
888 | set_root(nd); | |
443ed254 AV |
889 | path_put(&nd->path); |
890 | nd->path = nd->root; | |
891 | path_get(&nd->root); | |
892 | nd->flags |= LOOKUP_JUMPED; | |
893 | } | |
894 | nd->inode = nd->path.dentry->d_inode; | |
895 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
896 | if (unlikely(error)) |
897 | put_link(nd, link, *p); | |
1da177e4 | 898 | } |
6d7b5aae AV |
899 | |
900 | return error; | |
901 | ||
902 | out_put_nd_path: | |
98f6ef64 | 903 | *p = NULL; |
6d7b5aae | 904 | path_put(&nd->path); |
6d7b5aae | 905 | path_put(link); |
1da177e4 LT |
906 | return error; |
907 | } | |
908 | ||
31e6b01f NP |
909 | static int follow_up_rcu(struct path *path) |
910 | { | |
0714a533 AV |
911 | struct mount *mnt = real_mount(path->mnt); |
912 | struct mount *parent; | |
31e6b01f NP |
913 | struct dentry *mountpoint; |
914 | ||
0714a533 AV |
915 | parent = mnt->mnt_parent; |
916 | if (&parent->mnt == path->mnt) | |
31e6b01f | 917 | return 0; |
a73324da | 918 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 919 | path->dentry = mountpoint; |
0714a533 | 920 | path->mnt = &parent->mnt; |
31e6b01f NP |
921 | return 1; |
922 | } | |
923 | ||
f015f126 DH |
924 | /* |
925 | * follow_up - Find the mountpoint of path's vfsmount | |
926 | * | |
927 | * Given a path, find the mountpoint of its source file system. | |
928 | * Replace @path with the path of the mountpoint in the parent mount. | |
929 | * Up is towards /. | |
930 | * | |
931 | * Return 1 if we went up a level and 0 if we were already at the | |
932 | * root. | |
933 | */ | |
bab77ebf | 934 | int follow_up(struct path *path) |
1da177e4 | 935 | { |
0714a533 AV |
936 | struct mount *mnt = real_mount(path->mnt); |
937 | struct mount *parent; | |
1da177e4 | 938 | struct dentry *mountpoint; |
99b7db7b | 939 | |
48a066e7 | 940 | read_seqlock_excl(&mount_lock); |
0714a533 | 941 | parent = mnt->mnt_parent; |
3c0a6163 | 942 | if (parent == mnt) { |
48a066e7 | 943 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
944 | return 0; |
945 | } | |
0714a533 | 946 | mntget(&parent->mnt); |
a73324da | 947 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 948 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
949 | dput(path->dentry); |
950 | path->dentry = mountpoint; | |
951 | mntput(path->mnt); | |
0714a533 | 952 | path->mnt = &parent->mnt; |
1da177e4 LT |
953 | return 1; |
954 | } | |
4d359507 | 955 | EXPORT_SYMBOL(follow_up); |
1da177e4 | 956 | |
b5c84bf6 | 957 | /* |
9875cf80 DH |
958 | * Perform an automount |
959 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
960 | * were called with. | |
1da177e4 | 961 | */ |
9875cf80 DH |
962 | static int follow_automount(struct path *path, unsigned flags, |
963 | bool *need_mntput) | |
31e6b01f | 964 | { |
9875cf80 | 965 | struct vfsmount *mnt; |
ea5b778a | 966 | int err; |
9875cf80 DH |
967 | |
968 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
969 | return -EREMOTE; | |
970 | ||
0ec26fd0 MS |
971 | /* We don't want to mount if someone's just doing a stat - |
972 | * unless they're stat'ing a directory and appended a '/' to | |
973 | * the name. | |
974 | * | |
975 | * We do, however, want to mount if someone wants to open or | |
976 | * create a file of any type under the mountpoint, wants to | |
977 | * traverse through the mountpoint or wants to open the | |
978 | * mounted directory. Also, autofs may mark negative dentries | |
979 | * as being automount points. These will need the attentions | |
980 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 981 | */ |
0ec26fd0 | 982 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 983 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
984 | path->dentry->d_inode) |
985 | return -EISDIR; | |
986 | ||
9875cf80 DH |
987 | current->total_link_count++; |
988 | if (current->total_link_count >= 40) | |
989 | return -ELOOP; | |
990 | ||
991 | mnt = path->dentry->d_op->d_automount(path); | |
992 | if (IS_ERR(mnt)) { | |
993 | /* | |
994 | * The filesystem is allowed to return -EISDIR here to indicate | |
995 | * it doesn't want to automount. For instance, autofs would do | |
996 | * this so that its userspace daemon can mount on this dentry. | |
997 | * | |
998 | * However, we can only permit this if it's a terminal point in | |
999 | * the path being looked up; if it wasn't then the remainder of | |
1000 | * the path is inaccessible and we should say so. | |
1001 | */ | |
49084c3b | 1002 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
1003 | return -EREMOTE; |
1004 | return PTR_ERR(mnt); | |
31e6b01f | 1005 | } |
ea5b778a | 1006 | |
9875cf80 DH |
1007 | if (!mnt) /* mount collision */ |
1008 | return 0; | |
31e6b01f | 1009 | |
8aef1884 AV |
1010 | if (!*need_mntput) { |
1011 | /* lock_mount() may release path->mnt on error */ | |
1012 | mntget(path->mnt); | |
1013 | *need_mntput = true; | |
1014 | } | |
19a167af | 1015 | err = finish_automount(mnt, path); |
9875cf80 | 1016 | |
ea5b778a DH |
1017 | switch (err) { |
1018 | case -EBUSY: | |
1019 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 1020 | return 0; |
ea5b778a | 1021 | case 0: |
8aef1884 | 1022 | path_put(path); |
ea5b778a DH |
1023 | path->mnt = mnt; |
1024 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 1025 | return 0; |
19a167af AV |
1026 | default: |
1027 | return err; | |
ea5b778a | 1028 | } |
19a167af | 1029 | |
463ffb2e AV |
1030 | } |
1031 | ||
9875cf80 DH |
1032 | /* |
1033 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 1034 | * - Flagged for transit management (autofs) |
9875cf80 DH |
1035 | * - Flagged as mountpoint |
1036 | * - Flagged as automount point | |
1037 | * | |
1038 | * This may only be called in refwalk mode. | |
1039 | * | |
1040 | * Serialization is taken care of in namespace.c | |
1041 | */ | |
1042 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 1043 | { |
8aef1884 | 1044 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
1045 | unsigned managed; |
1046 | bool need_mntput = false; | |
8aef1884 | 1047 | int ret = 0; |
9875cf80 DH |
1048 | |
1049 | /* Given that we're not holding a lock here, we retain the value in a | |
1050 | * local variable for each dentry as we look at it so that we don't see | |
1051 | * the components of that value change under us */ | |
1052 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1053 | managed &= DCACHE_MANAGED_DENTRY, | |
1054 | unlikely(managed != 0)) { | |
cc53ce53 DH |
1055 | /* Allow the filesystem to manage the transit without i_mutex |
1056 | * being held. */ | |
1057 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1058 | BUG_ON(!path->dentry->d_op); | |
1059 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1060 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1061 | if (ret < 0) |
8aef1884 | 1062 | break; |
cc53ce53 DH |
1063 | } |
1064 | ||
9875cf80 DH |
1065 | /* Transit to a mounted filesystem. */ |
1066 | if (managed & DCACHE_MOUNTED) { | |
1067 | struct vfsmount *mounted = lookup_mnt(path); | |
1068 | if (mounted) { | |
1069 | dput(path->dentry); | |
1070 | if (need_mntput) | |
1071 | mntput(path->mnt); | |
1072 | path->mnt = mounted; | |
1073 | path->dentry = dget(mounted->mnt_root); | |
1074 | need_mntput = true; | |
1075 | continue; | |
1076 | } | |
1077 | ||
1078 | /* Something is mounted on this dentry in another | |
1079 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1080 | * namespace got unmounted before lookup_mnt() could |
1081 | * get it */ | |
9875cf80 DH |
1082 | } |
1083 | ||
1084 | /* Handle an automount point */ | |
1085 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1086 | ret = follow_automount(path, flags, &need_mntput); | |
1087 | if (ret < 0) | |
8aef1884 | 1088 | break; |
9875cf80 DH |
1089 | continue; |
1090 | } | |
1091 | ||
1092 | /* We didn't change the current path point */ | |
1093 | break; | |
1da177e4 | 1094 | } |
8aef1884 AV |
1095 | |
1096 | if (need_mntput && path->mnt == mnt) | |
1097 | mntput(path->mnt); | |
1098 | if (ret == -EISDIR) | |
1099 | ret = 0; | |
a3fbbde7 | 1100 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1101 | } |
1102 | ||
cc53ce53 | 1103 | int follow_down_one(struct path *path) |
1da177e4 LT |
1104 | { |
1105 | struct vfsmount *mounted; | |
1106 | ||
1c755af4 | 1107 | mounted = lookup_mnt(path); |
1da177e4 | 1108 | if (mounted) { |
9393bd07 AV |
1109 | dput(path->dentry); |
1110 | mntput(path->mnt); | |
1111 | path->mnt = mounted; | |
1112 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1113 | return 1; |
1114 | } | |
1115 | return 0; | |
1116 | } | |
4d359507 | 1117 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 | 1118 | |
b8faf035 | 1119 | static inline int managed_dentry_rcu(struct dentry *dentry) |
62a7375e | 1120 | { |
b8faf035 N |
1121 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? |
1122 | dentry->d_op->d_manage(dentry, true) : 0; | |
62a7375e IK |
1123 | } |
1124 | ||
9875cf80 | 1125 | /* |
287548e4 AV |
1126 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1127 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1128 | */ |
1129 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1130 | struct inode **inode) |
9875cf80 | 1131 | { |
62a7375e | 1132 | for (;;) { |
c7105365 | 1133 | struct mount *mounted; |
62a7375e IK |
1134 | /* |
1135 | * Don't forget we might have a non-mountpoint managed dentry | |
1136 | * that wants to block transit. | |
1137 | */ | |
b8faf035 N |
1138 | switch (managed_dentry_rcu(path->dentry)) { |
1139 | case -ECHILD: | |
1140 | default: | |
ab90911f | 1141 | return false; |
b8faf035 N |
1142 | case -EISDIR: |
1143 | return true; | |
1144 | case 0: | |
1145 | break; | |
1146 | } | |
62a7375e IK |
1147 | |
1148 | if (!d_mountpoint(path->dentry)) | |
b8faf035 | 1149 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
62a7375e | 1150 | |
474279dc | 1151 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1152 | if (!mounted) |
1153 | break; | |
c7105365 AV |
1154 | path->mnt = &mounted->mnt; |
1155 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1156 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1157 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1158 | /* |
1159 | * Update the inode too. We don't need to re-check the | |
1160 | * dentry sequence number here after this d_inode read, | |
1161 | * because a mount-point is always pinned. | |
1162 | */ | |
1163 | *inode = path->dentry->d_inode; | |
9875cf80 | 1164 | } |
f5be3e29 | 1165 | return !read_seqretry(&mount_lock, nd->m_seq) && |
b8faf035 | 1166 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); |
287548e4 AV |
1167 | } |
1168 | ||
31e6b01f NP |
1169 | static int follow_dotdot_rcu(struct nameidata *nd) |
1170 | { | |
4023bfc9 | 1171 | struct inode *inode = nd->inode; |
7bd88377 AV |
1172 | if (!nd->root.mnt) |
1173 | set_root_rcu(nd); | |
31e6b01f | 1174 | |
9875cf80 | 1175 | while (1) { |
31e6b01f NP |
1176 | if (nd->path.dentry == nd->root.dentry && |
1177 | nd->path.mnt == nd->root.mnt) { | |
1178 | break; | |
1179 | } | |
1180 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1181 | struct dentry *old = nd->path.dentry; | |
1182 | struct dentry *parent = old->d_parent; | |
1183 | unsigned seq; | |
1184 | ||
4023bfc9 | 1185 | inode = parent->d_inode; |
31e6b01f NP |
1186 | seq = read_seqcount_begin(&parent->d_seq); |
1187 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1188 | goto failed; |
31e6b01f NP |
1189 | nd->path.dentry = parent; |
1190 | nd->seq = seq; | |
1191 | break; | |
1192 | } | |
1193 | if (!follow_up_rcu(&nd->path)) | |
1194 | break; | |
4023bfc9 | 1195 | inode = nd->path.dentry->d_inode; |
31e6b01f | 1196 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
31e6b01f | 1197 | } |
b37199e6 AV |
1198 | while (d_mountpoint(nd->path.dentry)) { |
1199 | struct mount *mounted; | |
1200 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
1201 | if (!mounted) | |
1202 | break; | |
1203 | nd->path.mnt = &mounted->mnt; | |
1204 | nd->path.dentry = mounted->mnt.mnt_root; | |
4023bfc9 | 1205 | inode = nd->path.dentry->d_inode; |
b37199e6 | 1206 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
f5be3e29 | 1207 | if (read_seqretry(&mount_lock, nd->m_seq)) |
b37199e6 AV |
1208 | goto failed; |
1209 | } | |
4023bfc9 | 1210 | nd->inode = inode; |
31e6b01f | 1211 | return 0; |
ef7562d5 AV |
1212 | |
1213 | failed: | |
1214 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1215 | if (!(nd->flags & LOOKUP_ROOT)) |
1216 | nd->root.mnt = NULL; | |
8b61e74f | 1217 | rcu_read_unlock(); |
ef7562d5 | 1218 | return -ECHILD; |
31e6b01f NP |
1219 | } |
1220 | ||
cc53ce53 DH |
1221 | /* |
1222 | * Follow down to the covering mount currently visible to userspace. At each | |
1223 | * point, the filesystem owning that dentry may be queried as to whether the | |
1224 | * caller is permitted to proceed or not. | |
cc53ce53 | 1225 | */ |
7cc90cc3 | 1226 | int follow_down(struct path *path) |
cc53ce53 DH |
1227 | { |
1228 | unsigned managed; | |
1229 | int ret; | |
1230 | ||
1231 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1232 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1233 | /* Allow the filesystem to manage the transit without i_mutex | |
1234 | * being held. | |
1235 | * | |
1236 | * We indicate to the filesystem if someone is trying to mount | |
1237 | * something here. This gives autofs the chance to deny anyone | |
1238 | * other than its daemon the right to mount on its | |
1239 | * superstructure. | |
1240 | * | |
1241 | * The filesystem may sleep at this point. | |
1242 | */ | |
1243 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1244 | BUG_ON(!path->dentry->d_op); | |
1245 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1246 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1247 | path->dentry, false); |
cc53ce53 DH |
1248 | if (ret < 0) |
1249 | return ret == -EISDIR ? 0 : ret; | |
1250 | } | |
1251 | ||
1252 | /* Transit to a mounted filesystem. */ | |
1253 | if (managed & DCACHE_MOUNTED) { | |
1254 | struct vfsmount *mounted = lookup_mnt(path); | |
1255 | if (!mounted) | |
1256 | break; | |
1257 | dput(path->dentry); | |
1258 | mntput(path->mnt); | |
1259 | path->mnt = mounted; | |
1260 | path->dentry = dget(mounted->mnt_root); | |
1261 | continue; | |
1262 | } | |
1263 | ||
1264 | /* Don't handle automount points here */ | |
1265 | break; | |
1266 | } | |
1267 | return 0; | |
1268 | } | |
4d359507 | 1269 | EXPORT_SYMBOL(follow_down); |
cc53ce53 | 1270 | |
9875cf80 DH |
1271 | /* |
1272 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1273 | */ | |
1274 | static void follow_mount(struct path *path) | |
1275 | { | |
1276 | while (d_mountpoint(path->dentry)) { | |
1277 | struct vfsmount *mounted = lookup_mnt(path); | |
1278 | if (!mounted) | |
1279 | break; | |
1280 | dput(path->dentry); | |
1281 | mntput(path->mnt); | |
1282 | path->mnt = mounted; | |
1283 | path->dentry = dget(mounted->mnt_root); | |
1284 | } | |
1285 | } | |
1286 | ||
31e6b01f | 1287 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1288 | { |
7bd88377 AV |
1289 | if (!nd->root.mnt) |
1290 | set_root(nd); | |
e518ddb7 | 1291 | |
1da177e4 | 1292 | while(1) { |
4ac91378 | 1293 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1294 | |
2a737871 AV |
1295 | if (nd->path.dentry == nd->root.dentry && |
1296 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1297 | break; |
1298 | } | |
4ac91378 | 1299 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1300 | /* rare case of legitimate dget_parent()... */ |
1301 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1302 | dput(old); |
1303 | break; | |
1304 | } | |
3088dd70 | 1305 | if (!follow_up(&nd->path)) |
1da177e4 | 1306 | break; |
1da177e4 | 1307 | } |
79ed0226 | 1308 | follow_mount(&nd->path); |
31e6b01f | 1309 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1310 | } |
1311 | ||
baa03890 | 1312 | /* |
bad61189 MS |
1313 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1314 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1315 | * returns whether i_op->lookup is necessary. | |
1316 | * | |
1317 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1318 | */ |
bad61189 | 1319 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1320 | unsigned int flags, bool *need_lookup) |
baa03890 | 1321 | { |
baa03890 | 1322 | struct dentry *dentry; |
bad61189 | 1323 | int error; |
baa03890 | 1324 | |
bad61189 MS |
1325 | *need_lookup = false; |
1326 | dentry = d_lookup(dir, name); | |
1327 | if (dentry) { | |
39e3c955 | 1328 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1329 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1330 | if (unlikely(error <= 0)) { |
1331 | if (error < 0) { | |
1332 | dput(dentry); | |
1333 | return ERR_PTR(error); | |
5542aa2f EB |
1334 | } else { |
1335 | d_invalidate(dentry); | |
bad61189 MS |
1336 | dput(dentry); |
1337 | dentry = NULL; | |
1338 | } | |
