Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
630d9c47 | 18 | #include <linux/export.h> |
44696908 | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
1da177e4 | 23 | #include <linux/pagemap.h> |
0eeca283 | 24 | #include <linux/fsnotify.h> |
1da177e4 LT |
25 | #include <linux/personality.h> |
26 | #include <linux/security.h> | |
6146f0d5 | 27 | #include <linux/ima.h> |
1da177e4 LT |
28 | #include <linux/syscalls.h> |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
16f7e0fe | 31 | #include <linux/capability.h> |
834f2a4a | 32 | #include <linux/file.h> |
5590ff0d | 33 | #include <linux/fcntl.h> |
08ce5f16 | 34 | #include <linux/device_cgroup.h> |
5ad4e53b | 35 | #include <linux/fs_struct.h> |
e77819e5 | 36 | #include <linux/posix_acl.h> |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | ||
e81e3f4d | 39 | #include "internal.h" |
c7105365 | 40 | #include "mount.h" |
e81e3f4d | 41 | |
1da177e4 LT |
42 | /* [Feb-1997 T. Schoebel-Theuer] |
43 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
44 | * were necessary because of omirr. The reason is that omirr needs | |
45 | * to know the _real_ pathname, not the user-supplied one, in case | |
46 | * of symlinks (and also when transname replacements occur). | |
47 | * | |
48 | * The new code replaces the old recursive symlink resolution with | |
49 | * an iterative one (in case of non-nested symlink chains). It does | |
50 | * this with calls to <fs>_follow_link(). | |
51 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
52 | * replaced with a single function lookup_dentry() that can handle all | |
53 | * the special cases of the former code. | |
54 | * | |
55 | * With the new dcache, the pathname is stored at each inode, at least as | |
56 | * long as the refcount of the inode is positive. As a side effect, the | |
57 | * size of the dcache depends on the inode cache and thus is dynamic. | |
58 | * | |
59 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
60 | * resolution to correspond with current state of the code. | |
61 | * | |
62 | * Note that the symlink resolution is not *completely* iterative. | |
63 | * There is still a significant amount of tail- and mid- recursion in | |
64 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
65 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
66 | * may return different results than <fs>_follow_link(). Many virtual | |
67 | * filesystems (including /proc) exhibit this behavior. | |
68 | */ | |
69 | ||
70 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
71 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
72 | * and the name already exists in form of a symlink, try to create the new | |
73 | * name indicated by the symlink. The old code always complained that the | |
74 | * name already exists, due to not following the symlink even if its target | |
75 | * is nonexistent. The new semantics affects also mknod() and link() when | |
25985edc | 76 | * the name is a symlink pointing to a non-existent name. |
1da177e4 LT |
77 | * |
78 | * I don't know which semantics is the right one, since I have no access | |
79 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
80 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
81 | * "old" one. Personally, I think the new semantics is much more logical. | |
82 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
83 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
84 | * and in the old Linux semantics. | |
85 | */ | |
86 | ||
87 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
88 | * semantics. See the comments in "open_namei" and "do_link" below. | |
89 | * | |
90 | * [10-Sep-98 Alan Modra] Another symlink change. | |
91 | */ | |
92 | ||
93 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
94 | * inside the path - always follow. | |
95 | * in the last component in creation/removal/renaming - never follow. | |
96 | * if LOOKUP_FOLLOW passed - follow. | |
97 | * if the pathname has trailing slashes - follow. | |
98 | * otherwise - don't follow. | |
99 | * (applied in that order). | |
100 | * | |
101 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
102 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
103 | * During the 2.4 we need to fix the userland stuff depending on it - | |
104 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
105 | * XEmacs seems to be relying on it... | |
106 | */ | |
107 | /* | |
108 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
a11f3a05 | 109 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
1da177e4 LT |
110 | * any extra contention... |
111 | */ | |
112 | ||
113 | /* In order to reduce some races, while at the same time doing additional | |
114 | * checking and hopefully speeding things up, we copy filenames to the | |
115 | * kernel data space before using them.. | |
116 | * | |
117 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
118 | * PATH_MAX includes the nul terminator --RR. | |
119 | */ | |
91a27b2a | 120 | void final_putname(struct filename *name) |
1da177e4 | 121 | { |
7950e385 JL |
122 | if (name->separate) { |
123 | __putname(name->name); | |
124 | kfree(name); | |
125 | } else { | |
126 | __putname(name); | |
127 | } | |
91a27b2a JL |
128 | } |
129 | ||
7950e385 JL |
130 | #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) |
131 | ||
91a27b2a JL |
132 | static struct filename * |
133 | getname_flags(const char __user *filename, int flags, int *empty) | |
134 | { | |
135 | struct filename *result, *err; | |
3f9f0aa6 | 136 | int len; |
7950e385 JL |
137 | long max; |
138 | char *kname; | |
4043cde8 | 139 | |
7ac86265 JL |
140 | result = audit_reusename(filename); |
141 | if (result) | |
142 | return result; | |
143 | ||
7950e385 | 144 | result = __getname(); |
3f9f0aa6 | 145 | if (unlikely(!result)) |
4043cde8 EP |
146 | return ERR_PTR(-ENOMEM); |
147 | ||
7950e385 JL |
148 | /* |
149 | * First, try to embed the struct filename inside the names_cache | |
150 | * allocation | |
151 | */ | |
152 | kname = (char *)result + sizeof(*result); | |
91a27b2a | 153 | result->name = kname; |
7950e385 JL |
154 | result->separate = false; |
155 | max = EMBEDDED_NAME_MAX; | |
156 | ||
157 | recopy: | |
158 | len = strncpy_from_user(kname, filename, max); | |
91a27b2a JL |
159 | if (unlikely(len < 0)) { |
160 | err = ERR_PTR(len); | |
3f9f0aa6 | 161 | goto error; |
91a27b2a | 162 | } |
3f9f0aa6 | 163 | |
7950e385 JL |
164 | /* |
165 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
166 | * separate struct filename so we can dedicate the entire | |
167 | * names_cache allocation for the pathname, and re-do the copy from | |
168 | * userland. | |
169 | */ | |
170 | if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { | |
171 | kname = (char *)result; | |
172 | ||
173 | result = kzalloc(sizeof(*result), GFP_KERNEL); | |
174 | if (!result) { | |
175 | err = ERR_PTR(-ENOMEM); | |
176 | result = (struct filename *)kname; | |
177 | goto error; | |
178 | } | |
179 | result->name = kname; | |
180 | result->separate = true; | |
181 | max = PATH_MAX; | |
182 | goto recopy; | |
183 | } | |
184 | ||
3f9f0aa6 LT |
185 | /* The empty path is special. */ |
186 | if (unlikely(!len)) { | |
187 | if (empty) | |
4043cde8 | 188 | *empty = 1; |
3f9f0aa6 LT |
189 | err = ERR_PTR(-ENOENT); |
190 | if (!(flags & LOOKUP_EMPTY)) | |
191 | goto error; | |
1da177e4 | 192 | } |
3f9f0aa6 LT |
193 | |
194 | err = ERR_PTR(-ENAMETOOLONG); | |
7950e385 JL |
195 | if (unlikely(len >= PATH_MAX)) |
196 | goto error; | |
197 | ||
198 | result->uptr = filename; | |
199 | audit_getname(result); | |
200 | return result; | |
3f9f0aa6 LT |
201 | |
202 | error: | |
7950e385 | 203 | final_putname(result); |
3f9f0aa6 | 204 | return err; |
1da177e4 LT |
205 | } |
206 | ||
91a27b2a JL |
207 | struct filename * |
208 | getname(const char __user * filename) | |
f52e0c11 | 209 | { |
f7493e5d | 210 | return getname_flags(filename, 0, NULL); |
f52e0c11 AV |
211 | } |
212 | ||
1da177e4 | 213 | #ifdef CONFIG_AUDITSYSCALL |
91a27b2a | 214 | void putname(struct filename *name) |
1da177e4 | 215 | { |
5ac3a9c2 | 216 | if (unlikely(!audit_dummy_context())) |
91a27b2a JL |
217 | return audit_putname(name); |
218 | final_putname(name); | |
1da177e4 | 219 | } |
1da177e4 LT |
220 | #endif |
221 | ||
e77819e5 LT |
222 | static int check_acl(struct inode *inode, int mask) |
223 | { | |
84635d68 | 224 | #ifdef CONFIG_FS_POSIX_ACL |
e77819e5 LT |
225 | struct posix_acl *acl; |
226 | ||
e77819e5 | 227 | if (mask & MAY_NOT_BLOCK) { |
3567866b AV |
228 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
229 | if (!acl) | |
e77819e5 | 230 | return -EAGAIN; |
3567866b AV |
231 | /* no ->get_acl() calls in RCU mode... */ |
232 | if (acl == ACL_NOT_CACHED) | |
233 | return -ECHILD; | |
206b1d09 | 234 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
e77819e5 LT |
235 | } |
236 | ||
2982baa2 CH |
237 | acl = get_acl(inode, ACL_TYPE_ACCESS); |
238 | if (IS_ERR(acl)) | |
239 | return PTR_ERR(acl); | |
e77819e5 LT |
240 | if (acl) { |
241 | int error = posix_acl_permission(inode, acl, mask); | |
242 | posix_acl_release(acl); | |
243 | return error; | |
244 | } | |
84635d68 | 245 | #endif |
e77819e5 LT |
246 | |
247 | return -EAGAIN; | |
248 | } | |
249 | ||
5909ccaa | 250 | /* |
948409c7 | 251 | * This does the basic permission checking |
1da177e4 | 252 | */ |
7e40145e | 253 | static int acl_permission_check(struct inode *inode, int mask) |
1da177e4 | 254 | { |
26cf46be | 255 | unsigned int mode = inode->i_mode; |
1da177e4 | 256 | |
8e96e3b7 | 257 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) |
1da177e4 LT |
258 | mode >>= 6; |
259 | else { | |
e77819e5 | 260 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
7e40145e | 261 | int error = check_acl(inode, mask); |
b74c79e9 NP |
262 | if (error != -EAGAIN) |
263 | return error; | |
1da177e4 LT |
264 | } |
265 | ||
266 | if (in_group_p(inode->i_gid)) | |
267 | mode >>= 3; | |
268 | } | |
269 | ||
270 | /* | |
271 | * If the DACs are ok we don't need any capability check. | |
272 | */ | |
9c2c7039 | 273 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 | 274 | return 0; |
5909ccaa LT |
275 | return -EACCES; |
276 | } | |
277 | ||
278 | /** | |
b74c79e9 | 279 | * generic_permission - check for access rights on a Posix-like filesystem |
5909ccaa | 280 | * @inode: inode to check access rights for |
8fd90c8d | 281 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
5909ccaa LT |
282 | * |
283 | * Used to check for read/write/execute permissions on a file. | |
284 | * We use "fsuid" for this, letting us set arbitrary permissions | |
285 | * for filesystem access without changing the "normal" uids which | |
b74c79e9 NP |
286 | * are used for other things. |
287 | * | |
288 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
289 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
290 | * It would then be called again in ref-walk mode. | |
5909ccaa | 291 | */ |
2830ba7f | 292 | int generic_permission(struct inode *inode, int mask) |
5909ccaa LT |
293 | { |
294 | int ret; | |
295 | ||
296 | /* | |
948409c7 | 297 | * Do the basic permission checks. |
5909ccaa | 298 | */ |
7e40145e | 299 | ret = acl_permission_check(inode, mask); |
5909ccaa LT |
300 | if (ret != -EACCES) |
301 | return ret; | |
1da177e4 | 302 | |
d594e7ec AV |
303 | if (S_ISDIR(inode->i_mode)) { |
304 | /* DACs are overridable for directories */ | |
1a48e2ac | 305 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
d594e7ec AV |
306 | return 0; |
307 | if (!(mask & MAY_WRITE)) | |
1a48e2ac | 308 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
d594e7ec AV |
309 | return 0; |
310 | return -EACCES; | |
311 | } | |
1da177e4 LT |
312 | /* |
313 | * Read/write DACs are always overridable. | |
d594e7ec AV |
314 | * Executable DACs are overridable when there is |
315 | * at least one exec bit set. | |
1da177e4 | 316 | */ |
d594e7ec | 317 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
1a48e2ac | 318 | if (inode_capable(inode, CAP_DAC_OVERRIDE)) |
1da177e4 LT |
319 | return 0; |
320 | ||
321 | /* | |
322 | * Searching includes executable on directories, else just read. | |
323 | */ | |
7ea66001 | 324 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
d594e7ec | 325 | if (mask == MAY_READ) |
1a48e2ac | 326 | if (inode_capable(inode, CAP_DAC_READ_SEARCH)) |
1da177e4 LT |
327 | return 0; |
328 | ||
329 | return -EACCES; | |
330 | } | |
331 | ||
3ddcd056 LT |
332 | /* |
333 | * We _really_ want to just do "generic_permission()" without | |
334 | * even looking at the inode->i_op values. So we keep a cache | |
335 | * flag in inode->i_opflags, that says "this has not special | |
336 | * permission function, use the fast case". | |
337 | */ | |
338 | static inline int do_inode_permission(struct inode *inode, int mask) | |
339 | { | |
340 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
341 | if (likely(inode->i_op->permission)) | |
342 | return inode->i_op->permission(inode, mask); | |
343 | ||
344 | /* This gets set once for the inode lifetime */ | |
345 | spin_lock(&inode->i_lock); | |
346 | inode->i_opflags |= IOP_FASTPERM; | |
347 | spin_unlock(&inode->i_lock); | |
348 | } | |
349 | return generic_permission(inode, mask); | |
350 | } | |
351 | ||
cb23beb5 | 352 | /** |
0bdaea90 DH |
353 | * __inode_permission - Check for access rights to a given inode |
354 | * @inode: Inode to check permission on | |
355 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
cb23beb5 | 356 | * |
0bdaea90 | 357 | * Check for read/write/execute permissions on an inode. |
948409c7 AG |
358 | * |
359 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
0bdaea90 DH |
360 | * |
361 | * This does not check for a read-only file system. You probably want | |
362 | * inode_permission(). | |
cb23beb5 | 363 | */ |
0bdaea90 | 364 | int __inode_permission(struct inode *inode, int mask) |
1da177e4 | 365 | { |
e6305c43 | 366 | int retval; |
1da177e4 | 367 | |
3ddcd056 | 368 | if (unlikely(mask & MAY_WRITE)) { |
1da177e4 LT |
369 | /* |
370 | * Nobody gets write access to an immutable file. | |
371 | */ | |
372 | if (IS_IMMUTABLE(inode)) | |
373 | return -EACCES; | |
374 | } | |
375 | ||
3ddcd056 | 376 | retval = do_inode_permission(inode, mask); |
1da177e4 LT |
377 | if (retval) |
378 | return retval; | |
379 | ||
08ce5f16 SH |
380 | retval = devcgroup_inode_permission(inode, mask); |
381 | if (retval) | |
382 | return retval; | |
383 | ||
d09ca739 | 384 | return security_inode_permission(inode, mask); |
1da177e4 LT |
385 | } |
386 | ||
0bdaea90 DH |
387 | /** |
388 | * sb_permission - Check superblock-level permissions | |
389 | * @sb: Superblock of inode to check permission on | |
55852635 | 390 | * @inode: Inode to check permission on |
0bdaea90 DH |
391 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
392 | * | |
393 | * Separate out file-system wide checks from inode-specific permission checks. | |
394 | */ | |
395 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
396 | { | |
397 | if (unlikely(mask & MAY_WRITE)) { | |
398 | umode_t mode = inode->i_mode; | |
399 | ||
400 | /* Nobody gets write access to a read-only fs. */ | |
401 | if ((sb->s_flags & MS_RDONLY) && | |
402 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
403 | return -EROFS; | |
404 | } | |
405 | return 0; | |
406 | } | |
407 | ||
408 | /** | |
409 | * inode_permission - Check for access rights to a given inode | |
410 | * @inode: Inode to check permission on | |
411 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
412 | * | |
413 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
414 | * this, letting us set arbitrary permissions for filesystem access without | |
415 | * changing the "normal" UIDs which are used for other things. | |
416 | * | |
417 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
418 | */ | |
419 | int inode_permission(struct inode *inode, int mask) | |
420 | { | |
421 | int retval; | |
422 | ||
423 | retval = sb_permission(inode->i_sb, inode, mask); | |
424 | if (retval) | |
425 | return retval; | |
426 | return __inode_permission(inode, mask); | |
427 | } | |
428 | ||
5dd784d0 JB |
429 | /** |
430 | * path_get - get a reference to a path | |
431 | * @path: path to get the reference to | |
432 | * | |
433 | * Given a path increment the reference count to the dentry and the vfsmount. | |
434 | */ | |
dcf787f3 | 435 | void path_get(const struct path *path) |
5dd784d0 JB |
436 | { |
437 | mntget(path->mnt); | |
438 | dget(path->dentry); | |
439 | } | |
440 | EXPORT_SYMBOL(path_get); | |
441 | ||
1d957f9b JB |
442 | /** |
443 | * path_put - put a reference to a path | |
444 | * @path: path to put the reference to | |
445 | * | |
446 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
447 | */ | |
dcf787f3 | 448 | void path_put(const struct path *path) |
1da177e4 | 449 | { |
1d957f9b JB |
450 | dput(path->dentry); |
451 | mntput(path->mnt); | |
1da177e4 | 452 | } |
1d957f9b | 453 | EXPORT_SYMBOL(path_put); |
1da177e4 | 454 | |
19660af7 | 455 | /* |
31e6b01f | 456 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
19660af7 AV |
457 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
458 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
459 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk | |
460 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
461 | * got stuck, so ref-walk may continue from there. If this is not successful | |
462 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
463 | * to restart the path walk from the beginning in ref-walk mode. | |
31e6b01f | 464 | */ |
31e6b01f NP |
465 | |
466 | /** | |
19660af7 AV |
467 | * unlazy_walk - try to switch to ref-walk mode. |
468 | * @nd: nameidata pathwalk data | |
469 | * @dentry: child of nd->path.dentry or NULL | |
39191628 | 470 | * Returns: 0 on success, -ECHILD on failure |
31e6b01f | 471 | * |
19660af7 AV |
472 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
473 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
474 | * @nd or NULL. Must be called from rcu-walk context. | |
31e6b01f | 475 | */ |
19660af7 | 476 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
31e6b01f NP |
477 | { |
478 | struct fs_struct *fs = current->fs; | |
479 | struct dentry *parent = nd->path.dentry; | |
480 | ||
481 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
e5c832d5 LT |
482 | |
483 | /* | |
48a066e7 AV |
484 | * After legitimizing the bastards, terminate_walk() |
485 | * will do the right thing for non-RCU mode, and all our | |
486 | * subsequent exit cases should rcu_read_unlock() | |
487 | * before returning. Do vfsmount first; if dentry | |
488 | * can't be legitimized, just set nd->path.dentry to NULL | |
489 | * and rely on dput(NULL) being a no-op. | |
e5c832d5 | 490 | */ |
48a066e7 | 491 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) |
e5c832d5 | 492 | return -ECHILD; |
e5c832d5 | 493 | nd->flags &= ~LOOKUP_RCU; |
15570086 | 494 | |
48a066e7 AV |
495 | if (!lockref_get_not_dead(&parent->d_lockref)) { |
496 | nd->path.dentry = NULL; | |
d870b4a1 | 497 | goto out; |
48a066e7 AV |
498 | } |
499 | ||
15570086 LT |
500 | /* |
501 | * For a negative lookup, the lookup sequence point is the parents | |
502 | * sequence point, and it only needs to revalidate the parent dentry. | |
503 | * | |
504 | * For a positive lookup, we need to move both the parent and the | |
505 | * dentry from the RCU domain to be properly refcounted. And the | |
506 | * sequence number in the dentry validates *both* dentry counters, | |
507 | * since we checked the sequence number of the parent after we got | |
508 | * the child sequence number. So we know the parent must still | |
509 | * be valid if the child sequence number is still valid. | |
510 | */ | |
19660af7 | 511 | if (!dentry) { |
e5c832d5 LT |
512 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) |
513 | goto out; | |
19660af7 AV |
514 | BUG_ON(nd->inode != parent->d_inode); |
515 | } else { | |
e5c832d5 LT |
516 | if (!lockref_get_not_dead(&dentry->d_lockref)) |
517 | goto out; | |
518 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) | |
519 | goto drop_dentry; | |
19660af7 | 520 | } |
e5c832d5 LT |
521 | |
522 | /* | |
523 | * Sequence counts matched. Now make sure that the root is | |
524 | * still valid and get it if required. | |
525 | */ | |
526 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
527 | spin_lock(&fs->lock); | |
528 | if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) | |
529 | goto unlock_and_drop_dentry; | |
31e6b01f NP |
530 | path_get(&nd->root); |
531 | spin_unlock(&fs->lock); | |
532 | } | |
31e6b01f | 533 | |
8b61e74f | 534 | rcu_read_unlock(); |
31e6b01f | 535 | return 0; |
19660af7 | 536 | |
e5c832d5 LT |
537 | unlock_and_drop_dentry: |
538 | spin_unlock(&fs->lock); | |
539 | drop_dentry: | |
8b61e74f | 540 | rcu_read_unlock(); |
