FreeBSD/Linux Kernel Cross Reference
sys/fs/namespace.c
1 /*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11 #include <linux/config.h>
12 #include <linux/slab.h>
13 #include <linux/smp_lock.h>
14 #include <linux/init.h>
15 #include <linux/quotaops.h>
16 #include <linux/acct.h>
17 #include <linux/module.h>
18
19 #include <asm/uaccess.h>
20
21 #include <linux/seq_file.h>
22 #include <linux/namespace.h>
23
24 struct vfsmount *do_kern_mount(const char *type, int flags, char *name, void *data);
25 int do_remount_sb(struct super_block *sb, int flags, void * data);
26 void kill_super(struct super_block *sb);
27 extern int __init init_rootfs(void);
28
29 static struct list_head *mount_hashtable;
30 static int hash_mask, hash_bits;
31 static kmem_cache_t *mnt_cache;
32
33 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
34 {
35 unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
36 tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
37 tmp = tmp + (tmp >> hash_bits);
38 return tmp & hash_mask;
39 }
40
41 struct vfsmount *alloc_vfsmnt(char *name)
42 {
43 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
44 if (mnt) {
45 memset(mnt, 0, sizeof(struct vfsmount));
46 atomic_set(&mnt->mnt_count,1);
47 INIT_LIST_HEAD(&mnt->mnt_hash);
48 INIT_LIST_HEAD(&mnt->mnt_child);
49 INIT_LIST_HEAD(&mnt->mnt_mounts);
50 INIT_LIST_HEAD(&mnt->mnt_list);
51 if (name) {
52 int size = strlen(name)+1;
53 char * newname = kmalloc(size, GFP_KERNEL);
54 if (newname) {
55 memcpy(newname, name, size);
56 mnt->mnt_devname = newname;
57 }
58 }
59 }
60 return mnt;
61 }
62
63 void free_vfsmnt(struct vfsmount *mnt)
64 {
65 if (mnt->mnt_devname)
66 kfree(mnt->mnt_devname);
67 kmem_cache_free(mnt_cache, mnt);
68 }
69
70 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
71 {
72 struct list_head * head = mount_hashtable + hash(mnt, dentry);
73 struct list_head * tmp = head;
74 struct vfsmount *p;
75
76 for (;;) {
77 tmp = tmp->next;
78 p = NULL;
79 if (tmp == head)
80 break;
81 p = list_entry(tmp, struct vfsmount, mnt_hash);
82 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry)
83 break;
84 }
85 return p;
86 }
87
88 static int check_mnt(struct vfsmount *mnt)
89 {
90 spin_lock(&dcache_lock);
91 while (mnt->mnt_parent != mnt)
92 mnt = mnt->mnt_parent;
93 spin_unlock(&dcache_lock);
94 return mnt == current->namespace->root;
95 }
96
97 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
98 {
99 old_nd->dentry = mnt->mnt_mountpoint;
100 old_nd->mnt = mnt->mnt_parent;
101 mnt->mnt_parent = mnt;
102 mnt->mnt_mountpoint = mnt->mnt_root;
103 list_del_init(&mnt->mnt_child);
104 list_del_init(&mnt->mnt_hash);
105 old_nd->dentry->d_mounted--;
106 }
107
108 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
109 {
110 mnt->mnt_parent = mntget(nd->mnt);
111 mnt->mnt_mountpoint = dget(nd->dentry);
112 list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
113 list_add(&mnt->mnt_child, &nd->mnt->mnt_mounts);
114 nd->dentry->d_mounted++;
115 }
116
117 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
118 {
119 struct list_head *next = p->mnt_mounts.next;
120 if (next == &p->mnt_mounts) {
121 while (1) {
122 if (p == root)
123 return NULL;
124 next = p->mnt_child.next;
125 if (next != &p->mnt_parent->mnt_mounts)
126 break;
127 p = p->mnt_parent;
128 }
129 }