1339 | } | |
1340 | } | |
1341 | } | |
baa03890 | 1342 | |
bad61189 MS |
1343 | if (!dentry) { |
1344 | dentry = d_alloc(dir, name); | |
1345 | if (unlikely(!dentry)) | |
1346 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1347 | |
bad61189 | 1348 | *need_lookup = true; |
baa03890 NP |
1349 | } |
1350 | return dentry; | |
1351 | } | |
1352 | ||
44396f4b | 1353 | /* |
13a2c3be BF |
1354 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1355 | * unhashed. | |
bad61189 MS |
1356 | * |
1357 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1358 | */ |
bad61189 | 1359 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1360 | unsigned int flags) |
44396f4b | 1361 | { |
44396f4b JB |
1362 | struct dentry *old; |
1363 | ||
1364 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1365 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1366 | dput(dentry); |
44396f4b | 1367 | return ERR_PTR(-ENOENT); |
e188dc02 | 1368 | } |
44396f4b | 1369 | |
72bd866a | 1370 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1371 | if (unlikely(old)) { |
1372 | dput(dentry); | |
1373 | dentry = old; | |
1374 | } | |
1375 | return dentry; | |
1376 | } | |
1377 | ||
a3255546 | 1378 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1379 | struct dentry *base, unsigned int flags) |
a3255546 | 1380 | { |
bad61189 | 1381 | bool need_lookup; |
a3255546 AV |
1382 | struct dentry *dentry; |
1383 | ||
72bd866a | 1384 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1385 | if (!need_lookup) |
1386 | return dentry; | |
a3255546 | 1387 | |
72bd866a | 1388 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1389 | } |
1390 | ||
1da177e4 LT |
1391 | /* |
1392 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1393 | * small and for now I'd prefer to have fast path as straight as possible. | |
1394 | * It _is_ time-critical. | |
1395 | */ | |
e97cdc87 | 1396 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1397 | struct path *path, struct inode **inode) |
1da177e4 | 1398 | { |
4ac91378 | 1399 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1400 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1401 | int need_reval = 1; |
1402 | int status = 1; | |
9875cf80 DH |
1403 | int err; |
1404 | ||
b04f784e NP |
1405 | /* |
1406 | * Rename seqlock is not required here because in the off chance | |
1407 | * of a false negative due to a concurrent rename, we're going to | |
1408 | * do the non-racy lookup, below. | |
1409 | */ | |
31e6b01f NP |
1410 | if (nd->flags & LOOKUP_RCU) { |
1411 | unsigned seq; | |
da53be12 | 1412 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1413 | if (!dentry) |
1414 | goto unlazy; | |
1415 | ||
12f8ad4b LT |
1416 | /* |
1417 | * This sequence count validates that the inode matches | |
1418 | * the dentry name information from lookup. | |
1419 | */ | |
1420 | *inode = dentry->d_inode; | |
1421 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1422 | return -ECHILD; | |
1423 | ||
1424 | /* | |
1425 | * This sequence count validates that the parent had no | |
1426 | * changes while we did the lookup of the dentry above. | |
1427 | * | |
1428 | * The memory barrier in read_seqcount_begin of child is | |
1429 | * enough, we can use __read_seqcount_retry here. | |
1430 | */ | |
31e6b01f NP |
1431 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1432 | return -ECHILD; | |
31e6b01f | 1433 | nd->seq = seq; |
5a18fff2 | 1434 | |
24643087 | 1435 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1436 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1437 | if (unlikely(status <= 0)) { |
1438 | if (status != -ECHILD) | |
1439 | need_reval = 0; | |
1440 | goto unlazy; | |
1441 | } | |
24643087 | 1442 | } |
31e6b01f NP |
1443 | path->mnt = mnt; |
1444 | path->dentry = dentry; | |
b8faf035 N |
1445 | if (likely(__follow_mount_rcu(nd, path, inode))) |
1446 | return 0; | |
5a18fff2 | 1447 | unlazy: |
19660af7 AV |
1448 | if (unlazy_walk(nd, dentry)) |
1449 | return -ECHILD; | |
5a18fff2 | 1450 | } else { |
e97cdc87 | 1451 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1452 | } |
5a18fff2 | 1453 | |
81e6f520 AV |
1454 | if (unlikely(!dentry)) |
1455 | goto need_lookup; | |
1456 | ||
5a18fff2 | 1457 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1458 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1459 | if (unlikely(status <= 0)) { |
1460 | if (status < 0) { | |
1461 | dput(dentry); | |
1462 | return status; | |
1463 | } | |
5542aa2f EB |
1464 | d_invalidate(dentry); |
1465 | dput(dentry); | |
1466 | goto need_lookup; | |
24643087 | 1467 | } |
697f514d | 1468 | |
9875cf80 DH |
1469 | path->mnt = mnt; |
1470 | path->dentry = dentry; | |
1471 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1472 | if (unlikely(err < 0)) { |
1473 | path_put_conditional(path, nd); | |
9875cf80 | 1474 | return err; |
89312214 | 1475 | } |
a3fbbde7 AV |
1476 | if (err) |
1477 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1478 | *inode = path->dentry->d_inode; |
1da177e4 | 1479 | return 0; |
81e6f520 AV |
1480 | |
1481 | need_lookup: | |
697f514d MS |
1482 | return 1; |
1483 | } | |
1484 | ||
1485 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1486 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1487 | { |
1488 | struct dentry *dentry, *parent; | |
1489 | int err; | |
1490 | ||
1491 | parent = nd->path.dentry; | |
81e6f520 AV |
1492 | BUG_ON(nd->inode != parent->d_inode); |
1493 | ||
1494 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1495 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1496 | mutex_unlock(&parent->d_inode->i_mutex); |
1497 | if (IS_ERR(dentry)) | |
1498 | return PTR_ERR(dentry); | |
697f514d MS |
1499 | path->mnt = nd->path.mnt; |
1500 | path->dentry = dentry; | |
1501 | err = follow_managed(path, nd->flags); | |
1502 | if (unlikely(err < 0)) { | |
1503 | path_put_conditional(path, nd); | |
1504 | return err; | |
1505 | } | |
1506 | if (err) | |
1507 | nd->flags |= LOOKUP_JUMPED; | |
1508 | return 0; | |
1da177e4 LT |
1509 | } |
1510 | ||
52094c8a AV |
1511 | static inline int may_lookup(struct nameidata *nd) |
1512 | { | |
1513 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1514 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1515 | if (err != -ECHILD) |
1516 | return err; | |
19660af7 | 1517 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1518 | return -ECHILD; |
1519 | } | |
4ad5abb3 | 1520 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1521 | } |
1522 | ||
9856fa1b AV |
1523 | static inline int handle_dots(struct nameidata *nd, int type) |
1524 | { | |
1525 | if (type == LAST_DOTDOT) { | |
1526 | if (nd->flags & LOOKUP_RCU) { | |
1527 | if (follow_dotdot_rcu(nd)) | |
1528 | return -ECHILD; | |
1529 | } else | |
1530 | follow_dotdot(nd); | |
1531 | } | |
1532 | return 0; | |
1533 | } | |
1534 | ||
951361f9 AV |
1535 | static void terminate_walk(struct nameidata *nd) |
1536 | { | |
1537 | if (!(nd->flags & LOOKUP_RCU)) { | |
1538 | path_put(&nd->path); | |
1539 | } else { | |
1540 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1541 | if (!(nd->flags & LOOKUP_ROOT)) |
1542 | nd->root.mnt = NULL; | |
8b61e74f | 1543 | rcu_read_unlock(); |
951361f9 AV |
1544 | } |
1545 | } | |
1546 | ||
3ddcd056 LT |
1547 | /* |
1548 | * Do we need to follow links? We _really_ want to be able | |
1549 | * to do this check without having to look at inode->i_op, | |
1550 | * so we keep a cache of "no, this doesn't need follow_link" | |
1551 | * for the common case. | |
1552 | */ | |
b18825a7 | 1553 | static inline int should_follow_link(struct dentry *dentry, int follow) |
3ddcd056 | 1554 | { |
b18825a7 | 1555 | return unlikely(d_is_symlink(dentry)) ? follow : 0; |
3ddcd056 LT |
1556 | } |
1557 | ||
ce57dfc1 | 1558 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1559 | int follow) |
ce57dfc1 AV |
1560 | { |
1561 | struct inode *inode; | |
1562 | int err; | |
1563 | /* | |
1564 | * "." and ".." are special - ".." especially so because it has | |
1565 | * to be able to know about the current root directory and | |
1566 | * parent relationships. | |
1567 | */ | |
21b9b073 AV |
1568 | if (unlikely(nd->last_type != LAST_NORM)) |
1569 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1570 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1571 | if (unlikely(err)) { |
697f514d MS |
1572 | if (err < 0) |
1573 | goto out_err; | |
1574 | ||
cc2a5271 | 1575 | err = lookup_slow(nd, path); |
697f514d MS |
1576 | if (err < 0) |
1577 | goto out_err; | |
1578 | ||
1579 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1580 | } |
697f514d | 1581 | err = -ENOENT; |
22213318 | 1582 | if (!inode || d_is_negative(path->dentry)) |
697f514d MS |
1583 | goto out_path_put; |
1584 | ||
b18825a7 | 1585 | if (should_follow_link(path->dentry, follow)) { |
19660af7 AV |
1586 | if (nd->flags & LOOKUP_RCU) { |
1587 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1588 | err = -ECHILD; |
1589 | goto out_err; | |
19660af7 AV |
1590 | } |
1591 | } | |
ce57dfc1 AV |
1592 | BUG_ON(inode != path->dentry->d_inode); |
1593 | return 1; | |
1594 | } | |
1595 | path_to_nameidata(path, nd); | |
1596 | nd->inode = inode; | |
1597 | return 0; | |
697f514d MS |
1598 | |
1599 | out_path_put: | |
1600 | path_to_nameidata(path, nd); | |
1601 | out_err: | |
1602 | terminate_walk(nd); | |
1603 | return err; | |
ce57dfc1 AV |
1604 | } |
1605 | ||
b356379a AV |
1606 | /* |
1607 | * This limits recursive symlink follows to 8, while | |
1608 | * limiting consecutive symlinks to 40. | |
1609 | * | |
1610 | * Without that kind of total limit, nasty chains of consecutive | |
1611 | * symlinks can cause almost arbitrarily long lookups. | |
1612 | */ | |
1613 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1614 | { | |
1615 | int res; | |
1616 | ||
b356379a AV |
1617 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1618 | path_put_conditional(path, nd); | |
1619 | path_put(&nd->path); | |
1620 | return -ELOOP; | |
1621 | } | |
1a4022f8 | 1622 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1623 | |
1624 | nd->depth++; | |
1625 | current->link_count++; | |
1626 | ||
1627 | do { | |
1628 | struct path link = *path; | |
1629 | void *cookie; | |
574197e0 AV |
1630 | |
1631 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1632 | if (res) |
1633 | break; | |
21b9b073 | 1634 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1635 | put_link(nd, &link, cookie); |
b356379a AV |
1636 | } while (res > 0); |
1637 | ||
1638 | current->link_count--; | |
1639 | nd->depth--; | |
1640 | return res; | |
1641 | } | |
1642 | ||
bfcfaa77 LT |
1643 | /* |
1644 | * We can do the critical dentry name comparison and hashing | |
1645 | * operations one word at a time, but we are limited to: | |
1646 | * | |
1647 | * - Architectures with fast unaligned word accesses. We could | |
1648 | * do a "get_unaligned()" if this helps and is sufficiently | |
1649 | * fast. | |
1650 | * | |
bfcfaa77 LT |
1651 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1652 | * do not trap on the (extremely unlikely) case of a page | |
1653 | * crossing operation. | |
1654 | * | |
1655 | * - Furthermore, we need an efficient 64-bit compile for the | |
1656 | * 64-bit case in order to generate the "number of bytes in | |
1657 | * the final mask". Again, that could be replaced with a | |
1658 | * efficient population count instruction or similar. | |
1659 | */ | |
1660 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1661 | ||
f68e556e | 1662 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1663 | |
f68e556e | 1664 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1665 | |
1666 | static inline unsigned int fold_hash(unsigned long hash) | |
1667 | { | |
99d263d4 | 1668 | return hash_64(hash, 32); |
bfcfaa77 LT |
1669 | } |
1670 | ||
1671 | #else /* 32-bit case */ | |
1672 | ||
bfcfaa77 LT |
1673 | #define fold_hash(x) (x) |
1674 | ||
1675 | #endif | |
1676 | ||
1677 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1678 | { | |
1679 | unsigned long a, mask; | |
1680 | unsigned long hash = 0; | |
1681 | ||
1682 | for (;;) { | |
e419b4cc | 1683 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1684 | if (len < sizeof(unsigned long)) |
1685 | break; | |
1686 | hash += a; | |
f132c5be | 1687 | hash *= 9; |
bfcfaa77 LT |
1688 | name += sizeof(unsigned long); |
1689 | len -= sizeof(unsigned long); | |
1690 | if (!len) | |
1691 | goto done; | |
1692 | } | |
a5c21dce | 1693 | mask = bytemask_from_count(len); |
bfcfaa77 LT |
1694 | hash += mask & a; |
1695 | done: | |
1696 | return fold_hash(hash); | |
1697 | } | |
1698 | EXPORT_SYMBOL(full_name_hash); | |
1699 | ||
bfcfaa77 LT |
1700 | /* |
1701 | * Calculate the length and hash of the path component, and | |
d6bb3e90 | 1702 | * return the "hash_len" as the result. |
bfcfaa77 | 1703 | */ |
d6bb3e90 | 1704 | static inline u64 hash_name(const char *name) |
bfcfaa77 | 1705 | { |
36126f8f LT |
1706 | unsigned long a, b, adata, bdata, mask, hash, len; |
1707 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1708 | |
1709 | hash = a = 0; | |
1710 | len = -sizeof(unsigned long); | |
1711 | do { | |
1712 | hash = (hash + a) * 9; | |
1713 | len += sizeof(unsigned long); | |
e419b4cc | 1714 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1715 | b = a ^ REPEAT_BYTE('/'); |
1716 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1717 | ||
1718 | adata = prep_zero_mask(a, adata, &constants); | |
1719 | bdata = prep_zero_mask(b, bdata, &constants); | |
1720 | ||
1721 | mask = create_zero_mask(adata | bdata); | |
1722 | ||
1723 | hash += a & zero_bytemask(mask); | |
9226b5b4 | 1724 | len += find_zero(mask); |
d6bb3e90 | 1725 | return hashlen_create(fold_hash(hash), len); |
bfcfaa77 LT |
1726 | } |
1727 | ||
1728 | #else | |
1729 | ||
0145acc2 LT |
1730 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1731 | { | |
1732 | unsigned long hash = init_name_hash(); | |
1733 | while (len--) | |
1734 | hash = partial_name_hash(*name++, hash); | |
1735 | return end_name_hash(hash); | |
1736 | } | |
ae942ae7 | 1737 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1738 | |
200e9ef7 LT |
1739 | /* |
1740 | * We know there's a real path component here of at least | |
1741 | * one character. | |
1742 | */ | |
d6bb3e90 | 1743 | static inline u64 hash_name(const char *name) |
200e9ef7 LT |
1744 | { |
1745 | unsigned long hash = init_name_hash(); | |
1746 | unsigned long len = 0, c; | |
1747 | ||
1748 | c = (unsigned char)*name; | |
1749 | do { | |
1750 | len++; | |
1751 | hash = partial_name_hash(c, hash); | |
1752 | c = (unsigned char)name[len]; | |
1753 | } while (c && c != '/'); | |
d6bb3e90 | 1754 | return hashlen_create(end_name_hash(hash), len); |
200e9ef7 LT |
1755 | } |
1756 | ||
bfcfaa77 LT |
1757 | #endif |
1758 | ||
1da177e4 LT |
1759 | /* |
1760 | * Name resolution. | |
ea3834d9 PM |
1761 | * This is the basic name resolution function, turning a pathname into |
1762 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1763 | * |
ea3834d9 PM |
1764 | * Returns 0 and nd will have valid dentry and mnt on success. |
1765 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1766 | */ |
6de88d72 | 1767 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1768 | { |
1769 | struct path next; | |
1da177e4 | 1770 | int err; |
1da177e4 LT |
1771 | |
1772 | while (*name=='/') | |
1773 | name++; | |
1774 | if (!*name) | |
086e183a | 1775 | return 0; |
1da177e4 | 1776 | |
1da177e4 LT |
1777 | /* At this point we know we have a real path component. */ |
1778 | for(;;) { | |
d6bb3e90 | 1779 | u64 hash_len; |
fe479a58 | 1780 | int type; |
1da177e4 | 1781 | |
52094c8a | 1782 | err = may_lookup(nd); |
1da177e4 LT |
1783 | if (err) |
1784 | break; | |
1785 | ||
d6bb3e90 | 1786 | hash_len = hash_name(name); |
1da177e4 | 1787 | |
fe479a58 | 1788 | type = LAST_NORM; |
d6bb3e90 | 1789 | if (name[0] == '.') switch (hashlen_len(hash_len)) { |
fe479a58 | 1790 | case 2: |
200e9ef7 | 1791 | if (name[1] == '.') { |
fe479a58 | 1792 | type = LAST_DOTDOT; |
16c2cd71 AV |
1793 | nd->flags |= LOOKUP_JUMPED; |
1794 | } | |
fe479a58 AV |
1795 | break; |
1796 | case 1: | |
1797 | type = LAST_DOT; | |
1798 | } | |
5a202bcd AV |
1799 | if (likely(type == LAST_NORM)) { |
1800 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1801 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1802 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
a060dc50 | 1803 | struct qstr this = { { .hash_len = hash_len }, .name = name }; |
da53be12 | 1804 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1805 | if (err < 0) |
1806 | break; | |
d6bb3e90 LT |
1807 | hash_len = this.hash_len; |
1808 | name = this.name; | |
5a202bcd AV |
1809 | } |
1810 | } | |
fe479a58 | 1811 | |
d6bb3e90 LT |
1812 | nd->last.hash_len = hash_len; |
1813 | nd->last.name = name; | |
5f4a6a69 AV |
1814 | nd->last_type = type; |
1815 | ||
d6bb3e90 LT |
