Root/
1 | /* |
2 | * linux/fs/namei.c |
3 | * |
4 | * Copyright (C) 1991, 1992 Linus Torvalds |
5 | */ |
6 | |
7 | /* |
8 | * Some corrections by tytso. |
9 | */ |
10 | |
11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname |
12 | * lookup logic. |
13 | */ |
14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. |
15 | */ |
16 | |
17 | #include <linux/init.h> |
18 | #include <linux/module.h> |
19 | #include <linux/slab.h> |
20 | #include <linux/fs.h> |
21 | #include <linux/namei.h> |
22 | #include <linux/pagemap.h> |
23 | #include <linux/fsnotify.h> |
24 | #include <linux/personality.h> |
25 | #include <linux/security.h> |
26 | #include <linux/ima.h> |
27 | #include <linux/syscalls.h> |
28 | #include <linux/mount.h> |
29 | #include <linux/audit.h> |
30 | #include <linux/capability.h> |
31 | #include <linux/file.h> |
32 | #include <linux/fcntl.h> |
33 | #include <linux/device_cgroup.h> |
34 | #include <linux/fs_struct.h> |
35 | #include <asm/uaccess.h> |
36 | |
37 | #include "internal.h" |
38 | |
39 | /* [Feb-1997 T. Schoebel-Theuer] |
40 | * Fundamental changes in the pathname lookup mechanisms (namei) |
41 | * were necessary because of omirr. The reason is that omirr needs |
42 | * to know the _real_ pathname, not the user-supplied one, in case |
43 | * of symlinks (and also when transname replacements occur). |
44 | * |
45 | * The new code replaces the old recursive symlink resolution with |
46 | * an iterative one (in case of non-nested symlink chains). It does |
47 | * this with calls to <fs>_follow_link(). |
48 | * As a side effect, dir_namei(), _namei() and follow_link() are now |
49 | * replaced with a single function lookup_dentry() that can handle all |
50 | * the special cases of the former code. |
51 | * |
52 | * With the new dcache, the pathname is stored at each inode, at least as |
53 | * long as the refcount of the inode is positive. As a side effect, the |
54 | * size of the dcache depends on the inode cache and thus is dynamic. |
55 | * |
56 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink |
57 | * resolution to correspond with current state of the code. |
58 | * |
59 | * Note that the symlink resolution is not *completely* iterative. |
60 | * There is still a significant amount of tail- and mid- recursion in |
61 | * the algorithm. Also, note that <fs>_readlink() is not used in |
62 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() |
63 | * may return different results than <fs>_follow_link(). Many virtual |
64 | * filesystems (including /proc) exhibit this behavior. |
65 | */ |
66 | |
67 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: |
68 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL |
69 | * and the name already exists in form of a symlink, try to create the new |
70 | * name indicated by the symlink. The old code always complained that the |
71 | * name already exists, due to not following the symlink even if its target |
72 | * is nonexistent. The new semantics affects also mknod() and link() when |
73 | * the name is a symlink pointing to a non-existant name. |
74 | * |
75 | * I don't know which semantics is the right one, since I have no access |
76 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" |
77 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the |
78 | * "old" one. Personally, I think the new semantics is much more logical. |
79 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing |
80 | * file does succeed in both HP-UX and SunOs, but not in Solaris |
81 | * and in the old Linux semantics. |
82 | */ |
83 | |
84 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink |
85 | * semantics. See the comments in "open_namei" and "do_link" below. |
86 | * |
87 | * [10-Sep-98 Alan Modra] Another symlink change. |
88 | */ |
89 | |
90 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: |
91 | * inside the path - always follow. |
92 | * in the last component in creation/removal/renaming - never follow. |
93 | * if LOOKUP_FOLLOW passed - follow. |
94 | * if the pathname has trailing slashes - follow. |
95 | * otherwise - don't follow. |
96 | * (applied in that order). |
97 | * |
98 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT |
99 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. |
100 | * During the 2.4 we need to fix the userland stuff depending on it - |
101 | * hopefully we will be able to get rid of that wart in 2.5. So far only |
102 | * XEmacs seems to be relying on it... |
103 | */ |
104 | /* |
105 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) |
106 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
107 | * any extra contention... |
108 | */ |
109 | |
110 | /* In order to reduce some races, while at the same time doing additional |
111 | * checking and hopefully speeding things up, we copy filenames to the |
112 | * kernel data space before using them.. |
113 | * |
114 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). |
115 | * PATH_MAX includes the nul terminator --RR. |
116 | */ |
117 | static int do_getname(const char __user *filename, char *page) |
118 | { |
119 | int retval; |
120 | unsigned long len = PATH_MAX; |
121 | |
122 | if (!segment_eq(get_fs(), KERNEL_DS)) { |
123 | if ((unsigned long) filename >= TASK_SIZE) |
124 | return -EFAULT; |
125 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) |
126 | len = TASK_SIZE - (unsigned long) filename; |
127 | } |
128 | |
129 | retval = strncpy_from_user(page, filename, len); |
130 | if (retval > 0) { |
131 | if (retval < len) |
132 | return 0; |
133 | return -ENAMETOOLONG; |
134 | } else if (!retval) |
135 | retval = -ENOENT; |
136 | return retval; |
137 | } |
138 | |
139 | char * getname(const char __user * filename) |
140 | { |
141 | char *tmp, *result; |
142 | |
143 | result = ERR_PTR(-ENOMEM); |
144 | tmp = __getname(); |
145 | if (tmp) { |
146 | int retval = do_getname(filename, tmp); |
147 | |
148 | result = tmp; |
149 | if (retval < 0) { |
150 | __putname(tmp); |
151 | result = ERR_PTR(retval); |
152 | } |
153 | } |
154 | audit_getname(result); |
155 | return result; |
156 | } |
157 | |
158 | #ifdef CONFIG_AUDITSYSCALL |
159 | void putname(const char *name) |
160 | { |
161 | if (unlikely(!audit_dummy_context())) |
162 | audit_putname(name); |
163 | else |
164 | __putname(name); |
165 | } |
166 | EXPORT_SYMBOL(putname); |
167 | #endif |
168 | |
169 | /* |
170 | * This does basic POSIX ACL permission checking |
171 | */ |
172 | static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, |
173 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) |
174 | { |
175 | umode_t mode = inode->i_mode; |
176 | |
177 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
178 | |
179 | if (current_fsuid() == inode->i_uid) |
180 | mode >>= 6; |
181 | else { |
182 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { |
183 | int error = check_acl(inode, mask, flags); |
184 | if (error != -EAGAIN) |
185 | return error; |
186 | } |
187 | |
188 | if (in_group_p(inode->i_gid)) |
189 | mode >>= 3; |
190 | } |
191 | |
192 | /* |
193 | * If the DACs are ok we don't need any capability check. |
194 | */ |
195 | if ((mask & ~mode) == 0) |
196 | return 0; |
197 | return -EACCES; |
198 | } |
199 | |
200 | /** |
201 | * generic_permission - check for access rights on a Posix-like filesystem |
202 | * @inode: inode to check access rights for |
203 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
204 | * @check_acl: optional callback to check for Posix ACLs |
205 | * @flags: IPERM_FLAG_ flags. |
206 | * |
207 | * Used to check for read/write/execute permissions on a file. |
208 | * We use "fsuid" for this, letting us set arbitrary permissions |
209 | * for filesystem access without changing the "normal" uids which |
210 | * are used for other things. |
211 | * |
212 | * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk |
213 | * request cannot be satisfied (eg. requires blocking or too much complexity). |
214 | * It would then be called again in ref-walk mode. |
215 | */ |
216 | int generic_permission(struct inode *inode, int mask, unsigned int flags, |
217 | int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) |
218 | { |
219 | int ret; |
220 | |
221 | /* |
222 | * Do the basic POSIX ACL permission checks. |
223 | */ |
224 | ret = acl_permission_check(inode, mask, flags, check_acl); |
225 | if (ret != -EACCES) |
226 | return ret; |
227 | |
228 | /* |
229 | * Read/write DACs are always overridable. |
230 | * Executable DACs are overridable if at least one exec bit is set. |
231 | */ |
232 | if (!(mask & MAY_EXEC) || execute_ok(inode)) |
233 | if (capable(CAP_DAC_OVERRIDE)) |
234 | return 0; |
235 | |
236 | /* |
237 | * Searching includes executable on directories, else just read. |
238 | */ |
239 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
240 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) |
241 | if (capable(CAP_DAC_READ_SEARCH)) |
242 | return 0; |
243 | |
244 | return -EACCES; |
245 | } |
246 | |
247 | /** |
248 | * inode_permission - check for access rights to a given inode |
249 | * @inode: inode to check permission on |
250 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
251 | * |
252 | * Used to check for read/write/execute permissions on an inode. |
253 | * We use "fsuid" for this, letting us set arbitrary permissions |
254 | * for filesystem access without changing the "normal" uids which |
255 | * are used for other things. |
256 | */ |
257 | int inode_permission(struct inode *inode, int mask) |
258 | { |
259 | int retval; |
260 | |
261 | if (mask & MAY_WRITE) { |
262 | umode_t mode = inode->i_mode; |
263 | |
264 | /* |
265 | * Nobody gets write access to a read-only fs. |
266 | */ |
267 | if (IS_RDONLY(inode) && |
268 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
269 | return -EROFS; |
270 | |
271 | /* |
272 | * Nobody gets write access to an immutable file. |
273 | */ |
274 | if (IS_IMMUTABLE(inode)) |
275 | return -EACCES; |
276 | } |
277 | |
278 | if (inode->i_op->permission) |
279 | retval = inode->i_op->permission(inode, mask, 0); |
280 | else |
281 | retval = generic_permission(inode, mask, 0, |
282 | inode->i_op->check_acl); |
283 | |
284 | if (retval) |
285 | return retval; |
286 | |
287 | retval = devcgroup_inode_permission(inode, mask); |
288 | if (retval) |
289 | return retval; |
290 | |
291 | return security_inode_permission(inode, mask); |
292 | } |
293 | |
294 | /** |
295 | * file_permission - check for additional access rights to a given file |
296 | * @file: file to check access rights for |
297 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
298 | * |
299 | * Used to check for read/write/execute permissions on an already opened |
300 | * file. |
301 | * |
302 | * Note: |
303 | * Do not use this function in new code. All access checks should |
304 | * be done using inode_permission(). |
305 | */ |
306 | int file_permission(struct file *file, int mask) |
307 | { |
308 | return inode_permission(file->f_path.dentry->d_inode, mask); |
309 | } |
310 | |
311 | /* |
312 | * get_write_access() gets write permission for a file. |
313 | * put_write_access() releases this write permission. |
314 | * This is used for regular files. |
315 | * We cannot support write (and maybe mmap read-write shared) accesses and |
316 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode |
317 | * can have the following values: |
318 | * 0: no writers, no VM_DENYWRITE mappings |
319 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist |
320 | * > 0: (i_writecount) users are writing to the file. |
321 | * |
322 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe |
323 | * except for the cases where we don't hold i_writecount yet. Then we need to |
324 | * use {get,deny}_write_access() - these functions check the sign and refuse |
325 | * to do the change if sign is wrong. Exclusion between them is provided by |
326 | * the inode->i_lock spinlock. |
327 | */ |
328 | |
329 | int get_write_access(struct inode * inode) |
330 | { |
331 | spin_lock(&inode->i_lock); |
332 | if (atomic_read(&inode->i_writecount) < 0) { |
333 | spin_unlock(&inode->i_lock); |
334 | return -ETXTBSY; |
335 | } |
336 | atomic_inc(&inode->i_writecount); |
337 | spin_unlock(&inode->i_lock); |
338 | |
339 | return 0; |
340 | } |
341 | |
342 | int deny_write_access(struct file * file) |
343 | { |
344 | struct inode *inode = file->f_path.dentry->d_inode; |
345 | |
346 | spin_lock(&inode->i_lock); |
347 | if (atomic_read(&inode->i_writecount) > 0) { |
348 | spin_unlock(&inode->i_lock); |
349 | return -ETXTBSY; |
350 | } |
351 | atomic_dec(&inode->i_writecount); |
352 | spin_unlock(&inode->i_lock); |
353 | |
354 | return 0; |
355 | } |
356 | |
357 | /** |
358 | * path_get - get a reference to a path |
359 | * @path: path to get the reference to |
360 | * |
361 | * Given a path increment the reference count to the dentry and the vfsmount. |
362 | */ |
363 | void path_get(struct path *path) |
364 | { |
365 | mntget(path->mnt); |
366 | dget(path->dentry); |
367 | } |
368 | EXPORT_SYMBOL(path_get); |
369 | |
370 | /** |
371 | * path_put - put a reference to a path |
372 | * @path: path to put the reference to |
373 | * |
374 | * Given a path decrement the reference count to the dentry and the vfsmount. |
375 | */ |
376 | void path_put(struct path *path) |
377 | { |
378 | dput(path->dentry); |
379 | mntput(path->mnt); |
380 | } |
381 | EXPORT_SYMBOL(path_put); |
382 | |
383 | /** |
384 | * nameidata_drop_rcu - drop this nameidata out of rcu-walk |
385 | * @nd: nameidata pathwalk data to drop |
386 | * Returns: 0 on success, -ECHILD on failure |
387 | * |
388 | * Path walking has 2 modes, rcu-walk and ref-walk (see |
389 | * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt |
390 | * to drop out of rcu-walk mode and take normal reference counts on dentries |
391 | * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take |
392 | * refcounts at the last known good point before rcu-walk got stuck, so |
393 | * ref-walk may continue from there. If this is not successful (eg. a seqcount |
394 | * has changed), then failure is returned and path walk restarts from the |
395 | * beginning in ref-walk mode. |
396 | * |
397 | * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into |