130 return list_entry(next, struct vfsmount, mnt_child);
131 }
132
133 static struct vfsmount *
134 clone_mnt(struct vfsmount *old, struct dentry *root)
135 {
136 struct super_block *sb = old->mnt_sb;
137 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
138
139 if (mnt) {
140 mnt->mnt_flags = old->mnt_flags;
141 atomic_inc(&sb->s_active);
142 mnt->mnt_sb = sb;
143 mnt->mnt_root = dget(root);
144 mnt->mnt_mountpoint = mnt->mnt_root;
145 mnt->mnt_parent = mnt;
146 }
147 return mnt;
148 }
149
150 void __mntput(struct vfsmount *mnt)
151 {
152 struct super_block *sb = mnt->mnt_sb;
153 dput(mnt->mnt_root);
154 free_vfsmnt(mnt);
155 kill_super(sb);
156 }
157
158 /* iterator */
159 static void *m_start(struct seq_file *m, loff_t *pos)
160 {
161 struct namespace *n = m->private;
162 struct list_head *p;
163 loff_t l = *pos;
164
165 down_read(&n->sem);
166 list_for_each(p, &n->list)
167 if (!l--)
168 return list_entry(p, struct vfsmount, mnt_list);
169 return NULL;
170 }
171
172 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
173 {
174 struct namespace *n = m->private;
175 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
176 (*pos)++;
177 return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
178 }
179
180 static void m_stop(struct seq_file *m, void *v)
181 {
182 struct namespace *n = m->private;
183 up_read(&n->sem);
184 }
185
186 static inline void mangle(struct seq_file *m, const char *s)
187 {
188 seq_escape(m, s, " \t\n\\");
189 }
190
191 static int show_vfsmnt(struct seq_file *m, void *v)
192 {
193 struct vfsmount *mnt = v;
194 int err = 0;
195 static struct proc_fs_info {
196 int flag;
197 char *str;
198 } fs_info[] = {
199 { MS_SYNCHRONOUS, ",sync" },
200 { MS_MANDLOCK, ",mand" },
201 { MS_NOATIME, ",noatime" },
202 { MS_NODIRATIME, ",nodiratime" },
203 { 0, NULL }
204 };
205 static struct proc_fs_info mnt_info[] = {
206 { MNT_NOSUID, ",nosuid" },
207 { MNT_NODEV, ",nodev" },
208 { MNT_NOEXEC, ",noexec" },
209 { 0, NULL }
210 };
211 struct proc_fs_info *fs_infop;
212 char *path_buf, *path;
213
214 path_buf = (char *) __get_free_page(GFP_KERNEL);
215 if (!path_buf)
216 return -ENOMEM;
217 path = d_path(mnt->mnt_root, mnt, path_buf, PAGE_SIZE);
218 if (IS_ERR(path)) {
219 free_page((unsigned long) path_buf);
220 return PTR_ERR(path);
221 }
222
223 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
224 seq_putc(m, ' ');
225 mangle(m, path);
226 free_page((unsigned long) path_buf);
227 seq_putc(m, ' ');
228 mangle(m, mnt->mnt_sb->s_type->name);
229 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
230 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
231 if (mnt->mnt_sb->s_flags & fs_infop->flag)
232 seq_puts(m, fs_infop->str);
233 }
234 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
235 if (mnt->mnt_flags & fs_infop->flag)
236 seq_puts(m, fs_infop->str);
237 }
238 if (mnt->mnt_sb->s_op->show_options)
239 err = mnt->mnt_sb->s_op->show_options(m, mnt);
240 seq_puts(m, " 0 0\n");
241 return err;
242 }
243
244 struct seq_operations mounts_op = {
245 start: m_start,
246 next: m_next,
247 stop: m_stop,
248 show: show_vfsmnt
249 };
250
251 /*
252 * Doesn't take quota and stuff into account. IOW, in some cases it will
253 * give false negatives. The main reason why it's here is that we need
254 * a non-destructive way to look for easily umountable filesystems.
