]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/namei.h> | |
22 | #include <linux/quotaops.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 <asm/uaccess.h> | |
37 | ||
38 | #include "internal.h" | |
39 | ||
40 | /* [Feb-1997 T. Schoebel-Theuer] | |
41 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
42 | * were necessary because of omirr. The reason is that omirr needs | |
43 | * to know the _real_ pathname, not the user-supplied one, in case | |
44 | * of symlinks (and also when transname replacements occur). | |
45 | * | |
46 | * The new code replaces the old recursive symlink resolution with | |
47 | * an iterative one (in case of non-nested symlink chains). It does | |
48 | * this with calls to <fs>_follow_link(). | |
49 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
50 | * replaced with a single function lookup_dentry() that can handle all | |
51 | * the special cases of the former code. | |
52 | * | |
53 | * With the new dcache, the pathname is stored at each inode, at least as | |
54 | * long as the refcount of the inode is positive. As a side effect, the | |
55 | * size of the dcache depends on the inode cache and thus is dynamic. | |
56 | * | |
57 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
58 | * resolution to correspond with current state of the code. | |
59 | * | |
60 | * Note that the symlink resolution is not *completely* iterative. | |
61 | * There is still a significant amount of tail- and mid- recursion in | |
62 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
63 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
64 | * may return different results than <fs>_follow_link(). Many virtual | |
65 | * filesystems (including /proc) exhibit this behavior. | |
66 | */ | |
67 | ||
68 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
69 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
70 | * and the name already exists in form of a symlink, try to create the new | |
71 | * name indicated by the symlink. The old code always complained that the | |
72 | * name already exists, due to not following the symlink even if its target | |
73 | * is nonexistent. The new semantics affects also mknod() and link() when | |
74 | * the name is a symlink pointing to a non-existant name. | |
75 | * | |
76 | * I don't know which semantics is the right one, since I have no access | |
77 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
78 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
79 | * "old" one. Personally, I think the new semantics is much more logical. | |
80 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
81 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
82 | * and in the old Linux semantics. | |
83 | */ | |
84 | ||
85 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
86 | * semantics. See the comments in "open_namei" and "do_link" below. | |
87 | * | |
88 | * [10-Sep-98 Alan Modra] Another symlink change. | |
89 | */ | |
90 | ||
91 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
92 | * inside the path - always follow. | |
93 | * in the last component in creation/removal/renaming - never follow. | |
94 | * if LOOKUP_FOLLOW passed - follow. | |
95 | * if the pathname has trailing slashes - follow. | |
96 | * otherwise - don't follow. | |
97 | * (applied in that order). | |
98 | * | |
99 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
100 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
101 | * During the 2.4 we need to fix the userland stuff depending on it - | |
102 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
103 | * XEmacs seems to be relying on it... | |
104 | */ | |
105 | /* | |
106 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
107 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives | |
108 | * any extra contention... | |
109 | */ | |
110 | ||
111 | /* In order to reduce some races, while at the same time doing additional | |
112 | * checking and hopefully speeding things up, we copy filenames to the | |
113 | * kernel data space before using them.. | |
114 | * | |
115 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
116 | * PATH_MAX includes the nul terminator --RR. | |
117 | */ | |
118 | static int do_getname(const char __user *filename, char *page) | |
119 | { | |
120 | int retval; | |
121 | unsigned long len = PATH_MAX; | |
122 | ||
123 | if (!segment_eq(get_fs(), KERNEL_DS)) { | |
124 | if ((unsigned long) filename >= TASK_SIZE) | |
125 | return -EFAULT; | |
126 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) | |
127 | len = TASK_SIZE - (unsigned long) filename; | |
128 | } | |
129 | ||
130 | retval = strncpy_from_user(page, filename, len); | |
131 | if (retval > 0) { | |
132 | if (retval < len) | |
133 | return 0; | |
134 | return -ENAMETOOLONG; | |
135 | } else if (!retval) | |
136 | retval = -ENOENT; | |
137 | return retval; | |
138 | } | |
139 | ||
140 | char * getname(const char __user * filename) | |
141 | { | |
142 | char *tmp, *result; | |
143 | ||
144 | result = ERR_PTR(-ENOMEM); | |
145 | tmp = __getname(); | |
146 | if (tmp) { | |
147 | int retval = do_getname(filename, tmp); | |
148 | ||
149 | result = tmp; | |
150 | if (retval < 0) { | |
151 | __putname(tmp); | |
152 | result = ERR_PTR(retval); | |
153 | } | |
154 | } | |
155 | audit_getname(result); | |
156 | return result; | |
157 | } | |
158 | ||
159 | #ifdef CONFIG_AUDITSYSCALL | |
160 | void putname(const char *name) | |
161 | { | |
162 | if (unlikely(!audit_dummy_context())) | |
163 | audit_putname(name); | |
164 | else | |
165 | __putname(name); | |
166 | } | |
167 | EXPORT_SYMBOL(putname); | |
168 | #endif | |
169 | ||
170 | /* | |
171 | * This does basic POSIX ACL permission checking | |
172 | */ | |
173 | static int acl_permission_check(struct inode *inode, int mask, | |
174 | int (*check_acl)(struct inode *inode, int mask)) | |
175 | { | |
176 | umode_t mode = inode->i_mode; | |
177 | ||
178 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; | |
179 | ||
180 | if (current_fsuid() == inode->i_uid) | |
181 | mode >>= 6; | |
182 | else { | |
183 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { | |
184 | int error = check_acl(inode, mask); | |
185 | if (error != -EAGAIN) | |
186 | return error; | |
187 | } | |
188 | ||
189 | if (in_group_p(inode->i_gid)) | |
190 | mode >>= 3; | |
191 | } | |
192 | ||
193 | /* | |
194 | * If the DACs are ok we don't need any capability check. | |
195 | */ | |
196 | if ((mask & ~mode) == 0) | |
197 | return 0; | |
198 | return -EACCES; | |
199 | } | |
200 | ||
201 | /** | |
202 | * generic_permission - check for access rights on a Posix-like filesystem | |
203 | * @inode: inode to check access rights for | |
204 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
205 | * @check_acl: optional callback to check for Posix ACLs | |
206 | * | |
207 | * Used to check for read/write/execute permissions on a file. | |
208 | * We use "fsuid" for this, letting us set arbitrary permissions | |
209 | * for filesystem access without changing the "normal" uids which | |
210 | * are used for other things.. | |
211 | */ | |
212 | int generic_permission(struct inode *inode, int mask, | |
213 | int (*check_acl)(struct inode *inode, int mask)) | |
214 | { | |
215 | int ret; | |
216 | ||
217 | /* | |
218 | * Do the basic POSIX ACL permission checks. | |
219 | */ | |
220 | ret = acl_permission_check(inode, mask, check_acl); | |
221 | if (ret != -EACCES) | |
222 | return ret; | |
223 | ||
224 | /* | |
225 | * Read/write DACs are always overridable. | |
226 | * Executable DACs are overridable if at least one exec bit is set. | |
227 | */ | |
228 | if (!(mask & MAY_EXEC) || execute_ok(inode)) | |
229 | if (capable(CAP_DAC_OVERRIDE)) | |
230 | return 0; | |
231 | ||
232 | /* | |
233 | * Searching includes executable on directories, else just read. | |
234 | */ | |
235 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; | |
236 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) | |
237 | if (capable(CAP_DAC_READ_SEARCH)) | |
238 | return 0; | |
239 | ||
240 | return -EACCES; | |
241 | } | |
242 | ||
243 | /** | |
244 | * inode_permission - check for access rights to a given inode | |
245 | * @inode: inode to check permission on | |
246 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
247 | * | |
248 | * Used to check for read/write/execute permissions on an inode. | |
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 | int inode_permission(struct inode *inode, int mask) | |
254 | { | |
255 | int retval; | |
256 | ||
257 | if (mask & MAY_WRITE) { | |
258 | umode_t mode = inode->i_mode; | |
259 | ||
260 | /* | |
261 | * Nobody gets write access to a read-only fs. | |
262 | */ | |
263 | if (IS_RDONLY(inode) && | |
264 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
265 | return -EROFS; | |
266 | ||
267 | /* | |
268 | * Nobody gets write access to an immutable file. | |
269 | */ | |
270 | if (IS_IMMUTABLE(inode)) | |
271 | return -EACCES; | |
272 | } | |
273 | ||
274 | if (inode->i_op->permission) | |
275 | retval = inode->i_op->permission(inode, mask); | |
276 | else | |
277 | retval = generic_permission(inode, mask, inode->i_op->check_acl); | |
278 | ||
279 | if (retval) | |
280 | return retval; | |
281 | ||
282 | retval = devcgroup_inode_permission(inode, mask); | |
283 | if (retval) | |
284 | return retval; | |
285 | ||
286 | return security_inode_permission(inode, | |
287 | mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND)); | |
288 | } | |
289 | ||
290 | /** | |
291 | * file_permission - check for additional access rights to a given file | |
292 | * @file: file to check access rights for | |
293 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
294 | * | |
295 | * Used to check for read/write/execute permissions on an already opened | |
296 | * file. | |
297 | * | |
298 | * Note: | |
299 | * Do not use this function in new code. All access checks should | |
300 | * be done using inode_permission(). | |
301 | */ | |
302 | int file_permission(struct file *file, int mask) | |
303 | { | |
304 | return inode_permission(file->f_path.dentry->d_inode, mask); | |
305 | } | |
306 | ||
307 | /* | |
308 | * get_write_access() gets write permission for a file. | |
309 | * put_write_access() releases this write permission. | |
310 | * This is used for regular files. | |
311 | * We cannot support write (and maybe mmap read-write shared) accesses and | |
312 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode | |
313 | * can have the following values: | |
314 | * 0: no writers, no VM_DENYWRITE mappings | |
315 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist | |
316 | * > 0: (i_writecount) users are writing to the file. | |
317 | * | |
318 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe | |
319 | * except for the cases where we don't hold i_writecount yet. Then we need to | |
320 | * use {get,deny}_write_access() - these functions check the sign and refuse | |
321 | * to do the change if sign is wrong. Exclusion between them is provided by | |
322 | * the inode->i_lock spinlock. | |
323 | */ | |
324 | ||
325 | int get_write_access(struct inode * inode) | |
326 | { | |
327 | spin_lock(&inode->i_lock); | |
328 | if (atomic_read(&inode->i_writecount) < 0) { | |
329 | spin_unlock(&inode->i_lock); | |
330 | return -ETXTBSY; | |
331 | } | |
332 | atomic_inc(&inode->i_writecount); | |
333 | spin_unlock(&inode->i_lock); | |
334 | ||
335 | return 0; | |
336 | } | |
337 | ||
338 | int deny_write_access(struct file * file) | |
339 | { | |
340 | struct inode *inode = file->f_path.dentry->d_inode; | |
341 | ||
342 | spin_lock(&inode->i_lock); | |
343 | if (atomic_read(&inode->i_writecount) > 0) { | |
344 | spin_unlock(&inode->i_lock); | |
345 | return -ETXTBSY; | |
346 | } | |
347 | atomic_dec(&inode->i_writecount); | |
348 | spin_unlock(&inode->i_lock); | |
349 | ||
350 | return 0; | |
351 | } | |
352 | ||
353 | /** | |
354 | * path_get - get a reference to a path | |
355 | * @path: path to get the reference to | |
