]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/namei.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Some corrections by tytso. | |
9 | */ | |
10 | ||
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname | |
12 | * lookup logic. | |
13 | */ | |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. | |
15 | */ | |
16 | ||
17 | #include <linux/init.h> | |
18 | #include <linux/export.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/namei.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/fsnotify.h> | |
25 | #include <linux/personality.h> | |
26 | #include <linux/security.h> | |
27 | #include <linux/ima.h> | |
28 | #include <linux/syscalls.h> | |
29 | #include <linux/mount.h> | |
30 | #include <linux/audit.h> | |
31 | #include <linux/capability.h> | |
32 | #include <linux/file.h> | |
33 | #include <linux/fcntl.h> | |
34 | #include <linux/device_cgroup.h> | |
35 | #include <linux/fs_struct.h> | |
36 | #include <linux/posix_acl.h> | |
37 | #include <linux/hash.h> | |
38 | #include <asm/uaccess.h> | |
39 | ||
40 | #include "internal.h" | |
41 | #include "mount.h" | |
42 | ||
43 | /* [Feb-1997 T. Schoebel-Theuer] | |
44 | * Fundamental changes in the pathname lookup mechanisms (namei) | |
45 | * were necessary because of omirr. The reason is that omirr needs | |
46 | * to know the _real_ pathname, not the user-supplied one, in case | |
47 | * of symlinks (and also when transname replacements occur). | |
48 | * | |
49 | * The new code replaces the old recursive symlink resolution with | |
50 | * an iterative one (in case of non-nested symlink chains). It does | |
51 | * this with calls to <fs>_follow_link(). | |
52 | * As a side effect, dir_namei(), _namei() and follow_link() are now | |
53 | * replaced with a single function lookup_dentry() that can handle all | |
54 | * the special cases of the former code. | |
55 | * | |
56 | * With the new dcache, the pathname is stored at each inode, at least as | |
57 | * long as the refcount of the inode is positive. As a side effect, the | |
58 | * size of the dcache depends on the inode cache and thus is dynamic. | |
59 | * | |
60 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink | |
61 | * resolution to correspond with current state of the code. | |
62 | * | |
63 | * Note that the symlink resolution is not *completely* iterative. | |
64 | * There is still a significant amount of tail- and mid- recursion in | |
65 | * the algorithm. Also, note that <fs>_readlink() is not used in | |
66 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() | |
67 | * may return different results than <fs>_follow_link(). Many virtual | |
68 | * filesystems (including /proc) exhibit this behavior. | |
69 | */ | |
70 | ||
71 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: | |
72 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL | |
73 | * and the name already exists in form of a symlink, try to create the new | |
74 | * name indicated by the symlink. The old code always complained that the | |
75 | * name already exists, due to not following the symlink even if its target | |
76 | * is nonexistent. The new semantics affects also mknod() and link() when | |
77 | * the name is a symlink pointing to a non-existent name. | |
78 | * | |
79 | * I don't know which semantics is the right one, since I have no access | |
80 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" | |
81 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the | |
82 | * "old" one. Personally, I think the new semantics is much more logical. | |
83 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing | |
84 | * file does succeed in both HP-UX and SunOs, but not in Solaris | |
85 | * and in the old Linux semantics. | |
86 | */ | |
87 | ||
88 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink | |
89 | * semantics. See the comments in "open_namei" and "do_link" below. | |
90 | * | |
91 | * [10-Sep-98 Alan Modra] Another symlink change. | |
92 | */ | |
93 | ||
94 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: | |
95 | * inside the path - always follow. | |
96 | * in the last component in creation/removal/renaming - never follow. | |
97 | * if LOOKUP_FOLLOW passed - follow. | |
98 | * if the pathname has trailing slashes - follow. | |
99 | * otherwise - don't follow. | |
100 | * (applied in that order). | |
101 | * | |
102 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT | |
103 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. | |
104 | * During the 2.4 we need to fix the userland stuff depending on it - | |
105 | * hopefully we will be able to get rid of that wart in 2.5. So far only | |
106 | * XEmacs seems to be relying on it... | |
107 | */ | |
108 | /* | |
109 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) | |
110 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives | |
111 | * any extra contention... | |
112 | */ | |
113 | ||
114 | /* In order to reduce some races, while at the same time doing additional | |
115 | * checking and hopefully speeding things up, we copy filenames to the | |
116 | * kernel data space before using them.. | |
117 | * | |
118 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). | |
119 | * PATH_MAX includes the nul terminator --RR. | |
120 | */ | |
121 | ||
122 | #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) | |
123 | ||
124 | struct filename * | |
125 | getname_flags(const char __user *filename, int flags, int *empty) | |
126 | { | |
127 | struct filename *result; | |
128 | char *kname; | |
129 | int len; | |
130 | ||
131 | result = audit_reusename(filename); | |
132 | if (result) | |
133 | return result; | |
134 | ||
135 | result = __getname(); | |
136 | if (unlikely(!result)) | |
137 | return ERR_PTR(-ENOMEM); | |
138 | ||
139 | /* | |
140 | * First, try to embed the struct filename inside the names_cache | |
141 | * allocation | |
142 | */ | |
143 | kname = (char *)result->iname; | |
144 | result->name = kname; | |
145 | ||
146 | len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); | |
147 | if (unlikely(len < 0)) { | |
148 | __putname(result); | |
149 | return ERR_PTR(len); | |
150 | } | |
151 | ||
152 | /* | |
153 | * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a | |
154 | * separate struct filename so we can dedicate the entire | |
155 | * names_cache allocation for the pathname, and re-do the copy from | |
156 | * userland. | |
157 | */ | |
158 | if (unlikely(len == EMBEDDED_NAME_MAX)) { | |
159 | const size_t size = offsetof(struct filename, iname[1]); | |
160 | kname = (char *)result; | |
161 | ||
162 | /* | |
163 | * size is chosen that way we to guarantee that | |
164 | * result->iname[0] is within the same object and that | |
165 | * kname can't be equal to result->iname, no matter what. | |
166 | */ | |
167 | result = kzalloc(size, GFP_KERNEL); | |
168 | if (unlikely(!result)) { | |
169 | __putname(kname); | |
170 | return ERR_PTR(-ENOMEM); | |
171 | } | |
172 | result->name = kname; | |
173 | len = strncpy_from_user(kname, filename, PATH_MAX); | |
174 | if (unlikely(len < 0)) { | |
175 | __putname(kname); | |
176 | kfree(result); | |
177 | return ERR_PTR(len); | |
178 | } | |
179 | if (unlikely(len == PATH_MAX)) { | |
180 | __putname(kname); | |
181 | kfree(result); | |
182 | return ERR_PTR(-ENAMETOOLONG); | |
183 | } | |
184 | } | |
185 | ||
186 | result->refcnt = 1; | |
187 | /* The empty path is special. */ | |
188 | if (unlikely(!len)) { | |
189 | if (empty) | |
190 | *empty = 1; | |
191 | if (!(flags & LOOKUP_EMPTY)) { | |
192 | putname(result); | |
193 | return ERR_PTR(-ENOENT); | |
194 | } | |
195 | } | |
196 | ||
197 | result->uptr = filename; | |
198 | result->aname = NULL; | |
199 | audit_getname(result); | |
200 | return result; | |
201 | } | |
202 | ||
203 | struct filename * | |
204 | getname(const char __user * filename) | |
205 | { | |
206 | return getname_flags(filename, 0, NULL); | |
207 | } | |
208 | ||
209 | struct filename * | |
210 | getname_kernel(const char * filename) | |
211 | { | |
212 | struct filename *result; | |
213 | int len = strlen(filename) + 1; | |
214 | ||
215 | result = __getname(); | |
216 | if (unlikely(!result)) | |
217 | return ERR_PTR(-ENOMEM); | |
218 | ||
219 | if (len <= EMBEDDED_NAME_MAX) { | |
220 | result->name = (char *)result->iname; | |
221 | } else if (len <= PATH_MAX) { | |
222 | struct filename *tmp; | |
223 | ||
224 | tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); | |
225 | if (unlikely(!tmp)) { | |
226 | __putname(result); | |
227 | return ERR_PTR(-ENOMEM); | |
228 | } | |
229 | tmp->name = (char *)result; | |
230 | result = tmp; | |
231 | } else { | |
232 | __putname(result); | |
233 | return ERR_PTR(-ENAMETOOLONG); | |
234 | } | |
235 | memcpy((char *)result->name, filename, len); | |
236 | result->uptr = NULL; | |
237 | result->aname = NULL; | |
238 | result->refcnt = 1; | |
239 | audit_getname(result); | |
240 | ||
241 | return result; | |
242 | } | |
243 | ||
244 | void putname(struct filename *name) | |
245 | { | |
246 | BUG_ON(name->refcnt <= 0); | |
247 | ||
248 | if (--name->refcnt > 0) | |
249 | return; | |
250 | ||
251 | if (name->name != name->iname) { | |
252 | __putname(name->name); | |
253 | kfree(name); | |
254 | } else | |
255 | __putname(name); | |
256 | } | |
257 | ||
258 | static int check_acl(struct inode *inode, int mask) | |
259 | { | |
260 | #ifdef CONFIG_FS_POSIX_ACL | |
261 | struct posix_acl *acl; | |
262 | ||
263 | if (mask & MAY_NOT_BLOCK) { | |
264 | acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); | |
265 | if (!acl) | |
266 | return -EAGAIN; | |
267 | /* no ->get_acl() calls in RCU mode... */ | |
268 | if (acl == ACL_NOT_CACHED) | |
269 | return -ECHILD; | |
270 | return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); | |
271 | } | |
272 | ||
273 | acl = get_acl(inode, ACL_TYPE_ACCESS); | |
274 | if (IS_ERR(acl)) | |
275 | return PTR_ERR(acl); | |
276 | if (acl) { | |
277 | int error = posix_acl_permission(inode, acl, mask); | |
278 | posix_acl_release(acl); | |
279 | return error; | |
280 | } | |
281 | #endif | |
282 | ||
283 | return -EAGAIN; | |
284 | } | |
285 | ||
286 | /* | |
287 | * This does the basic permission checking | |
288 | */ | |
289 | static int acl_permission_check(struct inode *inode, int mask) | |
290 | { | |
291 | unsigned int mode = inode->i_mode; | |
292 | ||
293 | if (likely(uid_eq(current_fsuid(), inode->i_uid))) | |
294 | mode >>= 6; | |
295 | else { | |
296 | if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { | |
297 | int error = check_acl(inode, mask); | |
298 | if (error != -EAGAIN) | |
299 | return error; | |
300 | } | |
301 | ||
302 | if (in_group_p(inode->i_gid)) | |
303 | mode >>= 3; | |
304 | } | |
305 | ||
306 | /* | |
307 | * If the DACs are ok we don't need any capability check. | |
308 | */ | |
309 | if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) | |
310 | return 0; | |
311 | return -EACCES; | |
312 | } | |
313 | ||
314 | /** | |
315 | * generic_permission - check for access rights on a Posix-like filesystem | |
316 | * @inode: inode to check access rights for | |
317 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) | |
318 | * | |
319 | * Used to check for read/write/execute permissions on a file. | |
320 | * We use "fsuid" for this, letting us set arbitrary permissions | |
321 | * for filesystem access without changing the "normal" uids which | |
322 | * are used for other things. | |
323 | * | |
324 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk | |
325 | * request cannot be satisfied (eg. requires blocking or too much complexity). | |
326 | * It would then be called again in ref-walk mode. | |
327 | */ | |
328 | int generic_permission(struct inode *inode, int mask) | |
329 | { | |
330 | int ret; | |
331 | ||
332 | /* | |
333 | * Do the basic permission checks. | |
334 | */ | |
335 | ret = acl_permission_check(inode, mask); | |
336 | if (ret != -EACCES) | |
337 | return ret; | |
338 | ||
339 | if (S_ISDIR(inode->i_mode)) { | |
340 | /* DACs are overridable for directories */ | |
341 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
342 | return 0; | |
343 | if (!(mask & MAY_WRITE)) | |
344 | if (capable_wrt_inode_uidgid(inode, | |
345 | CAP_DAC_READ_SEARCH)) | |
346 | return 0; | |
347 | return -EACCES; | |
348 | } | |
349 | /* | |
350 | * Read/write DACs are always overridable. | |
351 | * Executable DACs are overridable when there is | |
352 | * at least one exec bit set. | |
353 | */ | |
354 | if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) | |
355 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) | |
356 | return 0; | |
357 | ||
358 | /* | |
359 | * Searching includes executable on directories, else just read. | |
360 | */ | |
361 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; | |
362 | if (mask == MAY_READ) | |
363 | if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) | |
364 | return 0; | |
365 | ||
366 | return -EACCES; | |
367 | } | |
368 | EXPORT_SYMBOL(generic_permission); | |
369 | ||
370 | /* | |
371 | * We _really_ want to just do "generic_permission()" without | |
372 | * even looking at the inode->i_op values. So we keep a cache | |
373 | * flag in inode->i_opflags, that says "this has not special | |
374 | * permission function, use the fast case". | |
375 | */ | |
376 | static inline int do_inode_permission(struct inode *inode, int mask) | |
377 | { | |
378 | if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { | |
379 | if (likely(inode->i_op->permission)) | |
380 | return inode->i_op->permission(inode, mask); | |
381 | ||
382 | /* This gets set once for the inode lifetime */ | |
383 | spin_lock(&inode->i_lock); | |
384 | inode->i_opflags |= IOP_FASTPERM; | |
385 | spin_unlock(&inode->i_lock); | |
386 | } | |
387 | return generic_permission(inode, mask); | |
388 | } | |
389 | ||
390 | /** | |
391 | * __inode_permission - Check for access rights to a given inode | |
392 | * @inode: Inode to check permission on | |
393 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
394 | * | |
395 | * Check for read/write/execute permissions on an inode. | |
396 | * | |
397 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
398 | * | |
399 | * This does not check for a read-only file system. You probably want | |
400 | * inode_permission(). | |
401 | */ | |
402 | int __inode_permission(struct inode *inode, int mask) | |
403 | { | |
404 | int retval; | |
405 | ||
406 | if (unlikely(mask & MAY_WRITE)) { | |
407 | /* | |
408 | * Nobody gets write access to an immutable file. | |
409 | */ | |
410 | if (IS_IMMUTABLE(inode)) | |
411 | return -EACCES; | |
412 | } | |
413 | ||
414 | retval = do_inode_permission(inode, mask); | |
415 | if (retval) | |
416 | return retval; | |
417 | ||
418 | retval = devcgroup_inode_permission(inode, mask); | |
419 | if (retval) | |
420 | return retval; | |
421 | ||
422 | return security_inode_permission(inode, mask); | |
423 | } | |
424 | EXPORT_SYMBOL(__inode_permission); | |
425 | ||
426 | /** | |
427 | * sb_permission - Check superblock-level permissions | |
428 | * @sb: Superblock of inode to check permission on | |
429 | * @inode: Inode to check permission on | |
430 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
431 | * | |
432 | * Separate out file-system wide checks from inode-specific permission checks. | |
433 | */ | |
434 | static int sb_permission(struct super_block *sb, struct inode *inode, int mask) | |
435 | { | |
436 | if (unlikely(mask & MAY_WRITE)) { | |
437 | umode_t mode = inode->i_mode; | |
438 | ||
439 | /* Nobody gets write access to a read-only fs. */ | |
440 | if ((sb->s_flags & MS_RDONLY) && | |
441 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) | |
442 | return -EROFS; | |
443 | } | |
444 | return 0; | |
445 | } | |
446 | ||
447 | /** | |
448 | * inode_permission - Check for access rights to a given inode | |
449 | * @inode: Inode to check permission on | |
450 | * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) | |
451 | * | |
452 | * Check for read/write/execute permissions on an inode. We use fs[ug]id for | |
453 | * this, letting us set arbitrary permissions for filesystem access without | |
454 | * changing the "normal" UIDs which are used for other things. | |
455 | * | |
456 | * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. | |
457 | */ | |
458 | int inode_permission(struct inode *inode, int mask) | |
459 | { | |
460 | int retval; | |
461 | ||
462 | retval = sb_permission(inode->i_sb, inode, mask); | |
463 | if (retval) | |
464 | return retval; | |
465 | return __inode_permission(inode, mask); | |
466 | } | |
467 | EXPORT_SYMBOL(inode_permission); | |
468 | ||
469 | /** | |
470 | * path_get - get a reference to a path | |
471 | * @path: path to get the reference to | |
472 | * | |
473 | * Given a path increment the reference count to the dentry and the vfsmount. | |
474 | */ | |
475 | void path_get(const struct path *path) | |
476 | { | |
477 | mntget(path->mnt); | |
478 | dget(path->dentry); | |
479 | } | |
480 | EXPORT_SYMBOL(path_get); | |
481 | ||
482 | /** | |
483 | * path_put - put a reference to a path | |
484 | * @path: path to put the reference to | |
485 | * | |
486 | * Given a path decrement the reference count to the dentry and the vfsmount. | |
487 | */ | |
488 | void path_put(const struct path *path) | |
489 | { | |
490 | dput(path->dentry); | |
491 | mntput(path->mnt); | |
492 | } | |
493 | EXPORT_SYMBOL(path_put); | |
494 | ||
495 | #define EMBEDDED_LEVELS 2 | |
496 | struct nameidata { | |
497 | struct path path; | |
498 | struct qstr last; | |
499 | struct path root; | |
500 | struct inode *inode; /* path.dentry.d_inode */ | |
501 | unsigned int flags; | |
502 | unsigned seq, m_seq; | |
503 | int last_type; | |
504 | unsigned depth; | |
505 | int total_link_count; | |
506 | struct saved { | |
507 | struct path link; | |
508 | struct delayed_call done; | |
509 | const char *name; | |
510 | unsigned seq; | |
511 | } *stack, internal[EMBEDDED_LEVELS]; | |
512 | struct filename *name; | |
513 | struct nameidata *saved; | |
514 | struct inode *link_inode; | |
515 | unsigned root_seq; | |
516 | int dfd; | |
517 | }; | |
518 | ||
519 | static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) | |
520 | { | |
521 | struct nameidata *old = current->nameidata; | |
522 | p->stack = p->internal; | |
523 | p->dfd = dfd; | |
524 | p->name = name; | |
525 | p->total_link_count = old ? old->total_link_count : 0; | |
526 | p->saved = old; | |
527 | current->nameidata = p; | |
528 | } | |
529 | ||
530 | static void restore_nameidata(void) | |
531 | { | |
532 | struct nameidata *now = current->nameidata, *old = now->saved; | |
533 | ||
534 | current->nameidata = old; | |
535 | if (old) | |
536 | old->total_link_count = now->total_link_count; | |
537 | if (now->stack != now->internal) | |
538 | kfree(now->stack); | |
539 | } | |
540 | ||
541 | static int __nd_alloc_stack(struct nameidata *nd) | |
542 | { | |
543 | struct saved *p; | |
544 | ||
545 | if (nd->flags & LOOKUP_RCU) { | |
546 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
547 | GFP_ATOMIC); | |
548 | if (unlikely(!p)) | |
549 | return -ECHILD; | |
550 | } else { | |
551 | p= kmalloc(MAXSYMLINKS * sizeof(struct saved), | |
552 | GFP_KERNEL); | |
553 | if (unlikely(!p)) | |
554 | return -ENOMEM; | |
555 | } | |
556 | memcpy(p, nd->internal, sizeof(nd->internal)); | |
557 | nd->stack = p; | |
558 | return 0; | |
559 | } | |
560 | ||
561 | /** | |
562 | * path_connected - Verify that a path->dentry is below path->mnt.mnt_root | |
563 | * @path: nameidate to verify | |
564 | * | |
565 | * Rename can sometimes move a file or directory outside of a bind | |
566 | * mount, path_connected allows those cases to be detected. | |
567 | */ | |
568 | static bool path_connected(const struct path *path) | |
569 | { | |
570 | struct vfsmount *mnt = path->mnt; | |
571 | ||