255 */
256 int may_umount(struct vfsmount *mnt)
257 {
258 if (atomic_read(&mnt->mnt_count) > 2)
259 return -EBUSY;
260 return 0;
261 }
262
263 void umount_tree(struct vfsmount *mnt)
264 {
265 struct vfsmount *p;
266 LIST_HEAD(kill);
267
268 for (p = mnt; p; p = next_mnt(p, mnt)) {
269 list_del(&p->mnt_list);
270 list_add(&p->mnt_list, &kill);
271 }
272
273 while (!list_empty(&kill)) {
274 mnt = list_entry(kill.next, struct vfsmount, mnt_list);
275 list_del_init(&mnt->mnt_list);
276 if (mnt->mnt_parent == mnt) {
277 spin_unlock(&dcache_lock);
278 } else {
279 struct nameidata old_nd;
280 detach_mnt(mnt, &old_nd);
281 spin_unlock(&dcache_lock);
282 path_release(&old_nd);
283 }
284 mntput(mnt);
285 spin_lock(&dcache_lock);
286 }
287 }
288
289 static int do_umount(struct vfsmount *mnt, int flags)
290 {
291 struct super_block * sb = mnt->mnt_sb;
292 int retval = 0;
293
294 /*
295 * If we may have to abort operations to get out of this
296 * mount, and they will themselves hold resources we must
297 * allow the fs to do things. In the Unix tradition of
298 * 'Gee thats tricky lets do it in userspace' the umount_begin
299 * might fail to complete on the first run through as other tasks
300 * must return, and the like. Thats for the mount program to worry
301 * about for the moment.
302 */
303
304 lock_kernel();
305 if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
306 sb->s_op->umount_begin(sb);
307 unlock_kernel();
308
309 /*
310 * No sense to grab the lock for this test, but test itself looks
311 * somewhat bogus. Suggestions for better replacement?
312 * Ho-hum... In principle, we might treat that as umount + switch
313 * to rootfs. GC would eventually take care of the old vfsmount.
314 * Actually it makes sense, especially if rootfs would contain a
315 * /reboot - static binary that would close all descriptors and
316 * call reboot(9). Then init(8) could umount root and exec /reboot.
317 */
318 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
319 /*
320 * Special case for "unmounting" root ...
321 * we just try to remount it readonly.
322 */
323 down_write(&sb->s_umount);
324 if (!(sb->s_flags & MS_RDONLY)) {
325 lock_kernel();
326 retval = do_remount_sb(sb, MS_RDONLY, 0);
327 unlock_kernel();
328 }
329 up_write(&sb->s_umount);
330 return retval;
331 }
332
333 down_write(¤t->namespace->sem);
334 spin_lock(&dcache_lock);
335
336 if (atomic_read(&sb->s_active) == 1) {
337 /* last instance - try to be smart */
338 spin_unlock(&dcache_lock);
339 lock_kernel();
340 DQUOT_OFF(sb);
341 acct_auto_close(sb->s_dev);
342 unlock_kernel();
343 spin_lock(&dcache_lock);
344 }
345 retval = -EBUSY;
346 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
347 if (!list_empty(&mnt->mnt_list))
348 umount_tree(mnt);
349 retval = 0;
350 }
351 spin_unlock(&dcache_lock);
352 up_write(¤t->namespace->sem);
353 return retval;
354 }
355
356 /*
357 * Now umount can handle mount points as well as block devices.
358 * This is important for filesystems which use unnamed block devices.
359 *
360 * We now support a flag for forced unmount like the other 'big iron'
361 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
362 */
363
364 asmlinkage long sys_umount(char * name, int flags)
365 {
366 struct nameidata nd;
367 int retval;
368
369 retval = __user_walk(name, LOOKUP_POSITIVE|LOOKUP_FOLLOW, &nd);
370 if (retval)
371 goto out;
372 retval = -EINVAL;
373 if (nd.dentry != nd.mnt->mnt_root)
374 goto dput_and_out;
375 if (!check_mnt(nd.mnt))
376 goto dput_and_out;
377
378 retval = -EPERM;
379 if (!capable(CAP_SYS_ADMIN))
380 goto dput_and_out;
381
382 retval = do_umount(nd.mnt, flags);
383 dput_and_out:
384 path_release(&nd);
385 out:
386 return retval;
387 }
388
389 /*
390 * The 2.0 compatible umount. No flags.