356 | * | |
357 | * Given a path increment the reference count to the dentry and the vfsmount. | |
358 | */ | |
359 | void path_get(struct path *path) | |
360 | { | |
361 | mntget(path->mnt); | |
362 | dget(path->dentry); | |
363 | } | |
364 | EXPORT_SYMBOL(path_get); | |
365 | ||
366 | /** | |
367 | * path_put - put a reference to a path | |
368 | * @path: path to put the reference to | |
369 | * | |
370 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
371 | */ | |
372 | void path_put(struct path *path) | |
373 | { | |
374 | dput(path->dentry); | |
375 | mntput(path->mnt); | |
376 | } | |
377 | EXPORT_SYMBOL(path_put); | |
378 | ||
379 | /** | |
380 | * release_open_intent - free up open intent resources | |
381 | * @nd: pointer to nameidata | |
382 | */ | |
383 | void release_open_intent(struct nameidata *nd) | |
384 | { | |
385 | if (nd->intent.open.file->f_path.dentry == NULL) | |
386 | put_filp(nd->intent.open.file); | |
387 | else | |
388 | fput(nd->intent.open.file); | |
389 | } | |
390 | ||
391 | static inline struct dentry * | |
392 | do_revalidate(struct dentry *dentry, struct nameidata *nd) | |
393 | { | |
394 | int status = dentry->d_op->d_revalidate(dentry, nd); | |
395 | if (unlikely(status <= 0)) { | |
396 | /* | |
397 | * The dentry failed validation. | |
398 | * If d_revalidate returned 0 attempt to invalidate | |
399 | * the dentry otherwise d_revalidate is asking us | |
400 | * to return a fail status. | |
401 | */ | |
402 | if (!status) { | |
403 | if (!d_invalidate(dentry)) { | |
404 | dput(dentry); | |
405 | dentry = NULL; | |
406 | } | |
407 | } else { | |
408 | dput(dentry); | |
409 | dentry = ERR_PTR(status); | |
410 | } | |
411 | } | |
412 | return dentry; | |
413 | } | |
414 | ||
415 | /* | |
416 | * force_reval_path - force revalidation of a dentry | |
417 | * | |
418 | * In some situations the path walking code will trust dentries without | |
419 | * revalidating them. This causes problems for filesystems that depend on | |
420 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set | |
421 | * (which indicates that it's possible for the dentry to go stale), force | |
422 | * a d_revalidate call before proceeding. | |
423 | * | |
424 | * Returns 0 if the revalidation was successful. If the revalidation fails, | |
425 | * either return the error returned by d_revalidate or -ESTALE if the | |
426 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to | |
427 | * invalidate the dentry. It's up to the caller to handle putting references | |
428 | * to the path if necessary. | |
429 | */ | |
430 | static int | |
431 | force_reval_path(struct path *path, struct nameidata *nd) | |
432 | { | |
433 | int status; | |
434 | struct dentry *dentry = path->dentry; | |
435 | ||
436 | /* | |
437 | * only check on filesystems where it's possible for the dentry to | |
438 | * become stale. It's assumed that if this flag is set then the | |
439 | * d_revalidate op will also be defined. | |
440 | */ | |
441 | if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) | |
442 | return 0; | |
443 | ||
444 | status = dentry->d_op->d_revalidate(dentry, nd); | |
445 | if (status > 0) | |
446 | return 0; | |
447 | ||
448 | if (!status) { | |
449 | d_invalidate(dentry); | |
450 | status = -ESTALE; | |
451 | } | |
452 | return status; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Short-cut version of permission(), for calling on directories | |
457 | * during pathname resolution. Combines parts of permission() | |
458 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. | |
459 | * | |
460 | * If appropriate, check DAC only. If not appropriate, or | |
461 | * short-cut DAC fails, then call ->permission() to do more | |
462 | * complete permission check. | |
463 | */ | |
464 | static int exec_permission(struct inode *inode) | |
465 | { | |
466 | int ret; | |
467 | ||
468 | if (inode->i_op->permission) { | |
469 | ret = inode->i_op->permission(inode, MAY_EXEC); | |
470 | if (!ret) | |
471 | goto ok; | |
472 | return ret; | |
473 | } | |
474 | ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl); | |
475 | if (!ret) | |
476 | goto ok; | |
477 | ||
478 | if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) | |
479 | goto ok; | |
480 | ||
481 | return ret; | |
482 | ok: | |
483 | return security_inode_permission(inode, MAY_EXEC); | |
484 | } | |
485 | ||
486 | static __always_inline void set_root(struct nameidata *nd) | |
487 | { | |
488 | if (!nd->root.mnt) { | |
489 | struct fs_struct *fs = current->fs; | |
490 | read_lock(&fs->lock); | |
491 | nd->root = fs->root; | |
492 | path_get(&nd->root); | |
493 | read_unlock(&fs->lock); | |
494 | } | |
495 | } | |
496 | ||
497 | static int link_path_walk(const char *, struct nameidata *); | |
498 | ||
499 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) | |
500 | { | |
501 | if (IS_ERR(link)) | |
502 | goto fail; | |
503 | ||
504 | if (*link == '/') { | |
505 | set_root(nd); | |
506 | path_put(&nd->path); | |
507 | nd->path = nd->root; | |
508 | path_get(&nd->root); | |
509 | } | |
510 | ||
511 | return link_path_walk(link, nd); | |
512 | fail: | |
513 | path_put(&nd->path); | |
514 | return PTR_ERR(link); | |
515 | } | |
516 | ||
517 | static void path_put_conditional(struct path *path, struct nameidata *nd) | |
518 | { | |
519 | dput(path->dentry); | |
520 | if (path->mnt != nd->path.mnt) | |
521 | mntput(path->mnt); | |
522 | } | |
523 | ||
524 | static inline void path_to_nameidata(struct path *path, struct nameidata *nd) | |
525 | { | |
526 | dput(nd->path.dentry); | |
527 | if (nd->path.mnt != path->mnt) | |
528 | mntput(nd->path.mnt); | |
529 | nd->path.mnt = path->mnt; | |
530 | nd->path.dentry = path->dentry; | |
531 | } | |
532 | ||
533 | static __always_inline int | |
534 | __do_follow_link(struct path *path, struct nameidata *nd, void **p) | |
535 | { | |
536 | int error; | |
537 | struct dentry *dentry = path->dentry; | |
538 | ||
539 | touch_atime(path->mnt, dentry); | |
540 | nd_set_link(nd, NULL); | |
541 | ||
542 | if (path->mnt != nd->path.mnt) { | |
543 | path_to_nameidata(path, nd); | |
544 | dget(dentry); | |
545 | } | |
546 | mntget(path->mnt); | |
547 | nd->last_type = LAST_BIND; | |
548 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); | |
549 | error = PTR_ERR(*p); | |
550 | if (!IS_ERR(*p)) { | |
551 | char *s = nd_get_link(nd); | |
552 | error = 0; | |
553 | if (s) | |
554 | error = __vfs_follow_link(nd, s); | |
555 | else if (nd->last_type == LAST_BIND) { | |
556 | error = force_reval_path(&nd->path, nd); | |
557 | if (error) | |
558 | path_put(&nd->path); | |
559 | } | |
560 | } | |
561 | return error; | |
562 | } | |
563 | ||
564 | /* | |
565 | * This limits recursive symlink follows to 8, while | |
566 | * limiting consecutive symlinks to 40. | |
567 | * | |
568 | * Without that kind of total limit, nasty chains of consecutive | |
569 | * symlinks can cause almost arbitrarily long lookups. | |
570 | */ | |
571 | static inline int do_follow_link(struct path *path, struct nameidata *nd) | |
572 | { | |
573 | void *cookie; | |
574 | int err = -ELOOP; | |
575 | if (current->link_count >= MAX_NESTED_LINKS) | |
576 | goto loop; | |
577 | if (current->total_link_count >= 40) | |
578 | goto loop; | |
579 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); | |
580 | cond_resched(); | |
581 | err = security_inode_follow_link(path->dentry, nd); | |
582 | if (err) | |
583 | goto loop; | |
584 | current->link_count++; | |
585 | current->total_link_count++; | |
586 | nd->depth++; | |
587 | err = __do_follow_link(path, nd, &cookie); | |
588 | if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) | |
589 | path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); | |
590 | path_put(path); | |
591 | current->link_count--; | |
592 | nd->depth--; | |
593 | return err; | |
594 | loop: | |
595 | path_put_conditional(path, nd); | |
596 | path_put(&nd->path); | |
597 | return err; | |
598 | } | |
599 | ||
600 | int follow_up(struct path *path) | |
601 | { | |
602 | struct vfsmount *parent; | |
603 | struct dentry *mountpoint; | |
604 | spin_lock(&vfsmount_lock); | |
605 | parent = path->mnt->mnt_parent; | |
606 | if (parent == path->mnt) { | |
607 | spin_unlock(&vfsmount_lock); | |
608 | return 0; | |
609 | } | |
610 | mntget(parent); | |
611 | mountpoint = dget(path->mnt->mnt_mountpoint); | |
612 | spin_unlock(&vfsmount_lock); | |
613 | dput(path->dentry); | |
614 | path->dentry = mountpoint; | |
615 | mntput(path->mnt); | |
616 | path->mnt = parent; | |
617 | return 1; | |
618 | } | |
619 | ||
620 | /* no need for dcache_lock, as serialization is taken care in | |
621 | * namespace.c | |
622 | */ | |
623 | static int __follow_mount(struct path *path) | |
624 | { | |
625 | int res = 0; | |
626 | while (d_mountpoint(path->dentry)) { | |
627 | struct vfsmount *mounted = lookup_mnt(path); | |
628 | if (!mounted) | |
629 | break; | |
630 | dput(path->dentry); | |
631 | if (res) | |
632 | mntput(path->mnt); | |
633 | path->mnt = mounted; | |
634 | path->dentry = dget(mounted->mnt_root); | |
635 | res = 1; | |
636 | } | |
637 | return res; | |
638 | } | |
639 | ||
640 | static void follow_mount(struct path *path) | |
641 | { | |
642 | while (d_mountpoint(path->dentry)) { | |
643 | struct vfsmount *mounted = lookup_mnt(path); | |
644 | if (!mounted) | |
645 | break; | |
646 | dput(path->dentry); | |
647 | mntput(path->mnt); | |
648 | path->mnt = mounted; | |
649 | path->dentry = dget(mounted->mnt_root); | |
650 | } | |
651 | } | |
652 | ||
653 | /* no need for dcache_lock, as serialization is taken care in | |
654 | * namespace.c | |
655 | */ | |
656 | int follow_down(struct path *path) | |
657 | { | |
658 | struct vfsmount *mounted; | |
659 | ||
660 | mounted = lookup_mnt(path); | |
661 | if (mounted) { | |
662 | dput(path->dentry); | |
663 | mntput(path->mnt); | |
664 | path->mnt = mounted; | |
665 | path->dentry = dget(mounted->mnt_root); | |
666 | return 1; | |
667 | } | |
668 | return 0; | |
669 | } | |
670 | ||
671 | static __always_inline void follow_dotdot(struct nameidata *nd) | |
672 | { | |
673 | set_root(nd); | |
674 | ||
675 | while(1) { | |
676 | struct dentry *old = nd->path.dentry; | |
677 | ||
678 | if (nd->path.dentry == nd->root.dentry && | |
679 | nd->path.mnt == nd->root.mnt) { | |
680 | break; | |
681 | } | |
682 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
683 | /* rare case of legitimate dget_parent()... */ | |
684 | nd->path.dentry = dget_parent(nd->path.dentry); | |
685 | dput(old); | |
686 | break; | |
687 | } | |
688 | if (!follow_up(&nd->path)) | |
689 | break; | |
690 | } | |
691 | follow_mount(&nd->path); | |
692 | } | |
693 | ||
694 | /* | |
695 | * It's more convoluted than I'd like it to be, but... it's still fairly | |
696 | * small and for now I'd prefer to have fast path as straight as possible. | |
697 | * It _is_ time-critical. | |
698 | */ | |
699 | static int do_lookup(struct nameidata *nd, struct qstr *name, | |
700 | struct path *path) | |
701 | { | |
702 | struct vfsmount *mnt = nd->path.mnt; | |
703 | struct dentry *dentry, *parent; | |
704 | struct inode *dir; | |
705 | /* | |
706 | * See if the low-level filesystem might want | |
707 | * to use its own hash.. | |
708 | */ | |
709 | if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) { | |
710 | int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name); | |
711 | if (err < 0) | |
712 | return err; | |
713 | } | |
714 | ||
715 | dentry = __d_lookup(nd->path.dentry, name); | |
716 | if (!dentry) | |
717 | goto need_lookup; | |
718 | if (dentry->d_op && dentry->d_op->d_revalidate) | |
719 | goto need_revalidate; | |
720 | done: | |
721 | path->mnt = mnt; | |
722 | path->dentry = dentry; | |
723 | __follow_mount(path); | |
724 | return 0; | |
725 | ||
726 | need_lookup: | |
727 | parent = nd->path.dentry; | |
728 | dir = parent->d_inode; | |
729 | ||
730 | mutex_lock(&dir->i_mutex); | |
731 | /* | |
732 | * First re-do the cached lookup just in case it was created | |
