Root/
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/module.h> |
19 | #include <linux/slab.h> |
20 | #include <linux/fs.h> |
21 | #include <linux/namei.h> |
22 | #include <linux/pagemap.h> |
23 | #include <linux/fsnotify.h> |
24 | #include <linux/personality.h> |
25 | #include <linux/security.h> |
26 | #include <linux/ima.h> |
27 | #include <linux/syscalls.h> |
28 | #include <linux/mount.h> |
29 | #include <linux/audit.h> |
30 | #include <linux/capability.h> |
31 | #include <linux/file.h> |
32 | #include <linux/fcntl.h> |
33 | #include <linux/device_cgroup.h> |
34 | #include <linux/fs_struct.h> |
35 | #include <linux/posix_acl.h> |
36 | #include <asm/uaccess.h> |
37 | |
38 | #include "internal.h" |
39 | #include "mount.h" |
40 | |
41 | /* [Feb-1997 T. Schoebel-Theuer] |
42 | * Fundamental changes in the pathname lookup mechanisms (namei) |
43 | * were necessary because of omirr. The reason is that omirr needs |
44 | * to know the _real_ pathname, not the user-supplied one, in case |
45 | * of symlinks (and also when transname replacements occur). |
46 | * |
47 | * The new code replaces the old recursive symlink resolution with |
48 | * an iterative one (in case of non-nested symlink chains). It does |
49 | * this with calls to <fs>_follow_link(). |
50 | * As a side effect, dir_namei(), _namei() and follow_link() are now |
51 | * replaced with a single function lookup_dentry() that can handle all |
52 | * the special cases of the former code. |
53 | * |
54 | * With the new dcache, the pathname is stored at each inode, at least as |
55 | * long as the refcount of the inode is positive. As a side effect, the |
56 | * size of the dcache depends on the inode cache and thus is dynamic. |
57 | * |
58 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink |
59 | * resolution to correspond with current state of the code. |
60 | * |
61 | * Note that the symlink resolution is not *completely* iterative. |
62 | * There is still a significant amount of tail- and mid- recursion in |
63 | * the algorithm. Also, note that <fs>_readlink() is not used in |
64 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() |
65 | * may return different results than <fs>_follow_link(). Many virtual |
66 | * filesystems (including /proc) exhibit this behavior. |
67 | */ |
68 | |
69 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: |
70 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL |
71 | * and the name already exists in form of a symlink, try to create the new |
72 | * name indicated by the symlink. The old code always complained that the |
73 | * name already exists, due to not following the symlink even if its target |
74 | * is nonexistent. The new semantics affects also mknod() and link() when |
75 | * the name is a symlink pointing to a non-existent name. |
76 | * |
77 | * I don't know which semantics is the right one, since I have no access |
78 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" |
79 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the |
80 | * "old" one. Personally, I think the new semantics is much more logical. |
81 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing |
82 | * file does succeed in both HP-UX and SunOs, but not in Solaris |
83 | * and in the old Linux semantics. |
84 | */ |
85 | |
86 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink |
87 | * semantics. See the comments in "open_namei" and "do_link" below. |
88 | * |
89 | * [10-Sep-98 Alan Modra] Another symlink change. |
90 | */ |
91 | |
92 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: |
93 | * inside the path - always follow. |
94 | * in the last component in creation/removal/renaming - never follow. |
95 | * if LOOKUP_FOLLOW passed - follow. |
96 | * if the pathname has trailing slashes - follow. |
97 | * otherwise - don't follow. |
98 | * (applied in that order). |
99 | * |
100 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT |
101 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. |
102 | * During the 2.4 we need to fix the userland stuff depending on it - |
103 | * hopefully we will be able to get rid of that wart in 2.5. So far only |
104 | * XEmacs seems to be relying on it... |
105 | */ |
106 | /* |
107 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) |
108 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
109 | * any extra contention... |
110 | */ |
111 | |
112 | /* In order to reduce some races, while at the same time doing additional |
113 | * checking and hopefully speeding things up, we copy filenames to the |
114 | * kernel data space before using them.. |
115 | * |
116 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). |
117 | * PATH_MAX includes the nul terminator --RR. |
118 | */ |
119 | static int do_getname(const char __user *filename, char *page) |
120 | { |
121 | int retval; |
122 | unsigned long len = PATH_MAX; |
123 | |
124 | if (!segment_eq(get_fs(), KERNEL_DS)) { |
125 | if ((unsigned long) filename >= TASK_SIZE) |
126 | return -EFAULT; |
127 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) |
128 | len = TASK_SIZE - (unsigned long) filename; |
129 | } |
130 | |
131 | retval = strncpy_from_user(page, filename, len); |
132 | if (retval > 0) { |
133 | if (retval < len) |
134 | return 0; |
135 | return -ENAMETOOLONG; |
136 | } else if (!retval) |
137 | retval = -ENOENT; |
138 | return retval; |
139 | } |
140 | |
141 | static char *getname_flags(const char __user *filename, int flags, int *empty) |
142 | { |
143 | char *result = __getname(); |
144 | int retval; |
145 | |
146 | if (!result) |
147 | return ERR_PTR(-ENOMEM); |
148 | |
149 | retval = do_getname(filename, result); |
150 | if (retval < 0) { |
151 | if (retval == -ENOENT && empty) |
152 | *empty = 1; |
153 | if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { |
154 | __putname(result); |
155 | return ERR_PTR(retval); |
156 | } |
157 | } |
158 | audit_getname(result); |
159 | return result; |
160 | } |
161 | |
162 | char *getname(const char __user * filename) |
163 | { |
164 | return getname_flags(filename, 0, 0); |
165 | } |
166 | |
167 | #ifdef CONFIG_AUDITSYSCALL |
168 | void putname(const char *name) |
169 | { |
170 | if (unlikely(!audit_dummy_context())) |
171 | audit_putname(name); |
172 | else |
173 | __putname(name); |
174 | } |
175 | EXPORT_SYMBOL(putname); |
176 | #endif |
177 | |
178 | static int check_acl(struct inode *inode, int mask) |
179 | { |
180 | #ifdef CONFIG_FS_POSIX_ACL |
181 | struct posix_acl *acl; |
182 | |
183 | if (mask & MAY_NOT_BLOCK) { |
184 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); |
185 | if (!acl) |
186 | return -EAGAIN; |
187 | /* no ->get_acl() calls in RCU mode... */ |
188 | if (acl == ACL_NOT_CACHED) |
189 | return -ECHILD; |
190 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); |
191 | } |
192 | |
193 | acl = get_cached_acl(inode, ACL_TYPE_ACCESS); |
194 | |
195 | /* |
196 | * A filesystem can force a ACL callback by just never filling the |
197 | * ACL cache. But normally you'd fill the cache either at inode |
198 | * instantiation time, or on the first ->get_acl call. |
199 | * |
200 | * If the filesystem doesn't have a get_acl() function at all, we'll |
201 | * just create the negative cache entry. |
202 | */ |
203 | if (acl == ACL_NOT_CACHED) { |
204 | if (inode->i_op->get_acl) { |
205 | acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); |
206 | if (IS_ERR(acl)) |
207 | return PTR_ERR(acl); |
208 | } else { |
209 | set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); |
210 | return -EAGAIN; |
211 | } |
212 | } |
213 | |
214 | if (acl) { |
215 | int error = posix_acl_permission(inode, acl, mask); |
216 | posix_acl_release(acl); |
217 | return error; |
218 | } |
219 | #endif |
220 | |
221 | return -EAGAIN; |
222 | } |
223 | |
224 | /* |
225 | * This does the basic permission checking |
226 | */ |
227 | static int acl_permission_check(struct inode *inode, int mask) |
228 | { |
229 | unsigned int mode = inode->i_mode; |
230 | |
231 | if (current_user_ns() != inode_userns(inode)) |
232 | goto other_perms; |
233 | |
234 | if (likely(current_fsuid() == inode->i_uid)) |
235 | mode >>= 6; |
236 | else { |
237 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { |
238 | int error = check_acl(inode, mask); |
239 | if (error != -EAGAIN) |
240 | return error; |
241 | } |
242 | |
243 | if (in_group_p(inode->i_gid)) |
244 | mode >>= 3; |
245 | } |
246 | |
247 | other_perms: |
248 | /* |
249 | * If the DACs are ok we don't need any capability check. |
250 | */ |
251 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
252 | return 0; |
253 | return -EACCES; |
254 | } |
255 | |
256 | /** |
257 | * generic_permission - check for access rights on a Posix-like filesystem |
258 | * @inode: inode to check access rights for |
259 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
260 | * |
261 | * Used to check for read/write/execute permissions on a file. |
262 | * We use "fsuid" for this, letting us set arbitrary permissions |
263 | * for filesystem access without changing the "normal" uids which |
264 | * are used for other things. |
265 | * |
266 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk |
267 | * request cannot be satisfied (eg. requires blocking or too much complexity). |
268 | * It would then be called again in ref-walk mode. |
269 | */ |
270 | int generic_permission(struct inode *inode, int mask) |
271 | { |
272 | int ret; |
273 | |
274 | /* |
275 | * Do the basic permission checks. |
276 | */ |
277 | ret = acl_permission_check(inode, mask); |
278 | if (ret != -EACCES) |
279 | return ret; |
280 | |
281 | if (S_ISDIR(inode->i_mode)) { |
282 | /* DACs are overridable for directories */ |
283 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) |
284 | return 0; |
285 | if (!(mask & MAY_WRITE)) |
286 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) |
287 | return 0; |
288 | return -EACCES; |
289 | } |
290 | /* |
291 | * Read/write DACs are always overridable. |
292 | * Executable DACs are overridable when there is |
293 | * at least one exec bit set. |
294 | */ |
295 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) |
296 | if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) |
297 | return 0; |
298 | |
299 | /* |
300 | * Searching includes executable on directories, else just read. |
301 | */ |
302 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
303 | if (mask == MAY_READ) |
304 | if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) |
305 | return 0; |
306 | |
307 | return -EACCES; |
308 | } |
309 | |
310 | /* |
311 | * We _really_ want to just do "generic_permission()" without |
312 | * even looking at the inode->i_op values. So we keep a cache |
313 | * flag in inode->i_opflags, that says "this has not special |
314 | * permission function, use the fast case". |
315 | */ |
316 | static inline int do_inode_permission(struct inode *inode, int mask) |
317 | { |
318 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { |
319 | if (likely(inode->i_op->permission)) |
320 | return inode->i_op->permission(inode, mask); |
321 | |
322 | /* This gets set once for the inode lifetime */ |
323 | spin_lock(&inode->i_lock); |
324 | inode->i_opflags |= IOP_FASTPERM; |
325 | spin_unlock(&inode->i_lock); |
326 | } |
327 | return generic_permission(inode, mask); |
328 | } |
329 | |
330 | /** |
331 | * inode_permission - check for access rights to a given inode |
332 | * @inode: inode to check permission on |
333 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) |
334 | * |
335 | * Used to check for read/write/execute permissions on an inode. |
336 | * We use "fsuid" for this, letting us set arbitrary permissions |
337 | * for filesystem access without changing the "normal" uids which |
338 | * are used for other things. |
339 | * |
340 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. |
341 | */ |
342 | int inode_permission(struct inode *inode, int mask) |
343 | { |
344 | int retval; |
345 | |
346 | if (unlikely(mask & MAY_WRITE)) { |
347 | umode_t mode = inode->i_mode; |
348 | |
349 | /* |
350 | * Nobody gets write access to a read-only fs. |
351 | */ |
352 | if (IS_RDONLY(inode) && |
353 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
354 | return -EROFS; |
355 | |
356 | /* |
357 | * Nobody gets write access to an immutable file. |
358 | */ |
359 | if (IS_IMMUTABLE(inode)) |
360 | return -EACCES; |
361 | } |
362 | |
363 | retval = do_inode_permission(inode, mask); |
364 | if (retval) |
365 | return retval; |
366 | |
367 | retval = devcgroup_inode_permission(inode, mask); |
368 | if (retval) |
369 | return retval; |
370 | |
371 | return security_inode_permission(inode, mask); |
372 | } |
373 | |
374 | /** |
375 | * path_get - get a reference to a path |
376 | * @path: path to get the reference to |
377 | * |
378 | * Given a path increment the reference count to the dentry and the vfsmount. |
379 | */ |
380 | void path_get(struct path *path) |
381 | { |
382 | mntget(path->mnt); |
383 | dget(path->dentry); |
384 | } |
385 | EXPORT_SYMBOL(path_get); |
386 | |
387 | /** |
388 | * path_put - put a reference to a path |
389 | * @path: path to put the reference to |
390 | * |
391 | * Given a path decrement the reference count to the dentry and the vfsmount. |
392 | */ |
393 | void path_put(struct path *path) |
394 | { |
395 | dput(path->dentry); |
396 | mntput(path->mnt); |
397 | } |
398 | EXPORT_SYMBOL(path_put); |