391 */
392
393 asmlinkage long sys_oldumount(char * name)
394 {
395 return sys_umount(name,0);
396 }
397
398 static int mount_is_safe(struct nameidata *nd)
399 {
400 if (capable(CAP_SYS_ADMIN))
401 return 0;
402 return -EPERM;
403 #ifdef notyet
404 if (S_ISLNK(nd->dentry->d_inode->i_mode))
405 return -EPERM;
406 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
407 if (current->uid != nd->dentry->d_inode->i_uid)
408 return -EPERM;
409 }
410 if (permission(nd->dentry->d_inode, MAY_WRITE))
411 return -EPERM;
412 return 0;
413 #endif
414 }
415
416 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
417 {
418 struct vfsmount *p, *next, *q, *res;
419 struct nameidata nd;
420
421 p = mnt;
422 res = nd.mnt = q = clone_mnt(p, dentry);
423 if (!q)
424 goto Enomem;
425 q->mnt_parent = q;
426 q->mnt_mountpoint = p->mnt_mountpoint;
427
428 while ( (next = next_mnt(p, mnt)) != NULL) {
429 while (p != next->mnt_parent) {
430 p = p->mnt_parent;
431 q = q->mnt_parent;
432 }
433 p = next;
434 nd.mnt = q;
435 nd.dentry = p->mnt_mountpoint;
436 q = clone_mnt(p, p->mnt_root);
437 if (!q)
438 goto Enomem;
439 spin_lock(&dcache_lock);
440 list_add_tail(&q->mnt_list, &res->mnt_list);
441 attach_mnt(q, &nd);
442 spin_unlock(&dcache_lock);
443 }
444 return res;
445 Enomem:
446 if (res) {
447 spin_lock(&dcache_lock);
448 umount_tree(res);
449 spin_unlock(&dcache_lock);
450 }
451 return NULL;
452 }
453
454 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
455 {
456 int err;
457 if (mnt->mnt_sb->s_flags & MS_NOUSER)
458 return -EINVAL;
459
460 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
461 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
462 return -ENOTDIR;
463
464 err = -ENOENT;
465 down(&nd->dentry->d_inode->i_zombie);
466 if (IS_DEADDIR(nd->dentry->d_inode))
467 goto out_unlock;
468
469 spin_lock(&dcache_lock);
470 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
471 struct list_head head;
472 attach_mnt(mnt, nd);
473 list_add_tail(&head, &mnt->mnt_list);
474 list_splice(&head, current->namespace->list.prev);
475 mntget(mnt);
476 err = 0;
477 }
478 spin_unlock(&dcache_lock);
479 out_unlock:
480 up(&nd->dentry->d_inode->i_zombie);
481 return err;
482 }
483
484 /*
485 * do loopback mount.
486 */
487 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
488 {
489 struct nameidata old_nd;
490 struct vfsmount *mnt = NULL;
491 int err = mount_is_safe(nd);
492 if (err)
493 return err;
494 if (!old_name || !*old_name)
495 return -EINVAL;
496 err = path_lookup(old_name, LOOKUP_POSITIVE|LOOKUP_FOLLOW, &old_nd);
497 if (err)
498 return err;
499
500 down_write(¤t->namespace->sem);
501 err = -EINVAL;
502 if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
503 err = -ENOMEM;
504 if (recurse)
505 mnt = copy_tree(old_nd.mnt, old_nd.dentry);
506 else
507 mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
508 }
509
510 if (mnt) {
511 err = graft_tree(mnt, nd);
512 if (err) {
513 spin_lock(&dcache_lock);
514 umount_tree(mnt);
515 spin_unlock(&dcache_lock);
516 } else
517 mntput(mnt);
518 }
519
520 up_write(¤t->namespace->sem);
521 path_release(&old_nd);
522 return err;
523 }
524
525 /*
526 * change filesystem flags. dir should be a physical root of filesystem.
527 * If you've mounted a non-root directory somewhere and want to do remount
528 * on it - tough luck.