398 | * ref-walk. Must be called from rcu-walk context. |
399 | */ |
400 | static int nameidata_drop_rcu(struct nameidata *nd) |
401 | { |
402 | struct fs_struct *fs = current->fs; |
403 | struct dentry *dentry = nd->path.dentry; |
404 | |
405 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
406 | if (nd->root.mnt) { |
407 | spin_lock(&fs->lock); |
408 | if (nd->root.mnt != fs->root.mnt || |
409 | nd->root.dentry != fs->root.dentry) |
410 | goto err_root; |
411 | } |
412 | spin_lock(&dentry->d_lock); |
413 | if (!__d_rcu_to_refcount(dentry, nd->seq)) |
414 | goto err; |
415 | BUG_ON(nd->inode != dentry->d_inode); |
416 | spin_unlock(&dentry->d_lock); |
417 | if (nd->root.mnt) { |
418 | path_get(&nd->root); |
419 | spin_unlock(&fs->lock); |
420 | } |
421 | mntget(nd->path.mnt); |
422 | |
423 | rcu_read_unlock(); |
424 | br_read_unlock(vfsmount_lock); |
425 | nd->flags &= ~LOOKUP_RCU; |
426 | return 0; |
427 | err: |
428 | spin_unlock(&dentry->d_lock); |
429 | err_root: |
430 | if (nd->root.mnt) |
431 | spin_unlock(&fs->lock); |
432 | return -ECHILD; |
433 | } |
434 | |
435 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ |
436 | static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) |
437 | { |
438 | if (nd->flags & LOOKUP_RCU) |
439 | return nameidata_drop_rcu(nd); |
440 | return 0; |
441 | } |
442 | |
443 | /** |
444 | * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk |
445 | * @nd: nameidata pathwalk data to drop |
446 | * @dentry: dentry to drop |
447 | * Returns: 0 on success, -ECHILD on failure |
448 | * |
449 | * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, |
450 | * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on |
451 | * @nd. Must be called from rcu-walk context. |
452 | */ |
453 | static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) |
454 | { |
455 | struct fs_struct *fs = current->fs; |
456 | struct dentry *parent = nd->path.dentry; |
457 | |
458 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
459 | if (nd->root.mnt) { |
460 | spin_lock(&fs->lock); |
461 | if (nd->root.mnt != fs->root.mnt || |
462 | nd->root.dentry != fs->root.dentry) |
463 | goto err_root; |
464 | } |
465 | spin_lock(&parent->d_lock); |
466 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
467 | if (!__d_rcu_to_refcount(dentry, nd->seq)) |
468 | goto err; |
469 | /* |
470 | * If the sequence check on the child dentry passed, then the child has |
471 | * not been removed from its parent. This means the parent dentry must |
472 | * be valid and able to take a reference at this point. |
473 | */ |
474 | BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); |
475 | BUG_ON(!parent->d_count); |
476 | parent->d_count++; |
477 | spin_unlock(&dentry->d_lock); |
478 | spin_unlock(&parent->d_lock); |
479 | if (nd->root.mnt) { |
480 | path_get(&nd->root); |
481 | spin_unlock(&fs->lock); |
482 | } |
483 | mntget(nd->path.mnt); |
484 | |
485 | rcu_read_unlock(); |
486 | br_read_unlock(vfsmount_lock); |
487 | nd->flags &= ~LOOKUP_RCU; |
488 | return 0; |
489 | err: |
490 | spin_unlock(&dentry->d_lock); |
491 | spin_unlock(&parent->d_lock); |
492 | err_root: |
493 | if (nd->root.mnt) |
494 | spin_unlock(&fs->lock); |
495 | return -ECHILD; |
496 | } |
497 | |
498 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ |
499 | static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) |
500 | { |
501 | if (nd->flags & LOOKUP_RCU) |
502 | return nameidata_dentry_drop_rcu(nd, dentry); |
503 | return 0; |
504 | } |
505 | |
506 | /** |
507 | * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk |
508 | * @nd: nameidata pathwalk data to drop |
509 | * Returns: 0 on success, -ECHILD on failure |
510 | * |
511 | * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. |
512 | * nd->path should be the final element of the lookup, so nd->root is discarded. |
513 | * Must be called from rcu-walk context. |
514 | */ |
515 | static int nameidata_drop_rcu_last(struct nameidata *nd) |
516 | { |
517 | struct dentry *dentry = nd->path.dentry; |
518 | |
519 | BUG_ON(!(nd->flags & LOOKUP_RCU)); |
520 | nd->flags &= ~LOOKUP_RCU; |
521 | nd->root.mnt = NULL; |
522 | spin_lock(&dentry->d_lock); |
523 | if (!__d_rcu_to_refcount(dentry, nd->seq)) |
524 | goto err_unlock; |
525 | BUG_ON(nd->inode != dentry->d_inode); |
526 | spin_unlock(&dentry->d_lock); |
527 | |
528 | mntget(nd->path.mnt); |
529 | |
530 | rcu_read_unlock(); |
531 | br_read_unlock(vfsmount_lock); |
532 | |
533 | return 0; |
534 | |
535 | err_unlock: |
536 | spin_unlock(&dentry->d_lock); |
537 | rcu_read_unlock(); |
538 | br_read_unlock(vfsmount_lock); |
539 | return -ECHILD; |
540 | } |
541 | |
542 | /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ |
543 | static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd) |
544 | { |
545 | if (likely(nd->flags & LOOKUP_RCU)) |
546 | return nameidata_drop_rcu_last(nd); |
547 | return 0; |
548 | } |
549 | |
550 | /** |
551 | * release_open_intent - free up open intent resources |
552 | * @nd: pointer to nameidata |
553 | */ |
554 | void release_open_intent(struct nameidata *nd) |
555 | { |
556 | struct file *file = nd->intent.open.file; |
557 | |
558 | if (file && !IS_ERR(file)) { |
559 | if (file->f_path.dentry == NULL) |
560 | put_filp(file); |
561 | else |
562 | fput(file); |
563 | } |
564 | } |
565 | |
566 | static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) |
567 | { |
568 | return dentry->d_op->d_revalidate(dentry, nd); |
569 | } |
570 | |
571 | static struct dentry * |
572 | do_revalidate(struct dentry *dentry, struct nameidata *nd) |
573 | { |
574 | int status = d_revalidate(dentry, nd); |
575 | if (unlikely(status <= 0)) { |
576 | /* |
577 | * The dentry failed validation. |
578 | * If d_revalidate returned 0 attempt to invalidate |
579 | * the dentry otherwise d_revalidate is asking us |
580 | * to return a fail status. |
581 | */ |
582 | if (status < 0) { |
583 | dput(dentry); |
584 | dentry = ERR_PTR(status); |
585 | } else if (!d_invalidate(dentry)) { |
586 | dput(dentry); |
587 | dentry = NULL; |
588 | } |
589 | } |
590 | return dentry; |
591 | } |
592 | |
593 | static inline struct dentry * |
594 | do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd) |
595 | { |
596 | int status = d_revalidate(dentry, nd); |
597 | if (likely(status > 0)) |
598 | return dentry; |
599 | if (status == -ECHILD) { |
600 | if (nameidata_dentry_drop_rcu(nd, dentry)) |
601 | return ERR_PTR(-ECHILD); |
602 | return do_revalidate(dentry, nd); |
603 | } |
604 | if (status < 0) |
605 | return ERR_PTR(status); |
606 | /* Don't d_invalidate in rcu-walk mode */ |
607 | if (nameidata_dentry_drop_rcu(nd, dentry)) |
608 | return ERR_PTR(-ECHILD); |
609 | if (!d_invalidate(dentry)) { |
610 | dput(dentry); |
611 | dentry = NULL; |
612 | } |
613 | return dentry; |
614 | } |
615 | |
616 | static inline int need_reval_dot(struct dentry *dentry) |
617 | { |
618 | if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) |
619 | return 0; |
620 | |
621 | if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) |
622 | return 0; |
623 | |
624 | return 1; |
625 | } |
626 | |
627 | /* |
628 | * force_reval_path - force revalidation of a dentry |
629 | * |
630 | * In some situations the path walking code will trust dentries without |
631 | * revalidating them. This causes problems for filesystems that depend on |
632 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set |
633 | * (which indicates that it's possible for the dentry to go stale), force |
634 | * a d_revalidate call before proceeding. |
635 | * |
636 | * Returns 0 if the revalidation was successful. If the revalidation fails, |
637 | * either return the error returned by d_revalidate or -ESTALE if the |
638 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to |
639 | * invalidate the dentry. It's up to the caller to handle putting references |
640 | * to the path if necessary. |
641 | */ |
642 | static int |
643 | force_reval_path(struct path *path, struct nameidata *nd) |
644 | { |
645 | int status; |
646 | struct dentry *dentry = path->dentry; |
647 | |
648 | /* |
649 | * only check on filesystems where it's possible for the dentry to |
650 | * become stale. |
651 | */ |
652 | if (!need_reval_dot(dentry)) |
653 | return 0; |
654 | |
655 | status = d_revalidate(dentry, nd); |
656 | if (status > 0) |
657 | return 0; |
658 | |
659 | if (!status) { |
660 | d_invalidate(dentry); |
661 | status = -ESTALE; |
662 | } |
663 | return status; |
664 | } |
665 | |
666 | /* |
667 | * Short-cut version of permission(), for calling on directories |
668 | * during pathname resolution. Combines parts of permission() |
669 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. |
670 | * |
671 | * If appropriate, check DAC only. If not appropriate, or |
672 | * short-cut DAC fails, then call ->permission() to do more |
673 | * complete permission check. |
674 | */ |
675 | static inline int exec_permission(struct inode *inode, unsigned int flags) |
676 | { |
677 | int ret; |
678 | |
679 | if (inode->i_op->permission) { |
680 | ret = inode->i_op->permission(inode, MAY_EXEC, flags); |
681 | } else { |
682 | ret = acl_permission_check(inode, MAY_EXEC, flags, |
683 | inode->i_op->check_acl); |
684 | } |
685 | if (likely(!ret)) |
686 | goto ok; |
687 | if (ret == -ECHILD) |
688 | return ret; |
689 | |
690 | if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) |
691 | goto ok; |
692 | |
693 | return ret; |
694 | ok: |
695 | return security_inode_exec_permission(inode, flags); |
696 | } |
697 | |
698 | static __always_inline void set_root(struct nameidata *nd) |
699 | { |
700 | if (!nd->root.mnt) |
701 | get_fs_root(current->fs, &nd->root); |
702 | } |
703 | |
704 | static int link_path_walk(const char *, struct nameidata *); |
705 | |
706 | static __always_inline void set_root_rcu(struct nameidata *nd) |
707 | { |
708 | if (!nd->root.mnt) { |
709 | struct fs_struct *fs = current->fs; |
710 | unsigned seq; |
711 | |
712 | do { |
713 | seq = read_seqcount_begin(&fs->seq); |
714 | nd->root = fs->root; |
715 | } while (read_seqcount_retry(&fs->seq, seq)); |
716 | } |
717 | } |
718 | |
719 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
720 | { |
721 | int ret; |
722 | |
723 | if (IS_ERR(link)) |
724 | goto fail; |
725 | |
726 | if (*link == '/') { |
727 | set_root(nd); |
728 | path_put(&nd->path); |
729 | nd->path = nd->root; |
730 | path_get(&nd->root); |
731 | } |
732 | nd->inode = nd->path.dentry->d_inode; |
733 | |
734 | ret = link_path_walk(link, nd); |
735 | return ret; |
736 | fail: |
737 | path_put(&nd->path); |
738 | return PTR_ERR(link); |
739 | } |
740 | |
741 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
742 | { |
743 | dput(path->dentry); |
744 | if (path->mnt != nd->path.mnt) |
745 | mntput(path->mnt); |
746 | } |
747 | |
748 | static inline void path_to_nameidata(const struct path *path, |
749 | struct nameidata *nd) |
750 | { |
751 | if (!(nd->flags & LOOKUP_RCU)) { |
752 | dput(nd->path.dentry); |
753 | if (nd->path.mnt != path->mnt) |
754 | mntput(nd->path.mnt); |
755 | } |
756 | nd->path.mnt = path->mnt; |
757 | nd->path.dentry = path->dentry; |
758 | } |
759 | |
760 | static __always_inline int |
761 | __do_follow_link(const struct path *link, struct nameidata *nd, void **p) |
762 | { |
763 | int error; |
764 | struct dentry *dentry = link->dentry; |
765 | |
766 | BUG_ON(nd->flags & LOOKUP_RCU); |
767 | |
768 | touch_atime(link->mnt, dentry); |
769 | nd_set_link(nd, NULL); |
770 | |
771 | if (link->mnt == nd->path.mnt) |
772 | mntget(link->mnt); |
773 | |
774 | nd->last_type = LAST_BIND; |
775 | *p = dentry->d_inode->i_op->follow_link(dentry, nd); |
776 | error = PTR_ERR(*p); |
777 | if (!IS_ERR(*p)) { |
778 | char *s = nd_get_link(nd); |
779 | error = 0; |
780 | if (s) |
781 | error = __vfs_follow_link(nd, s); |
782 | else if (nd->last_type == LAST_BIND) { |
783 | error = force_reval_path(&nd->path, nd); |
784 | if (error) |
785 | path_put(&nd->path); |
786 | } |
787 | } |
788 | return error; |
789 | } |
790 | |
791 | /* |
792 | * This limits recursive symlink follows to 8, while |
793 | * limiting consecutive symlinks to 40. |
794 | * |
795 | * Without that kind of total limit, nasty chains of consecutive |
796 | * symlinks can cause almost arbitrarily long lookups. |
797 | */ |
798 | static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd) |
799 | { |
800 | void *cookie; |
801 | int err = -ELOOP; |
802 | |
803 | /* We drop rcu-walk here */ |
804 | if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) |
805 | return -ECHILD; |
806 | BUG_ON(inode != path->dentry->d_inode); |
807 | |
808 | if (current->link_count >= MAX_NESTED_LINKS) |
809 | goto loop; |
810 | if (current->total_link_count >= 40) |
811 | goto loop; |
812 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
813 | cond_resched(); |
814 | err = security_inode_follow_link(path->dentry, nd); |
815 | if (err) |
816 | goto loop; |
817 | current->link_count++; |
818 | current->total_link_count++; |
819 | nd->depth++; |
820 | err = __do_follow_link(path, nd, &cookie); |
821 | if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) |
822 | path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); |
823 | path_put(path); |
824 | current->link_count--; |
825 | nd->depth--; |
826 | return err; |
827 | loop: |
828 | path_put_conditional(path, nd); |
829 | path_put(&nd->path); |
830 | return err; |
831 | } |
832 | |
833 | static int follow_up_rcu(struct path *path) |
834 | { |
835 | struct vfsmount *parent; |
836 | struct dentry *mountpoint; |
837 | |
838 | parent = path->mnt->mnt_parent; |
839 | if (parent == path->mnt) |
840 | return 0; |
841 | mountpoint = path->mnt->mnt_mountpoint; |
842 | path->dentry = mountpoint; |
843 | path->mnt = parent; |
844 | return 1; |
845 | } |
846 | |
847 | int follow_up(struct path *path) |
848 | { |
849 | struct vfsmount *parent; |
850 | struct dentry *mountpoint; |
851 | |
852 | br_read_lock(vfsmount_lock); |
853 | parent = path->mnt->mnt_parent; |
854 | if (parent == path->mnt) { |
855 | br_read_unlock(vfsmount_lock); |
856 | return 0; |
857 | } |
858 | mntget(parent); |
859 | mountpoint = dget(path->mnt->mnt_mountpoint); |
860 | br_read_unlock(vfsmount_lock); |
861 | dput(path->dentry); |