399 | |
400 | /* |
401 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
402 | * Documentation/filesystems/path-lookup.txt). In situations when we can't |
403 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab |
404 | * normal reference counts on dentries and vfsmounts to transition to rcu-walk |
405 | * mode. Refcounts are grabbed at the last known good point before rcu-walk |
406 | * got stuck, so ref-walk may continue from there. If this is not successful |
407 | * (eg. a seqcount has changed), then failure is returned and it's up to caller |
408 | * to restart the path walk from the beginning in ref-walk mode. |
409 | */ |
410 | |
411 | /** |
412 | * unlazy_walk - try to switch to ref-walk mode. |
413 | * @nd: nameidata pathwalk data |
414 | * @dentry: child of nd->path.dentry or NULL |
415 | * Returns: 0 on success, -ECHILD on failure |
416 | * |
417 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry |
418 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on |
419 | * @nd or NULL. Must be called from rcu-walk context. |
420 | */ |
421 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) |
422 | { |
423 | struct fs_struct *fs = current->fs; |
424 | struct dentry *parent = nd->path.dentry; |
425 | int want_root = 0; |
426 | |
427 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
428 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
429 | want_root = 1; |
430 | spin_lock(&fs->lock); |
431 | if (nd->root.mnt != fs->root.mnt || |
432 | nd->root.dentry != fs->root.dentry) |
433 | goto err_root; |
434 | } |
435 | spin_lock(&parent->d_lock); |
436 | if (!dentry) { |
437 | if (!__d_rcu_to_refcount(parent, nd->seq)) |
438 | goto err_parent; |
439 | BUG_ON(nd->inode != parent->d_inode); |
440 | } else { |
441 | if (dentry->d_parent != parent) |
442 | goto err_parent; |
443 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
444 | if (!__d_rcu_to_refcount(dentry, nd->seq)) |
445 | goto err_child; |
446 | /* |
447 | * If the sequence check on the child dentry passed, then |
448 | * the child has not been removed from its parent. This |
449 | * means the parent dentry must be valid and able to take |
450 | * a reference at this point. |
451 | */ |
452 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); |
453 | BUG_ON(!parent->d_count); |
454 | parent->d_count++; |
455 | spin_unlock(&dentry->d_lock); |
456 | } |
457 | spin_unlock(&parent->d_lock); |
458 | if (want_root) { |
459 | path_get(&nd->root); |
460 | spin_unlock(&fs->lock); |
461 | } |
462 | mntget(nd->path.mnt); |
463 | |
464 | rcu_read_unlock(); |
465 | br_read_unlock(vfsmount_lock); |
466 | nd->flags &= ~LOOKUP_RCU; |
467 | return 0; |
468 | |
469 | err_child: |
470 | spin_unlock(&dentry->d_lock); |
471 | err_parent: |
472 | spin_unlock(&parent->d_lock); |
473 | err_root: |
474 | if (want_root) |
475 | spin_unlock(&fs->lock); |
476 | return -ECHILD; |
477 | } |
478 | |
479 | /** |
480 | * release_open_intent - free up open intent resources |
481 | * @nd: pointer to nameidata |
482 | */ |
483 | void release_open_intent(struct nameidata *nd) |
484 | { |
485 | struct file *file = nd->intent.open.file; |
486 | |
487 | if (file && !IS_ERR(file)) { |
488 | if (file->f_path.dentry == NULL) |
489 | put_filp(file); |
490 | else |
491 | fput(file); |
492 | } |
493 | } |
494 | |
495 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
496 | { |
497 | return dentry->d_op->d_revalidate(dentry, nd); |
498 | } |
499 | |
500 | /** |
501 | * complete_walk - successful completion of path walk |
502 | * @nd: pointer nameidata |
503 | * |
504 | * If we had been in RCU mode, drop out of it and legitimize nd->path. |
505 | * Revalidate the final result, unless we'd already done that during |
506 | * the path walk or the filesystem doesn't ask for it. Return 0 on |
507 | * success, -error on failure. In case of failure caller does not |
508 | * need to drop nd->path. |
509 | */ |
510 | static int complete_walk(struct nameidata *nd) |
511 | { |
512 | struct dentry *dentry = nd->path.dentry; |
513 | int status; |
514 | |
515 | if (nd->flags & LOOKUP_RCU) { |
516 | nd->flags &= ~LOOKUP_RCU; |
517 | if (!(nd->flags & LOOKUP_ROOT)) |
518 | nd->root.mnt = NULL; |
519 | spin_lock(&dentry->d_lock); |
520 | if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { |
521 | spin_unlock(&dentry->d_lock); |
522 | rcu_read_unlock(); |
523 | br_read_unlock(vfsmount_lock); |
524 | return -ECHILD; |
525 | } |
526 | BUG_ON(nd->inode != dentry->d_inode); |
527 | spin_unlock(&dentry->d_lock); |
528 | mntget(nd->path.mnt); |
529 | rcu_read_unlock(); |
530 | br_read_unlock(vfsmount_lock); |
531 | } |
532 | |
533 | if (likely(!(nd->flags & LOOKUP_JUMPED))) |
534 | return 0; |
535 | |
536 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) |
537 | return 0; |
538 | |
539 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
540 | return 0; |
541 | |
542 | /* Note: we do not d_invalidate() */ |
543 | status = d_revalidate(dentry, nd); |
544 | if (status > 0) |
545 | return 0; |
546 | |
547 | if (!status) |
548 | status = -ESTALE; |
549 | |
550 | path_put(&nd->path); |
551 | return status; |
552 | } |
553 | |
554 | static __always_inline void set_root(struct nameidata *nd) |
555 | { |
556 | if (!nd->root.mnt) |
557 | get_fs_root(current->fs, &nd->root); |
558 | } |
559 | |
560 | static int link_path_walk(const char *, struct nameidata *); |
561 | |
562 | static __always_inline void set_root_rcu(struct nameidata *nd) |
563 | { |
564 | if (!nd->root.mnt) { |
565 | struct fs_struct *fs = current->fs; |
566 | unsigned seq; |
567 | |
568 | do { |
569 | seq = read_seqcount_begin(&fs->seq); |
570 | nd->root = fs->root; |
571 | nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); |
572 | } while (read_seqcount_retry(&fs->seq, seq)); |
573 | } |
574 | } |
575 | |
576 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
577 | { |
578 | int ret; |
579 | |
580 | if (IS_ERR(link)) |
581 | goto fail; |
582 | |
583 | if (*link == '/') { |
584 | set_root(nd); |
585 | path_put(&nd->path); |
586 | nd->path = nd->root; |
587 | path_get(&nd->root); |
588 | nd->flags |= LOOKUP_JUMPED; |
589 | } |
590 | nd->inode = nd->path.dentry->d_inode; |
591 | |
592 | ret = link_path_walk(link, nd); |
593 | return ret; |
594 | fail: |
595 | path_put(&nd->path); |
596 | return PTR_ERR(link); |
597 | } |
598 | |
599 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
600 | { |
601 | dput(path->dentry); |
602 | if (path->mnt != nd->path.mnt) |
603 | mntput(path->mnt); |
604 | } |
605 | |
606 | static inline void path_to_nameidata(const struct path *path, |
607 | struct nameidata *nd) |
608 | { |
609 | if (!(nd->flags & LOOKUP_RCU)) { |
610 | dput(nd->path.dentry); |
611 | if (nd->path.mnt != path->mnt) |
612 | mntput(nd->path.mnt); |
613 | } |
614 | nd->path.mnt = path->mnt; |
615 | nd->path.dentry = path->dentry; |
616 | } |
617 | |
618 | static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) |
619 | { |
620 | struct inode *inode = link->dentry->d_inode; |
621 | if (!IS_ERR(cookie) && inode->i_op->put_link) |
622 | inode->i_op->put_link(link->dentry, nd, cookie); |
623 | path_put(link); |
624 | } |
625 | |
626 | static __always_inline int |
627 | follow_link(struct path *link, struct nameidata *nd, void **p) |
628 | { |
629 | int error; |
630 | struct dentry *dentry = link->dentry; |
631 | |
632 | BUG_ON(nd->flags & LOOKUP_RCU); |
633 | |
634 | if (link->mnt == nd->path.mnt) |
635 | mntget(link->mnt); |
636 | |
637 | if (unlikely(current->total_link_count >= 40)) { |
638 | *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ |
639 | path_put(&nd->path); |
640 | return -ELOOP; |
641 | } |
642 | cond_resched(); |
643 | current->total_link_count++; |
644 | |
645 | touch_atime(link->mnt, dentry); |
646 | nd_set_link(nd, NULL); |
647 | |
648 | error = security_inode_follow_link(link->dentry, nd); |
649 | if (error) { |
650 | *p = ERR_PTR(error); /* no ->put_link(), please */ |
651 | path_put(&nd->path); |
652 | return error; |
653 | } |
654 | |
655 | nd->last_type = LAST_BIND; |
656 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
657 | error = PTR_ERR(*p); |
658 | if (!IS_ERR(*p)) { |
659 | char *s = nd_get_link(nd); |
660 | error = 0; |
661 | if (s) |
662 | error = __vfs_follow_link(nd, s); |
663 | else if (nd->last_type == LAST_BIND) { |
664 | nd->flags |= LOOKUP_JUMPED; |
665 | nd->inode = nd->path.dentry->d_inode; |
666 | if (nd->inode->i_op->follow_link) { |
667 | /* stepped on a _really_ weird one */ |
668 | path_put(&nd->path); |
669 | error = -ELOOP; |
670 | } |
671 | } |
672 | } |
673 | return error; |
674 | } |
675 | |
676 | static int follow_up_rcu(struct path *path) |
677 | { |
678 | struct mount *mnt = real_mount(path->mnt); |
679 | struct mount *parent; |
680 | struct dentry *mountpoint; |
681 | |
682 | parent = mnt->mnt_parent; |
683 | if (&parent->mnt == path->mnt) |
684 | return 0; |
685 | mountpoint = mnt->mnt_mountpoint; |
686 | path->dentry = mountpoint; |
687 | path->mnt = &parent->mnt; |
688 | return 1; |
689 | } |
690 | |
691 | int follow_up(struct path *path) |
692 | { |
693 | struct mount *mnt = real_mount(path->mnt); |
694 | struct mount *parent; |
695 | struct dentry *mountpoint; |
696 | |
697 | br_read_lock(vfsmount_lock); |
698 | parent = mnt->mnt_parent; |
699 | if (&parent->mnt == path->mnt) { |
700 | br_read_unlock(vfsmount_lock); |
701 | return 0; |
702 | } |
703 | mntget(&parent->mnt); |
704 | mountpoint = dget(mnt->mnt_mountpoint); |
705 | br_read_unlock(vfsmount_lock); |
706 | dput(path->dentry); |
707 | path->dentry = mountpoint; |
708 | mntput(path->mnt); |
709 | path->mnt = &parent->mnt; |
710 | return 1; |
711 | } |
712 | |
713 | /* |
714 | * Perform an automount |
715 | * - return -EISDIR to tell follow_managed() to stop and return the path we |
716 | * were called with. |
717 | */ |
718 | static int follow_automount(struct path *path, unsigned flags, |
719 | bool *need_mntput) |
720 | { |
721 | struct vfsmount *mnt; |
722 | int err; |
723 | |
724 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) |
725 | return -EREMOTE; |
726 | |
727 | /* We don't want to mount if someone's just doing a stat - |
728 | * unless they're stat'ing a directory and appended a '/' to |
729 | * the name. |
730 | * |
731 | * We do, however, want to mount if someone wants to open or |
732 | * create a file of any type under the mountpoint, wants to |
733 | * traverse through the mountpoint or wants to open the |
734 | * mounted directory. Also, autofs may mark negative dentries |
735 | * as being automount points. These will need the attentions |
736 | * of the daemon to instantiate them before they can be used. |
737 | */ |
738 | if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | |
739 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && |
740 | path->dentry->d_inode) |
741 | return -EISDIR; |
742 | |
743 | current->total_link_count++; |
744 | if (current->total_link_count >= 40) |
745 | return -ELOOP; |
746 | |
747 | mnt = path->dentry->d_op->d_automount(path); |
748 | if (IS_ERR(mnt)) { |
749 | /* |
750 | * The filesystem is allowed to return -EISDIR here to indicate |
751 | * it doesn't want to automount. For instance, autofs would do |
752 | * this so that its userspace daemon can mount on this dentry. |
753 | * |
754 | * However, we can only permit this if it's a terminal point in |
755 | * the path being looked up; if it wasn't then the remainder of |
756 | * the path is inaccessible and we should say so. |
757 | */ |
758 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) |
759 | return -EREMOTE; |
760 | return PTR_ERR(mnt); |
761 | } |
762 | |
763 | if (!mnt) /* mount collision */ |
764 | return 0; |
765 | |
766 | if (!*need_mntput) { |
767 | /* lock_mount() may release path->mnt on error */ |
768 | mntget(path->mnt); |
769 | *need_mntput = true; |
770 | } |
771 | err = finish_automount(mnt, path); |
772 | |
773 | switch (err) { |
774 | case -EBUSY: |
775 | /* Someone else made a mount here whilst we were busy */ |
776 | return 0; |
777 | case 0: |
778 | path_put(path); |
779 | path->mnt = mnt; |
780 | path->dentry = dget(mnt->mnt_root); |
781 | return 0; |
782 | default: |
783 | return err; |
784 | } |
785 | |
786 | } |
787 | |
788 | /* |
789 | * Handle a dentry that is managed in some way. |
790 | * - Flagged for transit management (autofs) |
791 | * - Flagged as mountpoint |
792 | * - Flagged as automount point |
793 | * |
794 | * This may only be called in refwalk mode. |
795 | * |
796 | * Serialization is taken care of in namespace.c |
797 | */ |
798 | static int follow_managed(struct path *path, unsigned flags) |
799 | { |
800 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ |
801 | unsigned managed; |
802 | bool need_mntput = false; |
803 | int ret = 0; |
804 | |
805 | /* Given that we're not holding a lock here, we retain the value in a |
806 | * local variable for each dentry as we look at it so that we don't see |
807 | * the components of that value change under us */ |
808 | while (managed = ACCESS_ONCE(path->dentry->d_flags), |
809 | managed &= DCACHE_MANAGED_DENTRY, |
810 | unlikely(managed != 0)) { |
811 | /* Allow the filesystem to manage the transit without i_mutex |
812 | * being held. */ |
813 | if (managed & DCACHE_MANAGE_TRANSIT) { |
814 | BUG_ON(!path->dentry->d_op); |
815 | BUG_ON(!path->dentry->d_op->d_manage); |
816 | ret = path->dentry->d_op->d_manage(path->dentry, false); |