572 | /* Only bind mounts can have disconnected paths */ | |
573 | if (mnt->mnt_root == mnt->mnt_sb->s_root) | |
574 | return true; | |
575 | ||
576 | return is_subdir(path->dentry, mnt->mnt_root); | |
577 | } | |
578 | ||
579 | static inline int nd_alloc_stack(struct nameidata *nd) | |
580 | { | |
581 | if (likely(nd->depth != EMBEDDED_LEVELS)) | |
582 | return 0; | |
583 | if (likely(nd->stack != nd->internal)) | |
584 | return 0; | |
585 | return __nd_alloc_stack(nd); | |
586 | } | |
587 | ||
588 | static void drop_links(struct nameidata *nd) | |
589 | { | |
590 | int i = nd->depth; | |
591 | while (i--) { | |
592 | struct saved *last = nd->stack + i; | |
593 | do_delayed_call(&last->done); | |
594 | clear_delayed_call(&last->done); | |
595 | } | |
596 | } | |
597 | ||
598 | static void terminate_walk(struct nameidata *nd) | |
599 | { | |
600 | drop_links(nd); | |
601 | if (!(nd->flags & LOOKUP_RCU)) { | |
602 | int i; | |
603 | path_put(&nd->path); | |
604 | for (i = 0; i < nd->depth; i++) | |
605 | path_put(&nd->stack[i].link); | |
606 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
607 | path_put(&nd->root); | |
608 | nd->root.mnt = NULL; | |
609 | } | |
610 | } else { | |
611 | nd->flags &= ~LOOKUP_RCU; | |
612 | if (!(nd->flags & LOOKUP_ROOT)) | |
613 | nd->root.mnt = NULL; | |
614 | rcu_read_unlock(); | |
615 | } | |
616 | nd->depth = 0; | |
617 | } | |
618 | ||
619 | /* path_put is needed afterwards regardless of success or failure */ | |
620 | static bool legitimize_path(struct nameidata *nd, | |
621 | struct path *path, unsigned seq) | |
622 | { | |
623 | int res = __legitimize_mnt(path->mnt, nd->m_seq); | |
624 | if (unlikely(res)) { | |
625 | if (res > 0) | |
626 | path->mnt = NULL; | |
627 | path->dentry = NULL; | |
628 | return false; | |
629 | } | |
630 | if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { | |
631 | path->dentry = NULL; | |
632 | return false; | |
633 | } | |
634 | return !read_seqcount_retry(&path->dentry->d_seq, seq); | |
635 | } | |
636 | ||
637 | static bool legitimize_links(struct nameidata *nd) | |
638 | { | |
639 | int i; | |
640 | for (i = 0; i < nd->depth; i++) { | |
641 | struct saved *last = nd->stack + i; | |
642 | if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { | |
643 | drop_links(nd); | |
644 | nd->depth = i + 1; | |
645 | return false; | |
646 | } | |
647 | } | |
648 | return true; | |
649 | } | |
650 | ||
651 | /* | |
652 | * Path walking has 2 modes, rcu-walk and ref-walk (see | |
653 | * Documentation/filesystems/path-lookup.txt). In situations when we can't | |
654 | * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab | |
655 | * normal reference counts on dentries and vfsmounts to transition to ref-walk | |
656 | * mode. Refcounts are grabbed at the last known good point before rcu-walk | |
657 | * got stuck, so ref-walk may continue from there. If this is not successful | |
658 | * (eg. a seqcount has changed), then failure is returned and it's up to caller | |
659 | * to restart the path walk from the beginning in ref-walk mode. | |
660 | */ | |
661 | ||
662 | /** | |
663 | * unlazy_walk - try to switch to ref-walk mode. | |
664 | * @nd: nameidata pathwalk data | |
665 | * @dentry: child of nd->path.dentry or NULL | |
666 | * @seq: seq number to check dentry against | |
667 | * Returns: 0 on success, -ECHILD on failure | |
668 | * | |
669 | * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry | |
670 | * for ref-walk mode. @dentry must be a path found by a do_lookup call on | |
671 | * @nd or NULL. Must be called from rcu-walk context. | |
672 | * Nothing should touch nameidata between unlazy_walk() failure and | |
673 | * terminate_walk(). | |
674 | */ | |
675 | static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) | |
676 | { | |
677 | struct dentry *parent = nd->path.dentry; | |
678 | ||
679 | BUG_ON(!(nd->flags & LOOKUP_RCU)); | |
680 | ||
681 | nd->flags &= ~LOOKUP_RCU; | |
682 | if (unlikely(!legitimize_links(nd))) | |
683 | goto out2; | |
684 | if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) | |
685 | goto out2; | |
686 | if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) | |
687 | goto out1; | |
688 | ||
689 | /* | |
690 | * For a negative lookup, the lookup sequence point is the parents | |
691 | * sequence point, and it only needs to revalidate the parent dentry. | |
692 | * | |
693 | * For a positive lookup, we need to move both the parent and the | |
694 | * dentry from the RCU domain to be properly refcounted. And the | |
695 | * sequence number in the dentry validates *both* dentry counters, | |
696 | * since we checked the sequence number of the parent after we got | |
697 | * the child sequence number. So we know the parent must still | |
698 | * be valid if the child sequence number is still valid. | |
699 | */ | |
700 | if (!dentry) { | |
701 | if (read_seqcount_retry(&parent->d_seq, nd->seq)) | |
702 | goto out; | |
703 | BUG_ON(nd->inode != parent->d_inode); | |
704 | } else { | |
705 | if (!lockref_get_not_dead(&dentry->d_lockref)) | |
706 | goto out; | |
707 | if (read_seqcount_retry(&dentry->d_seq, seq)) | |
708 | goto drop_dentry; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Sequence counts matched. Now make sure that the root is | |
713 | * still valid and get it if required. | |
714 | */ | |
715 | if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { | |
716 | if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { | |
717 | rcu_read_unlock(); | |
718 | dput(dentry); | |
719 | return -ECHILD; | |
720 | } | |
721 | } | |
722 | ||
723 | rcu_read_unlock(); | |
724 | return 0; | |
725 | ||
726 | drop_dentry: | |
727 | rcu_read_unlock(); | |
728 | dput(dentry); | |
729 | goto drop_root_mnt; | |
730 | out2: | |
731 | nd->path.mnt = NULL; | |
732 | out1: | |
733 | nd->path.dentry = NULL; | |
734 | out: | |
735 | rcu_read_unlock(); | |
736 | drop_root_mnt: | |
737 | if (!(nd->flags & LOOKUP_ROOT)) | |
738 | nd->root.mnt = NULL; | |
739 | return -ECHILD; | |
740 | } | |
741 | ||
742 | static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) | |
743 | { | |
744 | if (unlikely(!legitimize_path(nd, link, seq))) { | |
745 | drop_links(nd); | |
746 | nd->depth = 0; | |
747 | nd->flags &= ~LOOKUP_RCU; | |
748 | nd->path.mnt = NULL; | |
749 | nd->path.dentry = NULL; | |
750 | if (!(nd->flags & LOOKUP_ROOT)) | |
751 | nd->root.mnt = NULL; | |
752 | rcu_read_unlock(); | |
753 | } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { | |
754 | return 0; | |
755 | } | |
756 | path_put(link); | |
757 | return -ECHILD; | |
758 | } | |
759 | ||
760 | static inline int d_revalidate(struct dentry *dentry, unsigned int flags) | |
761 | { | |
762 | return dentry->d_op->d_revalidate(dentry, flags); | |
763 | } | |
764 | ||
765 | /** | |
766 | * complete_walk - successful completion of path walk | |
767 | * @nd: pointer nameidata | |
768 | * | |
769 | * If we had been in RCU mode, drop out of it and legitimize nd->path. | |
770 | * Revalidate the final result, unless we'd already done that during | |
771 | * the path walk or the filesystem doesn't ask for it. Return 0 on | |
772 | * success, -error on failure. In case of failure caller does not | |
773 | * need to drop nd->path. | |
774 | */ | |
775 | static int complete_walk(struct nameidata *nd) | |
776 | { | |
777 | struct dentry *dentry = nd->path.dentry; | |
778 | int status; | |
779 | ||
780 | if (nd->flags & LOOKUP_RCU) { | |
781 | if (!(nd->flags & LOOKUP_ROOT)) | |
782 | nd->root.mnt = NULL; | |
783 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
784 | return -ECHILD; | |
785 | } | |
786 | ||
787 | if (likely(!(nd->flags & LOOKUP_JUMPED))) | |
788 | return 0; | |
789 | ||
790 | if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) | |
791 | return 0; | |
792 | ||
793 | status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); | |
794 | if (status > 0) | |
795 | return 0; | |
796 | ||
797 | if (!status) | |
798 | status = -ESTALE; | |
799 | ||
800 | return status; | |
801 | } | |
802 | ||
803 | static void set_root(struct nameidata *nd) | |
804 | { | |
805 | struct fs_struct *fs = current->fs; | |
806 | ||
807 | if (nd->flags & LOOKUP_RCU) { | |
808 | unsigned seq; | |
809 | ||
810 | do { | |
811 | seq = read_seqcount_begin(&fs->seq); | |
812 | nd->root = fs->root; | |
813 | nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); | |
814 | } while (read_seqcount_retry(&fs->seq, seq)); | |
815 | } else { | |
816 | get_fs_root(fs, &nd->root); | |
817 | } | |
818 | } | |
819 | ||
820 | static void path_put_conditional(struct path *path, struct nameidata *nd) | |
821 | { | |
822 | dput(path->dentry); | |
823 | if (path->mnt != nd->path.mnt) | |
824 | mntput(path->mnt); | |
825 | } | |
826 | ||
827 | static inline void path_to_nameidata(const struct path *path, | |
828 | struct nameidata *nd) | |
829 | { | |
830 | if (!(nd->flags & LOOKUP_RCU)) { | |
831 | dput(nd->path.dentry); | |
832 | if (nd->path.mnt != path->mnt) | |
833 | mntput(nd->path.mnt); | |
834 | } | |
835 | nd->path.mnt = path->mnt; | |
836 | nd->path.dentry = path->dentry; | |
837 | } | |
838 | ||
839 | static int nd_jump_root(struct nameidata *nd) | |
840 | { | |
841 | if (nd->flags & LOOKUP_RCU) { | |
842 | struct dentry *d; | |
843 | nd->path = nd->root; | |
844 | d = nd->path.dentry; | |
845 | nd->inode = d->d_inode; | |
846 | nd->seq = nd->root_seq; | |
847 | if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) | |
848 | return -ECHILD; | |
849 | } else { | |
850 | path_put(&nd->path); | |
851 | nd->path = nd->root; | |
852 | path_get(&nd->path); | |
853 | nd->inode = nd->path.dentry->d_inode; | |
854 | } | |
855 | nd->flags |= LOOKUP_JUMPED; | |
856 | return 0; | |
857 | } | |
858 | ||
859 | /* | |
860 | * Helper to directly jump to a known parsed path from ->get_link, | |
861 | * caller must have taken a reference to path beforehand. | |
862 | */ | |
863 | void nd_jump_link(struct path *path) | |
864 | { | |
865 | struct nameidata *nd = current->nameidata; | |
866 | path_put(&nd->path); | |
867 | ||
868 | nd->path = *path; | |
869 | nd->inode = nd->path.dentry->d_inode; | |
870 | nd->flags |= LOOKUP_JUMPED; | |
871 | } | |
872 | ||
873 | static inline void put_link(struct nameidata *nd) | |
874 | { | |
875 | struct saved *last = nd->stack + --nd->depth; | |
876 | do_delayed_call(&last->done); | |
877 | if (!(nd->flags & LOOKUP_RCU)) | |
878 | path_put(&last->link); | |
879 | } | |
880 | ||
881 | int sysctl_protected_symlinks __read_mostly = 0; | |
882 | int sysctl_protected_hardlinks __read_mostly = 0; | |
883 | ||
884 | /** | |
885 | * may_follow_link - Check symlink following for unsafe situations | |
886 | * @nd: nameidata pathwalk data | |
887 | * | |
888 | * In the case of the sysctl_protected_symlinks sysctl being enabled, | |
889 | * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is | |
890 | * in a sticky world-writable directory. This is to protect privileged | |
891 | * processes from failing races against path names that may change out | |
892 | * from under them by way of other users creating malicious symlinks. | |
893 | * It will permit symlinks to be followed only when outside a sticky | |
894 | * world-writable directory, or when the uid of the symlink and follower | |
895 | * match, or when the directory owner matches the symlink's owner. | |
896 | * | |
897 | * Returns 0 if following the symlink is allowed, -ve on error. | |
898 | */ | |
899 | static inline int may_follow_link(struct nameidata *nd) | |
900 | { | |
901 | const struct inode *inode; | |
902 | const struct inode *parent; | |
903 | ||
904 | if (!sysctl_protected_symlinks) | |
905 | return 0; | |
906 | ||
907 | /* Allowed if owner and follower match. */ | |
908 | inode = nd->link_inode; | |
909 | if (uid_eq(current_cred()->fsuid, inode->i_uid)) | |
910 | return 0; | |
911 | ||
912 | /* Allowed if parent directory not sticky and world-writable. */ | |
913 | parent = nd->inode; | |
914 | if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) | |
915 | return 0; | |
916 | ||
917 | /* Allowed if parent directory and link owner match. */ | |
918 | if (uid_eq(parent->i_uid, inode->i_uid)) | |
919 | return 0; | |
920 | ||
921 | if (nd->flags & LOOKUP_RCU) | |
922 | return -ECHILD; | |
923 | ||
924 | audit_log_link_denied("follow_link", &nd->stack[0].link); | |
925 | return -EACCES; | |
926 | } | |
927 | ||
928 | /** | |
929 | * safe_hardlink_source - Check for safe hardlink conditions | |
930 | * @inode: the source inode to hardlink from | |
931 | * | |
932 | * Return false if at least one of the following conditions: | |
933 | * - inode is not a regular file | |
934 | * - inode is setuid | |
935 | * - inode is setgid and group-exec | |
936 | * - access failure for read and write | |
937 | * | |
938 | * Otherwise returns true. | |
939 | */ | |
940 | static bool safe_hardlink_source(struct inode *inode) | |
941 | { | |
942 | umode_t mode = inode->i_mode; | |
943 | ||
944 | /* Special files should not get pinned to the filesystem. */ | |
945 | if (!S_ISREG(mode)) | |
946 | return false; | |
947 | ||
948 | /* Setuid files should not get pinned to the filesystem. */ | |
949 | if (mode & S_ISUID) | |
950 | return false; | |
951 | ||
952 | /* Executable setgid files should not get pinned to the filesystem. */ | |
953 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
954 | return false; | |
955 | ||
956 | /* Hardlinking to unreadable or unwritable sources is dangerous. */ | |
957 | if (inode_permission(inode, MAY_READ | MAY_WRITE)) | |
958 | return false; | |
959 | ||
960 | return true; | |
961 | } | |
962 | ||
963 | /** | |
964 | * may_linkat - Check permissions for creating a hardlink | |
965 | * @link: the source to hardlink from | |
966 | * | |
967 | * Block hardlink when all of: | |
968 | * - sysctl_protected_hardlinks enabled | |
969 | * - fsuid does not match inode | |
970 | * - hardlink source is unsafe (see safe_hardlink_source() above) | |
971 | * - not CAP_FOWNER in a namespace with the inode owner uid mapped | |
972 | * | |
973 | * Returns 0 if successful, -ve on error. | |
974 | */ | |
975 | static int may_linkat(struct path *link) | |
976 | { | |
977 | struct inode *inode; | |
978 | ||
979 | if (!sysctl_protected_hardlinks) | |
980 | return 0; | |
981 | ||
982 | inode = link->dentry->d_inode; | |
983 | ||
984 | /* Source inode owner (or CAP_FOWNER) can hardlink all they like, | |
985 | * otherwise, it must be a safe source. | |
986 | */ | |
987 | if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) | |
988 | return 0; | |
989 | ||
990 | audit_log_link_denied("linkat", link); | |
991 | return -EPERM; | |
992 | } | |
993 | ||
994 | static __always_inline | |
995 | const char *get_link(struct nameidata *nd) | |
996 | { | |
997 | struct saved *last = nd->stack + nd->depth - 1; | |
998 | struct dentry *dentry = last->link.dentry; | |
999 | struct inode *inode = nd->link_inode; | |
1000 | int error; | |
1001 | const char *res; | |
1002 | ||
1003 | if (!(nd->flags & LOOKUP_RCU)) { | |
1004 | touch_atime(&last->link); | |
1005 | cond_resched(); | |
1006 | } else if (atime_needs_update(&last->link, inode)) { | |
1007 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1008 | return ERR_PTR(-ECHILD); | |
1009 | touch_atime(&last->link); | |
1010 | } | |
1011 | ||
1012 | error = security_inode_follow_link(dentry, inode, | |
1013 | nd->flags & LOOKUP_RCU); | |
1014 | if (unlikely(error)) | |
1015 | return ERR_PTR(error); | |
1016 | ||
1017 | nd->last_type = LAST_BIND; | |
1018 | res = inode->i_link; | |
1019 | if (!res) { | |
1020 | const char * (*get)(struct dentry *, struct inode *, | |
1021 | struct delayed_call *); | |
1022 | get = inode->i_op->get_link; | |
1023 | if (nd->flags & LOOKUP_RCU) { | |
1024 | res = get(NULL, inode, &last->done); | |
1025 | if (res == ERR_PTR(-ECHILD)) { | |
1026 | if (unlikely(unlazy_walk(nd, NULL, 0))) | |
1027 | return ERR_PTR(-ECHILD); | |
1028 | res = get(dentry, inode, &last->done); | |
1029 | } | |
1030 | } else { | |
1031 | res = get(dentry, inode, &last->done); | |
1032 | } | |
1033 | if (IS_ERR_OR_NULL(res)) | |
1034 | return res; | |
1035 | } | |
1036 | if (*res == '/') { | |
1037 | if (!nd->root.mnt) | |
1038 | set_root(nd); | |
1039 | if (unlikely(nd_jump_root(nd))) | |
1040 | return ERR_PTR(-ECHILD); | |
1041 | while (unlikely(*++res == '/')) | |
1042 | ; | |
1043 | } | |
1044 | if (!*res) | |
1045 | res = NULL; | |
1046 | return res; | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * follow_up - Find the mountpoint of path's vfsmount | |
1051 | * | |
1052 | * Given a path, find the mountpoint of its source file system. | |
1053 | * Replace @path with the path of the mountpoint in the parent mount. | |
1054 | * Up is towards /. | |
1055 | * | |
1056 | * Return 1 if we went up a level and 0 if we were already at the | |
1057 | * root. | |
1058 | */ | |
1059 | int follow_up(struct path *path) | |
1060 | { | |
1061 | struct mount *mnt = real_mount(path->mnt); | |
1062 | struct mount *parent; | |
1063 | struct dentry *mountpoint; | |
1064 | ||
1065 | read_seqlock_excl(&mount_lock); | |
1066 | parent = mnt->mnt_parent; | |
1067 | if (parent == mnt) { | |
1068 | read_sequnlock_excl(&mount_lock); | |
1069 | return 0; | |
1070 | } | |
1071 | mntget(&parent->mnt); | |
1072 | mountpoint = dget(mnt->mnt_mountpoint); | |
1073 | read_sequnlock_excl(&mount_lock); | |
1074 | dput(path->dentry); | |
1075 | path->dentry = mountpoint; | |
1076 | mntput(path->mnt); | |
1077 | path->mnt = &parent->mnt; | |
1078 | return 1; | |
1079 | } | |
1080 | EXPORT_SYMBOL(follow_up); | |
1081 | ||
1082 | /* | |
1083 | * Perform an automount | |
1084 | * - return -EISDIR to tell follow_managed() to stop and return the path we | |
1085 | * were called with. | |
1086 | */ | |
1087 | static int follow_automount(struct path *path, struct nameidata *nd, | |
1088 | bool *need_mntput) | |
1089 | { | |
1090 | struct vfsmount *mnt; | |
1091 | int err; | |
1092 | ||
1093 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) | |
1094 | return -EREMOTE; | |
1095 | ||
1096 | /* We don't want to mount if someone's just doing a stat - | |
1097 | * unless they're stat'ing a directory and appended a '/' to | |
1098 | * the name. | |
1099 | * | |
1100 | * We do, however, want to mount if someone wants to open or | |
1101 | * create a file of any type under the mountpoint, wants to | |
1102 | * traverse through the mountpoint or wants to open the | |
1103 | * mounted directory. Also, autofs may mark negative dentries | |
1104 | * as being automount points. These will need the attentions | |
1105 | * of the daemon to instantiate them before they can be used. | |
1106 | */ | |
1107 | if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | | |
1108 | LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && | |
1109 | path->dentry->d_inode) | |
1110 | return -EISDIR; | |
1111 | ||
1112 | nd->total_link_count++; | |
1113 | if (nd->total_link_count >= 40) | |
1114 | return -ELOOP; | |
1115 | ||
1116 | mnt = path->dentry->d_op->d_automount(path); | |
1117 | if (IS_ERR(mnt)) { | |
1118 | /* | |
1119 | * The filesystem is allowed to return -EISDIR here to indicate | |
1120 | * it doesn't want to automount. For instance, autofs would do | |
1121 | * this so that its userspace daemon can mount on this dentry. | |
1122 | * | |
1123 | * However, we can only permit this if it's a terminal point in | |
1124 | * the path being looked up; if it wasn't then the remainder of | |
1125 | * the path is inaccessible and we should say so. | |
1126 | */ | |
1127 | if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) | |
1128 | return -EREMOTE; | |
1129 | return PTR_ERR(mnt); | |
1130 | } | |
1131 | ||
1132 | if (!mnt) /* mount collision */ | |
1133 | return 0; | |
1134 | ||
1135 | if (!*need_mntput) { | |
1136 | /* lock_mount() may release path->mnt on error */ | |