1816 | name += hashlen_len(hash_len); |
1817 | if (!*name) | |
5f4a6a69 | 1818 | return 0; |
200e9ef7 LT |
1819 | /* |
1820 | * If it wasn't NUL, we know it was '/'. Skip that | |
1821 | * slash, and continue until no more slashes. | |
1822 | */ | |
1823 | do { | |
d6bb3e90 LT |
1824 | name++; |
1825 | } while (unlikely(*name == '/')); | |
1826 | if (!*name) | |
5f4a6a69 AV |
1827 | return 0; |
1828 | ||
21b9b073 | 1829 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1830 | if (err < 0) |
1831 | return err; | |
1da177e4 | 1832 | |
ce57dfc1 | 1833 | if (err) { |
b356379a | 1834 | err = nested_symlink(&next, nd); |
1da177e4 | 1835 | if (err) |
a7472bab | 1836 | return err; |
31e6b01f | 1837 | } |
44b1d530 | 1838 | if (!d_can_lookup(nd->path.dentry)) { |
5f4a6a69 AV |
1839 | err = -ENOTDIR; |
1840 | break; | |
1841 | } | |
1da177e4 | 1842 | } |
951361f9 | 1843 | terminate_walk(nd); |
1da177e4 LT |
1844 | return err; |
1845 | } | |
1846 | ||
70e9b357 AV |
1847 | static int path_init(int dfd, const char *name, unsigned int flags, |
1848 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1849 | { |
1850 | int retval = 0; | |
31e6b01f NP |
1851 | |
1852 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1853 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1854 | nd->depth = 0; |
5b6ca027 | 1855 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
1856 | struct dentry *root = nd->root.dentry; |
1857 | struct inode *inode = root->d_inode; | |
73d049a4 | 1858 | if (*name) { |
44b1d530 | 1859 | if (!d_can_lookup(root)) |
73d049a4 AV |
1860 | return -ENOTDIR; |
1861 | retval = inode_permission(inode, MAY_EXEC); | |
1862 | if (retval) | |
1863 | return retval; | |
1864 | } | |
5b6ca027 AV |
1865 | nd->path = nd->root; |
1866 | nd->inode = inode; | |
1867 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1868 | rcu_read_lock(); |
5b6ca027 | 1869 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1870 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1871 | } else { |
1872 | path_get(&nd->path); | |
1873 | } | |
1874 | return 0; | |
1875 | } | |
1876 | ||
31e6b01f | 1877 | nd->root.mnt = NULL; |
31e6b01f | 1878 | |
48a066e7 | 1879 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1880 | if (*name=='/') { |
e41f7d4e | 1881 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1882 | rcu_read_lock(); |
7bd88377 | 1883 | nd->seq = set_root_rcu(nd); |
e41f7d4e AV |
1884 | } else { |
1885 | set_root(nd); | |
1886 | path_get(&nd->root); | |
1887 | } | |
1888 | nd->path = nd->root; | |
31e6b01f | 1889 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1890 | if (flags & LOOKUP_RCU) { |
1891 | struct fs_struct *fs = current->fs; | |
1892 | unsigned seq; | |
31e6b01f | 1893 | |
8b61e74f | 1894 | rcu_read_lock(); |
c28cc364 | 1895 | |
e41f7d4e AV |
1896 | do { |
1897 | seq = read_seqcount_begin(&fs->seq); | |
1898 | nd->path = fs->pwd; | |
1899 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1900 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1901 | } else { | |
1902 | get_fs_pwd(current->fs, &nd->path); | |
1903 | } | |
31e6b01f | 1904 | } else { |
582aa64a | 1905 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1906 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1907 | struct dentry *dentry; |
1908 | ||
2903ff01 AV |
1909 | if (!f.file) |
1910 | return -EBADF; | |
31e6b01f | 1911 | |
2903ff01 | 1912 | dentry = f.file->f_path.dentry; |
31e6b01f | 1913 | |
f52e0c11 | 1914 | if (*name) { |
44b1d530 | 1915 | if (!d_can_lookup(dentry)) { |
2903ff01 AV |
1916 | fdput(f); |
1917 | return -ENOTDIR; | |
1918 | } | |
f52e0c11 | 1919 | } |
31e6b01f | 1920 | |
2903ff01 | 1921 | nd->path = f.file->f_path; |
e41f7d4e | 1922 | if (flags & LOOKUP_RCU) { |
9c225f26 | 1923 | if (f.flags & FDPUT_FPUT) |
2903ff01 | 1924 | *fp = f.file; |
e41f7d4e | 1925 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1926 | rcu_read_lock(); |
e41f7d4e | 1927 | } else { |
2903ff01 AV |
1928 | path_get(&nd->path); |
1929 | fdput(f); | |
e41f7d4e | 1930 | } |
31e6b01f | 1931 | } |
31e6b01f | 1932 | |
31e6b01f | 1933 | nd->inode = nd->path.dentry->d_inode; |
4023bfc9 AV |
1934 | if (!(flags & LOOKUP_RCU)) |
1935 | return 0; | |
1936 | if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) | |
1937 | return 0; | |
1938 | if (!(nd->flags & LOOKUP_ROOT)) | |
1939 | nd->root.mnt = NULL; | |
1940 | rcu_read_unlock(); | |
1941 | return -ECHILD; | |
9b4a9b14 AV |
1942 | } |
1943 | ||
bd92d7fe AV |
1944 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1945 | { | |
1946 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1947 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1948 | ||
1949 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1950 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1951 | } |
1952 | ||
9b4a9b14 | 1953 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1954 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1955 | unsigned int flags, struct nameidata *nd) |
1956 | { | |
70e9b357 | 1957 | struct file *base = NULL; |
bd92d7fe AV |
1958 | struct path path; |
1959 | int err; | |
31e6b01f NP |
1960 | |
1961 | /* | |
1962 | * Path walking is largely split up into 2 different synchronisation | |
1963 | * schemes, rcu-walk and ref-walk (explained in | |
1964 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1965 | * path walk code, but some things particularly setup, cleanup, and | |
1966 | * following mounts are sufficiently divergent that functions are | |
1967 | * duplicated. Typically there is a function foo(), and its RCU | |
1968 | * analogue, foo_rcu(). | |
1969 | * | |
1970 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1971 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1972 | * be handled by restarting a traditional ref-walk (which will always | |
1973 | * be able to complete). | |
1974 | */ | |
bd92d7fe | 1975 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1976 | |
bd92d7fe | 1977 | if (unlikely(err)) |
115cbfdc | 1978 | goto out; |
ee0827cd AV |
1979 | |
1980 | current->total_link_count = 0; | |
bd92d7fe AV |
1981 | err = link_path_walk(name, nd); |
1982 | ||
1983 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1984 | err = lookup_last(nd, &path); |
1985 | while (err > 0) { | |
1986 | void *cookie; | |
1987 | struct path link = path; | |
800179c9 KC |
1988 | err = may_follow_link(&link, nd); |
1989 | if (unlikely(err)) | |
1990 | break; | |
bd92d7fe | 1991 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1992 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1993 | if (err) |
1994 | break; | |
1995 | err = lookup_last(nd, &path); | |
574197e0 | 1996 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1997 | } |
1998 | } | |
ee0827cd | 1999 | |
9f1fafee AV |
2000 | if (!err) |
2001 | err = complete_walk(nd); | |
bd92d7fe AV |
2002 | |
2003 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
44b1d530 | 2004 | if (!d_can_lookup(nd->path.dentry)) { |
bd92d7fe | 2005 | path_put(&nd->path); |
bd23a539 | 2006 | err = -ENOTDIR; |
bd92d7fe AV |
2007 | } |
2008 | } | |
16c2cd71 | 2009 | |
115cbfdc | 2010 | out: |
70e9b357 AV |
2011 | if (base) |
2012 | fput(base); | |
ee0827cd | 2013 | |
5b6ca027 | 2014 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
2015 | path_put(&nd->root); |
2016 | nd->root.mnt = NULL; | |
2017 | } | |
bd92d7fe | 2018 | return err; |
ee0827cd | 2019 | } |
31e6b01f | 2020 | |
873f1eed | 2021 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
2022 | unsigned int flags, struct nameidata *nd) |
2023 | { | |
873f1eed | 2024 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 2025 | if (unlikely(retval == -ECHILD)) |
873f1eed | 2026 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 2027 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
2028 | retval = path_lookupat(dfd, name->name, |
2029 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 2030 | |
f78570dd | 2031 | if (likely(!retval)) |
adb5c247 | 2032 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 2033 | return retval; |
1da177e4 LT |
2034 | } |
2035 | ||
873f1eed JL |
2036 | static int do_path_lookup(int dfd, const char *name, |
2037 | unsigned int flags, struct nameidata *nd) | |
2038 | { | |
2039 | struct filename filename = { .name = name }; | |
2040 | ||
2041 | return filename_lookup(dfd, &filename, flags, nd); | |
2042 | } | |
2043 | ||
79714f72 AV |
2044 | /* does lookup, returns the object with parent locked */ |
2045 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 2046 | { |
79714f72 AV |
2047 | struct nameidata nd; |
2048 | struct dentry *d; | |
2049 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
2050 | if (err) | |
2051 | return ERR_PTR(err); | |
2052 | if (nd.last_type != LAST_NORM) { | |
2053 | path_put(&nd.path); | |
2054 | return ERR_PTR(-EINVAL); | |
2055 | } | |
2056 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 2057 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
2058 | if (IS_ERR(d)) { |
2059 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2060 | path_put(&nd.path); | |
2061 | return d; | |
2062 | } | |
2063 | *path = nd.path; | |
2064 | return d; | |
5590ff0d UD |
2065 | } |
2066 | ||
d1811465 AV |
2067 | int kern_path(const char *name, unsigned int flags, struct path *path) |
2068 | { | |
2069 | struct nameidata nd; | |
2070 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
2071 | if (!res) | |
2072 | *path = nd.path; | |
2073 | return res; | |
2074 | } | |
4d359507 | 2075 | EXPORT_SYMBOL(kern_path); |
d1811465 | 2076 | |
16f18200 JJS |
2077 | /** |
2078 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2079 | * @dentry: pointer to dentry of the base directory | |
2080 | * @mnt: pointer to vfs mount of the base directory | |
2081 | * @name: pointer to file name | |
2082 | * @flags: lookup flags | |
e0a01249 | 2083 | * @path: pointer to struct path to fill |
16f18200 JJS |
2084 | */ |
2085 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2086 | const char *name, unsigned int flags, | |
e0a01249 | 2087 | struct path *path) |
16f18200 | 2088 | { |
e0a01249 AV |
2089 | struct nameidata nd; |
2090 | int err; | |
2091 | nd.root.dentry = dentry; | |
2092 | nd.root.mnt = mnt; | |
2093 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2094 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2095 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2096 | if (!err) | |
2097 | *path = nd.path; | |
2098 | return err; | |
16f18200 | 2099 | } |
4d359507 | 2100 | EXPORT_SYMBOL(vfs_path_lookup); |
16f18200 | 2101 | |
057f6c01 JM |
2102 | /* |
2103 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2104 | * needs parent already locked. Doesn't follow mounts. | |
2105 | * SMP-safe. | |
2106 | */ | |
eead1911 | 2107 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2108 | { |
72bd866a | 2109 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2110 | } |
2111 | ||
eead1911 | 2112 | /** |
a6b91919 | 2113 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2114 | * @name: pathname component to lookup |
2115 | * @base: base directory to lookup from | |
2116 | * @len: maximum length @len should be interpreted to | |
2117 | * | |
a6b91919 RD |
2118 | * Note that this routine is purely a helper for filesystem usage and should |
2119 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2120 | * nameidata argument is passed to the filesystem methods and a filesystem |
2121 | * using this helper needs to be prepared for that. | |
2122 | */ | |
057f6c01 JM |
2123 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2124 | { | |
057f6c01 | 2125 | struct qstr this; |
6a96ba54 | 2126 | unsigned int c; |
cda309de | 2127 | int err; |
057f6c01 | 2128 | |
2f9092e1 DW |
2129 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2130 | ||
6a96ba54 AV |
2131 | this.name = name; |
2132 | this.len = len; | |
0145acc2 | 2133 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2134 | if (!len) |
2135 | return ERR_PTR(-EACCES); | |
2136 | ||
21d8a15a AV |
2137 | if (unlikely(name[0] == '.')) { |
2138 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2139 | return ERR_PTR(-EACCES); | |
2140 | } | |
2141 | ||
6a96ba54 AV |
2142 | while (len--) { |
2143 | c = *(const unsigned char *)name++; | |
2144 | if (c == '/' || c == '\0') | |
2145 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2146 | } |
5a202bcd AV |
2147 | /* |
2148 | * See if the low-level filesystem might want | |
2149 | * to use its own hash.. | |
2150 | */ | |
2151 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2152 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2153 | if (err < 0) |
2154 | return ERR_PTR(err); | |
2155 | } | |
eead1911 | 2156 | |
cda309de MS |
2157 | err = inode_permission(base->d_inode, MAY_EXEC); |
2158 | if (err) | |
2159 | return ERR_PTR(err); | |
2160 | ||
72bd866a | 2161 | return __lookup_hash(&this, base, 0); |
057f6c01 | 2162 | } |
4d359507 | 2163 | EXPORT_SYMBOL(lookup_one_len); |
057f6c01 | 2164 | |
1fa1e7f6 AW |
2165 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2166 | struct path *path, int *empty) | |
1da177e4 | 2167 | { |
2d8f3038 | 2168 | struct nameidata nd; |
91a27b2a | 2169 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2170 | int err = PTR_ERR(tmp); |
1da177e4 | 2171 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2172 | |
2173 | BUG_ON(flags & LOOKUP_PARENT); | |
2174 | ||
873f1eed | 2175 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2176 | putname(tmp); |
2d8f3038 AV |
2177 | if (!err) |
2178 | *path = nd.path; | |
1da177e4 LT |
2179 | } |
2180 | return err; | |
2181 | } | |
2182 | ||
1fa1e7f6 AW |
2183 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2184 | struct path *path) | |
2185 | { | |
f7493e5d | 2186 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 | 2187 | } |
4d359507 | 2188 | EXPORT_SYMBOL(user_path_at); |
1fa1e7f6 | 2189 | |
873f1eed JL |
2190 | /* |
2191 | * NB: most callers don't do anything directly with the reference to the | |
2192 | * to struct filename, but the nd->last pointer points into the name string | |
2193 | * allocated by getname. So we must hold the reference to it until all | |
2194 | * path-walking is complete. | |
2195 | */ | |
91a27b2a | 2196 | static struct filename * |
9e790bd6 JL |
2197 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2198 | unsigned int flags) | |
2ad94ae6 | 2199 | { |
91a27b2a | 2200 | struct filename *s = getname(path); |
2ad94ae6 AV |
2201 | int error; |
2202 | ||
9e790bd6 JL |
2203 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2204 | flags &= LOOKUP_REVAL; | |
2205 | ||
2ad94ae6 | 2206 | if (IS_ERR(s)) |
91a27b2a | 2207 | return s; |
2ad94ae6 | 2208 | |
9e790bd6 | 2209 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2210 | if (error) { |
2ad94ae6 | 2211 | putname(s); |
91a27b2a JL |
2212 | return ERR_PTR(error); |
2213 | } | |
2ad94ae6 | 2214 | |
91a27b2a | 2215 | return s; |
2ad94ae6 AV |
2216 | } |
2217 | ||
8033426e | 2218 | /** |
197df04c | 2219 | * mountpoint_last - look up last component for umount |
8033426e JL |
2220 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2221 | * @path: pointer to container for result | |
2222 | * | |
2223 | * This is a special lookup_last function just for umount. In this case, we | |
2224 | * need to resolve the path without doing any revalidation. | |
2225 | * | |
2226 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2227 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2228 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2229 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2230 | * bogus and it doesn't exist. | |
2231 | * | |
2232 | * Returns: | |
2233 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2234 | * lookup found a negative dentry. The nd->path reference will also be | |
2235 | * put in this case. | |
2236 | * | |
2237 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2238 | * symlink that needs to be followed. "path" will also be populated. | |
2239 | * The nd->path reference will also be put. | |
2240 | * | |
2241 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2242 | * that needs to be followed. "path" will be populated with the path | |
2243 | * to the link, and nd->path will *not* be put. | |
2244 | */ | |
2245 | static int | |
197df04c | 2246 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2247 | { |
2248 | int error = 0; | |
2249 | struct dentry *dentry; | |
2250 | struct dentry *dir = nd->path.dentry; | |
2251 | ||
35759521 AV |
2252 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2253 | if (nd->flags & LOOKUP_RCU) { | |
2254 | if (unlazy_walk(nd, NULL)) { | |
2255 | error = -ECHILD; | |
2256 | goto out; | |
2257 | } | |
8033426e JL |
2258 | } |
2259 | ||
2260 | nd->flags &= ~LOOKUP_PARENT; | |
2261 | ||
2262 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2263 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2264 | if (error) |