15570086 | 541 | dput(dentry); |
d0d27277 | 542 | goto drop_root_mnt; |
e5c832d5 | 543 | out: |
8b61e74f | 544 | rcu_read_unlock(); |
d0d27277 LT |
545 | drop_root_mnt: |
546 | if (!(nd->flags & LOOKUP_ROOT)) | |
547 | nd->root.mnt = NULL; | |
31e6b01f NP |
548 | return -ECHILD; |
549 | } | |
550 | ||
4ce16ef3 | 551 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) |
34286d66 | 552 | { |
4ce16ef3 | 553 | return dentry->d_op->d_revalidate(dentry, flags); |
34286d66 NP |
554 | } |
555 | ||
9f1fafee AV |
556 | /** |
557 | * complete_walk - successful completion of path walk | |
558 | * @nd: pointer nameidata | |
39159de2 | 559 | * |
9f1fafee AV |
560 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
561 | * Revalidate the final result, unless we'd already done that during | |
562 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
563 | * success, -error on failure. In case of failure caller does not | |
564 | * need to drop nd->path. | |
39159de2 | 565 | */ |
9f1fafee | 566 | static int complete_walk(struct nameidata *nd) |
39159de2 | 567 | { |
16c2cd71 | 568 | struct dentry *dentry = nd->path.dentry; |
39159de2 | 569 | int status; |
39159de2 | 570 | |
9f1fafee AV |
571 | if (nd->flags & LOOKUP_RCU) { |
572 | nd->flags &= ~LOOKUP_RCU; | |
573 | if (!(nd->flags & LOOKUP_ROOT)) | |
574 | nd->root.mnt = NULL; | |
15570086 | 575 | |
48a066e7 | 576 | if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { |
8b61e74f | 577 | rcu_read_unlock(); |
48a066e7 AV |
578 | return -ECHILD; |
579 | } | |
e5c832d5 | 580 | if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { |
8b61e74f | 581 | rcu_read_unlock(); |
48a066e7 | 582 | mntput(nd->path.mnt); |
e5c832d5 LT |
583 | return -ECHILD; |
584 | } | |
585 | if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { | |
8b61e74f | 586 | rcu_read_unlock(); |
e5c832d5 | 587 | dput(dentry); |
48a066e7 | 588 | mntput(nd->path.mnt); |
9f1fafee AV |
589 | return -ECHILD; |
590 | } | |
8b61e74f | 591 | rcu_read_unlock(); |
9f1fafee AV |
592 | } |
593 | ||
16c2cd71 AV |
594 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
595 | return 0; | |
596 | ||
ecf3d1f1 | 597 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) |
39159de2 JL |
598 | return 0; |
599 | ||
ecf3d1f1 | 600 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); |
39159de2 JL |
601 | if (status > 0) |
602 | return 0; | |
603 | ||
16c2cd71 | 604 | if (!status) |
39159de2 | 605 | status = -ESTALE; |
16c2cd71 | 606 | |
9f1fafee | 607 | path_put(&nd->path); |
39159de2 JL |
608 | return status; |
609 | } | |
610 | ||
2a737871 AV |
611 | static __always_inline void set_root(struct nameidata *nd) |
612 | { | |
f7ad3c6b MS |
613 | if (!nd->root.mnt) |
614 | get_fs_root(current->fs, &nd->root); | |
2a737871 AV |
615 | } |
616 | ||
6de88d72 AV |
617 | static int link_path_walk(const char *, struct nameidata *); |
618 | ||
31e6b01f NP |
619 | static __always_inline void set_root_rcu(struct nameidata *nd) |
620 | { | |
621 | if (!nd->root.mnt) { | |
622 | struct fs_struct *fs = current->fs; | |
c28cc364 NP |
623 | unsigned seq; |
624 | ||
625 | do { | |
626 | seq = read_seqcount_begin(&fs->seq); | |
627 | nd->root = fs->root; | |
c1530019 | 628 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
c28cc364 | 629 | } while (read_seqcount_retry(&fs->seq, seq)); |
31e6b01f NP |
630 | } |
631 | } | |
632 | ||
1d957f9b | 633 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
051d3812 IK |
634 | { |
635 | dput(path->dentry); | |
4ac91378 | 636 | if (path->mnt != nd->path.mnt) |
051d3812 IK |
637 | mntput(path->mnt); |
638 | } | |
639 | ||
7b9337aa NP |
640 | static inline void path_to_nameidata(const struct path *path, |
641 | struct nameidata *nd) | |
051d3812 | 642 | { |
31e6b01f NP |
643 | if (!(nd->flags & LOOKUP_RCU)) { |
644 | dput(nd->path.dentry); | |
645 | if (nd->path.mnt != path->mnt) | |
646 | mntput(nd->path.mnt); | |
9a229683 | 647 | } |
31e6b01f | 648 | nd->path.mnt = path->mnt; |
4ac91378 | 649 | nd->path.dentry = path->dentry; |
051d3812 IK |
650 | } |
651 | ||
b5fb63c1 CH |
652 | /* |
653 | * Helper to directly jump to a known parsed path from ->follow_link, | |
654 | * caller must have taken a reference to path beforehand. | |
655 | */ | |
656 | void nd_jump_link(struct nameidata *nd, struct path *path) | |
657 | { | |
658 | path_put(&nd->path); | |
659 | ||
660 | nd->path = *path; | |
661 | nd->inode = nd->path.dentry->d_inode; | |
662 | nd->flags |= LOOKUP_JUMPED; | |
b5fb63c1 CH |
663 | } |
664 | ||
574197e0 AV |
665 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
666 | { | |
667 | struct inode *inode = link->dentry->d_inode; | |
6d7b5aae | 668 | if (inode->i_op->put_link) |
574197e0 AV |
669 | inode->i_op->put_link(link->dentry, nd, cookie); |
670 | path_put(link); | |
671 | } | |
672 | ||
561ec64a LT |
673 | int sysctl_protected_symlinks __read_mostly = 0; |
674 | int sysctl_protected_hardlinks __read_mostly = 0; | |
800179c9 KC |
675 | |
676 | /** | |
677 | * may_follow_link - Check symlink following for unsafe situations | |
678 | * @link: The path of the symlink | |
55852635 | 679 | * @nd: nameidata pathwalk data |
800179c9 KC |
680 | * |
681 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
682 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
683 | * in a sticky world-writable directory. This is to protect privileged | |
684 | * processes from failing races against path names that may change out | |
685 | * from under them by way of other users creating malicious symlinks. | |
686 | * It will permit symlinks to be followed only when outside a sticky | |
687 | * world-writable directory, or when the uid of the symlink and follower | |
688 | * match, or when the directory owner matches the symlink's owner. | |
689 | * | |
690 | * Returns 0 if following the symlink is allowed, -ve on error. | |
691 | */ | |
692 | static inline int may_follow_link(struct path *link, struct nameidata *nd) | |
693 | { | |
694 | const struct inode *inode; | |
695 | const struct inode *parent; | |
696 | ||
697 | if (!sysctl_protected_symlinks) | |
698 | return 0; | |
699 | ||
700 | /* Allowed if owner and follower match. */ | |
701 | inode = link->dentry->d_inode; | |
81abe27b | 702 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) |
800179c9 KC |
703 | return 0; |
704 | ||
705 | /* Allowed if parent directory not sticky and world-writable. */ | |
706 | parent = nd->path.dentry->d_inode; | |
707 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
708 | return 0; | |
709 | ||
710 | /* Allowed if parent directory and link owner match. */ | |
81abe27b | 711 | if (uid_eq(parent->i_uid, inode->i_uid)) |
800179c9 KC |
712 | return 0; |
713 | ||
ffd8d101 | 714 | audit_log_link_denied("follow_link", link); |
800179c9 KC |
715 | path_put_conditional(link, nd); |
716 | path_put(&nd->path); | |
717 | return -EACCES; | |
718 | } | |
719 | ||
720 | /** | |
721 | * safe_hardlink_source - Check for safe hardlink conditions | |
722 | * @inode: the source inode to hardlink from | |
723 | * | |
724 | * Return false if at least one of the following conditions: | |
725 | * - inode is not a regular file | |
726 | * - inode is setuid | |
727 | * - inode is setgid and group-exec | |
728 | * - access failure for read and write | |
729 | * | |
730 | * Otherwise returns true. | |
731 | */ | |
732 | static bool safe_hardlink_source(struct inode *inode) | |
733 | { | |
734 | umode_t mode = inode->i_mode; | |
735 | ||
736 | /* Special files should not get pinned to the filesystem. */ | |
737 | if (!S_ISREG(mode)) | |
738 | return false; | |
739 | ||
740 | /* Setuid files should not get pinned to the filesystem. */ | |
741 | if (mode & S_ISUID) | |
742 | return false; | |
743 | ||
744 | /* Executable setgid files should not get pinned to the filesystem. */ | |
745 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
746 | return false; | |
747 | ||
748 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
749 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
750 | return false; | |
751 | ||
752 | return true; | |
753 | } | |
754 | ||
755 | /** | |
756 | * may_linkat - Check permissions for creating a hardlink | |
757 | * @link: the source to hardlink from | |
758 | * | |
759 | * Block hardlink when all of: | |
760 | * - sysctl_protected_hardlinks enabled | |
761 | * - fsuid does not match inode | |
762 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
763 | * - not CAP_FOWNER | |
764 | * | |
765 | * Returns 0 if successful, -ve on error. | |
766 | */ | |
767 | static int may_linkat(struct path *link) | |
768 | { | |
769 | const struct cred *cred; | |
770 | struct inode *inode; | |
771 | ||
772 | if (!sysctl_protected_hardlinks) | |
773 | return 0; | |
774 | ||
775 | cred = current_cred(); | |
776 | inode = link->dentry->d_inode; | |
777 | ||
778 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
779 | * otherwise, it must be a safe source. | |
780 | */ | |
81abe27b | 781 | if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || |
800179c9 KC |
782 | capable(CAP_FOWNER)) |
783 | return 0; | |
784 | ||
a51d9eaa | 785 | audit_log_link_denied("linkat", link); |
800179c9 KC |
786 | return -EPERM; |
787 | } | |
788 | ||
def4af30 | 789 | static __always_inline int |
574197e0 | 790 | follow_link(struct path *link, struct nameidata *nd, void **p) |
1da177e4 | 791 | { |
7b9337aa | 792 | struct dentry *dentry = link->dentry; |
6d7b5aae AV |
793 | int error; |
794 | char *s; | |
1da177e4 | 795 | |
844a3917 AV |
796 | BUG_ON(nd->flags & LOOKUP_RCU); |
797 | ||
0e794589 AV |
798 | if (link->mnt == nd->path.mnt) |
799 | mntget(link->mnt); | |
800 | ||
6d7b5aae AV |
801 | error = -ELOOP; |
802 | if (unlikely(current->total_link_count >= 40)) | |
803 | goto out_put_nd_path; | |
804 | ||
574197e0 AV |
805 | cond_resched(); |
806 | current->total_link_count++; | |
807 | ||
68ac1234 | 808 | touch_atime(link); |
1da177e4 | 809 | nd_set_link(nd, NULL); |
cd4e91d3 | 810 | |
36f3b4f6 | 811 | error = security_inode_follow_link(link->dentry, nd); |
6d7b5aae AV |
812 | if (error) |
813 | goto out_put_nd_path; | |
36f3b4f6 | 814 | |
86acdca1 | 815 | nd->last_type = LAST_BIND; |
def4af30 AV |
816 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
817 | error = PTR_ERR(*p); | |
6d7b5aae | 818 | if (IS_ERR(*p)) |
408ef013 | 819 | goto out_put_nd_path; |
6d7b5aae AV |
820 | |
821 | error = 0; | |
822 | s = nd_get_link(nd); | |
823 | if (s) { | |
443ed254 AV |
824 | if (unlikely(IS_ERR(s))) { |
825 | path_put(&nd->path); | |
826 | put_link(nd, link, *p); | |
827 | return PTR_ERR(s); | |
828 | } | |
829 | if (*s == '/') { | |
830 | set_root(nd); | |
831 | path_put(&nd->path); | |
832 | nd->path = nd->root; | |
833 | path_get(&nd->root); | |
834 | nd->flags |= LOOKUP_JUMPED; | |
835 | } | |
836 | nd->inode = nd->path.dentry->d_inode; | |
837 | error = link_path_walk(s, nd); | |
b5fb63c1 CH |
838 | if (unlikely(error)) |
839 | put_link(nd, link, *p); | |
1da177e4 | 840 | } |
6d7b5aae AV |
841 | |
842 | return error; | |
843 | ||
844 | out_put_nd_path: | |
98f6ef64 | 845 | *p = NULL; |
6d7b5aae | 846 | path_put(&nd->path); |
6d7b5aae | 847 | path_put(link); |
1da177e4 LT |
848 | return error; |
849 | } | |
850 | ||
31e6b01f NP |
851 | static int follow_up_rcu(struct path *path) |
852 | { | |
0714a533 AV |
853 | struct mount *mnt = real_mount(path->mnt); |
854 | struct mount *parent; | |
31e6b01f NP |
855 | struct dentry *mountpoint; |
856 | ||
0714a533 AV |
857 | parent = mnt->mnt_parent; |
858 | if (&parent->mnt == path->mnt) | |
31e6b01f | 859 | return 0; |
a73324da | 860 | mountpoint = mnt->mnt_mountpoint; |
31e6b01f | 861 | path->dentry = mountpoint; |
0714a533 | 862 | path->mnt = &parent->mnt; |
31e6b01f NP |
863 | return 1; |
864 | } | |
865 | ||
f015f126 DH |
866 | /* |
867 | * follow_up - Find the mountpoint of path's vfsmount | |
868 | * | |
869 | * Given a path, find the mountpoint of its source file system. | |
870 | * Replace @path with the path of the mountpoint in the parent mount. | |
871 | * Up is towards /. | |
872 | * | |
873 | * Return 1 if we went up a level and 0 if we were already at the | |
874 | * root. | |
875 | */ | |
bab77ebf | 876 | int follow_up(struct path *path) |
1da177e4 | 877 | { |
0714a533 AV |
878 | struct mount *mnt = real_mount(path->mnt); |
879 | struct mount *parent; | |
1da177e4 | 880 | struct dentry *mountpoint; |
99b7db7b | 881 | |
48a066e7 | 882 | read_seqlock_excl(&mount_lock); |
0714a533 | 883 | parent = mnt->mnt_parent; |
3c0a6163 | 884 | if (parent == mnt) { |
48a066e7 | 885 | read_sequnlock_excl(&mount_lock); |
1da177e4 LT |
886 | return 0; |
887 | } | |
0714a533 | 888 | mntget(&parent->mnt); |
a73324da | 889 | mountpoint = dget(mnt->mnt_mountpoint); |
48a066e7 | 890 | read_sequnlock_excl(&mount_lock); |
bab77ebf AV |
891 | dput(path->dentry); |
892 | path->dentry = mountpoint; | |
893 | mntput(path->mnt); | |
0714a533 | 894 | path->mnt = &parent->mnt; |
1da177e4 LT |
895 | return 1; |
896 | } | |
897 | ||
b5c84bf6 | 898 | /* |
9875cf80 DH |
899 | * Perform an automount |
900 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
901 | * were called with. | |
1da177e4 | 902 | */ |
9875cf80 DH |
903 | static int follow_automount(struct path *path, unsigned flags, |
904 | bool *need_mntput) | |
31e6b01f | 905 | { |
9875cf80 | 906 | struct vfsmount *mnt; |
ea5b778a | 907 | int err; |
9875cf80 DH |
908 | |
909 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
910 | return -EREMOTE; | |
911 | ||
0ec26fd0 MS |
912 | /* We don't want to mount if someone's just doing a stat - |
913 | * unless they're stat'ing a directory and appended a '/' to | |
914 | * the name. | |
915 | * | |
916 | * We do, however, want to mount if someone wants to open or | |
917 | * create a file of any type under the mountpoint, wants to | |
918 | * traverse through the mountpoint or wants to open the | |
919 | * mounted directory. Also, autofs may mark negative dentries | |
920 | * as being automount points. These will need the attentions | |
921 | * of the daemon to instantiate them before they can be used. | |
9875cf80 | 922 | */ |
0ec26fd0 | 923 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
d94c177b | 924 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
0ec26fd0 MS |
925 | path->dentry->d_inode) |
926 | return -EISDIR; | |
927 | ||
9875cf80 DH |
928 | current->total_link_count++; |
929 | if (current->total_link_count >= 40) | |
930 | return -ELOOP; | |
931 | ||
932 | mnt = path->dentry->d_op->d_automount(path); | |
933 | if (IS_ERR(mnt)) { | |
934 | /* | |
935 | * The filesystem is allowed to return -EISDIR here to indicate | |
936 | * it doesn't want to automount. For instance, autofs would do | |
937 | * this so that its userspace daemon can mount on this dentry. | |
938 | * | |
939 | * However, we can only permit this if it's a terminal point in | |
940 | * the path being looked up; if it wasn't then the remainder of | |
941 | * the path is inaccessible and we should say so. | |
942 | */ | |
49084c3b | 943 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
9875cf80 DH |
944 | return -EREMOTE; |
945 | return PTR_ERR(mnt); | |
31e6b01f | 946 | } |
ea5b778a | 947 | |
9875cf80 DH |
948 | if (!mnt) /* mount collision */ |
949 | return 0; | |
31e6b01f | 950 | |
8aef1884 AV |
951 | if (!*need_mntput) { |
952 | /* lock_mount() may release path->mnt on error */ | |
953 | mntget(path->mnt); | |
954 | *need_mntput = true; | |
955 | } | |
19a167af | 956 | err = finish_automount(mnt, path); |
9875cf80 | 957 | |
ea5b778a DH |
958 | switch (err) { |
959 | case -EBUSY: | |
960 | /* Someone else made a mount here whilst we were busy */ | |
19a167af | 961 | return 0; |
ea5b778a | 962 | case 0: |
8aef1884 | 963 | path_put(path); |
ea5b778a DH |
964 | path->mnt = mnt; |
965 | path->dentry = dget(mnt->mnt_root); | |
ea5b778a | 966 | return 0; |
19a167af AV |
967 | default: |
968 | return err; | |
ea5b778a | 969 | } |
19a167af | 970 | |
463ffb2e AV |
971 | } |
972 | ||
9875cf80 DH |
973 | /* |
974 | * Handle a dentry that is managed in some way. | |
cc53ce53 | 975 | * - Flagged for transit management (autofs) |
9875cf80 DH |
976 | * - Flagged as mountpoint |
977 | * - Flagged as automount point | |
978 | * | |
979 | * This may only be called in refwalk mode. | |
980 | * | |
981 | * Serialization is taken care of in namespace.c | |
982 | */ | |
983 | static int follow_managed(struct path *path, unsigned flags) | |
1da177e4 | 984 | { |
8aef1884 | 985 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
9875cf80 DH |
986 | unsigned managed; |
987 | bool need_mntput = false; | |
8aef1884 | 988 | int ret = 0; |
9875cf80 DH |
989 | |
990 | /* Given that we're not holding a lock here, we retain the value in a | |
991 | * local variable for each dentry as we look at it so that we don't see | |
992 | * the components of that value change under us */ | |
993 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
994 | managed &= DCACHE_MANAGED_DENTRY, | |
995 | unlikely(managed != 0)) { | |
cc53ce53 DH |
996 | /* Allow the filesystem to manage the transit without i_mutex |
997 | * being held. */ | |
998 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
999 | BUG_ON(!path->dentry->d_op); | |
1000 | BUG_ON(!path->dentry->d_op->d_manage); | |
1aed3e42 | 1001 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
cc53ce53 | 1002 | if (ret < 0) |
8aef1884 | 1003 | break; |
cc53ce53 DH |
1004 | } |
1005 | ||
9875cf80 DH |
1006 | /* Transit to a mounted filesystem. */ |
1007 | if (managed & DCACHE_MOUNTED) { | |
1008 | struct vfsmount *mounted = lookup_mnt(path); | |
1009 | if (mounted) { | |
1010 | dput(path->dentry); | |
1011 | if (need_mntput) | |
1012 | mntput(path->mnt); | |
1013 | path->mnt = mounted; | |
1014 | path->dentry = dget(mounted->mnt_root); | |
1015 | need_mntput = true; | |
1016 | continue; | |
1017 | } | |
1018 | ||
1019 | /* Something is mounted on this dentry in another | |
1020 | * namespace and/or whatever was mounted there in this | |
48a066e7 AV |
1021 | * namespace got unmounted before lookup_mnt() could |
1022 | * get it */ | |
9875cf80 DH |
1023 | } |
1024 | ||
1025 | /* Handle an automount point */ | |
1026 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1027 | ret = follow_automount(path, flags, &need_mntput); | |
1028 | if (ret < 0) | |
8aef1884 | 1029 | break; |
9875cf80 DH |
1030 | continue; |
1031 | } | |
1032 | ||
1033 | /* We didn't change the current path point */ | |
1034 | break; | |
1da177e4 | 1035 | } |
8aef1884 AV |
1036 | |
1037 | if (need_mntput && path->mnt == mnt) | |
1038 | mntput(path->mnt); | |
1039 | if (ret == -EISDIR) | |
1040 | ret = 0; | |
a3fbbde7 | 1041 | return ret < 0 ? ret : need_mntput; |
1da177e4 LT |
1042 | } |
1043 | ||
cc53ce53 | 1044 | int follow_down_one(struct path *path) |
1da177e4 LT |
1045 | { |
1046 | struct vfsmount *mounted; | |
1047 | ||
1c755af4 | 1048 | mounted = lookup_mnt(path); |
1da177e4 | 1049 | if (mounted) { |
9393bd07 AV |
1050 | dput(path->dentry); |
1051 | mntput(path->mnt); | |
1052 | path->mnt = mounted; | |
1053 | path->dentry = dget(mounted->mnt_root); | |
1da177e4 LT |
1054 | return 1; |
1055 | } | |
1056 | return 0; | |
1057 | } | |
1058 | ||
62a7375e IK |
1059 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
1060 | { | |
1061 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && | |
1062 | dentry->d_op->d_manage(dentry, true) < 0); | |
1063 | } | |
1064 | ||
9875cf80 | 1065 | /* |
287548e4 AV |
1066 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
1067 | * we meet a managed dentry that would need blocking. | |
9875cf80 DH |
1068 | */ |
1069 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
287548e4 | 1070 | struct inode **inode) |
9875cf80 | 1071 | { |
62a7375e | 1072 | for (;;) { |
c7105365 | 1073 | struct mount *mounted; |
62a7375e IK |
1074 | /* |
1075 | * Don't forget we might have a non-mountpoint managed dentry | |
1076 | * that wants to block transit. | |
1077 | */ | |
287548e4 | 1078 | if (unlikely(managed_dentry_might_block(path->dentry))) |
ab90911f | 1079 | return false; |
62a7375e IK |
1080 | |
1081 | if (!d_mountpoint(path->dentry)) | |
1082 | break; | |
1083 | ||
474279dc | 1084 | mounted = __lookup_mnt(path->mnt, path->dentry); |
9875cf80 DH |
1085 | if (!mounted) |
1086 | break; | |
c7105365 AV |
1087 | path->mnt = &mounted->mnt; |
1088 | path->dentry = mounted->mnt.mnt_root; | |
a3fbbde7 | 1089 | nd->flags |= LOOKUP_JUMPED; |
9875cf80 | 1090 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
59430262 LT |
1091 | /* |
1092 | * Update the inode too. We don't need to re-check the | |
1093 | * dentry sequence number here after this d_inode read, | |
1094 | * because a mount-point is always pinned. | |
1095 | */ | |
1096 | *inode = path->dentry->d_inode; | |
9875cf80 | 1097 | } |
9875cf80 DH |