1137 | mntget(path->mnt); | |
1138 | *need_mntput = true; | |
1139 | } | |
1140 | err = finish_automount(mnt, path); | |
1141 | ||
1142 | switch (err) { | |
1143 | case -EBUSY: | |
1144 | /* Someone else made a mount here whilst we were busy */ | |
1145 | return 0; | |
1146 | case 0: | |
1147 | path_put(path); | |
1148 | path->mnt = mnt; | |
1149 | path->dentry = dget(mnt->mnt_root); | |
1150 | return 0; | |
1151 | default: | |
1152 | return err; | |
1153 | } | |
1154 | ||
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * Handle a dentry that is managed in some way. | |
1159 | * - Flagged for transit management (autofs) | |
1160 | * - Flagged as mountpoint | |
1161 | * - Flagged as automount point | |
1162 | * | |
1163 | * This may only be called in refwalk mode. | |
1164 | * | |
1165 | * Serialization is taken care of in namespace.c | |
1166 | */ | |
1167 | static int follow_managed(struct path *path, struct nameidata *nd) | |
1168 | { | |
1169 | struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ | |
1170 | unsigned managed; | |
1171 | bool need_mntput = false; | |
1172 | int ret = 0; | |
1173 | ||
1174 | /* Given that we're not holding a lock here, we retain the value in a | |
1175 | * local variable for each dentry as we look at it so that we don't see | |
1176 | * the components of that value change under us */ | |
1177 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1178 | managed &= DCACHE_MANAGED_DENTRY, | |
1179 | unlikely(managed != 0)) { | |
1180 | /* Allow the filesystem to manage the transit without i_mutex | |
1181 | * being held. */ | |
1182 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1183 | BUG_ON(!path->dentry->d_op); | |
1184 | BUG_ON(!path->dentry->d_op->d_manage); | |
1185 | ret = path->dentry->d_op->d_manage(path->dentry, false); | |
1186 | if (ret < 0) | |
1187 | break; | |
1188 | } | |
1189 | ||
1190 | /* Transit to a mounted filesystem. */ | |
1191 | if (managed & DCACHE_MOUNTED) { | |
1192 | struct vfsmount *mounted = lookup_mnt(path); | |
1193 | if (mounted) { | |
1194 | dput(path->dentry); | |
1195 | if (need_mntput) | |
1196 | mntput(path->mnt); | |
1197 | path->mnt = mounted; | |
1198 | path->dentry = dget(mounted->mnt_root); | |
1199 | need_mntput = true; | |
1200 | continue; | |
1201 | } | |
1202 | ||
1203 | /* Something is mounted on this dentry in another | |
1204 | * namespace and/or whatever was mounted there in this | |
1205 | * namespace got unmounted before lookup_mnt() could | |
1206 | * get it */ | |
1207 | } | |
1208 | ||
1209 | /* Handle an automount point */ | |
1210 | if (managed & DCACHE_NEED_AUTOMOUNT) { | |
1211 | ret = follow_automount(path, nd, &need_mntput); | |
1212 | if (ret < 0) | |
1213 | break; | |
1214 | continue; | |
1215 | } | |
1216 | ||
1217 | /* We didn't change the current path point */ | |
1218 | break; | |
1219 | } | |
1220 | ||
1221 | if (need_mntput && path->mnt == mnt) | |
1222 | mntput(path->mnt); | |
1223 | if (ret == -EISDIR || !ret) | |
1224 | ret = 1; | |
1225 | if (need_mntput) | |
1226 | nd->flags |= LOOKUP_JUMPED; | |
1227 | if (unlikely(ret < 0)) | |
1228 | path_put_conditional(path, nd); | |
1229 | return ret; | |
1230 | } | |
1231 | ||
1232 | int follow_down_one(struct path *path) | |
1233 | { | |
1234 | struct vfsmount *mounted; | |
1235 | ||
1236 | mounted = lookup_mnt(path); | |
1237 | if (mounted) { | |
1238 | dput(path->dentry); | |
1239 | mntput(path->mnt); | |
1240 | path->mnt = mounted; | |
1241 | path->dentry = dget(mounted->mnt_root); | |
1242 | return 1; | |
1243 | } | |
1244 | return 0; | |
1245 | } | |
1246 | EXPORT_SYMBOL(follow_down_one); | |
1247 | ||
1248 | static inline int managed_dentry_rcu(struct dentry *dentry) | |
1249 | { | |
1250 | return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? | |
1251 | dentry->d_op->d_manage(dentry, true) : 0; | |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if | |
1256 | * we meet a managed dentry that would need blocking. | |
1257 | */ | |
1258 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, | |
1259 | struct inode **inode, unsigned *seqp) | |
1260 | { | |
1261 | for (;;) { | |
1262 | struct mount *mounted; | |
1263 | /* | |
1264 | * Don't forget we might have a non-mountpoint managed dentry | |
1265 | * that wants to block transit. | |
1266 | */ | |
1267 | switch (managed_dentry_rcu(path->dentry)) { | |
1268 | case -ECHILD: | |
1269 | default: | |
1270 | return false; | |
1271 | case -EISDIR: | |
1272 | return true; | |
1273 | case 0: | |
1274 | break; | |
1275 | } | |
1276 | ||
1277 | if (!d_mountpoint(path->dentry)) | |
1278 | return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); | |
1279 | ||
1280 | mounted = __lookup_mnt(path->mnt, path->dentry); | |
1281 | if (!mounted) | |
1282 | break; | |
1283 | path->mnt = &mounted->mnt; | |
1284 | path->dentry = mounted->mnt.mnt_root; | |
1285 | nd->flags |= LOOKUP_JUMPED; | |
1286 | *seqp = read_seqcount_begin(&path->dentry->d_seq); | |
1287 | /* | |
1288 | * Update the inode too. We don't need to re-check the | |
1289 | * dentry sequence number here after this d_inode read, | |
1290 | * because a mount-point is always pinned. | |
1291 | */ | |
1292 | *inode = path->dentry->d_inode; | |
1293 | } | |
1294 | return !read_seqretry(&mount_lock, nd->m_seq) && | |
1295 | !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); | |
1296 | } | |
1297 | ||
1298 | static int follow_dotdot_rcu(struct nameidata *nd) | |
1299 | { | |
1300 | struct inode *inode = nd->inode; | |
1301 | ||
1302 | while (1) { | |
1303 | if (path_equal(&nd->path, &nd->root)) | |
1304 | break; | |
1305 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1306 | struct dentry *old = nd->path.dentry; | |
1307 | struct dentry *parent = old->d_parent; | |
1308 | unsigned seq; | |
1309 | ||
1310 | inode = parent->d_inode; | |
1311 | seq = read_seqcount_begin(&parent->d_seq); | |
1312 | if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) | |
1313 | return -ECHILD; | |
1314 | nd->path.dentry = parent; | |
1315 | nd->seq = seq; | |
1316 | if (unlikely(!path_connected(&nd->path))) | |
1317 | return -ENOENT; | |
1318 | break; | |
1319 | } else { | |
1320 | struct mount *mnt = real_mount(nd->path.mnt); | |
1321 | struct mount *mparent = mnt->mnt_parent; | |
1322 | struct dentry *mountpoint = mnt->mnt_mountpoint; | |
1323 | struct inode *inode2 = mountpoint->d_inode; | |
1324 | unsigned seq = read_seqcount_begin(&mountpoint->d_seq); | |
1325 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1326 | return -ECHILD; | |
1327 | if (&mparent->mnt == nd->path.mnt) | |
1328 | break; | |
1329 | /* we know that mountpoint was pinned */ | |
1330 | nd->path.dentry = mountpoint; | |
1331 | nd->path.mnt = &mparent->mnt; | |
1332 | inode = inode2; | |
1333 | nd->seq = seq; | |
1334 | } | |
1335 | } | |
1336 | while (unlikely(d_mountpoint(nd->path.dentry))) { | |
1337 | struct mount *mounted; | |
1338 | mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); | |
1339 | if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) | |
1340 | return -ECHILD; | |
1341 | if (!mounted) | |
1342 | break; | |
1343 | nd->path.mnt = &mounted->mnt; | |
1344 | nd->path.dentry = mounted->mnt.mnt_root; | |
1345 | inode = nd->path.dentry->d_inode; | |
1346 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
1347 | } | |
1348 | nd->inode = inode; | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * Follow down to the covering mount currently visible to userspace. At each | |
1354 | * point, the filesystem owning that dentry may be queried as to whether the | |
1355 | * caller is permitted to proceed or not. | |
1356 | */ | |
1357 | int follow_down(struct path *path) | |
1358 | { | |
1359 | unsigned managed; | |
1360 | int ret; | |
1361 | ||
1362 | while (managed = ACCESS_ONCE(path->dentry->d_flags), | |
1363 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { | |
1364 | /* Allow the filesystem to manage the transit without i_mutex | |
1365 | * being held. | |
1366 | * | |
1367 | * We indicate to the filesystem if someone is trying to mount | |
1368 | * something here. This gives autofs the chance to deny anyone | |
1369 | * other than its daemon the right to mount on its | |
1370 | * superstructure. | |
1371 | * | |
1372 | * The filesystem may sleep at this point. | |
1373 | */ | |
1374 | if (managed & DCACHE_MANAGE_TRANSIT) { | |
1375 | BUG_ON(!path->dentry->d_op); | |
1376 | BUG_ON(!path->dentry->d_op->d_manage); | |
1377 | ret = path->dentry->d_op->d_manage( | |
1378 | path->dentry, false); | |
1379 | if (ret < 0) | |
1380 | return ret == -EISDIR ? 0 : ret; | |
1381 | } | |
1382 | ||
1383 | /* Transit to a mounted filesystem. */ | |
1384 | if (managed & DCACHE_MOUNTED) { | |
1385 | struct vfsmount *mounted = lookup_mnt(path); | |
1386 | if (!mounted) | |
1387 | break; | |
1388 | dput(path->dentry); | |
1389 | mntput(path->mnt); | |
1390 | path->mnt = mounted; | |
1391 | path->dentry = dget(mounted->mnt_root); | |
1392 | continue; | |
1393 | } | |
1394 | ||
1395 | /* Don't handle automount points here */ | |
1396 | break; | |
1397 | } | |
1398 | return 0; | |
1399 | } | |
1400 | EXPORT_SYMBOL(follow_down); | |
1401 | ||
1402 | /* | |
1403 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() | |
1404 | */ | |
1405 | static void follow_mount(struct path *path) | |
1406 | { | |
1407 | while (d_mountpoint(path->dentry)) { | |
1408 | struct vfsmount *mounted = lookup_mnt(path); | |
1409 | if (!mounted) | |
1410 | break; | |
1411 | dput(path->dentry); | |
1412 | mntput(path->mnt); | |
1413 | path->mnt = mounted; | |
1414 | path->dentry = dget(mounted->mnt_root); | |
1415 | } | |
1416 | } | |
1417 | ||
1418 | static int follow_dotdot(struct nameidata *nd) | |
1419 | { | |
1420 | while(1) { | |
1421 | struct dentry *old = nd->path.dentry; | |
1422 | ||
1423 | if (nd->path.dentry == nd->root.dentry && | |
1424 | nd->path.mnt == nd->root.mnt) { | |
1425 | break; | |
1426 | } | |
1427 | if (nd->path.dentry != nd->path.mnt->mnt_root) { | |
1428 | /* rare case of legitimate dget_parent()... */ | |
1429 | nd->path.dentry = dget_parent(nd->path.dentry); | |
1430 | dput(old); | |
1431 | if (unlikely(!path_connected(&nd->path))) | |
1432 | return -ENOENT; | |
1433 | break; | |
1434 | } | |
1435 | if (!follow_up(&nd->path)) | |
1436 | break; | |
1437 | } | |
1438 | follow_mount(&nd->path); | |
1439 | nd->inode = nd->path.dentry->d_inode; | |
1440 | return 0; | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * This looks up the name in dcache, possibly revalidates the old dentry and | |
1445 | * allocates a new one if not found or not valid. In the need_lookup argument | |
1446 | * returns whether i_op->lookup is necessary. | |
1447 | */ | |
1448 | static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, | |
1449 | unsigned int flags) | |
1450 | { | |
1451 | struct dentry *dentry; | |
1452 | int error; | |
1453 | ||
1454 | dentry = d_lookup(dir, name); | |
1455 | if (dentry) { | |
1456 | if (dentry->d_flags & DCACHE_OP_REVALIDATE) { | |
1457 | error = d_revalidate(dentry, flags); | |
1458 | if (unlikely(error <= 0)) { | |
1459 | if (!error) | |
1460 | d_invalidate(dentry); | |
1461 | dput(dentry); | |
1462 | return ERR_PTR(error); | |
1463 | } | |
1464 | } | |
1465 | } | |
1466 | return dentry; | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Call i_op->lookup on the dentry. The dentry must be negative and | |
1471 | * unhashed. | |
1472 | * | |
1473 | * dir->d_inode->i_mutex must be held | |
1474 | */ | |
1475 | static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, | |
1476 | unsigned int flags) | |
1477 | { | |
1478 | struct dentry *old; | |
1479 | ||
1480 | /* Don't create child dentry for a dead directory. */ | |
1481 | if (unlikely(IS_DEADDIR(dir))) { | |
1482 | dput(dentry); | |
1483 | return ERR_PTR(-ENOENT); | |
1484 | } | |
1485 | ||
1486 | old = dir->i_op->lookup(dir, dentry, flags); | |
1487 | if (unlikely(old)) { | |
1488 | dput(dentry); | |
1489 | dentry = old; | |
1490 | } | |
1491 | return dentry; | |
1492 | } | |
1493 | ||
1494 | static struct dentry *__lookup_hash(struct qstr *name, | |
1495 | struct dentry *base, unsigned int flags) | |
1496 | { | |
1497 | struct dentry *dentry = lookup_dcache(name, base, flags); | |
1498 | ||
1499 | if (dentry) | |
1500 | return dentry; | |
1501 | ||
1502 | dentry = d_alloc(base, name); | |
1503 | if (unlikely(!dentry)) | |
1504 | return ERR_PTR(-ENOMEM); | |
1505 | ||
1506 | return lookup_real(base->d_inode, dentry, flags); | |
1507 | } | |
1508 | ||
1509 | static int lookup_fast(struct nameidata *nd, | |
1510 | struct path *path, struct inode **inode, | |
1511 | unsigned *seqp) | |
1512 | { | |
1513 | struct vfsmount *mnt = nd->path.mnt; | |
1514 | struct dentry *dentry, *parent = nd->path.dentry; | |
1515 | int status = 1; | |
1516 | int err; | |
1517 | ||
1518 | /* | |
1519 | * Rename seqlock is not required here because in the off chance | |
1520 | * of a false negative due to a concurrent rename, the caller is | |
1521 | * going to fall back to non-racy lookup. | |
1522 | */ | |
1523 | if (nd->flags & LOOKUP_RCU) { | |
1524 | unsigned seq; | |
1525 | bool negative; | |
1526 | dentry = __d_lookup_rcu(parent, &nd->last, &seq); | |
1527 | if (unlikely(!dentry)) { | |
1528 | if (unlazy_walk(nd, NULL, 0)) | |
1529 | return -ECHILD; | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | /* | |
1534 | * This sequence count validates that the inode matches | |
1535 | * the dentry name information from lookup. | |
1536 | */ | |
1537 | *inode = d_backing_inode(dentry); | |
1538 | negative = d_is_negative(dentry); | |
1539 | if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) | |
1540 | return -ECHILD; | |
1541 | ||
1542 | /* | |
1543 | * This sequence count validates that the parent had no | |
1544 | * changes while we did the lookup of the dentry above. | |
1545 | * | |
1546 | * The memory barrier in read_seqcount_begin of child is | |
1547 | * enough, we can use __read_seqcount_retry here. | |
1548 | */ | |
1549 | if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) | |
1550 | return -ECHILD; | |
1551 | ||
1552 | *seqp = seq; | |
1553 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
1554 | status = d_revalidate(dentry, nd->flags); | |
1555 | if (unlikely(status <= 0)) { | |
1556 | if (unlazy_walk(nd, dentry, seq)) | |
1557 | return -ECHILD; | |
1558 | if (status == -ECHILD) | |
1559 | status = d_revalidate(dentry, nd->flags); | |
1560 | } else { | |
1561 | /* | |
1562 | * Note: do negative dentry check after revalidation in | |
1563 | * case that drops it. | |
1564 | */ | |
1565 | if (unlikely(negative)) | |
1566 | return -ENOENT; | |
1567 | path->mnt = mnt; | |
1568 | path->dentry = dentry; | |
1569 | if (likely(__follow_mount_rcu(nd, path, inode, seqp))) | |
1570 | return 1; | |
1571 | if (unlazy_walk(nd, dentry, seq)) | |
1572 | return -ECHILD; | |
1573 | } | |
1574 | } else { | |
1575 | dentry = __d_lookup(parent, &nd->last); | |
1576 | if (unlikely(!dentry)) | |
1577 | return 0; | |
1578 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) | |
1579 | status = d_revalidate(dentry, nd->flags); | |
1580 | } | |
1581 | if (unlikely(status <= 0)) { | |
1582 | if (!status) | |
1583 | d_invalidate(dentry); | |
1584 | dput(dentry); | |
1585 | return status; | |
1586 | } | |
1587 | if (unlikely(d_is_negative(dentry))) { | |
1588 | dput(dentry); | |
1589 | return -ENOENT; | |
1590 | } | |
1591 | ||
1592 | path->mnt = mnt; | |
1593 | path->dentry = dentry; | |
1594 | err = follow_managed(path, nd); | |
1595 | if (likely(err > 0)) | |
1596 | *inode = d_backing_inode(path->dentry); | |
1597 | return err; | |
1598 | } | |
1599 | ||
1600 | /* Fast lookup failed, do it the slow way */ | |
1601 | static int lookup_slow(struct nameidata *nd, struct path *path) | |
1602 | { | |
1603 | struct dentry *dentry, *parent; | |
1604 | ||
1605 | parent = nd->path.dentry; | |
1606 | BUG_ON(nd->inode != parent->d_inode); | |
1607 | ||
1608 | inode_lock(parent->d_inode); | |
1609 | dentry = __lookup_hash(&nd->last, parent, nd->flags); | |
1610 | inode_unlock(parent->d_inode); | |
1611 | if (IS_ERR(dentry)) | |
1612 | return PTR_ERR(dentry); | |
1613 | path->mnt = nd->path.mnt; | |
1614 | path->dentry = dentry; | |
1615 | return follow_managed(path, nd); | |
1616 | } | |
1617 | ||
1618 | static inline int may_lookup(struct nameidata *nd) | |
1619 | { | |
1620 | if (nd->flags & LOOKUP_RCU) { | |
1621 | int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); | |
1622 | if (err != -ECHILD) | |
1623 | return err; | |
1624 | if (unlazy_walk(nd, NULL, 0)) | |
1625 | return -ECHILD; | |
1626 | } | |
1627 | return inode_permission(nd->inode, MAY_EXEC); | |
1628 | } | |
1629 | ||
1630 | static inline int handle_dots(struct nameidata *nd, int type) | |
1631 | { | |
1632 | if (type == LAST_DOTDOT) { | |
1633 | if (!nd->root.mnt) | |
1634 | set_root(nd); | |
1635 | if (nd->flags & LOOKUP_RCU) { | |
1636 | return follow_dotdot_rcu(nd); | |
1637 | } else | |
1638 | return follow_dotdot(nd); | |
1639 | } | |
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | static int pick_link(struct nameidata *nd, struct path *link, | |
1644 | struct inode *inode, unsigned seq) | |
1645 | { | |
1646 | int error; | |
1647 | struct saved *last; | |
1648 | if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { | |
1649 | path_to_nameidata(link, nd); | |
1650 | return -ELOOP; | |
1651 | } | |
1652 | if (!(nd->flags & LOOKUP_RCU)) { | |
1653 | if (link->mnt == nd->path.mnt) | |
1654 | mntget(link->mnt); | |
1655 | } | |
1656 | error = nd_alloc_stack(nd); | |
1657 | if (unlikely(error)) { | |
1658 | if (error == -ECHILD) { | |
1659 | if (unlikely(unlazy_link(nd, link, seq))) | |
1660 | return -ECHILD; | |
1661 | error = nd_alloc_stack(nd); | |
1662 | } | |
1663 | if (error) { | |
1664 | path_put(link); | |
1665 | return error; | |
1666 | } | |
1667 | } | |
1668 | ||
1669 | last = nd->stack + nd->depth++; | |
1670 | last->link = *link; | |
1671 | clear_delayed_call(&last->done); | |
1672 | nd->link_inode = inode; | |
1673 | last->seq = seq; | |
1674 | return 1; | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * Do we need to follow links? We _really_ want to be able | |
1679 | * to do this check without having to look at inode->i_op, | |
1680 | * so we keep a cache of "no, this doesn't need follow_link" | |
1681 | * for the common case. | |
1682 | */ | |
1683 | static inline int should_follow_link(struct nameidata *nd, struct path *link, | |
1684 | int follow, | |
1685 | struct inode *inode, unsigned seq) | |
1686 | { | |
1687 | if (likely(!d_is_symlink(link->dentry))) | |
1688 | return 0; | |
1689 | if (!follow) | |
1690 | return 0; | |
1691 | /* make sure that d_is_symlink above matches inode */ | |
1692 | if (nd->flags & LOOKUP_RCU) { | |
1693 | if (read_seqcount_retry(&link->dentry->d_seq, seq)) | |
1694 | return -ECHILD; | |
1695 | } | |
1696 | return pick_link(nd, link, inode, seq); | |
1697 | } | |
1698 | ||
1699 | enum {WALK_GET = 1, WALK_PUT = 2}; | |
1700 | ||
1701 | static int walk_component(struct nameidata *nd, int flags) | |
1702 | { | |
1703 | struct path path; | |
1704 | struct inode *inode; | |
1705 | unsigned seq; | |
1706 | int err; | |
1707 | /* | |
1708 | * "." and ".." are special - ".." especially so because it has | |
1709 | * to be able to know about the current root directory and | |
1710 | * parent relationships. | |
1711 | */ | |
1712 | if (unlikely(nd->last_type != LAST_NORM)) { | |
1713 | err = handle_dots(nd, nd->last_type); | |
1714 | if (flags & WALK_PUT) | |
1715 | put_link(nd); | |
1716 | return err; | |
1717 | } | |
1718 | err = lookup_fast(nd, &path, &inode, &seq); | |
1719 | if (unlikely(err <= 0)) { | |
1720 | if (err < 0) | |
1721 | return err; | |
1722 | ||
1723 | err = lookup_slow(nd, &path); | |
1724 | if (err < 0) | |
1725 | return err; | |
1726 | ||
