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