2265 | goto out; | |
2266 | dentry = dget(nd->path.dentry); | |
2267 | goto done; | |
8033426e JL |
2268 | } |
2269 | ||
2270 | mutex_lock(&dir->d_inode->i_mutex); | |
2271 | dentry = d_lookup(dir, &nd->last); | |
2272 | if (!dentry) { | |
2273 | /* | |
2274 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2275 | * so that means that this dentry is probably a symlink or the | |
2276 | * path doesn't actually point to a mounted dentry. | |
2277 | */ | |
2278 | dentry = d_alloc(dir, &nd->last); | |
2279 | if (!dentry) { | |
2280 | error = -ENOMEM; | |
bcceeeba | 2281 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2282 | goto out; |
8033426e | 2283 | } |
35759521 AV |
2284 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2285 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2286 | if (IS_ERR(dentry)) { |
2287 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2288 | goto out; |
bcceeeba | 2289 | } |
8033426e JL |
2290 | } |
2291 | mutex_unlock(&dir->d_inode->i_mutex); | |
2292 | ||
35759521 | 2293 | done: |
22213318 | 2294 | if (!dentry->d_inode || d_is_negative(dentry)) { |
35759521 AV |
2295 | error = -ENOENT; |
2296 | dput(dentry); | |
2297 | goto out; | |
8033426e | 2298 | } |
35759521 | 2299 | path->dentry = dentry; |
295dc39d | 2300 | path->mnt = nd->path.mnt; |
b18825a7 | 2301 | if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) |
35759521 | 2302 | return 1; |
295dc39d | 2303 | mntget(path->mnt); |
35759521 AV |
2304 | follow_mount(path); |
2305 | error = 0; | |
2306 | out: | |
8033426e JL |
2307 | terminate_walk(nd); |
2308 | return error; | |
2309 | } | |
2310 | ||
2311 | /** | |
197df04c | 2312 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2313 | * @dfd: directory file descriptor to start walk from |
2314 | * @name: full pathname to walk | |
606d6fe3 | 2315 | * @path: pointer to container for result |
8033426e | 2316 | * @flags: lookup flags |
8033426e JL |
2317 | * |
2318 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2319 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2320 | */ |
2321 | static int | |
197df04c | 2322 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2323 | { |
2324 | struct file *base = NULL; | |
2325 | struct nameidata nd; | |
2326 | int err; | |
2327 | ||
2328 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2329 | if (unlikely(err)) | |
115cbfdc | 2330 | goto out; |
8033426e JL |
2331 | |
2332 | current->total_link_count = 0; | |
2333 | err = link_path_walk(name, &nd); | |
2334 | if (err) | |
2335 | goto out; | |
2336 | ||
197df04c | 2337 | err = mountpoint_last(&nd, path); |
8033426e JL |
2338 | while (err > 0) { |
2339 | void *cookie; | |
2340 | struct path link = *path; | |
2341 | err = may_follow_link(&link, &nd); | |
2342 | if (unlikely(err)) | |
2343 | break; | |
2344 | nd.flags |= LOOKUP_PARENT; | |
2345 | err = follow_link(&link, &nd, &cookie); | |
2346 | if (err) | |
2347 | break; | |
197df04c | 2348 | err = mountpoint_last(&nd, path); |
8033426e JL |
2349 | put_link(&nd, &link, cookie); |
2350 | } | |
2351 | out: | |
2352 | if (base) | |
2353 | fput(base); | |
2354 | ||
2355 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2356 | path_put(&nd.root); | |
2357 | ||
2358 | return err; | |
2359 | } | |
2360 | ||
2d864651 AV |
2361 | static int |
2362 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2363 | unsigned int flags) | |
2364 | { | |
2365 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2366 | if (unlikely(error == -ECHILD)) | |
2367 | error = path_mountpoint(dfd, s->name, path, flags); | |
2368 | if (unlikely(error == -ESTALE)) | |
2369 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2370 | if (likely(!error)) | |
2371 | audit_inode(s, path->dentry, 0); | |
2372 | return error; | |
2373 | } | |
2374 | ||
8033426e | 2375 | /** |
197df04c | 2376 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2377 | * @dfd: directory file descriptor |
2378 | * @name: pathname from userland | |
2379 | * @flags: lookup flags | |
2380 | * @path: pointer to container to hold result | |
2381 | * | |
2382 | * A umount is a special case for path walking. We're not actually interested | |
2383 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2384 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2385 | * and avoid revalidating the last component. | |
2386 | * | |
2387 | * Returns 0 and populates "path" on success. | |
2388 | */ | |
2389 | int | |
197df04c | 2390 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2391 | struct path *path) |
2392 | { | |
2393 | struct filename *s = getname(name); | |
2394 | int error; | |
8033426e JL |
2395 | if (IS_ERR(s)) |
2396 | return PTR_ERR(s); | |
2d864651 | 2397 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2398 | putname(s); |
2399 | return error; | |
2400 | } | |
2401 | ||
2d864651 AV |
2402 | int |
2403 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2404 | unsigned int flags) | |
2405 | { | |
2406 | struct filename s = {.name = name}; | |
2407 | return filename_mountpoint(dfd, &s, path, flags); | |
2408 | } | |
2409 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2410 | ||
cbdf35bc | 2411 | int __check_sticky(struct inode *dir, struct inode *inode) |
1da177e4 | 2412 | { |
8e96e3b7 | 2413 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2414 | |
8e96e3b7 | 2415 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2416 | return 0; |
8e96e3b7 | 2417 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2418 | return 0; |
23adbe12 | 2419 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); |
1da177e4 | 2420 | } |
cbdf35bc | 2421 | EXPORT_SYMBOL(__check_sticky); |
1da177e4 LT |
2422 | |
2423 | /* | |
2424 | * Check whether we can remove a link victim from directory dir, check | |
2425 | * whether the type of victim is right. | |
2426 | * 1. We can't do it if dir is read-only (done in permission()) | |
2427 | * 2. We should have write and exec permissions on dir | |
2428 | * 3. We can't remove anything from append-only dir | |
2429 | * 4. We can't do anything with immutable dir (done in permission()) | |
2430 | * 5. If the sticky bit on dir is set we should either | |
2431 | * a. be owner of dir, or | |
2432 | * b. be owner of victim, or | |
2433 | * c. have CAP_FOWNER capability | |
2434 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2435 | * links pointing to it. | |
2436 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2437 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2438 | * 9. We can't remove a root or mountpoint. | |
2439 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2440 | * nfs_async_unlink(). | |
2441 | */ | |
b18825a7 | 2442 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2443 | { |
b18825a7 | 2444 | struct inode *inode = victim->d_inode; |
1da177e4 LT |
2445 | int error; |
2446 | ||
b18825a7 | 2447 | if (d_is_negative(victim)) |
1da177e4 | 2448 | return -ENOENT; |
b18825a7 | 2449 | BUG_ON(!inode); |
1da177e4 LT |
2450 | |
2451 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2452 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2453 | |
f419a2e3 | 2454 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2455 | if (error) |
2456 | return error; | |
2457 | if (IS_APPEND(dir)) | |
2458 | return -EPERM; | |
b18825a7 DH |
2459 | |
2460 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2461 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2462 | return -EPERM; |
2463 | if (isdir) { | |
44b1d530 | 2464 | if (!d_is_dir(victim)) |
1da177e4 LT |
2465 | return -ENOTDIR; |
2466 | if (IS_ROOT(victim)) | |
2467 | return -EBUSY; | |
44b1d530 | 2468 | } else if (d_is_dir(victim)) |
1da177e4 LT |
2469 | return -EISDIR; |
2470 | if (IS_DEADDIR(dir)) | |
2471 | return -ENOENT; | |
2472 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2473 | return -EBUSY; | |
2474 | return 0; | |
2475 | } | |
2476 | ||
2477 | /* Check whether we can create an object with dentry child in directory | |
2478 | * dir. | |
2479 | * 1. We can't do it if child already exists (open has special treatment for | |
2480 | * this case, but since we are inlined it's OK) | |
2481 | * 2. We can't do it if dir is read-only (done in permission()) | |
2482 | * 3. We should have write and exec permissions on dir | |
2483 | * 4. We can't do it if dir is immutable (done in permission()) | |
2484 | */ | |
a95164d9 | 2485 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2486 | { |
14e972b4 | 2487 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2488 | if (child->d_inode) |
2489 | return -EEXIST; | |
2490 | if (IS_DEADDIR(dir)) | |
2491 | return -ENOENT; | |
f419a2e3 | 2492 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2493 | } |
2494 | ||
1da177e4 LT |
2495 | /* |
2496 | * p1 and p2 should be directories on the same fs. | |
2497 | */ | |
2498 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2499 | { | |
2500 | struct dentry *p; | |
2501 | ||
2502 | if (p1 == p2) { | |
f2eace23 | 2503 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2504 | return NULL; |
2505 | } | |
2506 | ||
a11f3a05 | 2507 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2508 | |
e2761a11 OH |
2509 | p = d_ancestor(p2, p1); |
2510 | if (p) { | |
2511 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2512 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2513 | return p; | |
1da177e4 LT |
2514 | } |
2515 | ||
e2761a11 OH |
2516 | p = d_ancestor(p1, p2); |
2517 | if (p) { | |
2518 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2519 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2520 | return p; | |
1da177e4 LT |
2521 | } |
2522 | ||
f2eace23 | 2523 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
d1b72cc6 | 2524 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); |
1da177e4 LT |
2525 | return NULL; |
2526 | } | |
4d359507 | 2527 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
2528 | |
2529 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2530 | { | |
1b1dcc1b | 2531 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2532 | if (p1 != p2) { |
1b1dcc1b | 2533 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2534 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2535 | } |
2536 | } | |
4d359507 | 2537 | EXPORT_SYMBOL(unlock_rename); |
1da177e4 | 2538 | |
4acdaf27 | 2539 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2540 | bool want_excl) |
1da177e4 | 2541 | { |
a95164d9 | 2542 | int error = may_create(dir, dentry); |
1da177e4 LT |
2543 | if (error) |
2544 | return error; | |
2545 | ||
acfa4380 | 2546 | if (!dir->i_op->create) |
1da177e4 LT |
2547 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2548 | mode &= S_IALLUGO; | |
2549 | mode |= S_IFREG; | |
2550 | error = security_inode_create(dir, dentry, mode); | |
2551 | if (error) | |
2552 | return error; | |
312b63fb | 2553 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2554 | if (!error) |
f38aa942 | 2555 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2556 | return error; |
2557 | } | |
4d359507 | 2558 | EXPORT_SYMBOL(vfs_create); |
1da177e4 | 2559 | |
73d049a4 | 2560 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2561 | { |
3fb64190 | 2562 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2563 | struct inode *inode = dentry->d_inode; |
2564 | int error; | |
2565 | ||
bcda7652 AV |
2566 | /* O_PATH? */ |
2567 | if (!acc_mode) | |
2568 | return 0; | |
2569 | ||
1da177e4 LT |
2570 | if (!inode) |
2571 | return -ENOENT; | |
2572 | ||
c8fe8f30 CH |
2573 | switch (inode->i_mode & S_IFMT) { |
2574 | case S_IFLNK: | |
1da177e4 | 2575 | return -ELOOP; |
c8fe8f30 CH |
2576 | case S_IFDIR: |
2577 | if (acc_mode & MAY_WRITE) | |
2578 | return -EISDIR; | |
2579 | break; | |
2580 | case S_IFBLK: | |
2581 | case S_IFCHR: | |
3fb64190 | 2582 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2583 | return -EACCES; |
c8fe8f30 CH |
2584 | /*FALLTHRU*/ |
2585 | case S_IFIFO: | |
2586 | case S_IFSOCK: | |
1da177e4 | 2587 | flag &= ~O_TRUNC; |
c8fe8f30 | 2588 | break; |
4a3fd211 | 2589 | } |
b41572e9 | 2590 | |
3fb64190 | 2591 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2592 | if (error) |
2593 | return error; | |
6146f0d5 | 2594 | |
1da177e4 LT |
2595 | /* |
2596 | * An append-only file must be opened in append mode for writing. | |
2597 | */ | |
2598 | if (IS_APPEND(inode)) { | |
8737c930 | 2599 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2600 | return -EPERM; |
1da177e4 | 2601 | if (flag & O_TRUNC) |
7715b521 | 2602 | return -EPERM; |
1da177e4 LT |
2603 | } |
2604 | ||
2605 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2606 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2607 | return -EPERM; |
1da177e4 | 2608 | |
f3c7691e | 2609 | return 0; |
7715b521 | 2610 | } |
1da177e4 | 2611 | |
e1181ee6 | 2612 | static int handle_truncate(struct file *filp) |
7715b521 | 2613 | { |
e1181ee6 | 2614 | struct path *path = &filp->f_path; |
7715b521 AV |
2615 | struct inode *inode = path->dentry->d_inode; |
2616 | int error = get_write_access(inode); | |
2617 | if (error) | |
2618 | return error; | |
2619 | /* | |
2620 | * Refuse to truncate files with mandatory locks held on them. | |
2621 | */ | |
d7a06983 | 2622 | error = locks_verify_locked(filp); |
7715b521 | 2623 | if (!error) |
ea0d3ab2 | 2624 | error = security_path_truncate(path); |
7715b521 AV |
2625 | if (!error) { |
2626 | error = do_truncate(path->dentry, 0, | |
2627 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2628 | filp); |
7715b521 AV |
2629 | } |
2630 | put_write_access(inode); | |
acd0c935 | 2631 | return error; |
1da177e4 LT |
2632 | } |
2633 | ||
d57999e1 DH |
2634 | static inline int open_to_namei_flags(int flag) |
2635 | { | |
8a5e929d AV |
2636 | if ((flag & O_ACCMODE) == 3) |
2637 | flag--; | |
d57999e1 DH |
2638 | return flag; |
2639 | } | |
2640 | ||
d18e9008 MS |
2641 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2642 | { | |
2643 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2644 | if (error) | |
2645 | return error; | |
2646 | ||
2647 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2648 | if (error) | |
2649 | return error; | |
2650 | ||
2651 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2652 | } | |
2653 | ||
1acf0af9 DH |
2654 | /* |
2655 | * Attempt to atomically look up, create and open a file from a negative | |
2656 | * dentry. | |
2657 | * | |
2658 | * Returns 0 if successful. The file will have been created and attached to | |
2659 | * @file by the filesystem calling finish_open(). | |
2660 | * | |
2661 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2662 | * caller will need to perform the open themselves. @path will have been | |
2663 | * updated to point to the new dentry. This may be negative. | |
2664 | * | |
2665 | * Returns an error code otherwise. | |
2666 | */ | |
2675a4eb AV |
2667 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2668 | struct path *path, struct file *file, | |
2669 | const struct open_flags *op, | |
64894cf8 | 2670 | bool got_write, bool need_lookup, |
2675a4eb | 2671 | int *opened) |
d18e9008 MS |
2672 | { |
2673 | struct inode *dir = nd->path.dentry->d_inode; | |
2674 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2675 | umode_t mode; | |
2676 | int error; | |
2677 | int acc_mode; | |
d18e9008 MS |
2678 | int create_error = 0; |
2679 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2680 | bool excl; |
d18e9008 MS |
2681 | |
2682 | BUG_ON(dentry->d_inode); | |
2683 | ||
2684 | /* Don't create child dentry for a dead directory. */ | |
2685 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2686 | error = -ENOENT; |
d18e9008 MS |
2687 | goto out; |
2688 | } | |
2689 | ||
62b259d8 | 2690 | mode = op->mode; |
d18e9008 MS |
2691 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2692 | mode &= ~current_umask(); | |
2693 | ||
116cc022 MS |
2694 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2695 | if (excl) | |
d18e9008 | 2696 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2697 | |
2698 | /* | |
2699 | * Checking write permission is tricky, bacuse we don't know if we are | |
2700 | * going to actually need it: O_CREAT opens should work as long as the | |
2701 | * file exists. But checking existence breaks atomicity. The trick is | |
2702 | * to check access and if not granted clear O_CREAT from the flags. | |
2703 | * | |
2704 | * Another problem is returing the "right" error value (e.g. for an | |
2705 | * O_EXCL open we want to return EEXIST not EROFS). | |
2706 | */ | |
64894cf8 AV |
2707 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2708 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2709 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2710 | /* |
2711 | * No O_CREATE -> atomicity not a requirement -> fall | |
2712 | * back to lookup + open | |
2713 | */ | |
2714 | goto no_open; | |