862 | path->dentry = mountpoint; |
863 | mntput(path->mnt); |
864 | path->mnt = parent; |
865 | return 1; |
866 | } |
867 | |
868 | /* |
869 | * Perform an automount |
870 | * - return -EISDIR to tell follow_managed() to stop and return the path we |
871 | * were called with. |
872 | */ |
873 | static int follow_automount(struct path *path, unsigned flags, |
874 | bool *need_mntput) |
875 | { |
876 | struct vfsmount *mnt; |
877 | int err; |
878 | |
879 | if (!path->dentry->d_op || !path->dentry->d_op->d_automount) |
880 | return -EREMOTE; |
881 | |
882 | /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT |
883 | * and this is the terminal part of the path. |
884 | */ |
885 | if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) |
886 | return -EISDIR; /* we actually want to stop here */ |
887 | |
888 | /* We want to mount if someone is trying to open/create a file of any |
889 | * type under the mountpoint, wants to traverse through the mountpoint |
890 | * or wants to open the mounted directory. |
891 | * |
892 | * We don't want to mount if someone's just doing a stat and they've |
893 | * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and |
894 | * appended a '/' to the name. |
895 | */ |
896 | if (!(flags & LOOKUP_FOLLOW) && |
897 | !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | |
898 | LOOKUP_OPEN | LOOKUP_CREATE))) |
899 | return -EISDIR; |
900 | |
901 | current->total_link_count++; |
902 | if (current->total_link_count >= 40) |
903 | return -ELOOP; |
904 | |
905 | mnt = path->dentry->d_op->d_automount(path); |
906 | if (IS_ERR(mnt)) { |
907 | /* |
908 | * The filesystem is allowed to return -EISDIR here to indicate |
909 | * it doesn't want to automount. For instance, autofs would do |
910 | * this so that its userspace daemon can mount on this dentry. |
911 | * |
912 | * However, we can only permit this if it's a terminal point in |
913 | * the path being looked up; if it wasn't then the remainder of |
914 | * the path is inaccessible and we should say so. |
915 | */ |
916 | if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) |
917 | return -EREMOTE; |
918 | return PTR_ERR(mnt); |
919 | } |
920 | |
921 | if (!mnt) /* mount collision */ |
922 | return 0; |
923 | |
924 | err = finish_automount(mnt, path); |
925 | |
926 | switch (err) { |
927 | case -EBUSY: |
928 | /* Someone else made a mount here whilst we were busy */ |
929 | return 0; |
930 | case 0: |
931 | dput(path->dentry); |
932 | if (*need_mntput) |
933 | mntput(path->mnt); |
934 | path->mnt = mnt; |
935 | path->dentry = dget(mnt->mnt_root); |
936 | *need_mntput = true; |
937 | return 0; |
938 | default: |
939 | return err; |
940 | } |
941 | |
942 | } |
943 | |
944 | /* |
945 | * Handle a dentry that is managed in some way. |
946 | * - Flagged for transit management (autofs) |
947 | * - Flagged as mountpoint |
948 | * - Flagged as automount point |
949 | * |
950 | * This may only be called in refwalk mode. |
951 | * |
952 | * Serialization is taken care of in namespace.c |
953 | */ |
954 | static int follow_managed(struct path *path, unsigned flags) |
955 | { |
956 | unsigned managed; |
957 | bool need_mntput = false; |
958 | int ret; |
959 | |
960 | /* Given that we're not holding a lock here, we retain the value in a |
961 | * local variable for each dentry as we look at it so that we don't see |
962 | * the components of that value change under us */ |
963 | while (managed = ACCESS_ONCE(path->dentry->d_flags), |
964 | managed &= DCACHE_MANAGED_DENTRY, |
965 | unlikely(managed != 0)) { |
966 | /* Allow the filesystem to manage the transit without i_mutex |
967 | * being held. */ |
968 | if (managed & DCACHE_MANAGE_TRANSIT) { |
969 | BUG_ON(!path->dentry->d_op); |
970 | BUG_ON(!path->dentry->d_op->d_manage); |
971 | ret = path->dentry->d_op->d_manage(path->dentry, |
972 | false, false); |
973 | if (ret < 0) |
974 | return ret == -EISDIR ? 0 : ret; |
975 | } |
976 | |
977 | /* Transit to a mounted filesystem. */ |
978 | if (managed & DCACHE_MOUNTED) { |
979 | struct vfsmount *mounted = lookup_mnt(path); |
980 | if (mounted) { |
981 | dput(path->dentry); |
982 | if (need_mntput) |
983 | mntput(path->mnt); |
984 | path->mnt = mounted; |
985 | path->dentry = dget(mounted->mnt_root); |
986 | need_mntput = true; |
987 | continue; |
988 | } |
989 | |
990 | /* Something is mounted on this dentry in another |
991 | * namespace and/or whatever was mounted there in this |
992 | * namespace got unmounted before we managed to get the |
993 | * vfsmount_lock */ |
994 | } |
995 | |
996 | /* Handle an automount point */ |
997 | if (managed & DCACHE_NEED_AUTOMOUNT) { |
998 | ret = follow_automount(path, flags, &need_mntput); |
999 | if (ret < 0) |
1000 | return ret == -EISDIR ? 0 : ret; |
1001 | continue; |
1002 | } |
1003 | |
1004 | /* We didn't change the current path point */ |
1005 | break; |
1006 | } |
1007 | return 0; |
1008 | } |
1009 | |
1010 | int follow_down_one(struct path *path) |
1011 | { |
1012 | struct vfsmount *mounted; |
1013 | |
1014 | mounted = lookup_mnt(path); |
1015 | if (mounted) { |
1016 | dput(path->dentry); |
1017 | mntput(path->mnt); |
1018 | path->mnt = mounted; |
1019 | path->dentry = dget(mounted->mnt_root); |
1020 | return 1; |
1021 | } |
1022 | return 0; |
1023 | } |
1024 | |
1025 | /* |
1026 | * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we |
1027 | * meet a managed dentry and we're not walking to "..". True is returned to |
1028 | * continue, false to abort. |
1029 | */ |
1030 | static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, |
1031 | struct inode **inode, bool reverse_transit) |
1032 | { |
1033 | while (d_mountpoint(path->dentry)) { |
1034 | struct vfsmount *mounted; |
1035 | if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && |
1036 | !reverse_transit && |
1037 | path->dentry->d_op->d_manage(path->dentry, false, true) < 0) |
1038 | return false; |
1039 | mounted = __lookup_mnt(path->mnt, path->dentry, 1); |
1040 | if (!mounted) |
1041 | break; |
1042 | path->mnt = mounted; |
1043 | path->dentry = mounted->mnt_root; |
1044 | nd->seq = read_seqcount_begin(&path->dentry->d_seq); |
1045 | *inode = path->dentry->d_inode; |
1046 | } |
1047 | |
1048 | if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) |
1049 | return reverse_transit; |
1050 | return true; |
1051 | } |
1052 | |
1053 | static int follow_dotdot_rcu(struct nameidata *nd) |
1054 | { |
1055 | struct inode *inode = nd->inode; |
1056 | |
1057 | set_root_rcu(nd); |
1058 | |
1059 | while (1) { |
1060 | if (nd->path.dentry == nd->root.dentry && |
1061 | nd->path.mnt == nd->root.mnt) { |
1062 | break; |
1063 | } |
1064 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1065 | struct dentry *old = nd->path.dentry; |
1066 | struct dentry *parent = old->d_parent; |
1067 | unsigned seq; |
1068 | |
1069 | seq = read_seqcount_begin(&parent->d_seq); |
1070 | if (read_seqcount_retry(&old->d_seq, nd->seq)) |
1071 | return -ECHILD; |
1072 | inode = parent->d_inode; |
1073 | nd->path.dentry = parent; |
1074 | nd->seq = seq; |
1075 | break; |
1076 | } |
1077 | if (!follow_up_rcu(&nd->path)) |
1078 | break; |
1079 | nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); |
1080 | inode = nd->path.dentry->d_inode; |
1081 | } |
1082 | __follow_mount_rcu(nd, &nd->path, &inode, true); |
1083 | nd->inode = inode; |
1084 | |
1085 | return 0; |
1086 | } |
1087 | |
1088 | /* |
1089 | * Follow down to the covering mount currently visible to userspace. At each |
1090 | * point, the filesystem owning that dentry may be queried as to whether the |
1091 | * caller is permitted to proceed or not. |
1092 | * |
1093 | * Care must be taken as namespace_sem may be held (indicated by mounting_here |
1094 | * being true). |
1095 | */ |
1096 | int follow_down(struct path *path, bool mounting_here) |
1097 | { |
1098 | unsigned managed; |
1099 | int ret; |
1100 | |
1101 | while (managed = ACCESS_ONCE(path->dentry->d_flags), |
1102 | unlikely(managed & DCACHE_MANAGED_DENTRY)) { |
1103 | /* Allow the filesystem to manage the transit without i_mutex |
1104 | * being held. |
1105 | * |
1106 | * We indicate to the filesystem if someone is trying to mount |
1107 | * something here. This gives autofs the chance to deny anyone |
1108 | * other than its daemon the right to mount on its |
1109 | * superstructure. |
1110 | * |
1111 | * The filesystem may sleep at this point. |
1112 | */ |
1113 | if (managed & DCACHE_MANAGE_TRANSIT) { |
1114 | BUG_ON(!path->dentry->d_op); |
1115 | BUG_ON(!path->dentry->d_op->d_manage); |
1116 | ret = path->dentry->d_op->d_manage( |
1117 | path->dentry, mounting_here, false); |
1118 | if (ret < 0) |
1119 | return ret == -EISDIR ? 0 : ret; |
1120 | } |
1121 | |
1122 | /* Transit to a mounted filesystem. */ |
1123 | if (managed & DCACHE_MOUNTED) { |
1124 | struct vfsmount *mounted = lookup_mnt(path); |
1125 | if (!mounted) |
1126 | break; |
1127 | dput(path->dentry); |
1128 | mntput(path->mnt); |
1129 | path->mnt = mounted; |
1130 | path->dentry = dget(mounted->mnt_root); |
1131 | continue; |
1132 | } |
1133 | |
1134 | /* Don't handle automount points here */ |
1135 | break; |
1136 | } |
1137 | return 0; |
1138 | } |
1139 | |
1140 | /* |
1141 | * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() |
1142 | */ |
1143 | static void follow_mount(struct path *path) |
1144 | { |
1145 | while (d_mountpoint(path->dentry)) { |
1146 | struct vfsmount *mounted = lookup_mnt(path); |
1147 | if (!mounted) |
1148 | break; |
1149 | dput(path->dentry); |
1150 | mntput(path->mnt); |
1151 | path->mnt = mounted; |
1152 | path->dentry = dget(mounted->mnt_root); |
1153 | } |
1154 | } |
1155 | |
1156 | static void follow_dotdot(struct nameidata *nd) |
1157 | { |
1158 | set_root(nd); |
1159 | |
1160 | while(1) { |
1161 | struct dentry *old = nd->path.dentry; |
1162 | |
1163 | if (nd->path.dentry == nd->root.dentry && |
1164 | nd->path.mnt == nd->root.mnt) { |
1165 | break; |
1166 | } |
1167 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
1168 | /* rare case of legitimate dget_parent()... */ |
1169 | nd->path.dentry = dget_parent(nd->path.dentry); |
1170 | dput(old); |
1171 | break; |
1172 | } |
1173 | if (!follow_up(&nd->path)) |
1174 | break; |
1175 | } |
1176 | follow_mount(&nd->path); |
1177 | nd->inode = nd->path.dentry->d_inode; |
1178 | } |
1179 | |
1180 | /* |
1181 | * Allocate a dentry with name and parent, and perform a parent |
1182 | * directory ->lookup on it. Returns the new dentry, or ERR_PTR |
1183 | * on error. parent->d_inode->i_mutex must be held. d_lookup must |
1184 | * have verified that no child exists while under i_mutex. |
1185 | */ |
1186 | static struct dentry *d_alloc_and_lookup(struct dentry *parent, |
1187 | struct qstr *name, struct nameidata *nd) |
1188 | { |
1189 | struct inode *inode = parent->d_inode; |
1190 | struct dentry *dentry; |
1191 | struct dentry *old; |
1192 | |
1193 | /* Don't create child dentry for a dead directory. */ |
1194 | if (unlikely(IS_DEADDIR(inode))) |
1195 | return ERR_PTR(-ENOENT); |
1196 | |
1197 | dentry = d_alloc(parent, name); |
1198 | if (unlikely(!dentry)) |
1199 | return ERR_PTR(-ENOMEM); |
1200 | |
1201 | old = inode->i_op->lookup(inode, dentry, nd); |
1202 | if (unlikely(old)) { |
1203 | dput(dentry); |
1204 | dentry = old; |
1205 | } |
1206 | return dentry; |
1207 | } |
1208 | |
1209 | /* |
1210 | * It's more convoluted than I'd like it to be, but... it's still fairly |
1211 | * small and for now I'd prefer to have fast path as straight as possible. |
1212 | * It _is_ time-critical. |
1213 | */ |
1214 | static int do_lookup(struct nameidata *nd, struct qstr *name, |
1215 | struct path *path, struct inode **inode) |
1216 | { |
1217 | struct vfsmount *mnt = nd->path.mnt; |
1218 | struct dentry *dentry, *parent = nd->path.dentry; |
1219 | struct inode *dir; |
1220 | int err; |
1221 | |
1222 | /* |
1223 | * See if the low-level filesystem might want |
1224 | * to use its own hash.. |
1225 | */ |
1226 | if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { |
1227 | err = parent->d_op->d_hash(parent, nd->inode, name); |
1228 | if (err < 0) |
1229 | return err; |
1230 | } |
1231 | |
1232 | /* |
1233 | * Rename seqlock is not required here because in the off chance |
1234 | * of a false negative due to a concurrent rename, we're going to |
1235 | * do the non-racy lookup, below. |
1236 | */ |
1237 | if (nd->flags & LOOKUP_RCU) { |
1238 | unsigned seq; |
1239 | |
1240 | *inode = nd->inode; |
1241 | dentry = __d_lookup_rcu(parent, name, &seq, inode); |
1242 | if (!dentry) { |
1243 | if (nameidata_drop_rcu(nd)) |
1244 | return -ECHILD; |
1245 | goto need_lookup; |
1246 | } |
1247 | /* Memory barrier in read_seqcount_begin of child is enough */ |
1248 | if (__read_seqcount_retry(&parent->d_seq, nd->seq)) |
1249 | return -ECHILD; |
1250 | |
1251 | nd->seq = seq; |
1252 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1253 | dentry = do_revalidate_rcu(dentry, nd); |
1254 | if (!dentry) |
1255 | goto need_lookup; |
1256 | if (IS_ERR(dentry)) |
1257 | goto fail; |
1258 | if (!(nd->flags & LOOKUP_RCU)) |
1259 | goto done; |
1260 | } |
1261 | path->mnt = mnt; |
1262 | path->dentry = dentry; |
1263 | if (likely(__follow_mount_rcu(nd, path, inode, false))) |
1264 | return 0; |
1265 | if (nameidata_drop_rcu(nd)) |
1266 | return -ECHILD; |
1267 | /* fallthru */ |
1268 | } |
1269 | dentry = __d_lookup(parent, name); |
1270 | if (!dentry) |
1271 | goto need_lookup; |
1272 | found: |
1273 | if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { |
1274 | dentry = do_revalidate(dentry, nd); |
1275 | if (!dentry) |
1276 | goto need_lookup; |
1277 | if (IS_ERR(dentry)) |
1278 | goto fail; |
1279 | } |
1280 | done: |
1281 | path->mnt = mnt; |
1282 | path->dentry = dentry; |
1283 | err = follow_managed(path, nd->flags); |
1284 | if (unlikely(err < 0)) { |
1285 | path_put_conditional(path, nd); |
1286 | return err; |
1287 | } |
1288 | *inode = path->dentry->d_inode; |
1289 | return 0; |
1290 | |
1291 | need_lookup: |
1292 | dir = parent->d_inode; |
1293 | BUG_ON(nd->inode != dir); |
1294 | |
1295 | mutex_lock(&dir->i_mutex); |
1296 | /* |
1297 | * First re-do the cached lookup just in case it was created |
1298 | * while we waited for the directory semaphore, or the first |