529 */
530
531 static int do_remount(struct nameidata *nd,int flags,int mnt_flags,void *data)
532 {
533 int err;
534 struct super_block * sb = nd->mnt->mnt_sb;
535
536 if (!capable(CAP_SYS_ADMIN))
537 return -EPERM;
538
539 if (!check_mnt(nd->mnt))
540 return -EINVAL;
541
542 if (nd->dentry != nd->mnt->mnt_root)
543 return -EINVAL;
544
545 down_write(&sb->s_umount);
546 err = do_remount_sb(sb, flags, data);
547 if (!err)
548 nd->mnt->mnt_flags=mnt_flags;
549 up_write(&sb->s_umount);
550 return err;
551 }
552
553 static int do_move_mount(struct nameidata *nd, char *old_name)
554 {
555 struct nameidata old_nd, parent_nd;
556 struct vfsmount *p;
557 int err = 0;
558 if (!capable(CAP_SYS_ADMIN))
559 return -EPERM;
560 if (!old_name || !*old_name)
561 return -EINVAL;
562 err = path_lookup(old_name, LOOKUP_POSITIVE|LOOKUP_FOLLOW, &old_nd);
563 if (err)
564 return err;
565
566 down_write(¤t->namespace->sem);
567 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
568 ;
569 err = -EINVAL;
570 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
571 goto out;
572
573 err = -ENOENT;
574 down(&nd->dentry->d_inode->i_zombie);
575 if (IS_DEADDIR(nd->dentry->d_inode))
576 goto out1;
577
578 spin_lock(&dcache_lock);
579 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
580 goto out2;
581
582 err = -EINVAL;
583 if (old_nd.dentry != old_nd.mnt->mnt_root)
584 goto out2;
585
586 if (old_nd.mnt == old_nd.mnt->mnt_parent)
587 goto out2;
588
589 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
590 S_ISDIR(old_nd.dentry->d_inode->i_mode))
591 goto out2;
592
593 err = -ELOOP;
594 for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
595 if (p == old_nd.mnt)
596 goto out2;
597 err = 0;
598
599 detach_mnt(old_nd.mnt, &parent_nd);
600 attach_mnt(old_nd.mnt, nd);
601 out2:
602 spin_unlock(&dcache_lock);
603 out1:
604 up(&nd->dentry->d_inode->i_zombie);
605 out:
606 up_write(¤t->namespace->sem);
607 if (!err)
608 path_release(&parent_nd);
609 path_release(&old_nd);
610 return err;
611 }
612
613 static int do_add_mount(struct nameidata *nd, char *type, int flags,
614 int mnt_flags, char *name, void *data)
615 {
616 struct vfsmount *mnt;
617 int err;
618
619 if (!type || !memchr(type, 0, PAGE_SIZE))
620 return -EINVAL;
621
622 /* we need capabilities... */
623 if (!capable(CAP_SYS_ADMIN))
624 return -EPERM;
625
626 mnt = do_kern_mount(type, flags, name, data);
627 err = PTR_ERR(mnt);
628 if (IS_ERR(mnt))
629 goto out;
630
631 down_write(¤t->namespace->sem);
632 /* Something was mounted here while we slept */
633 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
634 ;
635 err = -EINVAL;
636 if (!check_mnt(nd->mnt))
637 goto unlock;
638
639 /* Refuse the same filesystem on the same mount point */
640 err = -EBUSY;
641 if (nd->mnt->mnt_sb == mnt->mnt_sb && nd->mnt->mnt_root == nd->dentry)
642 goto unlock;
643
644 mnt->mnt_flags = mnt_flags;
645 err = graft_tree(mnt, nd);
646 unlock:
647 up_write(¤t->namespace->sem);
648 mntput(mnt);
649 out:
650 return err;
651 }
652
653 static int copy_mount_options (const void *data, unsigned long *where)
654 {
655 int i;
656 unsigned long page;
657 unsigned long size;
658
659 *where = 0;
660 if (!data)
661 return 0;
662
663 if (!(page = __get_free_page(GFP_KERNEL)))
664 return -ENOMEM;
665
666 /* We only care that *some* data at the address the user
667 * gave us is valid. Just in case, we'll zero
668 * the remainder of the page.
669 */
670 /* copy_from_user cannot cross TASK_SIZE ! */
671 size = TASK_SIZE - (unsigned long)data;
672 if (size > PAGE_SIZE)
673 size = PAGE_SIZE;
674
675 i = size - copy_from_user((void *)page, data, size);
676 if (!i) {
677 free_page(page);
678 return -EFAULT;
679 }
680 if (i != PAGE_SIZE)
681 memset((char *)page + i, 0, PAGE_SIZE - i);
682 *where = page;
683 return 0;
684 }
685
686 /*
687 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
688 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
689 *
690 * data is a (void *) that can point to any structure up to
691 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
692 * information (or be NULL).