1727 | seq = 0; /* we are already out of RCU mode */ | |
1728 | err = -ENOENT; | |
1729 | if (d_is_negative(path.dentry)) | |
1730 | goto out_path_put; | |
1731 | inode = d_backing_inode(path.dentry); | |
1732 | } | |
1733 | ||
1734 | if (flags & WALK_PUT) | |
1735 | put_link(nd); | |
1736 | err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); | |
1737 | if (unlikely(err)) | |
1738 | return err; | |
1739 | path_to_nameidata(&path, nd); | |
1740 | nd->inode = inode; | |
1741 | nd->seq = seq; | |
1742 | return 0; | |
1743 | ||
1744 | out_path_put: | |
1745 | path_to_nameidata(&path, nd); | |
1746 | return err; | |
1747 | } | |
1748 | ||
1749 | /* | |
1750 | * We can do the critical dentry name comparison and hashing | |
1751 | * operations one word at a time, but we are limited to: | |
1752 | * | |
1753 | * - Architectures with fast unaligned word accesses. We could | |
1754 | * do a "get_unaligned()" if this helps and is sufficiently | |
1755 | * fast. | |
1756 | * | |
1757 | * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we | |
1758 | * do not trap on the (extremely unlikely) case of a page | |
1759 | * crossing operation. | |
1760 | * | |
1761 | * - Furthermore, we need an efficient 64-bit compile for the | |
1762 | * 64-bit case in order to generate the "number of bytes in | |
1763 | * the final mask". Again, that could be replaced with a | |
1764 | * efficient population count instruction or similar. | |
1765 | */ | |
1766 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
1767 | ||
1768 | #include <asm/word-at-a-time.h> | |
1769 | ||
1770 | #ifdef CONFIG_64BIT | |
1771 | ||
1772 | static inline unsigned int fold_hash(unsigned long hash) | |
1773 | { | |
1774 | return hash_64(hash, 32); | |
1775 | } | |
1776 | ||
1777 | #else /* 32-bit case */ | |
1778 | ||
1779 | #define fold_hash(x) (x) | |
1780 | ||
1781 | #endif | |
1782 | ||
1783 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1784 | { | |
1785 | unsigned long a, mask; | |
1786 | unsigned long hash = 0; | |
1787 | ||
1788 | for (;;) { | |
1789 | a = load_unaligned_zeropad(name); | |
1790 | if (len < sizeof(unsigned long)) | |
1791 | break; | |
1792 | hash += a; | |
1793 | hash *= 9; | |
1794 | name += sizeof(unsigned long); | |
1795 | len -= sizeof(unsigned long); | |
1796 | if (!len) | |
1797 | goto done; | |
1798 | } | |
1799 | mask = bytemask_from_count(len); | |
1800 | hash += mask & a; | |
1801 | done: | |
1802 | return fold_hash(hash); | |
1803 | } | |
1804 | EXPORT_SYMBOL(full_name_hash); | |
1805 | ||
1806 | /* | |
1807 | * Calculate the length and hash of the path component, and | |
1808 | * return the "hash_len" as the result. | |
1809 | */ | |
1810 | static inline u64 hash_name(const char *name) | |
1811 | { | |
1812 | unsigned long a, b, adata, bdata, mask, hash, len; | |
1813 | const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; | |
1814 | ||
1815 | hash = a = 0; | |
1816 | len = -sizeof(unsigned long); | |
1817 | do { | |
1818 | hash = (hash + a) * 9; | |
1819 | len += sizeof(unsigned long); | |
1820 | a = load_unaligned_zeropad(name+len); | |
1821 | b = a ^ REPEAT_BYTE('/'); | |
1822 | } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); | |
1823 | ||
1824 | adata = prep_zero_mask(a, adata, &constants); | |
1825 | bdata = prep_zero_mask(b, bdata, &constants); | |
1826 | ||
1827 | mask = create_zero_mask(adata | bdata); | |
1828 | ||
1829 | hash += a & zero_bytemask(mask); | |
1830 | len += find_zero(mask); | |
1831 | return hashlen_create(fold_hash(hash), len); | |
1832 | } | |
1833 | ||
1834 | #else | |
1835 | ||
1836 | unsigned int full_name_hash(const unsigned char *name, unsigned int len) | |
1837 | { | |
1838 | unsigned long hash = init_name_hash(); | |
1839 | while (len--) | |
1840 | hash = partial_name_hash(*name++, hash); | |
1841 | return end_name_hash(hash); | |
1842 | } | |
1843 | EXPORT_SYMBOL(full_name_hash); | |
1844 | ||
1845 | /* | |
1846 | * We know there's a real path component here of at least | |
1847 | * one character. | |
1848 | */ | |
1849 | static inline u64 hash_name(const char *name) | |
1850 | { | |
1851 | unsigned long hash = init_name_hash(); | |
1852 | unsigned long len = 0, c; | |
1853 | ||
1854 | c = (unsigned char)*name; | |
1855 | do { | |
1856 | len++; | |
1857 | hash = partial_name_hash(c, hash); | |
1858 | c = (unsigned char)name[len]; | |
1859 | } while (c && c != '/'); | |
1860 | return hashlen_create(end_name_hash(hash), len); | |
1861 | } | |
1862 | ||
1863 | #endif | |
1864 | ||
1865 | /* | |
1866 | * Name resolution. | |
1867 | * This is the basic name resolution function, turning a pathname into | |
1868 | * the final dentry. We expect 'base' to be positive and a directory. | |
1869 | * | |
1870 | * Returns 0 and nd will have valid dentry and mnt on success. | |
1871 | * Returns error and drops reference to input namei data on failure. | |
1872 | */ | |
1873 | static int link_path_walk(const char *name, struct nameidata *nd) | |
1874 | { | |
1875 | int err; | |
1876 | ||
1877 | while (*name=='/') | |
1878 | name++; | |
1879 | if (!*name) | |
1880 | return 0; | |
1881 | ||
1882 | /* At this point we know we have a real path component. */ | |
1883 | for(;;) { | |
1884 | u64 hash_len; | |
1885 | int type; | |
1886 | ||
1887 | err = may_lookup(nd); | |
1888 | if (err) | |
1889 | return err; | |
1890 | ||
1891 | hash_len = hash_name(name); | |
1892 | ||
1893 | type = LAST_NORM; | |
1894 | if (name[0] == '.') switch (hashlen_len(hash_len)) { | |
1895 | case 2: | |
1896 | if (name[1] == '.') { | |
1897 | type = LAST_DOTDOT; | |
1898 | nd->flags |= LOOKUP_JUMPED; | |
1899 | } | |
1900 | break; | |
1901 | case 1: | |
1902 | type = LAST_DOT; | |
1903 | } | |
1904 | if (likely(type == LAST_NORM)) { | |
1905 | struct dentry *parent = nd->path.dentry; | |
1906 | nd->flags &= ~LOOKUP_JUMPED; | |
1907 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { | |
1908 | struct qstr this = { { .hash_len = hash_len }, .name = name }; | |
1909 | err = parent->d_op->d_hash(parent, &this); | |
1910 | if (err < 0) | |
1911 | return err; | |
1912 | hash_len = this.hash_len; | |
1913 | name = this.name; | |
1914 | } | |
1915 | } | |
1916 | ||
1917 | nd->last.hash_len = hash_len; | |
1918 | nd->last.name = name; | |
1919 | nd->last_type = type; | |
1920 | ||
1921 | name += hashlen_len(hash_len); | |
1922 | if (!*name) | |
1923 | goto OK; | |
1924 | /* | |
1925 | * If it wasn't NUL, we know it was '/'. Skip that | |
1926 | * slash, and continue until no more slashes. | |
1927 | */ | |
1928 | do { | |
1929 | name++; | |
1930 | } while (unlikely(*name == '/')); | |
1931 | if (unlikely(!*name)) { | |
1932 | OK: | |
1933 | /* pathname body, done */ | |
1934 | if (!nd->depth) | |
1935 | return 0; | |
1936 | name = nd->stack[nd->depth - 1].name; | |
1937 | /* trailing symlink, done */ | |
1938 | if (!name) | |
1939 | return 0; | |
1940 | /* last component of nested symlink */ | |
1941 | err = walk_component(nd, WALK_GET | WALK_PUT); | |
1942 | } else { | |
1943 | err = walk_component(nd, WALK_GET); | |
1944 | } | |
1945 | if (err < 0) | |
1946 | return err; | |
1947 | ||
1948 | if (err) { | |
1949 | const char *s = get_link(nd); | |
1950 | ||
1951 | if (IS_ERR(s)) | |
1952 | return PTR_ERR(s); | |
1953 | err = 0; | |
1954 | if (unlikely(!s)) { | |
1955 | /* jumped */ | |
1956 | put_link(nd); | |
1957 | } else { | |
1958 | nd->stack[nd->depth - 1].name = name; | |
1959 | name = s; | |
1960 | continue; | |
1961 | } | |
1962 | } | |
1963 | if (unlikely(!d_can_lookup(nd->path.dentry))) { | |
1964 | if (nd->flags & LOOKUP_RCU) { | |
1965 | if (unlazy_walk(nd, NULL, 0)) | |
1966 | return -ECHILD; | |
1967 | } | |
1968 | return -ENOTDIR; | |
1969 | } | |
1970 | } | |
1971 | } | |
1972 | ||
1973 | static const char *path_init(struct nameidata *nd, unsigned flags) | |
1974 | { | |
1975 | int retval = 0; | |
1976 | const char *s = nd->name->name; | |
1977 | ||
1978 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ | |
1979 | nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; | |
1980 | nd->depth = 0; | |
1981 | if (flags & LOOKUP_ROOT) { | |
1982 | struct dentry *root = nd->root.dentry; | |
1983 | struct inode *inode = root->d_inode; | |
1984 | if (*s) { | |
1985 | if (!d_can_lookup(root)) | |
1986 | return ERR_PTR(-ENOTDIR); | |
1987 | retval = inode_permission(inode, MAY_EXEC); | |
1988 | if (retval) | |
1989 | return ERR_PTR(retval); | |
1990 | } | |
1991 | nd->path = nd->root; | |
1992 | nd->inode = inode; | |
1993 | if (flags & LOOKUP_RCU) { | |
1994 | rcu_read_lock(); | |
1995 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
1996 | nd->root_seq = nd->seq; | |
1997 | nd->m_seq = read_seqbegin(&mount_lock); | |
1998 | } else { | |
1999 | path_get(&nd->path); | |
2000 | } | |
2001 | return s; | |
2002 | } | |
2003 | ||
2004 | nd->root.mnt = NULL; | |
2005 | nd->path.mnt = NULL; | |
2006 | nd->path.dentry = NULL; | |
2007 | ||
2008 | nd->m_seq = read_seqbegin(&mount_lock); | |
2009 | if (*s == '/') { | |
2010 | if (flags & LOOKUP_RCU) | |
2011 | rcu_read_lock(); | |
2012 | set_root(nd); | |
2013 | if (likely(!nd_jump_root(nd))) | |
2014 | return s; | |
2015 | nd->root.mnt = NULL; | |
2016 | rcu_read_unlock(); | |
2017 | return ERR_PTR(-ECHILD); | |
2018 | } else if (nd->dfd == AT_FDCWD) { | |
2019 | if (flags & LOOKUP_RCU) { | |
2020 | struct fs_struct *fs = current->fs; | |
2021 | unsigned seq; | |
2022 | ||
2023 | rcu_read_lock(); | |
2024 | ||
2025 | do { | |
2026 | seq = read_seqcount_begin(&fs->seq); | |
2027 | nd->path = fs->pwd; | |
2028 | nd->inode = nd->path.dentry->d_inode; | |
2029 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); | |
2030 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2031 | } else { | |
2032 | get_fs_pwd(current->fs, &nd->path); | |
2033 | nd->inode = nd->path.dentry->d_inode; | |
2034 | } | |
2035 | return s; | |
2036 | } else { | |
2037 | /* Caller must check execute permissions on the starting path component */ | |
2038 | struct fd f = fdget_raw(nd->dfd); | |
2039 | struct dentry *dentry; | |
2040 | ||
2041 | if (!f.file) | |
2042 | return ERR_PTR(-EBADF); | |
2043 | ||
2044 | dentry = f.file->f_path.dentry; | |
2045 | ||
2046 | if (*s) { | |
2047 | if (!d_can_lookup(dentry)) { | |
2048 | fdput(f); | |
2049 | return ERR_PTR(-ENOTDIR); | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | nd->path = f.file->f_path; | |
2054 | if (flags & LOOKUP_RCU) { | |
2055 | rcu_read_lock(); | |
2056 | nd->inode = nd->path.dentry->d_inode; | |
2057 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); | |
2058 | } else { | |
2059 | path_get(&nd->path); | |
2060 | nd->inode = nd->path.dentry->d_inode; | |
2061 | } | |
2062 | fdput(f); | |
2063 | return s; | |
2064 | } | |
2065 | } | |
2066 | ||
2067 | static const char *trailing_symlink(struct nameidata *nd) | |
2068 | { | |
2069 | const char *s; | |
2070 | int error = may_follow_link(nd); | |
2071 | if (unlikely(error)) | |
2072 | return ERR_PTR(error); | |
2073 | nd->flags |= LOOKUP_PARENT; | |
2074 | nd->stack[0].name = NULL; | |
2075 | s = get_link(nd); | |
2076 | return s ? s : ""; | |
2077 | } | |
2078 | ||
2079 | static inline int lookup_last(struct nameidata *nd) | |
2080 | { | |
2081 | if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) | |
2082 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
2083 | ||
2084 | nd->flags &= ~LOOKUP_PARENT; | |
2085 | return walk_component(nd, | |
2086 | nd->flags & LOOKUP_FOLLOW | |
2087 | ? nd->depth | |
2088 | ? WALK_PUT | WALK_GET | |
2089 | : WALK_GET | |
2090 | : 0); | |
2091 | } | |
2092 | ||
2093 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2094 | static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) | |
2095 | { | |
2096 | const char *s = path_init(nd, flags); | |
2097 | int err; | |
2098 | ||
2099 | if (IS_ERR(s)) | |
2100 | return PTR_ERR(s); | |
2101 | while (!(err = link_path_walk(s, nd)) | |
2102 | && ((err = lookup_last(nd)) > 0)) { | |
2103 | s = trailing_symlink(nd); | |
2104 | if (IS_ERR(s)) { | |
2105 | err = PTR_ERR(s); | |
2106 | break; | |
2107 | } | |
2108 | } | |
2109 | if (!err) | |
2110 | err = complete_walk(nd); | |
2111 | ||
2112 | if (!err && nd->flags & LOOKUP_DIRECTORY) | |
2113 | if (!d_can_lookup(nd->path.dentry)) | |
2114 | err = -ENOTDIR; | |
2115 | if (!err) { | |
2116 | *path = nd->path; | |
2117 | nd->path.mnt = NULL; | |
2118 | nd->path.dentry = NULL; | |
2119 | } | |
2120 | terminate_walk(nd); | |
2121 | return err; | |
2122 | } | |
2123 | ||
2124 | static int filename_lookup(int dfd, struct filename *name, unsigned flags, | |
2125 | struct path *path, struct path *root) | |
2126 | { | |
2127 | int retval; | |
2128 | struct nameidata nd; | |
2129 | if (IS_ERR(name)) | |
2130 | return PTR_ERR(name); | |
2131 | if (unlikely(root)) { | |
2132 | nd.root = *root; | |
2133 | flags |= LOOKUP_ROOT; | |
2134 | } | |
2135 | set_nameidata(&nd, dfd, name); | |
2136 | retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); | |
2137 | if (unlikely(retval == -ECHILD)) | |
2138 | retval = path_lookupat(&nd, flags, path); | |
2139 | if (unlikely(retval == -ESTALE)) | |
2140 | retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); | |
2141 | ||
2142 | if (likely(!retval)) | |
2143 | audit_inode(name, path->dentry, flags & LOOKUP_PARENT); | |
2144 | restore_nameidata(); | |
2145 | putname(name); | |
2146 | return retval; | |
2147 | } | |
2148 | ||
2149 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ | |
2150 | static int path_parentat(struct nameidata *nd, unsigned flags, | |
2151 | struct path *parent) | |
2152 | { | |
2153 | const char *s = path_init(nd, flags); | |
2154 | int err; | |
2155 | if (IS_ERR(s)) | |
2156 | return PTR_ERR(s); | |
2157 | err = link_path_walk(s, nd); | |
2158 | if (!err) | |
2159 | err = complete_walk(nd); | |
2160 | if (!err) { | |
2161 | *parent = nd->path; | |
2162 | nd->path.mnt = NULL; | |
2163 | nd->path.dentry = NULL; | |
2164 | } | |
2165 | terminate_walk(nd); | |
2166 | return err; | |
2167 | } | |
2168 | ||
2169 | static struct filename *filename_parentat(int dfd, struct filename *name, | |
2170 | unsigned int flags, struct path *parent, | |
2171 | struct qstr *last, int *type) | |
2172 | { | |
2173 | int retval; | |
2174 | struct nameidata nd; | |
2175 | ||
2176 | if (IS_ERR(name)) | |
2177 | return name; | |
2178 | set_nameidata(&nd, dfd, name); | |
2179 | retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); | |
2180 | if (unlikely(retval == -ECHILD)) | |
2181 | retval = path_parentat(&nd, flags, parent); | |
2182 | if (unlikely(retval == -ESTALE)) | |
2183 | retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); | |
2184 | if (likely(!retval)) { | |
2185 | *last = nd.last; | |
2186 | *type = nd.last_type; | |
2187 | audit_inode(name, parent->dentry, LOOKUP_PARENT); | |
2188 | } else { | |
2189 | putname(name); | |
2190 | name = ERR_PTR(retval); | |
2191 | } | |
2192 | restore_nameidata(); | |
2193 | return name; | |
2194 | } | |
2195 | ||
2196 | /* does lookup, returns the object with parent locked */ | |
2197 | struct dentry *kern_path_locked(const char *name, struct path *path) | |
2198 | { | |
2199 | struct filename *filename; | |
2200 | struct dentry *d; | |
2201 | struct qstr last; | |
2202 | int type; | |
2203 | ||
2204 | filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, | |
2205 | &last, &type); | |
2206 | if (IS_ERR(filename)) | |
2207 | return ERR_CAST(filename); | |
2208 | if (unlikely(type != LAST_NORM)) { | |
2209 | path_put(path); | |
2210 | putname(filename); | |
2211 | return ERR_PTR(-EINVAL); | |
2212 | } | |
2213 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
2214 | d = __lookup_hash(&last, path->dentry, 0); | |
2215 | if (IS_ERR(d)) { | |
2216 | inode_unlock(path->dentry->d_inode); | |
2217 | path_put(path); | |
2218 | } | |
2219 | putname(filename); | |
2220 | return d; | |
2221 | } | |
2222 | ||
2223 | int kern_path(const char *name, unsigned int flags, struct path *path) | |
2224 | { | |
2225 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2226 | flags, path, NULL); | |
2227 | } | |
2228 | EXPORT_SYMBOL(kern_path); | |
2229 | ||
2230 | /** | |
2231 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair | |
2232 | * @dentry: pointer to dentry of the base directory | |
2233 | * @mnt: pointer to vfs mount of the base directory | |
2234 | * @name: pointer to file name | |
2235 | * @flags: lookup flags | |
2236 | * @path: pointer to struct path to fill | |
2237 | */ | |
2238 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, | |
2239 | const char *name, unsigned int flags, | |
2240 | struct path *path) | |
2241 | { | |
2242 | struct path root = {.mnt = mnt, .dentry = dentry}; | |
2243 | /* the first argument of filename_lookup() is ignored with root */ | |
2244 | return filename_lookup(AT_FDCWD, getname_kernel(name), | |
2245 | flags , path, &root); | |
2246 | } | |
2247 | EXPORT_SYMBOL(vfs_path_lookup); | |
2248 | ||
2249 | /** | |
2250 | * lookup_one_len - filesystem helper to lookup single pathname component | |
2251 | * @name: pathname component to lookup | |
2252 | * @base: base directory to lookup from | |
2253 | * @len: maximum length @len should be interpreted to | |
2254 | * | |
2255 | * Note that this routine is purely a helper for filesystem usage and should | |
2256 | * not be called by generic code. | |
2257 | * | |
2258 | * The caller must hold base->i_mutex. | |
2259 | */ | |
2260 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) | |
2261 | { | |
2262 | struct qstr this; | |
2263 | unsigned int c; | |
2264 | int err; | |
2265 | ||
2266 | WARN_ON_ONCE(!inode_is_locked(base->d_inode)); | |
2267 | ||
2268 | this.name = name; | |
2269 | this.len = len; | |
2270 | this.hash = full_name_hash(name, len); | |
2271 | if (!len) | |
2272 | return ERR_PTR(-EACCES); | |
2273 | ||
2274 | if (unlikely(name[0] == '.')) { | |
2275 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2276 | return ERR_PTR(-EACCES); | |
2277 | } | |
2278 | ||
2279 | while (len--) { | |
2280 | c = *(const unsigned char *)name++; | |
2281 | if (c == '/' || c == '\0') | |
2282 | return ERR_PTR(-EACCES); | |
2283 | } | |
2284 | /* | |
2285 | * See if the low-level filesystem might want | |
2286 | * to use its own hash.. | |
2287 | */ | |
2288 | if (base->d_flags & DCACHE_OP_HASH) { | |
2289 | int err = base->d_op->d_hash(base, &this); | |
2290 | if (err < 0) | |
2291 | return ERR_PTR(err); | |
2292 | } | |
2293 | ||
2294 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2295 | if (err) | |
2296 | return ERR_PTR(err); | |
2297 | ||
2298 | return __lookup_hash(&this, base, 0); | |
2299 | } | |
2300 | EXPORT_SYMBOL(lookup_one_len); | |
2301 | ||
2302 | /** | |
2303 | * lookup_one_len_unlocked - filesystem helper to lookup single pathname component | |
2304 | * @name: pathname component to lookup | |
2305 | * @base: base directory to lookup from | |
2306 | * @len: maximum length @len should be interpreted to | |
2307 | * | |
2308 | * Note that this routine is purely a helper for filesystem usage and should | |
2309 | * not be called by generic code. | |
2310 | * | |
2311 | * Unlike lookup_one_len, it should be called without the parent | |
2312 | * i_mutex held, and will take the i_mutex itself if necessary. | |
2313 | */ | |
2314 | struct dentry *lookup_one_len_unlocked(const char *name, | |
2315 | struct dentry *base, int len) | |
2316 | { | |
2317 | struct qstr this; | |
2318 | unsigned int c; | |
2319 | int err; | |
2320 | struct dentry *ret; | |
2321 | ||
2322 | this.name = name; | |
2323 | this.len = len; | |
2324 | this.hash = full_name_hash(name, len); | |
2325 | if (!len) | |
2326 | return ERR_PTR(-EACCES); | |
2327 | ||
2328 | if (unlikely(name[0] == '.')) { | |
2329 | if (len < 2 || (len == 2 && name[1] == '.')) | |
2330 | return ERR_PTR(-EACCES); | |
2331 | } | |
2332 | ||
2333 | while (len--) { | |