1299 | * lookup failed due to an unrelated rename. |
1300 | * |
1301 | * This could use version numbering or similar to avoid unnecessary |
1302 | * cache lookups, but then we'd have to do the first lookup in the |
1303 | * non-racy way. However in the common case here, everything should |
1304 | * be hot in cache, so would it be a big win? |
1305 | */ |
1306 | dentry = d_lookup(parent, name); |
1307 | if (likely(!dentry)) { |
1308 | dentry = d_alloc_and_lookup(parent, name, nd); |
1309 | mutex_unlock(&dir->i_mutex); |
1310 | if (IS_ERR(dentry)) |
1311 | goto fail; |
1312 | goto done; |
1313 | } |
1314 | /* |
1315 | * Uhhuh! Nasty case: the cache was re-populated while |
1316 | * we waited on the semaphore. Need to revalidate. |
1317 | */ |
1318 | mutex_unlock(&dir->i_mutex); |
1319 | goto found; |
1320 | |
1321 | fail: |
1322 | return PTR_ERR(dentry); |
1323 | } |
1324 | |
1325 | /* |
1326 | * Name resolution. |
1327 | * This is the basic name resolution function, turning a pathname into |
1328 | * the final dentry. We expect 'base' to be positive and a directory. |
1329 | * |
1330 | * Returns 0 and nd will have valid dentry and mnt on success. |
1331 | * Returns error and drops reference to input namei data on failure. |
1332 | */ |
1333 | static int link_path_walk(const char *name, struct nameidata *nd) |
1334 | { |
1335 | struct path next; |
1336 | int err; |
1337 | unsigned int lookup_flags = nd->flags; |
1338 | |
1339 | while (*name=='/') |
1340 | name++; |
1341 | if (!*name) |
1342 | goto return_reval; |
1343 | |
1344 | if (nd->depth) |
1345 | lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); |
1346 | |
1347 | /* At this point we know we have a real path component. */ |
1348 | for(;;) { |
1349 | struct inode *inode; |
1350 | unsigned long hash; |
1351 | struct qstr this; |
1352 | unsigned int c; |
1353 | |
1354 | nd->flags |= LOOKUP_CONTINUE; |
1355 | if (nd->flags & LOOKUP_RCU) { |
1356 | err = exec_permission(nd->inode, IPERM_FLAG_RCU); |
1357 | if (err == -ECHILD) { |
1358 | if (nameidata_drop_rcu(nd)) |
1359 | return -ECHILD; |
1360 | goto exec_again; |
1361 | } |
1362 | } else { |
1363 | exec_again: |
1364 | err = exec_permission(nd->inode, 0); |
1365 | } |
1366 | if (err) |
1367 | break; |
1368 | |
1369 | this.name = name; |
1370 | c = *(const unsigned char *)name; |
1371 | |
1372 | hash = init_name_hash(); |
1373 | do { |
1374 | name++; |
1375 | hash = partial_name_hash(c, hash); |
1376 | c = *(const unsigned char *)name; |
1377 | } while (c && (c != '/')); |
1378 | this.len = name - (const char *) this.name; |
1379 | this.hash = end_name_hash(hash); |
1380 | |
1381 | /* remove trailing slashes? */ |
1382 | if (!c) |
1383 | goto last_component; |
1384 | while (*++name == '/'); |
1385 | if (!*name) |
1386 | goto last_with_slashes; |
1387 | |
1388 | /* |
1389 | * "." and ".." are special - ".." especially so because it has |
1390 | * to be able to know about the current root directory and |
1391 | * parent relationships. |
1392 | */ |
1393 | if (this.name[0] == '.') switch (this.len) { |
1394 | default: |
1395 | break; |
1396 | case 2: |
1397 | if (this.name[1] != '.') |
1398 | break; |
1399 | if (nd->flags & LOOKUP_RCU) { |
1400 | if (follow_dotdot_rcu(nd)) |
1401 | return -ECHILD; |
1402 | } else |
1403 | follow_dotdot(nd); |
1404 | /* fallthrough */ |
1405 | case 1: |
1406 | continue; |
1407 | } |
1408 | /* This does the actual lookups.. */ |
1409 | err = do_lookup(nd, &this, &next, &inode); |
1410 | if (err) |
1411 | break; |
1412 | err = -ENOENT; |
1413 | if (!inode) |
1414 | goto out_dput; |
1415 | |
1416 | if (inode->i_op->follow_link) { |
1417 | err = do_follow_link(inode, &next, nd); |
1418 | if (err) |
1419 | goto return_err; |
1420 | nd->inode = nd->path.dentry->d_inode; |
1421 | err = -ENOENT; |
1422 | if (!nd->inode) |
1423 | break; |
1424 | } else { |
1425 | path_to_nameidata(&next, nd); |
1426 | nd->inode = inode; |
1427 | } |
1428 | err = -ENOTDIR; |
1429 | if (!nd->inode->i_op->lookup) |
1430 | break; |
1431 | continue; |
1432 | /* here ends the main loop */ |
1433 | |
1434 | last_with_slashes: |
1435 | lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
1436 | last_component: |
1437 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
1438 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; |
1439 | if (lookup_flags & LOOKUP_PARENT) |
1440 | goto lookup_parent; |
1441 | if (this.name[0] == '.') switch (this.len) { |
1442 | default: |
1443 | break; |
1444 | case 2: |
1445 | if (this.name[1] != '.') |
1446 | break; |
1447 | if (nd->flags & LOOKUP_RCU) { |
1448 | if (follow_dotdot_rcu(nd)) |
1449 | return -ECHILD; |
1450 | } else |
1451 | follow_dotdot(nd); |
1452 | /* fallthrough */ |
1453 | case 1: |
1454 | goto return_reval; |
1455 | } |
1456 | err = do_lookup(nd, &this, &next, &inode); |
1457 | if (err) |
1458 | break; |
1459 | if (inode && unlikely(inode->i_op->follow_link) && |
1460 | (lookup_flags & LOOKUP_FOLLOW)) { |
1461 | err = do_follow_link(inode, &next, nd); |
1462 | if (err) |
1463 | goto return_err; |
1464 | nd->inode = nd->path.dentry->d_inode; |
1465 | } else { |
1466 | path_to_nameidata(&next, nd); |
1467 | nd->inode = inode; |
1468 | } |
1469 | err = -ENOENT; |
1470 | if (!nd->inode) |
1471 | break; |
1472 | if (lookup_flags & LOOKUP_DIRECTORY) { |
1473 | err = -ENOTDIR; |
1474 | if (!nd->inode->i_op->lookup) |
1475 | break; |
1476 | } |
1477 | goto return_base; |
1478 | lookup_parent: |
1479 | nd->last = this; |
1480 | nd->last_type = LAST_NORM; |
1481 | if (this.name[0] != '.') |
1482 | goto return_base; |
1483 | if (this.len == 1) |
1484 | nd->last_type = LAST_DOT; |
1485 | else if (this.len == 2 && this.name[1] == '.') |
1486 | nd->last_type = LAST_DOTDOT; |
1487 | else |
1488 | goto return_base; |
1489 | return_reval: |
1490 | /* |
1491 | * We bypassed the ordinary revalidation routines. |
1492 | * We may need to check the cached dentry for staleness. |
1493 | */ |
1494 | if (need_reval_dot(nd->path.dentry)) { |
1495 | if (nameidata_drop_rcu_last_maybe(nd)) |
1496 | return -ECHILD; |
1497 | /* Note: we do not d_invalidate() */ |
1498 | err = d_revalidate(nd->path.dentry, nd); |
1499 | if (!err) |
1500 | err = -ESTALE; |
1501 | if (err < 0) |
1502 | break; |
1503 | return 0; |
1504 | } |
1505 | return_base: |
1506 | if (nameidata_drop_rcu_last_maybe(nd)) |
1507 | return -ECHILD; |
1508 | return 0; |
1509 | out_dput: |
1510 | if (!(nd->flags & LOOKUP_RCU)) |
1511 | path_put_conditional(&next, nd); |
1512 | break; |
1513 | } |
1514 | if (!(nd->flags & LOOKUP_RCU)) |
1515 | path_put(&nd->path); |
1516 | return_err: |
1517 | return err; |
1518 | } |
1519 | |
1520 | static inline int path_walk_rcu(const char *name, struct nameidata *nd) |
1521 | { |
1522 | current->total_link_count = 0; |
1523 | |
1524 | return link_path_walk(name, nd); |
1525 | } |
1526 | |
1527 | static inline int path_walk_simple(const char *name, struct nameidata *nd) |
1528 | { |
1529 | current->total_link_count = 0; |
1530 | |
1531 | return link_path_walk(name, nd); |
1532 | } |
1533 | |
1534 | static int path_walk(const char *name, struct nameidata *nd) |
1535 | { |
1536 | struct path save = nd->path; |
1537 | int result; |
1538 | |
1539 | current->total_link_count = 0; |
1540 | |
1541 | /* make sure the stuff we saved doesn't go away */ |
1542 | path_get(&save); |
1543 | |
1544 | result = link_path_walk(name, nd); |
1545 | if (result == -ESTALE) { |
1546 | /* nd->path had been dropped */ |
1547 | current->total_link_count = 0; |
1548 | nd->path = save; |
1549 | nd->inode = save.dentry->d_inode; |
1550 | path_get(&nd->path); |
1551 | nd->flags |= LOOKUP_REVAL; |
1552 | result = link_path_walk(name, nd); |
1553 | } |
1554 | |
1555 | path_put(&save); |
1556 | |
1557 | return result; |
1558 | } |
1559 | |
1560 | static void path_finish_rcu(struct nameidata *nd) |
1561 | { |
1562 | if (nd->flags & LOOKUP_RCU) { |
1563 | /* RCU dangling. Cancel it. */ |
1564 | nd->flags &= ~LOOKUP_RCU; |
1565 | nd->root.mnt = NULL; |
1566 | rcu_read_unlock(); |
1567 | br_read_unlock(vfsmount_lock); |
1568 | } |
1569 | if (nd->file) |
1570 | fput(nd->file); |
1571 | } |
1572 | |
1573 | static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd) |
1574 | { |
1575 | int retval = 0; |
1576 | int fput_needed; |
1577 | struct file *file; |
1578 | |
1579 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ |
1580 | nd->flags = flags | LOOKUP_RCU; |
1581 | nd->depth = 0; |
1582 | nd->root.mnt = NULL; |
1583 | nd->file = NULL; |
1584 | |
1585 | if (*name=='/') { |
1586 | struct fs_struct *fs = current->fs; |
1587 | unsigned seq; |
1588 | |
1589 | br_read_lock(vfsmount_lock); |
1590 | rcu_read_lock(); |
1591 | |
1592 | do { |
1593 | seq = read_seqcount_begin(&fs->seq); |
1594 | nd->root = fs->root; |
1595 | nd->path = nd->root; |
1596 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1597 | } while (read_seqcount_retry(&fs->seq, seq)); |
1598 | |
1599 | } else if (dfd == AT_FDCWD) { |
1600 | struct fs_struct *fs = current->fs; |
1601 | unsigned seq; |
1602 | |
1603 | br_read_lock(vfsmount_lock); |
1604 | rcu_read_lock(); |
1605 | |
1606 | do { |
1607 | seq = read_seqcount_begin(&fs->seq); |
1608 | nd->path = fs->pwd; |
1609 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1610 | } while (read_seqcount_retry(&fs->seq, seq)); |
1611 | |
1612 | } else { |
1613 | struct dentry *dentry; |
1614 | |
1615 | file = fget_light(dfd, &fput_needed); |
1616 | retval = -EBADF; |
1617 | if (!file) |
1618 | goto out_fail; |
1619 | |
1620 | dentry = file->f_path.dentry; |
1621 | |
1622 | retval = -ENOTDIR; |
1623 | if (!S_ISDIR(dentry->d_inode->i_mode)) |
1624 | goto fput_fail; |
1625 | |
1626 | retval = file_permission(file, MAY_EXEC); |
1627 | if (retval) |
1628 | goto fput_fail; |
1629 | |
1630 | nd->path = file->f_path; |
1631 | if (fput_needed) |
1632 | nd->file = file; |
1633 | |
1634 | nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); |
1635 | br_read_lock(vfsmount_lock); |
1636 | rcu_read_lock(); |
1637 | } |
1638 | nd->inode = nd->path.dentry->d_inode; |
1639 | return 0; |
1640 | |
1641 | fput_fail: |
1642 | fput_light(file, fput_needed); |
1643 | out_fail: |
1644 | return retval; |
1645 | } |
1646 | |
1647 | static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) |
1648 | { |
1649 | int retval = 0; |
1650 | int fput_needed; |
1651 | struct file *file; |
1652 | |
1653 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ |
1654 | nd->flags = flags; |
1655 | nd->depth = 0; |
1656 | nd->root.mnt = NULL; |
1657 | |
1658 | if (*name=='/') { |
1659 | set_root(nd); |
1660 | nd->path = nd->root; |
1661 | path_get(&nd->root); |
1662 | } else if (dfd == AT_FDCWD) { |
1663 | get_fs_pwd(current->fs, &nd->path); |
1664 | } else { |
1665 | struct dentry *dentry; |
1666 | |
1667 | file = fget_light(dfd, &fput_needed); |
1668 | retval = -EBADF; |
1669 | if (!file) |
1670 | goto out_fail; |
1671 | |
1672 | dentry = file->f_path.dentry; |
1673 | |
1674 | retval = -ENOTDIR; |
1675 | if (!S_ISDIR(dentry->d_inode->i_mode)) |
1676 | goto fput_fail; |
1677 | |
1678 | retval = file_permission(file, MAY_EXEC); |
1679 | if (retval) |
1680 | goto fput_fail; |
1681 | |
1682 | nd->path = file->f_path; |
1683 | path_get(&file->f_path); |
1684 | |
1685 | fput_light(file, fput_needed); |
1686 | } |
1687 | nd->inode = nd->path.dentry->d_inode; |
1688 | return 0; |
1689 | |
1690 | fput_fail: |
1691 | fput_light(file, fput_needed); |
1692 | out_fail: |
1693 | return retval; |
1694 | } |
1695 | |
1696 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
1697 | static int do_path_lookup(int dfd, const char *name, |
1698 | unsigned int flags, struct nameidata *nd) |
1699 | { |
1700 | int retval; |
1701 | |
1702 | /* |
1703 | * Path walking is largely split up into 2 different synchronisation |
1704 | * schemes, rcu-walk and ref-walk (explained in |
1705 | * Documentation/filesystems/path-lookup.txt). These share much of the |
1706 | * path walk code, but some things particularly setup, cleanup, and |
1707 | * following mounts are sufficiently divergent that functions are |
1708 | * duplicated. Typically there is a function foo(), and its RCU |
1709 | * analogue, foo_rcu(). |
1710 | * |
1711 | * -ECHILD is the error number of choice (just to avoid clashes) that |
1712 | * is returned if some aspect of an rcu-walk fails. Such an error must |
1713 | * be handled by restarting a traditional ref-walk (which will always |
1714 | * be able to complete). |
1715 | */ |
1716 | retval = path_init_rcu(dfd, name, flags, nd); |
1717 | if (unlikely(retval)) |
1718 | return retval; |
1719 | retval = path_walk_rcu(name, nd); |
1720 | path_finish_rcu(nd); |
1721 | if (nd->root.mnt) { |
1722 | path_put(&nd->root); |
1723 | nd->root.mnt = NULL; |
1724 | } |
1725 | |
1726 | if (unlikely(retval == -ECHILD || retval == -ESTALE)) { |
1727 | /* slower, locked walk */ |
1728 | if (retval == -ESTALE) |
1729 | flags |= LOOKUP_REVAL; |
1730 | retval = path_init(dfd, name, flags, nd); |
1731 | if (unlikely(retval)) |
1732 | return retval; |
1733 | retval = path_walk(name, nd); |
1734 | if (nd->root.mnt) { |
1735 | path_put(&nd->root); |
1736 | nd->root.mnt = NULL; |
1737 | } |
1738 | } |
1739 | |
1740 | if (likely(!retval)) { |
1741 | if (unlikely(!audit_dummy_context())) { |
1742 | if (nd->path.dentry && nd->inode) |
1743 | audit_inode(name, nd->path.dentry); |
1744 | } |
1745 | } |
1746 | |
1747 | return retval; |
1748 | } |
1749 | |
1750 | int path_lookup(const char *name, unsigned int flags, |
1751 | struct nameidata *nd) |
1752 | { |
1753 | return do_path_lookup(AT_FDCWD, name, flags, nd); |
1754 | } |
1755 | |
1756 | int kern_path(const char *name, unsigned int flags, struct path *path) |
1757 | { |
1758 | struct nameidata nd; |
1759 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); |
1760 | if (!res) |
1761 | *path = nd.path; |
1762 | return res; |
1763 | } |
1764 | |
1765 | /** |
1766 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair |
1767 | * @dentry: pointer to dentry of the base directory |
1768 | * @mnt: pointer to vfs mount of the base directory |
1769 | * @name: pointer to file name |
1770 | * @flags: lookup flags |
1771 | * @nd: pointer to nameidata |