733 | * while we waited for the directory semaphore.. | |
734 | * | |
735 | * FIXME! This could use version numbering or similar to | |
736 | * avoid unnecessary cache lookups. | |
737 | * | |
738 | * The "dcache_lock" is purely to protect the RCU list walker | |
739 | * from concurrent renames at this point (we mustn't get false | |
740 | * negatives from the RCU list walk here, unlike the optimistic | |
741 | * fast walk). | |
742 | * | |
743 | * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup | |
744 | */ | |
745 | dentry = d_lookup(parent, name); | |
746 | if (!dentry) { | |
747 | struct dentry *new; | |
748 | ||
749 | /* Don't create child dentry for a dead directory. */ | |
750 | dentry = ERR_PTR(-ENOENT); | |
751 | if (IS_DEADDIR(dir)) | |
752 | goto out_unlock; | |
753 | ||
754 | new = d_alloc(parent, name); | |
755 | dentry = ERR_PTR(-ENOMEM); | |
756 | if (new) { | |
757 | dentry = dir->i_op->lookup(dir, new, nd); | |
758 | if (dentry) | |
759 | dput(new); | |
760 | else | |
761 | dentry = new; | |
762 | } | |
763 | out_unlock: | |
764 | mutex_unlock(&dir->i_mutex); | |
765 | if (IS_ERR(dentry)) | |
766 | goto fail; | |
767 | goto done; | |
768 | } | |
769 | ||
770 | /* | |
771 | * Uhhuh! Nasty case: the cache was re-populated while | |
772 | * we waited on the semaphore. Need to revalidate. | |
773 | */ | |
774 | mutex_unlock(&dir->i_mutex); | |
775 | if (dentry->d_op && dentry->d_op->d_revalidate) { | |
776 | dentry = do_revalidate(dentry, nd); | |
777 | if (!dentry) | |
778 | dentry = ERR_PTR(-ENOENT); | |
779 | } | |
780 | if (IS_ERR(dentry)) | |
781 | goto fail; | |
782 | goto done; | |
783 | ||
784 | need_revalidate: | |
785 | dentry = do_revalidate(dentry, nd); | |
786 | if (!dentry) | |
787 | goto need_lookup; | |
788 | if (IS_ERR(dentry)) | |
789 | goto fail; | |
790 | goto done; | |
791 | ||
792 | fail: | |
793 | return PTR_ERR(dentry); | |
794 | } | |
795 | ||
796 | /* | |
797 | * This is a temporary kludge to deal with "automount" symlinks; proper | |
798 | * solution is to trigger them on follow_mount(), so that do_lookup() | |
799 | * would DTRT. To be killed before 2.6.34-final. | |
800 | */ | |
801 | static inline int follow_on_final(struct inode *inode, unsigned lookup_flags) | |
802 | { | |
803 | return inode && unlikely(inode->i_op->follow_link) && | |
804 | ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode)); | |
805 | } | |
806 | ||
807 | /* | |
808 | * Name resolution. | |
809 | * This is the basic name resolution function, turning a pathname into | |
810 | * the final dentry. We expect 'base' to be positive and a directory. | |
811 | * | |
812 | * Returns 0 and nd will have valid dentry and mnt on success. | |
813 | * Returns error and drops reference to input namei data on failure. | |
814 | */ | |
815 | static int link_path_walk(const char *name, struct nameidata *nd) | |
816 | { | |
817 | struct path next; | |
818 | struct inode *inode; | |
819 | int err; | |
820 | unsigned int lookup_flags = nd->flags; | |
821 | ||
822 | while (*name=='/') | |
823 | name++; | |
824 | if (!*name) | |
825 | goto return_reval; | |
826 | ||
827 | inode = nd->path.dentry->d_inode; | |
828 | if (nd->depth) | |
829 | lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); | |
830 | ||
831 | /* At this point we know we have a real path component. */ | |
832 | for(;;) { | |
833 | unsigned long hash; | |
834 | struct qstr this; | |
835 | unsigned int c; | |
836 | ||
837 | nd->flags |= LOOKUP_CONTINUE; | |
838 | err = exec_permission(inode); | |
839 | if (err) | |
840 | break; | |
841 | ||
842 | this.name = name; | |
843 | c = *(const unsigned char *)name; | |
844 | ||
845 | hash = init_name_hash(); | |
846 | do { | |
847 | name++; | |
848 | hash = partial_name_hash(c, hash); | |
849 | c = *(const unsigned char *)name; | |
850 | } while (c && (c != '/')); | |
851 | this.len = name - (const char *) this.name; | |
852 | this.hash = end_name_hash(hash); | |
853 | ||
854 | /* remove trailing slashes? */ | |
855 | if (!c) | |
856 | goto last_component; | |
857 | while (*++name == '/'); | |
858 | if (!*name) | |
859 | goto last_with_slashes; | |
860 | ||
861 | /* | |
862 | * "." and ".." are special - ".." especially so because it has | |
863 | * to be able to know about the current root directory and | |
864 | * parent relationships. | |
865 | */ | |
866 | if (this.name[0] == '.') switch (this.len) { | |
867 | default: | |
868 | break; | |
869 | case 2: | |
870 | if (this.name[1] != '.') | |
871 | break; | |
872 | follow_dotdot(nd); | |
873 | inode = nd->path.dentry->d_inode; | |
874 | /* fallthrough */ | |
875 | case 1: | |
876 | continue; | |
877 | } | |
878 | /* This does the actual lookups.. */ | |
879 | err = do_lookup(nd, &this, &next); | |
880 | if (err) | |
881 | break; | |
882 | ||
883 | err = -ENOENT; | |
884 | inode = next.dentry->d_inode; | |
885 | if (!inode) | |
886 | goto out_dput; | |
887 | ||
888 | if (inode->i_op->follow_link) { | |
889 | err = do_follow_link(&next, nd); | |
890 | if (err) | |
891 | goto return_err; | |
892 | err = -ENOENT; | |
893 | inode = nd->path.dentry->d_inode; | |
894 | if (!inode) | |
895 | break; | |
896 | } else | |
897 | path_to_nameidata(&next, nd); | |
898 | err = -ENOTDIR; | |
899 | if (!inode->i_op->lookup) | |
900 | break; | |
901 | continue; | |
902 | /* here ends the main loop */ | |
903 | ||
904 | last_with_slashes: | |
905 | lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
906 | last_component: | |
907 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ | |
908 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; | |
909 | if (lookup_flags & LOOKUP_PARENT) | |
910 | goto lookup_parent; | |
911 | if (this.name[0] == '.') switch (this.len) { | |
912 | default: | |
913 | break; | |
914 | case 2: | |
915 | if (this.name[1] != '.') | |
916 | break; | |
917 | follow_dotdot(nd); | |
918 | inode = nd->path.dentry->d_inode; | |
919 | /* fallthrough */ | |
920 | case 1: | |
921 | goto return_reval; | |
922 | } | |
923 | err = do_lookup(nd, &this, &next); | |
924 | if (err) | |
925 | break; | |
926 | inode = next.dentry->d_inode; | |
927 | if (follow_on_final(inode, lookup_flags)) { | |
928 | err = do_follow_link(&next, nd); | |
929 | if (err) | |
930 | goto return_err; | |
931 | inode = nd->path.dentry->d_inode; | |
932 | } else | |
933 | path_to_nameidata(&next, nd); | |
934 | err = -ENOENT; | |
935 | if (!inode) | |
936 | break; | |
937 | if (lookup_flags & LOOKUP_DIRECTORY) { | |
938 | err = -ENOTDIR; | |
939 | if (!inode->i_op->lookup) | |
940 | break; | |
941 | } | |
942 | goto return_base; | |
943 | lookup_parent: | |
944 | nd->last = this; | |
945 | nd->last_type = LAST_NORM; | |
946 | if (this.name[0] != '.') | |
947 | goto return_base; | |
948 | if (this.len == 1) | |
949 | nd->last_type = LAST_DOT; | |
950 | else if (this.len == 2 && this.name[1] == '.') | |
951 | nd->last_type = LAST_DOTDOT; | |
952 | else | |
953 | goto return_base; | |
954 | return_reval: | |
955 | /* | |
956 | * We bypassed the ordinary revalidation routines. | |
957 | * We may need to check the cached dentry for staleness. | |
958 | */ | |
959 | if (nd->path.dentry && nd->path.dentry->d_sb && | |
960 | (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) { | |
961 | err = -ESTALE; | |
962 | /* Note: we do not d_invalidate() */ | |
963 | if (!nd->path.dentry->d_op->d_revalidate( | |
964 | nd->path.dentry, nd)) | |
965 | break; | |
966 | } | |
967 | return_base: | |
968 | return 0; | |
969 | out_dput: | |
970 | path_put_conditional(&next, nd); | |
971 | break; | |
972 | } | |
973 | path_put(&nd->path); | |
974 | return_err: | |
975 | return err; | |
976 | } | |
977 | ||
978 | static int path_walk(const char *name, struct nameidata *nd) | |
979 | { | |
980 | struct path save = nd->path; | |
981 | int result; | |
982 | ||
983 | current->total_link_count = 0; | |
984 | ||
985 | /* make sure the stuff we saved doesn't go away */ | |
986 | path_get(&save); | |
987 | ||
988 | result = link_path_walk(name, nd); | |
989 | if (result == -ESTALE) { | |
990 | /* nd->path had been dropped */ | |
991 | current->total_link_count = 0; | |
992 | nd->path = save; | |
993 | path_get(&nd->path); | |
994 | nd->flags |= LOOKUP_REVAL; | |
995 | result = link_path_walk(name, nd); | |
996 | } | |
997 | ||
998 | path_put(&save); | |
999 | ||
1000 | return result; | |
1001 | } | |
1002 | ||
1003 | static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) | |
1004 | { | |
1005 | int retval = 0; | |
1006 | int fput_needed; | |
1007 | struct file *file; | |
1008 | ||
1009 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
1010 | nd->flags = flags; | |
1011 | nd->depth = 0; | |
1012 | nd->root.mnt = NULL; | |
1013 | ||
1014 | if (*name=='/') { | |
1015 | set_root(nd); | |
1016 | nd->path = nd->root; | |
1017 | path_get(&nd->root); | |
1018 | } else if (dfd == AT_FDCWD) { | |
1019 | struct fs_struct *fs = current->fs; | |
1020 | read_lock(&fs->lock); | |
1021 | nd->path = fs->pwd; | |
1022 | path_get(&fs->pwd); | |
1023 | read_unlock(&fs->lock); | |
1024 | } else { | |
1025 | struct dentry *dentry; | |
1026 | ||
1027 | file = fget_light(dfd, &fput_needed); | |
1028 | retval = -EBADF; | |
1029 | if (!file) | |
1030 | goto out_fail; | |
1031 | ||
1032 | dentry = file->f_path.dentry; | |
1033 | ||
1034 | retval = -ENOTDIR; | |
1035 | if (!S_ISDIR(dentry->d_inode->i_mode)) | |
1036 | goto fput_fail; | |
1037 | ||
1038 | retval = file_permission(file, MAY_EXEC); | |
1039 | if (retval) | |
1040 | goto fput_fail; | |
1041 | ||
1042 | nd->path = file->f_path; | |
1043 | path_get(&file->f_path); | |
1044 | ||
1045 | fput_light(file, fput_needed); | |
1046 | } | |
1047 | return 0; | |
1048 | ||
1049 | fput_fail: | |
1050 | fput_light(file, fput_needed); | |
1051 | out_fail: | |
1052 | return retval; | |
1053 | } | |
1054 | ||
1055 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
1056 | static int do_path_lookup(int dfd, const char *name, | |
1057 | unsigned int flags, struct nameidata *nd) | |
1058 | { | |
1059 | int retval = path_init(dfd, name, flags, nd); | |
1060 | if (!retval) | |
1061 | retval = path_walk(name, nd); | |
1062 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && | |
1063 | nd->path.dentry->d_inode)) | |
1064 | audit_inode(name, nd->path.dentry); | |
1065 | if (nd->root.mnt) { | |
1066 | path_put(&nd->root); | |
1067 | nd->root.mnt = NULL; | |
1068 | } | |
1069 | return retval; | |
1070 | } | |
1071 | ||
1072 | int path_lookup(const char *name, unsigned int flags, | |
1073 | struct nameidata *nd) | |
1074 | { | |
1075 | return do_path_lookup(AT_FDCWD, name, flags, nd); | |
1076 | } | |
1077 | ||
1078 | int kern_path(const char *name, unsigned int flags, struct path *path) | |
1079 | { | |
1080 | struct nameidata nd; | |
1081 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); | |
1082 | if (!res) | |
1083 | *path = nd.path; | |
1084 | return res; | |
1085 | } | |
1086 | ||
1087 | /** | |
1088 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
1089 | * @dentry: pointer to dentry of the base directory | |
1090 | * @mnt: pointer to vfs mount of the base directory | |
1091 | * @name: pointer to file name | |
1092 | * @flags: lookup flags | |
1093 | * @nd: pointer to nameidata | |
1094 | */ | |
1095 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
1096 | const char *name, unsigned int flags, | |
1097 | struct nameidata *nd) | |
1098 | { | |
1099 | int retval; | |
1100 | ||
1101 | /* same as do_path_lookup */ | |
1102 | nd->last_type = LAST_ROOT; | |
1103 | nd->flags = flags; | |
1104 | nd->depth = 0; | |
1105 | ||
1106 | nd->path.dentry = dentry; | |
1107 | nd->path.mnt = mnt; | |
1108 | path_get(&nd->path); | |
1109 | nd->root = nd->path; | |
1110 | path_get(&nd->root); | |
1111 | ||
1112 | retval = path_walk(name, nd); | |
1113 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && | |
1114 | nd->path.dentry->d_inode)) | |
1115 | audit_inode(name, nd->path.dentry); | |
1116 | ||