2334 | c = *(const unsigned char *)name++; | |
2335 | if (c == '/' || c == '\0') | |
2336 | return ERR_PTR(-EACCES); | |
2337 | } | |
2338 | /* | |
2339 | * See if the low-level filesystem might want | |
2340 | * to use its own hash.. | |
2341 | */ | |
2342 | if (base->d_flags & DCACHE_OP_HASH) { | |
2343 | int err = base->d_op->d_hash(base, &this); | |
2344 | if (err < 0) | |
2345 | return ERR_PTR(err); | |
2346 | } | |
2347 | ||
2348 | err = inode_permission(base->d_inode, MAY_EXEC); | |
2349 | if (err) | |
2350 | return ERR_PTR(err); | |
2351 | ||
2352 | ret = lookup_dcache(&this, base, 0); | |
2353 | if (ret) | |
2354 | return ret; | |
2355 | ||
2356 | inode_lock(base->d_inode); | |
2357 | ret = __lookup_hash(&this, base, 0); | |
2358 | inode_unlock(base->d_inode); | |
2359 | return ret; | |
2360 | } | |
2361 | EXPORT_SYMBOL(lookup_one_len_unlocked); | |
2362 | ||
2363 | int user_path_at_empty(int dfd, const char __user *name, unsigned flags, | |
2364 | struct path *path, int *empty) | |
2365 | { | |
2366 | return filename_lookup(dfd, getname_flags(name, flags, empty), | |
2367 | flags, path, NULL); | |
2368 | } | |
2369 | EXPORT_SYMBOL(user_path_at_empty); | |
2370 | ||
2371 | /* | |
2372 | * NB: most callers don't do anything directly with the reference to the | |
2373 | * to struct filename, but the nd->last pointer points into the name string | |
2374 | * allocated by getname. So we must hold the reference to it until all | |
2375 | * path-walking is complete. | |
2376 | */ | |
2377 | static inline struct filename * | |
2378 | user_path_parent(int dfd, const char __user *path, | |
2379 | struct path *parent, | |
2380 | struct qstr *last, | |
2381 | int *type, | |
2382 | unsigned int flags) | |
2383 | { | |
2384 | /* only LOOKUP_REVAL is allowed in extra flags */ | |
2385 | return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, | |
2386 | parent, last, type); | |
2387 | } | |
2388 | ||
2389 | /** | |
2390 | * mountpoint_last - look up last component for umount | |
2391 | * @nd: pathwalk nameidata - currently pointing at parent directory of "last" | |
2392 | * @path: pointer to container for result | |
2393 | * | |
2394 | * This is a special lookup_last function just for umount. In this case, we | |
2395 | * need to resolve the path without doing any revalidation. | |
2396 | * | |
2397 | * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since | |
2398 | * mountpoints are always pinned in the dcache, their ancestors are too. Thus, | |
2399 | * in almost all cases, this lookup will be served out of the dcache. The only | |
2400 | * cases where it won't are if nd->last refers to a symlink or the path is | |
2401 | * bogus and it doesn't exist. | |
2402 | * | |
2403 | * Returns: | |
2404 | * -error: if there was an error during lookup. This includes -ENOENT if the | |
2405 | * lookup found a negative dentry. The nd->path reference will also be | |
2406 | * put in this case. | |
2407 | * | |
2408 | * 0: if we successfully resolved nd->path and found it to not to be a | |
2409 | * symlink that needs to be followed. "path" will also be populated. | |
2410 | * The nd->path reference will also be put. | |
2411 | * | |
2412 | * 1: if we successfully resolved nd->last and found it to be a symlink | |
2413 | * that needs to be followed. "path" will be populated with the path | |
2414 | * to the link, and nd->path will *not* be put. | |
2415 | */ | |
2416 | static int | |
2417 | mountpoint_last(struct nameidata *nd, struct path *path) | |
2418 | { | |
2419 | int error = 0; | |
2420 | struct dentry *dentry; | |
2421 | struct dentry *dir = nd->path.dentry; | |
2422 | ||
2423 | /* If we're in rcuwalk, drop out of it to handle last component */ | |
2424 | if (nd->flags & LOOKUP_RCU) { | |
2425 | if (unlazy_walk(nd, NULL, 0)) | |
2426 | return -ECHILD; | |
2427 | } | |
2428 | ||
2429 | nd->flags &= ~LOOKUP_PARENT; | |
2430 | ||
2431 | if (unlikely(nd->last_type != LAST_NORM)) { | |
2432 | error = handle_dots(nd, nd->last_type); | |
2433 | if (error) | |
2434 | return error; | |
2435 | dentry = dget(nd->path.dentry); | |
2436 | goto done; | |
2437 | } | |
2438 | ||
2439 | inode_lock(dir->d_inode); | |
2440 | dentry = d_lookup(dir, &nd->last); | |
2441 | if (!dentry) { | |
2442 | /* | |
2443 | * No cached dentry. Mounted dentries are pinned in the cache, | |
2444 | * so that means that this dentry is probably a symlink or the | |
2445 | * path doesn't actually point to a mounted dentry. | |
2446 | */ | |
2447 | dentry = d_alloc(dir, &nd->last); | |
2448 | if (!dentry) { | |
2449 | inode_unlock(dir->d_inode); | |
2450 | return -ENOMEM; | |
2451 | } | |
2452 | dentry = lookup_real(dir->d_inode, dentry, nd->flags); | |
2453 | if (IS_ERR(dentry)) { | |
2454 | inode_unlock(dir->d_inode); | |
2455 | return PTR_ERR(dentry); | |
2456 | } | |
2457 | } | |
2458 | inode_unlock(dir->d_inode); | |
2459 | ||
2460 | done: | |
2461 | if (d_is_negative(dentry)) { | |
2462 | dput(dentry); | |
2463 | return -ENOENT; | |
2464 | } | |
2465 | if (nd->depth) | |
2466 | put_link(nd); | |
2467 | path->dentry = dentry; | |
2468 | path->mnt = nd->path.mnt; | |
2469 | error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, | |
2470 | d_backing_inode(dentry), 0); | |
2471 | if (unlikely(error)) | |
2472 | return error; | |
2473 | mntget(path->mnt); | |
2474 | follow_mount(path); | |
2475 | return 0; | |
2476 | } | |
2477 | ||
2478 | /** | |
2479 | * path_mountpoint - look up a path to be umounted | |
2480 | * @nd: lookup context | |
2481 | * @flags: lookup flags | |
2482 | * @path: pointer to container for result | |
2483 | * | |
2484 | * Look up the given name, but don't attempt to revalidate the last component. | |
2485 | * Returns 0 and "path" will be valid on success; Returns error otherwise. | |
2486 | */ | |
2487 | static int | |
2488 | path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) | |
2489 | { | |
2490 | const char *s = path_init(nd, flags); | |
2491 | int err; | |
2492 | if (IS_ERR(s)) | |
2493 | return PTR_ERR(s); | |
2494 | while (!(err = link_path_walk(s, nd)) && | |
2495 | (err = mountpoint_last(nd, path)) > 0) { | |
2496 | s = trailing_symlink(nd); | |
2497 | if (IS_ERR(s)) { | |
2498 | err = PTR_ERR(s); | |
2499 | break; | |
2500 | } | |
2501 | } | |
2502 | terminate_walk(nd); | |
2503 | return err; | |
2504 | } | |
2505 | ||
2506 | static int | |
2507 | filename_mountpoint(int dfd, struct filename *name, struct path *path, | |
2508 | unsigned int flags) | |
2509 | { | |
2510 | struct nameidata nd; | |
2511 | int error; | |
2512 | if (IS_ERR(name)) | |
2513 | return PTR_ERR(name); | |
2514 | set_nameidata(&nd, dfd, name); | |
2515 | error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); | |
2516 | if (unlikely(error == -ECHILD)) | |
2517 | error = path_mountpoint(&nd, flags, path); | |
2518 | if (unlikely(error == -ESTALE)) | |
2519 | error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); | |
2520 | if (likely(!error)) | |
2521 | audit_inode(name, path->dentry, 0); | |
2522 | restore_nameidata(); | |
2523 | putname(name); | |
2524 | return error; | |
2525 | } | |
2526 | ||
2527 | /** | |
2528 | * user_path_mountpoint_at - lookup a path from userland in order to umount it | |
2529 | * @dfd: directory file descriptor | |
2530 | * @name: pathname from userland | |
2531 | * @flags: lookup flags | |
2532 | * @path: pointer to container to hold result | |
2533 | * | |
2534 | * A umount is a special case for path walking. We're not actually interested | |
2535 | * in the inode in this situation, and ESTALE errors can be a problem. We | |
2536 | * simply want track down the dentry and vfsmount attached at the mountpoint | |
2537 | * and avoid revalidating the last component. | |
2538 | * | |
2539 | * Returns 0 and populates "path" on success. | |
2540 | */ | |
2541 | int | |
2542 | user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, | |
2543 | struct path *path) | |
2544 | { | |
2545 | return filename_mountpoint(dfd, getname(name), path, flags); | |
2546 | } | |
2547 | ||
2548 | int | |
2549 | kern_path_mountpoint(int dfd, const char *name, struct path *path, | |
2550 | unsigned int flags) | |
2551 | { | |
2552 | return filename_mountpoint(dfd, getname_kernel(name), path, flags); | |
2553 | } | |
2554 | EXPORT_SYMBOL(kern_path_mountpoint); | |
2555 | ||
2556 | int __check_sticky(struct inode *dir, struct inode *inode) | |
2557 | { | |
2558 | kuid_t fsuid = current_fsuid(); | |
2559 | ||
2560 | if (uid_eq(inode->i_uid, fsuid)) | |
2561 | return 0; | |
2562 | if (uid_eq(dir->i_uid, fsuid)) | |
2563 | return 0; | |
2564 | return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); | |
2565 | } | |
2566 | EXPORT_SYMBOL(__check_sticky); | |
2567 | ||
2568 | /* | |
2569 | * Check whether we can remove a link victim from directory dir, check | |
2570 | * whether the type of victim is right. | |
2571 | * 1. We can't do it if dir is read-only (done in permission()) | |
2572 | * 2. We should have write and exec permissions on dir | |
2573 | * 3. We can't remove anything from append-only dir | |
2574 | * 4. We can't do anything with immutable dir (done in permission()) | |
2575 | * 5. If the sticky bit on dir is set we should either | |
2576 | * a. be owner of dir, or | |
2577 | * b. be owner of victim, or | |
2578 | * c. have CAP_FOWNER capability | |
2579 | * 6. If the victim is append-only or immutable we can't do antyhing with | |
2580 | * links pointing to it. | |
2581 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. | |
2582 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. | |
2583 | * 9. We can't remove a root or mountpoint. | |
2584 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by | |
2585 | * nfs_async_unlink(). | |
2586 | */ | |
2587 | static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) | |
2588 | { | |
2589 | struct inode *inode = d_backing_inode(victim); | |
2590 | int error; | |
2591 | ||
2592 | if (d_is_negative(victim)) | |
2593 | return -ENOENT; | |
2594 | BUG_ON(!inode); | |
2595 | ||
2596 | BUG_ON(victim->d_parent->d_inode != dir); | |
2597 | audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); | |
2598 | ||
2599 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
2600 | if (error) | |
2601 | return error; | |
2602 | if (IS_APPEND(dir)) | |
2603 | return -EPERM; | |
2604 | ||
2605 | if (check_sticky(dir, inode) || IS_APPEND(inode) || | |
2606 | IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) | |
2607 | return -EPERM; | |
2608 | if (isdir) { | |
2609 | if (!d_is_dir(victim)) | |
2610 | return -ENOTDIR; | |
2611 | if (IS_ROOT(victim)) | |
2612 | return -EBUSY; | |
2613 | } else if (d_is_dir(victim)) | |
2614 | return -EISDIR; | |
2615 | if (IS_DEADDIR(dir)) | |
2616 | return -ENOENT; | |
2617 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) | |
2618 | return -EBUSY; | |
2619 | return 0; | |
2620 | } | |
2621 | ||
2622 | /* Check whether we can create an object with dentry child in directory | |
2623 | * dir. | |
2624 | * 1. We can't do it if child already exists (open has special treatment for | |
2625 | * this case, but since we are inlined it's OK) | |
2626 | * 2. We can't do it if dir is read-only (done in permission()) | |
2627 | * 3. We should have write and exec permissions on dir | |
2628 | * 4. We can't do it if dir is immutable (done in permission()) | |
2629 | */ | |
2630 | static inline int may_create(struct inode *dir, struct dentry *child) | |
2631 | { | |
2632 | audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); | |
2633 | if (child->d_inode) | |
2634 | return -EEXIST; | |
2635 | if (IS_DEADDIR(dir)) | |
2636 | return -ENOENT; | |
2637 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * p1 and p2 should be directories on the same fs. | |
2642 | */ | |
2643 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) | |
2644 | { | |
2645 | struct dentry *p; | |
2646 | ||
2647 | if (p1 == p2) { | |
2648 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2649 | return NULL; | |
2650 | } | |
2651 | ||
2652 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); | |
2653 | ||
2654 | p = d_ancestor(p2, p1); | |
2655 | if (p) { | |
2656 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); | |
2657 | inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); | |
2658 | return p; | |
2659 | } | |
2660 | ||
2661 | p = d_ancestor(p1, p2); | |
2662 | if (p) { | |
2663 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2664 | inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); | |
2665 | return p; | |
2666 | } | |
2667 | ||
2668 | inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); | |
2669 | inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); | |
2670 | return NULL; | |
2671 | } | |
2672 | EXPORT_SYMBOL(lock_rename); | |
2673 | ||
2674 | void unlock_rename(struct dentry *p1, struct dentry *p2) | |
2675 | { | |
2676 | inode_unlock(p1->d_inode); | |
2677 | if (p1 != p2) { | |
2678 | inode_unlock(p2->d_inode); | |
2679 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); | |
2680 | } | |
2681 | } | |
2682 | EXPORT_SYMBOL(unlock_rename); | |
2683 | ||
2684 | int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, | |
2685 | bool want_excl) | |
2686 | { | |
2687 | int error = may_create(dir, dentry); | |
2688 | if (error) | |
2689 | return error; | |
2690 | ||
2691 | if (!dir->i_op->create) | |
2692 | return -EACCES; /* shouldn't it be ENOSYS? */ | |
2693 | mode &= S_IALLUGO; | |
2694 | mode |= S_IFREG; | |
2695 | error = security_inode_create(dir, dentry, mode); | |
2696 | if (error) | |
2697 | return error; | |
2698 | error = dir->i_op->create(dir, dentry, mode, want_excl); | |
2699 | if (!error) | |
2700 | fsnotify_create(dir, dentry); | |
2701 | return error; | |
2702 | } | |
2703 | EXPORT_SYMBOL(vfs_create); | |
2704 | ||
2705 | static int may_open(struct path *path, int acc_mode, int flag) | |
2706 | { | |
2707 | struct dentry *dentry = path->dentry; | |
2708 | struct inode *inode = dentry->d_inode; | |
2709 | int error; | |
2710 | ||
2711 | if (!inode) | |
2712 | return -ENOENT; | |
2713 | ||
2714 | switch (inode->i_mode & S_IFMT) { | |
2715 | case S_IFLNK: | |
2716 | return -ELOOP; | |
2717 | case S_IFDIR: | |
2718 | if (acc_mode & MAY_WRITE) | |
2719 | return -EISDIR; | |
2720 | break; | |
2721 | case S_IFBLK: | |
2722 | case S_IFCHR: | |
2723 | if (path->mnt->mnt_flags & MNT_NODEV) | |
2724 | return -EACCES; | |
2725 | /*FALLTHRU*/ | |
2726 | case S_IFIFO: | |
2727 | case S_IFSOCK: | |
2728 | flag &= ~O_TRUNC; | |
2729 | break; | |
2730 | } | |
2731 | ||
2732 | error = inode_permission(inode, MAY_OPEN | acc_mode); | |
2733 | if (error) | |
2734 | return error; | |
2735 | ||
2736 | /* | |
2737 | * An append-only file must be opened in append mode for writing. | |
2738 | */ | |
2739 | if (IS_APPEND(inode)) { | |
2740 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) | |
2741 | return -EPERM; | |
2742 | if (flag & O_TRUNC) | |
2743 | return -EPERM; | |
2744 | } | |
2745 | ||
2746 | /* O_NOATIME can only be set by the owner or superuser */ | |
2747 | if (flag & O_NOATIME && !inode_owner_or_capable(inode)) | |
2748 | return -EPERM; | |
2749 | ||
2750 | return 0; | |
2751 | } | |
2752 | ||
2753 | static int handle_truncate(struct file *filp) | |
2754 | { | |
2755 | struct path *path = &filp->f_path; | |
2756 | struct inode *inode = path->dentry->d_inode; | |
2757 | int error = get_write_access(inode); | |
2758 | if (error) | |
2759 | return error; | |
2760 | /* | |
2761 | * Refuse to truncate files with mandatory locks held on them. | |
2762 | */ | |
2763 | error = locks_verify_locked(filp); | |
2764 | if (!error) | |
2765 | error = security_path_truncate(path); | |
2766 | if (!error) { | |
2767 | error = do_truncate(path->dentry, 0, | |
2768 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, | |
2769 | filp); | |
2770 | } | |
2771 | put_write_access(inode); | |
2772 | return error; | |
2773 | } | |
2774 | ||
2775 | static inline int open_to_namei_flags(int flag) | |
2776 | { | |
2777 | if ((flag & O_ACCMODE) == 3) | |
2778 | flag--; | |
2779 | return flag; | |
2780 | } | |
2781 | ||
2782 | static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) | |
2783 | { | |
2784 | int error = security_path_mknod(dir, dentry, mode, 0); | |
2785 | if (error) | |
2786 | return error; | |
2787 | ||
2788 | error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); | |
2789 | if (error) | |
2790 | return error; | |
2791 | ||
2792 | return security_inode_create(dir->dentry->d_inode, dentry, mode); | |
2793 | } | |
2794 | ||
2795 | /* | |
2796 | * Attempt to atomically look up, create and open a file from a negative | |
2797 | * dentry. | |
2798 | * | |
2799 | * Returns 0 if successful. The file will have been created and attached to | |
2800 | * @file by the filesystem calling finish_open(). | |
2801 | * | |
2802 | * Returns 1 if the file was looked up only or didn't need creating. The | |
2803 | * caller will need to perform the open themselves. @path will have been | |
2804 | * updated to point to the new dentry. This may be negative. | |
2805 | * | |
2806 | * Returns an error code otherwise. | |
2807 | */ | |
2808 | static int atomic_open(struct nameidata *nd, struct dentry *dentry, | |
2809 | struct path *path, struct file *file, | |
2810 | const struct open_flags *op, | |
2811 | bool got_write, bool need_lookup, | |
2812 | int *opened) | |
2813 | { | |
2814 | struct inode *dir = nd->path.dentry->d_inode; | |
2815 | unsigned open_flag = open_to_namei_flags(op->open_flag); | |
2816 | umode_t mode; | |
2817 | int error; | |
2818 | int acc_mode; | |
2819 | int create_error = 0; | |
2820 | struct dentry *const DENTRY_NOT_SET = (void *) -1UL; | |
2821 | bool excl; | |
2822 | ||
2823 | BUG_ON(dentry->d_inode); | |
2824 | ||
2825 | /* Don't create child dentry for a dead directory. */ | |
2826 | if (unlikely(IS_DEADDIR(dir))) { | |
2827 | error = -ENOENT; | |
2828 | goto out; | |
2829 | } | |
2830 | ||
2831 | mode = op->mode; | |
2832 | if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) | |
2833 | mode &= ~current_umask(); | |
2834 | ||
2835 | excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); | |
2836 | if (excl) | |
2837 | open_flag &= ~O_TRUNC; | |
2838 | ||
2839 | /* | |
2840 | * Checking write permission is tricky, bacuse we don't know if we are | |
2841 | * going to actually need it: O_CREAT opens should work as long as the | |
2842 | * file exists. But checking existence breaks atomicity. The trick is | |
2843 | * to check access and if not granted clear O_CREAT from the flags. | |
2844 | * | |
2845 | * Another problem is returing the "right" error value (e.g. for an | |
2846 | * O_EXCL open we want to return EEXIST not EROFS). | |
2847 | */ | |
2848 | if (((open_flag & (O_CREAT | O_TRUNC)) || | |
2849 | (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { | |
2850 | if (!(open_flag & O_CREAT)) { | |
2851 | /* | |
2852 | * No O_CREATE -> atomicity not a requirement -> fall | |
2853 | * back to lookup + open | |
2854 | */ | |
2855 | goto no_open; | |
2856 | } else if (open_flag & (O_EXCL | O_TRUNC)) { | |
2857 | /* Fall back and fail with the right error */ | |
2858 | create_error = -EROFS; | |
2859 | goto no_open; | |
2860 | } else { | |
2861 | /* No side effects, safe to clear O_CREAT */ | |
2862 | create_error = -EROFS; | |
2863 | open_flag &= ~O_CREAT; | |
2864 | } | |
2865 | } | |
2866 | ||
2867 | if (open_flag & O_CREAT) { | |
2868 | error = may_o_create(&nd->path, dentry, mode); | |
2869 | if (error) { | |
2870 | create_error = error; | |
2871 | if (open_flag & O_EXCL) | |
2872 | goto no_open; | |
2873 | open_flag &= ~O_CREAT; | |
2874 | } | |
2875 | } | |
2876 | ||
2877 | if (nd->flags & LOOKUP_DIRECTORY) | |
2878 | open_flag |= O_DIRECTORY; | |
2879 | ||
2880 | file->f_path.dentry = DENTRY_NOT_SET; | |
2881 | file->f_path.mnt = nd->path.mnt; | |