2715 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2716 | /* Fall back and fail with the right error */ | |
64894cf8 | 2717 | create_error = -EROFS; |
d18e9008 MS |
2718 | goto no_open; |
2719 | } else { | |
2720 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2721 | create_error = -EROFS; |
d18e9008 MS |
2722 | open_flag &= ~O_CREAT; |
2723 | } | |
2724 | } | |
2725 | ||
2726 | if (open_flag & O_CREAT) { | |
38227f78 | 2727 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2728 | if (error) { |
2729 | create_error = error; | |
2730 | if (open_flag & O_EXCL) | |
2731 | goto no_open; | |
2732 | open_flag &= ~O_CREAT; | |
2733 | } | |
2734 | } | |
2735 | ||
2736 | if (nd->flags & LOOKUP_DIRECTORY) | |
2737 | open_flag |= O_DIRECTORY; | |
2738 | ||
30d90494 AV |
2739 | file->f_path.dentry = DENTRY_NOT_SET; |
2740 | file->f_path.mnt = nd->path.mnt; | |
2741 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2742 | opened); |
d9585277 | 2743 | if (error < 0) { |
d9585277 AV |
2744 | if (create_error && error == -ENOENT) |
2745 | error = create_error; | |
d18e9008 MS |
2746 | goto out; |
2747 | } | |
2748 | ||
d9585277 | 2749 | if (error) { /* returned 1, that is */ |
30d90494 | 2750 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2751 | error = -EIO; |
d18e9008 MS |
2752 | goto out; |
2753 | } | |
30d90494 | 2754 | if (file->f_path.dentry) { |
d18e9008 | 2755 | dput(dentry); |
30d90494 | 2756 | dentry = file->f_path.dentry; |
d18e9008 | 2757 | } |
03da633a AV |
2758 | if (*opened & FILE_CREATED) |
2759 | fsnotify_create(dir, dentry); | |
2760 | if (!dentry->d_inode) { | |
2761 | WARN_ON(*opened & FILE_CREATED); | |
2762 | if (create_error) { | |
2763 | error = create_error; | |
2764 | goto out; | |
2765 | } | |
2766 | } else { | |
2767 | if (excl && !(*opened & FILE_CREATED)) { | |
2768 | error = -EEXIST; | |
2769 | goto out; | |
2770 | } | |
62b2ce96 | 2771 | } |
d18e9008 MS |
2772 | goto looked_up; |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * We didn't have the inode before the open, so check open permission | |
2777 | * here. | |
2778 | */ | |
03da633a AV |
2779 | acc_mode = op->acc_mode; |
2780 | if (*opened & FILE_CREATED) { | |
2781 | WARN_ON(!(open_flag & O_CREAT)); | |
2782 | fsnotify_create(dir, dentry); | |
2783 | acc_mode = MAY_OPEN; | |
2784 | } | |
2675a4eb AV |
2785 | error = may_open(&file->f_path, acc_mode, open_flag); |
2786 | if (error) | |
2787 | fput(file); | |
d18e9008 MS |
2788 | |
2789 | out: | |
2790 | dput(dentry); | |
2675a4eb | 2791 | return error; |
d18e9008 | 2792 | |
d18e9008 MS |
2793 | no_open: |
2794 | if (need_lookup) { | |
72bd866a | 2795 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2796 | if (IS_ERR(dentry)) |
2675a4eb | 2797 | return PTR_ERR(dentry); |
d18e9008 MS |
2798 | |
2799 | if (create_error) { | |
2800 | int open_flag = op->open_flag; | |
2801 | ||
2675a4eb | 2802 | error = create_error; |
d18e9008 MS |
2803 | if ((open_flag & O_EXCL)) { |
2804 | if (!dentry->d_inode) | |
2805 | goto out; | |
2806 | } else if (!dentry->d_inode) { | |
2807 | goto out; | |
2808 | } else if ((open_flag & O_TRUNC) && | |
2809 | S_ISREG(dentry->d_inode->i_mode)) { | |
2810 | goto out; | |
2811 | } | |
2812 | /* will fail later, go on to get the right error */ | |
2813 | } | |
2814 | } | |
2815 | looked_up: | |
2816 | path->dentry = dentry; | |
2817 | path->mnt = nd->path.mnt; | |
2675a4eb | 2818 | return 1; |
d18e9008 MS |
2819 | } |
2820 | ||
d58ffd35 | 2821 | /* |
1acf0af9 | 2822 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2823 | * |
2824 | * Must be called with i_mutex held on parent. | |
2825 | * | |
1acf0af9 DH |
2826 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2827 | * opened. In this case the file will be returned attached to @file. | |
2828 | * | |
2829 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2830 | * and creations will have been performed and the dentry returned in @path will | |
2831 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2832 | * specified then a negative dentry may be returned. | |
2833 | * | |
2834 | * An error code is returned otherwise. | |
2835 | * | |
2836 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2837 | * cleared otherwise prior to returning. | |
d58ffd35 | 2838 | */ |
2675a4eb AV |
2839 | static int lookup_open(struct nameidata *nd, struct path *path, |
2840 | struct file *file, | |
2841 | const struct open_flags *op, | |
64894cf8 | 2842 | bool got_write, int *opened) |
d58ffd35 MS |
2843 | { |
2844 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2845 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2846 | struct dentry *dentry; |
2847 | int error; | |
54ef4872 | 2848 | bool need_lookup; |
d58ffd35 | 2849 | |
47237687 | 2850 | *opened &= ~FILE_CREATED; |
201f956e | 2851 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2852 | if (IS_ERR(dentry)) |
2675a4eb | 2853 | return PTR_ERR(dentry); |
d58ffd35 | 2854 | |
d18e9008 MS |
2855 | /* Cached positive dentry: will open in f_op->open */ |
2856 | if (!need_lookup && dentry->d_inode) | |
2857 | goto out_no_open; | |
2858 | ||
2859 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2860 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2861 | need_lookup, opened); |
d18e9008 MS |
2862 | } |
2863 | ||
54ef4872 MS |
2864 | if (need_lookup) { |
2865 | BUG_ON(dentry->d_inode); | |
2866 | ||
72bd866a | 2867 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2868 | if (IS_ERR(dentry)) |
2675a4eb | 2869 | return PTR_ERR(dentry); |
54ef4872 MS |
2870 | } |
2871 | ||
d58ffd35 MS |
2872 | /* Negative dentry, just create the file */ |
2873 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2874 | umode_t mode = op->mode; | |
2875 | if (!IS_POSIXACL(dir->d_inode)) | |
2876 | mode &= ~current_umask(); | |
2877 | /* | |
2878 | * This write is needed to ensure that a | |
2879 | * rw->ro transition does not occur between | |
2880 | * the time when the file is created and when | |
2881 | * a permanent write count is taken through | |
015c3bbc | 2882 | * the 'struct file' in finish_open(). |
d58ffd35 | 2883 | */ |
64894cf8 AV |
2884 | if (!got_write) { |
2885 | error = -EROFS; | |
d58ffd35 | 2886 | goto out_dput; |
64894cf8 | 2887 | } |
47237687 | 2888 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2889 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2890 | if (error) | |
2891 | goto out_dput; | |
312b63fb AV |
2892 | error = vfs_create(dir->d_inode, dentry, mode, |
2893 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2894 | if (error) |
2895 | goto out_dput; | |
2896 | } | |
d18e9008 | 2897 | out_no_open: |
d58ffd35 MS |
2898 | path->dentry = dentry; |
2899 | path->mnt = nd->path.mnt; | |
2675a4eb | 2900 | return 1; |
d58ffd35 MS |
2901 | |
2902 | out_dput: | |
2903 | dput(dentry); | |
2675a4eb | 2904 | return error; |
d58ffd35 MS |
2905 | } |
2906 | ||
31e6b01f | 2907 | /* |
fe2d35ff | 2908 | * Handle the last step of open() |
31e6b01f | 2909 | */ |
2675a4eb AV |
2910 | static int do_last(struct nameidata *nd, struct path *path, |
2911 | struct file *file, const struct open_flags *op, | |
669abf4e | 2912 | int *opened, struct filename *name) |
fb1cc555 | 2913 | { |
a1e28038 | 2914 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2915 | int open_flag = op->open_flag; |
77d660a8 | 2916 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2917 | bool got_write = false; |
bcda7652 | 2918 | int acc_mode = op->acc_mode; |
a1eb3315 | 2919 | struct inode *inode; |
77d660a8 | 2920 | bool symlink_ok = false; |
16b1c1cd MS |
2921 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2922 | bool retried = false; | |
16c2cd71 | 2923 | int error; |
1f36f774 | 2924 | |
c3e380b0 AV |
2925 | nd->flags &= ~LOOKUP_PARENT; |
2926 | nd->flags |= op->intent; | |
2927 | ||
bc77daa7 | 2928 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2929 | error = handle_dots(nd, nd->last_type); |
2930 | if (error) | |
2675a4eb | 2931 | return error; |
e83db167 | 2932 | goto finish_open; |
1f36f774 | 2933 | } |
67ee3ad2 | 2934 | |
ca344a89 | 2935 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2936 | if (nd->last.name[nd->last.len]) |
2937 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2938 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2939 | symlink_ok = true; |
fe2d35ff | 2940 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2941 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2942 | if (likely(!error)) |
2943 | goto finish_lookup; | |
2944 | ||
2945 | if (error < 0) | |
2675a4eb | 2946 | goto out; |
71574865 MS |
2947 | |
2948 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2949 | } else { |
2950 | /* create side of things */ | |
2951 | /* | |
2952 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2953 | * has been cleared when we got to the last component we are | |
2954 | * about to look up | |
2955 | */ | |
2956 | error = complete_walk(nd); | |
2957 | if (error) | |
2675a4eb | 2958 | return error; |
fe2d35ff | 2959 | |
33e2208a | 2960 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2961 | error = -EISDIR; |
2962 | /* trailing slashes? */ | |
2963 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2964 | goto out; |
b6183df7 | 2965 | } |
a2c36b45 | 2966 | |
16b1c1cd | 2967 | retry_lookup: |
64894cf8 AV |
2968 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2969 | error = mnt_want_write(nd->path.mnt); | |
2970 | if (!error) | |
2971 | got_write = true; | |
2972 | /* | |
2973 | * do _not_ fail yet - we might not need that or fail with | |
2974 | * a different error; let lookup_open() decide; we'll be | |
2975 | * dropping this one anyway. | |
2976 | */ | |
2977 | } | |
a1e28038 | 2978 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2979 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2980 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2981 | |
2675a4eb AV |
2982 | if (error <= 0) { |
2983 | if (error) | |
d18e9008 MS |
2984 | goto out; |
2985 | ||
47237687 | 2986 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2987 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2988 | will_truncate = false; |
d18e9008 | 2989 | |
adb5c247 | 2990 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2991 | goto opened; |
2992 | } | |
fb1cc555 | 2993 | |
47237687 | 2994 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2995 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2996 | open_flag &= ~O_TRUNC; |
77d660a8 | 2997 | will_truncate = false; |
bcda7652 | 2998 | acc_mode = MAY_OPEN; |
d58ffd35 | 2999 | path_to_nameidata(path, nd); |
e83db167 | 3000 | goto finish_open_created; |
fb1cc555 AV |
3001 | } |
3002 | ||
3003 | /* | |
3134f37e | 3004 | * create/update audit record if it already exists. |
fb1cc555 | 3005 | */ |
b18825a7 | 3006 | if (d_is_positive(path->dentry)) |
adb5c247 | 3007 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 3008 | |
d18e9008 MS |
3009 | /* |
3010 | * If atomic_open() acquired write access it is dropped now due to | |
3011 | * possible mount and symlink following (this might be optimized away if | |
3012 | * necessary...) | |
3013 | */ | |
64894cf8 | 3014 | if (got_write) { |
d18e9008 | 3015 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3016 | got_write = false; |
d18e9008 MS |
3017 | } |
3018 | ||
fb1cc555 | 3019 | error = -EEXIST; |
f8310c59 | 3020 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
3021 | goto exit_dput; |
3022 | ||
9875cf80 DH |
3023 | error = follow_managed(path, nd->flags); |
3024 | if (error < 0) | |
3025 | goto exit_dput; | |
fb1cc555 | 3026 | |
a3fbbde7 AV |
3027 | if (error) |
3028 | nd->flags |= LOOKUP_JUMPED; | |
3029 | ||
decf3400 MS |
3030 | BUG_ON(nd->flags & LOOKUP_RCU); |
3031 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
3032 | finish_lookup: |
3033 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 3034 | error = -ENOENT; |
22213318 | 3035 | if (!inode || d_is_negative(path->dentry)) { |
54c33e7f | 3036 | path_to_nameidata(path, nd); |
2675a4eb | 3037 | goto out; |
54c33e7f | 3038 | } |
9e67f361 | 3039 | |
b18825a7 | 3040 | if (should_follow_link(path->dentry, !symlink_ok)) { |
d45ea867 MS |
3041 | if (nd->flags & LOOKUP_RCU) { |
3042 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
3043 | error = -ECHILD; | |
2675a4eb | 3044 | goto out; |
d45ea867 MS |
3045 | } |
3046 | } | |
3047 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 3048 | return 1; |
d45ea867 | 3049 | } |
fb1cc555 | 3050 | |
16b1c1cd MS |
3051 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
3052 | path_to_nameidata(path, nd); | |
3053 | } else { | |
3054 | save_parent.dentry = nd->path.dentry; | |
3055 | save_parent.mnt = mntget(path->mnt); | |
3056 | nd->path.dentry = path->dentry; | |
3057 | ||
3058 | } | |
decf3400 | 3059 | nd->inode = inode; |
a3fbbde7 | 3060 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 3061 | finish_open: |
a3fbbde7 | 3062 | error = complete_walk(nd); |
16b1c1cd MS |
3063 | if (error) { |
3064 | path_put(&save_parent); | |
2675a4eb | 3065 | return error; |
16b1c1cd | 3066 | } |
bc77daa7 | 3067 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 3068 | error = -EISDIR; |
44b1d530 | 3069 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) |
2675a4eb | 3070 | goto out; |
af2f5542 | 3071 | error = -ENOTDIR; |
44b1d530 | 3072 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) |
2675a4eb | 3073 | goto out; |
6c0d46c4 | 3074 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 3075 | will_truncate = false; |
6c0d46c4 | 3076 | |
0f9d1a10 AV |
3077 | if (will_truncate) { |
3078 | error = mnt_want_write(nd->path.mnt); | |
3079 | if (error) | |
2675a4eb | 3080 | goto out; |
64894cf8 | 3081 | got_write = true; |
0f9d1a10 | 3082 | } |
e83db167 | 3083 | finish_open_created: |
bcda7652 | 3084 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3085 | if (error) |
2675a4eb | 3086 | goto out; |
4aa7c634 MS |
3087 | |
3088 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ | |
3089 | error = vfs_open(&nd->path, file, current_cred()); | |
3090 | if (!error) { | |
3091 | *opened |= FILE_OPENED; | |
3092 | } else { | |
30d90494 | 3093 | if (error == -EOPENSTALE) |
f60dc3db | 3094 | goto stale_open; |
015c3bbc | 3095 | goto out; |
f60dc3db | 3096 | } |
a8277b9b | 3097 | opened: |
2675a4eb | 3098 | error = open_check_o_direct(file); |
015c3bbc MS |
3099 | if (error) |
3100 | goto exit_fput; | |
3034a146 | 3101 | error = ima_file_check(file, op->acc_mode, *opened); |
aa4caadb MS |
3102 | if (error) |
3103 | goto exit_fput; | |
3104 | ||
3105 | if (will_truncate) { | |
2675a4eb | 3106 | error = handle_truncate(file); |
aa4caadb MS |
3107 | if (error) |
3108 | goto exit_fput; | |
0f9d1a10 | 3109 | } |
ca344a89 | 3110 | out: |
64894cf8 | 3111 | if (got_write) |
0f9d1a10 | 3112 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3113 | path_put(&save_parent); |
e276ae67 | 3114 | terminate_walk(nd); |
2675a4eb | 3115 | return error; |
fb1cc555 | 3116 | |
fb1cc555 AV |
3117 | exit_dput: |
3118 | path_put_conditional(path, nd); | |
ca344a89 | 3119 | goto out; |
015c3bbc | 3120 | exit_fput: |
2675a4eb AV |
3121 | fput(file); |
3122 | goto out; | |
015c3bbc | 3123 | |
f60dc3db MS |
3124 | stale_open: |
3125 | /* If no saved parent or already retried then can't retry */ | |
3126 | if (!save_parent.dentry || retried) | |
3127 | goto out; | |
3128 | ||
3129 | BUG_ON(save_parent.dentry != dir); | |
3130 | path_put(&nd->path); | |
3131 | nd->path = save_parent; | |
3132 | nd->inode = dir->d_inode; | |
3133 | save_parent.mnt = NULL; | |
3134 | save_parent.dentry = NULL; | |
64894cf8 | 3135 | if (got_write) { |
f60dc3db | 3136 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3137 | got_write = false; |
f60dc3db MS |
3138 | } |
3139 | retried = true; | |
3140 | goto retry_lookup; | |
fb1cc555 AV |
3141 | } |
3142 | ||
60545d0d AV |
3143 | static int do_tmpfile(int dfd, struct filename *pathname, |
3144 | struct nameidata *nd, int flags, | |
3145 | const struct open_flags *op, | |
3146 | struct file *file, int *opened) | |
3147 | { | |
3148 | static const struct qstr name = QSTR_INIT("/", 1); | |
3149 | struct dentry *dentry, *child; | |
3150 | struct inode *dir; | |
3151 | int error = path_lookupat(dfd, pathname->name, | |
3152 | flags | LOOKUP_DIRECTORY, nd); | |
3153 | if (unlikely(error)) | |
3154 | return error; | |
3155 | error = mnt_want_write(nd->path.mnt); | |
3156 | if (unlikely(error)) | |
3157 | goto out; | |
3158 | /* we want directory to be writable */ | |
3159 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3160 | if (error) | |
3161 | goto out2; | |
3162 | dentry = nd->path.dentry; | |
3163 | dir = dentry->d_inode; | |
3164 | if (!dir->i_op->tmpfile) { | |
3165 | error = -EOPNOTSUPP; | |
3166 | goto out2; | |
3167 | } | |
3168 | child = d_alloc(dentry, &name); | |
3169 | if (unlikely(!child)) { | |
3170 | error = -ENOMEM; | |