1772 | */ |
1773 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, |
1774 | const char *name, unsigned int flags, |
1775 | struct nameidata *nd) |
1776 | { |
1777 | int retval; |
1778 | |
1779 | /* same as do_path_lookup */ |
1780 | nd->last_type = LAST_ROOT; |
1781 | nd->flags = flags; |
1782 | nd->depth = 0; |
1783 | |
1784 | nd->path.dentry = dentry; |
1785 | nd->path.mnt = mnt; |
1786 | path_get(&nd->path); |
1787 | nd->root = nd->path; |
1788 | path_get(&nd->root); |
1789 | nd->inode = nd->path.dentry->d_inode; |
1790 | |
1791 | retval = path_walk(name, nd); |
1792 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && |
1793 | nd->inode)) |
1794 | audit_inode(name, nd->path.dentry); |
1795 | |
1796 | path_put(&nd->root); |
1797 | nd->root.mnt = NULL; |
1798 | |
1799 | return retval; |
1800 | } |
1801 | |
1802 | static struct dentry *__lookup_hash(struct qstr *name, |
1803 | struct dentry *base, struct nameidata *nd) |
1804 | { |
1805 | struct inode *inode = base->d_inode; |
1806 | struct dentry *dentry; |
1807 | int err; |
1808 | |
1809 | err = exec_permission(inode, 0); |
1810 | if (err) |
1811 | return ERR_PTR(err); |
1812 | |
1813 | /* |
1814 | * See if the low-level filesystem might want |
1815 | * to use its own hash.. |
1816 | */ |
1817 | if (base->d_flags & DCACHE_OP_HASH) { |
1818 | err = base->d_op->d_hash(base, inode, name); |
1819 | dentry = ERR_PTR(err); |
1820 | if (err < 0) |
1821 | goto out; |
1822 | } |
1823 | |
1824 | /* |
1825 | * Don't bother with __d_lookup: callers are for creat as |
1826 | * well as unlink, so a lot of the time it would cost |
1827 | * a double lookup. |
1828 | */ |
1829 | dentry = d_lookup(base, name); |
1830 | |
1831 | if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) |
1832 | dentry = do_revalidate(dentry, nd); |
1833 | |
1834 | if (!dentry) |
1835 | dentry = d_alloc_and_lookup(base, name, nd); |
1836 | out: |
1837 | return dentry; |
1838 | } |
1839 | |
1840 | /* |
1841 | * Restricted form of lookup. Doesn't follow links, single-component only, |
1842 | * needs parent already locked. Doesn't follow mounts. |
1843 | * SMP-safe. |
1844 | */ |
1845 | static struct dentry *lookup_hash(struct nameidata *nd) |
1846 | { |
1847 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
1848 | } |
1849 | |
1850 | static int __lookup_one_len(const char *name, struct qstr *this, |
1851 | struct dentry *base, int len) |
1852 | { |
1853 | unsigned long hash; |
1854 | unsigned int c; |
1855 | |
1856 | this->name = name; |
1857 | this->len = len; |
1858 | if (!len) |
1859 | return -EACCES; |
1860 | |
1861 | hash = init_name_hash(); |
1862 | while (len--) { |
1863 | c = *(const unsigned char *)name++; |
1864 | if (c == '/' || c == '\0') |
1865 | return -EACCES; |
1866 | hash = partial_name_hash(c, hash); |
1867 | } |
1868 | this->hash = end_name_hash(hash); |
1869 | return 0; |
1870 | } |
1871 | |
1872 | /** |
1873 | * lookup_one_len - filesystem helper to lookup single pathname component |
1874 | * @name: pathname component to lookup |
1875 | * @base: base directory to lookup from |
1876 | * @len: maximum length @len should be interpreted to |
1877 | * |
1878 | * Note that this routine is purely a helper for filesystem usage and should |
1879 | * not be called by generic code. Also note that by using this function the |
1880 | * nameidata argument is passed to the filesystem methods and a filesystem |
1881 | * using this helper needs to be prepared for that. |
1882 | */ |
1883 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
1884 | { |
1885 | int err; |
1886 | struct qstr this; |
1887 | |
1888 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
1889 | |
1890 | err = __lookup_one_len(name, &this, base, len); |
1891 | if (err) |
1892 | return ERR_PTR(err); |
1893 | |
1894 | return __lookup_hash(&this, base, NULL); |
1895 | } |
1896 | |
1897 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
1898 | struct path *path) |
1899 | { |
1900 | struct nameidata nd; |
1901 | char *tmp = getname(name); |
1902 | int err = PTR_ERR(tmp); |
1903 | if (!IS_ERR(tmp)) { |
1904 | |
1905 | BUG_ON(flags & LOOKUP_PARENT); |
1906 | |
1907 | err = do_path_lookup(dfd, tmp, flags, &nd); |
1908 | putname(tmp); |
1909 | if (!err) |
1910 | *path = nd.path; |
1911 | } |
1912 | return err; |
1913 | } |
1914 | |
1915 | static int user_path_parent(int dfd, const char __user *path, |
1916 | struct nameidata *nd, char **name) |
1917 | { |
1918 | char *s = getname(path); |
1919 | int error; |
1920 | |
1921 | if (IS_ERR(s)) |
1922 | return PTR_ERR(s); |
1923 | |
1924 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); |
1925 | if (error) |
1926 | putname(s); |
1927 | else |
1928 | *name = s; |
1929 | |
1930 | return error; |
1931 | } |
1932 | |
1933 | /* |
1934 | * It's inline, so penalty for filesystems that don't use sticky bit is |
1935 | * minimal. |
1936 | */ |
1937 | static inline int check_sticky(struct inode *dir, struct inode *inode) |
1938 | { |
1939 | uid_t fsuid = current_fsuid(); |
1940 | |
1941 | if (!(dir->i_mode & S_ISVTX)) |
1942 | return 0; |
1943 | if (inode->i_uid == fsuid) |
1944 | return 0; |
1945 | if (dir->i_uid == fsuid) |
1946 | return 0; |
1947 | return !capable(CAP_FOWNER); |
1948 | } |
1949 | |
1950 | /* |
1951 | * Check whether we can remove a link victim from directory dir, check |
1952 | * whether the type of victim is right. |
1953 | * 1. We can't do it if dir is read-only (done in permission()) |
1954 | * 2. We should have write and exec permissions on dir |
1955 | * 3. We can't remove anything from append-only dir |
1956 | * 4. We can't do anything with immutable dir (done in permission()) |
1957 | * 5. If the sticky bit on dir is set we should either |
1958 | * a. be owner of dir, or |
1959 | * b. be owner of victim, or |
1960 | * c. have CAP_FOWNER capability |
1961 | * 6. If the victim is append-only or immutable we can't do antyhing with |
1962 | * links pointing to it. |
1963 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. |
1964 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. |
1965 | * 9. We can't remove a root or mountpoint. |
1966 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by |
1967 | * nfs_async_unlink(). |
1968 | */ |
1969 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
1970 | { |
1971 | int error; |
1972 | |
1973 | if (!victim->d_inode) |
1974 | return -ENOENT; |
1975 | |
1976 | BUG_ON(victim->d_parent->d_inode != dir); |
1977 | audit_inode_child(victim, dir); |
1978 | |
1979 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
1980 | if (error) |
1981 | return error; |
1982 | if (IS_APPEND(dir)) |
1983 | return -EPERM; |
1984 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| |
1985 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
1986 | return -EPERM; |
1987 | if (isdir) { |
1988 | if (!S_ISDIR(victim->d_inode->i_mode)) |
1989 | return -ENOTDIR; |
1990 | if (IS_ROOT(victim)) |
1991 | return -EBUSY; |
1992 | } else if (S_ISDIR(victim->d_inode->i_mode)) |
1993 | return -EISDIR; |
1994 | if (IS_DEADDIR(dir)) |
1995 | return -ENOENT; |
1996 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) |
1997 | return -EBUSY; |
1998 | return 0; |
1999 | } |
2000 | |
2001 | /* Check whether we can create an object with dentry child in directory |
2002 | * dir. |
2003 | * 1. We can't do it if child already exists (open has special treatment for |
2004 | * this case, but since we are inlined it's OK) |
2005 | * 2. We can't do it if dir is read-only (done in permission()) |
2006 | * 3. We should have write and exec permissions on dir |
2007 | * 4. We can't do it if dir is immutable (done in permission()) |
2008 | */ |
2009 | static inline int may_create(struct inode *dir, struct dentry *child) |
2010 | { |
2011 | if (child->d_inode) |
2012 | return -EEXIST; |
2013 | if (IS_DEADDIR(dir)) |
2014 | return -ENOENT; |
2015 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
2016 | } |
2017 | |
2018 | /* |
2019 | * p1 and p2 should be directories on the same fs. |
2020 | */ |
2021 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) |
2022 | { |
2023 | struct dentry *p; |
2024 | |
2025 | if (p1 == p2) { |
2026 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2027 | return NULL; |
2028 | } |
2029 | |
2030 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
2031 | |
2032 | p = d_ancestor(p2, p1); |
2033 | if (p) { |
2034 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); |
2035 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); |
2036 | return p; |
2037 | } |
2038 | |
2039 | p = d_ancestor(p1, p2); |
2040 | if (p) { |
2041 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2042 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
2043 | return p; |
2044 | } |
2045 | |
2046 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
2047 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
2048 | return NULL; |
2049 | } |
2050 | |
2051 | void unlock_rename(struct dentry *p1, struct dentry *p2) |
2052 | { |
2053 | mutex_unlock(&p1->d_inode->i_mutex); |
2054 | if (p1 != p2) { |
2055 | mutex_unlock(&p2->d_inode->i_mutex); |
2056 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
2057 | } |
2058 | } |
2059 | |
2060 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, |
2061 | struct nameidata *nd) |
2062 | { |
2063 | int error = may_create(dir, dentry); |
2064 | |
2065 | if (error) |
2066 | return error; |
2067 | |
2068 | if (!dir->i_op->create) |
2069 | return -EACCES; /* shouldn't it be ENOSYS? */ |
2070 | mode &= S_IALLUGO; |
2071 | mode |= S_IFREG; |
2072 | error = security_inode_create(dir, dentry, mode); |
2073 | if (error) |
2074 | return error; |
2075 | error = dir->i_op->create(dir, dentry, mode, nd); |
2076 | if (!error) |
2077 | fsnotify_create(dir, dentry); |
2078 | return error; |
2079 | } |
2080 | |
2081 | int may_open(struct path *path, int acc_mode, int flag) |
2082 | { |
2083 | struct dentry *dentry = path->dentry; |
2084 | struct inode *inode = dentry->d_inode; |
2085 | int error; |
2086 | |
2087 | if (!inode) |
2088 | return -ENOENT; |
2089 | |
2090 | switch (inode->i_mode & S_IFMT) { |
2091 | case S_IFLNK: |
2092 | return -ELOOP; |
2093 | case S_IFDIR: |
2094 | if (acc_mode & MAY_WRITE) |
2095 | return -EISDIR; |
2096 | break; |
2097 | case S_IFBLK: |
2098 | case S_IFCHR: |
2099 | if (path->mnt->mnt_flags & MNT_NODEV) |
2100 | return -EACCES; |
2101 | /*FALLTHRU*/ |
2102 | case S_IFIFO: |
2103 | case S_IFSOCK: |
2104 | flag &= ~O_TRUNC; |
2105 | break; |
2106 | } |
2107 | |
2108 | error = inode_permission(inode, acc_mode); |
2109 | if (error) |
2110 | return error; |
2111 | |
2112 | /* |
2113 | * An append-only file must be opened in append mode for writing. |
2114 | */ |
2115 | if (IS_APPEND(inode)) { |
2116 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) |
2117 | return -EPERM; |
2118 | if (flag & O_TRUNC) |
2119 | return -EPERM; |
2120 | } |
2121 | |
2122 | /* O_NOATIME can only be set by the owner or superuser */ |
2123 | if (flag & O_NOATIME && !is_owner_or_cap(inode)) |
2124 | return -EPERM; |
2125 | |
2126 | /* |
2127 | * Ensure there are no outstanding leases on the file. |
2128 | */ |
2129 | return break_lease(inode, flag); |
2130 | } |
2131 | |
2132 | static int handle_truncate(struct file *filp) |
2133 | { |
2134 | struct path *path = &filp->f_path; |
2135 | struct inode *inode = path->dentry->d_inode; |
2136 | int error = get_write_access(inode); |
2137 | if (error) |
2138 | return error; |
2139 | /* |
2140 | * Refuse to truncate files with mandatory locks held on them. |
2141 | */ |
2142 | error = locks_verify_locked(inode); |
2143 | if (!error) |
2144 | error = security_path_truncate(path); |
2145 | if (!error) { |
2146 | error = do_truncate(path->dentry, 0, |
2147 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, |
2148 | filp); |
2149 | } |
2150 | put_write_access(inode); |
2151 | return error; |
2152 | } |
2153 | |
2154 | /* |
2155 | * Be careful about ever adding any more callers of this |
2156 | * function. Its flags must be in the namei format, not |
2157 | * what get passed to sys_open(). |
2158 | */ |
2159 | static int __open_namei_create(struct nameidata *nd, struct path *path, |
2160 | int open_flag, int mode) |
2161 | { |
2162 | int error; |
2163 | struct dentry *dir = nd->path.dentry; |
2164 | |
2165 | if (!IS_POSIXACL(dir->d_inode)) |
2166 | mode &= ~current_umask(); |
2167 | error = security_path_mknod(&nd->path, path->dentry, mode, 0); |
2168 | if (error) |
2169 | goto out_unlock; |
2170 | error = vfs_create(dir->d_inode, path->dentry, mode, nd); |
2171 | out_unlock: |
2172 | mutex_unlock(&dir->d_inode->i_mutex); |
2173 | dput(nd->path.dentry); |
2174 | nd->path.dentry = path->dentry; |
2175 | |
2176 | if (error) |
2177 | return error; |
2178 | /* Don't check for write permission, don't truncate */ |
2179 | return may_open(&nd->path, 0, open_flag & ~O_TRUNC); |
2180 | } |
2181 | |
2182 | /* |
2183 | * Note that while the flag value (low two bits) for sys_open means: |
2184 | * 00 - read-only |
2185 | * 01 - write-only |
2186 | * 10 - read-write |
2187 | * 11 - special |
2188 | * it is changed into |
2189 | * 00 - no permissions needed |
2190 | * 01 - read-permission |
2191 | * 10 - write-permission |
2192 | * 11 - read-write |
2193 | * for the internal routines (ie open_namei()/follow_link() etc) |
2194 | * This is more logical, and also allows the 00 "no perm needed" |
2195 | * to be used for symlinks (where the permissions are checked |
2196 | * later). |
2197 | * |
2198 | */ |
2199 | static inline int open_to_namei_flags(int flag) |
2200 | { |
2201 | if ((flag+1) & O_ACCMODE) |
2202 | flag++; |
2203 | return flag; |
2204 | } |
2205 | |
2206 | static int open_will_truncate(int flag, struct inode *inode) |
2207 | { |
2208 | /* |
2209 | * We'll never write to the fs underlying |
2210 | * a device file. |
2211 | */ |
2212 | if (special_file(inode->i_mode)) |
2213 | return 0; |
2214 | return (flag & O_TRUNC); |
2215 | } |
2216 | |
2217 | static struct file *finish_open(struct nameidata *nd, |
2218 | int open_flag, int acc_mode) |
2219 | { |
2220 | struct file *filp; |
2221 | int will_truncate; |
2222 | int error; |
2223 | |
2224 | will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); |