2882 | error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, | |
2883 | opened); | |
2884 | if (error < 0) { | |
2885 | if (create_error && error == -ENOENT) | |
2886 | error = create_error; | |
2887 | goto out; | |
2888 | } | |
2889 | ||
2890 | if (error) { /* returned 1, that is */ | |
2891 | if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { | |
2892 | error = -EIO; | |
2893 | goto out; | |
2894 | } | |
2895 | if (file->f_path.dentry) { | |
2896 | dput(dentry); | |
2897 | dentry = file->f_path.dentry; | |
2898 | } | |
2899 | if (*opened & FILE_CREATED) | |
2900 | fsnotify_create(dir, dentry); | |
2901 | if (!dentry->d_inode) { | |
2902 | WARN_ON(*opened & FILE_CREATED); | |
2903 | if (create_error) { | |
2904 | error = create_error; | |
2905 | goto out; | |
2906 | } | |
2907 | } else { | |
2908 | if (excl && !(*opened & FILE_CREATED)) { | |
2909 | error = -EEXIST; | |
2910 | goto out; | |
2911 | } | |
2912 | } | |
2913 | goto looked_up; | |
2914 | } | |
2915 | ||
2916 | /* | |
2917 | * We didn't have the inode before the open, so check open permission | |
2918 | * here. | |
2919 | */ | |
2920 | acc_mode = op->acc_mode; | |
2921 | if (*opened & FILE_CREATED) { | |
2922 | WARN_ON(!(open_flag & O_CREAT)); | |
2923 | fsnotify_create(dir, dentry); | |
2924 | acc_mode = 0; | |
2925 | } | |
2926 | error = may_open(&file->f_path, acc_mode, open_flag); | |
2927 | if (error) | |
2928 | fput(file); | |
2929 | ||
2930 | out: | |
2931 | dput(dentry); | |
2932 | return error; | |
2933 | ||
2934 | no_open: | |
2935 | if (need_lookup) { | |
2936 | dentry = lookup_real(dir, dentry, nd->flags); | |
2937 | if (IS_ERR(dentry)) | |
2938 | return PTR_ERR(dentry); | |
2939 | ||
2940 | if (create_error) { | |
2941 | int open_flag = op->open_flag; | |
2942 | ||
2943 | error = create_error; | |
2944 | if ((open_flag & O_EXCL)) { | |
2945 | if (!dentry->d_inode) | |
2946 | goto out; | |
2947 | } else if (!dentry->d_inode) { | |
2948 | goto out; | |
2949 | } else if ((open_flag & O_TRUNC) && | |
2950 | d_is_reg(dentry)) { | |
2951 | goto out; | |
2952 | } | |
2953 | /* will fail later, go on to get the right error */ | |
2954 | } | |
2955 | } | |
2956 | looked_up: | |
2957 | path->dentry = dentry; | |
2958 | path->mnt = nd->path.mnt; | |
2959 | return 1; | |
2960 | } | |
2961 | ||
2962 | /* | |
2963 | * Look up and maybe create and open the last component. | |
2964 | * | |
2965 | * Must be called with i_mutex held on parent. | |
2966 | * | |
2967 | * Returns 0 if the file was successfully atomically created (if necessary) and | |
2968 | * opened. In this case the file will be returned attached to @file. | |
2969 | * | |
2970 | * Returns 1 if the file was not completely opened at this time, though lookups | |
2971 | * and creations will have been performed and the dentry returned in @path will | |
2972 | * be positive upon return if O_CREAT was specified. If O_CREAT wasn't | |
2973 | * specified then a negative dentry may be returned. | |
2974 | * | |
2975 | * An error code is returned otherwise. | |
2976 | * | |
2977 | * FILE_CREATE will be set in @*opened if the dentry was created and will be | |
2978 | * cleared otherwise prior to returning. | |
2979 | */ | |
2980 | static int lookup_open(struct nameidata *nd, struct path *path, | |
2981 | struct file *file, | |
2982 | const struct open_flags *op, | |
2983 | bool got_write, int *opened) | |
2984 | { | |
2985 | struct dentry *dir = nd->path.dentry; | |
2986 | struct inode *dir_inode = dir->d_inode; | |
2987 | struct dentry *dentry; | |
2988 | int error; | |
2989 | bool need_lookup = false; | |
2990 | ||
2991 | *opened &= ~FILE_CREATED; | |
2992 | dentry = lookup_dcache(&nd->last, dir, nd->flags); | |
2993 | if (IS_ERR(dentry)) | |
2994 | return PTR_ERR(dentry); | |
2995 | ||
2996 | if (!dentry) { | |
2997 | dentry = d_alloc(dir, &nd->last); | |
2998 | if (unlikely(!dentry)) | |
2999 | return -ENOMEM; | |
3000 | need_lookup = true; | |
3001 | } else if (dentry->d_inode) { | |
3002 | /* Cached positive dentry: will open in f_op->open */ | |
3003 | goto out_no_open; | |
3004 | } | |
3005 | ||
3006 | if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { | |
3007 | return atomic_open(nd, dentry, path, file, op, got_write, | |
3008 | need_lookup, opened); | |
3009 | } | |
3010 | ||
3011 | if (need_lookup) { | |
3012 | BUG_ON(dentry->d_inode); | |
3013 | ||
3014 | dentry = lookup_real(dir_inode, dentry, nd->flags); | |
3015 | if (IS_ERR(dentry)) | |
3016 | return PTR_ERR(dentry); | |
3017 | } | |
3018 | ||
3019 | /* Negative dentry, just create the file */ | |
3020 | if (!dentry->d_inode && (op->open_flag & O_CREAT)) { | |
3021 | umode_t mode = op->mode; | |
3022 | if (!IS_POSIXACL(dir->d_inode)) | |
3023 | mode &= ~current_umask(); | |
3024 | /* | |
3025 | * This write is needed to ensure that a | |
3026 | * rw->ro transition does not occur between | |
3027 | * the time when the file is created and when | |
3028 | * a permanent write count is taken through | |
3029 | * the 'struct file' in finish_open(). | |
3030 | */ | |
3031 | if (!got_write) { | |
3032 | error = -EROFS; | |
3033 | goto out_dput; | |
3034 | } | |
3035 | *opened |= FILE_CREATED; | |
3036 | error = security_path_mknod(&nd->path, dentry, mode, 0); | |
3037 | if (error) | |
3038 | goto out_dput; | |
3039 | error = vfs_create(dir->d_inode, dentry, mode, | |
3040 | nd->flags & LOOKUP_EXCL); | |
3041 | if (error) | |
3042 | goto out_dput; | |
3043 | } | |
3044 | out_no_open: | |
3045 | path->dentry = dentry; | |
3046 | path->mnt = nd->path.mnt; | |
3047 | return 1; | |
3048 | ||
3049 | out_dput: | |
3050 | dput(dentry); | |
3051 | return error; | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * Handle the last step of open() | |
3056 | */ | |
3057 | static int do_last(struct nameidata *nd, | |
3058 | struct file *file, const struct open_flags *op, | |
3059 | int *opened) | |
3060 | { | |
3061 | struct dentry *dir = nd->path.dentry; | |
3062 | int open_flag = op->open_flag; | |
3063 | bool will_truncate = (open_flag & O_TRUNC) != 0; | |
3064 | bool got_write = false; | |
3065 | int acc_mode = op->acc_mode; | |
3066 | unsigned seq; | |
3067 | struct inode *inode; | |
3068 | struct path save_parent = { .dentry = NULL, .mnt = NULL }; | |
3069 | struct path path; | |
3070 | bool retried = false; | |
3071 | int error; | |
3072 | ||
3073 | nd->flags &= ~LOOKUP_PARENT; | |
3074 | nd->flags |= op->intent; | |
3075 | ||
3076 | if (nd->last_type != LAST_NORM) { | |
3077 | error = handle_dots(nd, nd->last_type); | |
3078 | if (unlikely(error)) | |
3079 | return error; | |
3080 | goto finish_open; | |
3081 | } | |
3082 | ||
3083 | if (!(open_flag & O_CREAT)) { | |
3084 | if (nd->last.name[nd->last.len]) | |
3085 | nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; | |
3086 | /* we _can_ be in RCU mode here */ | |
3087 | error = lookup_fast(nd, &path, &inode, &seq); | |
3088 | if (likely(error > 0)) | |
3089 | goto finish_lookup; | |
3090 | ||
3091 | if (error < 0) | |
3092 | return error; | |
3093 | ||
3094 | BUG_ON(nd->inode != dir->d_inode); | |
3095 | BUG_ON(nd->flags & LOOKUP_RCU); | |
3096 | } else { | |
3097 | /* create side of things */ | |
3098 | /* | |
3099 | * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED | |
3100 | * has been cleared when we got to the last component we are | |
3101 | * about to look up | |
3102 | */ | |
3103 | error = complete_walk(nd); | |
3104 | if (error) | |
3105 | return error; | |
3106 | ||
3107 | audit_inode(nd->name, dir, LOOKUP_PARENT); | |
3108 | /* trailing slashes? */ | |
3109 | if (unlikely(nd->last.name[nd->last.len])) | |
3110 | return -EISDIR; | |
3111 | } | |
3112 | ||
3113 | retry_lookup: | |
3114 | if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { | |
3115 | error = mnt_want_write(nd->path.mnt); | |
3116 | if (!error) | |
3117 | got_write = true; | |
3118 | /* | |
3119 | * do _not_ fail yet - we might not need that or fail with | |
3120 | * a different error; let lookup_open() decide; we'll be | |
3121 | * dropping this one anyway. | |
3122 | */ | |
3123 | } | |
3124 | inode_lock(dir->d_inode); | |
3125 | error = lookup_open(nd, &path, file, op, got_write, opened); | |
3126 | inode_unlock(dir->d_inode); | |
3127 | ||
3128 | if (error <= 0) { | |
3129 | if (error) | |
3130 | goto out; | |
3131 | ||
3132 | if ((*opened & FILE_CREATED) || | |
3133 | !S_ISREG(file_inode(file)->i_mode)) | |
3134 | will_truncate = false; | |
3135 | ||
3136 | audit_inode(nd->name, file->f_path.dentry, 0); | |
3137 | goto opened; | |
3138 | } | |
3139 | ||
3140 | if (*opened & FILE_CREATED) { | |
3141 | /* Don't check for write permission, don't truncate */ | |
3142 | open_flag &= ~O_TRUNC; | |
3143 | will_truncate = false; | |
3144 | acc_mode = 0; | |
3145 | path_to_nameidata(&path, nd); | |
3146 | goto finish_open_created; | |
3147 | } | |
3148 | ||
3149 | /* | |
3150 | * If atomic_open() acquired write access it is dropped now due to | |
3151 | * possible mount and symlink following (this might be optimized away if | |
3152 | * necessary...) | |
3153 | */ | |
3154 | if (got_write) { | |
3155 | mnt_drop_write(nd->path.mnt); | |
3156 | got_write = false; | |
3157 | } | |
3158 | ||
3159 | if (unlikely(d_is_negative(path.dentry))) { | |
3160 | path_to_nameidata(&path, nd); | |
3161 | return -ENOENT; | |
3162 | } | |
3163 | ||
3164 | /* | |
3165 | * create/update audit record if it already exists. | |
3166 | */ | |
3167 | audit_inode(nd->name, path.dentry, 0); | |
3168 | ||
3169 | if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { | |
3170 | path_to_nameidata(&path, nd); | |
3171 | return -EEXIST; | |
3172 | } | |
3173 | ||
3174 | error = follow_managed(&path, nd); | |
3175 | if (unlikely(error < 0)) | |
3176 | return error; | |
3177 | ||
3178 | seq = 0; /* out of RCU mode, so the value doesn't matter */ | |
3179 | inode = d_backing_inode(path.dentry); | |
3180 | finish_lookup: | |
3181 | if (nd->depth) | |
3182 | put_link(nd); | |
3183 | error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, | |
3184 | inode, seq); | |
3185 | if (unlikely(error)) | |
3186 | return error; | |
3187 | ||
3188 | if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { | |
3189 | path_to_nameidata(&path, nd); | |
3190 | } else { | |
3191 | save_parent.dentry = nd->path.dentry; | |
3192 | save_parent.mnt = mntget(path.mnt); | |
3193 | nd->path.dentry = path.dentry; | |
3194 | ||
3195 | } | |
3196 | nd->inode = inode; | |
3197 | nd->seq = seq; | |
3198 | /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ | |
3199 | finish_open: | |
3200 | error = complete_walk(nd); | |
3201 | if (error) { | |
3202 | path_put(&save_parent); | |
3203 | return error; | |
3204 | } | |
3205 | audit_inode(nd->name, nd->path.dentry, 0); | |
3206 | if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) { | |
3207 | error = -ELOOP; | |
3208 | goto out; | |
3209 | } | |
3210 | error = -EISDIR; | |
3211 | if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) | |
3212 | goto out; | |
3213 | error = -ENOTDIR; | |
3214 | if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) | |
3215 | goto out; | |
3216 | if (!d_is_reg(nd->path.dentry)) | |
3217 | will_truncate = false; | |
3218 | ||
3219 | if (will_truncate) { | |
3220 | error = mnt_want_write(nd->path.mnt); | |
3221 | if (error) | |
3222 | goto out; | |
3223 | got_write = true; | |
3224 | } | |
3225 | finish_open_created: | |
3226 | if (likely(!(open_flag & O_PATH))) { | |
3227 | error = may_open(&nd->path, acc_mode, open_flag); | |
3228 | if (error) | |
3229 | goto out; | |
3230 | } | |
3231 | BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ | |
3232 | error = vfs_open(&nd->path, file, current_cred()); | |
3233 | if (!error) { | |
3234 | *opened |= FILE_OPENED; | |
3235 | } else { | |
3236 | if (error == -EOPENSTALE) | |
3237 | goto stale_open; | |
3238 | goto out; | |
3239 | } | |
3240 | opened: | |
3241 | error = open_check_o_direct(file); | |
3242 | if (error) | |
3243 | goto exit_fput; | |
3244 | error = ima_file_check(file, op->acc_mode, *opened); | |
3245 | if (error) | |
3246 | goto exit_fput; | |
3247 | ||
3248 | if (will_truncate) { | |
3249 | error = handle_truncate(file); | |
3250 | if (error) | |
3251 | goto exit_fput; | |
3252 | } | |
3253 | out: | |
3254 | if (unlikely(error > 0)) { | |
3255 | WARN_ON(1); | |
3256 | error = -EINVAL; | |
3257 | } | |
3258 | if (got_write) | |
3259 | mnt_drop_write(nd->path.mnt); | |
3260 | path_put(&save_parent); | |
3261 | return error; | |
3262 | ||
3263 | exit_fput: | |
3264 | fput(file); | |
3265 | goto out; | |
3266 | ||
3267 | stale_open: | |
3268 | /* If no saved parent or already retried then can't retry */ | |
3269 | if (!save_parent.dentry || retried) | |
3270 | goto out; | |
3271 | ||
3272 | BUG_ON(save_parent.dentry != dir); | |
3273 | path_put(&nd->path); | |
3274 | nd->path = save_parent; | |
3275 | nd->inode = dir->d_inode; | |
3276 | save_parent.mnt = NULL; | |
3277 | save_parent.dentry = NULL; | |
3278 | if (got_write) { | |
3279 | mnt_drop_write(nd->path.mnt); | |
3280 | got_write = false; | |
3281 | } | |
3282 | retried = true; | |
3283 | goto retry_lookup; | |
3284 | } | |
3285 | ||
3286 | static int do_tmpfile(struct nameidata *nd, unsigned flags, | |
3287 | const struct open_flags *op, | |
3288 | struct file *file, int *opened) | |
3289 | { | |
3290 | static const struct qstr name = QSTR_INIT("/", 1); | |
3291 | struct dentry *child; | |
3292 | struct inode *dir; | |
3293 | struct path path; | |
3294 | int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); | |
3295 | if (unlikely(error)) | |
3296 | return error; | |
3297 | error = mnt_want_write(path.mnt); | |
3298 | if (unlikely(error)) | |
3299 | goto out; | |
3300 | dir = path.dentry->d_inode; | |
3301 | /* we want directory to be writable */ | |
3302 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); | |
3303 | if (error) | |
3304 | goto out2; | |
3305 | if (!dir->i_op->tmpfile) { | |
3306 | error = -EOPNOTSUPP; | |
3307 | goto out2; | |
3308 | } | |
3309 | child = d_alloc(path.dentry, &name); | |
3310 | if (unlikely(!child)) { | |
3311 | error = -ENOMEM; | |
3312 | goto out2; | |
3313 | } | |
3314 | dput(path.dentry); | |
3315 | path.dentry = child; | |
3316 | error = dir->i_op->tmpfile(dir, child, op->mode); | |
3317 | if (error) | |
3318 | goto out2; | |
3319 | audit_inode(nd->name, child, 0); | |
3320 | /* Don't check for other permissions, the inode was just created */ | |
3321 | error = may_open(&path, 0, op->open_flag); | |
3322 | if (error) | |
3323 | goto out2; | |
3324 | file->f_path.mnt = path.mnt; | |
3325 | error = finish_open(file, child, NULL, opened); | |
3326 | if (error) | |
3327 | goto out2; | |
3328 | error = open_check_o_direct(file); | |
3329 | if (error) { | |
3330 | fput(file); | |
3331 | } else if (!(op->open_flag & O_EXCL)) { | |
3332 | struct inode *inode = file_inode(file); | |
3333 | spin_lock(&inode->i_lock); | |
3334 | inode->i_state |= I_LINKABLE; | |
3335 | spin_unlock(&inode->i_lock); | |
3336 | } | |
3337 | out2: | |
3338 | mnt_drop_write(path.mnt); | |
3339 | out: | |
3340 | path_put(&path); | |
3341 | return error; | |
3342 | } | |
3343 | ||
3344 | static struct file *path_openat(struct nameidata *nd, | |
3345 | const struct open_flags *op, unsigned flags) | |
3346 | { | |
3347 | const char *s; | |
3348 | struct file *file; | |
3349 | int opened = 0; | |
3350 | int error; | |
3351 | ||
3352 | file = get_empty_filp(); | |
3353 | if (IS_ERR(file)) | |
3354 | return file; | |
3355 | ||
3356 | file->f_flags = op->open_flag; | |
3357 | ||
3358 | if (unlikely(file->f_flags & __O_TMPFILE)) { | |
3359 | error = do_tmpfile(nd, flags, op, file, &opened); | |
3360 | goto out2; | |
3361 | } | |
3362 | ||
3363 | s = path_init(nd, flags); | |
3364 | if (IS_ERR(s)) { | |
3365 | put_filp(file); | |
3366 | return ERR_CAST(s); | |
3367 | } | |
3368 | while (!(error = link_path_walk(s, nd)) && | |
3369 | (error = do_last(nd, file, op, &opened)) > 0) { | |
3370 | nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); | |
3371 | s = trailing_symlink(nd); | |
3372 | if (IS_ERR(s)) { | |
3373 | error = PTR_ERR(s); | |
3374 | break; | |
3375 | } | |
3376 | } | |
3377 | terminate_walk(nd); | |
3378 | out2: | |
3379 | if (!(opened & FILE_OPENED)) { | |
3380 | BUG_ON(!error); | |
3381 | put_filp(file); | |
3382 | } | |
3383 | if (unlikely(error)) { | |
3384 | if (error == -EOPENSTALE) { | |
3385 | if (flags & LOOKUP_RCU) | |
3386 | error = -ECHILD; | |
3387 | else | |
3388 | error = -ESTALE; | |
3389 | } | |
3390 | file = ERR_PTR(error); | |
3391 | } | |
3392 | return file; | |
3393 | } | |
3394 | ||
3395 | struct file *do_filp_open(int dfd, struct filename *pathname, | |
3396 | const struct open_flags *op) | |
3397 | { | |
3398 | struct nameidata nd; | |
3399 | int flags = op->lookup_flags; | |
3400 | struct file *filp; | |
3401 | ||
3402 | set_nameidata(&nd, dfd, pathname); | |
3403 | filp = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3404 | if (unlikely(filp == ERR_PTR(-ECHILD))) | |
3405 | filp = path_openat(&nd, op, flags); | |
3406 | if (unlikely(filp == ERR_PTR(-ESTALE))) | |
3407 | filp = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3408 | restore_nameidata(); | |
3409 | return filp; | |
3410 | } | |
3411 | ||
3412 | struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, | |
3413 | const char *name, const struct open_flags *op) | |
3414 | { | |
3415 | struct nameidata nd; | |
3416 | struct file *file; | |
3417 | struct filename *filename; | |
3418 | int flags = op->lookup_flags | LOOKUP_ROOT; | |
3419 | ||
3420 | nd.root.mnt = mnt; | |
3421 | nd.root.dentry = dentry; | |
3422 | ||
3423 | if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) | |
3424 | return ERR_PTR(-ELOOP); | |
3425 | ||
3426 | filename = getname_kernel(name); | |
3427 | if (IS_ERR(filename)) | |
3428 | return ERR_CAST(filename); | |
3429 | ||
3430 | set_nameidata(&nd, -1, filename); | |
3431 | file = path_openat(&nd, op, flags | LOOKUP_RCU); | |
3432 | if (unlikely(file == ERR_PTR(-ECHILD))) | |
3433 | file = path_openat(&nd, op, flags); | |
3434 | if (unlikely(file == ERR_PTR(-ESTALE))) | |
3435 | file = path_openat(&nd, op, flags | LOOKUP_REVAL); | |
3436 | restore_nameidata(); | |
3437 | putname(filename); | |
3438 | return file; | |
3439 | } | |
3440 | ||
3441 | static struct dentry *filename_create(int dfd, struct filename *name, | |
3442 | struct path *path, unsigned int lookup_flags) | |
3443 | { | |
3444 | struct dentry *dentry = ERR_PTR(-EEXIST); | |
3445 | struct qstr last; | |
3446 | int type; | |
3447 | int err2; | |
3448 | int error; | |
3449 | bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); | |
3450 | ||
3451 | /* | |
3452 | * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any | |
3453 | * other flags passed in are ignored! | |
3454 | */ | |
3455 | lookup_flags &= LOOKUP_REVAL; | |
3456 | ||
3457 | name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); | |
3458 | if (IS_ERR(name)) | |
3459 | return ERR_CAST(name); | |
3460 | ||
3461 | /* | |
3462 | * Yucky last component or no last component at all? | |
3463 | * (foo/., foo/.., /////) | |
3464 | */ | |
3465 | if (unlikely(type != LAST_NORM)) | |
3466 | goto out; | |
3467 | ||
3468 | /* don't fail immediately if it's r/o, at least try to report other errors */ | |
3469 | err2 = mnt_want_write(path->mnt); | |
3470 | /* | |
3471 | * Do the final lookup. | |
3472 | */ | |
3473 | lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; | |
3474 | inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); | |
3475 | dentry = __lookup_hash(&last, path->dentry, lookup_flags); | |
3476 | if (IS_ERR(dentry)) | |
3477 | goto unlock; | |
3478 | ||
3479 | error = -EEXIST; | |
3480 | if (d_is_positive(dentry)) | |
3481 | goto fail; | |
3482 | ||
3483 | /* | |
3484 | * Special case - lookup gave negative, but... we had foo/bar/ | |