3171 | goto out2; | |
3172 | } | |
3173 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3174 | nd->flags |= op->intent; | |
3175 | dput(nd->path.dentry); | |
3176 | nd->path.dentry = child; | |
3177 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3178 | if (error) | |
3179 | goto out2; | |
3180 | audit_inode(pathname, nd->path.dentry, 0); | |
69a91c23 ER |
3181 | /* Don't check for other permissions, the inode was just created */ |
3182 | error = may_open(&nd->path, MAY_OPEN, op->open_flag); | |
60545d0d AV |
3183 | if (error) |
3184 | goto out2; | |
3185 | file->f_path.mnt = nd->path.mnt; | |
3186 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3187 | if (error) | |
3188 | goto out2; | |
3189 | error = open_check_o_direct(file); | |
f4e0c30c | 3190 | if (error) { |
60545d0d | 3191 | fput(file); |
f4e0c30c AV |
3192 | } else if (!(op->open_flag & O_EXCL)) { |
3193 | struct inode *inode = file_inode(file); | |
3194 | spin_lock(&inode->i_lock); | |
3195 | inode->i_state |= I_LINKABLE; | |
3196 | spin_unlock(&inode->i_lock); | |
3197 | } | |
60545d0d AV |
3198 | out2: |
3199 | mnt_drop_write(nd->path.mnt); | |
3200 | out: | |
3201 | path_put(&nd->path); | |
3202 | return error; | |
3203 | } | |
3204 | ||
669abf4e | 3205 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3206 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3207 | { |
fe2d35ff | 3208 | struct file *base = NULL; |
30d90494 | 3209 | struct file *file; |
9850c056 | 3210 | struct path path; |
47237687 | 3211 | int opened = 0; |
13aab428 | 3212 | int error; |
31e6b01f | 3213 | |
30d90494 | 3214 | file = get_empty_filp(); |
1afc99be AV |
3215 | if (IS_ERR(file)) |
3216 | return file; | |
31e6b01f | 3217 | |
30d90494 | 3218 | file->f_flags = op->open_flag; |
31e6b01f | 3219 | |
bb458c64 | 3220 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3221 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3222 | goto out; | |
3223 | } | |
3224 | ||
669abf4e | 3225 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3226 | if (unlikely(error)) |
2675a4eb | 3227 | goto out; |
31e6b01f | 3228 | |
fe2d35ff | 3229 | current->total_link_count = 0; |
669abf4e | 3230 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3231 | if (unlikely(error)) |
2675a4eb | 3232 | goto out; |
1da177e4 | 3233 | |
2675a4eb AV |
3234 | error = do_last(nd, &path, file, op, &opened, pathname); |
3235 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3236 | struct path link = path; |
def4af30 | 3237 | void *cookie; |
574197e0 | 3238 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3239 | path_put_conditional(&path, nd); |
3240 | path_put(&nd->path); | |
2675a4eb | 3241 | error = -ELOOP; |
40b39136 AV |
3242 | break; |
3243 | } | |
800179c9 KC |
3244 | error = may_follow_link(&link, nd); |
3245 | if (unlikely(error)) | |
3246 | break; | |
73d049a4 AV |
3247 | nd->flags |= LOOKUP_PARENT; |
3248 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3249 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3250 | if (unlikely(error)) |
2675a4eb AV |
3251 | break; |
3252 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3253 | put_link(nd, &link, cookie); |
806b681c | 3254 | } |
10fa8e62 | 3255 | out: |
73d049a4 AV |
3256 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3257 | path_put(&nd->root); | |
fe2d35ff AV |
3258 | if (base) |
3259 | fput(base); | |
2675a4eb AV |
3260 | if (!(opened & FILE_OPENED)) { |
3261 | BUG_ON(!error); | |
30d90494 | 3262 | put_filp(file); |
16b1c1cd | 3263 | } |
2675a4eb AV |
3264 | if (unlikely(error)) { |
3265 | if (error == -EOPENSTALE) { | |
3266 | if (flags & LOOKUP_RCU) | |
3267 | error = -ECHILD; | |
3268 | else | |
3269 | error = -ESTALE; | |
3270 | } | |
3271 | file = ERR_PTR(error); | |
3272 | } | |
3273 | return file; | |
1da177e4 LT |
3274 | } |
3275 | ||
669abf4e | 3276 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3277 | const struct open_flags *op) |
13aab428 | 3278 | { |
73d049a4 | 3279 | struct nameidata nd; |
f9652e10 | 3280 | int flags = op->lookup_flags; |
13aab428 AV |
3281 | struct file *filp; |
3282 | ||
73d049a4 | 3283 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3284 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3285 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3286 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3287 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3288 | return filp; |
3289 | } | |
3290 | ||
73d049a4 | 3291 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3292 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3293 | { |
3294 | struct nameidata nd; | |
3295 | struct file *file; | |
669abf4e | 3296 | struct filename filename = { .name = name }; |
f9652e10 | 3297 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3298 | |
3299 | nd.root.mnt = mnt; | |
3300 | nd.root.dentry = dentry; | |
3301 | ||
b18825a7 | 3302 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3303 | return ERR_PTR(-ELOOP); |
3304 | ||
669abf4e | 3305 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3306 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3307 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3308 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3309 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3310 | return file; |
3311 | } | |
3312 | ||
1ac12b4b JL |
3313 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3314 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3315 | { |
c663e5d8 | 3316 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3317 | struct nameidata nd; |
c30dabfe | 3318 | int err2; |
1ac12b4b JL |
3319 | int error; |
3320 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3321 | ||
3322 | /* | |
3323 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3324 | * other flags passed in are ignored! | |
3325 | */ | |
3326 | lookup_flags &= LOOKUP_REVAL; | |
3327 | ||
3328 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3329 | if (error) |
3330 | return ERR_PTR(error); | |
1da177e4 | 3331 | |
c663e5d8 CH |
3332 | /* |
3333 | * Yucky last component or no last component at all? | |
3334 | * (foo/., foo/.., /////) | |
3335 | */ | |
ed75e95d AV |
3336 | if (nd.last_type != LAST_NORM) |
3337 | goto out; | |
3338 | nd.flags &= ~LOOKUP_PARENT; | |
3339 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3340 | |
c30dabfe JK |
3341 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3342 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3343 | /* |
3344 | * Do the final lookup. | |
3345 | */ | |
ed75e95d AV |
3346 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3347 | dentry = lookup_hash(&nd); | |
1da177e4 | 3348 | if (IS_ERR(dentry)) |
a8104a9f | 3349 | goto unlock; |
c663e5d8 | 3350 | |
a8104a9f | 3351 | error = -EEXIST; |
b18825a7 | 3352 | if (d_is_positive(dentry)) |
a8104a9f | 3353 | goto fail; |
b18825a7 | 3354 | |
c663e5d8 CH |
3355 | /* |
3356 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3357 | * From the vfs_mknod() POV we just have a negative dentry - | |
3358 | * all is fine. Let's be bastards - you had / on the end, you've | |
3359 | * been asking for (non-existent) directory. -ENOENT for you. | |
3360 | */ | |
ed75e95d | 3361 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3362 | error = -ENOENT; |
ed75e95d | 3363 | goto fail; |
e9baf6e5 | 3364 | } |
c30dabfe JK |
3365 | if (unlikely(err2)) { |
3366 | error = err2; | |
a8104a9f | 3367 | goto fail; |
c30dabfe | 3368 | } |
ed75e95d | 3369 | *path = nd.path; |
1da177e4 | 3370 | return dentry; |
1da177e4 | 3371 | fail: |
a8104a9f AV |
3372 | dput(dentry); |
3373 | dentry = ERR_PTR(error); | |
3374 | unlock: | |
ed75e95d | 3375 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3376 | if (!err2) |
3377 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3378 | out: |
3379 | path_put(&nd.path); | |
1da177e4 LT |
3380 | return dentry; |
3381 | } | |
dae6ad8f AV |
3382 | EXPORT_SYMBOL(kern_path_create); |
3383 | ||
921a1650 AV |
3384 | void done_path_create(struct path *path, struct dentry *dentry) |
3385 | { | |
3386 | dput(dentry); | |
3387 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3388 | mnt_drop_write(path->mnt); |
921a1650 AV |
3389 | path_put(path); |
3390 | } | |
3391 | EXPORT_SYMBOL(done_path_create); | |
3392 | ||
1ac12b4b JL |
3393 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3394 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3395 | { |
91a27b2a | 3396 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3397 | struct dentry *res; |
3398 | if (IS_ERR(tmp)) | |
3399 | return ERR_CAST(tmp); | |
1ac12b4b | 3400 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3401 | putname(tmp); |
3402 | return res; | |
3403 | } | |
3404 | EXPORT_SYMBOL(user_path_create); | |
3405 | ||
1a67aafb | 3406 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3407 | { |
a95164d9 | 3408 | int error = may_create(dir, dentry); |
1da177e4 LT |
3409 | |
3410 | if (error) | |
3411 | return error; | |
3412 | ||
975d6b39 | 3413 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3414 | return -EPERM; |
3415 | ||
acfa4380 | 3416 | if (!dir->i_op->mknod) |
1da177e4 LT |
3417 | return -EPERM; |
3418 | ||
08ce5f16 SH |
3419 | error = devcgroup_inode_mknod(mode, dev); |
3420 | if (error) | |
3421 | return error; | |
3422 | ||
1da177e4 LT |
3423 | error = security_inode_mknod(dir, dentry, mode, dev); |
3424 | if (error) | |
3425 | return error; | |
3426 | ||
1da177e4 | 3427 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3428 | if (!error) |
f38aa942 | 3429 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3430 | return error; |
3431 | } | |
4d359507 | 3432 | EXPORT_SYMBOL(vfs_mknod); |
1da177e4 | 3433 | |
f69aac00 | 3434 | static int may_mknod(umode_t mode) |
463c3197 DH |
3435 | { |
3436 | switch (mode & S_IFMT) { | |
3437 | case S_IFREG: | |
3438 | case S_IFCHR: | |
3439 | case S_IFBLK: | |
3440 | case S_IFIFO: | |
3441 | case S_IFSOCK: | |
3442 | case 0: /* zero mode translates to S_IFREG */ | |
3443 | return 0; | |
3444 | case S_IFDIR: | |
3445 | return -EPERM; | |
3446 | default: | |
3447 | return -EINVAL; | |
3448 | } | |
3449 | } | |
3450 | ||
8208a22b | 3451 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3452 | unsigned, dev) |
1da177e4 | 3453 | { |
2ad94ae6 | 3454 | struct dentry *dentry; |
dae6ad8f AV |
3455 | struct path path; |
3456 | int error; | |
972567f1 | 3457 | unsigned int lookup_flags = 0; |
1da177e4 | 3458 | |
8e4bfca1 AV |
3459 | error = may_mknod(mode); |
3460 | if (error) | |
3461 | return error; | |
972567f1 JL |
3462 | retry: |
3463 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3464 | if (IS_ERR(dentry)) |
3465 | return PTR_ERR(dentry); | |
2ad94ae6 | 3466 | |
dae6ad8f | 3467 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3468 | mode &= ~current_umask(); |
dae6ad8f | 3469 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3470 | if (error) |
a8104a9f | 3471 | goto out; |
463c3197 | 3472 | switch (mode & S_IFMT) { |
1da177e4 | 3473 | case 0: case S_IFREG: |
312b63fb | 3474 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3475 | break; |
3476 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3477 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3478 | new_decode_dev(dev)); |
3479 | break; | |
3480 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3481 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3482 | break; |
1da177e4 | 3483 | } |
a8104a9f | 3484 | out: |
921a1650 | 3485 | done_path_create(&path, dentry); |
972567f1 JL |
3486 | if (retry_estale(error, lookup_flags)) { |
3487 | lookup_flags |= LOOKUP_REVAL; | |
3488 | goto retry; | |
3489 | } | |
1da177e4 LT |
3490 | return error; |
3491 | } | |
3492 | ||
8208a22b | 3493 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3494 | { |
3495 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3496 | } | |
3497 | ||
18bb1db3 | 3498 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3499 | { |
a95164d9 | 3500 | int error = may_create(dir, dentry); |
8de52778 | 3501 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3502 | |
3503 | if (error) | |
3504 | return error; | |
3505 | ||
acfa4380 | 3506 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3507 | return -EPERM; |
3508 | ||
3509 | mode &= (S_IRWXUGO|S_ISVTX); | |
3510 | error = security_inode_mkdir(dir, dentry, mode); | |
3511 | if (error) | |
3512 | return error; | |
3513 | ||
8de52778 AV |
3514 | if (max_links && dir->i_nlink >= max_links) |
3515 | return -EMLINK; | |
3516 | ||
1da177e4 | 3517 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3518 | if (!error) |
f38aa942 | 3519 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3520 | return error; |
3521 | } | |
4d359507 | 3522 | EXPORT_SYMBOL(vfs_mkdir); |
1da177e4 | 3523 | |
a218d0fd | 3524 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3525 | { |
6902d925 | 3526 | struct dentry *dentry; |
dae6ad8f AV |
3527 | struct path path; |
3528 | int error; | |
b76d8b82 | 3529 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3530 | |
b76d8b82 JL |
3531 | retry: |
3532 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3533 | if (IS_ERR(dentry)) |
dae6ad8f | 3534 | return PTR_ERR(dentry); |
1da177e4 | 3535 | |
dae6ad8f | 3536 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3537 | mode &= ~current_umask(); |
dae6ad8f | 3538 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3539 | if (!error) |
3540 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3541 | done_path_create(&path, dentry); |
b76d8b82 JL |
3542 | if (retry_estale(error, lookup_flags)) { |
3543 | lookup_flags |= LOOKUP_REVAL; | |
3544 | goto retry; | |
3545 | } | |
1da177e4 LT |
3546 | return error; |
3547 | } | |
3548 | ||
a218d0fd | 3549 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3550 | { |
3551 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3552 | } | |
3553 | ||
1da177e4 | 3554 | /* |
a71905f0 | 3555 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3556 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3557 | * dentry, and if that is true (possibly after pruning the dcache), |
3558 | * then we drop the dentry now. | |
1da177e4 LT |
3559 | * |
3560 | * A low-level filesystem can, if it choses, legally | |
3561 | * do a | |
3562 | * | |
3563 | * if (!d_unhashed(dentry)) | |
3564 | * return -EBUSY; | |
3565 | * | |
3566 | * if it cannot handle the case of removing a directory | |
3567 | * that is still in use by something else.. | |
3568 | */ | |
3569 | void dentry_unhash(struct dentry *dentry) | |
3570 | { | |
dc168427 | 3571 | shrink_dcache_parent(dentry); |
1da177e4 | 3572 | spin_lock(&dentry->d_lock); |
98474236 | 3573 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3574 | __d_drop(dentry); |
3575 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 3576 | } |
4d359507 | 3577 | EXPORT_SYMBOL(dentry_unhash); |
1da177e4 LT |
3578 | |
3579 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3580 | { | |
3581 | int error = may_delete(dir, dentry, 1); | |
3582 | ||
3583 | if (error) | |
3584 | return error; | |
3585 | ||
acfa4380 | 3586 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3587 | return -EPERM; |
3588 | ||
1d2ef590 | 3589 | dget(dentry); |
1b1dcc1b | 3590 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3591 | |
3592 | error = -EBUSY; | |
7af1364f | 3593 | if (is_local_mountpoint(dentry)) |
912dbc15 SW |
3594 | goto out; |
3595 | ||
3596 | error = security_inode_rmdir(dir, dentry); | |
3597 | if (error) | |
3598 | goto out; | |
3599 | ||
3cebde24 | 3600 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3601 | error = dir->i_op->rmdir(dir, dentry); |
3602 | if (error) | |
3603 | goto out; | |
3604 | ||
3605 | dentry->d_inode->i_flags |= S_DEAD; | |
3606 | dont_mount(dentry); | |
8ed936b5 | 3607 | detach_mounts(dentry); |
912dbc15 SW |
3608 | |
3609 | out: | |
1b1dcc1b | 3610 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3611 | dput(dentry); |
912dbc15 | 3612 | if (!error) |
1da177e4 | 3613 | d_delete(dentry); |
1da177e4 LT |
3614 | return error; |
3615 | } | |
4d359507 | 3616 | EXPORT_SYMBOL(vfs_rmdir); |
1da177e4 | 3617 | |
5590ff0d | 3618 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3619 | { |
3620 | int error = 0; | |
91a27b2a | 3621 | struct filename *name; |
1da177e4 LT |
3622 | struct dentry *dentry; |
3623 | struct nameidata nd; | |
c6ee9206 JL |
3624 | unsigned int lookup_flags = 0; |
3625 | retry: | |
3626 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3627 | if (IS_ERR(name)) |
3628 | return PTR_ERR(name); | |
1da177e4 LT |
3629 | |
3630 | switch(nd.last_type) { | |
0612d9fb OH |
3631 | case LAST_DOTDOT: |
3632 | error = -ENOTEMPTY; | |
3633 | goto exit1; | |
3634 | case LAST_DOT: | |
3635 | error = -EINVAL; | |
3636 | goto exit1; | |
3637 | case LAST_ROOT: | |
3638 | error = -EBUSY; | |
3639 | goto exit1; | |
1da177e4 | 3640 | } |
0612d9fb OH |
3641 | |