1117 | path_put(&nd->root); | |
1118 | nd->root.mnt = NULL; | |
1119 | ||
1120 | return retval; | |
1121 | } | |
1122 | ||
1123 | static struct dentry *__lookup_hash(struct qstr *name, | |
1124 | struct dentry *base, struct nameidata *nd) | |
1125 | { | |
1126 | struct dentry *dentry; | |
1127 | struct inode *inode; | |
1128 | int err; | |
1129 | ||
1130 | inode = base->d_inode; | |
1131 | ||
1132 | /* | |
1133 | * See if the low-level filesystem might want | |
1134 | * to use its own hash.. | |
1135 | */ | |
1136 | if (base->d_op && base->d_op->d_hash) { | |
1137 | err = base->d_op->d_hash(base, name); | |
1138 | dentry = ERR_PTR(err); | |
1139 | if (err < 0) | |
1140 | goto out; | |
1141 | } | |
1142 | ||
1143 | dentry = __d_lookup(base, name); | |
1144 | ||
1145 | /* lockess __d_lookup may fail due to concurrent d_move() | |
1146 | * in some unrelated directory, so try with d_lookup | |
1147 | */ | |
1148 | if (!dentry) | |
1149 | dentry = d_lookup(base, name); | |
1150 | ||
1151 | if (dentry && dentry->d_op && dentry->d_op->d_revalidate) | |
1152 | dentry = do_revalidate(dentry, nd); | |
1153 | ||
1154 | if (!dentry) { | |
1155 | struct dentry *new; | |
1156 | ||
1157 | /* Don't create child dentry for a dead directory. */ | |
1158 | dentry = ERR_PTR(-ENOENT); | |
1159 | if (IS_DEADDIR(inode)) | |
1160 | goto out; | |
1161 | ||
1162 | new = d_alloc(base, name); | |
1163 | dentry = ERR_PTR(-ENOMEM); | |
1164 | if (!new) | |
1165 | goto out; | |
1166 | dentry = inode->i_op->lookup(inode, new, nd); | |
1167 | if (!dentry) | |
1168 | dentry = new; | |
1169 | else | |
1170 | dput(new); | |
1171 | } | |
1172 | out: | |
1173 | return dentry; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * Restricted form of lookup. Doesn't follow links, single-component only, | |
1178 | * needs parent already locked. Doesn't follow mounts. | |
1179 | * SMP-safe. | |
1180 | */ | |
1181 | static struct dentry *lookup_hash(struct nameidata *nd) | |
1182 | { | |
1183 | int err; | |
1184 | ||
1185 | err = exec_permission(nd->path.dentry->d_inode); | |
1186 | if (err) | |
1187 | return ERR_PTR(err); | |
1188 | return __lookup_hash(&nd->last, nd->path.dentry, nd); | |
1189 | } | |
1190 | ||
1191 | static int __lookup_one_len(const char *name, struct qstr *this, | |
1192 | struct dentry *base, int len) | |
1193 | { | |
1194 | unsigned long hash; | |
1195 | unsigned int c; | |
1196 | ||
1197 | this->name = name; | |
1198 | this->len = len; | |
1199 | if (!len) | |
1200 | return -EACCES; | |
1201 | ||
1202 | hash = init_name_hash(); | |
1203 | while (len--) { | |
1204 | c = *(const unsigned char *)name++; | |
1205 | if (c == '/' || c == '\0') | |
1206 | return -EACCES; | |
1207 | hash = partial_name_hash(c, hash); | |
1208 | } | |
1209 | this->hash = end_name_hash(hash); | |
1210 | return 0; | |
1211 | } | |
1212 | ||
1213 | /** | |
1214 | * lookup_one_len - filesystem helper to lookup single pathname component | |
1215 | * @name: pathname component to lookup | |
1216 | * @base: base directory to lookup from | |
1217 | * @len: maximum length @len should be interpreted to | |
1218 | * | |
1219 | * Note that this routine is purely a helper for filesystem usage and should | |
1220 | * not be called by generic code. Also note that by using this function the | |
1221 | * nameidata argument is passed to the filesystem methods and a filesystem | |
1222 | * using this helper needs to be prepared for that. | |
1223 | */ | |
1224 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
1225 | { | |
1226 | int err; | |
1227 | struct qstr this; | |
1228 | ||
1229 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); | |
1230 | ||
1231 | err = __lookup_one_len(name, &this, base, len); | |
1232 | if (err) | |
1233 | return ERR_PTR(err); | |
1234 | ||
1235 | err = exec_permission(base->d_inode); | |
1236 | if (err) | |
1237 | return ERR_PTR(err); | |
1238 | return __lookup_hash(&this, base, NULL); | |
1239 | } | |
1240 | ||
1241 | int user_path_at(int dfd, const char __user *name, unsigned flags, | |
1242 | struct path *path) | |
1243 | { | |
1244 | struct nameidata nd; | |
1245 | char *tmp = getname(name); | |
1246 | int err = PTR_ERR(tmp); | |
1247 | if (!IS_ERR(tmp)) { | |
1248 | ||
1249 | BUG_ON(flags & LOOKUP_PARENT); | |
1250 | ||
1251 | err = do_path_lookup(dfd, tmp, flags, &nd); | |
1252 | putname(tmp); | |
1253 | if (!err) | |
1254 | *path = nd.path; | |
1255 | } | |
1256 | return err; | |
1257 | } | |
1258 | ||
1259 | static int user_path_parent(int dfd, const char __user *path, | |
1260 | struct nameidata *nd, char **name) | |
1261 | { | |
1262 | char *s = getname(path); | |
1263 | int error; | |
1264 | ||
1265 | if (IS_ERR(s)) | |
1266 | return PTR_ERR(s); | |
1267 | ||
1268 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); | |
1269 | if (error) | |
1270 | putname(s); | |
1271 | else | |
1272 | *name = s; | |
1273 | ||
1274 | return error; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * It's inline, so penalty for filesystems that don't use sticky bit is | |
1279 | * minimal. | |
1280 | */ | |
1281 | static inline int check_sticky(struct inode *dir, struct inode *inode) | |
1282 | { | |
1283 | uid_t fsuid = current_fsuid(); | |
1284 | ||
1285 | if (!(dir->i_mode & S_ISVTX)) | |
1286 | return 0; | |
1287 | if (inode->i_uid == fsuid) | |
1288 | return 0; | |
1289 | if (dir->i_uid == fsuid) | |
1290 | return 0; | |
1291 | return !capable(CAP_FOWNER); | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Check whether we can remove a link victim from directory dir, check | |
1296 | * whether the type of victim is right. | |
1297 | * 1. We can't do it if dir is read-only (done in permission()) | |
1298 | * 2. We should have write and exec permissions on dir | |
1299 | * 3. We can't remove anything from append-only dir | |
1300 | * 4. We can't do anything with immutable dir (done in permission()) | |
1301 | * 5. If the sticky bit on dir is set we should either | |
1302 | * a. be owner of dir, or | |
1303 | * b. be owner of victim, or | |
1304 | * c. have CAP_FOWNER capability | |
1305 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
1306 | * links pointing to it. | |
1307 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
1308 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
1309 | * 9. We can't remove a root or mountpoint. | |
1310 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
1311 | * nfs_async_unlink(). | |
1312 | */ | |
1313 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) | |
1314 | { | |
1315 | int error; | |
1316 | ||
1317 | if (!victim->d_inode) | |
1318 | return -ENOENT; | |
1319 | ||
1320 | BUG_ON(victim->d_parent->d_inode != dir); | |
1321 | audit_inode_child(victim, dir); | |
1322 | ||
1323 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
1324 | if (error) | |
1325 | return error; | |
1326 | if (IS_APPEND(dir)) | |
1327 | return -EPERM; | |
1328 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| | |
1329 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) | |
1330 | return -EPERM; | |
1331 | if (isdir) { | |
1332 | if (!S_ISDIR(victim->d_inode->i_mode)) | |
1333 | return -ENOTDIR; | |
1334 | if (IS_ROOT(victim)) | |
1335 | return -EBUSY; | |
1336 | } else if (S_ISDIR(victim->d_inode->i_mode)) | |
1337 | return -EISDIR; | |
1338 | if (IS_DEADDIR(dir)) | |
1339 | return -ENOENT; | |
1340 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
1341 | return -EBUSY; | |
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | /* Check whether we can create an object with dentry child in directory | |
1346 | * dir. | |
1347 | * 1. We can't do it if child already exists (open has special treatment for | |
1348 | * this case, but since we are inlined it's OK) | |
1349 | * 2. We can't do it if dir is read-only (done in permission()) | |
1350 | * 3. We should have write and exec permissions on dir | |
1351 | * 4. We can't do it if dir is immutable (done in permission()) | |
1352 | */ | |
1353 | static inline int may_create(struct inode *dir, struct dentry *child) | |
1354 | { | |
1355 | if (child->d_inode) | |
1356 | return -EEXIST; | |
1357 | if (IS_DEADDIR(dir)) | |
1358 | return -ENOENT; | |
1359 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
1360 | } | |
1361 | ||
1362 | /* | |
1363 | * O_DIRECTORY translates into forcing a directory lookup. | |
1364 | */ | |
1365 | static inline int lookup_flags(unsigned int f) | |
1366 | { | |
1367 | unsigned long retval = LOOKUP_FOLLOW; | |
1368 | ||
1369 | if (f & O_NOFOLLOW) | |
1370 | retval &= ~LOOKUP_FOLLOW; | |
1371 | ||
1372 | if (f & O_DIRECTORY) | |
1373 | retval |= LOOKUP_DIRECTORY; | |
1374 | ||
1375 | return retval; | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * p1 and p2 should be directories on the same fs. | |
1380 | */ | |
1381 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
1382 | { | |
1383 | struct dentry *p; | |
1384 | ||
1385 | if (p1 == p2) { | |
1386 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1387 | return NULL; | |
1388 | } | |
1389 | ||
1390 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); | |
1391 | ||
1392 | p = d_ancestor(p2, p1); | |
1393 | if (p) { | |
1394 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); | |
1395 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); | |
1396 | return p; | |
1397 | } | |
1398 | ||
1399 | p = d_ancestor(p1, p2); | |
1400 | if (p) { | |
1401 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1402 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1403 | return p; | |
1404 | } | |
1405 | ||
1406 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); | |
1407 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); | |
1408 | return NULL; | |
1409 | } | |
1410 | ||
1411 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
1412 | { | |
1413 | mutex_unlock(&p1->d_inode->i_mutex); | |
1414 | if (p1 != p2) { | |
1415 | mutex_unlock(&p2->d_inode->i_mutex); | |
1416 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1421 | struct nameidata *nd) | |
1422 | { | |
1423 | int error = may_create(dir, dentry); | |
1424 | ||
1425 | if (error) | |
1426 | return error; | |
1427 | ||
1428 | if (!dir->i_op->create) | |
1429 | return -EACCES; /* shouldn't it be ENOSYS? */ | |
1430 | mode &= S_IALLUGO; | |
1431 | mode |= S_IFREG; | |
1432 | error = security_inode_create(dir, dentry, mode); | |
1433 | if (error) | |
1434 | return error; | |
1435 | vfs_dq_init(dir); | |
1436 | error = dir->i_op->create(dir, dentry, mode, nd); | |
1437 | if (!error) | |
1438 | fsnotify_create(dir, dentry); | |
1439 | return error; | |
1440 | } | |
1441 | ||
1442 | int may_open(struct path *path, int acc_mode, int flag) | |
1443 | { | |
1444 | struct dentry *dentry = path->dentry; | |
1445 | struct inode *inode = dentry->d_inode; | |
1446 | int error; | |
1447 | ||
1448 | if (!inode) | |
1449 | return -ENOENT; | |
1450 | ||
1451 | switch (inode->i_mode & S_IFMT) { | |
1452 | case S_IFLNK: | |
1453 | return -ELOOP; | |
1454 | case S_IFDIR: | |
1455 | if (acc_mode & MAY_WRITE) | |
1456 | return -EISDIR; | |
1457 | break; | |
1458 | case S_IFBLK: | |
1459 | case S_IFCHR: | |
1460 | if (path->mnt->mnt_flags & MNT_NODEV) | |
1461 | return -EACCES; | |
1462 | /*FALLTHRU*/ | |
1463 | case S_IFIFO: | |
1464 | case S_IFSOCK: | |
1465 | flag &= ~O_TRUNC; | |
1466 | break; | |
1467 | } | |
1468 | ||
1469 | error = inode_permission(inode, acc_mode); | |
1470 | if (error) | |
1471 | return error; | |
1472 | ||
1473 | /* | |
1474 | * An append-only file must be opened in append mode for writing. | |
1475 | */ | |
1476 | if (IS_APPEND(inode)) { | |
1477 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) | |
1478 | return -EPERM; | |
1479 | if (flag & O_TRUNC) | |
1480 | return -EPERM; | |
1481 | } | |
1482 | ||
1483 | /* O_NOATIME can only be set by the owner or superuser */ | |
1484 | if (flag & O_NOATIME && !is_owner_or_cap(inode)) | |
1485 | return -EPERM; | |
1486 | ||
1487 | /* | |
1488 | * Ensure there are no outstanding leases on the file. | |
