ia64/linux-2.6.18-xen.hg

view fs/namei.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
1 /*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
8 * Some corrections by tytso.
9 */
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 */
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/fs.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/namei.h>
35 #include <asm/namei.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105 /*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
109 */
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
118 static int do_getname(const char __user *filename, char *page)
119 {
120 int retval;
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138 }
140 char * getname(const char __user * filename)
141 {
142 char *tmp, *result;
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157 }
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
161 {
162 if (unlikely(!audit_dummy_context()))
163 audit_putname(name);
164 else
165 __putname(name);
166 }
167 EXPORT_SYMBOL(putname);
168 #endif
171 /**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184 {
185 umode_t mode = inode->i_mode;
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
225 return -EACCES;
226 }
228 int permission(struct inode *inode, int mask, struct nameidata *nd)
229 {
230 umode_t mode = inode->i_mode;
231 int retval, submask;
233 if (mask & MAY_WRITE) {
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
250 /*
251 * MAY_EXEC on regular files requires special handling: We override
252 * filesystem execute permissions if the mode bits aren't set.
253 */
254 if ((mask & MAY_EXEC) && S_ISREG(mode) && !(mode & S_IXUGO))
255 return -EACCES;
257 /* Ordinary permission routines do not understand MAY_APPEND. */
258 submask = mask & ~MAY_APPEND;
259 if (inode->i_op && inode->i_op->permission)
260 retval = inode->i_op->permission(inode, submask, nd);
261 else
262 retval = generic_permission(inode, submask, NULL);
263 if (retval)
264 return retval;
266 return security_inode_permission(inode, mask, nd);
267 }
269 /**
270 * vfs_permission - check for access rights to a given path
271 * @nd: lookup result that describes the path
272 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
273 *
274 * Used to check for read/write/execute permissions on a path.
275 * We use "fsuid" for this, letting us set arbitrary permissions
276 * for filesystem access without changing the "normal" uids which
277 * are used for other things.
278 */
279 int vfs_permission(struct nameidata *nd, int mask)
280 {
281 return permission(nd->dentry->d_inode, mask, nd);
282 }
284 /**
285 * file_permission - check for additional access rights to a given file
286 * @file: file to check access rights for
287 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
288 *
289 * Used to check for read/write/execute permissions on an already opened
290 * file.
291 *
292 * Note:
293 * Do not use this function in new code. All access checks should
294 * be done using vfs_permission().
295 */
296 int file_permission(struct file *file, int mask)
297 {
298 return permission(file->f_dentry->d_inode, mask, NULL);
299 }
301 /*
302 * get_write_access() gets write permission for a file.
303 * put_write_access() releases this write permission.
304 * This is used for regular files.
305 * We cannot support write (and maybe mmap read-write shared) accesses and
306 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
307 * can have the following values:
308 * 0: no writers, no VM_DENYWRITE mappings
309 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
310 * > 0: (i_writecount) users are writing to the file.
311 *
312 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
313 * except for the cases where we don't hold i_writecount yet. Then we need to
314 * use {get,deny}_write_access() - these functions check the sign and refuse
315 * to do the change if sign is wrong. Exclusion between them is provided by
316 * the inode->i_lock spinlock.
317 */
319 int get_write_access(struct inode * inode)
320 {
321 spin_lock(&inode->i_lock);
322 if (atomic_read(&inode->i_writecount) < 0) {
323 spin_unlock(&inode->i_lock);
324 return -ETXTBSY;
325 }
326 atomic_inc(&inode->i_writecount);
327 spin_unlock(&inode->i_lock);
329 return 0;
330 }
332 int deny_write_access(struct file * file)
333 {
334 struct inode *inode = file->f_dentry->d_inode;
336 spin_lock(&inode->i_lock);
337 if (atomic_read(&inode->i_writecount) > 0) {
338 spin_unlock(&inode->i_lock);
339 return -ETXTBSY;
340 }
341 atomic_dec(&inode->i_writecount);
342 spin_unlock(&inode->i_lock);
344 return 0;
345 }
347 void path_release(struct nameidata *nd)
348 {
349 dput(nd->dentry);
350 mntput(nd->mnt);
351 }
353 /*
354 * umount() mustn't call path_release()/mntput() as that would clear
355 * mnt_expiry_mark
356 */
357 void path_release_on_umount(struct nameidata *nd)
358 {
359 dput(nd->dentry);
360 mntput_no_expire(nd->mnt);
361 }
363 /**
364 * release_open_intent - free up open intent resources
365 * @nd: pointer to nameidata
366 */
367 void release_open_intent(struct nameidata *nd)
368 {
369 if (nd->intent.open.file->f_dentry == NULL)
370 put_filp(nd->intent.open.file);
371 else
372 fput(nd->intent.open.file);
373 }
375 /*
376 * Internal lookup() using the new generic dcache.
377 * SMP-safe
378 */
379 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
380 {
381 struct dentry * dentry = __d_lookup(parent, name);
383 /* lockess __d_lookup may fail due to concurrent d_move()
384 * in some unrelated directory, so try with d_lookup
385 */
386 if (!dentry)
387 dentry = d_lookup(parent, name);
389 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
390 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
391 dput(dentry);
392 dentry = NULL;
393 }
394 }
395 return dentry;
396 }
398 /*
399 * Short-cut version of permission(), for calling by
400 * path_walk(), when dcache lock is held. Combines parts
401 * of permission() and generic_permission(), and tests ONLY for
402 * MAY_EXEC permission.
403 *
404 * If appropriate, check DAC only. If not appropriate, or
405 * short-cut DAC fails, then call permission() to do more
406 * complete permission check.
407 */
408 static int exec_permission_lite(struct inode *inode,
409 struct nameidata *nd)
410 {
411 umode_t mode = inode->i_mode;
413 if (inode->i_op && inode->i_op->permission)
414 return -EAGAIN;
416 if (current->fsuid == inode->i_uid)
417 mode >>= 6;
418 else if (in_group_p(inode->i_gid))
419 mode >>= 3;
421 if (mode & MAY_EXEC)
422 goto ok;
424 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
425 goto ok;
427 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
428 goto ok;
430 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
431 goto ok;
433 return -EACCES;
434 ok:
435 return security_inode_permission(inode, MAY_EXEC, nd);
436 }
438 /*
439 * This is called when everything else fails, and we actually have
440 * to go to the low-level filesystem to find out what we should do..
441 *
442 * We get the directory semaphore, and after getting that we also
443 * make sure that nobody added the entry to the dcache in the meantime..
444 * SMP-safe
445 */
446 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
447 {
448 struct dentry * result;
449 struct inode *dir = parent->d_inode;
451 mutex_lock(&dir->i_mutex);
452 /*
453 * First re-do the cached lookup just in case it was created
454 * while we waited for the directory semaphore..
455 *
456 * FIXME! This could use version numbering or similar to
457 * avoid unnecessary cache lookups.
458 *
459 * The "dcache_lock" is purely to protect the RCU list walker
460 * from concurrent renames at this point (we mustn't get false
461 * negatives from the RCU list walk here, unlike the optimistic
462 * fast walk).
463 *
464 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
465 */
466 result = d_lookup(parent, name);
467 if (!result) {
468 struct dentry * dentry = d_alloc(parent, name);
469 result = ERR_PTR(-ENOMEM);
470 if (dentry) {
471 result = dir->i_op->lookup(dir, dentry, nd);
472 if (result)
473 dput(dentry);
474 else
475 result = dentry;
476 }
477 mutex_unlock(&dir->i_mutex);
478 return result;
479 }
481 /*
482 * Uhhuh! Nasty case: the cache was re-populated while
483 * we waited on the semaphore. Need to revalidate.
484 */
485 mutex_unlock(&dir->i_mutex);
486 if (result->d_op && result->d_op->d_revalidate) {
487 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
488 dput(result);
489 result = ERR_PTR(-ENOENT);
490 }
491 }
492 return result;
493 }
495 static int __emul_lookup_dentry(const char *, struct nameidata *);
497 /* SMP-safe */
498 static __always_inline int
499 walk_init_root(const char *name, struct nameidata *nd)
500 {
501 read_lock(&current->fs->lock);
502 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
503 nd->mnt = mntget(current->fs->altrootmnt);
504 nd->dentry = dget(current->fs->altroot);
505 read_unlock(&current->fs->lock);
506 if (__emul_lookup_dentry(name,nd))
507 return 0;
508 read_lock(&current->fs->lock);
509 }
510 nd->mnt = mntget(current->fs->rootmnt);
511 nd->dentry = dget(current->fs->root);
512 read_unlock(&current->fs->lock);
513 return 1;
514 }
516 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
517 {
518 int res = 0;
519 char *name;
520 if (IS_ERR(link))
521 goto fail;
523 if (*link == '/') {
524 path_release(nd);
525 if (!walk_init_root(link, nd))
526 /* weird __emul_prefix() stuff did it */
527 goto out;
528 }
529 res = link_path_walk(link, nd);
530 out:
531 if (nd->depth || res || nd->last_type!=LAST_NORM)
532 return res;
533 /*
534 * If it is an iterative symlinks resolution in open_namei() we
535 * have to copy the last component. And all that crap because of
536 * bloody create() on broken symlinks. Furrfu...