817 | if (ret < 0) |
818 | break; |
819 | } |
820 | |
821 | /* Transit to a mounted filesystem. */ |
822 | if (managed & DCACHE_MOUNTED) { |
823 | struct vfsmount *mounted = lookup_mnt(path); |
824 | if (mounted) { |
825 | dput(path->dentry); |
826 | if (need_mntput) |
827 | mntput(path->mnt); |
828 | path->mnt = mounted; |
829 | path->dentry = dget(mounted->mnt_root); |
830 | need_mntput = true; |
831 | continue; |
832 | } |
833 | |
834 | /* Something is mounted on this dentry in another |
835 | * namespace and/or whatever was mounted there in this |
836 | * namespace got unmounted before we managed to get the |
837 | * vfsmount_lock */ |
838 | } |
839 | |
840 | /* Handle an automount point */ |
841 | if (managed & DCACHE_NEED_AUTOMOUNT) { |
842 | ret = follow_automount(path, flags, &need_mntput); |
843 | if (ret < 0) |
844 | break; |
845 | continue; |
846 | } |
847 | |
848 | /* We didn't change the current path point */ |
849 | break; |
850 | } |
851 | |
852 | if (need_mntput && path->mnt == mnt) |
853 | mntput(path->mnt); |
854 | if (ret == -EISDIR) |
855 | ret = 0; |
856 | return ret < 0 ? ret : need_mntput; |
857 | } |
858 | |
859 | int follow_down_one(struct path *path) |
860 | { |
861 | struct vfsmount *mounted; |
862 | |
863 | mounted = lookup_mnt(path); |
864 | if (mounted) { |
865 | dput(path->dentry); |
866 | mntput(path->mnt); |
867 | path->mnt = mounted; |
868 | path->dentry = dget(mounted->mnt_root); |
869 | return 1; |
870 | } |
871 | return 0; |
872 | } |
873 | |
874 | static inline bool managed_dentry_might_block(struct dentry *dentry) |
875 | { |
876 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && |
877 | dentry->d_op->d_manage(dentry, true) < 0); |
878 | } |
879 | |
880 | /* |
881 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if |
882 | * we meet a managed dentry that would need blocking. |
883 | */ |
884 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, |
885 | struct inode **inode) |
886 | { |
887 | for (;;) { |
888 | struct mount *mounted; |
889 | /* |
890 | * Don't forget we might have a non-mountpoint managed dentry |
891 | * that wants to block transit. |
892 | */ |
893 | if (unlikely(managed_dentry_might_block(path->dentry))) |
894 | return false; |
895 | |
896 | if (!d_mountpoint(path->dentry)) |
897 | break; |
898 | |
899 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
900 | if (!mounted) |
901 | break; |
902 | path->mnt = &mounted->mnt; |
903 | path->dentry = mounted->mnt.mnt_root; |
904 | nd->flags |= LOOKUP_JUMPED; |
905 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
906 | /* |
907 | * Update the inode too. We don't need to re-check the |
908 | * dentry sequence number here after this d_inode read, |
909 | * because a mount-point is always pinned. |
910 | */ |
911 | *inode = path->dentry->d_inode; |
912 | } |
913 | return true; |
914 | } |
915 | |
916 | static void follow_mount_rcu(struct nameidata *nd) |
917 | { |
918 | while (d_mountpoint(nd->path.dentry)) { |
919 | struct mount *mounted; |
920 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); |
921 | if (!mounted) |
922 | break; |
923 | nd->path.mnt = &mounted->mnt; |
924 | nd->path.dentry = mounted->mnt.mnt_root; |
925 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
926 | } |
927 | } |
928 | |
929 | static int follow_dotdot_rcu(struct nameidata *nd) |
930 | { |
931 | set_root_rcu(nd); |
932 | |
933 | while (1) { |
934 | if (nd->path.dentry == nd->root.dentry && |
935 | nd->path.mnt == nd->root.mnt) { |
936 | break; |
937 | } |
938 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
939 | struct dentry *old = nd->path.dentry; |
940 | struct dentry *parent = old->d_parent; |
941 | unsigned seq; |
942 | |
943 | seq = read_seqcount_begin(&parent->d_seq); |
944 | if (read_seqcount_retry(&old->d_seq, nd->seq)) |
945 | goto failed; |
946 | nd->path.dentry = parent; |
947 | nd->seq = seq; |
948 | break; |
949 | } |
950 | if (!follow_up_rcu(&nd->path)) |
951 | break; |
952 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
953 | } |
954 | follow_mount_rcu(nd); |
955 | nd->inode = nd->path.dentry->d_inode; |
956 | return 0; |
957 | |
958 | failed: |
959 | nd->flags &= ~LOOKUP_RCU; |
960 | if (!(nd->flags & LOOKUP_ROOT)) |
961 | nd->root.mnt = NULL; |
962 | rcu_read_unlock(); |
963 | br_read_unlock(vfsmount_lock); |
964 | return -ECHILD; |
965 | } |
966 | |
967 | /* |
968 | * Follow down to the covering mount currently visible to userspace. At each |
969 | * point, the filesystem owning that dentry may be queried as to whether the |
970 | * caller is permitted to proceed or not. |
971 | */ |
972 | int follow_down(struct path *path) |
973 | { |
974 | unsigned managed; |
975 | int ret; |
976 | |
977 | while (managed = ACCESS_ONCE(path->dentry->d_flags), |
978 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { |
979 | /* Allow the filesystem to manage the transit without i_mutex |
980 | * being held. |
981 | * |
982 | * We indicate to the filesystem if someone is trying to mount |
983 | * something here. This gives autofs the chance to deny anyone |
984 | * other than its daemon the right to mount on its |
985 | * superstructure. |
986 | * |
987 | * The filesystem may sleep at this point. |
988 | */ |
989 | if (managed & DCACHE_MANAGE_TRANSIT) { |
990 | BUG_ON(!path->dentry->d_op); |
991 | BUG_ON(!path->dentry->d_op->d_manage); |
992 | ret = path->dentry->d_op->d_manage( |
993 | path->dentry, false); |
994 | if (ret < 0) |
995 | return ret == -EISDIR ? 0 : ret; |
996 | } |
997 | |
998 | /* Transit to a mounted filesystem. */ |
999 | if (managed & DCACHE_MOUNTED) { |
1000 | struct vfsmount *mounted = lookup_mnt(path); |
1001 | if (!mounted) |
1002 | break; |
1003 | dput(path->dentry); |
1004 | mntput(path->mnt); |
1005 | path->mnt = mounted; |
1006 | path->dentry = dget(mounted->mnt_root); |
1007 | continue; |
1008 | } |
1009 | |
1010 | /* Don't handle automount points here */ |
1011 | break; |
1012 | } |
1013 | return 0; |
1014 | } |
1015 | |
1016 | /* |
1017 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() |
1018 | */ |
1019 | static void follow_mount(struct path *path) |
1020 | { |
1021 | while (d_mountpoint(path->dentry)) { |
1022 | struct vfsmount *mounted = lookup_mnt(path); |
1023 | if (!mounted) |
1024 | break; |
1025 | dput(path->dentry); |
1026 | mntput(path->mnt); |
1027 | path->mnt = mounted; |
1028 | path->dentry = dget(mounted->mnt_root); |
1029 | } |
1030 | } |
1031 | |
1032 | static void follow_dotdot(struct nameidata *nd) |
1033 | { |
1034 | set_root(nd); |
1035 | |
1036 | while(1) { |
1037 | struct dentry *old = nd->path.dentry; |
1038 | |
1039 | if (nd->path.dentry == nd->root.dentry && |
1040 | nd->path.mnt == nd->root.mnt) { |
1041 | break; |
1042 | } |
1043 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1044 | /* rare case of legitimate dget_parent()... */ |
1045 | nd->path.dentry = dget_parent(nd->path.dentry); |
1046 | dput(old); |
1047 | break; |
1048 | } |
1049 | if (!follow_up(&nd->path)) |
1050 | break; |
1051 | } |
1052 | follow_mount(&nd->path); |
1053 | nd->inode = nd->path.dentry->d_inode; |
1054 | } |
1055 | |
1056 | /* |
1057 | * Allocate a dentry with name and parent, and perform a parent |
1058 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR |
1059 | * on error. parent->d_inode->i_mutex must be held. d_lookup must |
1060 | * have verified that no child exists while under i_mutex. |
1061 | */ |
1062 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, |
1063 | struct qstr *name, struct nameidata *nd) |
1064 | { |
1065 | struct inode *inode = parent->d_inode; |
1066 | struct dentry *dentry; |
1067 | struct dentry *old; |
1068 | |
1069 | /* Don't create child dentry for a dead directory. */ |
1070 | if (unlikely(IS_DEADDIR(inode))) |
1071 | return ERR_PTR(-ENOENT); |
1072 | |
1073 | dentry = d_alloc(parent, name); |
1074 | if (unlikely(!dentry)) |
1075 | return ERR_PTR(-ENOMEM); |
1076 | |
1077 | old = inode->i_op->lookup(inode, dentry, nd); |
1078 | if (unlikely(old)) { |
1079 | dput(dentry); |
1080 | dentry = old; |
1081 | } |
1082 | return dentry; |
1083 | } |
1084 | |
1085 | /* |
1086 | * We already have a dentry, but require a lookup to be performed on the parent |
1087 | * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error. |
1088 | * parent->d_inode->i_mutex must be held. d_lookup must have verified that no |
1089 | * child exists while under i_mutex. |
1090 | */ |
1091 | static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry, |
1092 | struct nameidata *nd) |
1093 | { |
1094 | struct inode *inode = parent->d_inode; |
1095 | struct dentry *old; |
1096 | |
1097 | /* Don't create child dentry for a dead directory. */ |
1098 | if (unlikely(IS_DEADDIR(inode))) { |
1099 | dput(dentry); |
1100 | return ERR_PTR(-ENOENT); |
1101 | } |
1102 | |
1103 | old = inode->i_op->lookup(inode, dentry, nd); |
1104 | if (unlikely(old)) { |
1105 | dput(dentry); |
1106 | dentry = old; |
1107 | } |
1108 | return dentry; |
1109 | } |
1110 | |
1111 | /* |
1112 | * It's more convoluted than I'd like it to be, but... it's still fairly |
1113 | * small and for now I'd prefer to have fast path as straight as possible. |
1114 | * It _is_ time-critical. |
1115 | */ |
1116 | static int do_lookup(struct nameidata *nd, struct qstr *name, |
1117 | struct path *path, struct inode **inode) |
1118 | { |
1119 | struct vfsmount *mnt = nd->path.mnt; |
1120 | struct dentry *dentry, *parent = nd->path.dentry; |
1121 | int need_reval = 1; |
1122 | int status = 1; |
1123 | int err; |
1124 | |
1125 | /* |
1126 | * Rename seqlock is not required here because in the off chance |
1127 | * of a false negative due to a concurrent rename, we're going to |
1128 | * do the non-racy lookup, below. |
1129 | */ |
1130 | if (nd->flags & LOOKUP_RCU) { |
1131 | unsigned seq; |
1132 | *inode = nd->inode; |
1133 | dentry = __d_lookup_rcu(parent, name, &seq, inode); |
1134 | if (!dentry) |
1135 | goto unlazy; |
1136 | |
1137 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1138 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1139 | return -ECHILD; |
1140 | nd->seq = seq; |
1141 | |
1142 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1143 | status = d_revalidate(dentry, nd); |
1144 | if (unlikely(status <= 0)) { |
1145 | if (status != -ECHILD) |
1146 | need_reval = 0; |
1147 | goto unlazy; |
1148 | } |
1149 | } |
1150 | if (unlikely(d_need_lookup(dentry))) |
1151 | goto unlazy; |
1152 | path->mnt = mnt; |
1153 | path->dentry = dentry; |
1154 | if (unlikely(!__follow_mount_rcu(nd, path, inode))) |
1155 | goto unlazy; |
1156 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) |
1157 | goto unlazy; |
1158 | return 0; |
1159 | unlazy: |
1160 | if (unlazy_walk(nd, dentry)) |
1161 | return -ECHILD; |
1162 | } else { |
1163 | dentry = __d_lookup(parent, name); |
1164 | } |
1165 | |
1166 | if (dentry && unlikely(d_need_lookup(dentry))) { |
1167 | dput(dentry); |
1168 | dentry = NULL; |
1169 | } |
1170 | retry: |
1171 | if (unlikely(!dentry)) { |
1172 | struct inode *dir = parent->d_inode; |
1173 | BUG_ON(nd->inode != dir); |
1174 | |
1175 | mutex_lock(&dir->i_mutex); |
1176 | dentry = d_lookup(parent, name); |
1177 | if (likely(!dentry)) { |
1178 | dentry = d_alloc_and_lookup(parent, name, nd); |
1179 | if (IS_ERR(dentry)) { |
1180 | mutex_unlock(&dir->i_mutex); |
1181 | return PTR_ERR(dentry); |
1182 | } |
1183 | /* known good */ |
1184 | need_reval = 0; |
1185 | status = 1; |
1186 | } else if (unlikely(d_need_lookup(dentry))) { |
1187 | dentry = d_inode_lookup(parent, dentry, nd); |
1188 | if (IS_ERR(dentry)) { |
1189 | mutex_unlock(&dir->i_mutex); |
1190 | return PTR_ERR(dentry); |
1191 | } |
1192 | /* known good */ |
1193 | need_reval = 0; |
1194 | status = 1; |
1195 | } |
1196 | mutex_unlock(&dir->i_mutex); |
1197 | } |
1198 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) |
1199 | status = d_revalidate(dentry, nd); |
1200 | if (unlikely(status <= 0)) { |
1201 | if (status < 0) { |
1202 | dput(dentry); |
1203 | return status; |
1204 | } |
1205 | if (!d_invalidate(dentry)) { |
1206 | dput(dentry); |
1207 | dentry = NULL; |
1208 | need_reval = 1; |
1209 | goto retry; |
1210 | } |
1211 | } |
1212 | |
1213 | path->mnt = mnt; |
1214 | path->dentry = dentry; |
1215 | err = follow_managed(path, nd->flags); |
1216 | if (unlikely(err < 0)) { |
1217 | path_put_conditional(path, nd); |
1218 | return err; |
1219 | } |
1220 | if (err) |
1221 | nd->flags |= LOOKUP_JUMPED; |
1222 | *inode = path->dentry->d_inode; |
1223 | return 0; |
1224 | } |
1225 | |
1226 | static inline int may_lookup(struct nameidata *nd) |
1227 | { |
1228 | if (nd->flags & LOOKUP_RCU) { |
1229 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); |
1230 | if (err != -ECHILD) |
1231 | return err; |
1232 | if (unlazy_walk(nd, NULL)) |
1233 | return -ECHILD; |
1234 | } |
1235 | return inode_permission(nd->inode, MAY_EXEC); |
1236 | } |
1237 | |
1238 | static inline int handle_dots(struct nameidata *nd, int type) |
1239 | { |
1240 | if (type == LAST_DOTDOT) { |
1241 | if (nd->flags & LOOKUP_RCU) { |
1242 | if (follow_dotdot_rcu(nd)) |
1243 | return -ECHILD; |
1244 | } else |
1245 | follow_dotdot(nd); |
1246 | } |
1247 | return 0; |
1248 | } |
1249 | |
1250 | static void terminate_walk(struct nameidata *nd) |
1251 | { |
1252 | if (!(nd->flags & LOOKUP_RCU)) { |
1253 | path_put(&nd->path); |
1254 | } else { |