1098 | return true; |
1099 | } | |
1100 | ||
dea39376 | 1101 | static void follow_mount_rcu(struct nameidata *nd) |
287548e4 | 1102 | { |
dea39376 | 1103 | while (d_mountpoint(nd->path.dentry)) { |
c7105365 | 1104 | struct mount *mounted; |
474279dc | 1105 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); |
287548e4 AV |
1106 | if (!mounted) |
1107 | break; | |
c7105365 AV |
1108 | nd->path.mnt = &mounted->mnt; |
1109 | nd->path.dentry = mounted->mnt.mnt_root; | |
dea39376 | 1110 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
287548e4 AV |
1111 | } |
1112 | } | |
1113 | ||
31e6b01f NP |
1114 | static int follow_dotdot_rcu(struct nameidata *nd) |
1115 | { | |
31e6b01f NP |
1116 | set_root_rcu(nd); |
1117 | ||
9875cf80 | 1118 | while (1) { |
31e6b01f NP |
1119 | if (nd->path.dentry == nd->root.dentry && |
1120 | nd->path.mnt == nd->root.mnt) { | |
1121 | break; | |
1122 | } | |
1123 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1124 | struct dentry *old = nd->path.dentry; | |
1125 | struct dentry *parent = old->d_parent; | |
1126 | unsigned seq; | |
1127 | ||
1128 | seq = read_seqcount_begin(&parent->d_seq); | |
1129 | if (read_seqcount_retry(&old->d_seq, nd->seq)) | |
ef7562d5 | 1130 | goto failed; |
31e6b01f NP |
1131 | nd->path.dentry = parent; |
1132 | nd->seq = seq; | |
1133 | break; | |
1134 | } | |
1135 | if (!follow_up_rcu(&nd->path)) | |
1136 | break; | |
1137 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
31e6b01f | 1138 | } |
dea39376 AV |
1139 | follow_mount_rcu(nd); |
1140 | nd->inode = nd->path.dentry->d_inode; | |
31e6b01f | 1141 | return 0; |
ef7562d5 AV |
1142 | |
1143 | failed: | |
1144 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1145 | if (!(nd->flags & LOOKUP_ROOT)) |
1146 | nd->root.mnt = NULL; | |
8b61e74f | 1147 | rcu_read_unlock(); |
ef7562d5 | 1148 | return -ECHILD; |
31e6b01f NP |
1149 | } |
1150 | ||
cc53ce53 DH |
1151 | /* |
1152 | * Follow down to the covering mount currently visible to userspace. At each | |
1153 | * point, the filesystem owning that dentry may be queried as to whether the | |
1154 | * caller is permitted to proceed or not. | |
cc53ce53 | 1155 | */ |
7cc90cc3 | 1156 | int follow_down(struct path *path) |
cc53ce53 DH |
1157 | { |
1158 | unsigned managed; | |
1159 | int ret; | |
1160 | ||
1161 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1162 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1163 | /* Allow the filesystem to manage the transit without i_mutex | |
1164 | * being held. | |
1165 | * | |
1166 | * We indicate to the filesystem if someone is trying to mount | |
1167 | * something here. This gives autofs the chance to deny anyone | |
1168 | * other than its daemon the right to mount on its | |
1169 | * superstructure. | |
1170 | * | |
1171 | * The filesystem may sleep at this point. | |
1172 | */ | |
1173 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1174 | BUG_ON(!path->dentry->d_op); | |
1175 | BUG_ON(!path->dentry->d_op->d_manage); | |
ab90911f | 1176 | ret = path->dentry->d_op->d_manage( |
1aed3e42 | 1177 | path->dentry, false); |
cc53ce53 DH |
1178 | if (ret < 0) |
1179 | return ret == -EISDIR ? 0 : ret; | |
1180 | } | |
1181 | ||
1182 | /* Transit to a mounted filesystem. */ | |
1183 | if (managed & DCACHE_MOUNTED) { | |
1184 | struct vfsmount *mounted = lookup_mnt(path); | |
1185 | if (!mounted) | |
1186 | break; | |
1187 | dput(path->dentry); | |
1188 | mntput(path->mnt); | |
1189 | path->mnt = mounted; | |
1190 | path->dentry = dget(mounted->mnt_root); | |
1191 | continue; | |
1192 | } | |
1193 | ||
1194 | /* Don't handle automount points here */ | |
1195 | break; | |
1196 | } | |
1197 | return 0; | |
1198 | } | |
1199 | ||
9875cf80 DH |
1200 | /* |
1201 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1202 | */ | |
1203 | static void follow_mount(struct path *path) | |
1204 | { | |
1205 | while (d_mountpoint(path->dentry)) { | |
1206 | struct vfsmount *mounted = lookup_mnt(path); | |
1207 | if (!mounted) | |
1208 | break; | |
1209 | dput(path->dentry); | |
1210 | mntput(path->mnt); | |
1211 | path->mnt = mounted; | |
1212 | path->dentry = dget(mounted->mnt_root); | |
1213 | } | |
1214 | } | |
1215 | ||
31e6b01f | 1216 | static void follow_dotdot(struct nameidata *nd) |
1da177e4 | 1217 | { |
2a737871 | 1218 | set_root(nd); |
e518ddb7 | 1219 | |
1da177e4 | 1220 | while(1) { |
4ac91378 | 1221 | struct dentry *old = nd->path.dentry; |
1da177e4 | 1222 | |
2a737871 AV |
1223 | if (nd->path.dentry == nd->root.dentry && |
1224 | nd->path.mnt == nd->root.mnt) { | |
1da177e4 LT |
1225 | break; |
1226 | } | |
4ac91378 | 1227 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
3088dd70 AV |
1228 | /* rare case of legitimate dget_parent()... */ |
1229 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1da177e4 LT |
1230 | dput(old); |
1231 | break; | |
1232 | } | |
3088dd70 | 1233 | if (!follow_up(&nd->path)) |
1da177e4 | 1234 | break; |
1da177e4 | 1235 | } |
79ed0226 | 1236 | follow_mount(&nd->path); |
31e6b01f | 1237 | nd->inode = nd->path.dentry->d_inode; |
1da177e4 LT |
1238 | } |
1239 | ||
baa03890 | 1240 | /* |
bad61189 MS |
1241 | * This looks up the name in dcache, possibly revalidates the old dentry and |
1242 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1243 | * returns whether i_op->lookup is necessary. | |
1244 | * | |
1245 | * dir->d_inode->i_mutex must be held | |
baa03890 | 1246 | */ |
bad61189 | 1247 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, |
201f956e | 1248 | unsigned int flags, bool *need_lookup) |
baa03890 | 1249 | { |
baa03890 | 1250 | struct dentry *dentry; |
bad61189 | 1251 | int error; |
baa03890 | 1252 | |
bad61189 MS |
1253 | *need_lookup = false; |
1254 | dentry = d_lookup(dir, name); | |
1255 | if (dentry) { | |
39e3c955 | 1256 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { |
201f956e | 1257 | error = d_revalidate(dentry, flags); |
bad61189 MS |
1258 | if (unlikely(error <= 0)) { |
1259 | if (error < 0) { | |
1260 | dput(dentry); | |
1261 | return ERR_PTR(error); | |
1262 | } else if (!d_invalidate(dentry)) { | |
1263 | dput(dentry); | |
1264 | dentry = NULL; | |
1265 | } | |
1266 | } | |
1267 | } | |
1268 | } | |
baa03890 | 1269 | |
bad61189 MS |
1270 | if (!dentry) { |
1271 | dentry = d_alloc(dir, name); | |
1272 | if (unlikely(!dentry)) | |
1273 | return ERR_PTR(-ENOMEM); | |
baa03890 | 1274 | |
bad61189 | 1275 | *need_lookup = true; |
baa03890 NP |
1276 | } |
1277 | return dentry; | |
1278 | } | |
1279 | ||
44396f4b | 1280 | /* |
13a2c3be BF |
1281 | * Call i_op->lookup on the dentry. The dentry must be negative and |
1282 | * unhashed. | |
bad61189 MS |
1283 | * |
1284 | * dir->d_inode->i_mutex must be held | |
44396f4b | 1285 | */ |
bad61189 | 1286 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, |
72bd866a | 1287 | unsigned int flags) |
44396f4b | 1288 | { |
44396f4b JB |
1289 | struct dentry *old; |
1290 | ||
1291 | /* Don't create child dentry for a dead directory. */ | |
bad61189 | 1292 | if (unlikely(IS_DEADDIR(dir))) { |
e188dc02 | 1293 | dput(dentry); |
44396f4b | 1294 | return ERR_PTR(-ENOENT); |
e188dc02 | 1295 | } |
44396f4b | 1296 | |
72bd866a | 1297 | old = dir->i_op->lookup(dir, dentry, flags); |
44396f4b JB |
1298 | if (unlikely(old)) { |
1299 | dput(dentry); | |
1300 | dentry = old; | |
1301 | } | |
1302 | return dentry; | |
1303 | } | |
1304 | ||
a3255546 | 1305 | static struct dentry *__lookup_hash(struct qstr *name, |
72bd866a | 1306 | struct dentry *base, unsigned int flags) |
a3255546 | 1307 | { |
bad61189 | 1308 | bool need_lookup; |
a3255546 AV |
1309 | struct dentry *dentry; |
1310 | ||
72bd866a | 1311 | dentry = lookup_dcache(name, base, flags, &need_lookup); |
bad61189 MS |
1312 | if (!need_lookup) |
1313 | return dentry; | |
a3255546 | 1314 | |
72bd866a | 1315 | return lookup_real(base->d_inode, dentry, flags); |
a3255546 AV |
1316 | } |
1317 | ||
1da177e4 LT |
1318 | /* |
1319 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
1320 | * small and for now I'd prefer to have fast path as straight as possible. | |
1321 | * It _is_ time-critical. | |
1322 | */ | |
e97cdc87 | 1323 | static int lookup_fast(struct nameidata *nd, |
697f514d | 1324 | struct path *path, struct inode **inode) |
1da177e4 | 1325 | { |
4ac91378 | 1326 | struct vfsmount *mnt = nd->path.mnt; |
31e6b01f | 1327 | struct dentry *dentry, *parent = nd->path.dentry; |
5a18fff2 AV |
1328 | int need_reval = 1; |
1329 | int status = 1; | |
9875cf80 DH |
1330 | int err; |
1331 | ||
b04f784e NP |
1332 | /* |
1333 | * Rename seqlock is not required here because in the off chance | |
1334 | * of a false negative due to a concurrent rename, we're going to | |
1335 | * do the non-racy lookup, below. | |
1336 | */ | |
31e6b01f NP |
1337 | if (nd->flags & LOOKUP_RCU) { |
1338 | unsigned seq; | |
da53be12 | 1339 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); |
5a18fff2 AV |
1340 | if (!dentry) |
1341 | goto unlazy; | |
1342 | ||
12f8ad4b LT |
1343 | /* |
1344 | * This sequence count validates that the inode matches | |
1345 | * the dentry name information from lookup. | |
1346 | */ | |
1347 | *inode = dentry->d_inode; | |
1348 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
1349 | return -ECHILD; | |
1350 | ||
1351 | /* | |
1352 | * This sequence count validates that the parent had no | |
1353 | * changes while we did the lookup of the dentry above. | |
1354 | * | |
1355 | * The memory barrier in read_seqcount_begin of child is | |
1356 | * enough, we can use __read_seqcount_retry here. | |
1357 | */ | |
31e6b01f NP |
1358 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1359 | return -ECHILD; | |
31e6b01f | 1360 | nd->seq = seq; |
5a18fff2 | 1361 | |
24643087 | 1362 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
4ce16ef3 | 1363 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1364 | if (unlikely(status <= 0)) { |
1365 | if (status != -ECHILD) | |
1366 | need_reval = 0; | |
1367 | goto unlazy; | |
1368 | } | |
24643087 | 1369 | } |
31e6b01f NP |
1370 | path->mnt = mnt; |
1371 | path->dentry = dentry; | |
d6e9bd25 AV |
1372 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1373 | goto unlazy; | |
1374 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) | |
1375 | goto unlazy; | |
1376 | return 0; | |
5a18fff2 | 1377 | unlazy: |
19660af7 AV |
1378 | if (unlazy_walk(nd, dentry)) |
1379 | return -ECHILD; | |
5a18fff2 | 1380 | } else { |
e97cdc87 | 1381 | dentry = __d_lookup(parent, &nd->last); |
9875cf80 | 1382 | } |
5a18fff2 | 1383 | |
81e6f520 AV |
1384 | if (unlikely(!dentry)) |
1385 | goto need_lookup; | |
1386 | ||
5a18fff2 | 1387 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
4ce16ef3 | 1388 | status = d_revalidate(dentry, nd->flags); |
5a18fff2 AV |
1389 | if (unlikely(status <= 0)) { |
1390 | if (status < 0) { | |
1391 | dput(dentry); | |
1392 | return status; | |
1393 | } | |
1394 | if (!d_invalidate(dentry)) { | |
1395 | dput(dentry); | |
81e6f520 | 1396 | goto need_lookup; |
5a18fff2 | 1397 | } |
24643087 | 1398 | } |
697f514d | 1399 | |
9875cf80 DH |
1400 | path->mnt = mnt; |
1401 | path->dentry = dentry; | |
1402 | err = follow_managed(path, nd->flags); | |
89312214 IK |
1403 | if (unlikely(err < 0)) { |
1404 | path_put_conditional(path, nd); | |
9875cf80 | 1405 | return err; |
89312214 | 1406 | } |
a3fbbde7 AV |
1407 | if (err) |
1408 | nd->flags |= LOOKUP_JUMPED; | |
9875cf80 | 1409 | *inode = path->dentry->d_inode; |
1da177e4 | 1410 | return 0; |
81e6f520 AV |
1411 | |
1412 | need_lookup: | |
697f514d MS |
1413 | return 1; |
1414 | } | |
1415 | ||
1416 | /* Fast lookup failed, do it the slow way */ | |
cc2a5271 | 1417 | static int lookup_slow(struct nameidata *nd, struct path *path) |
697f514d MS |
1418 | { |
1419 | struct dentry *dentry, *parent; | |
1420 | int err; | |
1421 | ||
1422 | parent = nd->path.dentry; | |
81e6f520 AV |
1423 | BUG_ON(nd->inode != parent->d_inode); |
1424 | ||
1425 | mutex_lock(&parent->d_inode->i_mutex); | |
cc2a5271 | 1426 | dentry = __lookup_hash(&nd->last, parent, nd->flags); |
81e6f520 AV |
1427 | mutex_unlock(&parent->d_inode->i_mutex); |
1428 | if (IS_ERR(dentry)) | |
1429 | return PTR_ERR(dentry); | |
697f514d MS |
1430 | path->mnt = nd->path.mnt; |
1431 | path->dentry = dentry; | |
1432 | err = follow_managed(path, nd->flags); | |
1433 | if (unlikely(err < 0)) { | |
1434 | path_put_conditional(path, nd); | |
1435 | return err; | |
1436 | } | |
1437 | if (err) | |
1438 | nd->flags |= LOOKUP_JUMPED; | |
1439 | return 0; | |
1da177e4 LT |
1440 | } |
1441 | ||
52094c8a AV |
1442 | static inline int may_lookup(struct nameidata *nd) |
1443 | { | |
1444 | if (nd->flags & LOOKUP_RCU) { | |
4ad5abb3 | 1445 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
52094c8a AV |
1446 | if (err != -ECHILD) |
1447 | return err; | |
19660af7 | 1448 | if (unlazy_walk(nd, NULL)) |
52094c8a AV |
1449 | return -ECHILD; |
1450 | } | |
4ad5abb3 | 1451 | return inode_permission(nd->inode, MAY_EXEC); |
52094c8a AV |
1452 | } |
1453 | ||
9856fa1b AV |
1454 | static inline int handle_dots(struct nameidata *nd, int type) |
1455 | { | |
1456 | if (type == LAST_DOTDOT) { | |
1457 | if (nd->flags & LOOKUP_RCU) { | |
1458 | if (follow_dotdot_rcu(nd)) | |
1459 | return -ECHILD; | |
1460 | } else | |
1461 | follow_dotdot(nd); | |
1462 | } | |
1463 | return 0; | |
1464 | } | |
1465 | ||
951361f9 AV |
1466 | static void terminate_walk(struct nameidata *nd) |
1467 | { | |
1468 | if (!(nd->flags & LOOKUP_RCU)) { | |
1469 | path_put(&nd->path); | |
1470 | } else { | |
1471 | nd->flags &= ~LOOKUP_RCU; | |
5b6ca027 AV |
1472 | if (!(nd->flags & LOOKUP_ROOT)) |
1473 | nd->root.mnt = NULL; | |
8b61e74f | 1474 | rcu_read_unlock(); |
951361f9 AV |
1475 | } |
1476 | } | |
1477 | ||
3ddcd056 LT |
1478 | /* |
1479 | * Do we need to follow links? We _really_ want to be able | |
1480 | * to do this check without having to look at inode->i_op, | |
1481 | * so we keep a cache of "no, this doesn't need follow_link" | |
1482 | * for the common case. | |
1483 | */ | |
b18825a7 | 1484 | static inline int should_follow_link(struct dentry *dentry, int follow) |
3ddcd056 | 1485 | { |
b18825a7 | 1486 | return unlikely(d_is_symlink(dentry)) ? follow : 0; |
3ddcd056 LT |
1487 | } |
1488 | ||
ce57dfc1 | 1489 | static inline int walk_component(struct nameidata *nd, struct path *path, |
21b9b073 | 1490 | int follow) |
ce57dfc1 AV |
1491 | { |
1492 | struct inode *inode; | |
1493 | int err; | |
1494 | /* | |
1495 | * "." and ".." are special - ".." especially so because it has | |
1496 | * to be able to know about the current root directory and | |
1497 | * parent relationships. | |
1498 | */ | |
21b9b073 AV |
1499 | if (unlikely(nd->last_type != LAST_NORM)) |
1500 | return handle_dots(nd, nd->last_type); | |
e97cdc87 | 1501 | err = lookup_fast(nd, path, &inode); |
ce57dfc1 | 1502 | if (unlikely(err)) { |
697f514d MS |
1503 | if (err < 0) |
1504 | goto out_err; | |
1505 | ||
cc2a5271 | 1506 | err = lookup_slow(nd, path); |
697f514d MS |
1507 | if (err < 0) |
1508 | goto out_err; | |
1509 | ||
1510 | inode = path->dentry->d_inode; | |
ce57dfc1 | 1511 | } |
697f514d MS |
1512 | err = -ENOENT; |
1513 | if (!inode) | |
1514 | goto out_path_put; | |
1515 | ||
b18825a7 | 1516 | if (should_follow_link(path->dentry, follow)) { |
19660af7 AV |
1517 | if (nd->flags & LOOKUP_RCU) { |
1518 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
697f514d MS |
1519 | err = -ECHILD; |
1520 | goto out_err; | |
19660af7 AV |
1521 | } |
1522 | } | |
ce57dfc1 AV |
1523 | BUG_ON(inode != path->dentry->d_inode); |
1524 | return 1; | |
1525 | } | |
1526 | path_to_nameidata(path, nd); | |
1527 | nd->inode = inode; | |
1528 | return 0; | |
697f514d MS |
1529 | |
1530 | out_path_put: | |
1531 | path_to_nameidata(path, nd); | |
1532 | out_err: | |
1533 | terminate_walk(nd); | |
1534 | return err; | |
ce57dfc1 AV |
1535 | } |
1536 | ||
b356379a AV |
1537 | /* |
1538 | * This limits recursive symlink follows to 8, while | |
1539 | * limiting consecutive symlinks to 40. | |
1540 | * | |
1541 | * Without that kind of total limit, nasty chains of consecutive | |
1542 | * symlinks can cause almost arbitrarily long lookups. | |
1543 | */ | |
1544 | static inline int nested_symlink(struct path *path, struct nameidata *nd) | |
1545 | { | |
1546 | int res; | |
1547 | ||
b356379a AV |
1548 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1549 | path_put_conditional(path, nd); | |
1550 | path_put(&nd->path); | |
1551 | return -ELOOP; | |
1552 | } | |
1a4022f8 | 1553 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
b356379a AV |
1554 | |
1555 | nd->depth++; | |
1556 | current->link_count++; | |
1557 | ||
1558 | do { | |
1559 | struct path link = *path; | |
1560 | void *cookie; | |
574197e0 AV |
1561 | |
1562 | res = follow_link(&link, nd, &cookie); | |
6d7b5aae AV |
1563 | if (res) |
1564 | break; | |
21b9b073 | 1565 | res = walk_component(nd, path, LOOKUP_FOLLOW); |
574197e0 | 1566 | put_link(nd, &link, cookie); |
b356379a AV |
1567 | } while (res > 0); |
1568 | ||
1569 | current->link_count--; | |
1570 | nd->depth--; | |
1571 | return res; | |
1572 | } | |
1573 | ||
bfcfaa77 LT |
1574 | /* |
1575 | * We can do the critical dentry name comparison and hashing | |
1576 | * operations one word at a time, but we are limited to: | |
1577 | * | |
1578 | * - Architectures with fast unaligned word accesses. We could | |
1579 | * do a "get_unaligned()" if this helps and is sufficiently | |
1580 | * fast. | |
1581 | * | |
bfcfaa77 LT |
1582 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we |
1583 | * do not trap on the (extremely unlikely) case of a page | |
1584 | * crossing operation. | |
1585 | * | |
1586 | * - Furthermore, we need an efficient 64-bit compile for the | |
1587 | * 64-bit case in order to generate the "number of bytes in | |
1588 | * the final mask". Again, that could be replaced with a | |
1589 | * efficient population count instruction or similar. | |
1590 | */ | |
1591 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1592 | ||
f68e556e | 1593 | #include <asm/word-at-a-time.h> |
bfcfaa77 | 1594 | |
f68e556e | 1595 | #ifdef CONFIG_64BIT |
bfcfaa77 LT |
1596 | |
1597 | static inline unsigned int fold_hash(unsigned long hash) | |
1598 | { | |
1599 | hash += hash >> (8*sizeof(int)); | |
1600 | return hash; | |
1601 | } | |
1602 | ||
1603 | #else /* 32-bit case */ | |
1604 | ||
bfcfaa77 LT |
1605 | #define fold_hash(x) (x) |
1606 | ||
1607 | #endif | |
1608 | ||
1609 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1610 | { | |
1611 | unsigned long a, mask; | |
1612 | unsigned long hash = 0; | |
1613 | ||
1614 | for (;;) { | |
e419b4cc | 1615 | a = load_unaligned_zeropad(name); |
bfcfaa77 LT |
1616 | if (len < sizeof(unsigned long)) |
1617 | break; | |
1618 | hash += a; | |
f132c5be | 1619 | hash *= 9; |
bfcfaa77 LT |
1620 | name += sizeof(unsigned long); |
1621 | len -= sizeof(unsigned long); | |
1622 | if (!len) | |
1623 | goto done; | |
1624 | } | |
a5c21dce | 1625 | mask = bytemask_from_count(len); |
bfcfaa77 LT |
1626 | hash += mask & a; |
1627 | done: | |
1628 | return fold_hash(hash); | |
1629 | } | |
1630 | EXPORT_SYMBOL(full_name_hash); | |
1631 | ||
bfcfaa77 LT |
1632 | /* |
1633 | * Calculate the length and hash of the path component, and | |
1634 | * return the length of the component; | |
1635 | */ | |
1636 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1637 | { | |
36126f8f LT |
1638 | unsigned long a, b, adata, bdata, mask, hash, len; |
1639 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
bfcfaa77 LT |
1640 | |
1641 | hash = a = 0; | |
1642 | len = -sizeof(unsigned long); | |
1643 | do { | |
1644 | hash = (hash + a) * 9; | |
1645 | len += sizeof(unsigned long); | |
e419b4cc | 1646 | a = load_unaligned_zeropad(name+len); |
36126f8f LT |
1647 | b = a ^ REPEAT_BYTE('/'); |
1648 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1649 | ||
1650 | adata = prep_zero_mask(a, adata, &constants); | |
1651 | bdata = prep_zero_mask(b, bdata, &constants); | |
1652 | ||
1653 | mask = create_zero_mask(adata | bdata); | |
1654 | ||
1655 | hash += a & zero_bytemask(mask); | |
bfcfaa77 LT |
1656 | *hashp = fold_hash(hash); |
1657 | ||
36126f8f | 1658 | return len + find_zero(mask); |
bfcfaa77 LT |
1659 | } |
1660 | ||
1661 | #else | |
1662 | ||
0145acc2 LT |
1663 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1664 | { | |
1665 | unsigned long hash = init_name_hash(); | |
1666 | while (len--) | |
1667 | hash = partial_name_hash(*name++, hash); | |
1668 | return end_name_hash(hash); | |
1669 | } | |
ae942ae7 | 1670 | EXPORT_SYMBOL(full_name_hash); |
0145acc2 | 1671 | |
200e9ef7 LT |
1672 | /* |
1673 | * We know there's a real path component here of at least | |
1674 | * one character. | |
1675 | */ | |