537 */
538 name = __getname();
539 if (unlikely(!name)) {
540 path_release(nd);
541 return -ENOMEM;
542 }
543 strcpy(name, nd->last.name);
544 nd->last.name = name;
545 return 0;
546 fail:
547 path_release(nd);
548 return PTR_ERR(link);
549 }
551 struct path {
552 struct vfsmount *mnt;
553 struct dentry *dentry;
554 };
556 static inline void dput_path(struct path *path, struct nameidata *nd)
557 {
558 dput(path->dentry);
559 if (path->mnt != nd->mnt)
560 mntput(path->mnt);
561 }
563 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
564 {
565 dput(nd->dentry);
566 if (nd->mnt != path->mnt)
567 mntput(nd->mnt);
568 nd->mnt = path->mnt;
569 nd->dentry = path->dentry;
570 }
572 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
573 {
574 int error;
575 void *cookie;
576 struct dentry *dentry = path->dentry;
578 touch_atime(path->mnt, dentry);
579 nd_set_link(nd, NULL);
581 if (path->mnt != nd->mnt) {
582 path_to_nameidata(path, nd);
583 dget(dentry);
584 }
585 mntget(path->mnt);
586 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
587 error = PTR_ERR(cookie);
588 if (!IS_ERR(cookie)) {
589 char *s = nd_get_link(nd);
590 error = 0;
591 if (s)
592 error = __vfs_follow_link(nd, s);
593 if (dentry->d_inode->i_op->put_link)
594 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
595 }
596 dput(dentry);
597 mntput(path->mnt);
599 return error;
600 }
602 /*
603 * This limits recursive symlink follows to 8, while
604 * limiting consecutive symlinks to 40.
605 *
606 * Without that kind of total limit, nasty chains of consecutive
607 * symlinks can cause almost arbitrarily long lookups.
608 */
609 static inline int do_follow_link(struct path *path, struct nameidata *nd)
610 {
611 int err = -ELOOP;
612 if (current->link_count >= MAX_NESTED_LINKS)
613 goto loop;
614 if (current->total_link_count >= 40)
615 goto loop;
616 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
617 cond_resched();
618 err = security_inode_follow_link(path->dentry, nd);
619 if (err)
620 goto loop;
621 current->link_count++;
622 current->total_link_count++;
623 nd->depth++;
624 err = __do_follow_link(path, nd);
625 current->link_count--;
626 nd->depth--;
627 return err;
628 loop:
629 dput_path(path, nd);
630 path_release(nd);
631 return err;
632 }
634 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
635 {
636 struct vfsmount *parent;
637 struct dentry *mountpoint;
638 spin_lock(&vfsmount_lock);
639 parent=(*mnt)->mnt_parent;
640 if (parent == *mnt) {
641 spin_unlock(&vfsmount_lock);
642 return 0;
643 }
644 mntget(parent);
645 mountpoint=dget((*mnt)->mnt_mountpoint);
646 spin_unlock(&vfsmount_lock);
647 dput(*dentry);
648 *dentry = mountpoint;
649 mntput(*mnt);
650 *mnt = parent;
651 return 1;
652 }
654 /* no need for dcache_lock, as serialization is taken care in
655 * namespace.c
656 */
657 static int __follow_mount(struct path *path)
658 {
659 int res = 0;
660 while (d_mountpoint(path->dentry)) {
661 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
662 if (!mounted)
663 break;
664 dput(path->dentry);
665 if (res)
666 mntput(path->mnt);
667 path->mnt = mounted;
668 path->dentry = dget(mounted->mnt_root);
669 res = 1;
670 }
671 return res;
672 }
674 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
675 {
676 while (d_mountpoint(*dentry)) {
677 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
678 if (!mounted)
679 break;
680 dput(*dentry);
681 mntput(*mnt);
682 *mnt = mounted;
683 *dentry = dget(mounted->mnt_root);
684 }
685 }
687 /* no need for dcache_lock, as serialization is taken care in
688 * namespace.c
689 */
690 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
691 {
692 struct vfsmount *mounted;
694 mounted = lookup_mnt(*mnt, *dentry);
695 if (mounted) {
696 dput(*dentry);
697 mntput(*mnt);
698 *mnt = mounted;
699 *dentry = dget(mounted->mnt_root);
700 return 1;
701 }
702 return 0;
703 }
705 static __always_inline void follow_dotdot(struct nameidata *nd)
706 {
707 while(1) {
708 struct vfsmount *parent;
709 struct dentry *old = nd->dentry;
711 read_lock(&current->fs->lock);
712 if (nd->dentry == current->fs->root &&
713 nd->mnt == current->fs->rootmnt) {
714 read_unlock(&current->fs->lock);
715 break;
716 }
717 read_unlock(&current->fs->lock);
718 spin_lock(&dcache_lock);
719 if (nd->dentry != nd->mnt->mnt_root) {
720 nd->dentry = dget(nd->dentry->d_parent);
721 spin_unlock(&dcache_lock);
722 dput(old);
723 break;
724 }
725 spin_unlock(&dcache_lock);
726 spin_lock(&vfsmount_lock);
727 parent = nd->mnt->mnt_parent;
728 if (parent == nd->mnt) {
729 spin_unlock(&vfsmount_lock);
730 break;
731 }
732 mntget(parent);
733 nd->dentry = dget(nd->mnt->mnt_mountpoint);
734 spin_unlock(&vfsmount_lock);
735 dput(old);
736 mntput(nd->mnt);
737 nd->mnt = parent;
738 }
739 follow_mount(&nd->mnt, &nd->dentry);
740 }
742 /*
743 * It's more convoluted than I'd like it to be, but... it's still fairly
744 * small and for now I'd prefer to have fast path as straight as possible.
745 * It _is_ time-critical.
746 */
747 static int do_lookup(struct nameidata *nd, struct qstr *name,
748 struct path *path)
749 {
750 struct vfsmount *mnt = nd->mnt;
751 struct dentry *dentry = __d_lookup(nd->dentry, name);
753 if (!dentry)
754 goto need_lookup;
755 if (dentry->d_op && dentry->d_op->d_revalidate)
756 goto need_revalidate;
757 done:
758 path->mnt = mnt;
759 path->dentry = dentry;
760 __follow_mount(path);
761 return 0;
763 need_lookup:
764 dentry = real_lookup(nd->dentry, name, nd);
765 if (IS_ERR(dentry))
766 goto fail;
767 goto done;
769 need_revalidate:
770 if (dentry->d_op->d_revalidate(dentry, nd))
771 goto done;
772 if (d_invalidate(dentry))
773 goto done;
774 dput(dentry);
775 goto need_lookup;
777 fail:
778 return PTR_ERR(dentry);
779 }
781 /*
782 * Name resolution.
783 * This is the basic name resolution function, turning a pathname into
784 * the final dentry. We expect 'base' to be positive and a directory.
785 *
786 * Returns 0 and nd will have valid dentry and mnt on success.
787 * Returns error and drops reference to input namei data on failure.
788 */
789 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
790 {
791 struct path next;
792 struct inode *inode;
793 int err;
794 unsigned int lookup_flags = nd->flags;
796 while (*name=='/')
797 name++;
798 if (!*name)
799 goto return_reval;
801 inode = nd->dentry->d_inode;
802 if (nd->depth)
803 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
805 /* At this point we know we have a real path component. */
806 for(;;) {
807 unsigned long hash;
808 struct qstr this;
809 unsigned int c;
811 nd->flags |= LOOKUP_CONTINUE;
812 err = exec_permission_lite(inode, nd);
813 if (err == -EAGAIN)
814 err = vfs_permission(nd, MAY_EXEC);
815 if (err)
816 break;
818 this.name = name;
819 c = *(const unsigned char *)name;
821 hash = init_name_hash();
822 do {
823 name++;
824 hash = partial_name_hash(c, hash);
825 c = *(const unsigned char *)name;
826 } while (c && (c != '/'));
827 this.len = name - (const char *) this.name;
828 this.hash = end_name_hash(hash);
830 /* remove trailing slashes? */
831 if (!c)
832 goto last_component;
833 while (*++name == '/');
834 if (!*name)
835 goto last_with_slashes;
837 /*
838 * "." and ".." are special - ".." especially so because it has
839 * to be able to know about the current root directory and
840 * parent relationships.