1255 | nd->flags &= ~LOOKUP_RCU; |
1256 | if (!(nd->flags & LOOKUP_ROOT)) |
1257 | nd->root.mnt = NULL; |
1258 | rcu_read_unlock(); |
1259 | br_read_unlock(vfsmount_lock); |
1260 | } |
1261 | } |
1262 | |
1263 | /* |
1264 | * Do we need to follow links? We _really_ want to be able |
1265 | * to do this check without having to look at inode->i_op, |
1266 | * so we keep a cache of "no, this doesn't need follow_link" |
1267 | * for the common case. |
1268 | */ |
1269 | static inline int should_follow_link(struct inode *inode, int follow) |
1270 | { |
1271 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { |
1272 | if (likely(inode->i_op->follow_link)) |
1273 | return follow; |
1274 | |
1275 | /* This gets set once for the inode lifetime */ |
1276 | spin_lock(&inode->i_lock); |
1277 | inode->i_opflags |= IOP_NOFOLLOW; |
1278 | spin_unlock(&inode->i_lock); |
1279 | } |
1280 | return 0; |
1281 | } |
1282 | |
1283 | static inline int walk_component(struct nameidata *nd, struct path *path, |
1284 | struct qstr *name, int type, int follow) |
1285 | { |
1286 | struct inode *inode; |
1287 | int err; |
1288 | /* |
1289 | * "." and ".." are special - ".." especially so because it has |
1290 | * to be able to know about the current root directory and |
1291 | * parent relationships. |
1292 | */ |
1293 | if (unlikely(type != LAST_NORM)) |
1294 | return handle_dots(nd, type); |
1295 | err = do_lookup(nd, name, path, &inode); |
1296 | if (unlikely(err)) { |
1297 | terminate_walk(nd); |
1298 | return err; |
1299 | } |
1300 | if (!inode) { |
1301 | path_to_nameidata(path, nd); |
1302 | terminate_walk(nd); |
1303 | return -ENOENT; |
1304 | } |
1305 | if (should_follow_link(inode, follow)) { |
1306 | if (nd->flags & LOOKUP_RCU) { |
1307 | if (unlikely(unlazy_walk(nd, path->dentry))) { |
1308 | terminate_walk(nd); |
1309 | return -ECHILD; |
1310 | } |
1311 | } |
1312 | BUG_ON(inode != path->dentry->d_inode); |
1313 | return 1; |
1314 | } |
1315 | path_to_nameidata(path, nd); |
1316 | nd->inode = inode; |
1317 | return 0; |
1318 | } |
1319 | |
1320 | /* |
1321 | * This limits recursive symlink follows to 8, while |
1322 | * limiting consecutive symlinks to 40. |
1323 | * |
1324 | * Without that kind of total limit, nasty chains of consecutive |
1325 | * symlinks can cause almost arbitrarily long lookups. |
1326 | */ |
1327 | static inline int nested_symlink(struct path *path, struct nameidata *nd) |
1328 | { |
1329 | int res; |
1330 | |
1331 | if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { |
1332 | path_put_conditional(path, nd); |
1333 | path_put(&nd->path); |
1334 | return -ELOOP; |
1335 | } |
1336 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
1337 | |
1338 | nd->depth++; |
1339 | current->link_count++; |
1340 | |
1341 | do { |
1342 | struct path link = *path; |
1343 | void *cookie; |
1344 | |
1345 | res = follow_link(&link, nd, &cookie); |
1346 | if (!res) |
1347 | res = walk_component(nd, path, &nd->last, |
1348 | nd->last_type, LOOKUP_FOLLOW); |
1349 | put_link(nd, &link, cookie); |
1350 | } while (res > 0); |
1351 | |
1352 | current->link_count--; |
1353 | nd->depth--; |
1354 | return res; |
1355 | } |
1356 | |
1357 | /* |
1358 | * We really don't want to look at inode->i_op->lookup |
1359 | * when we don't have to. So we keep a cache bit in |
1360 | * the inode ->i_opflags field that says "yes, we can |
1361 | * do lookup on this inode". |
1362 | */ |
1363 | static inline int can_lookup(struct inode *inode) |
1364 | { |
1365 | if (likely(inode->i_opflags & IOP_LOOKUP)) |
1366 | return 1; |
1367 | if (likely(!inode->i_op->lookup)) |
1368 | return 0; |
1369 | |
1370 | /* We do this once for the lifetime of the inode */ |
1371 | spin_lock(&inode->i_lock); |
1372 | inode->i_opflags |= IOP_LOOKUP; |
1373 | spin_unlock(&inode->i_lock); |
1374 | return 1; |
1375 | } |
1376 | |
1377 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) |
1378 | { |
1379 | unsigned long hash = init_name_hash(); |
1380 | while (len--) |
1381 | hash = partial_name_hash(*name++, hash); |
1382 | return end_name_hash(hash); |
1383 | } |
1384 | EXPORT_SYMBOL(full_name_hash); |
1385 | |
1386 | /* |
1387 | * We know there's a real path component here of at least |
1388 | * one character. |
1389 | */ |
1390 | static inline unsigned long hash_name(const char *name, unsigned int *hashp) |
1391 | { |
1392 | unsigned long hash = init_name_hash(); |
1393 | unsigned long len = 0, c; |
1394 | |
1395 | c = (unsigned char)*name; |
1396 | do { |
1397 | len++; |
1398 | hash = partial_name_hash(c, hash); |
1399 | c = (unsigned char)name[len]; |
1400 | } while (c && c != '/'); |
1401 | *hashp = end_name_hash(hash); |
1402 | return len; |
1403 | } |
1404 | |
1405 | /* |
1406 | * Name resolution. |
1407 | * This is the basic name resolution function, turning a pathname into |
1408 | * the final dentry. We expect 'base' to be positive and a directory. |
1409 | * |
1410 | * Returns 0 and nd will have valid dentry and mnt on success. |
1411 | * Returns error and drops reference to input namei data on failure. |
1412 | */ |
1413 | static int link_path_walk(const char *name, struct nameidata *nd) |
1414 | { |
1415 | struct path next; |
1416 | int err; |
1417 | |
1418 | while (*name=='/') |
1419 | name++; |
1420 | if (!*name) |
1421 | return 0; |
1422 | |
1423 | /* At this point we know we have a real path component. */ |
1424 | for(;;) { |
1425 | struct qstr this; |
1426 | long len; |
1427 | int type; |
1428 | |
1429 | err = may_lookup(nd); |
1430 | if (err) |
1431 | break; |
1432 | |
1433 | len = hash_name(name, &this.hash); |
1434 | this.name = name; |
1435 | this.len = len; |
1436 | |
1437 | type = LAST_NORM; |
1438 | if (name[0] == '.') switch (len) { |
1439 | case 2: |
1440 | if (name[1] == '.') { |
1441 | type = LAST_DOTDOT; |
1442 | nd->flags |= LOOKUP_JUMPED; |
1443 | } |
1444 | break; |
1445 | case 1: |
1446 | type = LAST_DOT; |
1447 | } |
1448 | if (likely(type == LAST_NORM)) { |
1449 | struct dentry *parent = nd->path.dentry; |
1450 | nd->flags &= ~LOOKUP_JUMPED; |
1451 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1452 | err = parent->d_op->d_hash(parent, nd->inode, |
1453 | &this); |
1454 | if (err < 0) |
1455 | break; |
1456 | } |
1457 | } |
1458 | |
1459 | if (!name[len]) |
1460 | goto last_component; |
1461 | /* |
1462 | * If it wasn't NUL, we know it was '/'. Skip that |
1463 | * slash, and continue until no more slashes. |
1464 | */ |
1465 | do { |
1466 | len++; |
1467 | } while (unlikely(name[len] == '/')); |
1468 | if (!name[len]) |
1469 | goto last_component; |
1470 | name += len; |
1471 | |
1472 | err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); |
1473 | if (err < 0) |
1474 | return err; |
1475 | |
1476 | if (err) { |
1477 | err = nested_symlink(&next, nd); |
1478 | if (err) |
1479 | return err; |
1480 | } |
1481 | if (can_lookup(nd->inode)) |
1482 | continue; |
1483 | err = -ENOTDIR; |
1484 | break; |
1485 | /* here ends the main loop */ |
1486 | |
1487 | last_component: |
1488 | nd->last = this; |
1489 | nd->last_type = type; |
1490 | return 0; |
1491 | } |
1492 | terminate_walk(nd); |
1493 | return err; |
1494 | } |
1495 | |
1496 | static int path_init(int dfd, const char *name, unsigned int flags, |
1497 | struct nameidata *nd, struct file **fp) |
1498 | { |
1499 | int retval = 0; |
1500 | int fput_needed; |
1501 | struct file *file; |
1502 | |
1503 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ |
1504 | nd->flags = flags | LOOKUP_JUMPED; |
1505 | nd->depth = 0; |
1506 | if (flags & LOOKUP_ROOT) { |
1507 | struct inode *inode = nd->root.dentry->d_inode; |
1508 | if (*name) { |
1509 | if (!inode->i_op->lookup) |
1510 | return -ENOTDIR; |
1511 | retval = inode_permission(inode, MAY_EXEC); |
1512 | if (retval) |
1513 | return retval; |
1514 | } |
1515 | nd->path = nd->root; |
1516 | nd->inode = inode; |
1517 | if (flags & LOOKUP_RCU) { |
1518 | br_read_lock(vfsmount_lock); |
1519 | rcu_read_lock(); |
1520 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1521 | } else { |
1522 | path_get(&nd->path); |
1523 | } |
1524 | return 0; |
1525 | } |
1526 | |
1527 | nd->root.mnt = NULL; |
1528 | |
1529 | if (*name=='/') { |
1530 | if (flags & LOOKUP_RCU) { |
1531 | br_read_lock(vfsmount_lock); |
1532 | rcu_read_lock(); |
1533 | set_root_rcu(nd); |
1534 | } else { |
1535 | set_root(nd); |
1536 | path_get(&nd->root); |
1537 | } |
1538 | nd->path = nd->root; |
1539 | } else if (dfd == AT_FDCWD) { |
1540 | if (flags & LOOKUP_RCU) { |
1541 | struct fs_struct *fs = current->fs; |
1542 | unsigned seq; |
1543 | |
1544 | br_read_lock(vfsmount_lock); |
1545 | rcu_read_lock(); |
1546 | |
1547 | do { |
1548 | seq = read_seqcount_begin(&fs->seq); |
1549 | nd->path = fs->pwd; |
1550 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1551 | } while (read_seqcount_retry(&fs->seq, seq)); |
1552 | } else { |
1553 | get_fs_pwd(current->fs, &nd->path); |
1554 | } |
1555 | } else { |
1556 | struct dentry *dentry; |
1557 | |
1558 | file = fget_raw_light(dfd, &fput_needed); |
1559 | retval = -EBADF; |
1560 | if (!file) |
1561 | goto out_fail; |
1562 | |
1563 | dentry = file->f_path.dentry; |
1564 | |
1565 | if (*name) { |
1566 | retval = -ENOTDIR; |
1567 | if (!S_ISDIR(dentry->d_inode->i_mode)) |
1568 | goto fput_fail; |
1569 | |
1570 | retval = inode_permission(dentry->d_inode, MAY_EXEC); |
1571 | if (retval) |
1572 | goto fput_fail; |
1573 | } |
1574 | |
1575 | nd->path = file->f_path; |
1576 | if (flags & LOOKUP_RCU) { |
1577 | if (fput_needed) |
1578 | *fp = file; |
1579 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1580 | br_read_lock(vfsmount_lock); |
1581 | rcu_read_lock(); |
1582 | } else { |
1583 | path_get(&file->f_path); |
1584 | fput_light(file, fput_needed); |
1585 | } |
1586 | } |
1587 | |
1588 | nd->inode = nd->path.dentry->d_inode; |
1589 | return 0; |
1590 | |
1591 | fput_fail: |
1592 | fput_light(file, fput_needed); |
1593 | out_fail: |
1594 | return retval; |
1595 | } |
1596 | |
1597 | static inline int lookup_last(struct nameidata *nd, struct path *path) |
1598 | { |
1599 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) |
1600 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
1601 | |
1602 | nd->flags &= ~LOOKUP_PARENT; |
1603 | return walk_component(nd, path, &nd->last, nd->last_type, |
1604 | nd->flags & LOOKUP_FOLLOW); |
1605 | } |
1606 | |
1607 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
1608 | static int path_lookupat(int dfd, const char *name, |
1609 | unsigned int flags, struct nameidata *nd) |
1610 | { |
1611 | struct file *base = NULL; |
1612 | struct path path; |
1613 | int err; |
1614 | |
1615 | /* |
1616 | * Path walking is largely split up into 2 different synchronisation |
1617 | * schemes, rcu-walk and ref-walk (explained in |
1618 | * Documentation/filesystems/path-lookup.txt). These share much of the |
1619 | * path walk code, but some things particularly setup, cleanup, and |
1620 | * following mounts are sufficiently divergent that functions are |
1621 | * duplicated. Typically there is a function foo(), and its RCU |
1622 | * analogue, foo_rcu(). |
1623 | * |
1624 | * -ECHILD is the error number of choice (just to avoid clashes) that |
1625 | * is returned if some aspect of an rcu-walk fails. Such an error must |
1626 | * be handled by restarting a traditional ref-walk (which will always |
1627 | * be able to complete). |
1628 | */ |
1629 | err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); |
1630 | |
1631 | if (unlikely(err)) |
1632 | return err; |
1633 | |
1634 | current->total_link_count = 0; |
1635 | err = link_path_walk(name, nd); |
1636 | |
1637 | if (!err && !(flags & LOOKUP_PARENT)) { |
1638 | err = lookup_last(nd, &path); |
1639 | while (err > 0) { |
1640 | void *cookie; |
1641 | struct path link = path; |
1642 | nd->flags |= LOOKUP_PARENT; |
1643 | err = follow_link(&link, nd, &cookie); |
1644 | if (!err) |
1645 | err = lookup_last(nd, &path); |
1646 | put_link(nd, &link, cookie); |
1647 | } |
1648 | } |
1649 | |
1650 | if (!err) |
1651 | err = complete_walk(nd); |
1652 | |
1653 | if (!err && nd->flags & LOOKUP_DIRECTORY) { |
1654 | if (!nd->inode->i_op->lookup) { |
1655 | path_put(&nd->path); |
1656 | err = -ENOTDIR; |
1657 | } |
1658 | } |
1659 | |
1660 | if (base) |
1661 | fput(base); |
1662 | |
1663 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { |
1664 | path_put(&nd->root); |
1665 | nd->root.mnt = NULL; |
1666 | } |
1667 | return err; |
1668 | } |
1669 | |
1670 | static int do_path_lookup(int dfd, const char *name, |
1671 | unsigned int flags, struct nameidata *nd) |
1672 | { |
1673 | int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); |
1674 | if (unlikely(retval == -ECHILD)) |
1675 | retval = path_lookupat(dfd, name, flags, nd); |
1676 | if (unlikely(retval == -ESTALE)) |
1677 | retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); |
1678 | |
1679 | if (likely(!retval)) { |
1680 | if (unlikely(!audit_dummy_context())) { |
1681 | if (nd->path.dentry && nd->inode) |
1682 | audit_inode(name, nd->path.dentry); |
1683 | } |
1684 | } |
1685 | return retval; |
1686 | } |
1687 | |
1688 | int kern_path_parent(const char *name, struct nameidata *nd) |
1689 | { |
1690 | return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); |
1691 | } |
1692 | |
1693 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1694 | { |
1695 | struct nameidata nd; |