2225 | if (will_truncate) { |
2226 | error = mnt_want_write(nd->path.mnt); |
2227 | if (error) |
2228 | goto exit; |
2229 | } |
2230 | error = may_open(&nd->path, acc_mode, open_flag); |
2231 | if (error) { |
2232 | if (will_truncate) |
2233 | mnt_drop_write(nd->path.mnt); |
2234 | goto exit; |
2235 | } |
2236 | filp = nameidata_to_filp(nd); |
2237 | if (!IS_ERR(filp)) { |
2238 | error = ima_file_check(filp, acc_mode); |
2239 | if (error) { |
2240 | fput(filp); |
2241 | filp = ERR_PTR(error); |
2242 | } |
2243 | } |
2244 | if (!IS_ERR(filp)) { |
2245 | if (will_truncate) { |
2246 | error = handle_truncate(filp); |
2247 | if (error) { |
2248 | fput(filp); |
2249 | filp = ERR_PTR(error); |
2250 | } |
2251 | } |
2252 | } |
2253 | /* |
2254 | * It is now safe to drop the mnt write |
2255 | * because the filp has had a write taken |
2256 | * on its behalf. |
2257 | */ |
2258 | if (will_truncate) |
2259 | mnt_drop_write(nd->path.mnt); |
2260 | path_put(&nd->path); |
2261 | return filp; |
2262 | |
2263 | exit: |
2264 | path_put(&nd->path); |
2265 | return ERR_PTR(error); |
2266 | } |
2267 | |
2268 | /* |
2269 | * Handle O_CREAT case for do_filp_open |
2270 | */ |
2271 | static struct file *do_last(struct nameidata *nd, struct path *path, |
2272 | int open_flag, int acc_mode, |
2273 | int mode, const char *pathname) |
2274 | { |
2275 | struct dentry *dir = nd->path.dentry; |
2276 | struct file *filp; |
2277 | int error = -EISDIR; |
2278 | |
2279 | switch (nd->last_type) { |
2280 | case LAST_DOTDOT: |
2281 | follow_dotdot(nd); |
2282 | dir = nd->path.dentry; |
2283 | case LAST_DOT: |
2284 | if (need_reval_dot(dir)) { |
2285 | int status = d_revalidate(nd->path.dentry, nd); |
2286 | if (!status) |
2287 | status = -ESTALE; |
2288 | if (status < 0) { |
2289 | error = status; |
2290 | goto exit; |
2291 | } |
2292 | } |
2293 | /* fallthrough */ |
2294 | case LAST_ROOT: |
2295 | goto exit; |
2296 | case LAST_BIND: |
2297 | audit_inode(pathname, dir); |
2298 | goto ok; |
2299 | } |
2300 | |
2301 | /* trailing slashes? */ |
2302 | if (nd->last.name[nd->last.len]) |
2303 | goto exit; |
2304 | |
2305 | mutex_lock(&dir->d_inode->i_mutex); |
2306 | |
2307 | path->dentry = lookup_hash(nd); |
2308 | path->mnt = nd->path.mnt; |
2309 | |
2310 | error = PTR_ERR(path->dentry); |
2311 | if (IS_ERR(path->dentry)) { |
2312 | mutex_unlock(&dir->d_inode->i_mutex); |
2313 | goto exit; |
2314 | } |
2315 | |
2316 | if (IS_ERR(nd->intent.open.file)) { |
2317 | error = PTR_ERR(nd->intent.open.file); |
2318 | goto exit_mutex_unlock; |
2319 | } |
2320 | |
2321 | /* Negative dentry, just create the file */ |
2322 | if (!path->dentry->d_inode) { |
2323 | /* |
2324 | * This write is needed to ensure that a |
2325 | * ro->rw transition does not occur between |
2326 | * the time when the file is created and when |
2327 | * a permanent write count is taken through |
2328 | * the 'struct file' in nameidata_to_filp(). |
2329 | */ |
2330 | error = mnt_want_write(nd->path.mnt); |
2331 | if (error) |
2332 | goto exit_mutex_unlock; |
2333 | error = __open_namei_create(nd, path, open_flag, mode); |
2334 | if (error) { |
2335 | mnt_drop_write(nd->path.mnt); |
2336 | goto exit; |
2337 | } |
2338 | filp = nameidata_to_filp(nd); |
2339 | mnt_drop_write(nd->path.mnt); |
2340 | path_put(&nd->path); |
2341 | if (!IS_ERR(filp)) { |
2342 | error = ima_file_check(filp, acc_mode); |
2343 | if (error) { |
2344 | fput(filp); |
2345 | filp = ERR_PTR(error); |
2346 | } |
2347 | } |
2348 | return filp; |
2349 | } |
2350 | |
2351 | /* |
2352 | * It already exists. |
2353 | */ |
2354 | mutex_unlock(&dir->d_inode->i_mutex); |
2355 | audit_inode(pathname, path->dentry); |
2356 | |
2357 | error = -EEXIST; |
2358 | if (open_flag & O_EXCL) |
2359 | goto exit_dput; |
2360 | |
2361 | error = follow_managed(path, nd->flags); |
2362 | if (error < 0) |
2363 | goto exit_dput; |
2364 | |
2365 | error = -ENOENT; |
2366 | if (!path->dentry->d_inode) |
2367 | goto exit_dput; |
2368 | |
2369 | if (path->dentry->d_inode->i_op->follow_link) |
2370 | return NULL; |
2371 | |
2372 | path_to_nameidata(path, nd); |
2373 | nd->inode = path->dentry->d_inode; |
2374 | error = -EISDIR; |
2375 | if (S_ISDIR(nd->inode->i_mode)) |
2376 | goto exit; |
2377 | ok: |
2378 | filp = finish_open(nd, open_flag, acc_mode); |
2379 | return filp; |
2380 | |
2381 | exit_mutex_unlock: |
2382 | mutex_unlock(&dir->d_inode->i_mutex); |
2383 | exit_dput: |
2384 | path_put_conditional(path, nd); |
2385 | exit: |
2386 | path_put(&nd->path); |
2387 | return ERR_PTR(error); |
2388 | } |
2389 | |
2390 | /* |
2391 | * Note that the low bits of the passed in "open_flag" |
2392 | * are not the same as in the local variable "flag". See |
2393 | * open_to_namei_flags() for more details. |
2394 | */ |
2395 | struct file *do_filp_open(int dfd, const char *pathname, |
2396 | int open_flag, int mode, int acc_mode) |
2397 | { |
2398 | struct file *filp; |
2399 | struct nameidata nd; |
2400 | int error; |
2401 | struct path path; |
2402 | int count = 0; |
2403 | int flag = open_to_namei_flags(open_flag); |
2404 | int flags; |
2405 | |
2406 | if (!(open_flag & O_CREAT)) |
2407 | mode = 0; |
2408 | |
2409 | /* Must never be set by userspace */ |
2410 | open_flag &= ~FMODE_NONOTIFY; |
2411 | |
2412 | /* |
2413 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only |
2414 | * check for O_DSYNC if the need any syncing at all we enforce it's |
2415 | * always set instead of having to deal with possibly weird behaviour |
2416 | * for malicious applications setting only __O_SYNC. |
2417 | */ |
2418 | if (open_flag & __O_SYNC) |
2419 | open_flag |= O_DSYNC; |
2420 | |
2421 | if (!acc_mode) |
2422 | acc_mode = MAY_OPEN | ACC_MODE(open_flag); |
2423 | |
2424 | /* O_TRUNC implies we need access checks for write permissions */ |
2425 | if (open_flag & O_TRUNC) |
2426 | acc_mode |= MAY_WRITE; |
2427 | |
2428 | /* Allow the LSM permission hook to distinguish append |
2429 | access from general write access. */ |
2430 | if (open_flag & O_APPEND) |
2431 | acc_mode |= MAY_APPEND; |
2432 | |
2433 | flags = LOOKUP_OPEN; |
2434 | if (open_flag & O_CREAT) { |
2435 | flags |= LOOKUP_CREATE; |
2436 | if (open_flag & O_EXCL) |
2437 | flags |= LOOKUP_EXCL; |
2438 | } |
2439 | if (open_flag & O_DIRECTORY) |
2440 | flags |= LOOKUP_DIRECTORY; |
2441 | if (!(open_flag & O_NOFOLLOW)) |
2442 | flags |= LOOKUP_FOLLOW; |
2443 | |
2444 | filp = get_empty_filp(); |
2445 | if (!filp) |
2446 | return ERR_PTR(-ENFILE); |
2447 | |
2448 | filp->f_flags = open_flag; |
2449 | nd.intent.open.file = filp; |
2450 | nd.intent.open.flags = flag; |
2451 | nd.intent.open.create_mode = mode; |
2452 | |
2453 | if (open_flag & O_CREAT) |
2454 | goto creat; |
2455 | |
2456 | /* !O_CREAT, simple open */ |
2457 | error = do_path_lookup(dfd, pathname, flags, &nd); |
2458 | if (unlikely(error)) |
2459 | goto out_filp2; |
2460 | error = -ELOOP; |
2461 | if (!(nd.flags & LOOKUP_FOLLOW)) { |
2462 | if (nd.inode->i_op->follow_link) |
2463 | goto out_path2; |
2464 | } |
2465 | error = -ENOTDIR; |
2466 | if (nd.flags & LOOKUP_DIRECTORY) { |
2467 | if (!nd.inode->i_op->lookup) |
2468 | goto out_path2; |
2469 | } |
2470 | audit_inode(pathname, nd.path.dentry); |
2471 | filp = finish_open(&nd, open_flag, acc_mode); |
2472 | out2: |
2473 | release_open_intent(&nd); |
2474 | return filp; |
2475 | |
2476 | out_path2: |
2477 | path_put(&nd.path); |
2478 | out_filp2: |
2479 | filp = ERR_PTR(error); |
2480 | goto out2; |
2481 | |
2482 | creat: |
2483 | /* OK, have to create the file. Find the parent. */ |
2484 | error = path_init_rcu(dfd, pathname, |
2485 | LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd); |
2486 | if (error) |
2487 | goto out_filp; |
2488 | error = path_walk_rcu(pathname, &nd); |
2489 | path_finish_rcu(&nd); |
2490 | if (unlikely(error == -ECHILD || error == -ESTALE)) { |
2491 | /* slower, locked walk */ |
2492 | if (error == -ESTALE) { |
2493 | reval: |
2494 | flags |= LOOKUP_REVAL; |
2495 | } |
2496 | error = path_init(dfd, pathname, |
2497 | LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd); |
2498 | if (error) |
2499 | goto out_filp; |
2500 | |
2501 | error = path_walk_simple(pathname, &nd); |
2502 | } |
2503 | if (unlikely(error)) |
2504 | goto out_filp; |
2505 | if (unlikely(!audit_dummy_context())) |
2506 | audit_inode(pathname, nd.path.dentry); |
2507 | |
2508 | /* |
2509 | * We have the parent and last component. |
2510 | */ |
2511 | nd.flags = flags; |
2512 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); |
2513 | while (unlikely(!filp)) { /* trailing symlink */ |
2514 | struct path link = path; |
2515 | struct inode *linki = link.dentry->d_inode; |
2516 | void *cookie; |
2517 | error = -ELOOP; |
2518 | if (!(nd.flags & LOOKUP_FOLLOW)) |
2519 | goto exit_dput; |
2520 | if (count++ == 32) |
2521 | goto exit_dput; |
2522 | /* |
2523 | * This is subtle. Instead of calling do_follow_link() we do |
2524 | * the thing by hands. The reason is that this way we have zero |
2525 | * link_count and path_walk() (called from ->follow_link) |
2526 | * honoring LOOKUP_PARENT. After that we have the parent and |
2527 | * last component, i.e. we are in the same situation as after |
2528 | * the first path_walk(). Well, almost - if the last component |
2529 | * is normal we get its copy stored in nd->last.name and we will |
2530 | * have to putname() it when we are done. Procfs-like symlinks |
2531 | * just set LAST_BIND. |
2532 | */ |
2533 | nd.flags |= LOOKUP_PARENT; |
2534 | error = security_inode_follow_link(link.dentry, &nd); |
2535 | if (error) |
2536 | goto exit_dput; |
2537 | error = __do_follow_link(&link, &nd, &cookie); |
2538 | if (unlikely(error)) { |
2539 | if (!IS_ERR(cookie) && linki->i_op->put_link) |
2540 | linki->i_op->put_link(link.dentry, &nd, cookie); |
2541 | /* nd.path had been dropped */ |
2542 | nd.path = link; |
2543 | goto out_path; |
2544 | } |
2545 | nd.flags &= ~LOOKUP_PARENT; |
2546 | filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); |
2547 | if (linki->i_op->put_link) |
2548 | linki->i_op->put_link(link.dentry, &nd, cookie); |
2549 | path_put(&link); |
2550 | } |
2551 | out: |
2552 | if (nd.root.mnt) |
2553 | path_put(&nd.root); |
2554 | if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL)) |
2555 | goto reval; |
2556 | release_open_intent(&nd); |
2557 | return filp; |
2558 | |
2559 | exit_dput: |
2560 | path_put_conditional(&path, &nd); |
2561 | out_path: |
2562 | path_put(&nd.path); |
2563 | out_filp: |
2564 | filp = ERR_PTR(error); |
2565 | goto out; |
2566 | } |
2567 | |
2568 | /** |
2569 | * filp_open - open file and return file pointer |
2570 | * |
2571 | * @filename: path to open |
2572 | * @flags: open flags as per the open(2) second argument |
2573 | * @mode: mode for the new file if O_CREAT is set, else ignored |
2574 | * |
2575 | * This is the helper to open a file from kernelspace if you really |
2576 | * have to. But in generally you should not do this, so please move |
2577 | * along, nothing to see here.. |
2578 | */ |
2579 | struct file *filp_open(const char *filename, int flags, int mode) |
2580 | { |
2581 | return do_filp_open(AT_FDCWD, filename, flags, mode, 0); |
2582 | } |
2583 | EXPORT_SYMBOL(filp_open); |
2584 | |
2585 | /** |
2586 | * lookup_create - lookup a dentry, creating it if it doesn't exist |
2587 | * @nd: nameidata info |
2588 | * @is_dir: directory flag |
2589 | * |
2590 | * Simple function to lookup and return a dentry and create it |
2591 | * if it doesn't exist. Is SMP-safe. |
2592 | * |
2593 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
2594 | */ |
2595 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) |
2596 | { |
2597 | struct dentry *dentry = ERR_PTR(-EEXIST); |
2598 | |
2599 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2600 | /* |
2601 | * Yucky last component or no last component at all? |
2602 | * (foo/., foo/.., /////) |
2603 | */ |
2604 | if (nd->last_type != LAST_NORM) |
2605 | goto fail; |
2606 | nd->flags &= ~LOOKUP_PARENT; |
2607 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
2608 | nd->intent.open.flags = O_EXCL; |
2609 | |
2610 | /* |
2611 | * Do the final lookup. |
2612 | */ |
2613 | dentry = lookup_hash(nd); |
2614 | if (IS_ERR(dentry)) |
2615 | goto fail; |
2616 | |
2617 | if (dentry->d_inode) |
2618 | goto eexist; |
2619 | /* |
2620 | * Special case - lookup gave negative, but... we had foo/bar/ |
2621 | * From the vfs_mknod() POV we just have a negative dentry - |
2622 | * all is fine. Let's be bastards - you had / on the end, you've |
2623 | * been asking for (non-existent) directory. -ENOENT for you. |
2624 | */ |
2625 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
2626 | dput(dentry); |
2627 | dentry = ERR_PTR(-ENOENT); |
2628 | } |
2629 | return dentry; |
2630 | eexist: |
2631 | dput(dentry); |
2632 | dentry = ERR_PTR(-EEXIST); |
2633 | fail: |
2634 | return dentry; |
2635 | } |
2636 | EXPORT_SYMBOL_GPL(lookup_create); |
2637 | |
2638 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) |
2639 | { |
2640 | int error = may_create(dir, dentry); |
2641 | |
2642 | if (error) |
2643 | return error; |
2644 | |
2645 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
2646 | return -EPERM; |
2647 | |
2648 | if (!dir->i_op->mknod) |
2649 | return -EPERM; |
2650 | |
2651 | error = devcgroup_inode_mknod(mode, dev); |
2652 | if (error) |
2653 | return error; |
2654 | |
2655 | error = security_inode_mknod(dir, dentry, mode, dev); |
2656 | if (error) |
2657 | return error; |
2658 | |
2659 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
2660 | if (!error) |
2661 | fsnotify_create(dir, dentry); |
2662 | return error; |
2663 | } |
2664 | |
2665 | static int may_mknod(mode_t mode) |
2666 | { |
2667 | switch (mode & S_IFMT) { |
2668 | case S_IFREG: |
2669 | case S_IFCHR: |
2670 | case S_IFBLK: |
2671 | case S_IFIFO: |
2672 | case S_IFSOCK: |
2673 | case 0: /* zero mode translates to S_IFREG */ |
2674 | return 0; |
2675 | case S_IFDIR: |
2676 | return -EPERM; |
2677 | default: |
2678 | return -EINVAL; |
2679 | } |
2680 | } |
2681 | |
2682 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
2683 | unsigned, dev) |
2684 | { |
2685 | int error; |
2686 | char *tmp; |
2687 | struct dentry *dentry; |