841 */
842 if (this.name[0] == '.') switch (this.len) {
843 default:
844 break;
845 case 2:
846 if (this.name[1] != '.')
847 break;
848 follow_dotdot(nd);
849 inode = nd->dentry->d_inode;
850 /* fallthrough */
851 case 1:
852 continue;
853 }
854 /*
855 * See if the low-level filesystem might want
856 * to use its own hash..
857 */
858 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
859 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
860 if (err < 0)
861 break;
862 }
863 /* This does the actual lookups.. */
864 err = do_lookup(nd, &this, &next);
865 if (err)
866 break;
868 err = -ENOENT;
869 inode = next.dentry->d_inode;
870 if (!inode)
871 goto out_dput;
872 err = -ENOTDIR;
873 if (!inode->i_op)
874 goto out_dput;
876 if (inode->i_op->follow_link) {
877 err = do_follow_link(&next, nd);
878 if (err)
879 goto return_err;
880 err = -ENOENT;
881 inode = nd->dentry->d_inode;
882 if (!inode)
883 break;
884 err = -ENOTDIR;
885 if (!inode->i_op)
886 break;
887 } else
888 path_to_nameidata(&next, nd);
889 err = -ENOTDIR;
890 if (!inode->i_op->lookup)
891 break;
892 continue;
893 /* here ends the main loop */
895 last_with_slashes:
896 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
897 last_component:
898 /* Clear LOOKUP_CONTINUE iff it was previously unset */
899 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
900 if (lookup_flags & LOOKUP_PARENT)
901 goto lookup_parent;
902 if (this.name[0] == '.') switch (this.len) {
903 default:
904 break;
905 case 2:
906 if (this.name[1] != '.')
907 break;
908 follow_dotdot(nd);
909 inode = nd->dentry->d_inode;
910 /* fallthrough */
911 case 1:
912 goto return_reval;
913 }
914 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
915 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
916 if (err < 0)
917 break;
918 }
919 err = do_lookup(nd, &this, &next);
920 if (err)
921 break;
922 inode = next.dentry->d_inode;
923 if ((lookup_flags & LOOKUP_FOLLOW)
924 && inode && inode->i_op && inode->i_op->follow_link) {
925 err = do_follow_link(&next, nd);
926 if (err)
927 goto return_err;
928 inode = nd->dentry->d_inode;
929 } else
930 path_to_nameidata(&next, nd);
931 err = -ENOENT;
932 if (!inode)
933 break;
934 if (lookup_flags & LOOKUP_DIRECTORY) {
935 err = -ENOTDIR;
936 if (!inode->i_op || !inode->i_op->lookup)
937 break;
938 }
939 goto return_base;
940 lookup_parent:
941 nd->last = this;
942 nd->last_type = LAST_NORM;
943 if (this.name[0] != '.')
944 goto return_base;
945 if (this.len == 1)
946 nd->last_type = LAST_DOT;
947 else if (this.len == 2 && this.name[1] == '.')
948 nd->last_type = LAST_DOTDOT;
949 else
950 goto return_base;
951 return_reval:
952 /*
953 * We bypassed the ordinary revalidation routines.
954 * We may need to check the cached dentry for staleness.
955 */
956 if (nd->dentry && nd->dentry->d_sb &&
957 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
958 err = -ESTALE;
959 /* Note: we do not d_invalidate() */
960 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
961 break;
962 }
963 return_base:
964 return 0;
965 out_dput:
966 dput_path(&next, nd);
967 break;
968 }
969 path_release(nd);
970 return_err:
971 return err;
972 }
974 /*
975 * Wrapper to retry pathname resolution whenever the underlying
976 * file system returns an ESTALE.
977 *
978 * Retry the whole path once, forcing real lookup requests
979 * instead of relying on the dcache.
980 */
981 int fastcall link_path_walk(const char *name, struct nameidata *nd)
982 {
983 struct nameidata save = *nd;
984 int result;
986 /* make sure the stuff we saved doesn't go away */
987 dget(save.dentry);
988 mntget(save.mnt);
990 result = __link_path_walk(name, nd);
991 if (result == -ESTALE) {
992 *nd = save;
993 dget(nd->dentry);
994 mntget(nd->mnt);
995 nd->flags |= LOOKUP_REVAL;
996 result = __link_path_walk(name, nd);
997 }
999 dput(save.dentry);
1000 mntput(save.mnt);
1002 return result;
1005 int fastcall path_walk(const char * name, struct nameidata *nd)
1007 current->total_link_count = 0;
1008 return link_path_walk(name, nd);
1011 /*
1012 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1013 * everything is done. Returns 0 and drops input nd, if lookup failed;
1014 */
1015 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1017 if (path_walk(name, nd))
1018 return 0; /* something went wrong... */
1020 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1021 struct dentry *old_dentry = nd->dentry;
1022 struct vfsmount *old_mnt = nd->mnt;
1023 struct qstr last = nd->last;
1024 int last_type = nd->last_type;
1025 /*
1026 * NAME was not found in alternate root or it's a directory. Try to find
1027 * it in the normal root:
1028 */
1029 nd->last_type = LAST_ROOT;
1030 read_lock(&current->fs->lock);
1031 nd->mnt = mntget(current->fs->rootmnt);
1032 nd->dentry = dget(current->fs->root);
1033 read_unlock(&current->fs->lock);
1034 if (path_walk(name, nd) == 0) {
1035 if (nd->dentry->d_inode) {
1036 dput(old_dentry);
1037 mntput(old_mnt);
1038 return 1;
1040 path_release(nd);
1042 nd->dentry = old_dentry;
1043 nd->mnt = old_mnt;
1044 nd->last = last;
1045 nd->last_type = last_type;
1047 return 1;
1050 void set_fs_altroot(void)
1052 char *emul = __emul_prefix();
1053 struct nameidata nd;
1054 struct vfsmount *mnt = NULL, *oldmnt;
1055 struct dentry *dentry = NULL, *olddentry;
1056 int err;
1058 if (!emul)
1059 goto set_it;
1060 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1061 if (!err) {
1062 mnt = nd.mnt;
1063 dentry = nd.dentry;
1065 set_it:
1066 write_lock(&current->fs->lock);
1067 oldmnt = current->fs->altrootmnt;
1068 olddentry = current->fs->altroot;
1069 current->fs->altrootmnt = mnt;
1070 current->fs->altroot = dentry;
1071 write_unlock(&current->fs->lock);
1072 if (olddentry) {
1073 dput(olddentry);
1074 mntput(oldmnt);
1078 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1079 static int fastcall do_path_lookup(int dfd, const char *name,
1080 unsigned int flags, struct nameidata *nd)
1082 int retval = 0;
1083 int fput_needed;
1084 struct file *file;
1086 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1087 nd->flags = flags;
1088 nd->depth = 0;
1090 if (*name=='/') {
1091 read_lock(&current->fs->lock);
1092 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1093 nd->mnt = mntget(current->fs->altrootmnt);
1094 nd->dentry = dget(current->fs->altroot);
1095 read_unlock(&current->fs->lock);
1096 if (__emul_lookup_dentry(name,nd))
1097 goto out; /* found in altroot */
1098 read_lock(&current->fs->lock);
1100 nd->mnt = mntget(current->fs->rootmnt);
1101 nd->dentry = dget(current->fs->root);
1102 read_unlock(&current->fs->lock);
1103 } else if (dfd == AT_FDCWD) {
1104 read_lock(&current->fs->lock);
1105 nd->mnt = mntget(current->fs->pwdmnt);
1106 nd->dentry = dget(current->fs->pwd);
1107 read_unlock(&current->fs->lock);
1108 } else {
1109 struct dentry *dentry;
1111 file = fget_light(dfd, &fput_needed);
1112 retval = -EBADF;
1113 if (!file)
1114 goto out_fail;
1116 dentry = file->f_dentry;
1118 retval = -ENOTDIR;
1119 if (!S_ISDIR(dentry->d_inode->i_mode))
1120 goto fput_fail;
1122 retval = file_permission(file, MAY_EXEC);
1123 if (retval)
1124 goto fput_fail;
1126 nd->mnt = mntget(file->f_vfsmnt);
1127 nd->dentry = dget(dentry);
1129 fput_light(file, fput_needed);
1131 current->total_link_count = 0;
1132 retval = link_path_walk(name, nd);
1133 out:
1134 if (likely(retval == 0)) {
1135 if (unlikely(!audit_dummy_context() && nd && nd->dentry &&
1136 nd->dentry->d_inode))
1137 audit_inode(name, nd->dentry->d_inode);
1139 out_fail:
1140 return retval;
1142 fput_fail:
1143 fput_light(file, fput_needed);
1144 goto out_fail;
1147 int fastcall path_lookup(const char *name, unsigned int flags,
1148 struct nameidata *nd)
1150 return do_path_lookup(AT_FDCWD, name, flags, nd);
1153 static int __path_lookup_intent_open(int dfd, const char *name,
1154 unsigned int lookup_flags, struct nameidata *nd,
1155 int open_flags, int create_mode)
1157 struct file *filp = get_empty_filp();
1158 int err;
1160 if (filp == NULL)
1161 return -ENFILE;
1162 nd->intent.open.file = filp;
1163 nd->intent.open.flags = open_flags;
1164 nd->intent.open.create_mode = create_mode;
1165 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1166 if (IS_ERR(nd->intent.open.file)) {
1167 if (err == 0) {
1168 err = PTR_ERR(nd->intent.open.file);
1169 path_release(nd);
1171 } else if (err != 0)
1172 release_open_intent(nd);
1173 return err;
1176 /**
1177 * path_lookup_open - lookup a file path with open intent
1178 * @dfd: the directory to use as base, or AT_FDCWD
1179 * @name: pointer to file name
1180 * @lookup_flags: lookup intent flags
1181 * @nd: pointer to nameidata
1182 * @open_flags: open intent flags
1183 */
1184 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1185 struct nameidata *nd, int open_flags)
1187 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1188 open_flags, 0);
1191 /**
1192 * path_lookup_create - lookup a file path with open + create intent
1193 * @dfd: the directory to use as base, or AT_FDCWD
1194 * @name: pointer to file name
1195 * @lookup_flags: lookup intent flags
1196 * @nd: pointer to nameidata
1197 * @open_flags: open intent flags
1198 * @create_mode: create intent flags
1199 */
1200 static int path_lookup_create(int dfd, const char *name,
1201 unsigned int lookup_flags, struct nameidata *nd,
1202 int open_flags, int create_mode)
1204 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1205 nd, open_flags, create_mode);
1208 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1209 struct nameidata *nd, int open_flags)
1211 char *tmp = getname(name);
1212 int err = PTR_ERR(tmp);
1214 if (!IS_ERR(tmp)) {
1215 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1216 putname(tmp);
1218 return err;
1221 /*
1222 * Restricted form of lookup. Doesn't follow links, single-component only,
1223 * needs parent already locked. Doesn't follow mounts.