1489 | */ | |
1490 | return break_lease(inode, flag); | |
1491 | } | |
1492 | ||
1493 | static int handle_truncate(struct path *path) | |
1494 | { | |
1495 | struct inode *inode = path->dentry->d_inode; | |
1496 | int error = get_write_access(inode); | |
1497 | if (error) | |
1498 | return error; | |
1499 | /* | |
1500 | * Refuse to truncate files with mandatory locks held on them. | |
1501 | */ | |
1502 | error = locks_verify_locked(inode); | |
1503 | if (!error) | |
1504 | error = security_path_truncate(path, 0, | |
1505 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN); | |
1506 | if (!error) { | |
1507 | error = do_truncate(path->dentry, 0, | |
1508 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
1509 | NULL); | |
1510 | } | |
1511 | put_write_access(inode); | |
1512 | return error; | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * Be careful about ever adding any more callers of this | |
1517 | * function. Its flags must be in the namei format, not | |
1518 | * what get passed to sys_open(). | |
1519 | */ | |
1520 | static int __open_namei_create(struct nameidata *nd, struct path *path, | |
1521 | int open_flag, int mode) | |
1522 | { | |
1523 | int error; | |
1524 | struct dentry *dir = nd->path.dentry; | |
1525 | ||
1526 | if (!IS_POSIXACL(dir->d_inode)) | |
1527 | mode &= ~current_umask(); | |
1528 | error = security_path_mknod(&nd->path, path->dentry, mode, 0); | |
1529 | if (error) | |
1530 | goto out_unlock; | |
1531 | error = vfs_create(dir->d_inode, path->dentry, mode, nd); | |
1532 | out_unlock: | |
1533 | mutex_unlock(&dir->d_inode->i_mutex); | |
1534 | dput(nd->path.dentry); | |
1535 | nd->path.dentry = path->dentry; | |
1536 | if (error) | |
1537 | return error; | |
1538 | /* Don't check for write permission, don't truncate */ | |
1539 | return may_open(&nd->path, 0, open_flag & ~O_TRUNC); | |
1540 | } | |
1541 | ||
1542 | /* | |
1543 | * Note that while the flag value (low two bits) for sys_open means: | |
1544 | * 00 - read-only | |
1545 | * 01 - write-only | |
1546 | * 10 - read-write | |
1547 | * 11 - special | |
1548 | * it is changed into | |
1549 | * 00 - no permissions needed | |
1550 | * 01 - read-permission | |
1551 | * 10 - write-permission | |
1552 | * 11 - read-write | |
1553 | * for the internal routines (ie open_namei()/follow_link() etc) | |
1554 | * This is more logical, and also allows the 00 "no perm needed" | |
1555 | * to be used for symlinks (where the permissions are checked | |
1556 | * later). | |
1557 | * | |
1558 | */ | |
1559 | static inline int open_to_namei_flags(int flag) | |
1560 | { | |
1561 | if ((flag+1) & O_ACCMODE) | |
1562 | flag++; | |
1563 | return flag; | |
1564 | } | |
1565 | ||
1566 | static int open_will_truncate(int flag, struct inode *inode) | |
1567 | { | |
1568 | /* | |
1569 | * We'll never write to the fs underlying | |
1570 | * a device file. | |
1571 | */ | |
1572 | if (special_file(inode->i_mode)) | |
1573 | return 0; | |
1574 | return (flag & O_TRUNC); | |
1575 | } | |
1576 | ||
1577 | static struct file *finish_open(struct nameidata *nd, | |
1578 | int open_flag, int acc_mode) | |
1579 | { | |
1580 | struct file *filp; | |
1581 | int will_truncate; | |
1582 | int error; | |
1583 | ||
1584 | will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); | |
1585 | if (will_truncate) { | |
1586 | error = mnt_want_write(nd->path.mnt); | |
1587 | if (error) | |
1588 | goto exit; | |
1589 | } | |
1590 | error = may_open(&nd->path, acc_mode, open_flag); | |
1591 | if (error) { | |
1592 | if (will_truncate) | |
1593 | mnt_drop_write(nd->path.mnt); | |
1594 | goto exit; | |
1595 | } | |
1596 | filp = nameidata_to_filp(nd); | |
1597 | if (!IS_ERR(filp)) { | |
1598 | error = ima_file_check(filp, acc_mode); | |
1599 | if (error) { | |
1600 | fput(filp); | |
1601 | filp = ERR_PTR(error); | |
1602 | } | |
1603 | } | |
1604 | if (!IS_ERR(filp)) { | |
1605 | if (acc_mode & MAY_WRITE) | |
1606 | vfs_dq_init(nd->path.dentry->d_inode); | |
1607 | ||
1608 | if (will_truncate) { | |
1609 | error = handle_truncate(&nd->path); | |
1610 | if (error) { | |
1611 | fput(filp); | |
1612 | filp = ERR_PTR(error); | |
1613 | } | |
1614 | } | |
1615 | } | |
1616 | /* | |
1617 | * It is now safe to drop the mnt write | |
1618 | * because the filp has had a write taken | |
1619 | * on its behalf. | |
1620 | */ | |
1621 | if (will_truncate) | |
1622 | mnt_drop_write(nd->path.mnt); | |
1623 | return filp; | |
1624 | ||
1625 | exit: | |
1626 | if (!IS_ERR(nd->intent.open.file)) | |
1627 | release_open_intent(nd); | |
1628 | path_put(&nd->path); | |
1629 | return ERR_PTR(error); | |
1630 | } | |
1631 | ||
1632 | static struct file *do_last(struct nameidata *nd, struct path *path, | |
1633 | int open_flag, int acc_mode, | |
1634 | int mode, const char *pathname) | |
1635 | { | |
1636 | struct dentry *dir = nd->path.dentry; | |
1637 | struct file *filp; | |
1638 | int error; | |
1639 | ||
1640 | if (nd->last_type == LAST_BIND) | |
1641 | goto ok; | |
1642 | ||
1643 | error = -EISDIR; | |
1644 | if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len]) | |
1645 | goto exit; | |
1646 | ||
1647 | mutex_lock(&dir->d_inode->i_mutex); | |
1648 | ||
1649 | path->dentry = lookup_hash(nd); | |
1650 | path->mnt = nd->path.mnt; | |
1651 | ||
1652 | error = PTR_ERR(path->dentry); | |
1653 | if (IS_ERR(path->dentry)) { | |
1654 | mutex_unlock(&dir->d_inode->i_mutex); | |
1655 | goto exit; | |
1656 | } | |
1657 | ||
1658 | if (IS_ERR(nd->intent.open.file)) { | |
1659 | error = PTR_ERR(nd->intent.open.file); | |
1660 | goto exit_mutex_unlock; | |
1661 | } | |
1662 | ||
1663 | /* Negative dentry, just create the file */ | |
1664 | if (!path->dentry->d_inode) { | |
1665 | /* | |
1666 | * This write is needed to ensure that a | |
1667 | * ro->rw transition does not occur between | |
1668 | * the time when the file is created and when | |
1669 | * a permanent write count is taken through | |
1670 | * the 'struct file' in nameidata_to_filp(). | |
1671 | */ | |
1672 | error = mnt_want_write(nd->path.mnt); | |
1673 | if (error) | |
1674 | goto exit_mutex_unlock; | |
1675 | error = __open_namei_create(nd, path, open_flag, mode); | |
1676 | if (error) { | |
1677 | mnt_drop_write(nd->path.mnt); | |
1678 | goto exit; | |
1679 | } | |
1680 | filp = nameidata_to_filp(nd); | |
1681 | mnt_drop_write(nd->path.mnt); | |
1682 | if (!IS_ERR(filp)) { | |
1683 | error = ima_file_check(filp, acc_mode); | |
1684 | if (error) { | |
1685 | fput(filp); | |
1686 | filp = ERR_PTR(error); | |
1687 | } | |
1688 | } | |
1689 | return filp; | |
1690 | } | |
1691 | ||
1692 | /* | |
1693 | * It already exists. | |
1694 | */ | |
1695 | mutex_unlock(&dir->d_inode->i_mutex); | |
1696 | audit_inode(pathname, path->dentry); | |
1697 | ||
1698 | error = -EEXIST; | |
1699 | if (open_flag & O_EXCL) | |
1700 | goto exit_dput; | |
1701 | ||
1702 | if (__follow_mount(path)) { | |
1703 | error = -ELOOP; | |
1704 | if (open_flag & O_NOFOLLOW) | |
1705 | goto exit_dput; | |
1706 | } | |
1707 | ||
1708 | error = -ENOENT; | |
1709 | if (!path->dentry->d_inode) | |
1710 | goto exit_dput; | |
1711 | ||
1712 | if (path->dentry->d_inode->i_op->follow_link) | |
1713 | return NULL; | |
1714 | ||
1715 | path_to_nameidata(path, nd); | |
1716 | error = -EISDIR; | |
1717 | if (S_ISDIR(path->dentry->d_inode->i_mode)) | |
1718 | goto exit; | |
1719 | ok: | |
1720 | filp = finish_open(nd, open_flag, acc_mode); | |
1721 | return filp; | |
1722 | ||
1723 | exit_mutex_unlock: | |
1724 | mutex_unlock(&dir->d_inode->i_mutex); | |
1725 | exit_dput: | |
1726 | path_put_conditional(path, nd); | |
1727 | exit: | |
1728 | if (!IS_ERR(nd->intent.open.file)) | |
1729 | release_open_intent(nd); | |
1730 | path_put(&nd->path); | |
1731 | return ERR_PTR(error); | |
1732 | } | |
1733 | ||
1734 | /* | |
1735 | * Note that the low bits of the passed in "open_flag" | |
1736 | * are not the same as in the local variable "flag". See | |
1737 | * open_to_namei_flags() for more details. | |
1738 | */ | |
1739 | struct file *do_filp_open(int dfd, const char *pathname, | |
1740 | int open_flag, int mode, int acc_mode) | |
1741 | { | |
1742 | struct file *filp; | |
1743 | struct nameidata nd; | |
1744 | int error; | |
1745 | struct path path; | |
1746 | int count = 0; | |
1747 | int flag = open_to_namei_flags(open_flag); | |
1748 | int force_reval = 0; | |
1749 | ||
1750 | /* | |
1751 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only | |
1752 | * check for O_DSYNC if the need any syncing at all we enforce it's | |
1753 | * always set instead of having to deal with possibly weird behaviour | |
1754 | * for malicious applications setting only __O_SYNC. | |
1755 | */ | |
1756 | if (open_flag & __O_SYNC) | |
1757 | open_flag |= O_DSYNC; | |
1758 | ||
1759 | if (!acc_mode) | |
1760 | acc_mode = MAY_OPEN | ACC_MODE(open_flag); | |
1761 | ||
1762 | /* O_TRUNC implies we need access checks for write permissions */ | |
1763 | if (open_flag & O_TRUNC) | |
1764 | acc_mode |= MAY_WRITE; | |
1765 | ||
1766 | /* Allow the LSM permission hook to distinguish append | |
1767 | access from general write access. */ | |
1768 | if (open_flag & O_APPEND) | |
1769 | acc_mode |= MAY_APPEND; | |
1770 | ||
1771 | /* | |
1772 | * The simplest case - just a plain lookup. | |
1773 | */ | |
1774 | if (!(open_flag & O_CREAT)) { | |
1775 | filp = get_empty_filp(); | |
1776 | ||
1777 | if (filp == NULL) | |
1778 | return ERR_PTR(-ENFILE); | |
1779 | nd.intent.open.file = filp; | |
1780 | filp->f_flags = open_flag; | |
1781 | nd.intent.open.flags = flag; | |
1782 | nd.intent.open.create_mode = 0; | |
1783 | error = do_path_lookup(dfd, pathname, | |
1784 | lookup_flags(open_flag)|LOOKUP_OPEN, &nd); | |
1785 | if (IS_ERR(nd.intent.open.file)) { | |
1786 | if (error == 0) { | |
1787 | error = PTR_ERR(nd.intent.open.file); | |
1788 | path_put(&nd.path); | |
1789 | } | |
1790 | } else if (error) | |
1791 | release_open_intent(&nd); | |
1792 | if (error) | |
1793 | return ERR_PTR(error); | |
1794 | return finish_open(&nd, open_flag, acc_mode); | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * Create - we need to know the parent. | |
1799 | */ | |
1800 | reval: | |
1801 | error = path_init(dfd, pathname, LOOKUP_PARENT, &nd); | |
1802 | if (error) | |
1803 | return ERR_PTR(error); | |
1804 | if (force_reval) | |
1805 | nd.flags |= LOOKUP_REVAL; | |
1806 | ||
1807 | current->total_link_count = 0; | |
1808 | error = link_path_walk(pathname, &nd); | |
1809 | if (error) { | |
1810 | filp = ERR_PTR(error); | |
1811 | goto out; | |
1812 | } | |
1813 | if (unlikely(!audit_dummy_context())) | |
1814 | audit_inode(pathname, nd.path.dentry); | |
1815 | ||
1816 | /* | |
1817 | * We have the parent and last component. | |
1818 | */ | |
1819 | ||
1820 | error = -ENFILE; | |
1821 | filp = get_empty_filp(); | |
1822 | if (filp == NULL) | |
1823 | goto exit_parent; | |
1824 | nd.intent.open.file = filp; | |
1825 | filp->f_flags = open_flag; | |
1826 | nd.intent.open.flags = flag; | |
1827 | nd.intent.open.create_mode = mode; | |
1828 | nd.flags &= ~LOOKUP_PARENT; | |
1829 | nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN; | |
1830 | if (open_flag & O_EXCL) | |
1831 | nd.flags |= LOOKUP_EXCL; | |
1832 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); | |
1833 | while (unlikely(!filp)) { /* trailing symlink */ | |
1834 | struct path holder; | |
1835 | struct inode *inode; | |
1836 | void *cookie; | |
1837 | error = -ELOOP; | |
1838 | if ((open_flag & O_NOFOLLOW) || count++ == 32) | |
1839 | goto exit_dput; | |
1840 | /* | |
1841 | * This is subtle. Instead of calling do_follow_link() we do | |
1842 | * the thing by hands. The reason is that this way we have zero | |
1843 | * link_count and path_walk() (called from ->follow_link) | |
1844 | * honoring LOOKUP_PARENT. After that we have the parent and | |
1845 | * last component, i.e. we are in the same situation as after | |
1846 | * the first path_walk(). Well, almost - if the last component | |