1696 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); |
1697 | if (!res) |
1698 | *path = nd.path; |
1699 | return res; |
1700 | } |
1701 | |
1702 | /** |
1703 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair |
1704 | * @dentry: pointer to dentry of the base directory |
1705 | * @mnt: pointer to vfs mount of the base directory |
1706 | * @name: pointer to file name |
1707 | * @flags: lookup flags |
1708 | * @path: pointer to struct path to fill |
1709 | */ |
1710 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, |
1711 | const char *name, unsigned int flags, |
1712 | struct path *path) |
1713 | { |
1714 | struct nameidata nd; |
1715 | int err; |
1716 | nd.root.dentry = dentry; |
1717 | nd.root.mnt = mnt; |
1718 | BUG_ON(flags & LOOKUP_PARENT); |
1719 | /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ |
1720 | err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); |
1721 | if (!err) |
1722 | *path = nd.path; |
1723 | return err; |
1724 | } |
1725 | |
1726 | static struct dentry *__lookup_hash(struct qstr *name, |
1727 | struct dentry *base, struct nameidata *nd) |
1728 | { |
1729 | struct inode *inode = base->d_inode; |
1730 | struct dentry *dentry; |
1731 | int err; |
1732 | |
1733 | err = inode_permission(inode, MAY_EXEC); |
1734 | if (err) |
1735 | return ERR_PTR(err); |
1736 | |
1737 | /* |
1738 | * Don't bother with __d_lookup: callers are for creat as |
1739 | * well as unlink, so a lot of the time it would cost |
1740 | * a double lookup. |
1741 | */ |
1742 | dentry = d_lookup(base, name); |
1743 | |
1744 | if (dentry && d_need_lookup(dentry)) { |
1745 | /* |
1746 | * __lookup_hash is called with the parent dir's i_mutex already |
1747 | * held, so we are good to go here. |
1748 | */ |
1749 | dentry = d_inode_lookup(base, dentry, nd); |
1750 | if (IS_ERR(dentry)) |
1751 | return dentry; |
1752 | } |
1753 | |
1754 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1755 | int status = d_revalidate(dentry, nd); |
1756 | if (unlikely(status <= 0)) { |
1757 | /* |
1758 | * The dentry failed validation. |
1759 | * If d_revalidate returned 0 attempt to invalidate |
1760 | * the dentry otherwise d_revalidate is asking us |
1761 | * to return a fail status. |
1762 | */ |
1763 | if (status < 0) { |
1764 | dput(dentry); |
1765 | return ERR_PTR(status); |
1766 | } else if (!d_invalidate(dentry)) { |
1767 | dput(dentry); |
1768 | dentry = NULL; |
1769 | } |
1770 | } |
1771 | } |
1772 | |
1773 | if (!dentry) |
1774 | dentry = d_alloc_and_lookup(base, name, nd); |
1775 | |
1776 | return dentry; |
1777 | } |
1778 | |
1779 | /* |
1780 | * Restricted form of lookup. Doesn't follow links, single-component only, |
1781 | * needs parent already locked. Doesn't follow mounts. |
1782 | * SMP-safe. |
1783 | */ |
1784 | static struct dentry *lookup_hash(struct nameidata *nd) |
1785 | { |
1786 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1787 | } |
1788 | |
1789 | /** |
1790 | * lookup_one_len - filesystem helper to lookup single pathname component |
1791 | * @name: pathname component to lookup |
1792 | * @base: base directory to lookup from |
1793 | * @len: maximum length @len should be interpreted to |
1794 | * |
1795 | * Note that this routine is purely a helper for filesystem usage and should |
1796 | * not be called by generic code. Also note that by using this function the |
1797 | * nameidata argument is passed to the filesystem methods and a filesystem |
1798 | * using this helper needs to be prepared for that. |
1799 | */ |
1800 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1801 | { |
1802 | struct qstr this; |
1803 | unsigned int c; |
1804 | |
1805 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1806 | |
1807 | this.name = name; |
1808 | this.len = len; |
1809 | this.hash = full_name_hash(name, len); |
1810 | if (!len) |
1811 | return ERR_PTR(-EACCES); |
1812 | |
1813 | while (len--) { |
1814 | c = *(const unsigned char *)name++; |
1815 | if (c == '/' || c == '\0') |
1816 | return ERR_PTR(-EACCES); |
1817 | } |
1818 | /* |
1819 | * See if the low-level filesystem might want |
1820 | * to use its own hash.. |
1821 | */ |
1822 | if (base->d_flags & DCACHE_OP_HASH) { |
1823 | int err = base->d_op->d_hash(base, base->d_inode, &this); |
1824 | if (err < 0) |
1825 | return ERR_PTR(err); |
1826 | } |
1827 | |
1828 | return __lookup_hash(&this, base, NULL); |
1829 | } |
1830 | |
1831 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, |
1832 | struct path *path, int *empty) |
1833 | { |
1834 | struct nameidata nd; |
1835 | char *tmp = getname_flags(name, flags, empty); |
1836 | int err = PTR_ERR(tmp); |
1837 | if (!IS_ERR(tmp)) { |
1838 | |
1839 | BUG_ON(flags & LOOKUP_PARENT); |
1840 | |
1841 | err = do_path_lookup(dfd, tmp, flags, &nd); |
1842 | putname(tmp); |
1843 | if (!err) |
1844 | *path = nd.path; |
1845 | } |
1846 | return err; |
1847 | } |
1848 | |
1849 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1850 | struct path *path) |
1851 | { |
1852 | return user_path_at_empty(dfd, name, flags, path, 0); |
1853 | } |
1854 | |
1855 | static int user_path_parent(int dfd, const char __user *path, |
1856 | struct nameidata *nd, char **name) |
1857 | { |
1858 | char *s = getname(path); |
1859 | int error; |
1860 | |
1861 | if (IS_ERR(s)) |
1862 | return PTR_ERR(s); |
1863 | |
1864 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); |
1865 | if (error) |
1866 | putname(s); |
1867 | else |
1868 | *name = s; |
1869 | |
1870 | return error; |
1871 | } |
1872 | |
1873 | /* |
1874 | * It's inline, so penalty for filesystems that don't use sticky bit is |
1875 | * minimal. |
1876 | */ |
1877 | static inline int check_sticky(struct inode *dir, struct inode *inode) |
1878 | { |
1879 | uid_t fsuid = current_fsuid(); |
1880 | |
1881 | if (!(dir->i_mode & S_ISVTX)) |
1882 | return 0; |
1883 | if (current_user_ns() != inode_userns(inode)) |
1884 | goto other_userns; |
1885 | if (inode->i_uid == fsuid) |
1886 | return 0; |
1887 | if (dir->i_uid == fsuid) |
1888 | return 0; |
1889 | |
1890 | other_userns: |
1891 | return !ns_capable(inode_userns(inode), CAP_FOWNER); |
1892 | } |
1893 | |
1894 | /* |
1895 | * Check whether we can remove a link victim from directory dir, check |
1896 | * whether the type of victim is right. |
1897 | * 1. We can't do it if dir is read-only (done in permission()) |
1898 | * 2. We should have write and exec permissions on dir |
1899 | * 3. We can't remove anything from append-only dir |
1900 | * 4. We can't do anything with immutable dir (done in permission()) |
1901 | * 5. If the sticky bit on dir is set we should either |
1902 | * a. be owner of dir, or |
1903 | * b. be owner of victim, or |
1904 | * c. have CAP_FOWNER capability |
1905 | * 6. If the victim is append-only or immutable we can't do antyhing with |
1906 | * links pointing to it. |
1907 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. |
1908 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. |
1909 | * 9. We can't remove a root or mountpoint. |
1910 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by |
1911 | * nfs_async_unlink(). |
1912 | */ |
1913 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1914 | { |
1915 | int error; |
1916 | |
1917 | if (!victim->d_inode) |
1918 | return -ENOENT; |
1919 | |
1920 | BUG_ON(victim->d_parent->d_inode != dir); |
1921 | audit_inode_child(victim, dir); |
1922 | |
1923 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1924 | if (error) |
1925 | return error; |
1926 | if (IS_APPEND(dir)) |
1927 | return -EPERM; |
1928 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| |
1929 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1930 | return -EPERM; |
1931 | if (isdir) { |
1932 | if (!S_ISDIR(victim->d_inode->i_mode)) |
1933 | return -ENOTDIR; |
1934 | if (IS_ROOT(victim)) |
1935 | return -EBUSY; |
1936 | } else if (S_ISDIR(victim->d_inode->i_mode)) |
1937 | return -EISDIR; |
1938 | if (IS_DEADDIR(dir)) |
1939 | return -ENOENT; |
1940 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) |
1941 | return -EBUSY; |
1942 | return 0; |
1943 | } |
1944 | |
1945 | /* Check whether we can create an object with dentry child in directory |
1946 | * dir. |
1947 | * 1. We can't do it if child already exists (open has special treatment for |
1948 | * this case, but since we are inlined it's OK) |
1949 | * 2. We can't do it if dir is read-only (done in permission()) |
1950 | * 3. We should have write and exec permissions on dir |
1951 | * 4. We can't do it if dir is immutable (done in permission()) |
1952 | */ |
1953 | static inline int may_create(struct inode *dir, struct dentry *child) |
1954 | { |
1955 | if (child->d_inode) |
1956 | return -EEXIST; |
1957 | if (IS_DEADDIR(dir)) |
1958 | return -ENOENT; |
1959 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1960 | } |
1961 | |
1962 | /* |
1963 | * p1 and p2 should be directories on the same fs. |
1964 | */ |
1965 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) |
1966 | { |
1967 | struct dentry *p; |
1968 | |
1969 | if (p1 == p2) { |
1970 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1971 | return NULL; |
1972 | } |
1973 | |
1974 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
1975 | |
1976 | p = d_ancestor(p2, p1); |
1977 | if (p) { |
1978 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); |
1979 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); |
1980 | return p; |
1981 | } |
1982 | |
1983 | p = d_ancestor(p1, p2); |
1984 | if (p) { |
1985 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1986 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
1987 | return p; |
1988 | } |
1989 | |
1990 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
1991 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
1992 | return NULL; |
1993 | } |
1994 | |
1995 | void unlock_rename(struct dentry *p1, struct dentry *p2) |
1996 | { |
1997 | mutex_unlock(&p1->d_inode->i_mutex); |
1998 | if (p1 != p2) { |
1999 | mutex_unlock(&p2->d_inode->i_mutex); |
2000 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
2001 | } |
2002 | } |
2003 | |
2004 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
2005 | struct nameidata *nd) |
2006 | { |
2007 | int error = may_create(dir, dentry); |
2008 | |
2009 | if (error) |
2010 | return error; |
2011 | |
2012 | if (!dir->i_op->create) |
2013 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2014 | mode &= S_IALLUGO; |
2015 | mode |= S_IFREG; |
2016 | error = security_inode_create(dir, dentry, mode); |
2017 | if (error) |
2018 | return error; |
2019 | error = dir->i_op->create(dir, dentry, mode, nd); |
2020 | if (!error) |
2021 | fsnotify_create(dir, dentry); |
2022 | return error; |
2023 | } |
2024 | |
2025 | static int may_open(struct path *path, int acc_mode, int flag) |
2026 | { |
2027 | struct dentry *dentry = path->dentry; |
2028 | struct inode *inode = dentry->d_inode; |
2029 | int error; |
2030 | |
2031 | /* O_PATH? */ |
2032 | if (!acc_mode) |
2033 | return 0; |
2034 | |
2035 | if (!inode) |
2036 | return -ENOENT; |
2037 | |
2038 | switch (inode->i_mode & S_IFMT) { |
2039 | case S_IFLNK: |
2040 | return -ELOOP; |
2041 | case S_IFDIR: |
2042 | if (acc_mode & MAY_WRITE) |
2043 | return -EISDIR; |
2044 | break; |
2045 | case S_IFBLK: |
2046 | case S_IFCHR: |
2047 | if (path->mnt->mnt_flags & MNT_NODEV) |
2048 | return -EACCES; |
2049 | /*FALLTHRU*/ |
2050 | case S_IFIFO: |
2051 | case S_IFSOCK: |
2052 | flag &= ~O_TRUNC; |
2053 | break; |
2054 | } |
2055 | |
2056 | error = inode_permission(inode, acc_mode); |
2057 | if (error) |
2058 | return error; |
2059 | |
2060 | /* |
2061 | * An append-only file must be opened in append mode for writing. |
2062 | */ |
2063 | if (IS_APPEND(inode)) { |
2064 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
2065 | return -EPERM; |
2066 | if (flag & O_TRUNC) |
2067 | return -EPERM; |
2068 | } |
2069 | |
2070 | /* O_NOATIME can only be set by the owner or superuser */ |
2071 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) |
2072 | return -EPERM; |
2073 | |
2074 | return 0; |
2075 | } |
2076 | |
2077 | static int handle_truncate(struct file *filp) |
2078 | { |
2079 | struct path *path = &filp->f_path; |
2080 | struct inode *inode = path->dentry->d_inode; |
2081 | int error = get_write_access(inode); |
2082 | if (error) |
2083 | return error; |
2084 | /* |
2085 | * Refuse to truncate files with mandatory locks held on them. |
2086 | */ |
2087 | error = locks_verify_locked(inode); |
2088 | if (!error) |
2089 | error = security_path_truncate(path); |
2090 | if (!error) { |
2091 | error = do_truncate(path->dentry, 0, |
2092 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, |
2093 | filp); |
2094 | } |
2095 | put_write_access(inode); |
2096 | return error; |
2097 | } |
2098 | |
2099 | static inline int open_to_namei_flags(int flag) |
2100 | { |
2101 | if ((flag & O_ACCMODE) == 3) |
2102 | flag--; |
2103 | return flag; |
2104 | } |
2105 | |
2106 | /* |
2107 | * Handle the last step of open() |
2108 | */ |
2109 | static struct file *do_last(struct nameidata *nd, struct path *path, |
2110 | const struct open_flags *op, const char *pathname) |
2111 | { |
2112 | struct dentry *dir = nd->path.dentry; |
2113 | struct dentry *dentry; |
2114 | int open_flag = op->open_flag; |
2115 | int will_truncate = open_flag & O_TRUNC; |
2116 | int want_write = 0; |
2117 | int acc_mode = op->acc_mode; |
2118 | struct file *filp; |
2119 | int error; |