3485 | * From the vfs_mknod() POV we just have a negative dentry - | |
3486 | * all is fine. Let's be bastards - you had / on the end, you've | |
3487 | * been asking for (non-existent) directory. -ENOENT for you. | |
3488 | */ | |
3489 | if (unlikely(!is_dir && last.name[last.len])) { | |
3490 | error = -ENOENT; | |
3491 | goto fail; | |
3492 | } | |
3493 | if (unlikely(err2)) { | |
3494 | error = err2; | |
3495 | goto fail; | |
3496 | } | |
3497 | putname(name); | |
3498 | return dentry; | |
3499 | fail: | |
3500 | dput(dentry); | |
3501 | dentry = ERR_PTR(error); | |
3502 | unlock: | |
3503 | inode_unlock(path->dentry->d_inode); | |
3504 | if (!err2) | |
3505 | mnt_drop_write(path->mnt); | |
3506 | out: | |
3507 | path_put(path); | |
3508 | putname(name); | |
3509 | return dentry; | |
3510 | } | |
3511 | ||
3512 | struct dentry *kern_path_create(int dfd, const char *pathname, | |
3513 | struct path *path, unsigned int lookup_flags) | |
3514 | { | |
3515 | return filename_create(dfd, getname_kernel(pathname), | |
3516 | path, lookup_flags); | |
3517 | } | |
3518 | EXPORT_SYMBOL(kern_path_create); | |
3519 | ||
3520 | void done_path_create(struct path *path, struct dentry *dentry) | |
3521 | { | |
3522 | dput(dentry); | |
3523 | inode_unlock(path->dentry->d_inode); | |
3524 | mnt_drop_write(path->mnt); | |
3525 | path_put(path); | |
3526 | } | |
3527 | EXPORT_SYMBOL(done_path_create); | |
3528 | ||
3529 | inline struct dentry *user_path_create(int dfd, const char __user *pathname, | |
3530 | struct path *path, unsigned int lookup_flags) | |
3531 | { | |
3532 | return filename_create(dfd, getname(pathname), path, lookup_flags); | |
3533 | } | |
3534 | EXPORT_SYMBOL(user_path_create); | |
3535 | ||
3536 | int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) | |
3537 | { | |
3538 | int error = may_create(dir, dentry); | |
3539 | ||
3540 | if (error) | |
3541 | return error; | |
3542 | ||
3543 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) | |
3544 | return -EPERM; | |
3545 | ||
3546 | if (!dir->i_op->mknod) | |
3547 | return -EPERM; | |
3548 | ||
3549 | error = devcgroup_inode_mknod(mode, dev); | |
3550 | if (error) | |
3551 | return error; | |
3552 | ||
3553 | error = security_inode_mknod(dir, dentry, mode, dev); | |
3554 | if (error) | |
3555 | return error; | |
3556 | ||
3557 | error = dir->i_op->mknod(dir, dentry, mode, dev); | |
3558 | if (!error) | |
3559 | fsnotify_create(dir, dentry); | |
3560 | return error; | |
3561 | } | |
3562 | EXPORT_SYMBOL(vfs_mknod); | |
3563 | ||
3564 | static int may_mknod(umode_t mode) | |
3565 | { | |
3566 | switch (mode & S_IFMT) { | |
3567 | case S_IFREG: | |
3568 | case S_IFCHR: | |
3569 | case S_IFBLK: | |
3570 | case S_IFIFO: | |
3571 | case S_IFSOCK: | |
3572 | case 0: /* zero mode translates to S_IFREG */ | |
3573 | return 0; | |
3574 | case S_IFDIR: | |
3575 | return -EPERM; | |
3576 | default: | |
3577 | return -EINVAL; | |
3578 | } | |
3579 | } | |
3580 | ||
3581 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, | |
3582 | unsigned, dev) | |
3583 | { | |
3584 | struct dentry *dentry; | |
3585 | struct path path; | |
3586 | int error; | |
3587 | unsigned int lookup_flags = 0; | |
3588 | ||
3589 | error = may_mknod(mode); | |
3590 | if (error) | |
3591 | return error; | |
3592 | retry: | |
3593 | dentry = user_path_create(dfd, filename, &path, lookup_flags); | |
3594 | if (IS_ERR(dentry)) | |
3595 | return PTR_ERR(dentry); | |
3596 | ||
3597 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3598 | mode &= ~current_umask(); | |
3599 | error = security_path_mknod(&path, dentry, mode, dev); | |
3600 | if (error) | |
3601 | goto out; | |
3602 | switch (mode & S_IFMT) { | |
3603 | case 0: case S_IFREG: | |
3604 | error = vfs_create(path.dentry->d_inode,dentry,mode,true); | |
3605 | break; | |
3606 | case S_IFCHR: case S_IFBLK: | |
3607 | error = vfs_mknod(path.dentry->d_inode,dentry,mode, | |
3608 | new_decode_dev(dev)); | |
3609 | break; | |
3610 | case S_IFIFO: case S_IFSOCK: | |
3611 | error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); | |
3612 | break; | |
3613 | } | |
3614 | out: | |
3615 | done_path_create(&path, dentry); | |
3616 | if (retry_estale(error, lookup_flags)) { | |
3617 | lookup_flags |= LOOKUP_REVAL; | |
3618 | goto retry; | |
3619 | } | |
3620 | return error; | |
3621 | } | |
3622 | ||
3623 | SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) | |
3624 | { | |
3625 | return sys_mknodat(AT_FDCWD, filename, mode, dev); | |
3626 | } | |
3627 | ||
3628 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) | |
3629 | { | |
3630 | int error = may_create(dir, dentry); | |
3631 | unsigned max_links = dir->i_sb->s_max_links; | |
3632 | ||
3633 | if (error) | |
3634 | return error; | |
3635 | ||
3636 | if (!dir->i_op->mkdir) | |
3637 | return -EPERM; | |
3638 | ||
3639 | mode &= (S_IRWXUGO|S_ISVTX); | |
3640 | error = security_inode_mkdir(dir, dentry, mode); | |
3641 | if (error) | |
3642 | return error; | |
3643 | ||
3644 | if (max_links && dir->i_nlink >= max_links) | |
3645 | return -EMLINK; | |
3646 | ||
3647 | error = dir->i_op->mkdir(dir, dentry, mode); | |
3648 | if (!error) | |
3649 | fsnotify_mkdir(dir, dentry); | |
3650 | return error; | |
3651 | } | |
3652 | EXPORT_SYMBOL(vfs_mkdir); | |
3653 | ||
3654 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) | |
3655 | { | |
3656 | struct dentry *dentry; | |
3657 | struct path path; | |
3658 | int error; | |
3659 | unsigned int lookup_flags = LOOKUP_DIRECTORY; | |
3660 | ||
3661 | retry: | |
3662 | dentry = user_path_create(dfd, pathname, &path, lookup_flags); | |
3663 | if (IS_ERR(dentry)) | |
3664 | return PTR_ERR(dentry); | |
3665 | ||
3666 | if (!IS_POSIXACL(path.dentry->d_inode)) | |
3667 | mode &= ~current_umask(); | |
3668 | error = security_path_mkdir(&path, dentry, mode); | |
3669 | if (!error) | |
3670 | error = vfs_mkdir(path.dentry->d_inode, dentry, mode); | |
3671 | done_path_create(&path, dentry); | |
3672 | if (retry_estale(error, lookup_flags)) { | |
3673 | lookup_flags |= LOOKUP_REVAL; | |
3674 | goto retry; | |
3675 | } | |
3676 | return error; | |
3677 | } | |
3678 | ||
3679 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) | |
3680 | { | |
3681 | return sys_mkdirat(AT_FDCWD, pathname, mode); | |
3682 | } | |
3683 | ||
3684 | /* | |
3685 | * The dentry_unhash() helper will try to drop the dentry early: we | |
3686 | * should have a usage count of 1 if we're the only user of this | |
3687 | * dentry, and if that is true (possibly after pruning the dcache), | |
3688 | * then we drop the dentry now. | |
3689 | * | |
3690 | * A low-level filesystem can, if it choses, legally | |
3691 | * do a | |
3692 | * | |
3693 | * if (!d_unhashed(dentry)) | |
3694 | * return -EBUSY; | |
3695 | * | |
3696 | * if it cannot handle the case of removing a directory | |
3697 | * that is still in use by something else.. | |
3698 | */ | |
3699 | void dentry_unhash(struct dentry *dentry) | |
3700 | { | |
3701 | shrink_dcache_parent(dentry); | |
3702 | spin_lock(&dentry->d_lock); | |
3703 | if (dentry->d_lockref.count == 1) | |
3704 | __d_drop(dentry); | |
3705 | spin_unlock(&dentry->d_lock); | |
3706 | } | |
3707 | EXPORT_SYMBOL(dentry_unhash); | |
3708 | ||
3709 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) | |
3710 | { | |
3711 | int error = may_delete(dir, dentry, 1); | |
3712 | ||
3713 | if (error) | |
3714 | return error; | |
3715 | ||
3716 | if (!dir->i_op->rmdir) | |
3717 | return -EPERM; | |
3718 | ||
3719 | dget(dentry); | |
3720 | inode_lock(dentry->d_inode); | |
3721 | ||
3722 | error = -EBUSY; | |
3723 | if (is_local_mountpoint(dentry)) | |
3724 | goto out; | |
3725 | ||
3726 | error = security_inode_rmdir(dir, dentry); | |
3727 | if (error) | |
3728 | goto out; | |
3729 | ||
3730 | shrink_dcache_parent(dentry); | |
3731 | error = dir->i_op->rmdir(dir, dentry); | |
3732 | if (error) | |
3733 | goto out; | |
3734 | ||
3735 | dentry->d_inode->i_flags |= S_DEAD; | |
3736 | dont_mount(dentry); | |
3737 | detach_mounts(dentry); | |
3738 | ||
3739 | out: | |
3740 | inode_unlock(dentry->d_inode); | |
3741 | dput(dentry); | |
3742 | if (!error) | |
3743 | d_delete(dentry); | |
3744 | return error; | |
3745 | } | |
3746 | EXPORT_SYMBOL(vfs_rmdir); | |
3747 | ||
3748 | static long do_rmdir(int dfd, const char __user *pathname) | |
3749 | { | |
3750 | int error = 0; | |
3751 | struct filename *name; | |
3752 | struct dentry *dentry; | |
3753 | struct path path; | |
3754 | struct qstr last; | |
3755 | int type; | |
3756 | unsigned int lookup_flags = 0; | |
3757 | retry: | |
3758 | name = user_path_parent(dfd, pathname, | |
3759 | &path, &last, &type, lookup_flags); | |
3760 | if (IS_ERR(name)) | |
3761 | return PTR_ERR(name); | |
3762 | ||
3763 | switch (type) { | |
3764 | case LAST_DOTDOT: | |
3765 | error = -ENOTEMPTY; | |
3766 | goto exit1; | |
3767 | case LAST_DOT: | |
3768 | error = -EINVAL; | |
3769 | goto exit1; | |
3770 | case LAST_ROOT: | |
3771 | error = -EBUSY; | |
3772 | goto exit1; | |
3773 | } | |
3774 | ||
3775 | error = mnt_want_write(path.mnt); | |
3776 | if (error) | |
3777 | goto exit1; | |
3778 | ||
3779 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
3780 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
3781 | error = PTR_ERR(dentry); | |
3782 | if (IS_ERR(dentry)) | |
3783 | goto exit2; | |
3784 | if (!dentry->d_inode) { | |
3785 | error = -ENOENT; | |
3786 | goto exit3; | |
3787 | } | |
3788 | error = security_path_rmdir(&path, dentry); | |
3789 | if (error) | |
3790 | goto exit3; | |
3791 | error = vfs_rmdir(path.dentry->d_inode, dentry); | |
3792 | exit3: | |
3793 | dput(dentry); | |
3794 | exit2: | |
3795 | inode_unlock(path.dentry->d_inode); | |
3796 | mnt_drop_write(path.mnt); | |
3797 | exit1: | |
3798 | path_put(&path); | |
3799 | putname(name); | |
3800 | if (retry_estale(error, lookup_flags)) { | |
3801 | lookup_flags |= LOOKUP_REVAL; | |
3802 | goto retry; | |
3803 | } | |
3804 | return error; | |
3805 | } | |
3806 | ||
3807 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) | |
3808 | { | |
3809 | return do_rmdir(AT_FDCWD, pathname); | |
3810 | } | |
3811 | ||
3812 | /** | |
3813 | * vfs_unlink - unlink a filesystem object | |
3814 | * @dir: parent directory | |
3815 | * @dentry: victim | |
3816 | * @delegated_inode: returns victim inode, if the inode is delegated. | |
3817 | * | |
3818 | * The caller must hold dir->i_mutex. | |
3819 | * | |
3820 | * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and | |
3821 | * return a reference to the inode in delegated_inode. The caller | |
3822 | * should then break the delegation on that inode and retry. Because | |
3823 | * breaking a delegation may take a long time, the caller should drop | |
3824 | * dir->i_mutex before doing so. | |
3825 | * | |
3826 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
3827 | * be appropriate for callers that expect the underlying filesystem not | |
3828 | * to be NFS exported. | |
3829 | */ | |
3830 | int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) | |
3831 | { | |
3832 | struct inode *target = dentry->d_inode; | |
3833 | int error = may_delete(dir, dentry, 0); | |
3834 | ||
3835 | if (error) | |
3836 | return error; | |
3837 | ||
3838 | if (!dir->i_op->unlink) | |
3839 | return -EPERM; | |
3840 | ||
3841 | inode_lock(target); | |
3842 | if (is_local_mountpoint(dentry)) | |
3843 | error = -EBUSY; | |
3844 | else { | |
3845 | error = security_inode_unlink(dir, dentry); | |
3846 | if (!error) { | |
3847 | error = try_break_deleg(target, delegated_inode); | |
3848 | if (error) | |
3849 | goto out; | |
3850 | error = dir->i_op->unlink(dir, dentry); | |
3851 | if (!error) { | |
3852 | dont_mount(dentry); | |
3853 | detach_mounts(dentry); | |
3854 | } | |
3855 | } | |
3856 | } | |
3857 | out: | |
3858 | inode_unlock(target); | |
3859 | ||
3860 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ | |
3861 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { | |
3862 | fsnotify_link_count(target); | |
3863 | d_delete(dentry); | |
3864 | } | |
3865 | ||
3866 | return error; | |
3867 | } | |
3868 | EXPORT_SYMBOL(vfs_unlink); | |
3869 | ||
3870 | /* | |
3871 | * Make sure that the actual truncation of the file will occur outside its | |
3872 | * directory's i_mutex. Truncate can take a long time if there is a lot of | |
3873 | * writeout happening, and we don't want to prevent access to the directory | |
3874 | * while waiting on the I/O. | |
3875 | */ | |
3876 | static long do_unlinkat(int dfd, const char __user *pathname) | |
3877 | { | |
3878 | int error; | |
3879 | struct filename *name; | |
3880 | struct dentry *dentry; | |
3881 | struct path path; | |
3882 | struct qstr last; | |
3883 | int type; | |
3884 | struct inode *inode = NULL; | |
3885 | struct inode *delegated_inode = NULL; | |
3886 | unsigned int lookup_flags = 0; | |
3887 | retry: | |
3888 | name = user_path_parent(dfd, pathname, | |
3889 | &path, &last, &type, lookup_flags); | |
3890 | if (IS_ERR(name)) | |
3891 | return PTR_ERR(name); | |
3892 | ||
3893 | error = -EISDIR; | |
3894 | if (type != LAST_NORM) | |
3895 | goto exit1; | |
3896 | ||
3897 | error = mnt_want_write(path.mnt); | |
3898 | if (error) | |
3899 | goto exit1; | |
3900 | retry_deleg: | |
3901 | inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); | |
3902 | dentry = __lookup_hash(&last, path.dentry, lookup_flags); | |
3903 | error = PTR_ERR(dentry); | |
3904 | if (!IS_ERR(dentry)) { | |
3905 | /* Why not before? Because we want correct error value */ | |
3906 | if (last.name[last.len]) | |
3907 | goto slashes; | |
3908 | inode = dentry->d_inode; | |
3909 | if (d_is_negative(dentry)) | |
3910 | goto slashes; | |
3911 | ihold(inode); | |
3912 | error = security_path_unlink(&path, dentry); | |
3913 | if (error) | |
3914 | goto exit2; | |
3915 | error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); | |
3916 | exit2: | |
3917 | dput(dentry); | |
3918 | } | |
3919 | inode_unlock(path.dentry->d_inode); | |
3920 | if (inode) | |
3921 | iput(inode); /* truncate the inode here */ | |
3922 | inode = NULL; | |
3923 | if (delegated_inode) { | |
3924 | error = break_deleg_wait(&delegated_inode); | |
3925 | if (!error) | |
3926 | goto retry_deleg; | |
3927 | } | |
3928 | mnt_drop_write(path.mnt); | |
3929 | exit1: | |
3930 | path_put(&path); | |
3931 | putname(name); | |
3932 | if (retry_estale(error, lookup_flags)) { | |
3933 | lookup_flags |= LOOKUP_REVAL; | |
3934 | inode = NULL; | |
3935 | goto retry; | |
3936 | } | |
3937 | return error; | |
3938 | ||
3939 | slashes: | |
3940 | if (d_is_negative(dentry)) | |
3941 | error = -ENOENT; | |
3942 | else if (d_is_dir(dentry)) | |
3943 | error = -EISDIR; | |
3944 | else | |
3945 | error = -ENOTDIR; | |
3946 | goto exit2; | |
3947 | } | |
3948 | ||
3949 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) | |
3950 | { | |
3951 | if ((flag & ~AT_REMOVEDIR) != 0) | |
3952 | return -EINVAL; | |
3953 | ||
3954 | if (flag & AT_REMOVEDIR) | |
3955 | return do_rmdir(dfd, pathname); | |
3956 | ||
3957 | return do_unlinkat(dfd, pathname); | |
3958 | } | |
3959 | ||
3960 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) | |
3961 | { | |
3962 | return do_unlinkat(AT_FDCWD, pathname); | |
3963 | } | |
3964 | ||
3965 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) | |
3966 | { | |
3967 | int error = may_create(dir, dentry); | |
3968 | ||
3969 | if (error) | |
3970 | return error; | |
3971 | ||
3972 | if (!dir->i_op->symlink) | |
3973 | return -EPERM; | |
3974 | ||
3975 | error = security_inode_symlink(dir, dentry, oldname); | |
3976 | if (error) | |
3977 | return error; | |
3978 | ||
3979 | error = dir->i_op->symlink(dir, dentry, oldname); | |
3980 | if (!error) | |
3981 | fsnotify_create(dir, dentry); | |
3982 | return error; | |
3983 | } | |
3984 | EXPORT_SYMBOL(vfs_symlink); | |
3985 | ||
3986 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, | |
3987 | int, newdfd, const char __user *, newname) | |
3988 | { | |
3989 | int error; | |
3990 | struct filename *from; | |
3991 | struct dentry *dentry; | |
3992 | struct path path; | |
3993 | unsigned int lookup_flags = 0; | |
3994 | ||
3995 | from = getname(oldname); | |
3996 | if (IS_ERR(from)) | |
3997 | return PTR_ERR(from); | |
3998 | retry: | |
3999 | dentry = user_path_create(newdfd, newname, &path, lookup_flags); | |
4000 | error = PTR_ERR(dentry); | |
4001 | if (IS_ERR(dentry)) | |
4002 | goto out_putname; | |
4003 | ||
4004 | error = security_path_symlink(&path, dentry, from->name); | |
4005 | if (!error) | |
4006 | error = vfs_symlink(path.dentry->d_inode, dentry, from->name); | |
4007 | done_path_create(&path, dentry); | |
4008 | if (retry_estale(error, lookup_flags)) { | |
4009 | lookup_flags |= LOOKUP_REVAL; | |
4010 | goto retry; | |
4011 | } | |
4012 | out_putname: | |
4013 | putname(from); | |
4014 | return error; | |
4015 | } | |
4016 | ||
4017 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) | |
4018 | { | |
4019 | return sys_symlinkat(oldname, AT_FDCWD, newname); | |
4020 | } | |
4021 | ||
4022 | /** | |
4023 | * vfs_link - create a new link | |
4024 | * @old_dentry: object to be linked | |
4025 | * @dir: new parent | |
4026 | * @new_dentry: where to create the new link | |
4027 | * @delegated_inode: returns inode needing a delegation break | |
4028 | * | |
4029 | * The caller must hold dir->i_mutex | |
4030 | * | |
4031 | * If vfs_link discovers a delegation on the to-be-linked file in need | |
4032 | * of breaking, it will return -EWOULDBLOCK and return a reference to the | |
4033 | * inode in delegated_inode. The caller should then break the delegation | |
4034 | * and retry. Because breaking a delegation may take a long time, the | |
4035 | * caller should drop the i_mutex before doing so. | |
4036 | * | |
4037 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4038 | * be appropriate for callers that expect the underlying filesystem not | |
4039 | * to be NFS exported. | |
4040 | */ | |
4041 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) | |
4042 | { | |
4043 | struct inode *inode = old_dentry->d_inode; | |
4044 | unsigned max_links = dir->i_sb->s_max_links; | |
4045 | int error; | |
4046 | ||
4047 | if (!inode) | |
4048 | return -ENOENT; | |
4049 | ||
4050 | error = may_create(dir, new_dentry); | |
4051 | if (error) | |
4052 | return error; | |
4053 | ||
4054 | if (dir->i_sb != inode->i_sb) | |
4055 | return -EXDEV; | |
4056 | ||
4057 | /* | |
4058 | * A link to an append-only or immutable file cannot be created. | |
4059 | */ | |
4060 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4061 | return -EPERM; | |
4062 | if (!dir->i_op->link) | |
4063 | return -EPERM; | |
4064 | if (S_ISDIR(inode->i_mode)) | |
4065 | return -EPERM; | |
4066 | ||
4067 | error = security_inode_link(old_dentry, dir, new_dentry); | |
4068 | if (error) | |
4069 | return error; | |
4070 | ||
4071 | inode_lock(inode); | |
4072 | /* Make sure we don't allow creating hardlink to an unlinked file */ | |
4073 | if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) | |
4074 | error = -ENOENT; | |
4075 | else if (max_links && inode->i_nlink >= max_links) | |
4076 | error = -EMLINK; | |
4077 | else { | |
4078 | error = try_break_deleg(inode, delegated_inode); | |
4079 | if (!error) | |
4080 | error = dir->i_op->link(old_dentry, dir, new_dentry); | |
4081 | } | |
4082 | ||
4083 | if (!error && (inode->i_state & I_LINKABLE)) { | |
4084 | spin_lock(&inode->i_lock); | |
4085 | inode->i_state &= ~I_LINKABLE; | |
4086 | spin_unlock(&inode->i_lock); | |
4087 | } | |
4088 | inode_unlock(inode); | |
4089 | if (!error) | |
4090 | fsnotify_link(dir, inode, new_dentry); | |
4091 | return error; | |
4092 | } | |
4093 | EXPORT_SYMBOL(vfs_link); | |
4094 | ||
4095 | /* | |
4096 | * Hardlinks are often used in delicate situations. We avoid | |