1676 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) | |
1677 | { | |
1678 | unsigned long hash = init_name_hash(); | |
1679 | unsigned long len = 0, c; | |
1680 | ||
1681 | c = (unsigned char)*name; | |
1682 | do { | |
1683 | len++; | |
1684 | hash = partial_name_hash(c, hash); | |
1685 | c = (unsigned char)name[len]; | |
1686 | } while (c && c != '/'); | |
1687 | *hashp = end_name_hash(hash); | |
1688 | return len; | |
1689 | } | |
1690 | ||
bfcfaa77 LT |
1691 | #endif |
1692 | ||
1da177e4 LT |
1693 | /* |
1694 | * Name resolution. | |
ea3834d9 PM |
1695 | * This is the basic name resolution function, turning a pathname into |
1696 | * the final dentry. We expect 'base' to be positive and a directory. | |
1da177e4 | 1697 | * |
ea3834d9 PM |
1698 | * Returns 0 and nd will have valid dentry and mnt on success. |
1699 | * Returns error and drops reference to input namei data on failure. | |
1da177e4 | 1700 | */ |
6de88d72 | 1701 | static int link_path_walk(const char *name, struct nameidata *nd) |
1da177e4 LT |
1702 | { |
1703 | struct path next; | |
1da177e4 | 1704 | int err; |
1da177e4 LT |
1705 | |
1706 | while (*name=='/') | |
1707 | name++; | |
1708 | if (!*name) | |
086e183a | 1709 | return 0; |
1da177e4 | 1710 | |
1da177e4 LT |
1711 | /* At this point we know we have a real path component. */ |
1712 | for(;;) { | |
1da177e4 | 1713 | struct qstr this; |
200e9ef7 | 1714 | long len; |
fe479a58 | 1715 | int type; |
1da177e4 | 1716 | |
52094c8a | 1717 | err = may_lookup(nd); |
1da177e4 LT |
1718 | if (err) |
1719 | break; | |
1720 | ||
200e9ef7 | 1721 | len = hash_name(name, &this.hash); |
1da177e4 | 1722 | this.name = name; |
200e9ef7 | 1723 | this.len = len; |
1da177e4 | 1724 | |
fe479a58 | 1725 | type = LAST_NORM; |
200e9ef7 | 1726 | if (name[0] == '.') switch (len) { |
fe479a58 | 1727 | case 2: |
200e9ef7 | 1728 | if (name[1] == '.') { |
fe479a58 | 1729 | type = LAST_DOTDOT; |
16c2cd71 AV |
1730 | nd->flags |= LOOKUP_JUMPED; |
1731 | } | |
fe479a58 AV |
1732 | break; |
1733 | case 1: | |
1734 | type = LAST_DOT; | |
1735 | } | |
5a202bcd AV |
1736 | if (likely(type == LAST_NORM)) { |
1737 | struct dentry *parent = nd->path.dentry; | |
16c2cd71 | 1738 | nd->flags &= ~LOOKUP_JUMPED; |
5a202bcd | 1739 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
da53be12 | 1740 | err = parent->d_op->d_hash(parent, &this); |
5a202bcd AV |
1741 | if (err < 0) |
1742 | break; | |
1743 | } | |
1744 | } | |
fe479a58 | 1745 | |
5f4a6a69 AV |
1746 | nd->last = this; |
1747 | nd->last_type = type; | |
1748 | ||
200e9ef7 | 1749 | if (!name[len]) |
5f4a6a69 | 1750 | return 0; |
200e9ef7 LT |
1751 | /* |
1752 | * If it wasn't NUL, we know it was '/'. Skip that | |
1753 | * slash, and continue until no more slashes. | |
1754 | */ | |
1755 | do { | |
1756 | len++; | |
1757 | } while (unlikely(name[len] == '/')); | |
1758 | if (!name[len]) | |
5f4a6a69 AV |
1759 | return 0; |
1760 | ||
200e9ef7 | 1761 | name += len; |
1da177e4 | 1762 | |
21b9b073 | 1763 | err = walk_component(nd, &next, LOOKUP_FOLLOW); |
ce57dfc1 AV |
1764 | if (err < 0) |
1765 | return err; | |
1da177e4 | 1766 | |
ce57dfc1 | 1767 | if (err) { |
b356379a | 1768 | err = nested_symlink(&next, nd); |
1da177e4 | 1769 | if (err) |
a7472bab | 1770 | return err; |
31e6b01f | 1771 | } |
b18825a7 | 1772 | if (!d_is_directory(nd->path.dentry)) { |
5f4a6a69 AV |
1773 | err = -ENOTDIR; |
1774 | break; | |
1775 | } | |
1da177e4 | 1776 | } |
951361f9 | 1777 | terminate_walk(nd); |
1da177e4 LT |
1778 | return err; |
1779 | } | |
1780 | ||
70e9b357 AV |
1781 | static int path_init(int dfd, const char *name, unsigned int flags, |
1782 | struct nameidata *nd, struct file **fp) | |
31e6b01f NP |
1783 | { |
1784 | int retval = 0; | |
31e6b01f NP |
1785 | |
1786 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
16c2cd71 | 1787 | nd->flags = flags | LOOKUP_JUMPED; |
31e6b01f | 1788 | nd->depth = 0; |
5b6ca027 | 1789 | if (flags & LOOKUP_ROOT) { |
b18825a7 DH |
1790 | struct dentry *root = nd->root.dentry; |
1791 | struct inode *inode = root->d_inode; | |
73d049a4 | 1792 | if (*name) { |
b18825a7 | 1793 | if (!d_is_directory(root)) |
73d049a4 AV |
1794 | return -ENOTDIR; |
1795 | retval = inode_permission(inode, MAY_EXEC); | |
1796 | if (retval) | |
1797 | return retval; | |
1798 | } | |
5b6ca027 AV |
1799 | nd->path = nd->root; |
1800 | nd->inode = inode; | |
1801 | if (flags & LOOKUP_RCU) { | |
8b61e74f | 1802 | rcu_read_lock(); |
5b6ca027 | 1803 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
48a066e7 | 1804 | nd->m_seq = read_seqbegin(&mount_lock); |
5b6ca027 AV |
1805 | } else { |
1806 | path_get(&nd->path); | |
1807 | } | |
1808 | return 0; | |
1809 | } | |
1810 | ||
31e6b01f | 1811 | nd->root.mnt = NULL; |
31e6b01f | 1812 | |
48a066e7 | 1813 | nd->m_seq = read_seqbegin(&mount_lock); |
31e6b01f | 1814 | if (*name=='/') { |
e41f7d4e | 1815 | if (flags & LOOKUP_RCU) { |
8b61e74f | 1816 | rcu_read_lock(); |
e41f7d4e AV |
1817 | set_root_rcu(nd); |
1818 | } else { | |
1819 | set_root(nd); | |
1820 | path_get(&nd->root); | |
1821 | } | |
1822 | nd->path = nd->root; | |
31e6b01f | 1823 | } else if (dfd == AT_FDCWD) { |
e41f7d4e AV |
1824 | if (flags & LOOKUP_RCU) { |
1825 | struct fs_struct *fs = current->fs; | |
1826 | unsigned seq; | |
31e6b01f | 1827 | |
8b61e74f | 1828 | rcu_read_lock(); |
c28cc364 | 1829 | |
e41f7d4e AV |
1830 | do { |
1831 | seq = read_seqcount_begin(&fs->seq); | |
1832 | nd->path = fs->pwd; | |
1833 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1834 | } while (read_seqcount_retry(&fs->seq, seq)); | |
1835 | } else { | |
1836 | get_fs_pwd(current->fs, &nd->path); | |
1837 | } | |
31e6b01f | 1838 | } else { |
582aa64a | 1839 | /* Caller must check execute permissions on the starting path component */ |
2903ff01 | 1840 | struct fd f = fdget_raw(dfd); |
31e6b01f NP |
1841 | struct dentry *dentry; |
1842 | ||
2903ff01 AV |
1843 | if (!f.file) |
1844 | return -EBADF; | |
31e6b01f | 1845 | |
2903ff01 | 1846 | dentry = f.file->f_path.dentry; |
31e6b01f | 1847 | |
f52e0c11 | 1848 | if (*name) { |
b18825a7 | 1849 | if (!d_is_directory(dentry)) { |
2903ff01 AV |
1850 | fdput(f); |
1851 | return -ENOTDIR; | |
1852 | } | |
f52e0c11 | 1853 | } |
31e6b01f | 1854 | |
2903ff01 | 1855 | nd->path = f.file->f_path; |
e41f7d4e | 1856 | if (flags & LOOKUP_RCU) { |
2903ff01 AV |
1857 | if (f.need_put) |
1858 | *fp = f.file; | |
e41f7d4e | 1859 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
8b61e74f | 1860 | rcu_read_lock(); |
e41f7d4e | 1861 | } else { |
2903ff01 AV |
1862 | path_get(&nd->path); |
1863 | fdput(f); | |
e41f7d4e | 1864 | } |
31e6b01f | 1865 | } |
31e6b01f | 1866 | |
31e6b01f | 1867 | nd->inode = nd->path.dentry->d_inode; |
9b4a9b14 | 1868 | return 0; |
9b4a9b14 AV |
1869 | } |
1870 | ||
bd92d7fe AV |
1871 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1872 | { | |
1873 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
1874 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
1875 | ||
1876 | nd->flags &= ~LOOKUP_PARENT; | |
21b9b073 | 1877 | return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); |
bd92d7fe AV |
1878 | } |
1879 | ||
9b4a9b14 | 1880 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
ee0827cd | 1881 | static int path_lookupat(int dfd, const char *name, |
9b4a9b14 AV |
1882 | unsigned int flags, struct nameidata *nd) |
1883 | { | |
70e9b357 | 1884 | struct file *base = NULL; |
bd92d7fe AV |
1885 | struct path path; |
1886 | int err; | |
31e6b01f NP |
1887 | |
1888 | /* | |
1889 | * Path walking is largely split up into 2 different synchronisation | |
1890 | * schemes, rcu-walk and ref-walk (explained in | |
1891 | * Documentation/filesystems/path-lookup.txt). These share much of the | |
1892 | * path walk code, but some things particularly setup, cleanup, and | |
1893 | * following mounts are sufficiently divergent that functions are | |
1894 | * duplicated. Typically there is a function foo(), and its RCU | |
1895 | * analogue, foo_rcu(). | |
1896 | * | |
1897 | * -ECHILD is the error number of choice (just to avoid clashes) that | |
1898 | * is returned if some aspect of an rcu-walk fails. Such an error must | |
1899 | * be handled by restarting a traditional ref-walk (which will always | |
1900 | * be able to complete). | |
1901 | */ | |
bd92d7fe | 1902 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
ee0827cd | 1903 | |
bd92d7fe AV |
1904 | if (unlikely(err)) |
1905 | return err; | |
ee0827cd AV |
1906 | |
1907 | current->total_link_count = 0; | |
bd92d7fe AV |
1908 | err = link_path_walk(name, nd); |
1909 | ||
1910 | if (!err && !(flags & LOOKUP_PARENT)) { | |
bd92d7fe AV |
1911 | err = lookup_last(nd, &path); |
1912 | while (err > 0) { | |
1913 | void *cookie; | |
1914 | struct path link = path; | |
800179c9 KC |
1915 | err = may_follow_link(&link, nd); |
1916 | if (unlikely(err)) | |
1917 | break; | |
bd92d7fe | 1918 | nd->flags |= LOOKUP_PARENT; |
574197e0 | 1919 | err = follow_link(&link, nd, &cookie); |
6d7b5aae AV |
1920 | if (err) |
1921 | break; | |
1922 | err = lookup_last(nd, &path); | |
574197e0 | 1923 | put_link(nd, &link, cookie); |
bd92d7fe AV |
1924 | } |
1925 | } | |
ee0827cd | 1926 | |
9f1fafee AV |
1927 | if (!err) |
1928 | err = complete_walk(nd); | |
bd92d7fe AV |
1929 | |
1930 | if (!err && nd->flags & LOOKUP_DIRECTORY) { | |
b18825a7 | 1931 | if (!d_is_directory(nd->path.dentry)) { |
bd92d7fe | 1932 | path_put(&nd->path); |
bd23a539 | 1933 | err = -ENOTDIR; |
bd92d7fe AV |
1934 | } |
1935 | } | |
16c2cd71 | 1936 | |
70e9b357 AV |
1937 | if (base) |
1938 | fput(base); | |
ee0827cd | 1939 | |
5b6ca027 | 1940 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
2a737871 AV |
1941 | path_put(&nd->root); |
1942 | nd->root.mnt = NULL; | |
1943 | } | |
bd92d7fe | 1944 | return err; |
ee0827cd | 1945 | } |
31e6b01f | 1946 | |
873f1eed | 1947 | static int filename_lookup(int dfd, struct filename *name, |
ee0827cd AV |
1948 | unsigned int flags, struct nameidata *nd) |
1949 | { | |
873f1eed | 1950 | int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); |
ee0827cd | 1951 | if (unlikely(retval == -ECHILD)) |
873f1eed | 1952 | retval = path_lookupat(dfd, name->name, flags, nd); |
ee0827cd | 1953 | if (unlikely(retval == -ESTALE)) |
873f1eed JL |
1954 | retval = path_lookupat(dfd, name->name, |
1955 | flags | LOOKUP_REVAL, nd); | |
31e6b01f | 1956 | |
f78570dd | 1957 | if (likely(!retval)) |
adb5c247 | 1958 | audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); |
170aa3d0 | 1959 | return retval; |
1da177e4 LT |
1960 | } |
1961 | ||
873f1eed JL |
1962 | static int do_path_lookup(int dfd, const char *name, |
1963 | unsigned int flags, struct nameidata *nd) | |
1964 | { | |
1965 | struct filename filename = { .name = name }; | |
1966 | ||
1967 | return filename_lookup(dfd, &filename, flags, nd); | |
1968 | } | |
1969 | ||
79714f72 AV |
1970 | /* does lookup, returns the object with parent locked */ |
1971 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
5590ff0d | 1972 | { |
79714f72 AV |
1973 | struct nameidata nd; |
1974 | struct dentry *d; | |
1975 | int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); | |
1976 | if (err) | |
1977 | return ERR_PTR(err); | |
1978 | if (nd.last_type != LAST_NORM) { | |
1979 | path_put(&nd.path); | |
1980 | return ERR_PTR(-EINVAL); | |
1981 | } | |
1982 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1e0ea001 | 1983 | d = __lookup_hash(&nd.last, nd.path.dentry, 0); |
79714f72 AV |
1984 | if (IS_ERR(d)) { |
1985 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
1986 | path_put(&nd.path); | |
1987 | return d; | |
1988 | } | |
1989 | *path = nd.path; | |
1990 | return d; | |
5590ff0d UD |
1991 | } |
1992 | ||
d1811465 AV |
1993 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1994 | { | |
1995 | struct nameidata nd; | |
1996 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1997 | if (!res) | |
1998 | *path = nd.path; | |
1999 | return res; | |
2000 | } | |
2001 | ||
16f18200 JJS |
2002 | /** |
2003 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2004 | * @dentry: pointer to dentry of the base directory | |
2005 | * @mnt: pointer to vfs mount of the base directory | |
2006 | * @name: pointer to file name | |
2007 | * @flags: lookup flags | |
e0a01249 | 2008 | * @path: pointer to struct path to fill |
16f18200 JJS |
2009 | */ |
2010 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2011 | const char *name, unsigned int flags, | |
e0a01249 | 2012 | struct path *path) |
16f18200 | 2013 | { |
e0a01249 AV |
2014 | struct nameidata nd; |
2015 | int err; | |
2016 | nd.root.dentry = dentry; | |
2017 | nd.root.mnt = mnt; | |
2018 | BUG_ON(flags & LOOKUP_PARENT); | |
5b6ca027 | 2019 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
e0a01249 AV |
2020 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
2021 | if (!err) | |
2022 | *path = nd.path; | |
2023 | return err; | |
16f18200 JJS |
2024 | } |
2025 | ||
057f6c01 JM |
2026 | /* |
2027 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
2028 | * needs parent already locked. Doesn't follow mounts. | |
2029 | * SMP-safe. | |
2030 | */ | |
eead1911 | 2031 | static struct dentry *lookup_hash(struct nameidata *nd) |
057f6c01 | 2032 | { |
72bd866a | 2033 | return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); |
1da177e4 LT |
2034 | } |
2035 | ||
eead1911 | 2036 | /** |
a6b91919 | 2037 | * lookup_one_len - filesystem helper to lookup single pathname component |
eead1911 CH |
2038 | * @name: pathname component to lookup |
2039 | * @base: base directory to lookup from | |
2040 | * @len: maximum length @len should be interpreted to | |
2041 | * | |
a6b91919 RD |
2042 | * Note that this routine is purely a helper for filesystem usage and should |
2043 | * not be called by generic code. Also note that by using this function the | |
eead1911 CH |
2044 | * nameidata argument is passed to the filesystem methods and a filesystem |
2045 | * using this helper needs to be prepared for that. | |
2046 | */ | |
057f6c01 JM |
2047 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
2048 | { | |
057f6c01 | 2049 | struct qstr this; |
6a96ba54 | 2050 | unsigned int c; |
cda309de | 2051 | int err; |
057f6c01 | 2052 | |
2f9092e1 DW |
2053 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
2054 | ||
6a96ba54 AV |
2055 | this.name = name; |
2056 | this.len = len; | |
0145acc2 | 2057 | this.hash = full_name_hash(name, len); |
6a96ba54 AV |
2058 | if (!len) |
2059 | return ERR_PTR(-EACCES); | |
2060 | ||
21d8a15a AV |
2061 | if (unlikely(name[0] == '.')) { |
2062 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2063 | return ERR_PTR(-EACCES); | |
2064 | } | |
2065 | ||
6a96ba54 AV |
2066 | while (len--) { |
2067 | c = *(const unsigned char *)name++; | |
2068 | if (c == '/' || c == '\0') | |
2069 | return ERR_PTR(-EACCES); | |
6a96ba54 | 2070 | } |
5a202bcd AV |
2071 | /* |
2072 | * See if the low-level filesystem might want | |
2073 | * to use its own hash.. | |
2074 | */ | |
2075 | if (base->d_flags & DCACHE_OP_HASH) { | |
da53be12 | 2076 | int err = base->d_op->d_hash(base, &this); |
5a202bcd AV |
2077 | if (err < 0) |
2078 | return ERR_PTR(err); | |
2079 | } | |
eead1911 | 2080 | |
cda309de MS |
2081 | err = inode_permission(base->d_inode, MAY_EXEC); |
2082 | if (err) | |
2083 | return ERR_PTR(err); | |
2084 | ||
72bd866a | 2085 | return __lookup_hash(&this, base, 0); |
057f6c01 JM |
2086 | } |
2087 | ||
1fa1e7f6 AW |
2088 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
2089 | struct path *path, int *empty) | |
1da177e4 | 2090 | { |
2d8f3038 | 2091 | struct nameidata nd; |
91a27b2a | 2092 | struct filename *tmp = getname_flags(name, flags, empty); |
1da177e4 | 2093 | int err = PTR_ERR(tmp); |
1da177e4 | 2094 | if (!IS_ERR(tmp)) { |
2d8f3038 AV |
2095 | |
2096 | BUG_ON(flags & LOOKUP_PARENT); | |
2097 | ||
873f1eed | 2098 | err = filename_lookup(dfd, tmp, flags, &nd); |
1da177e4 | 2099 | putname(tmp); |
2d8f3038 AV |
2100 | if (!err) |
2101 | *path = nd.path; | |
1da177e4 LT |
2102 | } |
2103 | return err; | |
2104 | } | |
2105 | ||
1fa1e7f6 AW |
2106 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
2107 | struct path *path) | |
2108 | { | |
f7493e5d | 2109 | return user_path_at_empty(dfd, name, flags, path, NULL); |
1fa1e7f6 AW |
2110 | } |
2111 | ||
873f1eed JL |
2112 | /* |
2113 | * NB: most callers don't do anything directly with the reference to the | |
2114 | * to struct filename, but the nd->last pointer points into the name string | |
2115 | * allocated by getname. So we must hold the reference to it until all | |
2116 | * path-walking is complete. | |
2117 | */ | |
91a27b2a | 2118 | static struct filename * |
9e790bd6 JL |
2119 | user_path_parent(int dfd, const char __user *path, struct nameidata *nd, |
2120 | unsigned int flags) | |
2ad94ae6 | 2121 | { |
91a27b2a | 2122 | struct filename *s = getname(path); |
2ad94ae6 AV |
2123 | int error; |
2124 | ||
9e790bd6 JL |
2125 | /* only LOOKUP_REVAL is allowed in extra flags */ |
2126 | flags &= LOOKUP_REVAL; | |
2127 | ||
2ad94ae6 | 2128 | if (IS_ERR(s)) |
91a27b2a | 2129 | return s; |
2ad94ae6 | 2130 | |
9e790bd6 | 2131 | error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); |
91a27b2a | 2132 | if (error) { |
2ad94ae6 | 2133 | putname(s); |
91a27b2a JL |
2134 | return ERR_PTR(error); |
2135 | } | |
2ad94ae6 | 2136 | |
91a27b2a | 2137 | return s; |
2ad94ae6 AV |
2138 | } |
2139 | ||
8033426e | 2140 | /** |
197df04c | 2141 | * mountpoint_last - look up last component for umount |
8033426e JL |
2142 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" |
2143 | * @path: pointer to container for result | |
2144 | * | |
2145 | * This is a special lookup_last function just for umount. In this case, we | |
2146 | * need to resolve the path without doing any revalidation. | |
2147 | * | |
2148 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2149 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2150 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2151 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2152 | * bogus and it doesn't exist. | |
2153 | * | |
2154 | * Returns: | |
2155 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2156 | * lookup found a negative dentry. The nd->path reference will also be | |
2157 | * put in this case. | |
2158 | * | |
2159 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2160 | * symlink that needs to be followed. "path" will also be populated. | |
2161 | * The nd->path reference will also be put. | |
2162 | * | |
2163 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2164 | * that needs to be followed. "path" will be populated with the path | |
2165 | * to the link, and nd->path will *not* be put. | |
2166 | */ | |
2167 | static int | |
197df04c | 2168 | mountpoint_last(struct nameidata *nd, struct path *path) |
8033426e JL |
2169 | { |
2170 | int error = 0; | |
2171 | struct dentry *dentry; | |
2172 | struct dentry *dir = nd->path.dentry; | |
2173 | ||
35759521 AV |
2174 | /* If we're in rcuwalk, drop out of it to handle last component */ |
2175 | if (nd->flags & LOOKUP_RCU) { | |
2176 | if (unlazy_walk(nd, NULL)) { | |
2177 | error = -ECHILD; | |
2178 | goto out; | |
2179 | } | |
8033426e JL |
2180 | } |
2181 | ||
2182 | nd->flags &= ~LOOKUP_PARENT; | |
2183 | ||
2184 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2185 | error = handle_dots(nd, nd->last_type); | |
35759521 AV |
2186 | if (error) |
2187 | goto out; | |
2188 | dentry = dget(nd->path.dentry); | |
2189 | goto done; | |
8033426e JL |
2190 | } |
2191 | ||
2192 | mutex_lock(&dir->d_inode->i_mutex); | |
2193 | dentry = d_lookup(dir, &nd->last); | |
2194 | if (!dentry) { | |
2195 | /* | |
2196 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2197 | * so that means that this dentry is probably a symlink or the | |
2198 | * path doesn't actually point to a mounted dentry. | |
2199 | */ | |
2200 | dentry = d_alloc(dir, &nd->last); | |
2201 | if (!dentry) { | |
2202 | error = -ENOMEM; | |
bcceeeba | 2203 | mutex_unlock(&dir->d_inode->i_mutex); |
35759521 | 2204 | goto out; |
8033426e | 2205 | } |
35759521 AV |
2206 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); |
2207 | error = PTR_ERR(dentry); | |
bcceeeba DJ |
2208 | if (IS_ERR(dentry)) { |
2209 | mutex_unlock(&dir->d_inode->i_mutex); | |
35759521 | 2210 | goto out; |
bcceeeba | 2211 | } |
8033426e JL |
2212 | } |
2213 | mutex_unlock(&dir->d_inode->i_mutex); | |
2214 | ||
35759521 AV |
2215 | done: |
2216 | if (!dentry->d_inode) { | |
2217 | error = -ENOENT; | |
2218 | dput(dentry); | |
2219 | goto out; | |
8033426e | 2220 | } |
35759521 AV |
2221 | path->dentry = dentry; |
2222 | path->mnt = mntget(nd->path.mnt); | |
b18825a7 | 2223 | if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) |
35759521 AV |
2224 | return 1; |