1224 * SMP-safe.
1225 */
1226 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1228 struct dentry * dentry;
1229 struct inode *inode;
1230 int err;
1232 inode = base->d_inode;
1233 err = permission(inode, MAY_EXEC, nd);
1234 dentry = ERR_PTR(err);
1235 if (err)
1236 goto out;
1238 /*
1239 * See if the low-level filesystem might want
1240 * to use its own hash..
1241 */
1242 if (base->d_op && base->d_op->d_hash) {
1243 err = base->d_op->d_hash(base, name);
1244 dentry = ERR_PTR(err);
1245 if (err < 0)
1246 goto out;
1249 dentry = cached_lookup(base, name, nd);
1250 if (!dentry) {
1251 struct dentry *new = d_alloc(base, name);
1252 dentry = ERR_PTR(-ENOMEM);
1253 if (!new)
1254 goto out;
1255 dentry = inode->i_op->lookup(inode, new, nd);
1256 if (!dentry)
1257 dentry = new;
1258 else
1259 dput(new);
1261 out:
1262 return dentry;
1265 static struct dentry *lookup_hash(struct nameidata *nd)
1267 return __lookup_hash(&nd->last, nd->dentry, nd);
1270 /* SMP-safe */
1271 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1273 unsigned long hash;
1274 struct qstr this;
1275 unsigned int c;
1277 this.name = name;
1278 this.len = len;
1279 if (!len)
1280 goto access;
1282 hash = init_name_hash();
1283 while (len--) {
1284 c = *(const unsigned char *)name++;
1285 if (c == '/' || c == '\0')
1286 goto access;
1287 hash = partial_name_hash(c, hash);
1289 this.hash = end_name_hash(hash);
1291 return __lookup_hash(&this, base, NULL);
1292 access:
1293 return ERR_PTR(-EACCES);
1296 /*
1297 * namei()
1299 * is used by most simple commands to get the inode of a specified name.
1300 * Open, link etc use their own routines, but this is enough for things
1301 * like 'chmod' etc.
1303 * namei exists in two versions: namei/lnamei. The only difference is
1304 * that namei follows links, while lnamei does not.
1305 * SMP-safe
1306 */
1307 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1308 struct nameidata *nd)
1310 char *tmp = getname(name);
1311 int err = PTR_ERR(tmp);
1313 if (!IS_ERR(tmp)) {
1314 err = do_path_lookup(dfd, tmp, flags, nd);
1315 putname(tmp);
1317 return err;
1320 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1322 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1325 /*
1326 * It's inline, so penalty for filesystems that don't use sticky bit is
1327 * minimal.
1328 */
1329 static inline int check_sticky(struct inode *dir, struct inode *inode)
1331 if (!(dir->i_mode & S_ISVTX))
1332 return 0;
1333 if (inode->i_uid == current->fsuid)
1334 return 0;
1335 if (dir->i_uid == current->fsuid)
1336 return 0;
1337 return !capable(CAP_FOWNER);
1340 /*
1341 * Check whether we can remove a link victim from directory dir, check
1342 * whether the type of victim is right.
1343 * 1. We can't do it if dir is read-only (done in permission())
1344 * 2. We should have write and exec permissions on dir
1345 * 3. We can't remove anything from append-only dir
1346 * 4. We can't do anything with immutable dir (done in permission())
1347 * 5. If the sticky bit on dir is set we should either
1348 * a. be owner of dir, or
1349 * b. be owner of victim, or
1350 * c. have CAP_FOWNER capability
1351 * 6. If the victim is append-only or immutable we can't do antyhing with
1352 * links pointing to it.
1353 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1354 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1355 * 9. We can't remove a root or mountpoint.
1356 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1357 * nfs_async_unlink().
1358 */
1359 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1361 int error;
1363 if (!victim->d_inode)
1364 return -ENOENT;
1366 BUG_ON(victim->d_parent->d_inode != dir);
1367 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
1369 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1370 if (error)
1371 return error;
1372 if (IS_APPEND(dir))
1373 return -EPERM;
1374 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1375 IS_IMMUTABLE(victim->d_inode))
1376 return -EPERM;
1377 if (isdir) {
1378 if (!S_ISDIR(victim->d_inode->i_mode))
1379 return -ENOTDIR;
1380 if (IS_ROOT(victim))
1381 return -EBUSY;
1382 } else if (S_ISDIR(victim->d_inode->i_mode))
1383 return -EISDIR;
1384 if (IS_DEADDIR(dir))
1385 return -ENOENT;
1386 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1387 return -EBUSY;
1388 return 0;
1391 /* Check whether we can create an object with dentry child in directory
1392 * dir.
1393 * 1. We can't do it if child already exists (open has special treatment for
1394 * this case, but since we are inlined it's OK)
1395 * 2. We can't do it if dir is read-only (done in permission())
1396 * 3. We should have write and exec permissions on dir
1397 * 4. We can't do it if dir is immutable (done in permission())
1398 */
1399 static inline int may_create(struct inode *dir, struct dentry *child,
1400 struct nameidata *nd)
1402 if (child->d_inode)
1403 return -EEXIST;
1404 if (IS_DEADDIR(dir))
1405 return -ENOENT;
1406 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1409 /*
1410 * O_DIRECTORY translates into forcing a directory lookup.
1411 */
1412 static inline int lookup_flags(unsigned int f)
1414 unsigned long retval = LOOKUP_FOLLOW;
1416 if (f & O_NOFOLLOW)
1417 retval &= ~LOOKUP_FOLLOW;
1419 if (f & O_DIRECTORY)
1420 retval |= LOOKUP_DIRECTORY;
1422 return retval;
1425 /*
1426 * p1 and p2 should be directories on the same fs.