1847 | * is normal we get its copy stored in nd->last.name and we will | |
1848 | * have to putname() it when we are done. Procfs-like symlinks | |
1849 | * just set LAST_BIND. | |
1850 | */ | |
1851 | nd.flags |= LOOKUP_PARENT; | |
1852 | error = security_inode_follow_link(path.dentry, &nd); | |
1853 | if (error) | |
1854 | goto exit_dput; | |
1855 | error = __do_follow_link(&path, &nd, &cookie); | |
1856 | if (unlikely(error)) { | |
1857 | /* nd.path had been dropped */ | |
1858 | inode = path.dentry->d_inode; | |
1859 | if (!IS_ERR(cookie) && inode->i_op->put_link) | |
1860 | inode->i_op->put_link(path.dentry, &nd, cookie); | |
1861 | path_put(&path); | |
1862 | release_open_intent(&nd); | |
1863 | filp = ERR_PTR(error); | |
1864 | goto out; | |
1865 | } | |
1866 | holder = path; | |
1867 | nd.flags &= ~LOOKUP_PARENT; | |
1868 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); | |
1869 | inode = holder.dentry->d_inode; | |
1870 | if (inode->i_op->put_link) | |
1871 | inode->i_op->put_link(holder.dentry, &nd, cookie); | |
1872 | path_put(&holder); | |
1873 | } | |
1874 | out: | |
1875 | if (nd.root.mnt) | |
1876 | path_put(&nd.root); | |
1877 | if (filp == ERR_PTR(-ESTALE) && !force_reval) { | |
1878 | force_reval = 1; | |
1879 | goto reval; | |
1880 | } | |
1881 | return filp; | |
1882 | ||
1883 | exit_dput: | |
1884 | path_put_conditional(&path, &nd); | |
1885 | if (!IS_ERR(nd.intent.open.file)) | |
1886 | release_open_intent(&nd); | |
1887 | exit_parent: | |
1888 | path_put(&nd.path); | |
1889 | filp = ERR_PTR(error); | |
1890 | goto out; | |
1891 | } | |
1892 | ||
1893 | /** | |
1894 | * filp_open - open file and return file pointer | |
1895 | * | |
1896 | * @filename: path to open | |
1897 | * @flags: open flags as per the open(2) second argument | |
1898 | * @mode: mode for the new file if O_CREAT is set, else ignored | |
1899 | * | |
1900 | * This is the helper to open a file from kernelspace if you really | |
1901 | * have to. But in generally you should not do this, so please move | |
1902 | * along, nothing to see here.. | |
1903 | */ | |
1904 | struct file *filp_open(const char *filename, int flags, int mode) | |
1905 | { | |
1906 | return do_filp_open(AT_FDCWD, filename, flags, mode, 0); | |
1907 | } | |
1908 | EXPORT_SYMBOL(filp_open); | |
1909 | ||
1910 | /** | |
1911 | * lookup_create - lookup a dentry, creating it if it doesn't exist | |
1912 | * @nd: nameidata info | |
1913 | * @is_dir: directory flag | |
1914 | * | |
1915 | * Simple function to lookup and return a dentry and create it | |
1916 | * if it doesn't exist. Is SMP-safe. | |
1917 | * | |
1918 | * Returns with nd->path.dentry->d_inode->i_mutex locked. | |
1919 | */ | |
1920 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) | |
1921 | { | |
1922 | struct dentry *dentry = ERR_PTR(-EEXIST); | |
1923 | ||
1924 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
1925 | /* | |
1926 | * Yucky last component or no last component at all? | |
1927 | * (foo/., foo/.., /////) | |
1928 | */ | |
1929 | if (nd->last_type != LAST_NORM) | |
1930 | goto fail; | |
1931 | nd->flags &= ~LOOKUP_PARENT; | |
1932 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
1933 | nd->intent.open.flags = O_EXCL; | |
1934 | ||
1935 | /* | |
1936 | * Do the final lookup. | |
1937 | */ | |
1938 | dentry = lookup_hash(nd); | |
1939 | if (IS_ERR(dentry)) | |
1940 | goto fail; | |
1941 | ||
1942 | if (dentry->d_inode) | |
1943 | goto eexist; | |
1944 | /* | |
1945 | * Special case - lookup gave negative, but... we had foo/bar/ | |
1946 | * From the vfs_mknod() POV we just have a negative dentry - | |
1947 | * all is fine. Let's be bastards - you had / on the end, you've | |
1948 | * been asking for (non-existent) directory. -ENOENT for you. | |
1949 | */ | |
1950 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { | |
1951 | dput(dentry); | |
1952 | dentry = ERR_PTR(-ENOENT); | |
1953 | } | |
1954 | return dentry; | |
1955 | eexist: | |
1956 | dput(dentry); | |
1957 | dentry = ERR_PTR(-EEXIST); | |
1958 | fail: | |
1959 | return dentry; | |
1960 | } | |
1961 | EXPORT_SYMBOL_GPL(lookup_create); | |
1962 | ||
1963 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) | |
1964 | { | |
1965 | int error = may_create(dir, dentry); | |
1966 | ||
1967 | if (error) | |
1968 | return error; | |
1969 | ||
1970 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) | |
1971 | return -EPERM; | |
1972 | ||
1973 | if (!dir->i_op->mknod) | |
1974 | return -EPERM; | |
1975 | ||
1976 | error = devcgroup_inode_mknod(mode, dev); | |
1977 | if (error) | |
1978 | return error; | |
1979 | ||
1980 | error = security_inode_mknod(dir, dentry, mode, dev); | |
1981 | if (error) | |
1982 | return error; | |
1983 | ||
1984 | vfs_dq_init(dir); | |
1985 | error = dir->i_op->mknod(dir, dentry, mode, dev); | |
1986 | if (!error) | |
1987 | fsnotify_create(dir, dentry); | |
1988 | return error; | |
1989 | } | |
1990 | ||
1991 | static int may_mknod(mode_t mode) | |
1992 | { | |
1993 | switch (mode & S_IFMT) { | |
1994 | case S_IFREG: | |
1995 | case S_IFCHR: | |
1996 | case S_IFBLK: | |
1997 | case S_IFIFO: | |
1998 | case S_IFSOCK: | |
1999 | case 0: /* zero mode translates to S_IFREG */ | |
2000 | return 0; | |
2001 | case S_IFDIR: | |
2002 | return -EPERM; | |
2003 | default: | |
2004 | return -EINVAL; | |
2005 | } | |
2006 | } | |
2007 | ||
2008 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, | |
2009 | unsigned, dev) | |
2010 | { | |
2011 | int error; | |
2012 | char *tmp; | |
2013 | struct dentry *dentry; | |
2014 | struct nameidata nd; | |
2015 | ||
2016 | if (S_ISDIR(mode)) | |
2017 | return -EPERM; | |
2018 | ||
2019 | error = user_path_parent(dfd, filename, &nd, &tmp); | |
2020 | if (error) | |
2021 | return error; | |
2022 | ||
2023 | dentry = lookup_create(&nd, 0); | |
2024 | if (IS_ERR(dentry)) { | |
2025 | error = PTR_ERR(dentry); | |
2026 | goto out_unlock; | |
2027 | } | |
2028 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) | |
2029 | mode &= ~current_umask(); | |
2030 | error = may_mknod(mode); | |
2031 | if (error) | |
2032 | goto out_dput; | |
2033 | error = mnt_want_write(nd.path.mnt); | |
2034 | if (error) | |
2035 | goto out_dput; | |
2036 | error = security_path_mknod(&nd.path, dentry, mode, dev); | |
2037 | if (error) | |
2038 | goto out_drop_write; | |
2039 | switch (mode & S_IFMT) { | |
2040 | case 0: case S_IFREG: | |
2041 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); | |
2042 | break; | |
2043 | case S_IFCHR: case S_IFBLK: | |
2044 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, | |
2045 | new_decode_dev(dev)); | |
2046 | break; | |
2047 | case S_IFIFO: case S_IFSOCK: | |
2048 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); | |
2049 | break; | |
2050 | } | |
2051 | out_drop_write: | |
2052 | mnt_drop_write(nd.path.mnt); | |
2053 | out_dput: | |
2054 | dput(dentry); | |
2055 | out_unlock: | |
2056 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2057 | path_put(&nd.path); | |
2058 | putname(tmp); | |
2059 | ||
2060 | return error; | |
2061 | } | |
2062 | ||
2063 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) | |
2064 | { | |
2065 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
2066 | } | |
2067 | ||
2068 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
2069 | { | |
2070 | int error = may_create(dir, dentry); | |
2071 | ||
2072 | if (error) | |
2073 | return error; | |
2074 | ||
2075 | if (!dir->i_op->mkdir) | |
2076 | return -EPERM; | |
2077 | ||
2078 | mode &= (S_IRWXUGO|S_ISVTX); | |
2079 | error = security_inode_mkdir(dir, dentry, mode); | |
2080 | if (error) | |
2081 | return error; | |
2082 | ||
2083 | vfs_dq_init(dir); | |
2084 | error = dir->i_op->mkdir(dir, dentry, mode); | |
2085 | if (!error) | |
2086 | fsnotify_mkdir(dir, dentry); | |
2087 | return error; | |
2088 | } | |
2089 | ||
2090 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) | |
2091 | { | |
2092 | int error = 0; | |
2093 | char * tmp; | |
2094 | struct dentry *dentry; | |
2095 | struct nameidata nd; | |
2096 | ||
2097 | error = user_path_parent(dfd, pathname, &nd, &tmp); | |
2098 | if (error) | |
2099 | goto out_err; | |
2100 | ||
2101 | dentry = lookup_create(&nd, 1); | |
2102 | error = PTR_ERR(dentry); | |
2103 | if (IS_ERR(dentry)) | |
2104 | goto out_unlock; | |
2105 | ||
2106 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) | |
2107 | mode &= ~current_umask(); | |
2108 | error = mnt_want_write(nd.path.mnt); | |
2109 | if (error) | |
2110 | goto out_dput; | |
2111 | error = security_path_mkdir(&nd.path, dentry, mode); | |
2112 | if (error) | |
2113 | goto out_drop_write; | |
2114 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); | |
2115 | out_drop_write: | |
2116 | mnt_drop_write(nd.path.mnt); | |
2117 | out_dput: | |
2118 | dput(dentry); | |
2119 | out_unlock: | |
2120 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2121 | path_put(&nd.path); | |
2122 | putname(tmp); | |
2123 | out_err: | |
2124 | return error; | |
2125 | } | |
2126 | ||
2127 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) | |
2128 | { | |
2129 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * We try to drop the dentry early: we should have | |
2134 | * a usage count of 2 if we're the only user of this | |
2135 | * dentry, and if that is true (possibly after pruning | |
2136 | * the dcache), then we drop the dentry now. | |
2137 | * | |
2138 | * A low-level filesystem can, if it choses, legally | |
2139 | * do a | |
2140 | * | |
2141 | * if (!d_unhashed(dentry)) | |
2142 | * return -EBUSY; | |
2143 | * | |
2144 | * if it cannot handle the case of removing a directory | |
2145 | * that is still in use by something else.. | |
2146 | */ | |
2147 | void dentry_unhash(struct dentry *dentry) | |
2148 | { | |
2149 | dget(dentry); | |
2150 | shrink_dcache_parent(dentry); | |
2151 | spin_lock(&dcache_lock); | |
2152 | spin_lock(&dentry->d_lock); | |
2153 | if (atomic_read(&dentry->d_count) == 2) | |
2154 | __d_drop(dentry); | |
2155 | spin_unlock(&dentry->d_lock); | |
2156 | spin_unlock(&dcache_lock); | |
2157 | } | |
2158 | ||
2159 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2160 | { | |
2161 | int error = may_delete(dir, dentry, 1); | |
2162 | ||
2163 | if (error) | |
2164 | return error; | |
2165 | ||
2166 | if (!dir->i_op->rmdir) | |
2167 | return -EPERM; | |
2168 | ||
2169 | vfs_dq_init(dir); | |
2170 | ||
2171 | mutex_lock(&dentry->d_inode->i_mutex); | |
2172 | dentry_unhash(dentry); | |
2173 | if (d_mountpoint(dentry)) | |
2174 | error = -EBUSY; | |
2175 | else { | |
2176 | error = security_inode_rmdir(dir, dentry); | |
2177 | if (!error) { | |
2178 | error = dir->i_op->rmdir(dir, dentry); | |
2179 | if (!error) | |
2180 | dentry->d_inode->i_flags |= S_DEAD; | |
2181 | } | |
2182 | } | |
2183 | mutex_unlock(&dentry->d_inode->i_mutex); | |
2184 | if (!error) { | |
2185 | d_delete(dentry); | |
2186 | } | |
2187 | dput(dentry); | |
2188 | ||
2189 | return error; | |
2190 | } | |
2191 | ||
2192 | static long do_rmdir(int dfd, const char __user *pathname) | |
2193 | { | |
2194 | int error = 0; | |
2195 | char * name; | |
2196 | struct dentry *dentry; | |
2197 | struct nameidata nd; | |
2198 | ||
2199 | error = user_path_parent(dfd, pathname, &nd, &name); | |
2200 | if (error) | |
2201 | return error; | |
2202 | ||
2203 | switch(nd.last_type) { | |
2204 | case LAST_DOTDOT: | |
2205 | error = -ENOTEMPTY; | |
2206 | goto exit1; | |
2207 | case LAST_DOT: | |
2208 | error = -EINVAL; | |
2209 | goto exit1; | |
2210 | case LAST_ROOT: | |
2211 | error = -EBUSY; | |
2212 | goto exit1; | |
2213 | } | |
2214 | ||
2215 | nd.flags &= ~LOOKUP_PARENT; | |
2216 | ||
2217 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
2218 | dentry = lookup_hash(&nd); | |
2219 | error = PTR_ERR(dentry); | |
2220 | if (IS_ERR(dentry)) | |
2221 | goto exit2; | |
2222 | error = mnt_want_write(nd.path.mnt); | |
2223 | if (error) | |
2224 | goto exit3; | |
2225 | error = security_path_rmdir(&nd.path, dentry); | |