2225 | follow_mount(path); | |
2226 | error = 0; | |
2227 | out: | |
8033426e JL |
2228 | terminate_walk(nd); |
2229 | return error; | |
2230 | } | |
2231 | ||
2232 | /** | |
197df04c | 2233 | * path_mountpoint - look up a path to be umounted |
8033426e JL |
2234 | * @dfd: directory file descriptor to start walk from |
2235 | * @name: full pathname to walk | |
606d6fe3 | 2236 | * @path: pointer to container for result |
8033426e | 2237 | * @flags: lookup flags |
8033426e JL |
2238 | * |
2239 | * Look up the given name, but don't attempt to revalidate the last component. | |
606d6fe3 | 2240 | * Returns 0 and "path" will be valid on success; Returns error otherwise. |
8033426e JL |
2241 | */ |
2242 | static int | |
197df04c | 2243 | path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) |
8033426e JL |
2244 | { |
2245 | struct file *base = NULL; | |
2246 | struct nameidata nd; | |
2247 | int err; | |
2248 | ||
2249 | err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); | |
2250 | if (unlikely(err)) | |
2251 | return err; | |
2252 | ||
2253 | current->total_link_count = 0; | |
2254 | err = link_path_walk(name, &nd); | |
2255 | if (err) | |
2256 | goto out; | |
2257 | ||
197df04c | 2258 | err = mountpoint_last(&nd, path); |
8033426e JL |
2259 | while (err > 0) { |
2260 | void *cookie; | |
2261 | struct path link = *path; | |
2262 | err = may_follow_link(&link, &nd); | |
2263 | if (unlikely(err)) | |
2264 | break; | |
2265 | nd.flags |= LOOKUP_PARENT; | |
2266 | err = follow_link(&link, &nd, &cookie); | |
2267 | if (err) | |
2268 | break; | |
197df04c | 2269 | err = mountpoint_last(&nd, path); |
8033426e JL |
2270 | put_link(&nd, &link, cookie); |
2271 | } | |
2272 | out: | |
2273 | if (base) | |
2274 | fput(base); | |
2275 | ||
2276 | if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) | |
2277 | path_put(&nd.root); | |
2278 | ||
2279 | return err; | |
2280 | } | |
2281 | ||
2d864651 AV |
2282 | static int |
2283 | filename_mountpoint(int dfd, struct filename *s, struct path *path, | |
2284 | unsigned int flags) | |
2285 | { | |
2286 | int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); | |
2287 | if (unlikely(error == -ECHILD)) | |
2288 | error = path_mountpoint(dfd, s->name, path, flags); | |
2289 | if (unlikely(error == -ESTALE)) | |
2290 | error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); | |
2291 | if (likely(!error)) | |
2292 | audit_inode(s, path->dentry, 0); | |
2293 | return error; | |
2294 | } | |
2295 | ||
8033426e | 2296 | /** |
197df04c | 2297 | * user_path_mountpoint_at - lookup a path from userland in order to umount it |
8033426e JL |
2298 | * @dfd: directory file descriptor |
2299 | * @name: pathname from userland | |
2300 | * @flags: lookup flags | |
2301 | * @path: pointer to container to hold result | |
2302 | * | |
2303 | * A umount is a special case for path walking. We're not actually interested | |
2304 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2305 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2306 | * and avoid revalidating the last component. | |
2307 | * | |
2308 | * Returns 0 and populates "path" on success. | |
2309 | */ | |
2310 | int | |
197df04c | 2311 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, |
8033426e JL |
2312 | struct path *path) |
2313 | { | |
2314 | struct filename *s = getname(name); | |
2315 | int error; | |
8033426e JL |
2316 | if (IS_ERR(s)) |
2317 | return PTR_ERR(s); | |
2d864651 | 2318 | error = filename_mountpoint(dfd, s, path, flags); |
8033426e JL |
2319 | putname(s); |
2320 | return error; | |
2321 | } | |
2322 | ||
2d864651 AV |
2323 | int |
2324 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2325 | unsigned int flags) | |
2326 | { | |
2327 | struct filename s = {.name = name}; | |
2328 | return filename_mountpoint(dfd, &s, path, flags); | |
2329 | } | |
2330 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2331 | ||
1da177e4 LT |
2332 | /* |
2333 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
2334 | * minimal. | |
2335 | */ | |
2336 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
2337 | { | |
8e96e3b7 | 2338 | kuid_t fsuid = current_fsuid(); |
da9592ed | 2339 | |
1da177e4 LT |
2340 | if (!(dir->i_mode & S_ISVTX)) |
2341 | return 0; | |
8e96e3b7 | 2342 | if (uid_eq(inode->i_uid, fsuid)) |
1da177e4 | 2343 | return 0; |
8e96e3b7 | 2344 | if (uid_eq(dir->i_uid, fsuid)) |
1da177e4 | 2345 | return 0; |
1a48e2ac | 2346 | return !inode_capable(inode, CAP_FOWNER); |
1da177e4 LT |
2347 | } |
2348 | ||
2349 | /* | |
2350 | * Check whether we can remove a link victim from directory dir, check | |
2351 | * whether the type of victim is right. | |
2352 | * 1. We can't do it if dir is read-only (done in permission()) | |
2353 | * 2. We should have write and exec permissions on dir | |
2354 | * 3. We can't remove anything from append-only dir | |
2355 | * 4. We can't do anything with immutable dir (done in permission()) | |
2356 | * 5. If the sticky bit on dir is set we should either | |
2357 | * a. be owner of dir, or | |
2358 | * b. be owner of victim, or | |
2359 | * c. have CAP_FOWNER capability | |
2360 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2361 | * links pointing to it. | |
2362 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2363 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2364 | * 9. We can't remove a root or mountpoint. | |
2365 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2366 | * nfs_async_unlink(). | |
2367 | */ | |
b18825a7 | 2368 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) |
1da177e4 | 2369 | { |
b18825a7 | 2370 | struct inode *inode = victim->d_inode; |
1da177e4 LT |
2371 | int error; |
2372 | ||
b18825a7 | 2373 | if (d_is_negative(victim)) |
1da177e4 | 2374 | return -ENOENT; |
b18825a7 | 2375 | BUG_ON(!inode); |
1da177e4 LT |
2376 | |
2377 | BUG_ON(victim->d_parent->d_inode != dir); | |
4fa6b5ec | 2378 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); |
1da177e4 | 2379 | |
f419a2e3 | 2380 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2381 | if (error) |
2382 | return error; | |
2383 | if (IS_APPEND(dir)) | |
2384 | return -EPERM; | |
b18825a7 DH |
2385 | |
2386 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2387 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
1da177e4 LT |
2388 | return -EPERM; |
2389 | if (isdir) { | |
b18825a7 | 2390 | if (!d_is_directory(victim) && !d_is_autodir(victim)) |
1da177e4 LT |
2391 | return -ENOTDIR; |
2392 | if (IS_ROOT(victim)) | |
2393 | return -EBUSY; | |
b18825a7 | 2394 | } else if (d_is_directory(victim) || d_is_autodir(victim)) |
1da177e4 LT |
2395 | return -EISDIR; |
2396 | if (IS_DEADDIR(dir)) | |
2397 | return -ENOENT; | |
2398 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2399 | return -EBUSY; | |
2400 | return 0; | |
2401 | } | |
2402 | ||
2403 | /* Check whether we can create an object with dentry child in directory | |
2404 | * dir. | |
2405 | * 1. We can't do it if child already exists (open has special treatment for | |
2406 | * this case, but since we are inlined it's OK) | |
2407 | * 2. We can't do it if dir is read-only (done in permission()) | |
2408 | * 3. We should have write and exec permissions on dir | |
2409 | * 4. We can't do it if dir is immutable (done in permission()) | |
2410 | */ | |
a95164d9 | 2411 | static inline int may_create(struct inode *dir, struct dentry *child) |
1da177e4 | 2412 | { |
14e972b4 | 2413 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); |
1da177e4 LT |
2414 | if (child->d_inode) |
2415 | return -EEXIST; | |
2416 | if (IS_DEADDIR(dir)) | |
2417 | return -ENOENT; | |
f419a2e3 | 2418 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1da177e4 LT |
2419 | } |
2420 | ||
1da177e4 LT |
2421 | /* |
2422 | * p1 and p2 should be directories on the same fs. | |
2423 | */ | |
2424 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2425 | { | |
2426 | struct dentry *p; | |
2427 | ||
2428 | if (p1 == p2) { | |
f2eace23 | 2429 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1da177e4 LT |
2430 | return NULL; |
2431 | } | |
2432 | ||
a11f3a05 | 2433 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 | 2434 | |
e2761a11 OH |
2435 | p = d_ancestor(p2, p1); |
2436 | if (p) { | |
2437 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
2438 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
2439 | return p; | |
1da177e4 LT |
2440 | } |
2441 | ||
e2761a11 OH |
2442 | p = d_ancestor(p1, p2); |
2443 | if (p) { | |
2444 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
2445 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
2446 | return p; | |
1da177e4 LT |
2447 | } |
2448 | ||
f2eace23 IM |
2449 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2450 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1da177e4 LT |
2451 | return NULL; |
2452 | } | |
2453 | ||
2454 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2455 | { | |
1b1dcc1b | 2456 | mutex_unlock(&p1->d_inode->i_mutex); |
1da177e4 | 2457 | if (p1 != p2) { |
1b1dcc1b | 2458 | mutex_unlock(&p2->d_inode->i_mutex); |
a11f3a05 | 2459 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1da177e4 LT |
2460 | } |
2461 | } | |
2462 | ||
4acdaf27 | 2463 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
312b63fb | 2464 | bool want_excl) |
1da177e4 | 2465 | { |
a95164d9 | 2466 | int error = may_create(dir, dentry); |
1da177e4 LT |
2467 | if (error) |
2468 | return error; | |
2469 | ||
acfa4380 | 2470 | if (!dir->i_op->create) |
1da177e4 LT |
2471 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2472 | mode &= S_IALLUGO; | |
2473 | mode |= S_IFREG; | |
2474 | error = security_inode_create(dir, dentry, mode); | |
2475 | if (error) | |
2476 | return error; | |
312b63fb | 2477 | error = dir->i_op->create(dir, dentry, mode, want_excl); |
a74574aa | 2478 | if (!error) |
f38aa942 | 2479 | fsnotify_create(dir, dentry); |
1da177e4 LT |
2480 | return error; |
2481 | } | |
2482 | ||
73d049a4 | 2483 | static int may_open(struct path *path, int acc_mode, int flag) |
1da177e4 | 2484 | { |
3fb64190 | 2485 | struct dentry *dentry = path->dentry; |
1da177e4 LT |
2486 | struct inode *inode = dentry->d_inode; |
2487 | int error; | |
2488 | ||
bcda7652 AV |
2489 | /* O_PATH? */ |
2490 | if (!acc_mode) | |
2491 | return 0; | |
2492 | ||
1da177e4 LT |
2493 | if (!inode) |
2494 | return -ENOENT; | |
2495 | ||
c8fe8f30 CH |
2496 | switch (inode->i_mode & S_IFMT) { |
2497 | case S_IFLNK: | |
1da177e4 | 2498 | return -ELOOP; |
c8fe8f30 CH |
2499 | case S_IFDIR: |
2500 | if (acc_mode & MAY_WRITE) | |
2501 | return -EISDIR; | |
2502 | break; | |
2503 | case S_IFBLK: | |
2504 | case S_IFCHR: | |
3fb64190 | 2505 | if (path->mnt->mnt_flags & MNT_NODEV) |
1da177e4 | 2506 | return -EACCES; |
c8fe8f30 CH |
2507 | /*FALLTHRU*/ |
2508 | case S_IFIFO: | |
2509 | case S_IFSOCK: | |
1da177e4 | 2510 | flag &= ~O_TRUNC; |
c8fe8f30 | 2511 | break; |
4a3fd211 | 2512 | } |
b41572e9 | 2513 | |
3fb64190 | 2514 | error = inode_permission(inode, acc_mode); |
b41572e9 DH |
2515 | if (error) |
2516 | return error; | |
6146f0d5 | 2517 | |
1da177e4 LT |
2518 | /* |
2519 | * An append-only file must be opened in append mode for writing. | |
2520 | */ | |
2521 | if (IS_APPEND(inode)) { | |
8737c930 | 2522 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
7715b521 | 2523 | return -EPERM; |
1da177e4 | 2524 | if (flag & O_TRUNC) |
7715b521 | 2525 | return -EPERM; |
1da177e4 LT |
2526 | } |
2527 | ||
2528 | /* O_NOATIME can only be set by the owner or superuser */ | |
2e149670 | 2529 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
7715b521 | 2530 | return -EPERM; |
1da177e4 | 2531 | |
f3c7691e | 2532 | return 0; |
7715b521 | 2533 | } |
1da177e4 | 2534 | |
e1181ee6 | 2535 | static int handle_truncate(struct file *filp) |
7715b521 | 2536 | { |
e1181ee6 | 2537 | struct path *path = &filp->f_path; |
7715b521 AV |
2538 | struct inode *inode = path->dentry->d_inode; |
2539 | int error = get_write_access(inode); | |
2540 | if (error) | |
2541 | return error; | |
2542 | /* | |
2543 | * Refuse to truncate files with mandatory locks held on them. | |
2544 | */ | |
2545 | error = locks_verify_locked(inode); | |
2546 | if (!error) | |
ea0d3ab2 | 2547 | error = security_path_truncate(path); |
7715b521 AV |
2548 | if (!error) { |
2549 | error = do_truncate(path->dentry, 0, | |
2550 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
e1181ee6 | 2551 | filp); |
7715b521 AV |
2552 | } |
2553 | put_write_access(inode); | |
acd0c935 | 2554 | return error; |
1da177e4 LT |
2555 | } |
2556 | ||
d57999e1 DH |
2557 | static inline int open_to_namei_flags(int flag) |
2558 | { | |
8a5e929d AV |
2559 | if ((flag & O_ACCMODE) == 3) |
2560 | flag--; | |
d57999e1 DH |
2561 | return flag; |
2562 | } | |
2563 | ||
d18e9008 MS |
2564 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) |
2565 | { | |
2566 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2567 | if (error) | |
2568 | return error; | |
2569 | ||
2570 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2571 | if (error) | |
2572 | return error; | |
2573 | ||
2574 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2575 | } | |
2576 | ||
1acf0af9 DH |
2577 | /* |
2578 | * Attempt to atomically look up, create and open a file from a negative | |
2579 | * dentry. | |
2580 | * | |
2581 | * Returns 0 if successful. The file will have been created and attached to | |
2582 | * @file by the filesystem calling finish_open(). | |
2583 | * | |
2584 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2585 | * caller will need to perform the open themselves. @path will have been | |
2586 | * updated to point to the new dentry. This may be negative. | |
2587 | * | |
2588 | * Returns an error code otherwise. | |
2589 | */ | |
2675a4eb AV |
2590 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, |
2591 | struct path *path, struct file *file, | |
2592 | const struct open_flags *op, | |
64894cf8 | 2593 | bool got_write, bool need_lookup, |
2675a4eb | 2594 | int *opened) |
d18e9008 MS |
2595 | { |
2596 | struct inode *dir = nd->path.dentry->d_inode; | |
2597 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2598 | umode_t mode; | |
2599 | int error; | |
2600 | int acc_mode; | |
d18e9008 MS |
2601 | int create_error = 0; |
2602 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
116cc022 | 2603 | bool excl; |
d18e9008 MS |
2604 | |
2605 | BUG_ON(dentry->d_inode); | |
2606 | ||
2607 | /* Don't create child dentry for a dead directory. */ | |
2608 | if (unlikely(IS_DEADDIR(dir))) { | |
2675a4eb | 2609 | error = -ENOENT; |
d18e9008 MS |
2610 | goto out; |
2611 | } | |
2612 | ||
62b259d8 | 2613 | mode = op->mode; |
d18e9008 MS |
2614 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) |
2615 | mode &= ~current_umask(); | |
2616 | ||
116cc022 MS |
2617 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); |
2618 | if (excl) | |
d18e9008 | 2619 | open_flag &= ~O_TRUNC; |
d18e9008 MS |
2620 | |
2621 | /* | |
2622 | * Checking write permission is tricky, bacuse we don't know if we are | |
2623 | * going to actually need it: O_CREAT opens should work as long as the | |
2624 | * file exists. But checking existence breaks atomicity. The trick is | |
2625 | * to check access and if not granted clear O_CREAT from the flags. | |
2626 | * | |
2627 | * Another problem is returing the "right" error value (e.g. for an | |
2628 | * O_EXCL open we want to return EEXIST not EROFS). | |
2629 | */ | |
64894cf8 AV |
2630 | if (((open_flag & (O_CREAT | O_TRUNC)) || |
2631 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2632 | if (!(open_flag & O_CREAT)) { | |
d18e9008 MS |
2633 | /* |
2634 | * No O_CREATE -> atomicity not a requirement -> fall | |
2635 | * back to lookup + open | |
2636 | */ | |
2637 | goto no_open; | |
2638 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2639 | /* Fall back and fail with the right error */ | |
64894cf8 | 2640 | create_error = -EROFS; |
d18e9008 MS |
2641 | goto no_open; |
2642 | } else { | |
2643 | /* No side effects, safe to clear O_CREAT */ | |
64894cf8 | 2644 | create_error = -EROFS; |
d18e9008 MS |
2645 | open_flag &= ~O_CREAT; |
2646 | } | |
2647 | } | |
2648 | ||
2649 | if (open_flag & O_CREAT) { | |
38227f78 | 2650 | error = may_o_create(&nd->path, dentry, mode); |
d18e9008 MS |
2651 | if (error) { |
2652 | create_error = error; | |
2653 | if (open_flag & O_EXCL) | |
2654 | goto no_open; | |
2655 | open_flag &= ~O_CREAT; | |
2656 | } | |
2657 | } | |
2658 | ||
2659 | if (nd->flags & LOOKUP_DIRECTORY) | |
2660 | open_flag |= O_DIRECTORY; | |
2661 | ||
30d90494 AV |
2662 | file->f_path.dentry = DENTRY_NOT_SET; |
2663 | file->f_path.mnt = nd->path.mnt; | |
2664 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
47237687 | 2665 | opened); |
d9585277 | 2666 | if (error < 0) { |
d9585277 AV |
2667 | if (create_error && error == -ENOENT) |
2668 | error = create_error; | |
d18e9008 MS |
2669 | goto out; |
2670 | } | |
2671 | ||
d9585277 | 2672 | if (error) { /* returned 1, that is */ |
30d90494 | 2673 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { |
2675a4eb | 2674 | error = -EIO; |
d18e9008 MS |
2675 | goto out; |
2676 | } | |
30d90494 | 2677 | if (file->f_path.dentry) { |
d18e9008 | 2678 | dput(dentry); |
30d90494 | 2679 | dentry = file->f_path.dentry; |
d18e9008 | 2680 | } |
03da633a AV |
2681 | if (*opened & FILE_CREATED) |
2682 | fsnotify_create(dir, dentry); | |
2683 | if (!dentry->d_inode) { | |
2684 | WARN_ON(*opened & FILE_CREATED); | |
2685 | if (create_error) { | |
2686 | error = create_error; | |
2687 | goto out; | |
2688 | } | |
2689 | } else { | |
2690 | if (excl && !(*opened & FILE_CREATED)) { | |
2691 | error = -EEXIST; | |
2692 | goto out; | |
2693 | } | |
62b2ce96 | 2694 | } |
d18e9008 MS |
2695 | goto looked_up; |
2696 | } | |
2697 | ||
2698 | /* | |
2699 | * We didn't have the inode before the open, so check open permission | |
2700 | * here. | |
2701 | */ | |
03da633a AV |
2702 | acc_mode = op->acc_mode; |
2703 | if (*opened & FILE_CREATED) { | |
2704 | WARN_ON(!(open_flag & O_CREAT)); | |
2705 | fsnotify_create(dir, dentry); | |
2706 | acc_mode = MAY_OPEN; | |
2707 | } | |
2675a4eb AV |
2708 | error = may_open(&file->f_path, acc_mode, open_flag); |
2709 | if (error) | |
2710 | fput(file); | |
d18e9008 MS |
2711 | |
2712 | out: | |
2713 | dput(dentry); | |
2675a4eb | 2714 | return error; |
d18e9008 | 2715 | |
d18e9008 MS |
2716 | no_open: |
2717 | if (need_lookup) { | |
72bd866a | 2718 | dentry = lookup_real(dir, dentry, nd->flags); |
d18e9008 | 2719 | if (IS_ERR(dentry)) |
2675a4eb | 2720 | return PTR_ERR(dentry); |
d18e9008 MS |
2721 | |
2722 | if (create_error) { | |
2723 | int open_flag = op->open_flag; | |
2724 | ||
2675a4eb | 2725 | error = create_error; |
d18e9008 MS |
2726 | if ((open_flag & O_EXCL)) { |
2727 | if (!dentry->d_inode) | |
2728 | goto out; | |
2729 | } else if (!dentry->d_inode) { | |
2730 | goto out; | |
2731 | } else if ((open_flag & O_TRUNC) && | |
2732 | S_ISREG(dentry->d_inode->i_mode)) { | |
2733 | goto out; | |
2734 | } | |
2735 | /* will fail later, go on to get the right error */ | |
2736 | } | |
2737 | } | |
2738 | looked_up: | |
2739 | path->dentry = dentry; | |
2740 | path->mnt = nd->path.mnt; | |
2675a4eb | 2741 | return 1; |
d18e9008 MS |
2742 | } |
2743 | ||
d58ffd35 | 2744 | /* |
1acf0af9 | 2745 | * Look up and maybe create and open the last component. |
d58ffd35 MS |
2746 | * |
2747 | * Must be called with i_mutex held on parent. | |
2748 | * | |
1acf0af9 DH |
2749 | * Returns 0 if the file was successfully atomically created (if necessary) and |
2750 | * opened. In this case the file will be returned attached to @file. | |
2751 | * | |
2752 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2753 | * and creations will have been performed and the dentry returned in @path will | |
2754 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2755 | * specified then a negative dentry may be returned. | |
2756 | * | |
2757 | * An error code is returned otherwise. | |
2758 | * | |
2759 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2760 | * cleared otherwise prior to returning. | |
d58ffd35 | 2761 | */ |
2675a4eb AV |
2762 | static int lookup_open(struct nameidata *nd, struct path *path, |
2763 | struct file *file, | |
2764 | const struct open_flags *op, | |
64894cf8 | 2765 | bool got_write, int *opened) |
d58ffd35 MS |
2766 | { |
2767 | struct dentry *dir = nd->path.dentry; | |
54ef4872 | 2768 | struct inode *dir_inode = dir->d_inode; |
d58ffd35 MS |
2769 | struct dentry *dentry; |
2770 | int error; | |
54ef4872 | 2771 | bool need_lookup; |
d58ffd35 | 2772 | |
47237687 | 2773 | *opened &= ~FILE_CREATED; |
201f956e | 2774 | dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); |
d58ffd35 | 2775 | if (IS_ERR(dentry)) |
2675a4eb | 2776 | return PTR_ERR(dentry); |