1427 */
1428 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1430 struct dentry *p;
1432 if (p1 == p2) {
1433 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1434 return NULL;
1437 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1439 for (p = p1; p->d_parent != p; p = p->d_parent) {
1440 if (p->d_parent == p2) {
1441 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1442 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1443 return p;
1447 for (p = p2; p->d_parent != p; p = p->d_parent) {
1448 if (p->d_parent == p1) {
1449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1450 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1451 return p;
1455 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1456 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1457 return NULL;
1460 void unlock_rename(struct dentry *p1, struct dentry *p2)
1462 mutex_unlock(&p1->d_inode->i_mutex);
1463 if (p1 != p2) {
1464 mutex_unlock(&p2->d_inode->i_mutex);
1465 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1469 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1470 struct nameidata *nd)
1472 int error = may_create(dir, dentry, nd);
1474 if (error)
1475 return error;
1477 if (!dir->i_op || !dir->i_op->create)
1478 return -EACCES; /* shouldn't it be ENOSYS? */
1479 mode &= S_IALLUGO;
1480 mode |= S_IFREG;
1481 error = security_inode_create(dir, dentry, mode);
1482 if (error)
1483 return error;
1484 DQUOT_INIT(dir);
1485 error = dir->i_op->create(dir, dentry, mode, nd);
1486 if (!error)
1487 fsnotify_create(dir, dentry);
1488 return error;
1491 int may_open(struct nameidata *nd, int acc_mode, int flag)
1493 struct dentry *dentry = nd->dentry;
1494 struct inode *inode = dentry->d_inode;
1495 int error;
1497 if (!inode)
1498 return -ENOENT;
1500 if (S_ISLNK(inode->i_mode))
1501 return -ELOOP;
1503 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1504 return -EISDIR;
1506 error = vfs_permission(nd, acc_mode);
1507 if (error)
1508 return error;
1510 /*
1511 * FIFO's, sockets and device files are special: they don't
1512 * actually live on the filesystem itself, and as such you
1513 * can write to them even if the filesystem is read-only.
1514 */
1515 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1516 flag &= ~O_TRUNC;
1517 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1518 if (nd->mnt->mnt_flags & MNT_NODEV)
1519 return -EACCES;
1521 flag &= ~O_TRUNC;
1522 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1523 return -EROFS;
1524 /*
1525 * An append-only file must be opened in append mode for writing.
1526 */
1527 if (IS_APPEND(inode)) {
1528 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1529 return -EPERM;
1530 if (flag & O_TRUNC)
1531 return -EPERM;
1534 /* O_NOATIME can only be set by the owner or superuser */
1535 if (flag & O_NOATIME)
1536 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1537 return -EPERM;
1539 /*
1540 * Ensure there are no outstanding leases on the file.
1541 */
1542 error = break_lease(inode, flag);
1543 if (error)
1544 return error;
1546 if (flag & O_TRUNC) {
1547 error = get_write_access(inode);
1548 if (error)
1549 return error;
1551 /*
1552 * Refuse to truncate files with mandatory locks held on them.
1553 */
1554 error = locks_verify_locked(inode);
1555 if (!error) {
1556 DQUOT_INIT(inode);
1558 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1560 put_write_access(inode);
1561 if (error)
1562 return error;
1563 } else
1564 if (flag & FMODE_WRITE)
1565 DQUOT_INIT(inode);
1567 return 0;
1570 /*
1571 * open_namei()
1573 * namei for open - this is in fact almost the whole open-routine.
1575 * Note that the low bits of "flag" aren't the same as in the open
1576 * system call - they are 00 - no permissions needed
1577 * 01 - read permission needed
1578 * 10 - write permission needed
1579 * 11 - read/write permissions needed
1580 * which is a lot more logical, and also allows the "no perm" needed
1581 * for symlinks (where the permissions are checked later).
1582 * SMP-safe
1583 */
1584 int open_namei(int dfd, const char *pathname, int flag,
1585 int mode, struct nameidata *nd)
1587 int acc_mode, error;
1588 struct path path;
1589 struct dentry *dir;
1590 int count = 0;
1592 acc_mode = ACC_MODE(flag);
1594 /* O_TRUNC implies we need access checks for write permissions */
1595 if (flag & O_TRUNC)
1596 acc_mode |= MAY_WRITE;
1598 /* Allow the LSM permission hook to distinguish append
1599 access from general write access. */
1600 if (flag & O_APPEND)
1601 acc_mode |= MAY_APPEND;
1603 /*
1604 * The simplest case - just a plain lookup.
1605 */
1606 if (!(flag & O_CREAT)) {
1607 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1608 nd, flag);
1609 if (error)
1610 return error;
1611 goto ok;
1614 /*
1615 * Create - we need to know the parent.
1616 */
1617 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1618 if (error)
1619 return error;
1621 /*
1622 * We have the parent and last component. First of all, check
1623 * that we are not asked to creat(2) an obvious directory - that
1624 * will not do.
1625 */
1626 error = -EISDIR;
1627 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1628 goto exit;
1630 dir = nd->dentry;
1631 nd->flags &= ~LOOKUP_PARENT;
1632 mutex_lock(&dir->d_inode->i_mutex);
1633 path.dentry = lookup_hash(nd);
1634 path.mnt = nd->mnt;
1636 do_last:
1637 error = PTR_ERR(path.dentry);
1638 if (IS_ERR(path.dentry)) {
1639 mutex_unlock(&dir->d_inode->i_mutex);
1640 goto exit;
1643 if (IS_ERR(nd->intent.open.file)) {
1644 mutex_unlock(&dir->d_inode->i_mutex);
1645 error = PTR_ERR(nd->intent.open.file);
1646 goto exit_dput;
1649 /* Negative dentry, just create the file */
1650 if (!path.dentry->d_inode) {
1651 if (!IS_POSIXACL(dir->d_inode))
1652 mode &= ~current->fs->umask;
1653 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1654 mutex_unlock(&dir->d_inode->i_mutex);
1655 dput(nd->dentry);
1656 nd->dentry = path.dentry;
1657 if (error)
1658 goto exit;
1659 /* Don't check for write permission, don't truncate */
1660 acc_mode = 0;
1661 flag &= ~O_TRUNC;
1662 goto ok;
1665 /*
1666 * It already exists.
1667 */
1668 mutex_unlock(&dir->d_inode->i_mutex);
1669 audit_inode_update(path.dentry->d_inode);
1671 error = -EEXIST;
1672 if (flag & O_EXCL)
1673 goto exit_dput;
1675 if (__follow_mount(&path)) {
1676 error = -ELOOP;
1677 if (flag & O_NOFOLLOW)
1678 goto exit_dput;
1681 error = -ENOENT;
1682 if (!path.dentry->d_inode)
1683 goto exit_dput;
1684 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1685 goto do_link;
1687 path_to_nameidata(&path, nd);
1688 error = -EISDIR;
1689 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1690 goto exit;
1691 ok:
1692 error = may_open(nd, acc_mode, flag);
1693 if (error)
1694 goto exit;
1695 return 0;
1697 exit_dput:
1698 dput_path(&path, nd);
1699 exit:
1700 if (!IS_ERR(nd->intent.open.file))
1701 release_open_intent(nd);
1702 path_release(nd);
1703 return error;
1705 do_link:
1706 error = -ELOOP;
1707 if (flag & O_NOFOLLOW)
1708 goto exit_dput;
1709 /*
1710 * This is subtle. Instead of calling do_follow_link() we do the
1711 * thing by hands. The reason is that this way we have zero link_count
1712 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1713 * After that we have the parent and last component, i.e.
1714 * we are in the same situation as after the first path_walk().
1715 * Well, almost - if the last component is normal we get its copy
1716 * stored in nd->last.name and we will have to putname() it when we
1717 * are done. Procfs-like symlinks just set LAST_BIND.
1718 */
1719 nd->flags |= LOOKUP_PARENT;
1720 error = security_inode_follow_link(path.dentry, nd);
1721 if (error)
1722 goto exit_dput;
1723 error = __do_follow_link(&path, nd);
1724 if (error) {
1725 /* Does someone understand code flow here? Or it is only
1726 * me so stupid? Anathema to whoever designed this non-sense
1727 * with "intent.open".
1728 */
1729 release_open_intent(nd);
1730 return error;
1732 nd->flags &= ~LOOKUP_PARENT;
1733 if (nd->last_type == LAST_BIND)
1734 goto ok;
1735 error = -EISDIR;
1736 if (nd->last_type != LAST_NORM)
1737 goto exit;
1738 if (nd->last.name[nd->last.len]) {
1739 __putname(nd->last.name);
1740 goto exit;
1742 error = -ELOOP;
1743 if (count++==32) {
1744 __putname(nd->last.name);
1745 goto exit;
1747 dir = nd->dentry;
1748 mutex_lock(&dir->d_inode->i_mutex);
1749 path.dentry = lookup_hash(nd);
1750 path.mnt = nd->mnt;
1751 __putname(nd->last.name);
1752 goto do_last;
1755 /**
1756 * lookup_create - lookup a dentry, creating it if it doesn't exist
1757 * @nd: nameidata info
1758 * @is_dir: directory flag
1760 * Simple function to lookup and return a dentry and create it
1761 * if it doesn't exist. Is SMP-safe.
1763 * Returns with nd->dentry->d_inode->i_mutex locked.
1764 */
1765 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1767 struct dentry *dentry = ERR_PTR(-EEXIST);
1769 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1770 /*
1771 * Yucky last component or no last component at all?
1772 * (foo/., foo/.., /////)
1773 */
1774 if (nd->last_type != LAST_NORM)
1775 goto fail;
1776 nd->flags &= ~LOOKUP_PARENT;
1777 nd->flags |= LOOKUP_CREATE;
1778 nd->intent.open.flags = O_EXCL;
1780 /*
1781 * Do the final lookup.