2120 | |
2121 | nd->flags &= ~LOOKUP_PARENT; |
2122 | nd->flags |= op->intent; |
2123 | |
2124 | switch (nd->last_type) { |
2125 | case LAST_DOTDOT: |
2126 | case LAST_DOT: |
2127 | error = handle_dots(nd, nd->last_type); |
2128 | if (error) |
2129 | return ERR_PTR(error); |
2130 | /* fallthrough */ |
2131 | case LAST_ROOT: |
2132 | error = complete_walk(nd); |
2133 | if (error) |
2134 | return ERR_PTR(error); |
2135 | audit_inode(pathname, nd->path.dentry); |
2136 | if (open_flag & O_CREAT) { |
2137 | error = -EISDIR; |
2138 | goto exit; |
2139 | } |
2140 | goto ok; |
2141 | case LAST_BIND: |
2142 | error = complete_walk(nd); |
2143 | if (error) |
2144 | return ERR_PTR(error); |
2145 | audit_inode(pathname, dir); |
2146 | goto ok; |
2147 | } |
2148 | |
2149 | if (!(open_flag & O_CREAT)) { |
2150 | int symlink_ok = 0; |
2151 | if (nd->last.name[nd->last.len]) |
2152 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
2153 | if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) |
2154 | symlink_ok = 1; |
2155 | /* we _can_ be in RCU mode here */ |
2156 | error = walk_component(nd, path, &nd->last, LAST_NORM, |
2157 | !symlink_ok); |
2158 | if (error < 0) |
2159 | return ERR_PTR(error); |
2160 | if (error) /* symlink */ |
2161 | return NULL; |
2162 | /* sayonara */ |
2163 | error = complete_walk(nd); |
2164 | if (error) |
2165 | return ERR_PTR(error); |
2166 | |
2167 | error = -ENOTDIR; |
2168 | if (nd->flags & LOOKUP_DIRECTORY) { |
2169 | if (!nd->inode->i_op->lookup) |
2170 | goto exit; |
2171 | } |
2172 | audit_inode(pathname, nd->path.dentry); |
2173 | goto ok; |
2174 | } |
2175 | |
2176 | /* create side of things */ |
2177 | /* |
2178 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been |
2179 | * cleared when we got to the last component we are about to look up |
2180 | */ |
2181 | error = complete_walk(nd); |
2182 | if (error) |
2183 | return ERR_PTR(error); |
2184 | |
2185 | audit_inode(pathname, dir); |
2186 | error = -EISDIR; |
2187 | /* trailing slashes? */ |
2188 | if (nd->last.name[nd->last.len]) |
2189 | goto exit; |
2190 | |
2191 | mutex_lock(&dir->d_inode->i_mutex); |
2192 | |
2193 | dentry = lookup_hash(nd); |
2194 | error = PTR_ERR(dentry); |
2195 | if (IS_ERR(dentry)) { |
2196 | mutex_unlock(&dir->d_inode->i_mutex); |
2197 | goto exit; |
2198 | } |
2199 | |
2200 | path->dentry = dentry; |
2201 | path->mnt = nd->path.mnt; |
2202 | |
2203 | /* Negative dentry, just create the file */ |
2204 | if (!dentry->d_inode) { |
2205 | umode_t mode = op->mode; |
2206 | if (!IS_POSIXACL(dir->d_inode)) |
2207 | mode &= ~current_umask(); |
2208 | /* |
2209 | * This write is needed to ensure that a |
2210 | * rw->ro transition does not occur between |
2211 | * the time when the file is created and when |
2212 | * a permanent write count is taken through |
2213 | * the 'struct file' in nameidata_to_filp(). |
2214 | */ |
2215 | error = mnt_want_write(nd->path.mnt); |
2216 | if (error) |
2217 | goto exit_mutex_unlock; |
2218 | want_write = 1; |
2219 | /* Don't check for write permission, don't truncate */ |
2220 | open_flag &= ~O_TRUNC; |
2221 | will_truncate = 0; |
2222 | acc_mode = MAY_OPEN; |
2223 | error = security_path_mknod(&nd->path, dentry, mode, 0); |
2224 | if (error) |
2225 | goto exit_mutex_unlock; |
2226 | error = vfs_create(dir->d_inode, dentry, mode, nd); |
2227 | if (error) |
2228 | goto exit_mutex_unlock; |
2229 | mutex_unlock(&dir->d_inode->i_mutex); |
2230 | dput(nd->path.dentry); |
2231 | nd->path.dentry = dentry; |
2232 | goto common; |
2233 | } |
2234 | |
2235 | /* |
2236 | * It already exists. |
2237 | */ |
2238 | mutex_unlock(&dir->d_inode->i_mutex); |
2239 | audit_inode(pathname, path->dentry); |
2240 | |
2241 | error = -EEXIST; |
2242 | if (open_flag & O_EXCL) |
2243 | goto exit_dput; |
2244 | |
2245 | error = follow_managed(path, nd->flags); |
2246 | if (error < 0) |
2247 | goto exit_dput; |
2248 | |
2249 | if (error) |
2250 | nd->flags |= LOOKUP_JUMPED; |
2251 | |
2252 | error = -ENOENT; |
2253 | if (!path->dentry->d_inode) |
2254 | goto exit_dput; |
2255 | |
2256 | if (path->dentry->d_inode->i_op->follow_link) |
2257 | return NULL; |
2258 | |
2259 | path_to_nameidata(path, nd); |
2260 | nd->inode = path->dentry->d_inode; |
2261 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ |
2262 | error = complete_walk(nd); |
2263 | if (error) |
2264 | return ERR_PTR(error); |
2265 | error = -EISDIR; |
2266 | if (S_ISDIR(nd->inode->i_mode)) |
2267 | goto exit; |
2268 | ok: |
2269 | if (!S_ISREG(nd->inode->i_mode)) |
2270 | will_truncate = 0; |
2271 | |
2272 | if (will_truncate) { |
2273 | error = mnt_want_write(nd->path.mnt); |
2274 | if (error) |
2275 | goto exit; |
2276 | want_write = 1; |
2277 | } |
2278 | common: |
2279 | error = may_open(&nd->path, acc_mode, open_flag); |
2280 | if (error) |
2281 | goto exit; |
2282 | filp = nameidata_to_filp(nd); |
2283 | if (!IS_ERR(filp)) { |
2284 | error = ima_file_check(filp, op->acc_mode); |
2285 | if (error) { |
2286 | fput(filp); |
2287 | filp = ERR_PTR(error); |
2288 | } |
2289 | } |
2290 | if (!IS_ERR(filp)) { |
2291 | if (will_truncate) { |
2292 | error = handle_truncate(filp); |
2293 | if (error) { |
2294 | fput(filp); |
2295 | filp = ERR_PTR(error); |
2296 | } |
2297 | } |
2298 | } |
2299 | out: |
2300 | if (want_write) |
2301 | mnt_drop_write(nd->path.mnt); |
2302 | path_put(&nd->path); |
2303 | return filp; |
2304 | |
2305 | exit_mutex_unlock: |
2306 | mutex_unlock(&dir->d_inode->i_mutex); |
2307 | exit_dput: |
2308 | path_put_conditional(path, nd); |
2309 | exit: |
2310 | filp = ERR_PTR(error); |
2311 | goto out; |
2312 | } |
2313 | |
2314 | static struct file *path_openat(int dfd, const char *pathname, |
2315 | struct nameidata *nd, const struct open_flags *op, int flags) |
2316 | { |
2317 | struct file *base = NULL; |
2318 | struct file *filp; |
2319 | struct path path; |
2320 | int error; |
2321 | |
2322 | filp = get_empty_filp(); |
2323 | if (!filp) |
2324 | return ERR_PTR(-ENFILE); |
2325 | |
2326 | filp->f_flags = op->open_flag; |
2327 | nd->intent.open.file = filp; |
2328 | nd->intent.open.flags = open_to_namei_flags(op->open_flag); |
2329 | nd->intent.open.create_mode = op->mode; |
2330 | |
2331 | error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); |
2332 | if (unlikely(error)) |
2333 | goto out_filp; |
2334 | |
2335 | current->total_link_count = 0; |
2336 | error = link_path_walk(pathname, nd); |
2337 | if (unlikely(error)) |
2338 | goto out_filp; |
2339 | |
2340 | filp = do_last(nd, &path, op, pathname); |
2341 | while (unlikely(!filp)) { /* trailing symlink */ |
2342 | struct path link = path; |
2343 | void *cookie; |
2344 | if (!(nd->flags & LOOKUP_FOLLOW)) { |
2345 | path_put_conditional(&path, nd); |
2346 | path_put(&nd->path); |
2347 | filp = ERR_PTR(-ELOOP); |
2348 | break; |
2349 | } |
2350 | nd->flags |= LOOKUP_PARENT; |
2351 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); |
2352 | error = follow_link(&link, nd, &cookie); |
2353 | if (unlikely(error)) |
2354 | filp = ERR_PTR(error); |
2355 | else |
2356 | filp = do_last(nd, &path, op, pathname); |
2357 | put_link(nd, &link, cookie); |
2358 | } |
2359 | out: |
2360 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) |
2361 | path_put(&nd->root); |
2362 | if (base) |
2363 | fput(base); |
2364 | release_open_intent(nd); |
2365 | return filp; |
2366 | |
2367 | out_filp: |
2368 | filp = ERR_PTR(error); |
2369 | goto out; |
2370 | } |
2371 | |
2372 | struct file *do_filp_open(int dfd, const char *pathname, |
2373 | const struct open_flags *op, int flags) |
2374 | { |
2375 | struct nameidata nd; |
2376 | struct file *filp; |
2377 | |
2378 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); |
2379 | if (unlikely(filp == ERR_PTR(-ECHILD))) |
2380 | filp = path_openat(dfd, pathname, &nd, op, flags); |
2381 | if (unlikely(filp == ERR_PTR(-ESTALE))) |
2382 | filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); |
2383 | return filp; |
2384 | } |
2385 | |
2386 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, |
2387 | const char *name, const struct open_flags *op, int flags) |
2388 | { |
2389 | struct nameidata nd; |
2390 | struct file *file; |
2391 | |
2392 | nd.root.mnt = mnt; |
2393 | nd.root.dentry = dentry; |
2394 | |
2395 | flags |= LOOKUP_ROOT; |
2396 | |
2397 | if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) |
2398 | return ERR_PTR(-ELOOP); |
2399 | |
2400 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); |
2401 | if (unlikely(file == ERR_PTR(-ECHILD))) |
2402 | file = path_openat(-1, name, &nd, op, flags); |
2403 | if (unlikely(file == ERR_PTR(-ESTALE))) |
2404 | file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); |
2405 | return file; |
2406 | } |
2407 | |
2408 | struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) |
2409 | { |
2410 | struct dentry *dentry = ERR_PTR(-EEXIST); |
2411 | struct nameidata nd; |
2412 | int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); |
2413 | if (error) |
2414 | return ERR_PTR(error); |
2415 | |
2416 | /* |
2417 | * Yucky last component or no last component at all? |
2418 | * (foo/., foo/.., /////) |
2419 | */ |
2420 | if (nd.last_type != LAST_NORM) |
2421 | goto out; |
2422 | nd.flags &= ~LOOKUP_PARENT; |
2423 | nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
2424 | nd.intent.open.flags = O_EXCL; |
2425 | |
2426 | /* |
2427 | * Do the final lookup. |
2428 | */ |
2429 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2430 | dentry = lookup_hash(&nd); |
2431 | if (IS_ERR(dentry)) |
2432 | goto fail; |
2433 | |
2434 | if (dentry->d_inode) |
2435 | goto eexist; |
2436 | /* |
2437 | * Special case - lookup gave negative, but... we had foo/bar/ |
2438 | * From the vfs_mknod() POV we just have a negative dentry - |
2439 | * all is fine. Let's be bastards - you had / on the end, you've |
2440 | * been asking for (non-existent) directory. -ENOENT for you. |
2441 | */ |
2442 | if (unlikely(!is_dir && nd.last.name[nd.last.len])) { |
2443 | dput(dentry); |
2444 | dentry = ERR_PTR(-ENOENT); |
2445 | goto fail; |
2446 | } |
2447 | *path = nd.path; |
2448 | return dentry; |
2449 | eexist: |
2450 | dput(dentry); |
2451 | dentry = ERR_PTR(-EEXIST); |
2452 | fail: |
2453 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2454 | out: |
2455 | path_put(&nd.path); |
2456 | return dentry; |
2457 | } |
2458 | EXPORT_SYMBOL(kern_path_create); |
2459 | |
2460 | struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) |
2461 | { |
2462 | char *tmp = getname(pathname); |
2463 | struct dentry *res; |
2464 | if (IS_ERR(tmp)) |
2465 | return ERR_CAST(tmp); |
2466 | res = kern_path_create(dfd, tmp, path, is_dir); |
2467 | putname(tmp); |
2468 | return res; |
2469 | } |
2470 | EXPORT_SYMBOL(user_path_create); |
2471 | |
2472 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
2473 | { |
2474 | int error = may_create(dir, dentry); |
2475 | |
2476 | if (error) |
2477 | return error; |
2478 | |
2479 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && |
2480 | !ns_capable(inode_userns(dir), CAP_MKNOD)) |
2481 | return -EPERM; |
2482 | |
2483 | if (!dir->i_op->mknod) |
2484 | return -EPERM; |
2485 | |
2486 | error = devcgroup_inode_mknod(mode, dev); |
2487 | if (error) |
2488 | return error; |
2489 | |
2490 | error = security_inode_mknod(dir, dentry, mode, dev); |
2491 | if (error) |
2492 | return error; |
2493 | |
2494 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
2495 | if (!error) |
2496 | fsnotify_create(dir, dentry); |
2497 | return error; |
2498 | } |
2499 | |
2500 | static int may_mknod(umode_t mode) |
2501 | { |
2502 | switch (mode & S_IFMT) { |
2503 | case S_IFREG: |
2504 | case S_IFCHR: |
2505 | case S_IFBLK: |
2506 | case S_IFIFO: |
2507 | case S_IFSOCK: |
2508 | case 0: /* zero mode translates to S_IFREG */ |
2509 | return 0; |
2510 | case S_IFDIR: |
2511 | return -EPERM; |
2512 | default: |
2513 | return -EINVAL; |
2514 | } |
2515 | } |
2516 | |
2517 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, |
2518 | unsigned, dev) |
2519 | { |
2520 | struct dentry *dentry; |
2521 | struct path path; |
2522 | int error; |
2523 | |
2524 | if (S_ISDIR(mode)) |
2525 | return -EPERM; |
2526 | |
2527 | dentry = user_path_create(dfd, filename, &path, 0); |
2528 | if (IS_ERR(dentry)) |
2529 | return PTR_ERR(dentry); |
2530 | |
2531 | if (!IS_POSIXACL(path.dentry->d_inode)) |
2532 | mode &= ~current_umask(); |
2533 | error = may_mknod(mode); |
2534 | if (error) |
2535 | goto out_dput; |
2536 | error = mnt_want_write(path.mnt); |
2537 | if (error) |
2538 | goto out_dput; |
2539 | error = security_path_mknod(&path, dentry, mode, dev); |
2540 | if (error) |
2541 | goto out_drop_write; |
2542 | switch (mode & S_IFMT) { |
2543 | case 0: case S_IFREG: |
2544 | error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); |
2545 | break; |
2546 | case S_IFCHR: case S_IFBLK: |
2547 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, |
2548 | new_decode_dev(dev)); |
2549 | break; |
2550 | case S_IFIFO: case S_IFSOCK: |
2551 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); |
2552 | break; |
2553 | } |
2554 | out_drop_write: |
2555 | mnt_drop_write(path.mnt); |
2556 | out_dput: |
2557 | dput(dentry); |
2558 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2559 | path_put(&path); |
2560 | |
2561 | return error; |
2562 | } |
2563 | |
2564 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) |