d58ffd35 | 2777 | |
d18e9008 MS |
2778 | /* Cached positive dentry: will open in f_op->open */ |
2779 | if (!need_lookup && dentry->d_inode) | |
2780 | goto out_no_open; | |
2781 | ||
2782 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
64894cf8 | 2783 | return atomic_open(nd, dentry, path, file, op, got_write, |
47237687 | 2784 | need_lookup, opened); |
d18e9008 MS |
2785 | } |
2786 | ||
54ef4872 MS |
2787 | if (need_lookup) { |
2788 | BUG_ON(dentry->d_inode); | |
2789 | ||
72bd866a | 2790 | dentry = lookup_real(dir_inode, dentry, nd->flags); |
54ef4872 | 2791 | if (IS_ERR(dentry)) |
2675a4eb | 2792 | return PTR_ERR(dentry); |
54ef4872 MS |
2793 | } |
2794 | ||
d58ffd35 MS |
2795 | /* Negative dentry, just create the file */ |
2796 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
2797 | umode_t mode = op->mode; | |
2798 | if (!IS_POSIXACL(dir->d_inode)) | |
2799 | mode &= ~current_umask(); | |
2800 | /* | |
2801 | * This write is needed to ensure that a | |
2802 | * rw->ro transition does not occur between | |
2803 | * the time when the file is created and when | |
2804 | * a permanent write count is taken through | |
015c3bbc | 2805 | * the 'struct file' in finish_open(). |
d58ffd35 | 2806 | */ |
64894cf8 AV |
2807 | if (!got_write) { |
2808 | error = -EROFS; | |
d58ffd35 | 2809 | goto out_dput; |
64894cf8 | 2810 | } |
47237687 | 2811 | *opened |= FILE_CREATED; |
d58ffd35 MS |
2812 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2813 | if (error) | |
2814 | goto out_dput; | |
312b63fb AV |
2815 | error = vfs_create(dir->d_inode, dentry, mode, |
2816 | nd->flags & LOOKUP_EXCL); | |
d58ffd35 MS |
2817 | if (error) |
2818 | goto out_dput; | |
2819 | } | |
d18e9008 | 2820 | out_no_open: |
d58ffd35 MS |
2821 | path->dentry = dentry; |
2822 | path->mnt = nd->path.mnt; | |
2675a4eb | 2823 | return 1; |
d58ffd35 MS |
2824 | |
2825 | out_dput: | |
2826 | dput(dentry); | |
2675a4eb | 2827 | return error; |
d58ffd35 MS |
2828 | } |
2829 | ||
31e6b01f | 2830 | /* |
fe2d35ff | 2831 | * Handle the last step of open() |
31e6b01f | 2832 | */ |
2675a4eb AV |
2833 | static int do_last(struct nameidata *nd, struct path *path, |
2834 | struct file *file, const struct open_flags *op, | |
669abf4e | 2835 | int *opened, struct filename *name) |
fb1cc555 | 2836 | { |
a1e28038 | 2837 | struct dentry *dir = nd->path.dentry; |
ca344a89 | 2838 | int open_flag = op->open_flag; |
77d660a8 | 2839 | bool will_truncate = (open_flag & O_TRUNC) != 0; |
64894cf8 | 2840 | bool got_write = false; |
bcda7652 | 2841 | int acc_mode = op->acc_mode; |
a1eb3315 | 2842 | struct inode *inode; |
77d660a8 | 2843 | bool symlink_ok = false; |
16b1c1cd MS |
2844 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; |
2845 | bool retried = false; | |
16c2cd71 | 2846 | int error; |
1f36f774 | 2847 | |
c3e380b0 AV |
2848 | nd->flags &= ~LOOKUP_PARENT; |
2849 | nd->flags |= op->intent; | |
2850 | ||
bc77daa7 | 2851 | if (nd->last_type != LAST_NORM) { |
fe2d35ff AV |
2852 | error = handle_dots(nd, nd->last_type); |
2853 | if (error) | |
2675a4eb | 2854 | return error; |
e83db167 | 2855 | goto finish_open; |
1f36f774 | 2856 | } |
67ee3ad2 | 2857 | |
ca344a89 | 2858 | if (!(open_flag & O_CREAT)) { |
fe2d35ff AV |
2859 | if (nd->last.name[nd->last.len]) |
2860 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
bcda7652 | 2861 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
77d660a8 | 2862 | symlink_ok = true; |
fe2d35ff | 2863 | /* we _can_ be in RCU mode here */ |
e97cdc87 | 2864 | error = lookup_fast(nd, path, &inode); |
71574865 MS |
2865 | if (likely(!error)) |
2866 | goto finish_lookup; | |
2867 | ||
2868 | if (error < 0) | |
2675a4eb | 2869 | goto out; |
71574865 MS |
2870 | |
2871 | BUG_ON(nd->inode != dir->d_inode); | |
b6183df7 MS |
2872 | } else { |
2873 | /* create side of things */ | |
2874 | /* | |
2875 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
2876 | * has been cleared when we got to the last component we are | |
2877 | * about to look up | |
2878 | */ | |
2879 | error = complete_walk(nd); | |
2880 | if (error) | |
2675a4eb | 2881 | return error; |
fe2d35ff | 2882 | |
33e2208a | 2883 | audit_inode(name, dir, LOOKUP_PARENT); |
b6183df7 MS |
2884 | error = -EISDIR; |
2885 | /* trailing slashes? */ | |
2886 | if (nd->last.name[nd->last.len]) | |
2675a4eb | 2887 | goto out; |
b6183df7 | 2888 | } |
a2c36b45 | 2889 | |
16b1c1cd | 2890 | retry_lookup: |
64894cf8 AV |
2891 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { |
2892 | error = mnt_want_write(nd->path.mnt); | |
2893 | if (!error) | |
2894 | got_write = true; | |
2895 | /* | |
2896 | * do _not_ fail yet - we might not need that or fail with | |
2897 | * a different error; let lookup_open() decide; we'll be | |
2898 | * dropping this one anyway. | |
2899 | */ | |
2900 | } | |
a1e28038 | 2901 | mutex_lock(&dir->d_inode->i_mutex); |
64894cf8 | 2902 | error = lookup_open(nd, path, file, op, got_write, opened); |
d58ffd35 | 2903 | mutex_unlock(&dir->d_inode->i_mutex); |
a1e28038 | 2904 | |
2675a4eb AV |
2905 | if (error <= 0) { |
2906 | if (error) | |
d18e9008 MS |
2907 | goto out; |
2908 | ||
47237687 | 2909 | if ((*opened & FILE_CREATED) || |
496ad9aa | 2910 | !S_ISREG(file_inode(file)->i_mode)) |
77d660a8 | 2911 | will_truncate = false; |
d18e9008 | 2912 | |
adb5c247 | 2913 | audit_inode(name, file->f_path.dentry, 0); |
d18e9008 MS |
2914 | goto opened; |
2915 | } | |
fb1cc555 | 2916 | |
47237687 | 2917 | if (*opened & FILE_CREATED) { |
9b44f1b3 | 2918 | /* Don't check for write permission, don't truncate */ |
ca344a89 | 2919 | open_flag &= ~O_TRUNC; |
77d660a8 | 2920 | will_truncate = false; |
bcda7652 | 2921 | acc_mode = MAY_OPEN; |
d58ffd35 | 2922 | path_to_nameidata(path, nd); |
e83db167 | 2923 | goto finish_open_created; |
fb1cc555 AV |
2924 | } |
2925 | ||
2926 | /* | |
3134f37e | 2927 | * create/update audit record if it already exists. |
fb1cc555 | 2928 | */ |
b18825a7 | 2929 | if (d_is_positive(path->dentry)) |
adb5c247 | 2930 | audit_inode(name, path->dentry, 0); |
fb1cc555 | 2931 | |
d18e9008 MS |
2932 | /* |
2933 | * If atomic_open() acquired write access it is dropped now due to | |
2934 | * possible mount and symlink following (this might be optimized away if | |
2935 | * necessary...) | |
2936 | */ | |
64894cf8 | 2937 | if (got_write) { |
d18e9008 | 2938 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 2939 | got_write = false; |
d18e9008 MS |
2940 | } |
2941 | ||
fb1cc555 | 2942 | error = -EEXIST; |
f8310c59 | 2943 | if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) |
fb1cc555 AV |
2944 | goto exit_dput; |
2945 | ||
9875cf80 DH |
2946 | error = follow_managed(path, nd->flags); |
2947 | if (error < 0) | |
2948 | goto exit_dput; | |
fb1cc555 | 2949 | |
a3fbbde7 AV |
2950 | if (error) |
2951 | nd->flags |= LOOKUP_JUMPED; | |
2952 | ||
decf3400 MS |
2953 | BUG_ON(nd->flags & LOOKUP_RCU); |
2954 | inode = path->dentry->d_inode; | |
5f5daac1 MS |
2955 | finish_lookup: |
2956 | /* we _can_ be in RCU mode here */ | |
fb1cc555 | 2957 | error = -ENOENT; |
b18825a7 | 2958 | if (d_is_negative(path->dentry)) { |
54c33e7f | 2959 | path_to_nameidata(path, nd); |
2675a4eb | 2960 | goto out; |
54c33e7f | 2961 | } |
9e67f361 | 2962 | |
b18825a7 | 2963 | if (should_follow_link(path->dentry, !symlink_ok)) { |
d45ea867 MS |
2964 | if (nd->flags & LOOKUP_RCU) { |
2965 | if (unlikely(unlazy_walk(nd, path->dentry))) { | |
2966 | error = -ECHILD; | |
2675a4eb | 2967 | goto out; |
d45ea867 MS |
2968 | } |
2969 | } | |
2970 | BUG_ON(inode != path->dentry->d_inode); | |
2675a4eb | 2971 | return 1; |
d45ea867 | 2972 | } |
fb1cc555 | 2973 | |
16b1c1cd MS |
2974 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { |
2975 | path_to_nameidata(path, nd); | |
2976 | } else { | |
2977 | save_parent.dentry = nd->path.dentry; | |
2978 | save_parent.mnt = mntget(path->mnt); | |
2979 | nd->path.dentry = path->dentry; | |
2980 | ||
2981 | } | |
decf3400 | 2982 | nd->inode = inode; |
a3fbbde7 | 2983 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
bc77daa7 | 2984 | finish_open: |
a3fbbde7 | 2985 | error = complete_walk(nd); |
16b1c1cd MS |
2986 | if (error) { |
2987 | path_put(&save_parent); | |
2675a4eb | 2988 | return error; |
16b1c1cd | 2989 | } |
bc77daa7 | 2990 | audit_inode(name, nd->path.dentry, 0); |
fb1cc555 | 2991 | error = -EISDIR; |
b18825a7 DH |
2992 | if ((open_flag & O_CREAT) && |
2993 | (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry))) | |
2675a4eb | 2994 | goto out; |
af2f5542 | 2995 | error = -ENOTDIR; |
b18825a7 | 2996 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry)) |
2675a4eb | 2997 | goto out; |
6c0d46c4 | 2998 | if (!S_ISREG(nd->inode->i_mode)) |
77d660a8 | 2999 | will_truncate = false; |
6c0d46c4 | 3000 | |
0f9d1a10 AV |
3001 | if (will_truncate) { |
3002 | error = mnt_want_write(nd->path.mnt); | |
3003 | if (error) | |
2675a4eb | 3004 | goto out; |
64894cf8 | 3005 | got_write = true; |
0f9d1a10 | 3006 | } |
e83db167 | 3007 | finish_open_created: |
bcda7652 | 3008 | error = may_open(&nd->path, acc_mode, open_flag); |
ca344a89 | 3009 | if (error) |
2675a4eb | 3010 | goto out; |
30d90494 AV |
3011 | file->f_path.mnt = nd->path.mnt; |
3012 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3013 | if (error) { | |
30d90494 | 3014 | if (error == -EOPENSTALE) |
f60dc3db | 3015 | goto stale_open; |
015c3bbc | 3016 | goto out; |
f60dc3db | 3017 | } |
a8277b9b | 3018 | opened: |
2675a4eb | 3019 | error = open_check_o_direct(file); |
015c3bbc MS |
3020 | if (error) |
3021 | goto exit_fput; | |
2675a4eb | 3022 | error = ima_file_check(file, op->acc_mode); |
aa4caadb MS |
3023 | if (error) |
3024 | goto exit_fput; | |
3025 | ||
3026 | if (will_truncate) { | |
2675a4eb | 3027 | error = handle_truncate(file); |
aa4caadb MS |
3028 | if (error) |
3029 | goto exit_fput; | |
0f9d1a10 | 3030 | } |
ca344a89 | 3031 | out: |
64894cf8 | 3032 | if (got_write) |
0f9d1a10 | 3033 | mnt_drop_write(nd->path.mnt); |
16b1c1cd | 3034 | path_put(&save_parent); |
e276ae67 | 3035 | terminate_walk(nd); |
2675a4eb | 3036 | return error; |
fb1cc555 | 3037 | |
fb1cc555 AV |
3038 | exit_dput: |
3039 | path_put_conditional(path, nd); | |
ca344a89 | 3040 | goto out; |
015c3bbc | 3041 | exit_fput: |
2675a4eb AV |
3042 | fput(file); |
3043 | goto out; | |
015c3bbc | 3044 | |
f60dc3db MS |
3045 | stale_open: |
3046 | /* If no saved parent or already retried then can't retry */ | |
3047 | if (!save_parent.dentry || retried) | |
3048 | goto out; | |
3049 | ||
3050 | BUG_ON(save_parent.dentry != dir); | |
3051 | path_put(&nd->path); | |
3052 | nd->path = save_parent; | |
3053 | nd->inode = dir->d_inode; | |
3054 | save_parent.mnt = NULL; | |
3055 | save_parent.dentry = NULL; | |
64894cf8 | 3056 | if (got_write) { |
f60dc3db | 3057 | mnt_drop_write(nd->path.mnt); |
64894cf8 | 3058 | got_write = false; |
f60dc3db MS |
3059 | } |
3060 | retried = true; | |
3061 | goto retry_lookup; | |
fb1cc555 AV |
3062 | } |
3063 | ||
60545d0d AV |
3064 | static int do_tmpfile(int dfd, struct filename *pathname, |
3065 | struct nameidata *nd, int flags, | |
3066 | const struct open_flags *op, | |
3067 | struct file *file, int *opened) | |
3068 | { | |
3069 | static const struct qstr name = QSTR_INIT("/", 1); | |
3070 | struct dentry *dentry, *child; | |
3071 | struct inode *dir; | |
3072 | int error = path_lookupat(dfd, pathname->name, | |
3073 | flags | LOOKUP_DIRECTORY, nd); | |
3074 | if (unlikely(error)) | |
3075 | return error; | |
3076 | error = mnt_want_write(nd->path.mnt); | |
3077 | if (unlikely(error)) | |
3078 | goto out; | |
3079 | /* we want directory to be writable */ | |
3080 | error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); | |
3081 | if (error) | |
3082 | goto out2; | |
3083 | dentry = nd->path.dentry; | |
3084 | dir = dentry->d_inode; | |
3085 | if (!dir->i_op->tmpfile) { | |
3086 | error = -EOPNOTSUPP; | |
3087 | goto out2; | |
3088 | } | |
3089 | child = d_alloc(dentry, &name); | |
3090 | if (unlikely(!child)) { | |
3091 | error = -ENOMEM; | |
3092 | goto out2; | |
3093 | } | |
3094 | nd->flags &= ~LOOKUP_DIRECTORY; | |
3095 | nd->flags |= op->intent; | |
3096 | dput(nd->path.dentry); | |
3097 | nd->path.dentry = child; | |
3098 | error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); | |
3099 | if (error) | |
3100 | goto out2; | |
3101 | audit_inode(pathname, nd->path.dentry, 0); | |
3102 | error = may_open(&nd->path, op->acc_mode, op->open_flag); | |
3103 | if (error) | |
3104 | goto out2; | |
3105 | file->f_path.mnt = nd->path.mnt; | |
3106 | error = finish_open(file, nd->path.dentry, NULL, opened); | |
3107 | if (error) | |
3108 | goto out2; | |
3109 | error = open_check_o_direct(file); | |
f4e0c30c | 3110 | if (error) { |
60545d0d | 3111 | fput(file); |
f4e0c30c AV |
3112 | } else if (!(op->open_flag & O_EXCL)) { |
3113 | struct inode *inode = file_inode(file); | |
3114 | spin_lock(&inode->i_lock); | |
3115 | inode->i_state |= I_LINKABLE; | |
3116 | spin_unlock(&inode->i_lock); | |
3117 | } | |
60545d0d AV |
3118 | out2: |
3119 | mnt_drop_write(nd->path.mnt); | |
3120 | out: | |
3121 | path_put(&nd->path); | |
3122 | return error; | |
3123 | } | |
3124 | ||
669abf4e | 3125 | static struct file *path_openat(int dfd, struct filename *pathname, |
73d049a4 | 3126 | struct nameidata *nd, const struct open_flags *op, int flags) |
1da177e4 | 3127 | { |
fe2d35ff | 3128 | struct file *base = NULL; |
30d90494 | 3129 | struct file *file; |
9850c056 | 3130 | struct path path; |
47237687 | 3131 | int opened = 0; |
13aab428 | 3132 | int error; |
31e6b01f | 3133 | |
30d90494 | 3134 | file = get_empty_filp(); |
1afc99be AV |
3135 | if (IS_ERR(file)) |
3136 | return file; | |
31e6b01f | 3137 | |
30d90494 | 3138 | file->f_flags = op->open_flag; |
31e6b01f | 3139 | |
bb458c64 | 3140 | if (unlikely(file->f_flags & __O_TMPFILE)) { |
60545d0d AV |
3141 | error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); |
3142 | goto out; | |
3143 | } | |
3144 | ||
669abf4e | 3145 | error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); |
31e6b01f | 3146 | if (unlikely(error)) |
2675a4eb | 3147 | goto out; |
31e6b01f | 3148 | |
fe2d35ff | 3149 | current->total_link_count = 0; |
669abf4e | 3150 | error = link_path_walk(pathname->name, nd); |
31e6b01f | 3151 | if (unlikely(error)) |
2675a4eb | 3152 | goto out; |
1da177e4 | 3153 | |
2675a4eb AV |
3154 | error = do_last(nd, &path, file, op, &opened, pathname); |
3155 | while (unlikely(error > 0)) { /* trailing symlink */ | |
7b9337aa | 3156 | struct path link = path; |
def4af30 | 3157 | void *cookie; |
574197e0 | 3158 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
73d049a4 AV |
3159 | path_put_conditional(&path, nd); |
3160 | path_put(&nd->path); | |
2675a4eb | 3161 | error = -ELOOP; |
40b39136 AV |
3162 | break; |
3163 | } | |
800179c9 KC |
3164 | error = may_follow_link(&link, nd); |
3165 | if (unlikely(error)) | |
3166 | break; | |
73d049a4 AV |
3167 | nd->flags |= LOOKUP_PARENT; |
3168 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
574197e0 | 3169 | error = follow_link(&link, nd, &cookie); |
c3e380b0 | 3170 | if (unlikely(error)) |
2675a4eb AV |
3171 | break; |
3172 | error = do_last(nd, &path, file, op, &opened, pathname); | |
574197e0 | 3173 | put_link(nd, &link, cookie); |
806b681c | 3174 | } |
10fa8e62 | 3175 | out: |
73d049a4 AV |
3176 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
3177 | path_put(&nd->root); | |
fe2d35ff AV |
3178 | if (base) |
3179 | fput(base); | |
2675a4eb AV |
3180 | if (!(opened & FILE_OPENED)) { |
3181 | BUG_ON(!error); | |
30d90494 | 3182 | put_filp(file); |
16b1c1cd | 3183 | } |
2675a4eb AV |
3184 | if (unlikely(error)) { |
3185 | if (error == -EOPENSTALE) { | |
3186 | if (flags & LOOKUP_RCU) | |
3187 | error = -ECHILD; | |
3188 | else | |
3189 | error = -ESTALE; | |
3190 | } | |
3191 | file = ERR_PTR(error); | |
3192 | } | |
3193 | return file; | |
1da177e4 LT |
3194 | } |
3195 | ||
669abf4e | 3196 | struct file *do_filp_open(int dfd, struct filename *pathname, |
f9652e10 | 3197 | const struct open_flags *op) |
13aab428 | 3198 | { |
73d049a4 | 3199 | struct nameidata nd; |
f9652e10 | 3200 | int flags = op->lookup_flags; |
13aab428 AV |
3201 | struct file *filp; |
3202 | ||
73d049a4 | 3203 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
13aab428 | 3204 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
73d049a4 | 3205 | filp = path_openat(dfd, pathname, &nd, op, flags); |
13aab428 | 3206 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
73d049a4 | 3207 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
13aab428 AV |
3208 | return filp; |
3209 | } | |
3210 | ||
73d049a4 | 3211 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
f9652e10 | 3212 | const char *name, const struct open_flags *op) |
73d049a4 AV |
3213 | { |
3214 | struct nameidata nd; | |
3215 | struct file *file; | |
669abf4e | 3216 | struct filename filename = { .name = name }; |
f9652e10 | 3217 | int flags = op->lookup_flags | LOOKUP_ROOT; |
73d049a4 AV |
3218 | |
3219 | nd.root.mnt = mnt; | |
3220 | nd.root.dentry = dentry; | |
3221 | ||
b18825a7 | 3222 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) |
73d049a4 AV |
3223 | return ERR_PTR(-ELOOP); |
3224 | ||
669abf4e | 3225 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); |
73d049a4 | 3226 | if (unlikely(file == ERR_PTR(-ECHILD))) |
669abf4e | 3227 | file = path_openat(-1, &filename, &nd, op, flags); |
73d049a4 | 3228 | if (unlikely(file == ERR_PTR(-ESTALE))) |
669abf4e | 3229 | file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); |
73d049a4 AV |
3230 | return file; |
3231 | } | |
3232 | ||
1ac12b4b JL |
3233 | struct dentry *kern_path_create(int dfd, const char *pathname, |
3234 | struct path *path, unsigned int lookup_flags) | |
1da177e4 | 3235 | { |
c663e5d8 | 3236 | struct dentry *dentry = ERR_PTR(-EEXIST); |
ed75e95d | 3237 | struct nameidata nd; |
c30dabfe | 3238 | int err2; |
1ac12b4b JL |
3239 | int error; |
3240 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3241 | ||
3242 | /* | |
3243 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3244 | * other flags passed in are ignored! | |
3245 | */ | |
3246 | lookup_flags &= LOOKUP_REVAL; | |
3247 | ||
3248 | error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); | |
ed75e95d AV |
3249 | if (error) |
3250 | return ERR_PTR(error); | |
1da177e4 | 3251 | |
c663e5d8 CH |
3252 | /* |
3253 | * Yucky last component or no last component at all? | |
3254 | * (foo/., foo/.., /////) | |
3255 | */ | |
ed75e95d AV |
3256 | if (nd.last_type != LAST_NORM) |
3257 | goto out; | |
3258 | nd.flags &= ~LOOKUP_PARENT; | |
3259 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
c663e5d8 | 3260 | |
c30dabfe JK |
3261 | /* don't fail immediately if it's r/o, at least try to report other errors */ |
3262 | err2 = mnt_want_write(nd.path.mnt); | |
c663e5d8 CH |
3263 | /* |
3264 | * Do the final lookup. | |
3265 | */ | |
ed75e95d AV |
3266 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
3267 | dentry = lookup_hash(&nd); | |
1da177e4 | 3268 | if (IS_ERR(dentry)) |
a8104a9f | 3269 | goto unlock; |
c663e5d8 | 3270 | |
a8104a9f | 3271 | error = -EEXIST; |
b18825a7 | 3272 | if (d_is_positive(dentry)) |
a8104a9f | 3273 | goto fail; |
b18825a7 | 3274 | |
c663e5d8 CH |
3275 | /* |
3276 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3277 | * From the vfs_mknod() POV we just have a negative dentry - | |
3278 | * all is fine. Let's be bastards - you had / on the end, you've | |
3279 | * been asking for (non-existent) directory. -ENOENT for you. | |
3280 | */ | |
ed75e95d | 3281 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
a8104a9f | 3282 | error = -ENOENT; |
ed75e95d | 3283 | goto fail; |
e9baf6e5 | 3284 | } |
c30dabfe JK |
3285 | if (unlikely(err2)) { |
3286 | error = err2; | |
a8104a9f | 3287 | goto fail; |
c30dabfe | 3288 | } |
ed75e95d | 3289 | *path = nd.path; |
1da177e4 | 3290 | return dentry; |
1da177e4 | 3291 | fail: |
a8104a9f AV |
3292 | dput(dentry); |
3293 | dentry = ERR_PTR(error); | |
3294 | unlock: | |
ed75e95d | 3295 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe JK |
3296 | if (!err2) |
3297 | mnt_drop_write(nd.path.mnt); | |
ed75e95d AV |
3298 | out: |
3299 | path_put(&nd.path); | |
1da177e4 LT |
3300 | return dentry; |
3301 | } | |
dae6ad8f AV |
3302 | EXPORT_SYMBOL(kern_path_create); |
3303 | ||
921a1650 AV |
3304 | void done_path_create(struct path *path, struct dentry *dentry) |
3305 | { | |
3306 | dput(dentry); | |
3307 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
a8104a9f | 3308 | mnt_drop_write(path->mnt); |
921a1650 AV |
3309 | path_put(path); |
3310 | } | |
3311 | EXPORT_SYMBOL(done_path_create); | |
3312 | ||
1ac12b4b JL |
3313 | struct dentry *user_path_create(int dfd, const char __user *pathname, |