2226 | if (error) | |
2227 | goto exit4; | |
2228 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); | |
2229 | exit4: | |
2230 | mnt_drop_write(nd.path.mnt); | |
2231 | exit3: | |
2232 | dput(dentry); | |
2233 | exit2: | |
2234 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2235 | exit1: | |
2236 | path_put(&nd.path); | |
2237 | putname(name); | |
2238 | return error; | |
2239 | } | |
2240 | ||
2241 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) | |
2242 | { | |
2243 | return do_rmdir(AT_FDCWD, pathname); | |
2244 | } | |
2245 | ||
2246 | int vfs_unlink(struct inode *dir, struct dentry *dentry) | |
2247 | { | |
2248 | int error = may_delete(dir, dentry, 0); | |
2249 | ||
2250 | if (error) | |
2251 | return error; | |
2252 | ||
2253 | if (!dir->i_op->unlink) | |
2254 | return -EPERM; | |
2255 | ||
2256 | vfs_dq_init(dir); | |
2257 | ||
2258 | mutex_lock(&dentry->d_inode->i_mutex); | |
2259 | if (d_mountpoint(dentry)) | |
2260 | error = -EBUSY; | |
2261 | else { | |
2262 | error = security_inode_unlink(dir, dentry); | |
2263 | if (!error) { | |
2264 | error = dir->i_op->unlink(dir, dentry); | |
2265 | if (!error) | |
2266 | dentry->d_inode->i_flags |= S_DEAD; | |
2267 | } | |
2268 | } | |
2269 | mutex_unlock(&dentry->d_inode->i_mutex); | |
2270 | ||
2271 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
2272 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
2273 | fsnotify_link_count(dentry->d_inode); | |
2274 | d_delete(dentry); | |
2275 | } | |
2276 | ||
2277 | return error; | |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * Make sure that the actual truncation of the file will occur outside its | |
2282 | * directory's i_mutex. Truncate can take a long time if there is a lot of | |
2283 | * writeout happening, and we don't want to prevent access to the directory | |
2284 | * while waiting on the I/O. | |
2285 | */ | |
2286 | static long do_unlinkat(int dfd, const char __user *pathname) | |
2287 | { | |
2288 | int error; | |
2289 | char *name; | |
2290 | struct dentry *dentry; | |
2291 | struct nameidata nd; | |
2292 | struct inode *inode = NULL; | |
2293 | ||
2294 | error = user_path_parent(dfd, pathname, &nd, &name); | |
2295 | if (error) | |
2296 | return error; | |
2297 | ||
2298 | error = -EISDIR; | |
2299 | if (nd.last_type != LAST_NORM) | |
2300 | goto exit1; | |
2301 | ||
2302 | nd.flags &= ~LOOKUP_PARENT; | |
2303 | ||
2304 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); | |
2305 | dentry = lookup_hash(&nd); | |
2306 | error = PTR_ERR(dentry); | |
2307 | if (!IS_ERR(dentry)) { | |
2308 | /* Why not before? Because we want correct error value */ | |
2309 | if (nd.last.name[nd.last.len]) | |
2310 | goto slashes; | |
2311 | inode = dentry->d_inode; | |
2312 | if (inode) | |
2313 | atomic_inc(&inode->i_count); | |
2314 | error = mnt_want_write(nd.path.mnt); | |
2315 | if (error) | |
2316 | goto exit2; | |
2317 | error = security_path_unlink(&nd.path, dentry); | |
2318 | if (error) | |
2319 | goto exit3; | |
2320 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); | |
2321 | exit3: | |
2322 | mnt_drop_write(nd.path.mnt); | |
2323 | exit2: | |
2324 | dput(dentry); | |
2325 | } | |
2326 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2327 | if (inode) | |
2328 | iput(inode); /* truncate the inode here */ | |
2329 | exit1: | |
2330 | path_put(&nd.path); | |
2331 | putname(name); | |
2332 | return error; | |
2333 | ||
2334 | slashes: | |
2335 | error = !dentry->d_inode ? -ENOENT : | |
2336 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; | |
2337 | goto exit2; | |
2338 | } | |
2339 | ||
2340 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) | |
2341 | { | |
2342 | if ((flag & ~AT_REMOVEDIR) != 0) | |
2343 | return -EINVAL; | |
2344 | ||
2345 | if (flag & AT_REMOVEDIR) | |
2346 | return do_rmdir(dfd, pathname); | |
2347 | ||
2348 | return do_unlinkat(dfd, pathname); | |
2349 | } | |
2350 | ||
2351 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) | |
2352 | { | |
2353 | return do_unlinkat(AT_FDCWD, pathname); | |
2354 | } | |
2355 | ||
2356 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) | |
2357 | { | |
2358 | int error = may_create(dir, dentry); | |
2359 | ||
2360 | if (error) | |
2361 | return error; | |
2362 | ||
2363 | if (!dir->i_op->symlink) | |
2364 | return -EPERM; | |
2365 | ||
2366 | error = security_inode_symlink(dir, dentry, oldname); | |
2367 | if (error) | |
2368 | return error; | |
2369 | ||
2370 | vfs_dq_init(dir); | |
2371 | error = dir->i_op->symlink(dir, dentry, oldname); | |
2372 | if (!error) | |
2373 | fsnotify_create(dir, dentry); | |
2374 | return error; | |
2375 | } | |
2376 | ||
2377 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, | |
2378 | int, newdfd, const char __user *, newname) | |
2379 | { | |
2380 | int error; | |
2381 | char *from; | |
2382 | char *to; | |
2383 | struct dentry *dentry; | |
2384 | struct nameidata nd; | |
2385 | ||
2386 | from = getname(oldname); | |
2387 | if (IS_ERR(from)) | |
2388 | return PTR_ERR(from); | |
2389 | ||
2390 | error = user_path_parent(newdfd, newname, &nd, &to); | |
2391 | if (error) | |
2392 | goto out_putname; | |
2393 | ||
2394 | dentry = lookup_create(&nd, 0); | |
2395 | error = PTR_ERR(dentry); | |
2396 | if (IS_ERR(dentry)) | |
2397 | goto out_unlock; | |
2398 | ||
2399 | error = mnt_want_write(nd.path.mnt); | |
2400 | if (error) | |
2401 | goto out_dput; | |
2402 | error = security_path_symlink(&nd.path, dentry, from); | |
2403 | if (error) | |
2404 | goto out_drop_write; | |
2405 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); | |
2406 | out_drop_write: | |
2407 | mnt_drop_write(nd.path.mnt); | |
2408 | out_dput: | |
2409 | dput(dentry); | |
2410 | out_unlock: | |
2411 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2412 | path_put(&nd.path); | |
2413 | putname(to); | |
2414 | out_putname: | |
2415 | putname(from); | |
2416 | return error; | |
2417 | } | |
2418 | ||
2419 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) | |
2420 | { | |
2421 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
2422 | } | |
2423 | ||
2424 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) | |
2425 | { | |
2426 | struct inode *inode = old_dentry->d_inode; | |
2427 | int error; | |
2428 | ||
2429 | if (!inode) | |
2430 | return -ENOENT; | |
2431 | ||
2432 | error = may_create(dir, new_dentry); | |
2433 | if (error) | |
2434 | return error; | |
2435 | ||
2436 | if (dir->i_sb != inode->i_sb) | |
2437 | return -EXDEV; | |
2438 | ||
2439 | /* | |
2440 | * A link to an append-only or immutable file cannot be created. | |
2441 | */ | |
2442 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2443 | return -EPERM; | |
2444 | if (!dir->i_op->link) | |
2445 | return -EPERM; | |
2446 | if (S_ISDIR(inode->i_mode)) | |
2447 | return -EPERM; | |
2448 | ||
2449 | error = security_inode_link(old_dentry, dir, new_dentry); | |
2450 | if (error) | |
2451 | return error; | |
2452 | ||
2453 | mutex_lock(&inode->i_mutex); | |
2454 | vfs_dq_init(dir); | |
2455 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
2456 | mutex_unlock(&inode->i_mutex); | |
2457 | if (!error) | |
2458 | fsnotify_link(dir, inode, new_dentry); | |
2459 | return error; | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * Hardlinks are often used in delicate situations. We avoid | |
2464 | * security-related surprises by not following symlinks on the | |
2465 | * newname. --KAB | |
2466 | * | |
2467 | * We don't follow them on the oldname either to be compatible | |
2468 | * with linux 2.0, and to avoid hard-linking to directories | |
2469 | * and other special files. --ADM | |
2470 | */ | |
2471 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, | |
2472 | int, newdfd, const char __user *, newname, int, flags) | |
2473 | { | |
2474 | struct dentry *new_dentry; | |
2475 | struct nameidata nd; | |
2476 | struct path old_path; | |
2477 | int error; | |
2478 | char *to; | |
2479 | ||
2480 | if ((flags & ~AT_SYMLINK_FOLLOW) != 0) | |
2481 | return -EINVAL; | |
2482 | ||
2483 | error = user_path_at(olddfd, oldname, | |
2484 | flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, | |
2485 | &old_path); | |
2486 | if (error) | |
2487 | return error; | |
2488 | ||
2489 | error = user_path_parent(newdfd, newname, &nd, &to); | |
2490 | if (error) | |
2491 | goto out; | |
2492 | error = -EXDEV; | |
2493 | if (old_path.mnt != nd.path.mnt) | |
2494 | goto out_release; | |
2495 | new_dentry = lookup_create(&nd, 0); | |
2496 | error = PTR_ERR(new_dentry); | |
2497 | if (IS_ERR(new_dentry)) | |
2498 | goto out_unlock; | |
2499 | error = mnt_want_write(nd.path.mnt); | |
2500 | if (error) | |
2501 | goto out_dput; | |
2502 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); | |
2503 | if (error) | |
2504 | goto out_drop_write; | |
2505 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); | |
2506 | out_drop_write: | |
2507 | mnt_drop_write(nd.path.mnt); | |
2508 | out_dput: | |
2509 | dput(new_dentry); | |
2510 | out_unlock: | |
2511 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); | |
2512 | out_release: | |
2513 | path_put(&nd.path); | |
2514 | putname(to); | |
2515 | out: | |
2516 | path_put(&old_path); | |
2517 | ||
2518 | return error; | |
2519 | } | |
2520 | ||
2521 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) | |
2522 | { | |
2523 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
2524 | } | |
2525 | ||
2526 | /* | |
2527 | * The worst of all namespace operations - renaming directory. "Perverted" | |
2528 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
2529 | * Problems: | |
2530 | * a) we can get into loop creation. Check is done in is_subdir(). | |
2531 | * b) race potential - two innocent renames can create a loop together. | |
2532 | * That's where 4.4 screws up. Current fix: serialization on | |
2533 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another | |
2534 | * story. | |
2535 | * c) we have to lock _three_ objects - parents and victim (if it exists). | |
2536 | * And that - after we got ->i_mutex on parents (until then we don't know | |
2537 | * whether the target exists). Solution: try to be smart with locking | |
2538 | * order for inodes. We rely on the fact that tree topology may change | |
2539 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we | |
2540 | * move will be locked. Thus we can rank directories by the tree | |
2541 | * (ancestors first) and rank all non-directories after them. | |
2542 | * That works since everybody except rename does "lock parent, lookup, | |
2543 | * lock child" and rename is under ->s_vfs_rename_mutex. | |
2544 | * HOWEVER, it relies on the assumption that any object with ->lookup() | |
2545 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
2546 | * we'd better make sure that there's no link(2) for them. | |
2547 | * d) some filesystems don't support opened-but-unlinked directories, | |
2548 | * either because of layout or because they are not ready to deal with | |
2549 | * all cases correctly. The latter will be fixed (taking this sort of | |
2550 | * stuff into VFS), but the former is not going away. Solution: the same | |
2551 | * trick as in rmdir(). | |
2552 | * e) conversion from fhandle to dentry may come in the wrong moment - when | |
2553 | * we are removing the target. Solution: we will have to grab ->i_mutex | |
2554 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on | |
2555 | * ->i_mutex on parents, which works but leads to some truely excessive | |
2556 | * locking]. | |
2557 | */ | |
2558 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, | |
2559 | struct inode *new_dir, struct dentry *new_dentry) | |
2560 | { | |
2561 | int error = 0; | |
2562 | struct inode *target; | |
2563 | ||
2564 | /* | |
2565 | * If we are going to change the parent - check write permissions, | |
2566 | * we'll need to flip '..'. | |
2567 | */ | |
2568 | if (new_dir != old_dir) { | |
2569 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); | |