3642 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3643 | error = mnt_want_write(nd.path.mnt); |
3644 | if (error) | |
3645 | goto exit1; | |
0612d9fb | 3646 | |
4ac91378 | 3647 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3648 | dentry = lookup_hash(&nd); |
1da177e4 | 3649 | error = PTR_ERR(dentry); |
6902d925 DH |
3650 | if (IS_ERR(dentry)) |
3651 | goto exit2; | |
e6bc45d6 TT |
3652 | if (!dentry->d_inode) { |
3653 | error = -ENOENT; | |
3654 | goto exit3; | |
3655 | } | |
be6d3e56 KT |
3656 | error = security_path_rmdir(&nd.path, dentry); |
3657 | if (error) | |
c30dabfe | 3658 | goto exit3; |
4ac91378 | 3659 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3660 | exit3: |
6902d925 DH |
3661 | dput(dentry); |
3662 | exit2: | |
4ac91378 | 3663 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3664 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3665 | exit1: |
1d957f9b | 3666 | path_put(&nd.path); |
1da177e4 | 3667 | putname(name); |
c6ee9206 JL |
3668 | if (retry_estale(error, lookup_flags)) { |
3669 | lookup_flags |= LOOKUP_REVAL; | |
3670 | goto retry; | |
3671 | } | |
1da177e4 LT |
3672 | return error; |
3673 | } | |
3674 | ||
3cdad428 | 3675 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3676 | { |
3677 | return do_rmdir(AT_FDCWD, pathname); | |
3678 | } | |
3679 | ||
b21996e3 BF |
3680 | /** |
3681 | * vfs_unlink - unlink a filesystem object | |
3682 | * @dir: parent directory | |
3683 | * @dentry: victim | |
3684 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3685 | * | |
3686 | * The caller must hold dir->i_mutex. | |
3687 | * | |
3688 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3689 | * return a reference to the inode in delegated_inode. The caller | |
3690 | * should then break the delegation on that inode and retry. Because | |
3691 | * breaking a delegation may take a long time, the caller should drop | |
3692 | * dir->i_mutex before doing so. | |
3693 | * | |
3694 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3695 | * be appropriate for callers that expect the underlying filesystem not | |
3696 | * to be NFS exported. | |
3697 | */ | |
3698 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3699 | { |
9accbb97 | 3700 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3701 | int error = may_delete(dir, dentry, 0); |
3702 | ||
3703 | if (error) | |
3704 | return error; | |
3705 | ||
acfa4380 | 3706 | if (!dir->i_op->unlink) |
1da177e4 LT |
3707 | return -EPERM; |
3708 | ||
9accbb97 | 3709 | mutex_lock(&target->i_mutex); |
8ed936b5 | 3710 | if (is_local_mountpoint(dentry)) |
1da177e4 LT |
3711 | error = -EBUSY; |
3712 | else { | |
3713 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3714 | if (!error) { |
5a14696c BF |
3715 | error = try_break_deleg(target, delegated_inode); |
3716 | if (error) | |
b21996e3 | 3717 | goto out; |
1da177e4 | 3718 | error = dir->i_op->unlink(dir, dentry); |
8ed936b5 | 3719 | if (!error) { |
d83c49f3 | 3720 | dont_mount(dentry); |
8ed936b5 EB |
3721 | detach_mounts(dentry); |
3722 | } | |
bec1052e | 3723 | } |
1da177e4 | 3724 | } |
b21996e3 | 3725 | out: |
9accbb97 | 3726 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3727 | |
3728 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3729 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3730 | fsnotify_link_count(target); |
e234f35c | 3731 | d_delete(dentry); |
1da177e4 | 3732 | } |
0eeca283 | 3733 | |
1da177e4 LT |
3734 | return error; |
3735 | } | |
4d359507 | 3736 | EXPORT_SYMBOL(vfs_unlink); |
1da177e4 LT |
3737 | |
3738 | /* | |
3739 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3740 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3741 | * writeout happening, and we don't want to prevent access to the directory |
3742 | * while waiting on the I/O. | |
3743 | */ | |
5590ff0d | 3744 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3745 | { |
2ad94ae6 | 3746 | int error; |
91a27b2a | 3747 | struct filename *name; |
1da177e4 LT |
3748 | struct dentry *dentry; |
3749 | struct nameidata nd; | |
3750 | struct inode *inode = NULL; | |
b21996e3 | 3751 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3752 | unsigned int lookup_flags = 0; |
3753 | retry: | |
3754 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3755 | if (IS_ERR(name)) |
3756 | return PTR_ERR(name); | |
2ad94ae6 | 3757 | |
1da177e4 LT |
3758 | error = -EISDIR; |
3759 | if (nd.last_type != LAST_NORM) | |
3760 | goto exit1; | |
0612d9fb OH |
3761 | |
3762 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3763 | error = mnt_want_write(nd.path.mnt); |
3764 | if (error) | |
3765 | goto exit1; | |
b21996e3 | 3766 | retry_deleg: |
4ac91378 | 3767 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3768 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3769 | error = PTR_ERR(dentry); |
3770 | if (!IS_ERR(dentry)) { | |
3771 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3772 | if (nd.last.name[nd.last.len]) |
3773 | goto slashes; | |
1da177e4 | 3774 | inode = dentry->d_inode; |
b18825a7 | 3775 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3776 | goto slashes; |
3777 | ihold(inode); | |
be6d3e56 KT |
3778 | error = security_path_unlink(&nd.path, dentry); |
3779 | if (error) | |
c30dabfe | 3780 | goto exit2; |
b21996e3 | 3781 | error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3782 | exit2: |
1da177e4 LT |
3783 | dput(dentry); |
3784 | } | |
4ac91378 | 3785 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3786 | if (inode) |
3787 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3788 | inode = NULL; |
3789 | if (delegated_inode) { | |
5a14696c | 3790 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3791 | if (!error) |
3792 | goto retry_deleg; | |
3793 | } | |
c30dabfe | 3794 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3795 | exit1: |
1d957f9b | 3796 | path_put(&nd.path); |
1da177e4 | 3797 | putname(name); |
5d18f813 JL |
3798 | if (retry_estale(error, lookup_flags)) { |
3799 | lookup_flags |= LOOKUP_REVAL; | |
3800 | inode = NULL; | |
3801 | goto retry; | |
3802 | } | |
1da177e4 LT |
3803 | return error; |
3804 | ||
3805 | slashes: | |
b18825a7 DH |
3806 | if (d_is_negative(dentry)) |
3807 | error = -ENOENT; | |
44b1d530 | 3808 | else if (d_is_dir(dentry)) |
b18825a7 DH |
3809 | error = -EISDIR; |
3810 | else | |
3811 | error = -ENOTDIR; | |
1da177e4 LT |
3812 | goto exit2; |
3813 | } | |
3814 | ||
2e4d0924 | 3815 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3816 | { |
3817 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3818 | return -EINVAL; | |
3819 | ||
3820 | if (flag & AT_REMOVEDIR) | |
3821 | return do_rmdir(dfd, pathname); | |
3822 | ||
3823 | return do_unlinkat(dfd, pathname); | |
3824 | } | |
3825 | ||
3480b257 | 3826 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3827 | { |
3828 | return do_unlinkat(AT_FDCWD, pathname); | |
3829 | } | |
3830 | ||
db2e747b | 3831 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3832 | { |
a95164d9 | 3833 | int error = may_create(dir, dentry); |
1da177e4 LT |
3834 | |
3835 | if (error) | |
3836 | return error; | |
3837 | ||
acfa4380 | 3838 | if (!dir->i_op->symlink) |
1da177e4 LT |
3839 | return -EPERM; |
3840 | ||
3841 | error = security_inode_symlink(dir, dentry, oldname); | |
3842 | if (error) | |
3843 | return error; | |
3844 | ||
1da177e4 | 3845 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3846 | if (!error) |
f38aa942 | 3847 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3848 | return error; |
3849 | } | |
4d359507 | 3850 | EXPORT_SYMBOL(vfs_symlink); |
1da177e4 | 3851 | |
2e4d0924 HC |
3852 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3853 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3854 | { |
2ad94ae6 | 3855 | int error; |
91a27b2a | 3856 | struct filename *from; |
6902d925 | 3857 | struct dentry *dentry; |
dae6ad8f | 3858 | struct path path; |
f46d3567 | 3859 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3860 | |
3861 | from = getname(oldname); | |
2ad94ae6 | 3862 | if (IS_ERR(from)) |
1da177e4 | 3863 | return PTR_ERR(from); |
f46d3567 JL |
3864 | retry: |
3865 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3866 | error = PTR_ERR(dentry); |
3867 | if (IS_ERR(dentry)) | |
dae6ad8f | 3868 | goto out_putname; |
6902d925 | 3869 | |
91a27b2a | 3870 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3871 | if (!error) |
91a27b2a | 3872 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3873 | done_path_create(&path, dentry); |
f46d3567 JL |
3874 | if (retry_estale(error, lookup_flags)) { |
3875 | lookup_flags |= LOOKUP_REVAL; | |
3876 | goto retry; | |
3877 | } | |
6902d925 | 3878 | out_putname: |
1da177e4 LT |
3879 | putname(from); |
3880 | return error; | |
3881 | } | |
3882 | ||
3480b257 | 3883 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3884 | { |
3885 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3886 | } | |
3887 | ||
146a8595 BF |
3888 | /** |
3889 | * vfs_link - create a new link | |
3890 | * @old_dentry: object to be linked | |
3891 | * @dir: new parent | |
3892 | * @new_dentry: where to create the new link | |
3893 | * @delegated_inode: returns inode needing a delegation break | |
3894 | * | |
3895 | * The caller must hold dir->i_mutex | |
3896 | * | |
3897 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3898 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3899 | * inode in delegated_inode. The caller should then break the delegation | |
3900 | * and retry. Because breaking a delegation may take a long time, the | |
3901 | * caller should drop the i_mutex before doing so. | |
3902 | * | |
3903 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3904 | * be appropriate for callers that expect the underlying filesystem not | |
3905 | * to be NFS exported. | |
3906 | */ | |
3907 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
3908 | { |
3909 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3910 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3911 | int error; |
3912 | ||
3913 | if (!inode) | |
3914 | return -ENOENT; | |
3915 | ||
a95164d9 | 3916 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3917 | if (error) |
3918 | return error; | |
3919 | ||
3920 | if (dir->i_sb != inode->i_sb) | |
3921 | return -EXDEV; | |
3922 | ||
3923 | /* | |
3924 | * A link to an append-only or immutable file cannot be created. | |
3925 | */ | |
3926 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3927 | return -EPERM; | |
acfa4380 | 3928 | if (!dir->i_op->link) |
1da177e4 | 3929 | return -EPERM; |
7e79eedb | 3930 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3931 | return -EPERM; |
3932 | ||
3933 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3934 | if (error) | |
3935 | return error; | |
3936 | ||
7e79eedb | 3937 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3938 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3939 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3940 | error = -ENOENT; |
8de52778 AV |
3941 | else if (max_links && inode->i_nlink >= max_links) |
3942 | error = -EMLINK; | |
146a8595 BF |
3943 | else { |
3944 | error = try_break_deleg(inode, delegated_inode); | |
3945 | if (!error) | |
3946 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
3947 | } | |
f4e0c30c AV |
3948 | |
3949 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3950 | spin_lock(&inode->i_lock); | |
3951 | inode->i_state &= ~I_LINKABLE; | |
3952 | spin_unlock(&inode->i_lock); | |
3953 | } | |
7e79eedb | 3954 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3955 | if (!error) |
7e79eedb | 3956 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3957 | return error; |
3958 | } | |
4d359507 | 3959 | EXPORT_SYMBOL(vfs_link); |
1da177e4 LT |
3960 | |
3961 | /* | |
3962 | * Hardlinks are often used in delicate situations. We avoid | |
3963 | * security-related surprises by not following symlinks on the | |
3964 | * newname. --KAB | |
3965 | * | |
3966 | * We don't follow them on the oldname either to be compatible | |
3967 | * with linux 2.0, and to avoid hard-linking to directories | |
3968 | * and other special files. --ADM | |
3969 | */ | |
2e4d0924 HC |
3970 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3971 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3972 | { |
3973 | struct dentry *new_dentry; | |
dae6ad8f | 3974 | struct path old_path, new_path; |
146a8595 | 3975 | struct inode *delegated_inode = NULL; |
11a7b371 | 3976 | int how = 0; |
1da177e4 | 3977 | int error; |
1da177e4 | 3978 | |
11a7b371 | 3979 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3980 | return -EINVAL; |
11a7b371 | 3981 | /* |
f0cc6ffb LT |
3982 | * To use null names we require CAP_DAC_READ_SEARCH |
3983 | * This ensures that not everyone will be able to create | |
3984 | * handlink using the passed filedescriptor. | |
11a7b371 | 3985 | */ |
f0cc6ffb LT |
3986 | if (flags & AT_EMPTY_PATH) { |
3987 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3988 | return -ENOENT; | |
11a7b371 | 3989 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3990 | } |
11a7b371 AK |
3991 | |
3992 | if (flags & AT_SYMLINK_FOLLOW) | |
3993 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3994 | retry: |
11a7b371 | 3995 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3996 | if (error) |
2ad94ae6 AV |
3997 | return error; |
3998 | ||
442e31ca JL |
3999 | new_dentry = user_path_create(newdfd, newname, &new_path, |
4000 | (how & LOOKUP_REVAL)); | |
1da177e4 | 4001 | error = PTR_ERR(new_dentry); |
6902d925 | 4002 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
4003 | goto out; |
4004 | ||
4005 | error = -EXDEV; | |
4006 | if (old_path.mnt != new_path.mnt) | |
4007 | goto out_dput; | |
800179c9 KC |
4008 | error = may_linkat(&old_path); |
4009 | if (unlikely(error)) | |
4010 | goto out_dput; | |
dae6ad8f | 4011 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 4012 | if (error) |
a8104a9f | 4013 | goto out_dput; |
146a8595 | 4014 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 4015 | out_dput: |
921a1650 | 4016 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
4017 | if (delegated_inode) { |
4018 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
4019 | if (!error) { |
4020 | path_put(&old_path); | |
146a8595 | 4021 | goto retry; |
d22e6338 | 4022 | } |
146a8595 | 4023 | } |
442e31ca | 4024 | if (retry_estale(error, how)) { |
d22e6338 | 4025 | path_put(&old_path); |
442e31ca JL |
4026 | how |= LOOKUP_REVAL; |
4027 | goto retry; | |
4028 | } | |
1da177e4 | 4029 | out: |
2d8f3038 | 4030 | path_put(&old_path); |
1da177e4 LT |
4031 | |
4032 | return error; | |
4033 | } | |
4034 | ||
3480b257 | 4035 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4036 | { |
c04030e1 | 4037 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4038 | } |
4039 | ||
bc27027a MS |
4040 | /** |
4041 | * vfs_rename - rename a filesystem object | |
4042 | * @old_dir: parent of source | |
4043 | * @old_dentry: source | |
4044 | * @new_dir: parent of destination | |
4045 | * @new_dentry: destination | |
4046 | * @delegated_inode: returns an inode needing a delegation break | |
520c8b16 | 4047 | * @flags: rename flags |
bc27027a MS |
4048 | * |
4049 | * The caller must hold multiple mutexes--see lock_rename()). | |
4050 | * | |
4051 | * If vfs_rename discovers a delegation in need of breaking at either | |
4052 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4053 | * reference to the inode in delegated_inode. The caller should then | |
4054 | * break the delegation and retry. Because breaking a delegation may | |
4055 | * take a long time, the caller should drop all locks before doing | |
4056 | * so. | |
4057 | * | |
4058 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4059 | * be appropriate for callers that expect the underlying filesystem not | |
4060 | * to be NFS exported. | |
4061 | * | |
1da177e4 LT |
4062 | * The worst of all namespace operations - renaming directory. "Perverted" |
4063 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4064 | * Problems: | |
d03b29a2 | 4065 | * a) we can get into loop creation. |
1da177e4 LT |
4066 | * b) race potential - two innocent renames can create a loop together. |
4067 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 4068 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 4069 | * story. |
6cedba89 BF |
4070 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
4071 | * and source (if it is not a directory). | |
1b1dcc1b | 4072 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
4073 | * whether the target exists). Solution: try to be smart with locking |
4074 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 4075 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
4076 | * move will be locked. Thus we can rank directories by the tree |
4077 | * (ancestors first) and rank all non-directories after them. | |