4097 | * security-related surprises by not following symlinks on the | |
4098 | * newname. --KAB | |
4099 | * | |
4100 | * We don't follow them on the oldname either to be compatible | |
4101 | * with linux 2.0, and to avoid hard-linking to directories | |
4102 | * and other special files. --ADM | |
4103 | */ | |
4104 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, | |
4105 | int, newdfd, const char __user *, newname, int, flags) | |
4106 | { | |
4107 | struct dentry *new_dentry; | |
4108 | struct path old_path, new_path; | |
4109 | struct inode *delegated_inode = NULL; | |
4110 | int how = 0; | |
4111 | int error; | |
4112 | ||
4113 | if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) | |
4114 | return -EINVAL; | |
4115 | /* | |
4116 | * To use null names we require CAP_DAC_READ_SEARCH | |
4117 | * This ensures that not everyone will be able to create | |
4118 | * handlink using the passed filedescriptor. | |
4119 | */ | |
4120 | if (flags & AT_EMPTY_PATH) { | |
4121 | if (!capable(CAP_DAC_READ_SEARCH)) | |
4122 | return -ENOENT; | |
4123 | how = LOOKUP_EMPTY; | |
4124 | } | |
4125 | ||
4126 | if (flags & AT_SYMLINK_FOLLOW) | |
4127 | how |= LOOKUP_FOLLOW; | |
4128 | retry: | |
4129 | error = user_path_at(olddfd, oldname, how, &old_path); | |
4130 | if (error) | |
4131 | return error; | |
4132 | ||
4133 | new_dentry = user_path_create(newdfd, newname, &new_path, | |
4134 | (how & LOOKUP_REVAL)); | |
4135 | error = PTR_ERR(new_dentry); | |
4136 | if (IS_ERR(new_dentry)) | |
4137 | goto out; | |
4138 | ||
4139 | error = -EXDEV; | |
4140 | if (old_path.mnt != new_path.mnt) | |
4141 | goto out_dput; | |
4142 | error = may_linkat(&old_path); | |
4143 | if (unlikely(error)) | |
4144 | goto out_dput; | |
4145 | error = security_path_link(old_path.dentry, &new_path, new_dentry); | |
4146 | if (error) | |
4147 | goto out_dput; | |
4148 | error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); | |
4149 | out_dput: | |
4150 | done_path_create(&new_path, new_dentry); | |
4151 | if (delegated_inode) { | |
4152 | error = break_deleg_wait(&delegated_inode); | |
4153 | if (!error) { | |
4154 | path_put(&old_path); | |
4155 | goto retry; | |
4156 | } | |
4157 | } | |
4158 | if (retry_estale(error, how)) { | |
4159 | path_put(&old_path); | |
4160 | how |= LOOKUP_REVAL; | |
4161 | goto retry; | |
4162 | } | |
4163 | out: | |
4164 | path_put(&old_path); | |
4165 | ||
4166 | return error; | |
4167 | } | |
4168 | ||
4169 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) | |
4170 | { | |
4171 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
4172 | } | |
4173 | ||
4174 | /** | |
4175 | * vfs_rename - rename a filesystem object | |
4176 | * @old_dir: parent of source | |
4177 | * @old_dentry: source | |
4178 | * @new_dir: parent of destination | |
4179 | * @new_dentry: destination | |
4180 | * @delegated_inode: returns an inode needing a delegation break | |
4181 | * @flags: rename flags | |
4182 | * | |
4183 | * The caller must hold multiple mutexes--see lock_rename()). | |
4184 | * | |
4185 | * If vfs_rename discovers a delegation in need of breaking at either | |
4186 | * the source or destination, it will return -EWOULDBLOCK and return a | |
4187 | * reference to the inode in delegated_inode. The caller should then | |
4188 | * break the delegation and retry. Because breaking a delegation may | |
4189 | * take a long time, the caller should drop all locks before doing | |
4190 | * so. | |
4191 | * | |
4192 | * Alternatively, a caller may pass NULL for delegated_inode. This may | |
4193 | * be appropriate for callers that expect the underlying filesystem not | |
4194 | * to be NFS exported. | |
4195 | * | |
4196 | * The worst of all namespace operations - renaming directory. "Perverted" | |
4197 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... | |
4198 | * Problems: | |
4199 | * a) we can get into loop creation. | |
4200 | * b) race potential - two innocent renames can create a loop together. | |
4201 | * That's where 4.4 screws up. Current fix: serialization on | |
4202 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another | |
4203 | * story. | |
4204 | * c) we have to lock _four_ objects - parents and victim (if it exists), | |
4205 | * and source (if it is not a directory). | |
4206 | * And that - after we got ->i_mutex on parents (until then we don't know | |
4207 | * whether the target exists). Solution: try to be smart with locking | |
4208 | * order for inodes. We rely on the fact that tree topology may change | |
4209 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we | |
4210 | * move will be locked. Thus we can rank directories by the tree | |
4211 | * (ancestors first) and rank all non-directories after them. | |
4212 | * That works since everybody except rename does "lock parent, lookup, | |
4213 | * lock child" and rename is under ->s_vfs_rename_mutex. | |
4214 | * HOWEVER, it relies on the assumption that any object with ->lookup() | |
4215 | * has no more than 1 dentry. If "hybrid" objects will ever appear, | |
4216 | * we'd better make sure that there's no link(2) for them. | |
4217 | * d) conversion from fhandle to dentry may come in the wrong moment - when | |
4218 | * we are removing the target. Solution: we will have to grab ->i_mutex | |
4219 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on | |
4220 | * ->i_mutex on parents, which works but leads to some truly excessive | |
4221 | * locking]. | |
4222 | */ | |
4223 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
4224 | struct inode *new_dir, struct dentry *new_dentry, | |
4225 | struct inode **delegated_inode, unsigned int flags) | |
4226 | { | |
4227 | int error; | |
4228 | bool is_dir = d_is_dir(old_dentry); | |
4229 | const unsigned char *old_name; | |
4230 | struct inode *source = old_dentry->d_inode; | |
4231 | struct inode *target = new_dentry->d_inode; | |
4232 | bool new_is_dir = false; | |
4233 | unsigned max_links = new_dir->i_sb->s_max_links; | |
4234 | ||
4235 | if (source == target) | |
4236 | return 0; | |
4237 | ||
4238 | error = may_delete(old_dir, old_dentry, is_dir); | |
4239 | if (error) | |
4240 | return error; | |
4241 | ||
4242 | if (!target) { | |
4243 | error = may_create(new_dir, new_dentry); | |
4244 | } else { | |
4245 | new_is_dir = d_is_dir(new_dentry); | |
4246 | ||
4247 | if (!(flags & RENAME_EXCHANGE)) | |
4248 | error = may_delete(new_dir, new_dentry, is_dir); | |
4249 | else | |
4250 | error = may_delete(new_dir, new_dentry, new_is_dir); | |
4251 | } | |
4252 | if (error) | |
4253 | return error; | |
4254 | ||
4255 | if (!old_dir->i_op->rename && !old_dir->i_op->rename2) | |
4256 | return -EPERM; | |
4257 | ||
4258 | if (flags && !old_dir->i_op->rename2) | |
4259 | return -EINVAL; | |
4260 | ||
4261 | /* | |
4262 | * If we are going to change the parent - check write permissions, | |
4263 | * we'll need to flip '..'. | |
4264 | */ | |
4265 | if (new_dir != old_dir) { | |
4266 | if (is_dir) { | |
4267 | error = inode_permission(source, MAY_WRITE); | |
4268 | if (error) | |
4269 | return error; | |
4270 | } | |
4271 | if ((flags & RENAME_EXCHANGE) && new_is_dir) { | |
4272 | error = inode_permission(target, MAY_WRITE); | |
4273 | if (error) | |
4274 | return error; | |
4275 | } | |
4276 | } | |
4277 | ||
4278 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, | |
4279 | flags); | |
4280 | if (error) | |
4281 | return error; | |
4282 | ||
4283 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); | |
4284 | dget(new_dentry); | |
4285 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4286 | lock_two_nondirectories(source, target); | |
4287 | else if (target) | |
4288 | inode_lock(target); | |
4289 | ||
4290 | error = -EBUSY; | |
4291 | if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) | |
4292 | goto out; | |
4293 | ||
4294 | if (max_links && new_dir != old_dir) { | |
4295 | error = -EMLINK; | |
4296 | if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) | |
4297 | goto out; | |
4298 | if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && | |
4299 | old_dir->i_nlink >= max_links) | |
4300 | goto out; | |
4301 | } | |
4302 | if (is_dir && !(flags & RENAME_EXCHANGE) && target) | |
4303 | shrink_dcache_parent(new_dentry); | |
4304 | if (!is_dir) { | |
4305 | error = try_break_deleg(source, delegated_inode); | |
4306 | if (error) | |
4307 | goto out; | |
4308 | } | |
4309 | if (target && !new_is_dir) { | |
4310 | error = try_break_deleg(target, delegated_inode); | |
4311 | if (error) | |
4312 | goto out; | |
4313 | } | |
4314 | if (!old_dir->i_op->rename2) { | |
4315 | error = old_dir->i_op->rename(old_dir, old_dentry, | |
4316 | new_dir, new_dentry); | |
4317 | } else { | |
4318 | WARN_ON(old_dir->i_op->rename != NULL); | |
4319 | error = old_dir->i_op->rename2(old_dir, old_dentry, | |
4320 | new_dir, new_dentry, flags); | |
4321 | } | |
4322 | if (error) | |
4323 | goto out; | |
4324 | ||
4325 | if (!(flags & RENAME_EXCHANGE) && target) { | |
4326 | if (is_dir) | |
4327 | target->i_flags |= S_DEAD; | |
4328 | dont_mount(new_dentry); | |
4329 | detach_mounts(new_dentry); | |
4330 | } | |
4331 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { | |
4332 | if (!(flags & RENAME_EXCHANGE)) | |
4333 | d_move(old_dentry, new_dentry); | |
4334 | else | |
4335 | d_exchange(old_dentry, new_dentry); | |
4336 | } | |
4337 | out: | |
4338 | if (!is_dir || (flags & RENAME_EXCHANGE)) | |
4339 | unlock_two_nondirectories(source, target); | |
4340 | else if (target) | |
4341 | inode_unlock(target); | |
4342 | dput(new_dentry); | |
4343 | if (!error) { | |
4344 | fsnotify_move(old_dir, new_dir, old_name, is_dir, | |
4345 | !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); | |
4346 | if (flags & RENAME_EXCHANGE) { | |
4347 | fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, | |
4348 | new_is_dir, NULL, new_dentry); | |
4349 | } | |
4350 | } | |
4351 | fsnotify_oldname_free(old_name); | |
4352 | ||
4353 | return error; | |
4354 | } | |
4355 | EXPORT_SYMBOL(vfs_rename); | |
4356 | ||
4357 | SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, | |
4358 | int, newdfd, const char __user *, newname, unsigned int, flags) | |
4359 | { | |
4360 | struct dentry *old_dentry, *new_dentry; | |
4361 | struct dentry *trap; | |
4362 | struct path old_path, new_path; | |
4363 | struct qstr old_last, new_last; | |
4364 | int old_type, new_type; | |
4365 | struct inode *delegated_inode = NULL; | |
4366 | struct filename *from; | |
4367 | struct filename *to; | |
4368 | unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; | |
4369 | bool should_retry = false; | |
4370 | int error; | |
4371 | ||
4372 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) | |
4373 | return -EINVAL; | |
4374 | ||
4375 | if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && | |
4376 | (flags & RENAME_EXCHANGE)) | |
4377 | return -EINVAL; | |
4378 | ||
4379 | if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) | |
4380 | return -EPERM; | |
4381 | ||
4382 | if (flags & RENAME_EXCHANGE) | |
4383 | target_flags = 0; | |
4384 | ||
4385 | retry: | |
4386 | from = user_path_parent(olddfd, oldname, | |
4387 | &old_path, &old_last, &old_type, lookup_flags); | |
4388 | if (IS_ERR(from)) { | |
4389 | error = PTR_ERR(from); | |
4390 | goto exit; | |
4391 | } | |
4392 | ||
4393 | to = user_path_parent(newdfd, newname, | |
4394 | &new_path, &new_last, &new_type, lookup_flags); | |
4395 | if (IS_ERR(to)) { | |
4396 | error = PTR_ERR(to); | |
4397 | goto exit1; | |
4398 | } | |
4399 | ||
4400 | error = -EXDEV; | |
4401 | if (old_path.mnt != new_path.mnt) | |
4402 | goto exit2; | |
4403 | ||
4404 | error = -EBUSY; | |
4405 | if (old_type != LAST_NORM) | |
4406 | goto exit2; | |
4407 | ||
4408 | if (flags & RENAME_NOREPLACE) | |
4409 | error = -EEXIST; | |
4410 | if (new_type != LAST_NORM) | |
4411 | goto exit2; | |
4412 | ||
4413 | error = mnt_want_write(old_path.mnt); | |
4414 | if (error) | |
4415 | goto exit2; | |
4416 | ||
4417 | retry_deleg: | |
4418 | trap = lock_rename(new_path.dentry, old_path.dentry); | |
4419 | ||
4420 | old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); | |
4421 | error = PTR_ERR(old_dentry); | |
4422 | if (IS_ERR(old_dentry)) | |
4423 | goto exit3; | |
4424 | /* source must exist */ | |
4425 | error = -ENOENT; | |
4426 | if (d_is_negative(old_dentry)) | |
4427 | goto exit4; | |
4428 | new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); | |
4429 | error = PTR_ERR(new_dentry); | |
4430 | if (IS_ERR(new_dentry)) | |
4431 | goto exit4; | |
4432 | error = -EEXIST; | |
4433 | if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) | |
4434 | goto exit5; | |
4435 | if (flags & RENAME_EXCHANGE) { | |
4436 | error = -ENOENT; | |
4437 | if (d_is_negative(new_dentry)) | |
4438 | goto exit5; | |
4439 | ||
4440 | if (!d_is_dir(new_dentry)) { | |
4441 | error = -ENOTDIR; | |
4442 | if (new_last.name[new_last.len]) | |
4443 | goto exit5; | |
4444 | } | |
4445 | } | |
4446 | /* unless the source is a directory trailing slashes give -ENOTDIR */ | |
4447 | if (!d_is_dir(old_dentry)) { | |
4448 | error = -ENOTDIR; | |
4449 | if (old_last.name[old_last.len]) | |
4450 | goto exit5; | |
4451 | if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) | |
4452 | goto exit5; | |
4453 | } | |
4454 | /* source should not be ancestor of target */ | |
4455 | error = -EINVAL; | |
4456 | if (old_dentry == trap) | |
4457 | goto exit5; | |
4458 | /* target should not be an ancestor of source */ | |
4459 | if (!(flags & RENAME_EXCHANGE)) | |
4460 | error = -ENOTEMPTY; | |
4461 | if (new_dentry == trap) | |
4462 | goto exit5; | |
4463 | ||
4464 | error = security_path_rename(&old_path, old_dentry, | |
4465 | &new_path, new_dentry, flags); | |
4466 | if (error) | |
4467 | goto exit5; | |
4468 | error = vfs_rename(old_path.dentry->d_inode, old_dentry, | |
4469 | new_path.dentry->d_inode, new_dentry, | |
4470 | &delegated_inode, flags); | |
4471 | exit5: | |
4472 | dput(new_dentry); | |
4473 | exit4: | |
4474 | dput(old_dentry); | |
4475 | exit3: | |
4476 | unlock_rename(new_path.dentry, old_path.dentry); | |
4477 | if (delegated_inode) { | |
4478 | error = break_deleg_wait(&delegated_inode); | |
4479 | if (!error) | |
4480 | goto retry_deleg; | |
4481 | } | |
4482 | mnt_drop_write(old_path.mnt); | |
4483 | exit2: | |
4484 | if (retry_estale(error, lookup_flags)) | |
4485 | should_retry = true; | |
4486 | path_put(&new_path); | |
4487 | putname(to); | |
4488 | exit1: | |
4489 | path_put(&old_path); | |
4490 | putname(from); | |
4491 | if (should_retry) { | |
4492 | should_retry = false; | |
4493 | lookup_flags |= LOOKUP_REVAL; | |
4494 | goto retry; | |
4495 | } | |
4496 | exit: | |
4497 | return error; | |
4498 | } | |
4499 | ||
4500 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, | |
4501 | int, newdfd, const char __user *, newname) | |
4502 | { | |
4503 | return sys_renameat2(olddfd, oldname, newdfd, newname, 0); | |
4504 | } | |
4505 | ||
4506 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) | |
4507 | { | |
4508 | return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); | |
4509 | } | |
4510 | ||
4511 | int vfs_whiteout(struct inode *dir, struct dentry *dentry) | |
4512 | { | |
4513 | int error = may_create(dir, dentry); | |
4514 | if (error) | |
4515 | return error; | |
4516 | ||
4517 | if (!dir->i_op->mknod) | |
4518 | return -EPERM; | |
4519 | ||
4520 | return dir->i_op->mknod(dir, dentry, | |
4521 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
4522 | } | |
4523 | EXPORT_SYMBOL(vfs_whiteout); | |
4524 | ||
4525 | int readlink_copy(char __user *buffer, int buflen, const char *link) | |
4526 | { | |
4527 | int len = PTR_ERR(link); | |
4528 | if (IS_ERR(link)) | |
4529 | goto out; | |
4530 | ||
4531 | len = strlen(link); | |
4532 | if (len > (unsigned) buflen) | |
4533 | len = buflen; | |
4534 | if (copy_to_user(buffer, link, len)) | |
4535 | len = -EFAULT; | |
4536 | out: | |
4537 | return len; | |
4538 | } | |
4539 | EXPORT_SYMBOL(readlink_copy); | |
4540 | ||
4541 | /* | |
4542 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that | |
4543 | * have ->get_link() not calling nd_jump_link(). Using (or not using) it | |
4544 | * for any given inode is up to filesystem. | |
4545 | */ | |
4546 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4547 | { | |
4548 | DEFINE_DELAYED_CALL(done); | |
4549 | struct inode *inode = d_inode(dentry); | |
4550 | const char *link = inode->i_link; | |
4551 | int res; | |
4552 | ||
4553 | if (!link) { | |
4554 | link = inode->i_op->get_link(dentry, inode, &done); | |
4555 | if (IS_ERR(link)) | |
4556 | return PTR_ERR(link); | |
4557 | } | |
4558 | res = readlink_copy(buffer, buflen, link); | |
4559 | do_delayed_call(&done); | |
4560 | return res; | |
4561 | } | |
4562 | EXPORT_SYMBOL(generic_readlink); | |
4563 | ||
4564 | /* get the link contents into pagecache */ | |
4565 | const char *page_get_link(struct dentry *dentry, struct inode *inode, | |
4566 | struct delayed_call *callback) | |
4567 | { | |
4568 | char *kaddr; | |
4569 | struct page *page; | |
4570 | struct address_space *mapping = inode->i_mapping; | |
4571 | ||
4572 | if (!dentry) { | |
4573 | page = find_get_page(mapping, 0); | |
4574 | if (!page) | |
4575 | return ERR_PTR(-ECHILD); | |
4576 | if (!PageUptodate(page)) { | |
4577 | put_page(page); | |
4578 | return ERR_PTR(-ECHILD); | |
4579 | } | |
4580 | } else { | |
4581 | page = read_mapping_page(mapping, 0, NULL); | |
4582 | if (IS_ERR(page)) | |
4583 | return (char*)page; | |
4584 | } | |
4585 | set_delayed_call(callback, page_put_link, page); | |
4586 | BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); | |
4587 | kaddr = page_address(page); | |
4588 | nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); | |
4589 | return kaddr; | |
4590 | } | |
4591 | ||
4592 | EXPORT_SYMBOL(page_get_link); | |
4593 | ||
4594 | void page_put_link(void *arg) | |
4595 | { | |
4596 | put_page(arg); | |
4597 | } | |
4598 | EXPORT_SYMBOL(page_put_link); | |
4599 | ||
4600 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) | |
4601 | { | |
4602 | DEFINE_DELAYED_CALL(done); | |
4603 | int res = readlink_copy(buffer, buflen, | |
4604 | page_get_link(dentry, d_inode(dentry), | |
4605 | &done)); | |
4606 | do_delayed_call(&done); | |
4607 | return res; | |
4608 | } | |
4609 | EXPORT_SYMBOL(page_readlink); | |
4610 | ||
4611 | /* | |
4612 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS | |
4613 | */ | |
4614 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) | |
4615 | { | |
4616 | struct address_space *mapping = inode->i_mapping; | |
4617 | struct page *page; | |
4618 | void *fsdata; | |
4619 | int err; | |
4620 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; | |
4621 | if (nofs) | |
4622 | flags |= AOP_FLAG_NOFS; | |
4623 | ||
4624 | retry: | |
4625 | err = pagecache_write_begin(NULL, mapping, 0, len-1, | |
4626 | flags, &page, &fsdata); | |
4627 | if (err) | |
4628 | goto fail; | |
4629 | ||
4630 | memcpy(page_address(page), symname, len-1); | |
4631 | ||
4632 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, | |
4633 | page, fsdata); | |
4634 | if (err < 0) | |
4635 | goto fail; | |
4636 | if (err < len-1) | |
4637 | goto retry; | |
4638 | ||
4639 | mark_inode_dirty(inode); | |
4640 | return 0; | |
4641 | fail: | |
4642 | return err; | |
4643 | } | |
4644 | EXPORT_SYMBOL(__page_symlink); | |
4645 | ||
4646 | int page_symlink(struct inode *inode, const char *symname, int len) | |
4647 | { | |
4648 | return __page_symlink(inode, symname, len, | |
4649 | !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); | |
4650 | } | |
4651 | EXPORT_SYMBOL(page_symlink); | |
4652 | ||
4653 | const struct inode_operations page_symlink_inode_operations = { | |
4654 | .readlink = generic_readlink, | |
4655 | .get_link = page_get_link, | |
4656 | }; | |
4657 | EXPORT_SYMBOL(page_symlink_inode_operations); |