2688 | struct nameidata nd; |
2689 | |
2690 | if (S_ISDIR(mode)) |
2691 | return -EPERM; |
2692 | |
2693 | error = user_path_parent(dfd, filename, &nd, &tmp); |
2694 | if (error) |
2695 | return error; |
2696 | |
2697 | dentry = lookup_create(&nd, 0); |
2698 | if (IS_ERR(dentry)) { |
2699 | error = PTR_ERR(dentry); |
2700 | goto out_unlock; |
2701 | } |
2702 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
2703 | mode &= ~current_umask(); |
2704 | error = may_mknod(mode); |
2705 | if (error) |
2706 | goto out_dput; |
2707 | error = mnt_want_write(nd.path.mnt); |
2708 | if (error) |
2709 | goto out_dput; |
2710 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
2711 | if (error) |
2712 | goto out_drop_write; |
2713 | switch (mode & S_IFMT) { |
2714 | case 0: case S_IFREG: |
2715 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
2716 | break; |
2717 | case S_IFCHR: case S_IFBLK: |
2718 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
2719 | new_decode_dev(dev)); |
2720 | break; |
2721 | case S_IFIFO: case S_IFSOCK: |
2722 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
2723 | break; |
2724 | } |
2725 | out_drop_write: |
2726 | mnt_drop_write(nd.path.mnt); |
2727 | out_dput: |
2728 | dput(dentry); |
2729 | out_unlock: |
2730 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2731 | path_put(&nd.path); |
2732 | putname(tmp); |
2733 | |
2734 | return error; |
2735 | } |
2736 | |
2737 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
2738 | { |
2739 | return sys_mknodat(AT_FDCWD, filename, mode, dev); |
2740 | } |
2741 | |
2742 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
2743 | { |
2744 | int error = may_create(dir, dentry); |
2745 | |
2746 | if (error) |
2747 | return error; |
2748 | |
2749 | if (!dir->i_op->mkdir) |
2750 | return -EPERM; |
2751 | |
2752 | mode &= (S_IRWXUGO|S_ISVTX); |
2753 | error = security_inode_mkdir(dir, dentry, mode); |
2754 | if (error) |
2755 | return error; |
2756 | |
2757 | error = dir->i_op->mkdir(dir, dentry, mode); |
2758 | if (!error) |
2759 | fsnotify_mkdir(dir, dentry); |
2760 | return error; |
2761 | } |
2762 | |
2763 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
2764 | { |
2765 | int error = 0; |
2766 | char * tmp; |
2767 | struct dentry *dentry; |
2768 | struct nameidata nd; |
2769 | |
2770 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
2771 | if (error) |
2772 | goto out_err; |
2773 | |
2774 | dentry = lookup_create(&nd, 1); |
2775 | error = PTR_ERR(dentry); |
2776 | if (IS_ERR(dentry)) |
2777 | goto out_unlock; |
2778 | |
2779 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
2780 | mode &= ~current_umask(); |
2781 | error = mnt_want_write(nd.path.mnt); |
2782 | if (error) |
2783 | goto out_dput; |
2784 | error = security_path_mkdir(&nd.path, dentry, mode); |
2785 | if (error) |
2786 | goto out_drop_write; |
2787 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
2788 | out_drop_write: |
2789 | mnt_drop_write(nd.path.mnt); |
2790 | out_dput: |
2791 | dput(dentry); |
2792 | out_unlock: |
2793 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2794 | path_put(&nd.path); |
2795 | putname(tmp); |
2796 | out_err: |
2797 | return error; |
2798 | } |
2799 | |
2800 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
2801 | { |
2802 | return sys_mkdirat(AT_FDCWD, pathname, mode); |
2803 | } |
2804 | |
2805 | /* |
2806 | * We try to drop the dentry early: we should have |
2807 | * a usage count of 2 if we're the only user of this |
2808 | * dentry, and if that is true (possibly after pruning |
2809 | * the dcache), then we drop the dentry now. |
2810 | * |
2811 | * A low-level filesystem can, if it choses, legally |
2812 | * do a |
2813 | * |
2814 | * if (!d_unhashed(dentry)) |
2815 | * return -EBUSY; |
2816 | * |
2817 | * if it cannot handle the case of removing a directory |
2818 | * that is still in use by something else.. |
2819 | */ |
2820 | void dentry_unhash(struct dentry *dentry) |
2821 | { |
2822 | dget(dentry); |
2823 | shrink_dcache_parent(dentry); |
2824 | spin_lock(&dentry->d_lock); |
2825 | if (dentry->d_count == 2) |
2826 | __d_drop(dentry); |
2827 | spin_unlock(&dentry->d_lock); |
2828 | } |
2829 | |
2830 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
2831 | { |
2832 | int error = may_delete(dir, dentry, 1); |
2833 | |
2834 | if (error) |
2835 | return error; |
2836 | |
2837 | if (!dir->i_op->rmdir) |
2838 | return -EPERM; |
2839 | |
2840 | mutex_lock(&dentry->d_inode->i_mutex); |
2841 | dentry_unhash(dentry); |
2842 | if (d_mountpoint(dentry)) |
2843 | error = -EBUSY; |
2844 | else { |
2845 | error = security_inode_rmdir(dir, dentry); |
2846 | if (!error) { |
2847 | error = dir->i_op->rmdir(dir, dentry); |
2848 | if (!error) { |
2849 | dentry->d_inode->i_flags |= S_DEAD; |
2850 | dont_mount(dentry); |
2851 | } |
2852 | } |
2853 | } |
2854 | mutex_unlock(&dentry->d_inode->i_mutex); |
2855 | if (!error) { |
2856 | d_delete(dentry); |
2857 | } |
2858 | dput(dentry); |
2859 | |
2860 | return error; |
2861 | } |
2862 | |
2863 | static long do_rmdir(int dfd, const char __user *pathname) |
2864 | { |
2865 | int error = 0; |
2866 | char * name; |
2867 | struct dentry *dentry; |
2868 | struct nameidata nd; |
2869 | |
2870 | error = user_path_parent(dfd, pathname, &nd, &name); |
2871 | if (error) |
2872 | return error; |
2873 | |
2874 | switch(nd.last_type) { |
2875 | case LAST_DOTDOT: |
2876 | error = -ENOTEMPTY; |
2877 | goto exit1; |
2878 | case LAST_DOT: |
2879 | error = -EINVAL; |
2880 | goto exit1; |
2881 | case LAST_ROOT: |
2882 | error = -EBUSY; |
2883 | goto exit1; |
2884 | } |
2885 | |
2886 | nd.flags &= ~LOOKUP_PARENT; |
2887 | |
2888 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2889 | dentry = lookup_hash(&nd); |
2890 | error = PTR_ERR(dentry); |
2891 | if (IS_ERR(dentry)) |
2892 | goto exit2; |
2893 | error = mnt_want_write(nd.path.mnt); |
2894 | if (error) |
2895 | goto exit3; |
2896 | error = security_path_rmdir(&nd.path, dentry); |
2897 | if (error) |
2898 | goto exit4; |
2899 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
2900 | exit4: |
2901 | mnt_drop_write(nd.path.mnt); |
2902 | exit3: |
2903 | dput(dentry); |
2904 | exit2: |
2905 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2906 | exit1: |
2907 | path_put(&nd.path); |
2908 | putname(name); |
2909 | return error; |
2910 | } |
2911 | |
2912 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
2913 | { |
2914 | return do_rmdir(AT_FDCWD, pathname); |
2915 | } |
2916 | |
2917 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
2918 | { |
2919 | int error = may_delete(dir, dentry, 0); |
2920 | |
2921 | if (error) |
2922 | return error; |
2923 | |
2924 | if (!dir->i_op->unlink) |
2925 | return -EPERM; |
2926 | |
2927 | mutex_lock(&dentry->d_inode->i_mutex); |
2928 | if (d_mountpoint(dentry)) |
2929 | error = -EBUSY; |
2930 | else { |
2931 | error = security_inode_unlink(dir, dentry); |
2932 | if (!error) { |
2933 | error = dir->i_op->unlink(dir, dentry); |
2934 | if (!error) |
2935 | dont_mount(dentry); |
2936 | } |
2937 | } |
2938 | mutex_unlock(&dentry->d_inode->i_mutex); |
2939 | |
2940 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ |
2941 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { |
2942 | fsnotify_link_count(dentry->d_inode); |
2943 | d_delete(dentry); |
2944 | } |
2945 | |
2946 | return error; |
2947 | } |
2948 | |
2949 | /* |
2950 | * Make sure that the actual truncation of the file will occur outside its |
2951 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
2952 | * writeout happening, and we don't want to prevent access to the directory |
2953 | * while waiting on the I/O. |
2954 | */ |
2955 | static long do_unlinkat(int dfd, const char __user *pathname) |
2956 | { |
2957 | int error; |
2958 | char *name; |
2959 | struct dentry *dentry; |
2960 | struct nameidata nd; |
2961 | struct inode *inode = NULL; |
2962 | |
2963 | error = user_path_parent(dfd, pathname, &nd, &name); |
2964 | if (error) |
2965 | return error; |
2966 | |
2967 | error = -EISDIR; |
2968 | if (nd.last_type != LAST_NORM) |
2969 | goto exit1; |
2970 | |
2971 | nd.flags &= ~LOOKUP_PARENT; |
2972 | |
2973 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
2974 | dentry = lookup_hash(&nd); |
2975 | error = PTR_ERR(dentry); |
2976 | if (!IS_ERR(dentry)) { |
2977 | /* Why not before? Because we want correct error value */ |
2978 | if (nd.last.name[nd.last.len]) |
2979 | goto slashes; |
2980 | inode = dentry->d_inode; |
2981 | if (inode) |
2982 | ihold(inode); |
2983 | error = mnt_want_write(nd.path.mnt); |
2984 | if (error) |
2985 | goto exit2; |
2986 | error = security_path_unlink(&nd.path, dentry); |
2987 | if (error) |
2988 | goto exit3; |
2989 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
2990 | exit3: |
2991 | mnt_drop_write(nd.path.mnt); |
2992 | exit2: |
2993 | dput(dentry); |
2994 | } |
2995 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
2996 | if (inode) |
2997 | iput(inode); /* truncate the inode here */ |
2998 | exit1: |
2999 | path_put(&nd.path); |
3000 | putname(name); |
3001 | return error; |
3002 | |
3003 | slashes: |
3004 | error = !dentry->d_inode ? -ENOENT : |
3005 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; |
3006 | goto exit2; |
3007 | } |
3008 | |
3009 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
3010 | { |
3011 | if ((flag & ~AT_REMOVEDIR) != 0) |
3012 | return -EINVAL; |
3013 | |
3014 | if (flag & AT_REMOVEDIR) |
3015 | return do_rmdir(dfd, pathname); |
3016 | |
3017 | return do_unlinkat(dfd, pathname); |
3018 | } |
3019 | |
3020 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
3021 | { |
3022 | return do_unlinkat(AT_FDCWD, pathname); |
3023 | } |
3024 | |
3025 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
3026 | { |
3027 | int error = may_create(dir, dentry); |
3028 | |
3029 | if (error) |
3030 | return error; |
3031 | |
3032 | if (!dir->i_op->symlink) |
3033 | return -EPERM; |
3034 | |
3035 | error = security_inode_symlink(dir, dentry, oldname); |
3036 | if (error) |
3037 | return error; |
3038 | |
3039 | error = dir->i_op->symlink(dir, dentry, oldname); |
3040 | if (!error) |
3041 | fsnotify_create(dir, dentry); |
3042 | return error; |
3043 | } |
3044 | |
3045 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
3046 | int, newdfd, const char __user *, newname) |
3047 | { |
3048 | int error; |
3049 | char *from; |
3050 | char *to; |
3051 | struct dentry *dentry; |
3052 | struct nameidata nd; |
3053 | |
3054 | from = getname(oldname); |
3055 | if (IS_ERR(from)) |
3056 | return PTR_ERR(from); |
3057 | |
3058 | error = user_path_parent(newdfd, newname, &nd, &to); |
3059 | if (error) |
3060 | goto out_putname; |
3061 | |
3062 | dentry = lookup_create(&nd, 0); |
3063 | error = PTR_ERR(dentry); |
3064 | if (IS_ERR(dentry)) |
3065 | goto out_unlock; |
3066 | |
3067 | error = mnt_want_write(nd.path.mnt); |
3068 | if (error) |
3069 | goto out_dput; |
3070 | error = security_path_symlink(&nd.path, dentry, from); |
3071 | if (error) |
3072 | goto out_drop_write; |
3073 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
3074 | out_drop_write: |
3075 | mnt_drop_write(nd.path.mnt); |
3076 | out_dput: |
3077 | dput(dentry); |
3078 | out_unlock: |
3079 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
3080 | path_put(&nd.path); |
3081 | putname(to); |
3082 | out_putname: |
3083 | putname(from); |
3084 | return error; |
3085 | } |
3086 | |
3087 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
3088 | { |
3089 | return sys_symlinkat(oldname, AT_FDCWD, newname); |
3090 | } |
3091 | |
3092 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
3093 | { |
3094 | struct inode *inode = old_dentry->d_inode; |
3095 | int error; |
3096 | |
3097 | if (!inode) |
3098 | return -ENOENT; |
3099 | |
3100 | error = may_create(dir, new_dentry); |
3101 | if (error) |
3102 | return error; |
3103 | |
3104 | if (dir->i_sb != inode->i_sb) |
3105 | return -EXDEV; |
3106 | |
3107 | /* |
3108 | * A link to an append-only or immutable file cannot be created. |
3109 | */ |
3110 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) |
3111 | return -EPERM; |
3112 | if (!dir->i_op->link) |
3113 | return -EPERM; |
3114 | if (S_ISDIR(inode->i_mode)) |
3115 | return -EPERM; |
3116 | |
3117 | error = security_inode_link(old_dentry, dir, new_dentry); |
3118 | if (error) |
3119 | return error; |
3120 | |
3121 | mutex_lock(&inode->i_mutex); |
3122 | error = dir->i_op->link(old_dentry, dir, new_dentry); |
3123 | mutex_unlock(&inode->i_mutex); |
3124 | if (!error) |
3125 | fsnotify_link(dir, inode, new_dentry); |
3126 | return error; |
3127 | } |
3128 | |
3129 | /* |
3130 | * Hardlinks are often used in delicate situations. We avoid |
3131 | * security-related surprises by not following symlinks on the |
3132 | * newname. --KAB |
3133 | * |
3134 | * We don't follow them on the oldname either to be compatible |
3135 | * with linux 2.0, and to avoid hard-linking to directories |
3136 | * and other special files. --ADM |
3137 | */ |
3138 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
3139 | int, newdfd, const char __user *, newname, int, flags) |
3140 | { |
3141 | struct dentry *new_dentry; |
3142 | struct nameidata nd; |
3143 | struct path old_path; |
3144 | int error; |
3145 | char *to; |
3146 | |
3147 | if ((flags & ~AT_SYMLINK_FOLLOW) != 0) |
3148 | return -EINVAL; |
3149 | |
3150 | error = user_path_at(olddfd, oldname, |
3151 | flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, |
3152 | &old_path); |
3153 | if (error) |
3154 | return error; |
3155 | |
3156 | error = user_path_parent(newdfd, newname, &nd, &to); |
3157 | if (error) |
3158 | goto out; |
3159 | error = -EXDEV; |
3160 | if (old_path.mnt != nd.path.mnt) |
3161 | goto out_release; |
3162 | new_dentry = lookup_create(&nd, 0); |
3163 | error = PTR_ERR(new_dentry); |
3164 | if (IS_ERR(new_dentry)) |
3165 | goto out_unlock; |
3166 | error = mnt_want_write(nd.path.mnt); |
3167 | if (error) |
3168 | goto out_dput; |
3169 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
3170 | if (error) |
3171 | goto out_drop_write; |
3172 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
3173 | out_drop_write: |
3174 | mnt_drop_write(nd.path.mnt); |
3175 | out_dput: |