4078 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 4079 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
4080 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
4081 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4082 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 4083 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 4084 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 4085 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 4086 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
4087 | * locking]. |
4088 | */ | |
bc27027a MS |
4089 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4090 | struct inode *new_dir, struct dentry *new_dentry, | |
520c8b16 | 4091 | struct inode **delegated_inode, unsigned int flags) |
1da177e4 | 4092 | { |
bc27027a MS |
4093 | int error; |
4094 | bool is_dir = d_is_dir(old_dentry); | |
4095 | const unsigned char *old_name; | |
4096 | struct inode *source = old_dentry->d_inode; | |
9055cba7 | 4097 | struct inode *target = new_dentry->d_inode; |
da1ce067 MS |
4098 | bool new_is_dir = false; |
4099 | unsigned max_links = new_dir->i_sb->s_max_links; | |
bc27027a MS |
4100 | |
4101 | if (source == target) | |
4102 | return 0; | |
4103 | ||
4104 | error = may_delete(old_dir, old_dentry, is_dir); | |
4105 | if (error) | |
4106 | return error; | |
4107 | ||
da1ce067 | 4108 | if (!target) { |
bc27027a | 4109 | error = may_create(new_dir, new_dentry); |
da1ce067 MS |
4110 | } else { |
4111 | new_is_dir = d_is_dir(new_dentry); | |
4112 | ||
4113 | if (!(flags & RENAME_EXCHANGE)) | |
4114 | error = may_delete(new_dir, new_dentry, is_dir); | |
4115 | else | |
4116 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4117 | } | |
bc27027a MS |
4118 | if (error) |
4119 | return error; | |
4120 | ||
7177a9c4 | 4121 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) |
bc27027a | 4122 | return -EPERM; |
1da177e4 | 4123 | |
520c8b16 MS |
4124 | if (flags && !old_dir->i_op->rename2) |
4125 | return -EINVAL; | |
4126 | ||
1da177e4 LT |
4127 | /* |
4128 | * If we are going to change the parent - check write permissions, | |
4129 | * we'll need to flip '..'. | |
4130 | */ | |
da1ce067 MS |
4131 | if (new_dir != old_dir) { |
4132 | if (is_dir) { | |
4133 | error = inode_permission(source, MAY_WRITE); | |
4134 | if (error) | |
4135 | return error; | |
4136 | } | |
4137 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4138 | error = inode_permission(target, MAY_WRITE); | |
4139 | if (error) | |
4140 | return error; | |
4141 | } | |
1da177e4 LT |
4142 | } |
4143 | ||
0b3974eb MS |
4144 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, |
4145 | flags); | |
1da177e4 LT |
4146 | if (error) |
4147 | return error; | |
4148 | ||
bc27027a | 4149 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
1d2ef590 | 4150 | dget(new_dentry); |
da1ce067 | 4151 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4152 | lock_two_nondirectories(source, target); |
4153 | else if (target) | |
1b1dcc1b | 4154 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4155 | |
4156 | error = -EBUSY; | |
7af1364f | 4157 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) |
9055cba7 SW |
4158 | goto out; |
4159 | ||
da1ce067 | 4160 | if (max_links && new_dir != old_dir) { |
bc27027a | 4161 | error = -EMLINK; |
da1ce067 | 4162 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) |
bc27027a | 4163 | goto out; |
da1ce067 MS |
4164 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && |
4165 | old_dir->i_nlink >= max_links) | |
4166 | goto out; | |
4167 | } | |
4168 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4169 | shrink_dcache_parent(new_dentry); | |
4170 | if (!is_dir) { | |
bc27027a | 4171 | error = try_break_deleg(source, delegated_inode); |
8e6d782c BF |
4172 | if (error) |
4173 | goto out; | |
da1ce067 MS |
4174 | } |
4175 | if (target && !new_is_dir) { | |
4176 | error = try_break_deleg(target, delegated_inode); | |
4177 | if (error) | |
4178 | goto out; | |
8e6d782c | 4179 | } |
7177a9c4 | 4180 | if (!old_dir->i_op->rename2) { |
520c8b16 MS |
4181 | error = old_dir->i_op->rename(old_dir, old_dentry, |
4182 | new_dir, new_dentry); | |
4183 | } else { | |
7177a9c4 | 4184 | WARN_ON(old_dir->i_op->rename != NULL); |
520c8b16 MS |
4185 | error = old_dir->i_op->rename2(old_dir, old_dentry, |
4186 | new_dir, new_dentry, flags); | |
4187 | } | |
51892bbb SW |
4188 | if (error) |
4189 | goto out; | |
4190 | ||
da1ce067 | 4191 | if (!(flags & RENAME_EXCHANGE) && target) { |
bc27027a MS |
4192 | if (is_dir) |
4193 | target->i_flags |= S_DEAD; | |
51892bbb | 4194 | dont_mount(new_dentry); |
8ed936b5 | 4195 | detach_mounts(new_dentry); |
bc27027a | 4196 | } |
da1ce067 MS |
4197 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { |
4198 | if (!(flags & RENAME_EXCHANGE)) | |
4199 | d_move(old_dentry, new_dentry); | |
4200 | else | |
4201 | d_exchange(old_dentry, new_dentry); | |
4202 | } | |
51892bbb | 4203 | out: |
da1ce067 | 4204 | if (!is_dir || (flags & RENAME_EXCHANGE)) |
bc27027a MS |
4205 | unlock_two_nondirectories(source, target); |
4206 | else if (target) | |
4207 | mutex_unlock(&target->i_mutex); | |
1da177e4 | 4208 | dput(new_dentry); |
da1ce067 | 4209 | if (!error) { |
123df294 | 4210 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
da1ce067 MS |
4211 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); |
4212 | if (flags & RENAME_EXCHANGE) { | |
4213 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4214 | new_is_dir, NULL, new_dentry); | |
4215 | } | |
4216 | } | |
0eeca283 RL |
4217 | fsnotify_oldname_free(old_name); |
4218 | ||
1da177e4 LT |
4219 | return error; |
4220 | } | |
4d359507 | 4221 | EXPORT_SYMBOL(vfs_rename); |
1da177e4 | 4222 | |
520c8b16 MS |
4223 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, |
4224 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
1da177e4 | 4225 | { |
2ad94ae6 AV |
4226 | struct dentry *old_dir, *new_dir; |
4227 | struct dentry *old_dentry, *new_dentry; | |
4228 | struct dentry *trap; | |
1da177e4 | 4229 | struct nameidata oldnd, newnd; |
8e6d782c | 4230 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4231 | struct filename *from; |
4232 | struct filename *to; | |
c6a94284 JL |
4233 | unsigned int lookup_flags = 0; |
4234 | bool should_retry = false; | |
2ad94ae6 | 4235 | int error; |
520c8b16 | 4236 | |
0d7a8555 | 4237 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
da1ce067 MS |
4238 | return -EINVAL; |
4239 | ||
0d7a8555 MS |
4240 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && |
4241 | (flags & RENAME_EXCHANGE)) | |
520c8b16 MS |
4242 | return -EINVAL; |
4243 | ||
0d7a8555 MS |
4244 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) |
4245 | return -EPERM; | |
4246 | ||
c6a94284 JL |
4247 | retry: |
4248 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4249 | if (IS_ERR(from)) { |
4250 | error = PTR_ERR(from); | |
1da177e4 | 4251 | goto exit; |
91a27b2a | 4252 | } |
1da177e4 | 4253 | |
c6a94284 | 4254 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4255 | if (IS_ERR(to)) { |
4256 | error = PTR_ERR(to); | |
1da177e4 | 4257 | goto exit1; |
91a27b2a | 4258 | } |
1da177e4 LT |
4259 | |
4260 | error = -EXDEV; | |
4ac91378 | 4261 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4262 | goto exit2; |
4263 | ||
4ac91378 | 4264 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4265 | error = -EBUSY; |
4266 | if (oldnd.last_type != LAST_NORM) | |
4267 | goto exit2; | |
4268 | ||
4ac91378 | 4269 | new_dir = newnd.path.dentry; |
0a7c3937 MS |
4270 | if (flags & RENAME_NOREPLACE) |
4271 | error = -EEXIST; | |
1da177e4 LT |
4272 | if (newnd.last_type != LAST_NORM) |
4273 | goto exit2; | |
4274 | ||
c30dabfe JK |
4275 | error = mnt_want_write(oldnd.path.mnt); |
4276 | if (error) | |
4277 | goto exit2; | |
4278 | ||
0612d9fb OH |
4279 | oldnd.flags &= ~LOOKUP_PARENT; |
4280 | newnd.flags &= ~LOOKUP_PARENT; | |
da1ce067 MS |
4281 | if (!(flags & RENAME_EXCHANGE)) |
4282 | newnd.flags |= LOOKUP_RENAME_TARGET; | |
0612d9fb | 4283 | |
8e6d782c | 4284 | retry_deleg: |
1da177e4 LT |
4285 | trap = lock_rename(new_dir, old_dir); |
4286 | ||
49705b77 | 4287 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4288 | error = PTR_ERR(old_dentry); |
4289 | if (IS_ERR(old_dentry)) | |
4290 | goto exit3; | |
4291 | /* source must exist */ | |
4292 | error = -ENOENT; | |
b18825a7 | 4293 | if (d_is_negative(old_dentry)) |
1da177e4 | 4294 | goto exit4; |
0a7c3937 MS |
4295 | new_dentry = lookup_hash(&newnd); |
4296 | error = PTR_ERR(new_dentry); | |
4297 | if (IS_ERR(new_dentry)) | |
4298 | goto exit4; | |
4299 | error = -EEXIST; | |
4300 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4301 | goto exit5; | |
da1ce067 MS |
4302 | if (flags & RENAME_EXCHANGE) { |
4303 | error = -ENOENT; | |
4304 | if (d_is_negative(new_dentry)) | |
4305 | goto exit5; | |
4306 | ||
4307 | if (!d_is_dir(new_dentry)) { | |
4308 | error = -ENOTDIR; | |
4309 | if (newnd.last.name[newnd.last.len]) | |
4310 | goto exit5; | |
4311 | } | |
4312 | } | |
1da177e4 | 4313 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
44b1d530 | 4314 | if (!d_is_dir(old_dentry)) { |
1da177e4 LT |
4315 | error = -ENOTDIR; |
4316 | if (oldnd.last.name[oldnd.last.len]) | |
0a7c3937 | 4317 | goto exit5; |
da1ce067 | 4318 | if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) |
0a7c3937 | 4319 | goto exit5; |
1da177e4 LT |
4320 | } |
4321 | /* source should not be ancestor of target */ | |
4322 | error = -EINVAL; | |
4323 | if (old_dentry == trap) | |
0a7c3937 | 4324 | goto exit5; |
1da177e4 | 4325 | /* target should not be an ancestor of source */ |
da1ce067 MS |
4326 | if (!(flags & RENAME_EXCHANGE)) |
4327 | error = -ENOTEMPTY; | |
1da177e4 LT |
4328 | if (new_dentry == trap) |
4329 | goto exit5; | |
4330 | ||
be6d3e56 | 4331 | error = security_path_rename(&oldnd.path, old_dentry, |
0b3974eb | 4332 | &newnd.path, new_dentry, flags); |
be6d3e56 | 4333 | if (error) |
c30dabfe | 4334 | goto exit5; |
1da177e4 | 4335 | error = vfs_rename(old_dir->d_inode, old_dentry, |
520c8b16 MS |
4336 | new_dir->d_inode, new_dentry, |
4337 | &delegated_inode, flags); | |
1da177e4 LT |
4338 | exit5: |
4339 | dput(new_dentry); | |
4340 | exit4: | |
4341 | dput(old_dentry); | |
4342 | exit3: | |
4343 | unlock_rename(new_dir, old_dir); | |
8e6d782c BF |
4344 | if (delegated_inode) { |
4345 | error = break_deleg_wait(&delegated_inode); | |
4346 | if (!error) | |
4347 | goto retry_deleg; | |
4348 | } | |
c30dabfe | 4349 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4350 | exit2: |
c6a94284 JL |
4351 | if (retry_estale(error, lookup_flags)) |
4352 | should_retry = true; | |
1d957f9b | 4353 | path_put(&newnd.path); |
2ad94ae6 | 4354 | putname(to); |
1da177e4 | 4355 | exit1: |
1d957f9b | 4356 | path_put(&oldnd.path); |
1da177e4 | 4357 | putname(from); |
c6a94284 JL |
4358 | if (should_retry) { |
4359 | should_retry = false; | |
4360 | lookup_flags |= LOOKUP_REVAL; | |
4361 | goto retry; | |
4362 | } | |
2ad94ae6 | 4363 | exit: |
1da177e4 LT |
4364 | return error; |
4365 | } | |
4366 | ||
520c8b16 MS |
4367 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4368 | int, newdfd, const char __user *, newname) | |
4369 | { | |
4370 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4371 | } | |
4372 | ||
a26eab24 | 4373 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 4374 | { |
520c8b16 | 4375 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
4376 | } |
4377 | ||
787fb6bc MS |
4378 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) |
4379 | { | |
4380 | int error = may_create(dir, dentry); | |
4381 | if (error) | |
4382 | return error; | |
4383 | ||
4384 | if (!dir->i_op->mknod) | |
4385 | return -EPERM; | |
4386 | ||
4387 | return dir->i_op->mknod(dir, dentry, | |
4388 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4389 | } | |
4390 | EXPORT_SYMBOL(vfs_whiteout); | |
4391 | ||
5d826c84 | 4392 | int readlink_copy(char __user *buffer, int buflen, const char *link) |
1da177e4 | 4393 | { |
5d826c84 | 4394 | int len = PTR_ERR(link); |
1da177e4 LT |
4395 | if (IS_ERR(link)) |
4396 | goto out; | |
4397 | ||
4398 | len = strlen(link); | |
4399 | if (len > (unsigned) buflen) | |
4400 | len = buflen; | |
4401 | if (copy_to_user(buffer, link, len)) | |
4402 | len = -EFAULT; | |
4403 | out: | |
4404 | return len; | |
4405 | } | |
5d826c84 | 4406 | EXPORT_SYMBOL(readlink_copy); |
1da177e4 LT |
4407 | |
4408 | /* | |
4409 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4410 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4411 | * using) it for any given inode is up to filesystem. | |
4412 | */ | |
4413 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4414 | { | |
4415 | struct nameidata nd; | |
cc314eef | 4416 | void *cookie; |
694a1764 | 4417 | int res; |
cc314eef | 4418 | |
1da177e4 | 4419 | nd.depth = 0; |
cc314eef | 4420 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4421 | if (IS_ERR(cookie)) |
4422 | return PTR_ERR(cookie); | |
4423 | ||
5d826c84 | 4424 | res = readlink_copy(buffer, buflen, nd_get_link(&nd)); |
694a1764 MS |
4425 | if (dentry->d_inode->i_op->put_link) |
4426 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4427 | return res; | |
1da177e4 | 4428 | } |
4d359507 | 4429 | EXPORT_SYMBOL(generic_readlink); |
1da177e4 | 4430 | |
1da177e4 LT |
4431 | /* get the link contents into pagecache */ |
4432 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4433 | { | |
ebd09abb DG |
4434 | char *kaddr; |
4435 | struct page *page; | |
1da177e4 | 4436 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4437 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4438 | if (IS_ERR(page)) |
6fe6900e | 4439 | return (char*)page; |
1da177e4 | 4440 | *ppage = page; |
ebd09abb DG |
4441 | kaddr = kmap(page); |
4442 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4443 | return kaddr; | |
1da177e4 LT |
4444 | } |
4445 | ||
4446 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4447 | { | |
4448 | struct page *page = NULL; | |
5d826c84 | 4449 | int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); |
1da177e4 LT |
4450 | if (page) { |
4451 | kunmap(page); | |
4452 | page_cache_release(page); | |
4453 | } | |
4454 | return res; | |
4455 | } | |
4d359507 | 4456 | EXPORT_SYMBOL(page_readlink); |
1da177e4 | 4457 | |
cc314eef | 4458 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4459 | { |
cc314eef | 4460 | struct page *page = NULL; |
1da177e4 | 4461 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4462 | return page; |
1da177e4 | 4463 | } |
4d359507 | 4464 | EXPORT_SYMBOL(page_follow_link_light); |
1da177e4 | 4465 | |
cc314eef | 4466 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4467 | { |
cc314eef LT |
4468 | struct page *page = cookie; |
4469 | ||
4470 | if (page) { | |
1da177e4 LT |
4471 | kunmap(page); |
4472 | page_cache_release(page); | |
1da177e4 LT |
4473 | } |
4474 | } | |
4d359507 | 4475 | EXPORT_SYMBOL(page_put_link); |
1da177e4 | 4476 | |
54566b2c NP |
4477 | /* |
4478 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4479 | */ | |
4480 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4481 | { |
4482 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4483 | struct page *page; |
afddba49 | 4484 | void *fsdata; |
beb497ab | 4485 | int err; |
1da177e4 | 4486 | char *kaddr; |
54566b2c NP |
4487 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4488 | if (nofs) | |
4489 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4490 | |
7e53cac4 | 4491 | retry: |
afddba49 | 4492 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4493 | flags, &page, &fsdata); |
1da177e4 | 4494 | if (err) |
afddba49 NP |
4495 | goto fail; |
4496 | ||
e8e3c3d6 | 4497 | kaddr = kmap_atomic(page); |
1da177e4 | 4498 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4499 | kunmap_atomic(kaddr); |
afddba49 NP |
4500 | |
4501 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4502 | page, fsdata); | |
1da177e4 LT |
4503 | if (err < 0) |
4504 | goto fail; | |
afddba49 NP |
4505 | if (err < len-1) |
4506 | goto retry; | |
4507 | ||
1da177e4 LT |
4508 | mark_inode_dirty(inode); |
4509 | return 0; | |
1da177e4 LT |
4510 | fail: |
4511 | return err; | |
4512 | } | |
4d359507 | 4513 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 | 4514 | |
0adb25d2 KK |
4515 | int page_symlink(struct inode *inode, const char *symname, int len) |
4516 | { | |
4517 | return __page_symlink(inode, symname, len, | |
54566b2c | 4518 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 | 4519 | } |
4d359507 | 4520 | EXPORT_SYMBOL(page_symlink); |
0adb25d2 | 4521 | |
92e1d5be | 4522 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4523 | .readlink = generic_readlink, |
4524 | .follow_link = page_follow_link_light, | |
4525 | .put_link = page_put_link, | |
4526 | }; | |
1da177e4 | 4527 | EXPORT_SYMBOL(page_symlink_inode_operations); |