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