3176 | dput(new_dentry); |
3177 | out_unlock: |
3178 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
3179 | out_release: |
3180 | path_put(&nd.path); |
3181 | putname(to); |
3182 | out: |
3183 | path_put(&old_path); |
3184 | |
3185 | return error; |
3186 | } |
3187 | |
3188 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
3189 | { |
3190 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
3191 | } |
3192 | |
3193 | /* |
3194 | * The worst of all namespace operations - renaming directory. "Perverted" |
3195 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... |
3196 | * Problems: |
3197 | * a) we can get into loop creation. Check is done in is_subdir(). |
3198 | * b) race potential - two innocent renames can create a loop together. |
3199 | * That's where 4.4 screws up. Current fix: serialization on |
3200 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
3201 | * story. |
3202 | * c) we have to lock _three_ objects - parents and victim (if it exists). |
3203 | * And that - after we got ->i_mutex on parents (until then we don't know |
3204 | * whether the target exists). Solution: try to be smart with locking |
3205 | * order for inodes. We rely on the fact that tree topology may change |
3206 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
3207 | * move will be locked. Thus we can rank directories by the tree |
3208 | * (ancestors first) and rank all non-directories after them. |
3209 | * That works since everybody except rename does "lock parent, lookup, |
3210 | * lock child" and rename is under ->s_vfs_rename_mutex. |
3211 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
3212 | * has no more than 1 dentry. If "hybrid" objects will ever appear, |
3213 | * we'd better make sure that there's no link(2) for them. |
3214 | * d) some filesystems don't support opened-but-unlinked directories, |
3215 | * either because of layout or because they are not ready to deal with |
3216 | * all cases correctly. The latter will be fixed (taking this sort of |
3217 | * stuff into VFS), but the former is not going away. Solution: the same |
3218 | * trick as in rmdir(). |
3219 | * e) conversion from fhandle to dentry may come in the wrong moment - when |
3220 | * we are removing the target. Solution: we will have to grab ->i_mutex |
3221 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
3222 | * ->i_mutex on parents, which works but leads to some truly excessive |
3223 | * locking]. |
3224 | */ |
3225 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
3226 | struct inode *new_dir, struct dentry *new_dentry) |
3227 | { |
3228 | int error = 0; |
3229 | struct inode *target; |
3230 | |
3231 | /* |
3232 | * If we are going to change the parent - check write permissions, |
3233 | * we'll need to flip '..'. |
3234 | */ |
3235 | if (new_dir != old_dir) { |
3236 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
3237 | if (error) |
3238 | return error; |
3239 | } |
3240 | |
3241 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
3242 | if (error) |
3243 | return error; |
3244 | |
3245 | target = new_dentry->d_inode; |
3246 | if (target) |
3247 | mutex_lock(&target->i_mutex); |
3248 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3249 | error = -EBUSY; |
3250 | else { |
3251 | if (target) |
3252 | dentry_unhash(new_dentry); |
3253 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3254 | } |
3255 | if (target) { |
3256 | if (!error) { |
3257 | target->i_flags |= S_DEAD; |
3258 | dont_mount(new_dentry); |
3259 | } |
3260 | mutex_unlock(&target->i_mutex); |
3261 | if (d_unhashed(new_dentry)) |
3262 | d_rehash(new_dentry); |
3263 | dput(new_dentry); |
3264 | } |
3265 | if (!error) |
3266 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3267 | d_move(old_dentry,new_dentry); |
3268 | return error; |
3269 | } |
3270 | |
3271 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
3272 | struct inode *new_dir, struct dentry *new_dentry) |
3273 | { |
3274 | struct inode *target; |
3275 | int error; |
3276 | |
3277 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
3278 | if (error) |
3279 | return error; |
3280 | |
3281 | dget(new_dentry); |
3282 | target = new_dentry->d_inode; |
3283 | if (target) |
3284 | mutex_lock(&target->i_mutex); |
3285 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
3286 | error = -EBUSY; |
3287 | else |
3288 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
3289 | if (!error) { |
3290 | if (target) |
3291 | dont_mount(new_dentry); |
3292 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
3293 | d_move(old_dentry, new_dentry); |
3294 | } |
3295 | if (target) |
3296 | mutex_unlock(&target->i_mutex); |
3297 | dput(new_dentry); |
3298 | return error; |
3299 | } |
3300 | |
3301 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
3302 | struct inode *new_dir, struct dentry *new_dentry) |
3303 | { |
3304 | int error; |
3305 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); |
3306 | const unsigned char *old_name; |
3307 | |
3308 | if (old_dentry->d_inode == new_dentry->d_inode) |
3309 | return 0; |
3310 | |
3311 | error = may_delete(old_dir, old_dentry, is_dir); |
3312 | if (error) |
3313 | return error; |
3314 | |
3315 | if (!new_dentry->d_inode) |
3316 | error = may_create(new_dir, new_dentry); |
3317 | else |
3318 | error = may_delete(new_dir, new_dentry, is_dir); |
3319 | if (error) |
3320 | return error; |
3321 | |
3322 | if (!old_dir->i_op->rename) |
3323 | return -EPERM; |
3324 | |
3325 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
3326 | |
3327 | if (is_dir) |
3328 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); |
3329 | else |
3330 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); |
3331 | if (!error) |
3332 | fsnotify_move(old_dir, new_dir, old_name, is_dir, |
3333 | new_dentry->d_inode, old_dentry); |
3334 | fsnotify_oldname_free(old_name); |
3335 | |
3336 | return error; |
3337 | } |
3338 | |
3339 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
3340 | int, newdfd, const char __user *, newname) |
3341 | { |
3342 | struct dentry *old_dir, *new_dir; |
3343 | struct dentry *old_dentry, *new_dentry; |
3344 | struct dentry *trap; |
3345 | struct nameidata oldnd, newnd; |
3346 | char *from; |
3347 | char *to; |
3348 | int error; |
3349 | |
3350 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
3351 | if (error) |
3352 | goto exit; |
3353 | |
3354 | error = user_path_parent(newdfd, newname, &newnd, &to); |
3355 | if (error) |
3356 | goto exit1; |
3357 | |
3358 | error = -EXDEV; |
3359 | if (oldnd.path.mnt != newnd.path.mnt) |
3360 | goto exit2; |
3361 | |
3362 | old_dir = oldnd.path.dentry; |
3363 | error = -EBUSY; |
3364 | if (oldnd.last_type != LAST_NORM) |
3365 | goto exit2; |
3366 | |
3367 | new_dir = newnd.path.dentry; |
3368 | if (newnd.last_type != LAST_NORM) |
3369 | goto exit2; |
3370 | |
3371 | oldnd.flags &= ~LOOKUP_PARENT; |
3372 | newnd.flags &= ~LOOKUP_PARENT; |
3373 | newnd.flags |= LOOKUP_RENAME_TARGET; |
3374 | |
3375 | trap = lock_rename(new_dir, old_dir); |
3376 | |
3377 | old_dentry = lookup_hash(&oldnd); |
3378 | error = PTR_ERR(old_dentry); |
3379 | if (IS_ERR(old_dentry)) |
3380 | goto exit3; |
3381 | /* source must exist */ |
3382 | error = -ENOENT; |
3383 | if (!old_dentry->d_inode) |
3384 | goto exit4; |
3385 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
3386 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { |
3387 | error = -ENOTDIR; |
3388 | if (oldnd.last.name[oldnd.last.len]) |
3389 | goto exit4; |
3390 | if (newnd.last.name[newnd.last.len]) |
3391 | goto exit4; |
3392 | } |
3393 | /* source should not be ancestor of target */ |
3394 | error = -EINVAL; |
3395 | if (old_dentry == trap) |
3396 | goto exit4; |
3397 | new_dentry = lookup_hash(&newnd); |
3398 | error = PTR_ERR(new_dentry); |
3399 | if (IS_ERR(new_dentry)) |
3400 | goto exit4; |
3401 | /* target should not be an ancestor of source */ |
3402 | error = -ENOTEMPTY; |
3403 | if (new_dentry == trap) |
3404 | goto exit5; |
3405 | |
3406 | error = mnt_want_write(oldnd.path.mnt); |
3407 | if (error) |
3408 | goto exit5; |
3409 | error = security_path_rename(&oldnd.path, old_dentry, |
3410 | &newnd.path, new_dentry); |
3411 | if (error) |
3412 | goto exit6; |
3413 | error = vfs_rename(old_dir->d_inode, old_dentry, |
3414 | new_dir->d_inode, new_dentry); |
3415 | exit6: |
3416 | mnt_drop_write(oldnd.path.mnt); |
3417 | exit5: |
3418 | dput(new_dentry); |
3419 | exit4: |
3420 | dput(old_dentry); |
3421 | exit3: |
3422 | unlock_rename(new_dir, old_dir); |
3423 | exit2: |
3424 | path_put(&newnd.path); |
3425 | putname(to); |
3426 | exit1: |
3427 | path_put(&oldnd.path); |
3428 | putname(from); |
3429 | exit: |
3430 | return error; |
3431 | } |
3432 | |
3433 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
3434 | { |
3435 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); |
3436 | } |
3437 | |
3438 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
3439 | { |
3440 | int len; |
3441 | |
3442 | len = PTR_ERR(link); |
3443 | if (IS_ERR(link)) |
3444 | goto out; |
3445 | |
3446 | len = strlen(link); |
3447 | if (len > (unsigned) buflen) |
3448 | len = buflen; |
3449 | if (copy_to_user(buffer, link, len)) |
3450 | len = -EFAULT; |
3451 | out: |
3452 | return len; |
3453 | } |
3454 | |
3455 | /* |
3456 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that |
3457 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not |
3458 | * using) it for any given inode is up to filesystem. |
3459 | */ |
3460 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
3461 | { |
3462 | struct nameidata nd; |
3463 | void *cookie; |
3464 | int res; |
3465 | |
3466 | nd.depth = 0; |
3467 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
3468 | if (IS_ERR(cookie)) |
3469 | return PTR_ERR(cookie); |
3470 | |
3471 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); |
3472 | if (dentry->d_inode->i_op->put_link) |
3473 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); |
3474 | return res; |
3475 | } |
3476 | |
3477 | int vfs_follow_link(struct nameidata *nd, const char *link) |
3478 | { |
3479 | return __vfs_follow_link(nd, link); |
3480 | } |
3481 | |
3482 | /* get the link contents into pagecache */ |
3483 | static char *page_getlink(struct dentry * dentry, struct page **ppage) |
3484 | { |
3485 | char *kaddr; |
3486 | struct page *page; |
3487 | struct address_space *mapping = dentry->d_inode->i_mapping; |
3488 | page = read_mapping_page(mapping, 0, NULL); |
3489 | if (IS_ERR(page)) |
3490 | return (char*)page; |
3491 | *ppage = page; |
3492 | kaddr = kmap(page); |
3493 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); |
3494 | return kaddr; |
3495 | } |
3496 | |
3497 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
3498 | { |
3499 | struct page *page = NULL; |
3500 | char *s = page_getlink(dentry, &page); |
3501 | int res = vfs_readlink(dentry,buffer,buflen,s); |
3502 | if (page) { |
3503 | kunmap(page); |
3504 | page_cache_release(page); |
3505 | } |
3506 | return res; |
3507 | } |
3508 | |
3509 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
3510 | { |
3511 | struct page *page = NULL; |
3512 | nd_set_link(nd, page_getlink(dentry, &page)); |
3513 | return page; |
3514 | } |
3515 | |
3516 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
3517 | { |
3518 | struct page *page = cookie; |
3519 | |
3520 | if (page) { |
3521 | kunmap(page); |
3522 | page_cache_release(page); |
3523 | } |
3524 | } |
3525 | |
3526 | /* |
3527 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS |
3528 | */ |
3529 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) |
3530 | { |
3531 | struct address_space *mapping = inode->i_mapping; |
3532 | struct page *page; |
3533 | void *fsdata; |
3534 | int err; |
3535 | char *kaddr; |
3536 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
3537 | if (nofs) |
3538 | flags |= AOP_FLAG_NOFS; |
3539 | |
3540 | retry: |
3541 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
3542 | flags, &page, &fsdata); |
3543 | if (err) |
3544 | goto fail; |
3545 | |
3546 | kaddr = kmap_atomic(page, KM_USER0); |
3547 | memcpy(kaddr, symname, len-1); |
3548 | kunmap_atomic(kaddr, KM_USER0); |
3549 | |
3550 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, |
3551 | page, fsdata); |
3552 | if (err < 0) |
3553 | goto fail; |
3554 | if (err < len-1) |
3555 | goto retry; |
3556 | |
3557 | mark_inode_dirty(inode); |
3558 | return 0; |
3559 | fail: |
3560 | return err; |
3561 | } |
3562 | |
3563 | int page_symlink(struct inode *inode, const char *symname, int len) |
3564 | { |
3565 | return __page_symlink(inode, symname, len, |
3566 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
3567 | } |
3568 | |
3569 | const struct inode_operations page_symlink_inode_operations = { |
3570 | .readlink = generic_readlink, |
3571 | .follow_link = page_follow_link_light, |
3572 | .put_link = page_put_link, |
3573 | }; |
3574 | |
3575 | EXPORT_SYMBOL(user_path_at); |
3576 | EXPORT_SYMBOL(follow_down_one); |
3577 | EXPORT_SYMBOL(follow_down); |
3578 | EXPORT_SYMBOL(follow_up); |
3579 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ |
3580 | EXPORT_SYMBOL(getname); |
3581 | EXPORT_SYMBOL(lock_rename); |
3582 | EXPORT_SYMBOL(lookup_one_len); |
3583 | EXPORT_SYMBOL(page_follow_link_light); |
3584 | EXPORT_SYMBOL(page_put_link); |
3585 | EXPORT_SYMBOL(page_readlink); |
3586 | EXPORT_SYMBOL(__page_symlink); |
3587 | EXPORT_SYMBOL(page_symlink); |
3588 | EXPORT_SYMBOL(page_symlink_inode_operations); |
3589 | EXPORT_SYMBOL(path_lookup); |
3590 | EXPORT_SYMBOL(kern_path); |
3591 | EXPORT_SYMBOL(vfs_path_lookup); |
3592 | EXPORT_SYMBOL(inode_permission); |
3593 | EXPORT_SYMBOL(file_permission); |
3594 | EXPORT_SYMBOL(unlock_rename); |
3595 | EXPORT_SYMBOL(vfs_create); |
3596 | EXPORT_SYMBOL(vfs_follow_link); |
3597 | EXPORT_SYMBOL(vfs_link); |
3598 | EXPORT_SYMBOL(vfs_mkdir); |
3599 | EXPORT_SYMBOL(vfs_mknod); |
3600 | EXPORT_SYMBOL(generic_permission); |
3601 | EXPORT_SYMBOL(vfs_readlink); |
3602 | EXPORT_SYMBOL(vfs_rename); |
3603 | EXPORT_SYMBOL(vfs_rmdir); |
3604 | EXPORT_SYMBOL(vfs_symlink); |
3605 | EXPORT_SYMBOL(vfs_unlink); |
3606 | EXPORT_SYMBOL(dentry_unhash); |
3607 | EXPORT_SYMBOL(generic_readlink); |
3608 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9