2565 | { |
2566 | return sys_mknodat(AT_FDCWD, filename, mode, dev); |
2567 | } |
2568 | |
2569 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
2570 | { |
2571 | int error = may_create(dir, dentry); |
2572 | |
2573 | if (error) |
2574 | return error; |
2575 | |
2576 | if (!dir->i_op->mkdir) |
2577 | return -EPERM; |
2578 | |
2579 | mode &= (S_IRWXUGO|S_ISVTX); |
2580 | error = security_inode_mkdir(dir, dentry, mode); |
2581 | if (error) |
2582 | return error; |
2583 | |
2584 | error = dir->i_op->mkdir(dir, dentry, mode); |
2585 | if (!error) |
2586 | fsnotify_mkdir(dir, dentry); |
2587 | return error; |
2588 | } |
2589 | |
2590 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) |
2591 | { |
2592 | struct dentry *dentry; |
2593 | struct path path; |
2594 | int error; |
2595 | |
2596 | dentry = user_path_create(dfd, pathname, &path, 1); |
2597 | if (IS_ERR(dentry)) |
2598 | return PTR_ERR(dentry); |
2599 | |
2600 | if (!IS_POSIXACL(path.dentry->d_inode)) |
2601 | mode &= ~current_umask(); |
2602 | error = mnt_want_write(path.mnt); |
2603 | if (error) |
2604 | goto out_dput; |
2605 | error = security_path_mkdir(&path, dentry, mode); |
2606 | if (error) |
2607 | goto out_drop_write; |
2608 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); |
2609 | out_drop_write: |
2610 | mnt_drop_write(path.mnt); |
2611 | out_dput: |
2612 | dput(dentry); |
2613 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2614 | path_put(&path); |
2615 | return error; |
2616 | } |
2617 | |
2618 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) |
2619 | { |
2620 | return sys_mkdirat(AT_FDCWD, pathname, mode); |
2621 | } |
2622 | |
2623 | /* |
2624 | * The dentry_unhash() helper will try to drop the dentry early: we |
2625 | * should have a usage count of 2 if we're the only user of this |
2626 | * dentry, and if that is true (possibly after pruning the dcache), |
2627 | * then we drop the dentry now. |
2628 | * |
2629 | * A low-level filesystem can, if it choses, legally |
2630 | * do a |
2631 | * |
2632 | * if (!d_unhashed(dentry)) |
2633 | * return -EBUSY; |
2634 | * |
2635 | * if it cannot handle the case of removing a directory |
2636 | * that is still in use by something else.. |
2637 | */ |
2638 | void dentry_unhash(struct dentry *dentry) |
2639 | { |
2640 | shrink_dcache_parent(dentry); |
2641 | spin_lock(&dentry->d_lock); |
2642 | if (dentry->d_count == 1) |
2643 | __d_drop(dentry); |
2644 | spin_unlock(&dentry->d_lock); |
2645 | } |
2646 | |
2647 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
2648 | { |
2649 | int error = may_delete(dir, dentry, 1); |
2650 | |
2651 | if (error) |
2652 | return error; |
2653 | |
2654 | if (!dir->i_op->rmdir) |
2655 | return -EPERM; |
2656 | |
2657 | dget(dentry); |
2658 | mutex_lock(&dentry->d_inode->i_mutex); |
2659 | |
2660 | error = -EBUSY; |
2661 | if (d_mountpoint(dentry)) |
2662 | goto out; |
2663 | |
2664 | error = security_inode_rmdir(dir, dentry); |
2665 | if (error) |
2666 | goto out; |
2667 | |
2668 | shrink_dcache_parent(dentry); |
2669 | error = dir->i_op->rmdir(dir, dentry); |
2670 | if (error) |
2671 | goto out; |
2672 | |
2673 | dentry->d_inode->i_flags |= S_DEAD; |
2674 | dont_mount(dentry); |
2675 | |
2676 | out: |
2677 | mutex_unlock(&dentry->d_inode->i_mutex); |
2678 | dput(dentry); |
2679 | if (!error) |
2680 | d_delete(dentry); |
2681 | return error; |
2682 | } |
2683 | |
2684 | static long do_rmdir(int dfd, const char __user *pathname) |
2685 | { |
2686 | int error = 0; |
2687 | char * name; |
2688 | struct dentry *dentry; |
2689 | struct nameidata nd; |
2690 | |
2691 | error = user_path_parent(dfd, pathname, &nd, &name); |
2692 | if (error) |
2693 | return error; |
2694 | |
2695 | switch(nd.last_type) { |
2696 | case LAST_DOTDOT: |
2697 | error = -ENOTEMPTY; |
2698 | goto exit1; |
2699 | case LAST_DOT: |
2700 | error = -EINVAL; |
2701 | goto exit1; |
2702 | case LAST_ROOT: |
2703 | error = -EBUSY; |
2704 | goto exit1; |
2705 | } |
2706 | |
2707 | nd.flags &= ~LOOKUP_PARENT; |
2708 | |
2709 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2710 | dentry = lookup_hash(&nd); |
2711 | error = PTR_ERR(dentry); |
2712 | if (IS_ERR(dentry)) |
2713 | goto exit2; |
2714 | if (!dentry->d_inode) { |
2715 | error = -ENOENT; |
2716 | goto exit3; |
2717 | } |
2718 | error = mnt_want_write(nd.path.mnt); |
2719 | if (error) |
2720 | goto exit3; |
2721 | error = security_path_rmdir(&nd.path, dentry); |
2722 | if (error) |
2723 | goto exit4; |
2724 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
2725 | exit4: |
2726 | mnt_drop_write(nd.path.mnt); |
2727 | exit3: |
2728 | dput(dentry); |
2729 | exit2: |
2730 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2731 | exit1: |
2732 | path_put(&nd.path); |
2733 | putname(name); |
2734 | return error; |
2735 | } |
2736 | |
2737 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
2738 | { |
2739 | return do_rmdir(AT_FDCWD, pathname); |
2740 | } |
2741 | |
2742 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2743 | { |
2744 | int error = may_delete(dir, dentry, 0); |
2745 | |
2746 | if (error) |
2747 | return error; |
2748 | |
2749 | if (!dir->i_op->unlink) |
2750 | return -EPERM; |
2751 | |
2752 | mutex_lock(&dentry->d_inode->i_mutex); |
2753 | if (d_mountpoint(dentry)) |
2754 | error = -EBUSY; |
2755 | else { |
2756 | error = security_inode_unlink(dir, dentry); |
2757 | if (!error) { |
2758 | error = dir->i_op->unlink(dir, dentry); |
2759 | if (!error) |
2760 | dont_mount(dentry); |
2761 | } |
2762 | } |
2763 | mutex_unlock(&dentry->d_inode->i_mutex); |
2764 | |
2765 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ |
2766 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { |
2767 | fsnotify_link_count(dentry->d_inode); |
2768 | d_delete(dentry); |
2769 | } |
2770 | |
2771 | return error; |
2772 | } |
2773 | |
2774 | /* |
2775 | * Make sure that the actual truncation of the file will occur outside its |
2776 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
2777 | * writeout happening, and we don't want to prevent access to the directory |
2778 | * while waiting on the I/O. |
2779 | */ |
2780 | static long do_unlinkat(int dfd, const char __user *pathname) |
2781 | { |
2782 | int error; |
2783 | char *name; |
2784 | struct dentry *dentry; |
2785 | struct nameidata nd; |
2786 | struct inode *inode = NULL; |
2787 | |
2788 | error = user_path_parent(dfd, pathname, &nd, &name); |
2789 | if (error) |
2790 | return error; |
2791 | |
2792 | error = -EISDIR; |
2793 | if (nd.last_type != LAST_NORM) |
2794 | goto exit1; |
2795 | |
2796 | nd.flags &= ~LOOKUP_PARENT; |
2797 | |
2798 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2799 | dentry = lookup_hash(&nd); |
2800 | error = PTR_ERR(dentry); |
2801 | if (!IS_ERR(dentry)) { |
2802 | /* Why not before? Because we want correct error value */ |
2803 | if (nd.last.name[nd.last.len]) |
2804 | goto slashes; |
2805 | inode = dentry->d_inode; |
2806 | if (!inode) |
2807 | goto slashes; |
2808 | ihold(inode); |
2809 | error = mnt_want_write(nd.path.mnt); |
2810 | if (error) |
2811 | goto exit2; |
2812 | error = security_path_unlink(&nd.path, dentry); |
2813 | if (error) |
2814 | goto exit3; |
2815 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
2816 | exit3: |
2817 | mnt_drop_write(nd.path.mnt); |
2818 | exit2: |
2819 | dput(dentry); |
2820 | } |
2821 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2822 | if (inode) |
2823 | iput(inode); /* truncate the inode here */ |
2824 | exit1: |
2825 | path_put(&nd.path); |
2826 | putname(name); |
2827 | return error; |
2828 | |
2829 | slashes: |
2830 | error = !dentry->d_inode ? -ENOENT : |
2831 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; |
2832 | goto exit2; |
2833 | } |
2834 | |
2835 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
2836 | { |
2837 | if ((flag & ~AT_REMOVEDIR) != 0) |
2838 | return -EINVAL; |
2839 | |
2840 | if (flag & AT_REMOVEDIR) |
2841 | return do_rmdir(dfd, pathname); |
2842 | |
2843 | return do_unlinkat(dfd, pathname); |
2844 | } |
2845 | |
2846 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
2847 | { |
2848 | return do_unlinkat(AT_FDCWD, pathname); |
2849 | } |
2850 | |
2851 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
2852 | { |
2853 | int error = may_create(dir, dentry); |
2854 | |
2855 | if (error) |
2856 | return error; |
2857 | |
2858 | if (!dir->i_op->symlink) |
2859 | return -EPERM; |
2860 | |
2861 | error = security_inode_symlink(dir, dentry, oldname); |
2862 | if (error) |
2863 | return error; |
2864 | |
2865 | error = dir->i_op->symlink(dir, dentry, oldname); |
2866 | if (!error) |
2867 | fsnotify_create(dir, dentry); |
2868 | return error; |
2869 | } |
2870 | |
2871 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
2872 | int, newdfd, const char __user *, newname) |
2873 | { |
2874 | int error; |
2875 | char *from; |
2876 | struct dentry *dentry; |
2877 | struct path path; |
2878 | |
2879 | from = getname(oldname); |
2880 | if (IS_ERR(from)) |
2881 | return PTR_ERR(from); |
2882 | |
2883 | dentry = user_path_create(newdfd, newname, &path, 0); |
2884 | error = PTR_ERR(dentry); |
2885 | if (IS_ERR(dentry)) |
2886 | goto out_putname; |
2887 | |
2888 | error = mnt_want_write(path.mnt); |
2889 | if (error) |
2890 | goto out_dput; |
2891 | error = security_path_symlink(&path, dentry, from); |
2892 | if (error) |
2893 | goto out_drop_write; |
2894 | error = vfs_symlink(path.dentry->d_inode, dentry, from); |
2895 | out_drop_write: |
2896 | mnt_drop_write(path.mnt); |
2897 | out_dput: |
2898 | dput(dentry); |
2899 | mutex_unlock(&path.dentry->d_inode->i_mutex); |
2900 | path_put(&path); |
2901 | out_putname: |
2902 | putname(from); |
2903 | return error; |
2904 | } |
2905 | |
2906 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
2907 | { |
2908 | return sys_symlinkat(oldname, AT_FDCWD, newname); |
2909 | } |
2910 | |
2911 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
2912 | { |
2913 | struct inode *inode = old_dentry->d_inode; |
2914 | int error; |
2915 | |
2916 | if (!inode) |
2917 | return -ENOENT; |
2918 | |
2919 | error = may_create(dir, new_dentry); |
2920 | if (error) |
2921 | return error; |
2922 | |
2923 | if (dir->i_sb != inode->i_sb) |
2924 | return -EXDEV; |
2925 | |
2926 | /* |
2927 | * A link to an append-only or immutable file cannot be created. |
2928 | */ |
2929 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) |
2930 | return -EPERM; |
2931 | if (!dir->i_op->link) |
2932 | return -EPERM; |
2933 | if (S_ISDIR(inode->i_mode)) |
2934 | return -EPERM; |
2935 | |
2936 | error = security_inode_link(old_dentry, dir, new_dentry); |
2937 | if (error) |
2938 | return error; |
2939 | |
2940 | mutex_lock(&inode->i_mutex); |
2941 | /* Make sure we don't allow creating hardlink to an unlinked file */ |
2942 | if (inode->i_nlink == 0) |
2943 | error = -ENOENT; |
2944 | else |
2945 | error = dir->i_op->link(old_dentry, dir, new_dentry); |
2946 | mutex_unlock(&inode->i_mutex); |
2947 | if (!error) |
2948 | fsnotify_link(dir, inode, new_dentry); |
2949 | return error; |
2950 | } |
2951 | |
2952 | /* |
2953 | * Hardlinks are often used in delicate situations. We avoid |
2954 | * security-related surprises by not following symlinks on the |
2955 | * newname. --KAB |
2956 | * |
2957 | * We don't follow them on the oldname either to be compatible |
2958 | * with linux 2.0, and to avoid hard-linking to directories |
2959 | * and other special files. --ADM |
2960 | */ |
2961 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
2962 | int, newdfd, const char __user *, newname, int, flags) |
2963 | { |
2964 | struct dentry *new_dentry; |
2965 | struct path old_path, new_path; |
2966 | int how = 0; |
2967 | int error; |
2968 | |
2969 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) |
2970 | return -EINVAL; |
2971 | /* |
2972 | * To use null names we require CAP_DAC_READ_SEARCH |
2973 | * This ensures that not everyone will be able to create |
2974 | * handlink using the passed filedescriptor. |
2975 | */ |
2976 | if (flags & AT_EMPTY_PATH) { |
2977 | if (!capable(CAP_DAC_READ_SEARCH)) |
2978 | return -ENOENT; |
2979 | how = LOOKUP_EMPTY; |
2980 | } |
2981 | |
2982 | if (flags & AT_SYMLINK_FOLLOW) |
2983 | how |= LOOKUP_FOLLOW; |
2984 | |
2985 | error = user_path_at(olddfd, oldname, how, &old_path); |
2986 | if (error) |
2987 | return error; |
2988 | |
2989 | new_dentry = user_path_create(newdfd, newname, &new_path, 0); |
2990 | error = PTR_ERR(new_dentry); |
2991 | if (IS_ERR(new_dentry)) |
2992 | goto out; |
2993 | |
2994 | error = -EXDEV; |
2995 | if (old_path.mnt != new_path.mnt) |
2996 | goto out_dput; |
2997 | error = mnt_want_write(new_path.mnt); |
2998 | if (error) |
2999 | goto out_dput; |
3000 | error = security_path_link(old_path.dentry, &new_path, new_dentry); |
3001 | if (error) |
3002 | goto out_drop_write; |
3003 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); |
3004 | out_drop_write: |
3005 | mnt_drop_write(new_path.mnt); |
3006 | out_dput: |
3007 | dput(new_dentry); |
3008 | mutex_unlock(&new_path.dentry->d_inode->i_mutex); |
3009 | path_put(&new_path); |
3010 | out: |
3011 | path_put(&old_path); |
3012 | |
3013 | return error; |
3014 | } |
3015 | |
3016 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
3017 | { |
3018 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
3019 | } |
3020 | |
3021 | /* |