3314 | struct path *path, unsigned int lookup_flags) | |
dae6ad8f | 3315 | { |
91a27b2a | 3316 | struct filename *tmp = getname(pathname); |
dae6ad8f AV |
3317 | struct dentry *res; |
3318 | if (IS_ERR(tmp)) | |
3319 | return ERR_CAST(tmp); | |
1ac12b4b | 3320 | res = kern_path_create(dfd, tmp->name, path, lookup_flags); |
dae6ad8f AV |
3321 | putname(tmp); |
3322 | return res; | |
3323 | } | |
3324 | EXPORT_SYMBOL(user_path_create); | |
3325 | ||
1a67aafb | 3326 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3327 | { |
a95164d9 | 3328 | int error = may_create(dir, dentry); |
1da177e4 LT |
3329 | |
3330 | if (error) | |
3331 | return error; | |
3332 | ||
975d6b39 | 3333 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
1da177e4 LT |
3334 | return -EPERM; |
3335 | ||
acfa4380 | 3336 | if (!dir->i_op->mknod) |
1da177e4 LT |
3337 | return -EPERM; |
3338 | ||
08ce5f16 SH |
3339 | error = devcgroup_inode_mknod(mode, dev); |
3340 | if (error) | |
3341 | return error; | |
3342 | ||
1da177e4 LT |
3343 | error = security_inode_mknod(dir, dentry, mode, dev); |
3344 | if (error) | |
3345 | return error; | |
3346 | ||
1da177e4 | 3347 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
a74574aa | 3348 | if (!error) |
f38aa942 | 3349 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3350 | return error; |
3351 | } | |
3352 | ||
f69aac00 | 3353 | static int may_mknod(umode_t mode) |
463c3197 DH |
3354 | { |
3355 | switch (mode & S_IFMT) { | |
3356 | case S_IFREG: | |
3357 | case S_IFCHR: | |
3358 | case S_IFBLK: | |
3359 | case S_IFIFO: | |
3360 | case S_IFSOCK: | |
3361 | case 0: /* zero mode translates to S_IFREG */ | |
3362 | return 0; | |
3363 | case S_IFDIR: | |
3364 | return -EPERM; | |
3365 | default: | |
3366 | return -EINVAL; | |
3367 | } | |
3368 | } | |
3369 | ||
8208a22b | 3370 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2e4d0924 | 3371 | unsigned, dev) |
1da177e4 | 3372 | { |
2ad94ae6 | 3373 | struct dentry *dentry; |
dae6ad8f AV |
3374 | struct path path; |
3375 | int error; | |
972567f1 | 3376 | unsigned int lookup_flags = 0; |
1da177e4 | 3377 | |
8e4bfca1 AV |
3378 | error = may_mknod(mode); |
3379 | if (error) | |
3380 | return error; | |
972567f1 JL |
3381 | retry: |
3382 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
dae6ad8f AV |
3383 | if (IS_ERR(dentry)) |
3384 | return PTR_ERR(dentry); | |
2ad94ae6 | 3385 | |
dae6ad8f | 3386 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3387 | mode &= ~current_umask(); |
dae6ad8f | 3388 | error = security_path_mknod(&path, dentry, mode, dev); |
be6d3e56 | 3389 | if (error) |
a8104a9f | 3390 | goto out; |
463c3197 | 3391 | switch (mode & S_IFMT) { |
1da177e4 | 3392 | case 0: case S_IFREG: |
312b63fb | 3393 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); |
1da177e4 LT |
3394 | break; |
3395 | case S_IFCHR: case S_IFBLK: | |
dae6ad8f | 3396 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
1da177e4 LT |
3397 | new_decode_dev(dev)); |
3398 | break; | |
3399 | case S_IFIFO: case S_IFSOCK: | |
dae6ad8f | 3400 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
1da177e4 | 3401 | break; |
1da177e4 | 3402 | } |
a8104a9f | 3403 | out: |
921a1650 | 3404 | done_path_create(&path, dentry); |
972567f1 JL |
3405 | if (retry_estale(error, lookup_flags)) { |
3406 | lookup_flags |= LOOKUP_REVAL; | |
3407 | goto retry; | |
3408 | } | |
1da177e4 LT |
3409 | return error; |
3410 | } | |
3411 | ||
8208a22b | 3412 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
5590ff0d UD |
3413 | { |
3414 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3415 | } | |
3416 | ||
18bb1db3 | 3417 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 | 3418 | { |
a95164d9 | 3419 | int error = may_create(dir, dentry); |
8de52778 | 3420 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3421 | |
3422 | if (error) | |
3423 | return error; | |
3424 | ||
acfa4380 | 3425 | if (!dir->i_op->mkdir) |
1da177e4 LT |
3426 | return -EPERM; |
3427 | ||
3428 | mode &= (S_IRWXUGO|S_ISVTX); | |
3429 | error = security_inode_mkdir(dir, dentry, mode); | |
3430 | if (error) | |
3431 | return error; | |
3432 | ||
8de52778 AV |
3433 | if (max_links && dir->i_nlink >= max_links) |
3434 | return -EMLINK; | |
3435 | ||
1da177e4 | 3436 | error = dir->i_op->mkdir(dir, dentry, mode); |
a74574aa | 3437 | if (!error) |
f38aa942 | 3438 | fsnotify_mkdir(dir, dentry); |
1da177e4 LT |
3439 | return error; |
3440 | } | |
3441 | ||
a218d0fd | 3442 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
1da177e4 | 3443 | { |
6902d925 | 3444 | struct dentry *dentry; |
dae6ad8f AV |
3445 | struct path path; |
3446 | int error; | |
b76d8b82 | 3447 | unsigned int lookup_flags = LOOKUP_DIRECTORY; |
1da177e4 | 3448 | |
b76d8b82 JL |
3449 | retry: |
3450 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
6902d925 | 3451 | if (IS_ERR(dentry)) |
dae6ad8f | 3452 | return PTR_ERR(dentry); |
1da177e4 | 3453 | |
dae6ad8f | 3454 | if (!IS_POSIXACL(path.dentry->d_inode)) |
ce3b0f8d | 3455 | mode &= ~current_umask(); |
dae6ad8f | 3456 | error = security_path_mkdir(&path, dentry, mode); |
a8104a9f AV |
3457 | if (!error) |
3458 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
921a1650 | 3459 | done_path_create(&path, dentry); |
b76d8b82 JL |
3460 | if (retry_estale(error, lookup_flags)) { |
3461 | lookup_flags |= LOOKUP_REVAL; | |
3462 | goto retry; | |
3463 | } | |
1da177e4 LT |
3464 | return error; |
3465 | } | |
3466 | ||
a218d0fd | 3467 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
5590ff0d UD |
3468 | { |
3469 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3470 | } | |
3471 | ||
1da177e4 | 3472 | /* |
a71905f0 | 3473 | * The dentry_unhash() helper will try to drop the dentry early: we |
c0d02594 | 3474 | * should have a usage count of 1 if we're the only user of this |
a71905f0 SW |
3475 | * dentry, and if that is true (possibly after pruning the dcache), |
3476 | * then we drop the dentry now. | |
1da177e4 LT |
3477 | * |
3478 | * A low-level filesystem can, if it choses, legally | |
3479 | * do a | |
3480 | * | |
3481 | * if (!d_unhashed(dentry)) | |
3482 | * return -EBUSY; | |
3483 | * | |
3484 | * if it cannot handle the case of removing a directory | |
3485 | * that is still in use by something else.. | |
3486 | */ | |
3487 | void dentry_unhash(struct dentry *dentry) | |
3488 | { | |
dc168427 | 3489 | shrink_dcache_parent(dentry); |
1da177e4 | 3490 | spin_lock(&dentry->d_lock); |
98474236 | 3491 | if (dentry->d_lockref.count == 1) |
1da177e4 LT |
3492 | __d_drop(dentry); |
3493 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
3494 | } |
3495 | ||
3496 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3497 | { | |
3498 | int error = may_delete(dir, dentry, 1); | |
3499 | ||
3500 | if (error) | |
3501 | return error; | |
3502 | ||
acfa4380 | 3503 | if (!dir->i_op->rmdir) |
1da177e4 LT |
3504 | return -EPERM; |
3505 | ||
1d2ef590 | 3506 | dget(dentry); |
1b1dcc1b | 3507 | mutex_lock(&dentry->d_inode->i_mutex); |
912dbc15 SW |
3508 | |
3509 | error = -EBUSY; | |
1da177e4 | 3510 | if (d_mountpoint(dentry)) |
912dbc15 SW |
3511 | goto out; |
3512 | ||
3513 | error = security_inode_rmdir(dir, dentry); | |
3514 | if (error) | |
3515 | goto out; | |
3516 | ||
3cebde24 | 3517 | shrink_dcache_parent(dentry); |
912dbc15 SW |
3518 | error = dir->i_op->rmdir(dir, dentry); |
3519 | if (error) | |
3520 | goto out; | |
3521 | ||
3522 | dentry->d_inode->i_flags |= S_DEAD; | |
3523 | dont_mount(dentry); | |
3524 | ||
3525 | out: | |
1b1dcc1b | 3526 | mutex_unlock(&dentry->d_inode->i_mutex); |
1d2ef590 | 3527 | dput(dentry); |
912dbc15 | 3528 | if (!error) |
1da177e4 | 3529 | d_delete(dentry); |
1da177e4 LT |
3530 | return error; |
3531 | } | |
3532 | ||
5590ff0d | 3533 | static long do_rmdir(int dfd, const char __user *pathname) |
1da177e4 LT |
3534 | { |
3535 | int error = 0; | |
91a27b2a | 3536 | struct filename *name; |
1da177e4 LT |
3537 | struct dentry *dentry; |
3538 | struct nameidata nd; | |
c6ee9206 JL |
3539 | unsigned int lookup_flags = 0; |
3540 | retry: | |
3541 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3542 | if (IS_ERR(name)) |
3543 | return PTR_ERR(name); | |
1da177e4 LT |
3544 | |
3545 | switch(nd.last_type) { | |
0612d9fb OH |
3546 | case LAST_DOTDOT: |
3547 | error = -ENOTEMPTY; | |
3548 | goto exit1; | |
3549 | case LAST_DOT: | |
3550 | error = -EINVAL; | |
3551 | goto exit1; | |
3552 | case LAST_ROOT: | |
3553 | error = -EBUSY; | |
3554 | goto exit1; | |
1da177e4 | 3555 | } |
0612d9fb OH |
3556 | |
3557 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3558 | error = mnt_want_write(nd.path.mnt); |
3559 | if (error) | |
3560 | goto exit1; | |
0612d9fb | 3561 | |
4ac91378 | 3562 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3563 | dentry = lookup_hash(&nd); |
1da177e4 | 3564 | error = PTR_ERR(dentry); |
6902d925 DH |
3565 | if (IS_ERR(dentry)) |
3566 | goto exit2; | |
e6bc45d6 TT |
3567 | if (!dentry->d_inode) { |
3568 | error = -ENOENT; | |
3569 | goto exit3; | |
3570 | } | |
be6d3e56 KT |
3571 | error = security_path_rmdir(&nd.path, dentry); |
3572 | if (error) | |
c30dabfe | 3573 | goto exit3; |
4ac91378 | 3574 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
0622753b | 3575 | exit3: |
6902d925 DH |
3576 | dput(dentry); |
3577 | exit2: | |
4ac91378 | 3578 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
c30dabfe | 3579 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3580 | exit1: |
1d957f9b | 3581 | path_put(&nd.path); |
1da177e4 | 3582 | putname(name); |
c6ee9206 JL |
3583 | if (retry_estale(error, lookup_flags)) { |
3584 | lookup_flags |= LOOKUP_REVAL; | |
3585 | goto retry; | |
3586 | } | |
1da177e4 LT |
3587 | return error; |
3588 | } | |
3589 | ||
3cdad428 | 3590 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
5590ff0d UD |
3591 | { |
3592 | return do_rmdir(AT_FDCWD, pathname); | |
3593 | } | |
3594 | ||
b21996e3 BF |
3595 | /** |
3596 | * vfs_unlink - unlink a filesystem object | |
3597 | * @dir: parent directory | |
3598 | * @dentry: victim | |
3599 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3600 | * | |
3601 | * The caller must hold dir->i_mutex. | |
3602 | * | |
3603 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3604 | * return a reference to the inode in delegated_inode. The caller | |
3605 | * should then break the delegation on that inode and retry. Because | |
3606 | * breaking a delegation may take a long time, the caller should drop | |
3607 | * dir->i_mutex before doing so. | |
3608 | * | |
3609 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3610 | * be appropriate for callers that expect the underlying filesystem not | |
3611 | * to be NFS exported. | |
3612 | */ | |
3613 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
1da177e4 | 3614 | { |
9accbb97 | 3615 | struct inode *target = dentry->d_inode; |
1da177e4 LT |
3616 | int error = may_delete(dir, dentry, 0); |
3617 | ||
3618 | if (error) | |
3619 | return error; | |
3620 | ||
acfa4380 | 3621 | if (!dir->i_op->unlink) |
1da177e4 LT |
3622 | return -EPERM; |
3623 | ||
9accbb97 | 3624 | mutex_lock(&target->i_mutex); |
1da177e4 LT |
3625 | if (d_mountpoint(dentry)) |
3626 | error = -EBUSY; | |
3627 | else { | |
3628 | error = security_inode_unlink(dir, dentry); | |
bec1052e | 3629 | if (!error) { |
5a14696c BF |
3630 | error = try_break_deleg(target, delegated_inode); |
3631 | if (error) | |
b21996e3 | 3632 | goto out; |
1da177e4 | 3633 | error = dir->i_op->unlink(dir, dentry); |
bec1052e | 3634 | if (!error) |
d83c49f3 | 3635 | dont_mount(dentry); |
bec1052e | 3636 | } |
1da177e4 | 3637 | } |
b21996e3 | 3638 | out: |
9accbb97 | 3639 | mutex_unlock(&target->i_mutex); |
1da177e4 LT |
3640 | |
3641 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3642 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
9accbb97 | 3643 | fsnotify_link_count(target); |
e234f35c | 3644 | d_delete(dentry); |
1da177e4 | 3645 | } |
0eeca283 | 3646 | |
1da177e4 LT |
3647 | return error; |
3648 | } | |
3649 | ||
3650 | /* | |
3651 | * Make sure that the actual truncation of the file will occur outside its | |
1b1dcc1b | 3652 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
1da177e4 LT |
3653 | * writeout happening, and we don't want to prevent access to the directory |
3654 | * while waiting on the I/O. | |
3655 | */ | |
5590ff0d | 3656 | static long do_unlinkat(int dfd, const char __user *pathname) |
1da177e4 | 3657 | { |
2ad94ae6 | 3658 | int error; |
91a27b2a | 3659 | struct filename *name; |
1da177e4 LT |
3660 | struct dentry *dentry; |
3661 | struct nameidata nd; | |
3662 | struct inode *inode = NULL; | |
b21996e3 | 3663 | struct inode *delegated_inode = NULL; |
5d18f813 JL |
3664 | unsigned int lookup_flags = 0; |
3665 | retry: | |
3666 | name = user_path_parent(dfd, pathname, &nd, lookup_flags); | |
91a27b2a JL |
3667 | if (IS_ERR(name)) |
3668 | return PTR_ERR(name); | |
2ad94ae6 | 3669 | |
1da177e4 LT |
3670 | error = -EISDIR; |
3671 | if (nd.last_type != LAST_NORM) | |
3672 | goto exit1; | |
0612d9fb OH |
3673 | |
3674 | nd.flags &= ~LOOKUP_PARENT; | |
c30dabfe JK |
3675 | error = mnt_want_write(nd.path.mnt); |
3676 | if (error) | |
3677 | goto exit1; | |
b21996e3 | 3678 | retry_deleg: |
4ac91378 | 3679 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
49705b77 | 3680 | dentry = lookup_hash(&nd); |
1da177e4 LT |
3681 | error = PTR_ERR(dentry); |
3682 | if (!IS_ERR(dentry)) { | |
3683 | /* Why not before? Because we want correct error value */ | |
50338b88 TE |
3684 | if (nd.last.name[nd.last.len]) |
3685 | goto slashes; | |
1da177e4 | 3686 | inode = dentry->d_inode; |
b18825a7 | 3687 | if (d_is_negative(dentry)) |
e6bc45d6 TT |
3688 | goto slashes; |
3689 | ihold(inode); | |
be6d3e56 KT |
3690 | error = security_path_unlink(&nd.path, dentry); |
3691 | if (error) | |
c30dabfe | 3692 | goto exit2; |
b21996e3 | 3693 | error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); |
c30dabfe | 3694 | exit2: |
1da177e4 LT |
3695 | dput(dentry); |
3696 | } | |
4ac91378 | 3697 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
1da177e4 LT |
3698 | if (inode) |
3699 | iput(inode); /* truncate the inode here */ | |
b21996e3 BF |
3700 | inode = NULL; |
3701 | if (delegated_inode) { | |
5a14696c | 3702 | error = break_deleg_wait(&delegated_inode); |
b21996e3 BF |
3703 | if (!error) |
3704 | goto retry_deleg; | |
3705 | } | |
c30dabfe | 3706 | mnt_drop_write(nd.path.mnt); |
1da177e4 | 3707 | exit1: |
1d957f9b | 3708 | path_put(&nd.path); |
1da177e4 | 3709 | putname(name); |
5d18f813 JL |
3710 | if (retry_estale(error, lookup_flags)) { |
3711 | lookup_flags |= LOOKUP_REVAL; | |
3712 | inode = NULL; | |
3713 | goto retry; | |
3714 | } | |
1da177e4 LT |
3715 | return error; |
3716 | ||
3717 | slashes: | |
b18825a7 DH |
3718 | if (d_is_negative(dentry)) |
3719 | error = -ENOENT; | |
3720 | else if (d_is_directory(dentry) || d_is_autodir(dentry)) | |
3721 | error = -EISDIR; | |
3722 | else | |
3723 | error = -ENOTDIR; | |
1da177e4 LT |
3724 | goto exit2; |
3725 | } | |
3726 | ||
2e4d0924 | 3727 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
5590ff0d UD |
3728 | { |
3729 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3730 | return -EINVAL; | |
3731 | ||
3732 | if (flag & AT_REMOVEDIR) | |
3733 | return do_rmdir(dfd, pathname); | |
3734 | ||
3735 | return do_unlinkat(dfd, pathname); | |
3736 | } | |
3737 | ||
3480b257 | 3738 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
5590ff0d UD |
3739 | { |
3740 | return do_unlinkat(AT_FDCWD, pathname); | |
3741 | } | |
3742 | ||
db2e747b | 3743 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
1da177e4 | 3744 | { |
a95164d9 | 3745 | int error = may_create(dir, dentry); |
1da177e4 LT |
3746 | |
3747 | if (error) | |
3748 | return error; | |
3749 | ||
acfa4380 | 3750 | if (!dir->i_op->symlink) |
1da177e4 LT |
3751 | return -EPERM; |
3752 | ||
3753 | error = security_inode_symlink(dir, dentry, oldname); | |
3754 | if (error) | |
3755 | return error; | |
3756 | ||
1da177e4 | 3757 | error = dir->i_op->symlink(dir, dentry, oldname); |
a74574aa | 3758 | if (!error) |
f38aa942 | 3759 | fsnotify_create(dir, dentry); |
1da177e4 LT |
3760 | return error; |
3761 | } | |
3762 | ||
2e4d0924 HC |
3763 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3764 | int, newdfd, const char __user *, newname) | |
1da177e4 | 3765 | { |
2ad94ae6 | 3766 | int error; |
91a27b2a | 3767 | struct filename *from; |
6902d925 | 3768 | struct dentry *dentry; |
dae6ad8f | 3769 | struct path path; |
f46d3567 | 3770 | unsigned int lookup_flags = 0; |
1da177e4 LT |
3771 | |
3772 | from = getname(oldname); | |
2ad94ae6 | 3773 | if (IS_ERR(from)) |
1da177e4 | 3774 | return PTR_ERR(from); |
f46d3567 JL |
3775 | retry: |
3776 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
6902d925 DH |
3777 | error = PTR_ERR(dentry); |
3778 | if (IS_ERR(dentry)) | |
dae6ad8f | 3779 | goto out_putname; |
6902d925 | 3780 | |
91a27b2a | 3781 | error = security_path_symlink(&path, dentry, from->name); |
a8104a9f | 3782 | if (!error) |
91a27b2a | 3783 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); |
921a1650 | 3784 | done_path_create(&path, dentry); |
f46d3567 JL |
3785 | if (retry_estale(error, lookup_flags)) { |
3786 | lookup_flags |= LOOKUP_REVAL; | |
3787 | goto retry; | |
3788 | } | |
6902d925 | 3789 | out_putname: |
1da177e4 LT |
3790 | putname(from); |
3791 | return error; | |
3792 | } | |
3793 | ||
3480b257 | 3794 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
3795 | { |
3796 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
3797 | } | |
3798 | ||
146a8595 BF |
3799 | /** |
3800 | * vfs_link - create a new link | |
3801 | * @old_dentry: object to be linked | |
3802 | * @dir: new parent | |
3803 | * @new_dentry: where to create the new link | |
3804 | * @delegated_inode: returns inode needing a delegation break | |
3805 | * | |
3806 | * The caller must hold dir->i_mutex | |
3807 | * | |
3808 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
3809 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
3810 | * inode in delegated_inode. The caller should then break the delegation | |
3811 | * and retry. Because breaking a delegation may take a long time, the | |
3812 | * caller should drop the i_mutex before doing so. | |
3813 | * | |
3814 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3815 | * be appropriate for callers that expect the underlying filesystem not | |
3816 | * to be NFS exported. | |
3817 | */ | |
3818 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
1da177e4 LT |
3819 | { |
3820 | struct inode *inode = old_dentry->d_inode; | |
8de52778 | 3821 | unsigned max_links = dir->i_sb->s_max_links; |
1da177e4 LT |
3822 | int error; |
3823 | ||
3824 | if (!inode) | |
3825 | return -ENOENT; | |
3826 | ||
a95164d9 | 3827 | error = may_create(dir, new_dentry); |
1da177e4 LT |
3828 | if (error) |
3829 | return error; | |
3830 | ||
3831 | if (dir->i_sb != inode->i_sb) | |
3832 | return -EXDEV; | |
3833 | ||
3834 | /* | |
3835 | * A link to an append-only or immutable file cannot be created. | |
3836 | */ | |
3837 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3838 | return -EPERM; | |
acfa4380 | 3839 | if (!dir->i_op->link) |
1da177e4 | 3840 | return -EPERM; |
7e79eedb | 3841 | if (S_ISDIR(inode->i_mode)) |
1da177e4 LT |
3842 | return -EPERM; |
3843 | ||
3844 | error = security_inode_link(old_dentry, dir, new_dentry); | |
3845 | if (error) | |
3846 | return error; | |
3847 | ||
7e79eedb | 3848 | mutex_lock(&inode->i_mutex); |
aae8a97d | 3849 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
f4e0c30c | 3850 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) |
aae8a97d | 3851 | error = -ENOENT; |
8de52778 AV |
3852 | else if (max_links && inode->i_nlink >= max_links) |
3853 | error = -EMLINK; | |
146a8595 BF |
3854 | else { |
3855 | error = try_break_deleg(inode, delegated_inode); | |
3856 | if (!error) | |
3857 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
3858 | } | |
f4e0c30c AV |
3859 | |
3860 | if (!error && (inode->i_state & I_LINKABLE)) { | |
3861 | spin_lock(&inode->i_lock); | |
3862 | inode->i_state &= ~I_LINKABLE; | |
3863 | spin_unlock(&inode->i_lock); | |
3864 | } | |
7e79eedb | 3865 | mutex_unlock(&inode->i_mutex); |
e31e14ec | 3866 | if (!error) |
7e79eedb | 3867 | fsnotify_link(dir, inode, new_dentry); |
1da177e4 LT |
3868 | return error; |
3869 | } | |
3870 | ||
3871 | /* | |
3872 | * Hardlinks are often used in delicate situations. We avoid | |
3873 | * security-related surprises by not following symlinks on the | |
3874 | * newname. --KAB | |
3875 | * | |
3876 | * We don't follow them on the oldname either to be compatible | |
3877 | * with linux 2.0, and to avoid hard-linking to directories | |
3878 | * and other special files. --ADM | |
3879 | */ | |
2e4d0924 HC |