1782 */
1783 dentry = lookup_hash(nd);
1784 if (IS_ERR(dentry))
1785 goto fail;
1787 /*
1788 * Special case - lookup gave negative, but... we had foo/bar/
1789 * From the vfs_mknod() POV we just have a negative dentry -
1790 * all is fine. Let's be bastards - you had / on the end, you've
1791 * been asking for (non-existent) directory. -ENOENT for you.
1792 */
1793 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1794 goto enoent;
1795 return dentry;
1796 enoent:
1797 dput(dentry);
1798 dentry = ERR_PTR(-ENOENT);
1799 fail:
1800 return dentry;
1802 EXPORT_SYMBOL_GPL(lookup_create);
1804 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1806 int error = may_create(dir, dentry, NULL);
1808 if (error)
1809 return error;
1811 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1812 return -EPERM;
1814 if (!dir->i_op || !dir->i_op->mknod)
1815 return -EPERM;
1817 error = security_inode_mknod(dir, dentry, mode, dev);
1818 if (error)
1819 return error;
1821 DQUOT_INIT(dir);
1822 error = dir->i_op->mknod(dir, dentry, mode, dev);
1823 if (!error)
1824 fsnotify_create(dir, dentry);
1825 return error;
1828 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1829 unsigned dev)
1831 int error = 0;
1832 char * tmp;
1833 struct dentry * dentry;
1834 struct nameidata nd;
1836 if (S_ISDIR(mode))
1837 return -EPERM;
1838 tmp = getname(filename);
1839 if (IS_ERR(tmp))
1840 return PTR_ERR(tmp);
1842 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1843 if (error)
1844 goto out;
1845 dentry = lookup_create(&nd, 0);
1846 error = PTR_ERR(dentry);
1848 if (!IS_POSIXACL(nd.dentry->d_inode))
1849 mode &= ~current->fs->umask;
1850 if (!IS_ERR(dentry)) {
1851 switch (mode & S_IFMT) {
1852 case 0: case S_IFREG:
1853 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1854 break;
1855 case S_IFCHR: case S_IFBLK:
1856 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1857 new_decode_dev(dev));
1858 break;
1859 case S_IFIFO: case S_IFSOCK:
1860 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1861 break;
1862 case S_IFDIR:
1863 error = -EPERM;
1864 break;
1865 default:
1866 error = -EINVAL;
1868 dput(dentry);
1870 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1871 path_release(&nd);
1872 out:
1873 putname(tmp);
1875 return error;
1878 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1880 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1883 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1885 int error = may_create(dir, dentry, NULL);
1887 if (error)
1888 return error;
1890 if (!dir->i_op || !dir->i_op->mkdir)
1891 return -EPERM;
1893 mode &= (S_IRWXUGO|S_ISVTX);
1894 error = security_inode_mkdir(dir, dentry, mode);
1895 if (error)
1896 return error;
1898 DQUOT_INIT(dir);
1899 error = dir->i_op->mkdir(dir, dentry, mode);
1900 if (!error)
1901 fsnotify_mkdir(dir, dentry);
1902 return error;
1905 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1907 int error = 0;
1908 char * tmp;
1910 tmp = getname(pathname);
1911 error = PTR_ERR(tmp);
1912 if (!IS_ERR(tmp)) {
1913 struct dentry *dentry;
1914 struct nameidata nd;
1916 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1917 if (error)
1918 goto out;
1919 dentry = lookup_create(&nd, 1);
1920 error = PTR_ERR(dentry);
1921 if (!IS_ERR(dentry)) {
1922 if (!IS_POSIXACL(nd.dentry->d_inode))
1923 mode &= ~current->fs->umask;
1924 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1925 dput(dentry);
1927 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1928 path_release(&nd);
1929 out:
1930 putname(tmp);
1933 return error;
1936 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1938 return sys_mkdirat(AT_FDCWD, pathname, mode);
1941 /*
1942 * We try to drop the dentry early: we should have
1943 * a usage count of 2 if we're the only user of this
1944 * dentry, and if that is true (possibly after pruning
1945 * the dcache), then we drop the dentry now.
1947 * A low-level filesystem can, if it choses, legally
1948 * do a
1950 * if (!d_unhashed(dentry))
1951 * return -EBUSY;
1953 * if it cannot handle the case of removing a directory
1954 * that is still in use by something else..
1955 */
1956 void dentry_unhash(struct dentry *dentry)
1958 dget(dentry);
1959 if (atomic_read(&dentry->d_count))
1960 shrink_dcache_parent(dentry);
1961 spin_lock(&dcache_lock);
1962 spin_lock(&dentry->d_lock);
1963 if (atomic_read(&dentry->d_count) == 2)
1964 __d_drop(dentry);
1965 spin_unlock(&dentry->d_lock);
1966 spin_unlock(&dcache_lock);
1969 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1971 int error = may_delete(dir, dentry, 1);
1973 if (error)
1974 return error;
1976 if (!dir->i_op || !dir->i_op->rmdir)
1977 return -EPERM;
1979 DQUOT_INIT(dir);
1981 mutex_lock(&dentry->d_inode->i_mutex);
1982 dentry_unhash(dentry);
1983 if (d_mountpoint(dentry))
1984 error = -EBUSY;
1985 else {
1986 error = security_inode_rmdir(dir, dentry);
1987 if (!error) {
1988 error = dir->i_op->rmdir(dir, dentry);
1989 if (!error)
1990 dentry->d_inode->i_flags |= S_DEAD;
1993 mutex_unlock(&dentry->d_inode->i_mutex);
1994 if (!error) {
1995 d_delete(dentry);
1997 dput(dentry);
1999 return error;
2002 static long do_rmdir(int dfd, const char __user *pathname)
2004 int error = 0;
2005 char * name;
2006 struct dentry *dentry;
2007 struct nameidata nd;
2009 name = getname(pathname);
2010 if(IS_ERR(name))
2011 return PTR_ERR(name);
2013 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2014 if (error)
2015 goto exit;
2017 switch(nd.last_type) {
2018 case LAST_DOTDOT:
2019 error = -ENOTEMPTY;
2020 goto exit1;
2021 case LAST_DOT:
2022 error = -EINVAL;
2023 goto exit1;
2024 case LAST_ROOT:
2025 error = -EBUSY;
2026 goto exit1;
2028 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2029 dentry = lookup_hash(&nd);
2030 error = PTR_ERR(dentry);
2031 if (!IS_ERR(dentry)) {
2032 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2033 dput(dentry);
2035 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2036 exit1:
2037 path_release(&nd);
2038 exit:
2039 putname(name);
2040 return error;
2043 asmlinkage long sys_rmdir(const char __user *pathname)
2045 return do_rmdir(AT_FDCWD, pathname);
2048 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2050 int error = may_delete(dir, dentry, 0);
2052 if (error)
2053 return error;
2055 if (!dir->i_op || !dir->i_op->unlink)
2056 return -EPERM;
2058 DQUOT_INIT(dir);
2060 mutex_lock(&dentry->d_inode->i_mutex);
2061 if (d_mountpoint(dentry))
2062 error = -EBUSY;
2063 else {
2064 error = security_inode_unlink(dir, dentry);
2065 if (!error)
2066 error = dir->i_op->unlink(dir, dentry);
2068 mutex_unlock(&dentry->d_inode->i_mutex);
2070 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2071 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2072 d_delete(dentry);
2075 return error;
2078 /*
2079 * Make sure that the actual truncation of the file will occur outside its
2080 * directory's i_mutex. Truncate can take a long time if there is a lot of
2081 * writeout happening, and we don't want to prevent access to the directory
2082 * while waiting on the I/O.