3022 | * The worst of all namespace operations - renaming directory. "Perverted" |
3023 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... |
3024 | * Problems: |
3025 | * a) we can get into loop creation. Check is done in is_subdir(). |
3026 | * b) race potential - two innocent renames can create a loop together. |
3027 | * That's where 4.4 screws up. Current fix: serialization on |
3028 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
3029 | * story. |
3030 | * c) we have to lock _three_ objects - parents and victim (if it exists). |
3031 | * And that - after we got ->i_mutex on parents (until then we don't know |
3032 | * whether the target exists). Solution: try to be smart with locking |
3033 | * order for inodes. We rely on the fact that tree topology may change |
3034 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
3035 | * move will be locked. Thus we can rank directories by the tree |
3036 | * (ancestors first) and rank all non-directories after them. |
3037 | * That works since everybody except rename does "lock parent, lookup, |
3038 | * lock child" and rename is under ->s_vfs_rename_mutex. |
3039 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3040 | * has no more than 1 dentry. If "hybrid" objects will ever appear, |
3041 | * we'd better make sure that there's no link(2) for them. |
3042 | * d) conversion from fhandle to dentry may come in the wrong moment - when |
3043 | * we are removing the target. Solution: we will have to grab ->i_mutex |
3044 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
3045 | * ->i_mutex on parents, which works but leads to some truly excessive |
3046 | * locking]. |
3047 | */ |
3048 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3049 | struct inode *new_dir, struct dentry *new_dentry) |
3050 | { |
3051 | int error = 0; |
3052 | struct inode *target = new_dentry->d_inode; |
3053 | |
3054 | /* |
3055 | * If we are going to change the parent - check write permissions, |
3056 | * we'll need to flip '..'. |
3057 | */ |
3058 | if (new_dir != old_dir) { |
3059 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
3060 | if (error) |
3061 | return error; |
3062 | } |
3063 | |
3064 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
3065 | if (error) |
3066 | return error; |
3067 | |
3068 | dget(new_dentry); |
3069 | if (target) |
3070 | mutex_lock(&target->i_mutex); |
3071 | |
3072 | error = -EBUSY; |
3073 | if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) |
3074 | goto out; |
3075 | |
3076 | if (target) |
3077 | shrink_dcache_parent(new_dentry); |
3078 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3079 | if (error) |
3080 | goto out; |
3081 | |
3082 | if (target) { |
3083 | target->i_flags |= S_DEAD; |
3084 | dont_mount(new_dentry); |
3085 | } |
3086 | out: |
3087 | if (target) |
3088 | mutex_unlock(&target->i_mutex); |
3089 | dput(new_dentry); |
3090 | if (!error) |
3091 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3092 | d_move(old_dentry,new_dentry); |
3093 | return error; |
3094 | } |
3095 | |
3096 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3097 | struct inode *new_dir, struct dentry *new_dentry) |
3098 | { |
3099 | struct inode *target = new_dentry->d_inode; |
3100 | int error; |
3101 | |
3102 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
3103 | if (error) |
3104 | return error; |
3105 | |
3106 | dget(new_dentry); |
3107 | if (target) |
3108 | mutex_lock(&target->i_mutex); |
3109 | |
3110 | error = -EBUSY; |
3111 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3112 | goto out; |
3113 | |
3114 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3115 | if (error) |
3116 | goto out; |
3117 | |
3118 | if (target) |
3119 | dont_mount(new_dentry); |
3120 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3121 | d_move(old_dentry, new_dentry); |
3122 | out: |
3123 | if (target) |
3124 | mutex_unlock(&target->i_mutex); |
3125 | dput(new_dentry); |
3126 | return error; |
3127 | } |
3128 | |
3129 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
3130 | struct inode *new_dir, struct dentry *new_dentry) |
3131 | { |
3132 | int error; |
3133 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); |
3134 | const unsigned char *old_name; |
3135 | |
3136 | if (old_dentry->d_inode == new_dentry->d_inode) |
3137 | return 0; |
3138 | |
3139 | error = may_delete(old_dir, old_dentry, is_dir); |
3140 | if (error) |
3141 | return error; |
3142 | |
3143 | if (!new_dentry->d_inode) |
3144 | error = may_create(new_dir, new_dentry); |
3145 | else |
3146 | error = may_delete(new_dir, new_dentry, is_dir); |
3147 | if (error) |
3148 | return error; |
3149 | |
3150 | if (!old_dir->i_op->rename) |
3151 | return -EPERM; |
3152 | |
3153 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3154 | |
3155 | if (is_dir) |
3156 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); |
3157 | else |
3158 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); |
3159 | if (!error) |
3160 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
3161 | new_dentry->d_inode, old_dentry); |
3162 | fsnotify_oldname_free(old_name); |
3163 | |
3164 | return error; |
3165 | } |
3166 | |
3167 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3168 | int, newdfd, const char __user *, newname) |
3169 | { |
3170 | struct dentry *old_dir, *new_dir; |
3171 | struct dentry *old_dentry, *new_dentry; |
3172 | struct dentry *trap; |
3173 | struct nameidata oldnd, newnd; |
3174 | char *from; |
3175 | char *to; |
3176 | int error; |
3177 | |
3178 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
3179 | if (error) |
3180 | goto exit; |
3181 | |
3182 | error = user_path_parent(newdfd, newname, &newnd, &to); |
3183 | if (error) |
3184 | goto exit1; |
3185 | |
3186 | error = -EXDEV; |
3187 | if (oldnd.path.mnt != newnd.path.mnt) |
3188 | goto exit2; |
3189 | |
3190 | old_dir = oldnd.path.dentry; |
3191 | error = -EBUSY; |
3192 | if (oldnd.last_type != LAST_NORM) |
3193 | goto exit2; |
3194 | |
3195 | new_dir = newnd.path.dentry; |
3196 | if (newnd.last_type != LAST_NORM) |
3197 | goto exit2; |
3198 | |
3199 | oldnd.flags &= ~LOOKUP_PARENT; |
3200 | newnd.flags &= ~LOOKUP_PARENT; |
3201 | newnd.flags |= LOOKUP_RENAME_TARGET; |
3202 | |
3203 | trap = lock_rename(new_dir, old_dir); |
3204 | |
3205 | old_dentry = lookup_hash(&oldnd); |
3206 | error = PTR_ERR(old_dentry); |
3207 | if (IS_ERR(old_dentry)) |
3208 | goto exit3; |
3209 | /* source must exist */ |
3210 | error = -ENOENT; |
3211 | if (!old_dentry->d_inode) |
3212 | goto exit4; |
3213 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
3214 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { |
3215 | error = -ENOTDIR; |
3216 | if (oldnd.last.name[oldnd.last.len]) |
3217 | goto exit4; |
3218 | if (newnd.last.name[newnd.last.len]) |
3219 | goto exit4; |
3220 | } |
3221 | /* source should not be ancestor of target */ |
3222 | error = -EINVAL; |
3223 | if (old_dentry == trap) |
3224 | goto exit4; |
3225 | new_dentry = lookup_hash(&newnd); |
3226 | error = PTR_ERR(new_dentry); |
3227 | if (IS_ERR(new_dentry)) |
3228 | goto exit4; |
3229 | /* target should not be an ancestor of source */ |
3230 | error = -ENOTEMPTY; |
3231 | if (new_dentry == trap) |
3232 | goto exit5; |
3233 | |
3234 | error = mnt_want_write(oldnd.path.mnt); |
3235 | if (error) |
3236 | goto exit5; |
3237 | error = security_path_rename(&oldnd.path, old_dentry, |
3238 | &newnd.path, new_dentry); |
3239 | if (error) |
3240 | goto exit6; |
3241 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3242 | new_dir->d_inode, new_dentry); |
3243 | exit6: |
3244 | mnt_drop_write(oldnd.path.mnt); |
3245 | exit5: |
3246 | dput(new_dentry); |
3247 | exit4: |
3248 | dput(old_dentry); |
3249 | exit3: |
3250 | unlock_rename(new_dir, old_dir); |
3251 | exit2: |
3252 | path_put(&newnd.path); |
3253 | putname(to); |
3254 | exit1: |
3255 | path_put(&oldnd.path); |
3256 | putname(from); |
3257 | exit: |
3258 | return error; |
3259 | } |
3260 | |
3261 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
3262 | { |
3263 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); |
3264 | } |
3265 | |
3266 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3267 | { |
3268 | int len; |
3269 | |
3270 | len = PTR_ERR(link); |
3271 | if (IS_ERR(link)) |
3272 | goto out; |
3273 | |
3274 | len = strlen(link); |
3275 | if (len > (unsigned) buflen) |
3276 | len = buflen; |
3277 | if (copy_to_user(buffer, link, len)) |
3278 | len = -EFAULT; |
3279 | out: |
3280 | return len; |
3281 | } |
3282 | |
3283 | /* |
3284 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that |
3285 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not |
3286 | * using) it for any given inode is up to filesystem. |
3287 | */ |
3288 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
3289 | { |
3290 | struct nameidata nd; |
3291 | void *cookie; |
3292 | int res; |
3293 | |
3294 | nd.depth = 0; |
3295 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
3296 | if (IS_ERR(cookie)) |
3297 | return PTR_ERR(cookie); |
3298 | |
3299 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); |
3300 | if (dentry->d_inode->i_op->put_link) |
3301 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); |
3302 | return res; |
3303 | } |
3304 | |
3305 | int vfs_follow_link(struct nameidata *nd, const char *link) |
3306 | { |
3307 | return __vfs_follow_link(nd, link); |
3308 | } |
3309 | |
3310 | /* get the link contents into pagecache */ |
3311 | static char *page_getlink(struct dentry * dentry, struct page **ppage) |
3312 | { |
3313 | char *kaddr; |
3314 | struct page *page; |
3315 | struct address_space *mapping = dentry->d_inode->i_mapping; |
3316 | page = read_mapping_page(mapping, 0, NULL); |
3317 | if (IS_ERR(page)) |
3318 | return (char*)page; |
3319 | *ppage = page; |
3320 | kaddr = kmap(page); |
3321 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); |
3322 | return kaddr; |
3323 | } |
3324 | |
3325 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
3326 | { |
3327 | struct page *page = NULL; |
3328 | char *s = page_getlink(dentry, &page); |
3329 | int res = vfs_readlink(dentry,buffer,buflen,s); |
3330 | if (page) { |
3331 | kunmap(page); |
3332 | page_cache_release(page); |
3333 | } |
3334 | return res; |
3335 | } |
3336 | |
3337 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
3338 | { |
3339 | struct page *page = NULL; |
3340 | nd_set_link(nd, page_getlink(dentry, &page)); |
3341 | return page; |
3342 | } |
3343 | |
3344 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
3345 | { |
3346 | struct page *page = cookie; |
3347 | |
3348 | if (page) { |
3349 | kunmap(page); |
3350 | page_cache_release(page); |
3351 | } |
3352 | } |
3353 | |
3354 | /* |
3355 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS |
3356 | */ |
3357 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) |
3358 | { |
3359 | struct address_space *mapping = inode->i_mapping; |
3360 | struct page *page; |
3361 | void *fsdata; |
3362 | int err; |
3363 | char *kaddr; |
3364 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3365 | if (nofs) |
3366 | flags |= AOP_FLAG_NOFS; |
3367 | |
3368 | retry: |
3369 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
3370 | flags, &page, &fsdata); |
3371 | if (err) |
3372 | goto fail; |
3373 | |
3374 | kaddr = kmap_atomic(page, KM_USER0); |
3375 | memcpy(kaddr, symname, len-1); |
3376 | kunmap_atomic(kaddr, KM_USER0); |
3377 | |
3378 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, |
3379 | page, fsdata); |
3380 | if (err < 0) |
3381 | goto fail; |
3382 | if (err < len-1) |
3383 | goto retry; |
3384 | |
3385 | mark_inode_dirty(inode); |
3386 | return 0; |
3387 | fail: |
3388 | return err; |
3389 | } |
3390 | |
3391 | int page_symlink(struct inode *inode, const char *symname, int len) |
3392 | { |
3393 | return __page_symlink(inode, symname, len, |
3394 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
3395 | } |
3396 | |
3397 | const struct inode_operations page_symlink_inode_operations = { |
3398 | .readlink = generic_readlink, |
3399 | .follow_link = page_follow_link_light, |
3400 | .put_link = page_put_link, |
3401 | }; |
3402 | |
3403 | EXPORT_SYMBOL(user_path_at); |
3404 | EXPORT_SYMBOL(follow_down_one); |
3405 | EXPORT_SYMBOL(follow_down); |
3406 | EXPORT_SYMBOL(follow_up); |
3407 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ |
3408 | EXPORT_SYMBOL(getname); |
3409 | EXPORT_SYMBOL(lock_rename); |
3410 | EXPORT_SYMBOL(lookup_one_len); |
3411 | EXPORT_SYMBOL(page_follow_link_light); |
3412 | EXPORT_SYMBOL(page_put_link); |
3413 | EXPORT_SYMBOL(page_readlink); |
3414 | EXPORT_SYMBOL(__page_symlink); |
3415 | EXPORT_SYMBOL(page_symlink); |
3416 | EXPORT_SYMBOL(page_symlink_inode_operations); |
3417 | EXPORT_SYMBOL(kern_path); |
3418 | EXPORT_SYMBOL(vfs_path_lookup); |
3419 | EXPORT_SYMBOL(inode_permission); |
3420 | EXPORT_SYMBOL(unlock_rename); |
3421 | EXPORT_SYMBOL(vfs_create); |
3422 | EXPORT_SYMBOL(vfs_follow_link); |
3423 | EXPORT_SYMBOL(vfs_link); |
3424 | EXPORT_SYMBOL(vfs_mkdir); |
3425 | EXPORT_SYMBOL(vfs_mknod); |
3426 | EXPORT_SYMBOL(generic_permission); |
3427 | EXPORT_SYMBOL(vfs_readlink); |
3428 | EXPORT_SYMBOL(vfs_rename); |
3429 | EXPORT_SYMBOL(vfs_rmdir); |
3430 | EXPORT_SYMBOL(vfs_symlink); |
3431 | EXPORT_SYMBOL(vfs_unlink); |
3432 | EXPORT_SYMBOL(dentry_unhash); |
3433 | EXPORT_SYMBOL(generic_readlink); |
3434 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9