2570 | if (error) | |
2571 | return error; | |
2572 | } | |
2573 | ||
2574 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2575 | if (error) | |
2576 | return error; | |
2577 | ||
2578 | target = new_dentry->d_inode; | |
2579 | if (target) { | |
2580 | mutex_lock(&target->i_mutex); | |
2581 | dentry_unhash(new_dentry); | |
2582 | } | |
2583 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) | |
2584 | error = -EBUSY; | |
2585 | else | |
2586 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
2587 | if (target) { | |
2588 | if (!error) | |
2589 | target->i_flags |= S_DEAD; | |
2590 | mutex_unlock(&target->i_mutex); | |
2591 | if (d_unhashed(new_dentry)) | |
2592 | d_rehash(new_dentry); | |
2593 | dput(new_dentry); | |
2594 | } | |
2595 | if (!error) | |
2596 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
2597 | d_move(old_dentry,new_dentry); | |
2598 | return error; | |
2599 | } | |
2600 | ||
2601 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, | |
2602 | struct inode *new_dir, struct dentry *new_dentry) | |
2603 | { | |
2604 | struct inode *target; | |
2605 | int error; | |
2606 | ||
2607 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2608 | if (error) | |
2609 | return error; | |
2610 | ||
2611 | dget(new_dentry); | |
2612 | target = new_dentry->d_inode; | |
2613 | if (target) | |
2614 | mutex_lock(&target->i_mutex); | |
2615 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) | |
2616 | error = -EBUSY; | |
2617 | else | |
2618 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); | |
2619 | if (!error) { | |
2620 | if (target) | |
2621 | target->i_flags |= S_DEAD; | |
2622 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) | |
2623 | d_move(old_dentry, new_dentry); | |
2624 | } | |
2625 | if (target) | |
2626 | mutex_unlock(&target->i_mutex); | |
2627 | dput(new_dentry); | |
2628 | return error; | |
2629 | } | |
2630 | ||
2631 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2632 | struct inode *new_dir, struct dentry *new_dentry) | |
2633 | { | |
2634 | int error; | |
2635 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); | |
2636 | const char *old_name; | |
2637 | ||
2638 | if (old_dentry->d_inode == new_dentry->d_inode) | |
2639 | return 0; | |
2640 | ||
2641 | error = may_delete(old_dir, old_dentry, is_dir); | |
2642 | if (error) | |
2643 | return error; | |
2644 | ||
2645 | if (!new_dentry->d_inode) | |
2646 | error = may_create(new_dir, new_dentry); | |
2647 | else | |
2648 | error = may_delete(new_dir, new_dentry, is_dir); | |
2649 | if (error) | |
2650 | return error; | |
2651 | ||
2652 | if (!old_dir->i_op->rename) | |
2653 | return -EPERM; | |
2654 | ||
2655 | vfs_dq_init(old_dir); | |
2656 | vfs_dq_init(new_dir); | |
2657 | ||
2658 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); | |
2659 | ||
2660 | if (is_dir) | |
2661 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); | |
2662 | else | |
2663 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); | |
2664 | if (!error) | |
2665 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
2666 | new_dentry->d_inode, old_dentry); | |
2667 | fsnotify_oldname_free(old_name); | |
2668 | ||
2669 | return error; | |
2670 | } | |
2671 | ||
2672 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, | |
2673 | int, newdfd, const char __user *, newname) | |
2674 | { | |
2675 | struct dentry *old_dir, *new_dir; | |
2676 | struct dentry *old_dentry, *new_dentry; | |
2677 | struct dentry *trap; | |
2678 | struct nameidata oldnd, newnd; | |
2679 | char *from; | |
2680 | char *to; | |
2681 | int error; | |
2682 | ||
2683 | error = user_path_parent(olddfd, oldname, &oldnd, &from); | |
2684 | if (error) | |
2685 | goto exit; | |
2686 | ||
2687 | error = user_path_parent(newdfd, newname, &newnd, &to); | |
2688 | if (error) | |
2689 | goto exit1; | |
2690 | ||
2691 | error = -EXDEV; | |
2692 | if (oldnd.path.mnt != newnd.path.mnt) | |
2693 | goto exit2; | |
2694 | ||
2695 | old_dir = oldnd.path.dentry; | |
2696 | error = -EBUSY; | |
2697 | if (oldnd.last_type != LAST_NORM) | |
2698 | goto exit2; | |
2699 | ||
2700 | new_dir = newnd.path.dentry; | |
2701 | if (newnd.last_type != LAST_NORM) | |
2702 | goto exit2; | |
2703 | ||
2704 | oldnd.flags &= ~LOOKUP_PARENT; | |
2705 | newnd.flags &= ~LOOKUP_PARENT; | |
2706 | newnd.flags |= LOOKUP_RENAME_TARGET; | |
2707 | ||
2708 | trap = lock_rename(new_dir, old_dir); | |
2709 | ||
2710 | old_dentry = lookup_hash(&oldnd); | |
2711 | error = PTR_ERR(old_dentry); | |
2712 | if (IS_ERR(old_dentry)) | |
2713 | goto exit3; | |
2714 | /* source must exist */ | |
2715 | error = -ENOENT; | |
2716 | if (!old_dentry->d_inode) | |
2717 | goto exit4; | |
2718 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
2719 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { | |
2720 | error = -ENOTDIR; | |
2721 | if (oldnd.last.name[oldnd.last.len]) | |
2722 | goto exit4; | |
2723 | if (newnd.last.name[newnd.last.len]) | |
2724 | goto exit4; | |
2725 | } | |
2726 | /* source should not be ancestor of target */ | |
2727 | error = -EINVAL; | |
2728 | if (old_dentry == trap) | |
2729 | goto exit4; | |
2730 | new_dentry = lookup_hash(&newnd); | |
2731 | error = PTR_ERR(new_dentry); | |
2732 | if (IS_ERR(new_dentry)) | |
2733 | goto exit4; | |
2734 | /* target should not be an ancestor of source */ | |
2735 | error = -ENOTEMPTY; | |
2736 | if (new_dentry == trap) | |
2737 | goto exit5; | |
2738 | ||
2739 | error = mnt_want_write(oldnd.path.mnt); | |
2740 | if (error) | |
2741 | goto exit5; | |
2742 | error = security_path_rename(&oldnd.path, old_dentry, | |
2743 | &newnd.path, new_dentry); | |
2744 | if (error) | |
2745 | goto exit6; | |
2746 | error = vfs_rename(old_dir->d_inode, old_dentry, | |
2747 | new_dir->d_inode, new_dentry); | |
2748 | exit6: | |
2749 | mnt_drop_write(oldnd.path.mnt); | |
2750 | exit5: | |
2751 | dput(new_dentry); | |
2752 | exit4: | |
2753 | dput(old_dentry); | |
2754 | exit3: | |
2755 | unlock_rename(new_dir, old_dir); | |
2756 | exit2: | |
2757 | path_put(&newnd.path); | |
2758 | putname(to); | |
2759 | exit1: | |
2760 | path_put(&oldnd.path); | |
2761 | putname(from); | |
2762 | exit: | |
2763 | return error; | |
2764 | } | |
2765 | ||
2766 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) | |
2767 | { | |
2768 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); | |
2769 | } | |
2770 | ||
2771 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) | |
2772 | { | |
2773 | int len; | |
2774 | ||
2775 | len = PTR_ERR(link); | |
2776 | if (IS_ERR(link)) | |
2777 | goto out; | |
2778 | ||
2779 | len = strlen(link); | |
2780 | if (len > (unsigned) buflen) | |
2781 | len = buflen; | |
2782 | if (copy_to_user(buffer, link, len)) | |
2783 | len = -EFAULT; | |
2784 | out: | |
2785 | return len; | |
2786 | } | |
2787 | ||
2788 | /* | |
2789 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
2790 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not | |
2791 | * using) it for any given inode is up to filesystem. | |
2792 | */ | |
2793 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
2794 | { | |
2795 | struct nameidata nd; | |
2796 | void *cookie; | |
2797 | int res; | |
2798 | ||
2799 | nd.depth = 0; | |
2800 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); | |
2801 | if (IS_ERR(cookie)) | |
2802 | return PTR_ERR(cookie); | |
2803 | ||
2804 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); | |
2805 | if (dentry->d_inode->i_op->put_link) | |
2806 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); | |
2807 | return res; | |
2808 | } | |
2809 | ||
2810 | int vfs_follow_link(struct nameidata *nd, const char *link) | |
2811 | { | |
2812 | return __vfs_follow_link(nd, link); | |
2813 | } | |
2814 | ||
2815 | /* get the link contents into pagecache */ | |
2816 | static char *page_getlink(struct dentry * dentry, struct page **ppage) | |
2817 | { | |
2818 | char *kaddr; | |
2819 | struct page *page; | |
2820 | struct address_space *mapping = dentry->d_inode->i_mapping; | |
2821 | page = read_mapping_page(mapping, 0, NULL); | |
2822 | if (IS_ERR(page)) | |
2823 | return (char*)page; | |
2824 | *ppage = page; | |
2825 | kaddr = kmap(page); | |
2826 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); | |
2827 | return kaddr; | |
2828 | } | |
2829 | ||
2830 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
2831 | { | |
2832 | struct page *page = NULL; | |
2833 | char *s = page_getlink(dentry, &page); | |
2834 | int res = vfs_readlink(dentry,buffer,buflen,s); | |
2835 | if (page) { | |
2836 | kunmap(page); | |
2837 | page_cache_release(page); | |
2838 | } | |
2839 | return res; | |
2840 | } | |
2841 | ||
2842 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) | |
2843 | { | |
2844 | struct page *page = NULL; | |
2845 | nd_set_link(nd, page_getlink(dentry, &page)); | |
2846 | return page; | |
2847 | } | |
2848 | ||
2849 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) | |
2850 | { | |
2851 | struct page *page = cookie; | |
2852 | ||
2853 | if (page) { | |
2854 | kunmap(page); | |
2855 | page_cache_release(page); | |
2856 | } | |
2857 | } | |
2858 | ||
2859 | /* | |
2860 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
2861 | */ | |
2862 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
2863 | { | |
2864 | struct address_space *mapping = inode->i_mapping; | |
2865 | struct page *page; | |
2866 | void *fsdata; | |
2867 | int err; | |
2868 | char *kaddr; | |
2869 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; | |
2870 | if (nofs) | |
2871 | flags |= AOP_FLAG_NOFS; | |
2872 | ||
2873 | retry: | |
2874 | err = pagecache_write_begin(NULL, mapping, 0, len-1, | |
2875 | flags, &page, &fsdata); | |
2876 | if (err) | |
2877 | goto fail; | |
2878 | ||
2879 | kaddr = kmap_atomic(page, KM_USER0); | |
2880 | memcpy(kaddr, symname, len-1); | |
2881 | kunmap_atomic(kaddr, KM_USER0); | |
2882 | ||
2883 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
2884 | page, fsdata); | |
2885 | if (err < 0) | |
2886 | goto fail; | |
2887 | if (err < len-1) | |
2888 | goto retry; | |
2889 | ||
2890 | mark_inode_dirty(inode); | |
2891 | return 0; | |
2892 | fail: | |
2893 | return err; | |
2894 | } | |
2895 | ||
2896 | int page_symlink(struct inode *inode, const char *symname, int len) | |
2897 | { | |
2898 | return __page_symlink(inode, symname, len, | |
2899 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); | |
2900 | } | |
2901 | ||
2902 | const struct inode_operations page_symlink_inode_operations = { | |
2903 | .readlink = generic_readlink, | |
2904 | .follow_link = page_follow_link_light, | |
2905 | .put_link = page_put_link, | |
2906 | }; | |
2907 | ||
2908 | EXPORT_SYMBOL(user_path_at); | |
2909 | EXPORT_SYMBOL(follow_down); | |
2910 | EXPORT_SYMBOL(follow_up); | |
2911 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ | |
2912 | EXPORT_SYMBOL(getname); | |
2913 | EXPORT_SYMBOL(lock_rename); | |
2914 | EXPORT_SYMBOL(lookup_one_len); | |
2915 | EXPORT_SYMBOL(page_follow_link_light); | |
2916 | EXPORT_SYMBOL(page_put_link); | |
2917 | EXPORT_SYMBOL(page_readlink); | |
2918 | EXPORT_SYMBOL(__page_symlink); | |
2919 | EXPORT_SYMBOL(page_symlink); | |
2920 | EXPORT_SYMBOL(page_symlink_inode_operations); | |
2921 | EXPORT_SYMBOL(path_lookup); | |
2922 | EXPORT_SYMBOL(kern_path); | |
2923 | EXPORT_SYMBOL(vfs_path_lookup); | |
2924 | EXPORT_SYMBOL(inode_permission); | |
2925 | EXPORT_SYMBOL(file_permission); | |
2926 | EXPORT_SYMBOL(unlock_rename); | |
2927 | EXPORT_SYMBOL(vfs_create); | |
2928 | EXPORT_SYMBOL(vfs_follow_link); | |
2929 | EXPORT_SYMBOL(vfs_link); | |
2930 | EXPORT_SYMBOL(vfs_mkdir); | |
2931 | EXPORT_SYMBOL(vfs_mknod); | |
2932 | EXPORT_SYMBOL(generic_permission); | |
2933 | EXPORT_SYMBOL(vfs_readlink); | |
2934 | EXPORT_SYMBOL(vfs_rename); | |
2935 | EXPORT_SYMBOL(vfs_rmdir); | |
2936 | EXPORT_SYMBOL(vfs_symlink); | |
2937 | EXPORT_SYMBOL(vfs_unlink); | |
2938 | EXPORT_SYMBOL(dentry_unhash); | |
2939 | EXPORT_SYMBOL(generic_readlink); |