2083 */
2084 static long do_unlinkat(int dfd, const char __user *pathname)
2086 int error = 0;
2087 char * name;
2088 struct dentry *dentry;
2089 struct nameidata nd;
2090 struct inode *inode = NULL;
2092 name = getname(pathname);
2093 if(IS_ERR(name))
2094 return PTR_ERR(name);
2096 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2097 if (error)
2098 goto exit;
2099 error = -EISDIR;
2100 if (nd.last_type != LAST_NORM)
2101 goto exit1;
2102 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2103 dentry = lookup_hash(&nd);
2104 error = PTR_ERR(dentry);
2105 if (!IS_ERR(dentry)) {
2106 /* Why not before? Because we want correct error value */
2107 if (nd.last.name[nd.last.len])
2108 goto slashes;
2109 inode = dentry->d_inode;
2110 if (inode)
2111 atomic_inc(&inode->i_count);
2112 error = vfs_unlink(nd.dentry->d_inode, dentry);
2113 exit2:
2114 dput(dentry);
2116 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2117 if (inode)
2118 iput(inode); /* truncate the inode here */
2119 exit1:
2120 path_release(&nd);
2121 exit:
2122 putname(name);
2123 return error;
2125 slashes:
2126 error = !dentry->d_inode ? -ENOENT :
2127 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2128 goto exit2;
2131 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2133 if ((flag & ~AT_REMOVEDIR) != 0)
2134 return -EINVAL;
2136 if (flag & AT_REMOVEDIR)
2137 return do_rmdir(dfd, pathname);
2139 return do_unlinkat(dfd, pathname);
2142 asmlinkage long sys_unlink(const char __user *pathname)
2144 return do_unlinkat(AT_FDCWD, pathname);
2147 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2149 int error = may_create(dir, dentry, NULL);
2151 if (error)
2152 return error;
2154 if (!dir->i_op || !dir->i_op->symlink)
2155 return -EPERM;
2157 error = security_inode_symlink(dir, dentry, oldname);
2158 if (error)
2159 return error;
2161 DQUOT_INIT(dir);
2162 error = dir->i_op->symlink(dir, dentry, oldname);
2163 if (!error)
2164 fsnotify_create(dir, dentry);
2165 return error;
2168 asmlinkage long sys_symlinkat(const char __user *oldname,
2169 int newdfd, const char __user *newname)
2171 int error = 0;
2172 char * from;
2173 char * to;
2175 from = getname(oldname);
2176 if(IS_ERR(from))
2177 return PTR_ERR(from);
2178 to = getname(newname);
2179 error = PTR_ERR(to);
2180 if (!IS_ERR(to)) {
2181 struct dentry *dentry;
2182 struct nameidata nd;
2184 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2185 if (error)
2186 goto out;
2187 dentry = lookup_create(&nd, 0);
2188 error = PTR_ERR(dentry);
2189 if (!IS_ERR(dentry)) {
2190 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2191 dput(dentry);
2193 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2194 path_release(&nd);
2195 out:
2196 putname(to);
2198 putname(from);
2199 return error;
2202 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2204 return sys_symlinkat(oldname, AT_FDCWD, newname);
2207 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2209 struct inode *inode = old_dentry->d_inode;
2210 int error;
2212 if (!inode)
2213 return -ENOENT;
2215 error = may_create(dir, new_dentry, NULL);
2216 if (error)
2217 return error;
2219 if (dir->i_sb != inode->i_sb)
2220 return -EXDEV;
2222 /*
2223 * A link to an append-only or immutable file cannot be created.
2224 */
2225 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2226 return -EPERM;
2227 if (!dir->i_op || !dir->i_op->link)
2228 return -EPERM;
2229 if (S_ISDIR(old_dentry->d_inode->i_mode))
2230 return -EPERM;
2232 error = security_inode_link(old_dentry, dir, new_dentry);
2233 if (error)
2234 return error;
2236 mutex_lock(&old_dentry->d_inode->i_mutex);
2237 DQUOT_INIT(dir);
2238 error = dir->i_op->link(old_dentry, dir, new_dentry);
2239 mutex_unlock(&old_dentry->d_inode->i_mutex);
2240 if (!error)
2241 fsnotify_create(dir, new_dentry);
2242 return error;
2245 /*
2246 * Hardlinks are often used in delicate situations. We avoid
2247 * security-related surprises by not following symlinks on the
2248 * newname. --KAB
2250 * We don't follow them on the oldname either to be compatible
2251 * with linux 2.0, and to avoid hard-linking to directories
2252 * and other special files. --ADM
2253 */
2254 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2255 int newdfd, const char __user *newname,
2256 int flags)
2258 struct dentry *new_dentry;
2259 struct nameidata nd, old_nd;
2260 int error;
2261 char * to;
2263 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2264 return -EINVAL;
2266 to = getname(newname);
2267 if (IS_ERR(to))
2268 return PTR_ERR(to);
2270 error = __user_walk_fd(olddfd, oldname,
2271 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2272 &old_nd);
2273 if (error)
2274 goto exit;
2275 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2276 if (error)
2277 goto out;
2278 error = -EXDEV;
2279 if (old_nd.mnt != nd.mnt)
2280 goto out_release;
2281 new_dentry = lookup_create(&nd, 0);
2282 error = PTR_ERR(new_dentry);
2283 if (!IS_ERR(new_dentry)) {
2284 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2285 dput(new_dentry);
2287 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2288 out_release:
2289 path_release(&nd);
2290 out:
2291 path_release(&old_nd);
2292 exit:
2293 putname(to);
2295 return error;
2298 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2300 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2303 /*
2304 * The worst of all namespace operations - renaming directory. "Perverted"
2305 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2306 * Problems:
2307 * a) we can get into loop creation. Check is done in is_subdir().
2308 * b) race potential - two innocent renames can create a loop together.
2309 * That's where 4.4 screws up. Current fix: serialization on
2310 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2311 * story.
2312 * c) we have to lock _three_ objects - parents and victim (if it exists).
2313 * And that - after we got ->i_mutex on parents (until then we don't know
2314 * whether the target exists). Solution: try to be smart with locking
2315 * order for inodes. We rely on the fact that tree topology may change
2316 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2317 * move will be locked. Thus we can rank directories by the tree
2318 * (ancestors first) and rank all non-directories after them.
2319 * That works since everybody except rename does "lock parent, lookup,
2320 * lock child" and rename is under ->s_vfs_rename_mutex.
2321 * HOWEVER, it relies on the assumption that any object with ->lookup()
2322 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2323 * we'd better make sure that there's no link(2) for them.
2324 * d) some filesystems don't support opened-but-unlinked directories,
2325 * either because of layout or because they are not ready to deal with
2326 * all cases correctly. The latter will be fixed (taking this sort of
2327 * stuff into VFS), but the former is not going away. Solution: the same
2328 * trick as in rmdir().
2329 * e) conversion from fhandle to dentry may come in the wrong moment - when
2330 * we are removing the target. Solution: we will have to grab ->i_mutex
2331 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2332 * ->i_mutex on parents, which works but leads to some truely excessive
2333 * locking].
2334 */
2335 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2336 struct inode *new_dir, struct dentry *new_dentry)
2338 int error = 0;
2339 struct inode *target;
2341 /*
2342 * If we are going to change the parent - check write permissions,
2343 * we'll need to flip '..'.