3880 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3881 | int, newdfd, const char __user *, newname, int, flags) | |
1da177e4 LT |
3882 | { |
3883 | struct dentry *new_dentry; | |
dae6ad8f | 3884 | struct path old_path, new_path; |
146a8595 | 3885 | struct inode *delegated_inode = NULL; |
11a7b371 | 3886 | int how = 0; |
1da177e4 | 3887 | int error; |
1da177e4 | 3888 | |
11a7b371 | 3889 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
c04030e1 | 3890 | return -EINVAL; |
11a7b371 | 3891 | /* |
f0cc6ffb LT |
3892 | * To use null names we require CAP_DAC_READ_SEARCH |
3893 | * This ensures that not everyone will be able to create | |
3894 | * handlink using the passed filedescriptor. | |
11a7b371 | 3895 | */ |
f0cc6ffb LT |
3896 | if (flags & AT_EMPTY_PATH) { |
3897 | if (!capable(CAP_DAC_READ_SEARCH)) | |
3898 | return -ENOENT; | |
11a7b371 | 3899 | how = LOOKUP_EMPTY; |
f0cc6ffb | 3900 | } |
11a7b371 AK |
3901 | |
3902 | if (flags & AT_SYMLINK_FOLLOW) | |
3903 | how |= LOOKUP_FOLLOW; | |
442e31ca | 3904 | retry: |
11a7b371 | 3905 | error = user_path_at(olddfd, oldname, how, &old_path); |
1da177e4 | 3906 | if (error) |
2ad94ae6 AV |
3907 | return error; |
3908 | ||
442e31ca JL |
3909 | new_dentry = user_path_create(newdfd, newname, &new_path, |
3910 | (how & LOOKUP_REVAL)); | |
1da177e4 | 3911 | error = PTR_ERR(new_dentry); |
6902d925 | 3912 | if (IS_ERR(new_dentry)) |
dae6ad8f AV |
3913 | goto out; |
3914 | ||
3915 | error = -EXDEV; | |
3916 | if (old_path.mnt != new_path.mnt) | |
3917 | goto out_dput; | |
800179c9 KC |
3918 | error = may_linkat(&old_path); |
3919 | if (unlikely(error)) | |
3920 | goto out_dput; | |
dae6ad8f | 3921 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
be6d3e56 | 3922 | if (error) |
a8104a9f | 3923 | goto out_dput; |
146a8595 | 3924 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); |
75c3f29d | 3925 | out_dput: |
921a1650 | 3926 | done_path_create(&new_path, new_dentry); |
146a8595 BF |
3927 | if (delegated_inode) { |
3928 | error = break_deleg_wait(&delegated_inode); | |
d22e6338 OD |
3929 | if (!error) { |
3930 | path_put(&old_path); | |
146a8595 | 3931 | goto retry; |
d22e6338 | 3932 | } |
146a8595 | 3933 | } |
442e31ca | 3934 | if (retry_estale(error, how)) { |
d22e6338 | 3935 | path_put(&old_path); |
442e31ca JL |
3936 | how |= LOOKUP_REVAL; |
3937 | goto retry; | |
3938 | } | |
1da177e4 | 3939 | out: |
2d8f3038 | 3940 | path_put(&old_path); |
1da177e4 LT |
3941 | |
3942 | return error; | |
3943 | } | |
3944 | ||
3480b257 | 3945 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
5590ff0d | 3946 | { |
c04030e1 | 3947 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
5590ff0d UD |
3948 | } |
3949 | ||
1da177e4 LT |
3950 | /* |
3951 | * The worst of all namespace operations - renaming directory. "Perverted" | |
3952 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
3953 | * Problems: | |
3954 | * a) we can get into loop creation. Check is done in is_subdir(). | |
3955 | * b) race potential - two innocent renames can create a loop together. | |
3956 | * That's where 4.4 screws up. Current fix: serialization on | |
a11f3a05 | 3957 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
1da177e4 | 3958 | * story. |
6cedba89 BF |
3959 | * c) we have to lock _four_ objects - parents and victim (if it exists), |
3960 | * and source (if it is not a directory). | |
1b1dcc1b | 3961 | * And that - after we got ->i_mutex on parents (until then we don't know |
1da177e4 LT |
3962 | * whether the target exists). Solution: try to be smart with locking |
3963 | * order for inodes. We rely on the fact that tree topology may change | |
a11f3a05 | 3964 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
1da177e4 LT |
3965 | * move will be locked. Thus we can rank directories by the tree |
3966 | * (ancestors first) and rank all non-directories after them. | |
3967 | * That works since everybody except rename does "lock parent, lookup, | |
a11f3a05 | 3968 | * lock child" and rename is under ->s_vfs_rename_mutex. |
1da177e4 LT |
3969 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3970 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
3971 | * we'd better make sure that there's no link(2) for them. | |
e4eaac06 | 3972 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
1b1dcc1b | 3973 | * we are removing the target. Solution: we will have to grab ->i_mutex |
1da177e4 | 3974 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
c41b20e7 | 3975 | * ->i_mutex on parents, which works but leads to some truly excessive |
1da177e4 LT |
3976 | * locking]. |
3977 | */ | |
75c96f85 AB |
3978 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3979 | struct inode *new_dir, struct dentry *new_dentry) | |
1da177e4 LT |
3980 | { |
3981 | int error = 0; | |
9055cba7 | 3982 | struct inode *target = new_dentry->d_inode; |
8de52778 | 3983 | unsigned max_links = new_dir->i_sb->s_max_links; |
1da177e4 LT |
3984 | |
3985 | /* | |
3986 | * If we are going to change the parent - check write permissions, | |
3987 | * we'll need to flip '..'. | |
3988 | */ | |
3989 | if (new_dir != old_dir) { | |
f419a2e3 | 3990 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
1da177e4 LT |
3991 | if (error) |
3992 | return error; | |
3993 | } | |
3994 | ||
3995 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
3996 | if (error) | |
3997 | return error; | |
3998 | ||
1d2ef590 | 3999 | dget(new_dentry); |
d83c49f3 | 4000 | if (target) |
1b1dcc1b | 4001 | mutex_lock(&target->i_mutex); |
9055cba7 SW |
4002 | |
4003 | error = -EBUSY; | |
4004 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) | |
4005 | goto out; | |
4006 | ||
8de52778 AV |
4007 | error = -EMLINK; |
4008 | if (max_links && !target && new_dir != old_dir && | |
4009 | new_dir->i_nlink >= max_links) | |
4010 | goto out; | |
4011 | ||
3cebde24 SW |
4012 | if (target) |
4013 | shrink_dcache_parent(new_dentry); | |
9055cba7 SW |
4014 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4015 | if (error) | |
4016 | goto out; | |
4017 | ||
1da177e4 | 4018 | if (target) { |
9055cba7 SW |
4019 | target->i_flags |= S_DEAD; |
4020 | dont_mount(new_dentry); | |
1da177e4 | 4021 | } |
9055cba7 SW |
4022 | out: |
4023 | if (target) | |
4024 | mutex_unlock(&target->i_mutex); | |
1d2ef590 | 4025 | dput(new_dentry); |
e31e14ec | 4026 | if (!error) |
349457cc MF |
4027 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
4028 | d_move(old_dentry,new_dentry); | |
1da177e4 LT |
4029 | return error; |
4030 | } | |
4031 | ||
75c96f85 | 4032 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4033 | struct inode *new_dir, struct dentry *new_dentry, |
4034 | struct inode **delegated_inode) | |
1da177e4 | 4035 | { |
51892bbb | 4036 | struct inode *target = new_dentry->d_inode; |
6cedba89 | 4037 | struct inode *source = old_dentry->d_inode; |
1da177e4 LT |
4038 | int error; |
4039 | ||
4040 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
4041 | if (error) | |
4042 | return error; | |
4043 | ||
4044 | dget(new_dentry); | |
6cedba89 | 4045 | lock_two_nondirectories(source, target); |
51892bbb SW |
4046 | |
4047 | error = -EBUSY; | |
1da177e4 | 4048 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
51892bbb SW |
4049 | goto out; |
4050 | ||
8e6d782c BF |
4051 | error = try_break_deleg(source, delegated_inode); |
4052 | if (error) | |
4053 | goto out; | |
4054 | if (target) { | |
4055 | error = try_break_deleg(target, delegated_inode); | |
4056 | if (error) | |
4057 | goto out; | |
4058 | } | |
51892bbb SW |
4059 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
4060 | if (error) | |
4061 | goto out; | |
4062 | ||
4063 | if (target) | |
4064 | dont_mount(new_dentry); | |
4065 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
4066 | d_move(old_dentry, new_dentry); | |
4067 | out: | |
6cedba89 | 4068 | unlock_two_nondirectories(source, target); |
1da177e4 LT |
4069 | dput(new_dentry); |
4070 | return error; | |
4071 | } | |
4072 | ||
8e6d782c BF |
4073 | /** |
4074 | * vfs_rename - rename a filesystem object | |
4075 | * @old_dir: parent of source | |
4076 | * @old_dentry: source | |
4077 | * @new_dir: parent of destination | |
4078 | * @new_dentry: destination | |
4079 | * @delegated_inode: returns an inode needing a delegation break | |
4080 | * | |
4081 | * The caller must hold multiple mutexes--see lock_rename()). | |
4082 | * | |
4083 | * If vfs_rename discovers a delegation in need of breaking at either | |
4084 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4085 | * reference to the inode in delegated_inode. The caller should then | |
4086 | * break the delegation and retry. Because breaking a delegation may | |
4087 | * take a long time, the caller should drop all locks before doing | |
4088 | * so. | |
4089 | * | |
4090 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4091 | * be appropriate for callers that expect the underlying filesystem not | |
4092 | * to be NFS exported. | |
4093 | */ | |
1da177e4 | 4094 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8e6d782c BF |
4095 | struct inode *new_dir, struct dentry *new_dentry, |
4096 | struct inode **delegated_inode) | |
1da177e4 LT |
4097 | { |
4098 | int error; | |
b18825a7 | 4099 | int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry); |
59b0df21 | 4100 | const unsigned char *old_name; |
1da177e4 LT |
4101 | |
4102 | if (old_dentry->d_inode == new_dentry->d_inode) | |
4103 | return 0; | |
4104 | ||
4105 | error = may_delete(old_dir, old_dentry, is_dir); | |
4106 | if (error) | |
4107 | return error; | |
4108 | ||
4109 | if (!new_dentry->d_inode) | |
a95164d9 | 4110 | error = may_create(new_dir, new_dentry); |
1da177e4 LT |
4111 | else |
4112 | error = may_delete(new_dir, new_dentry, is_dir); | |
4113 | if (error) | |
4114 | return error; | |
4115 | ||
acfa4380 | 4116 | if (!old_dir->i_op->rename) |
1da177e4 LT |
4117 | return -EPERM; |
4118 | ||
0eeca283 RL |
4119 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
4120 | ||
1da177e4 LT |
4121 | if (is_dir) |
4122 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
4123 | else | |
8e6d782c | 4124 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode); |
123df294 AV |
4125 | if (!error) |
4126 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
5a190ae6 | 4127 | new_dentry->d_inode, old_dentry); |
0eeca283 RL |
4128 | fsnotify_oldname_free(old_name); |
4129 | ||
1da177e4 LT |
4130 | return error; |
4131 | } | |
4132 | ||
2e4d0924 HC |
4133 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
4134 | int, newdfd, const char __user *, newname) | |
1da177e4 | 4135 | { |
2ad94ae6 AV |
4136 | struct dentry *old_dir, *new_dir; |
4137 | struct dentry *old_dentry, *new_dentry; | |
4138 | struct dentry *trap; | |
1da177e4 | 4139 | struct nameidata oldnd, newnd; |
8e6d782c | 4140 | struct inode *delegated_inode = NULL; |
91a27b2a JL |
4141 | struct filename *from; |
4142 | struct filename *to; | |
c6a94284 JL |
4143 | unsigned int lookup_flags = 0; |
4144 | bool should_retry = false; | |
2ad94ae6 | 4145 | int error; |
c6a94284 JL |
4146 | retry: |
4147 | from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); | |
91a27b2a JL |
4148 | if (IS_ERR(from)) { |
4149 | error = PTR_ERR(from); | |
1da177e4 | 4150 | goto exit; |
91a27b2a | 4151 | } |
1da177e4 | 4152 | |
c6a94284 | 4153 | to = user_path_parent(newdfd, newname, &newnd, lookup_flags); |
91a27b2a JL |
4154 | if (IS_ERR(to)) { |
4155 | error = PTR_ERR(to); | |
1da177e4 | 4156 | goto exit1; |
91a27b2a | 4157 | } |
1da177e4 LT |
4158 | |
4159 | error = -EXDEV; | |
4ac91378 | 4160 | if (oldnd.path.mnt != newnd.path.mnt) |
1da177e4 LT |
4161 | goto exit2; |
4162 | ||
4ac91378 | 4163 | old_dir = oldnd.path.dentry; |
1da177e4 LT |
4164 | error = -EBUSY; |
4165 | if (oldnd.last_type != LAST_NORM) | |
4166 | goto exit2; | |
4167 | ||
4ac91378 | 4168 | new_dir = newnd.path.dentry; |
1da177e4 LT |
4169 | if (newnd.last_type != LAST_NORM) |
4170 | goto exit2; | |
4171 | ||
c30dabfe JK |
4172 | error = mnt_want_write(oldnd.path.mnt); |
4173 | if (error) | |
4174 | goto exit2; | |
4175 | ||
0612d9fb OH |
4176 | oldnd.flags &= ~LOOKUP_PARENT; |
4177 | newnd.flags &= ~LOOKUP_PARENT; | |
4e9ed2f8 | 4178 | newnd.flags |= LOOKUP_RENAME_TARGET; |
0612d9fb | 4179 | |
8e6d782c | 4180 | retry_deleg: |
1da177e4 LT |
4181 | trap = lock_rename(new_dir, old_dir); |
4182 | ||
49705b77 | 4183 | old_dentry = lookup_hash(&oldnd); |
1da177e4 LT |
4184 | error = PTR_ERR(old_dentry); |
4185 | if (IS_ERR(old_dentry)) | |
4186 | goto exit3; | |
4187 | /* source must exist */ | |
4188 | error = -ENOENT; | |
b18825a7 | 4189 | if (d_is_negative(old_dentry)) |
1da177e4 LT |
4190 | goto exit4; |
4191 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
b18825a7 | 4192 | if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) { |
1da177e4 LT |
4193 | error = -ENOTDIR; |
4194 | if (oldnd.last.name[oldnd.last.len]) | |
4195 | goto exit4; | |
4196 | if (newnd.last.name[newnd.last.len]) | |
4197 | goto exit4; | |
4198 | } | |
4199 | /* source should not be ancestor of target */ | |
4200 | error = -EINVAL; | |
4201 | if (old_dentry == trap) | |
4202 | goto exit4; | |
49705b77 | 4203 | new_dentry = lookup_hash(&newnd); |
1da177e4 LT |
4204 | error = PTR_ERR(new_dentry); |
4205 | if (IS_ERR(new_dentry)) | |
4206 | goto exit4; | |
4207 | /* target should not be an ancestor of source */ | |
4208 | error = -ENOTEMPTY; | |
4209 | if (new_dentry == trap) | |
4210 | goto exit5; | |
4211 | ||
be6d3e56 KT |
4212 | error = security_path_rename(&oldnd.path, old_dentry, |
4213 | &newnd.path, new_dentry); | |
4214 | if (error) | |
c30dabfe | 4215 | goto exit5; |
1da177e4 | 4216 | error = vfs_rename(old_dir->d_inode, old_dentry, |
8e6d782c BF |
4217 | new_dir->d_inode, new_dentry, |
4218 | &delegated_inode); | |
1da177e4 LT |
4219 | exit5: |
4220 | dput(new_dentry); | |
4221 | exit4: | |
4222 | dput(old_dentry); | |
4223 | exit3: | |
4224 | unlock_rename(new_dir, old_dir); | |
8e6d782c BF |
4225 | if (delegated_inode) { |
4226 | error = break_deleg_wait(&delegated_inode); | |
4227 | if (!error) | |
4228 | goto retry_deleg; | |
4229 | } | |
c30dabfe | 4230 | mnt_drop_write(oldnd.path.mnt); |
1da177e4 | 4231 | exit2: |
c6a94284 JL |
4232 | if (retry_estale(error, lookup_flags)) |
4233 | should_retry = true; | |
1d957f9b | 4234 | path_put(&newnd.path); |
2ad94ae6 | 4235 | putname(to); |
1da177e4 | 4236 | exit1: |
1d957f9b | 4237 | path_put(&oldnd.path); |
1da177e4 | 4238 | putname(from); |
c6a94284 JL |
4239 | if (should_retry) { |
4240 | should_retry = false; | |
4241 | lookup_flags |= LOOKUP_REVAL; | |
4242 | goto retry; | |
4243 | } | |
2ad94ae6 | 4244 | exit: |
1da177e4 LT |
4245 | return error; |
4246 | } | |
4247 | ||
a26eab24 | 4248 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
5590ff0d UD |
4249 | { |
4250 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
4251 | } | |
4252 | ||
1da177e4 LT |
4253 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
4254 | { | |
4255 | int len; | |
4256 | ||
4257 | len = PTR_ERR(link); | |
4258 | if (IS_ERR(link)) | |
4259 | goto out; | |
4260 | ||
4261 | len = strlen(link); | |
4262 | if (len > (unsigned) buflen) | |
4263 | len = buflen; | |
4264 | if (copy_to_user(buffer, link, len)) | |
4265 | len = -EFAULT; | |
4266 | out: | |
4267 | return len; | |
4268 | } | |
4269 | ||
4270 | /* | |
4271 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4272 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
4273 | * using) it for any given inode is up to filesystem. | |
4274 | */ | |
4275 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4276 | { | |
4277 | struct nameidata nd; | |
cc314eef | 4278 | void *cookie; |
694a1764 | 4279 | int res; |
cc314eef | 4280 | |
1da177e4 | 4281 | nd.depth = 0; |
cc314eef | 4282 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
694a1764 MS |
4283 | if (IS_ERR(cookie)) |
4284 | return PTR_ERR(cookie); | |
4285 | ||
4286 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
4287 | if (dentry->d_inode->i_op->put_link) | |
4288 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
4289 | return res; | |
1da177e4 LT |
4290 | } |
4291 | ||
1da177e4 LT |
4292 | /* get the link contents into pagecache */ |
4293 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
4294 | { | |
ebd09abb DG |
4295 | char *kaddr; |
4296 | struct page *page; | |
1da177e4 | 4297 | struct address_space *mapping = dentry->d_inode->i_mapping; |
090d2b18 | 4298 | page = read_mapping_page(mapping, 0, NULL); |
1da177e4 | 4299 | if (IS_ERR(page)) |
6fe6900e | 4300 | return (char*)page; |
1da177e4 | 4301 | *ppage = page; |
ebd09abb DG |
4302 | kaddr = kmap(page); |
4303 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
4304 | return kaddr; | |
1da177e4 LT |
4305 | } |
4306 | ||
4307 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4308 | { | |
4309 | struct page *page = NULL; | |
4310 | char *s = page_getlink(dentry, &page); | |
4311 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
4312 | if (page) { | |
4313 | kunmap(page); | |
4314 | page_cache_release(page); | |
4315 | } | |
4316 | return res; | |
4317 | } | |
4318 | ||
cc314eef | 4319 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
1da177e4 | 4320 | { |
cc314eef | 4321 | struct page *page = NULL; |
1da177e4 | 4322 | nd_set_link(nd, page_getlink(dentry, &page)); |
cc314eef | 4323 | return page; |
1da177e4 LT |
4324 | } |
4325 | ||
cc314eef | 4326 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
1da177e4 | 4327 | { |
cc314eef LT |
4328 | struct page *page = cookie; |
4329 | ||
4330 | if (page) { | |
1da177e4 LT |
4331 | kunmap(page); |
4332 | page_cache_release(page); | |
1da177e4 LT |
4333 | } |
4334 | } | |
4335 | ||
54566b2c NP |
4336 | /* |
4337 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4338 | */ | |
4339 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
1da177e4 LT |
4340 | { |
4341 | struct address_space *mapping = inode->i_mapping; | |
0adb25d2 | 4342 | struct page *page; |
afddba49 | 4343 | void *fsdata; |
beb497ab | 4344 | int err; |
1da177e4 | 4345 | char *kaddr; |
54566b2c NP |
4346 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
4347 | if (nofs) | |
4348 | flags |= AOP_FLAG_NOFS; | |
1da177e4 | 4349 | |
7e53cac4 | 4350 | retry: |
afddba49 | 4351 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
54566b2c | 4352 | flags, &page, &fsdata); |
1da177e4 | 4353 | if (err) |
afddba49 NP |
4354 | goto fail; |
4355 | ||
e8e3c3d6 | 4356 | kaddr = kmap_atomic(page); |
1da177e4 | 4357 | memcpy(kaddr, symname, len-1); |
e8e3c3d6 | 4358 | kunmap_atomic(kaddr); |
afddba49 NP |
4359 | |
4360 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4361 | page, fsdata); | |
1da177e4 LT |
4362 | if (err < 0) |
4363 | goto fail; | |
afddba49 NP |
4364 | if (err < len-1) |
4365 | goto retry; | |
4366 | ||
1da177e4 LT |
4367 | mark_inode_dirty(inode); |
4368 | return 0; | |
1da177e4 LT |
4369 | fail: |
4370 | return err; | |
4371 | } | |
4372 | ||
0adb25d2 KK |
4373 | int page_symlink(struct inode *inode, const char *symname, int len) |
4374 | { | |
4375 | return __page_symlink(inode, symname, len, | |
54566b2c | 4376 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
0adb25d2 KK |
4377 | } |
4378 | ||
92e1d5be | 4379 | const struct inode_operations page_symlink_inode_operations = { |
1da177e4 LT |
4380 | .readlink = generic_readlink, |
4381 | .follow_link = page_follow_link_light, | |
4382 | .put_link = page_put_link, | |
4383 | }; | |
4384 | ||
2d8f3038 | 4385 | EXPORT_SYMBOL(user_path_at); |
cc53ce53 | 4386 | EXPORT_SYMBOL(follow_down_one); |
1da177e4 LT |
4387 | EXPORT_SYMBOL(follow_down); |
4388 | EXPORT_SYMBOL(follow_up); | |
f6d2ac5c | 4389 | EXPORT_SYMBOL(get_write_access); /* nfsd */ |
1da177e4 | 4390 | EXPORT_SYMBOL(lock_rename); |
1da177e4 LT |
4391 | EXPORT_SYMBOL(lookup_one_len); |
4392 | EXPORT_SYMBOL(page_follow_link_light); | |
4393 | EXPORT_SYMBOL(page_put_link); | |
4394 | EXPORT_SYMBOL(page_readlink); | |
0adb25d2 | 4395 | EXPORT_SYMBOL(__page_symlink); |
1da177e4 LT |
4396 | EXPORT_SYMBOL(page_symlink); |
4397 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
d1811465 | 4398 | EXPORT_SYMBOL(kern_path); |
16f18200 | 4399 | EXPORT_SYMBOL(vfs_path_lookup); |
f419a2e3 | 4400 | EXPORT_SYMBOL(inode_permission); |
1da177e4 LT |
4401 | EXPORT_SYMBOL(unlock_rename); |
4402 | EXPORT_SYMBOL(vfs_create); | |
1da177e4 LT |
4403 | EXPORT_SYMBOL(vfs_link); |
4404 | EXPORT_SYMBOL(vfs_mkdir); | |
4405 | EXPORT_SYMBOL(vfs_mknod); | |
4406 | EXPORT_SYMBOL(generic_permission); | |
4407 | EXPORT_SYMBOL(vfs_readlink); | |
4408 | EXPORT_SYMBOL(vfs_rename); | |
4409 | EXPORT_SYMBOL(vfs_rmdir); | |
4410 | EXPORT_SYMBOL(vfs_symlink); | |
4411 | EXPORT_SYMBOL(vfs_unlink); | |
4412 | EXPORT_SYMBOL(dentry_unhash); | |
4413 | EXPORT_SYMBOL(generic_readlink); |