2344 */
2345 if (new_dir != old_dir) {
2346 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2347 if (error)
2348 return error;
2351 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2352 if (error)
2353 return error;
2355 target = new_dentry->d_inode;
2356 if (target) {
2357 mutex_lock(&target->i_mutex);
2358 dentry_unhash(new_dentry);
2360 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2361 error = -EBUSY;
2362 else
2363 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2364 if (target) {
2365 if (!error)
2366 target->i_flags |= S_DEAD;
2367 mutex_unlock(&target->i_mutex);
2368 if (d_unhashed(new_dentry))
2369 d_rehash(new_dentry);
2370 dput(new_dentry);
2372 if (!error)
2373 d_move(old_dentry,new_dentry);
2374 return error;
2377 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2378 struct inode *new_dir, struct dentry *new_dentry)
2380 struct inode *target;
2381 int error;
2383 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2384 if (error)
2385 return error;
2387 dget(new_dentry);
2388 target = new_dentry->d_inode;
2389 if (target)
2390 mutex_lock(&target->i_mutex);
2391 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2392 error = -EBUSY;
2393 else
2394 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2395 if (!error) {
2396 /* The following d_move() should become unconditional */
2397 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2398 d_move(old_dentry, new_dentry);
2400 if (target)
2401 mutex_unlock(&target->i_mutex);
2402 dput(new_dentry);
2403 return error;
2406 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2407 struct inode *new_dir, struct dentry *new_dentry)
2409 int error;
2410 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2411 const char *old_name;
2413 if (old_dentry->d_inode == new_dentry->d_inode)
2414 return 0;
2416 error = may_delete(old_dir, old_dentry, is_dir);
2417 if (error)
2418 return error;
2420 if (!new_dentry->d_inode)
2421 error = may_create(new_dir, new_dentry, NULL);
2422 else
2423 error = may_delete(new_dir, new_dentry, is_dir);
2424 if (error)
2425 return error;
2427 if (!old_dir->i_op || !old_dir->i_op->rename)
2428 return -EPERM;
2430 DQUOT_INIT(old_dir);
2431 DQUOT_INIT(new_dir);
2433 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2435 if (is_dir)
2436 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2437 else
2438 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2439 if (!error) {
2440 const char *new_name = old_dentry->d_name.name;
2441 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2442 new_dentry->d_inode, old_dentry->d_inode);
2444 fsnotify_oldname_free(old_name);
2446 return error;
2449 static int do_rename(int olddfd, const char *oldname,
2450 int newdfd, const char *newname)
2452 int error = 0;
2453 struct dentry * old_dir, * new_dir;
2454 struct dentry * old_dentry, *new_dentry;
2455 struct dentry * trap;
2456 struct nameidata oldnd, newnd;
2458 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2459 if (error)
2460 goto exit;
2462 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2463 if (error)
2464 goto exit1;
2466 error = -EXDEV;
2467 if (oldnd.mnt != newnd.mnt)
2468 goto exit2;
2470 old_dir = oldnd.dentry;
2471 error = -EBUSY;
2472 if (oldnd.last_type != LAST_NORM)
2473 goto exit2;
2475 new_dir = newnd.dentry;
2476 if (newnd.last_type != LAST_NORM)
2477 goto exit2;
2479 trap = lock_rename(new_dir, old_dir);
2481 old_dentry = lookup_hash(&oldnd);
2482 error = PTR_ERR(old_dentry);
2483 if (IS_ERR(old_dentry))
2484 goto exit3;
2485 /* source must exist */
2486 error = -ENOENT;
2487 if (!old_dentry->d_inode)
2488 goto exit4;
2489 /* unless the source is a directory trailing slashes give -ENOTDIR */
2490 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2491 error = -ENOTDIR;
2492 if (oldnd.last.name[oldnd.last.len])
2493 goto exit4;
2494 if (newnd.last.name[newnd.last.len])
2495 goto exit4;
2497 /* source should not be ancestor of target */
2498 error = -EINVAL;
2499 if (old_dentry == trap)
2500 goto exit4;
2501 new_dentry = lookup_hash(&newnd);
2502 error = PTR_ERR(new_dentry);
2503 if (IS_ERR(new_dentry))
2504 goto exit4;
2505 /* target should not be an ancestor of source */
2506 error = -ENOTEMPTY;
2507 if (new_dentry == trap)
2508 goto exit5;
2510 error = vfs_rename(old_dir->d_inode, old_dentry,
2511 new_dir->d_inode, new_dentry);
2512 exit5:
2513 dput(new_dentry);
2514 exit4:
2515 dput(old_dentry);
2516 exit3:
2517 unlock_rename(new_dir, old_dir);
2518 exit2:
2519 path_release(&newnd);
2520 exit1:
2521 path_release(&oldnd);
2522 exit:
2523 return error;
2526 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2527 int newdfd, const char __user *newname)
2529 int error;
2530 char * from;
2531 char * to;
2533 from = getname(oldname);
2534 if(IS_ERR(from))
2535 return PTR_ERR(from);
2536 to = getname(newname);
2537 error = PTR_ERR(to);
2538 if (!IS_ERR(to)) {
2539 error = do_rename(olddfd, from, newdfd, to);
2540 putname(to);
2542 putname(from);
2543 return error;
2546 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2548 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2551 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2553 int len;
2555 len = PTR_ERR(link);
2556 if (IS_ERR(link))
2557 goto out;
2559 len = strlen(link);
2560 if (len > (unsigned) buflen)
2561 len = buflen;
2562 if (copy_to_user(buffer, link, len))
2563 len = -EFAULT;
2564 out:
2565 return len;
2568 /*
2569 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2570 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2571 * using) it for any given inode is up to filesystem.
2572 */
2573 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2575 struct nameidata nd;
2576 void *cookie;
2578 nd.depth = 0;
2579 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2580 if (!IS_ERR(cookie)) {
2581 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2582 if (dentry->d_inode->i_op->put_link)
2583 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2584 cookie = ERR_PTR(res);
2586 return PTR_ERR(cookie);
2589 int vfs_follow_link(struct nameidata *nd, const char *link)
2591 return __vfs_follow_link(nd, link);
2594 /* get the link contents into pagecache */
2595 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2597 struct page * page;
2598 struct address_space *mapping = dentry->d_inode->i_mapping;
2599 page = read_mapping_page(mapping, 0, NULL);
2600 if (IS_ERR(page))
2601 goto sync_fail;
2602 wait_on_page_locked(page);
2603 if (!PageUptodate(page))
2604 goto async_fail;
2605 *ppage = page;
2606 return kmap(page);
2608 async_fail:
2609 page_cache_release(page);
2610 return ERR_PTR(-EIO);
2612 sync_fail:
2613 return (char*)page;
2616 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2618 struct page *page = NULL;
2619 char *s = page_getlink(dentry, &page);
2620 int res = vfs_readlink(dentry,buffer,buflen,s);
2621 if (page) {
2622 kunmap(page);
2623 page_cache_release(page);
2625 return res;
2628 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2630 struct page *page = NULL;
2631 nd_set_link(nd, page_getlink(dentry, &page));
2632 return page;
2635 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2637 struct page *page = cookie;
2639 if (page) {
2640 kunmap(page);
2641 page_cache_release(page);
2645 int __page_symlink(struct inode *inode, const char *symname, int len,
2646 gfp_t gfp_mask)
2648 struct address_space *mapping = inode->i_mapping;
2649 struct page *page;
2650 int err = -ENOMEM;
2651 char *kaddr;
2653 retry:
2654 page = find_or_create_page(mapping, 0, gfp_mask);
2655 if (!page)
2656 goto fail;
2657 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2658 if (err == AOP_TRUNCATED_PAGE) {
2659 page_cache_release(page);
2660 goto retry;
2662 if (err)
2663 goto fail_map;
2664 kaddr = kmap_atomic(page, KM_USER0);
2665 memcpy(kaddr, symname, len-1);
2666 kunmap_atomic(kaddr, KM_USER0);
2667 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2668 if (err == AOP_TRUNCATED_PAGE) {
2669 page_cache_release(page);
2670 goto retry;
2672 if (err)
2673 goto fail_map;
2674 /*
2675 * Notice that we are _not_ going to block here - end of page is
2676 * unmapped, so this will only try to map the rest of page, see
2677 * that it is unmapped (typically even will not look into inode -
2678 * ->i_size will be enough for everything) and zero it out.
2679 * OTOH it's obviously correct and should make the page up-to-date.
2680 */
2681 if (!PageUptodate(page)) {
2682 err = mapping->a_ops->readpage(NULL, page);
2683 if (err != AOP_TRUNCATED_PAGE)
2684 wait_on_page_locked(page);
2685 } else {
2686 unlock_page(page);
2688 page_cache_release(page);
2689 if (err < 0)
2690 goto fail;
2691 mark_inode_dirty(inode);
2692 return 0;
2693 fail_map:
2694 unlock_page(page);
2695 page_cache_release(page);
2696 fail:
2697 return err;
2700 int page_symlink(struct inode *inode, const char *symname, int len)
2702 return __page_symlink(inode, symname, len,
2703 mapping_gfp_mask(inode->i_mapping));
2706 struct inode_operations page_symlink_inode_operations = {
2707 .readlink = generic_readlink,
2708 .follow_link = page_follow_link_light,
2709 .put_link = page_put_link,
2710 };
2712 EXPORT_SYMBOL(__user_walk);
2713 EXPORT_SYMBOL(__user_walk_fd);
2714 EXPORT_SYMBOL(follow_down);
2715 EXPORT_SYMBOL(follow_up);
2716 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2717 EXPORT_SYMBOL(getname);
2718 EXPORT_SYMBOL(lock_rename);
2719 EXPORT_SYMBOL(lookup_one_len);
2720 EXPORT_SYMBOL(page_follow_link_light);
2721 EXPORT_SYMBOL(page_put_link);
2722 EXPORT_SYMBOL(page_readlink);
2723 EXPORT_SYMBOL(__page_symlink);
2724 EXPORT_SYMBOL(page_symlink);
2725 EXPORT_SYMBOL(page_symlink_inode_operations);
2726 EXPORT_SYMBOL(path_lookup);
2727 EXPORT_SYMBOL(path_release);
2728 EXPORT_SYMBOL(path_walk);
2729 EXPORT_SYMBOL(permission);
2730 EXPORT_SYMBOL(vfs_permission);
2731 EXPORT_SYMBOL(file_permission);
2732 EXPORT_SYMBOL(unlock_rename);
2733 EXPORT_SYMBOL(vfs_create);
2734 EXPORT_SYMBOL(vfs_follow_link);
2735 EXPORT_SYMBOL(vfs_link);
2736 EXPORT_SYMBOL(vfs_mkdir);
2737 EXPORT_SYMBOL(vfs_mknod);
2738 EXPORT_SYMBOL(generic_permission);
2739 EXPORT_SYMBOL(vfs_readlink);
2740 EXPORT_SYMBOL(vfs_rename);
2741 EXPORT_SYMBOL(vfs_rmdir);
2742 EXPORT_SYMBOL(vfs_symlink);
2743 EXPORT_SYMBOL(vfs_unlink);
2744 EXPORT_SYMBOL(dentry_unhash);
2745 EXPORT_SYMBOL(generic_readlink);