693 *
694 * Pre-0.97 versions of mount() didn't have a flags word.
695 * When the flags word was introduced its top half was required
696 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
697 * Therefore, if this magic number is present, it carries no information
698 * and must be discarded.
699 */
700 long do_mount(char * dev_name, char * dir_name, char *type_page,
701 unsigned long flags, void *data_page)
702 {
703 struct nameidata nd;
704 int retval = 0;
705 int mnt_flags = 0;
706
707 /* Discard magic */
708 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
709 flags &= ~MS_MGC_MSK;
710
711 /* Basic sanity checks */
712
713 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
714 return -EINVAL;
715 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
716 return -EINVAL;
717
718 /* Separate the per-mountpoint flags */
719 if (flags & MS_NOSUID)
720 mnt_flags |= MNT_NOSUID;
721 if (flags & MS_NODEV)
722 mnt_flags |= MNT_NODEV;
723 if (flags & MS_NOEXEC)
724 mnt_flags |= MNT_NOEXEC;
725 flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV);
726
727 /* ... and get the mountpoint */
728 retval = path_lookup(dir_name, LOOKUP_FOLLOW|LOOKUP_POSITIVE, &nd);
729 if (retval)
730 return retval;
731
732 if (flags & MS_REMOUNT)
733 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
734 data_page);
735 else if (flags & MS_BIND)
736 retval = do_loopback(&nd, dev_name, flags & MS_REC);
737 else if (flags & MS_MOVE)
738 retval = do_move_mount(&nd, dev_name);
739 else
740 retval = do_add_mount(&nd, type_page, flags, mnt_flags,
741 dev_name, data_page);
742 path_release(&nd);
743 return retval;
744 }
745
746 int copy_namespace(int flags, struct task_struct *tsk)
747 {
748 struct namespace *namespace = tsk->namespace;
749 struct namespace *new_ns;
750 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
751 struct fs_struct *fs = tsk->fs;
752
753 if (!namespace)
754 return 0;
755
756 get_namespace(namespace);
757
758 if (! (flags & CLONE_NEWNS))
759 return 0;
760
761 if (!capable(CAP_SYS_ADMIN)) {
762 put_namespace(namespace);
763 return -EPERM;
764 }
765
766 new_ns = kmalloc(sizeof(struct namespace *), GFP_KERNEL);
767 if (!new_ns)
768 goto out;
769
770 atomic_set(&new_ns->count, 1);
771 init_rwsem(&new_ns->sem);
772 new_ns->root = NULL;
773 INIT_LIST_HEAD(&new_ns->list);
774
775 down_write(&tsk->namespace->sem);
776 /* First pass: copy the tree topology */
777 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
778 spin_lock(&dcache_lock);
779 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
780 spin_unlock(&dcache_lock);
781
782 /* Second pass: switch the tsk->fs->* elements */
783 if (fs) {
784 struct vfsmount *p, *q;
785 write_lock(&fs->lock);
786
787 p = namespace->root;
788 q = new_ns->root;
789 while (p) {
790 if (p == fs->rootmnt) {
791 rootmnt = p;
792 fs->rootmnt = mntget(q);
793 }
794 if (p == fs->pwdmnt) {
795 pwdmnt = p;
796 fs->pwdmnt = mntget(q);
797 }
798 if (p == fs->altrootmnt) {
799 altrootmnt = p;
800 fs->altrootmnt = mntget(q);
801 }
802 p = next_mnt(p, namespace->root);
803 q = next_mnt(q, new_ns->root);
804 }
805 write_unlock(&fs->lock);
806 }
807 up_write(&tsk->namespace->sem);
808
809 tsk->namespace = new_ns;
810
811 if (rootmnt)
812 mntput(rootmnt);
813 if (pwdmnt)
814 mntput(pwdmnt);
815 if (altrootmnt)
816 mntput(altrootmnt);
817
818 put_namespace(namespace);
819 return 0;
820
821 out:
822 put_namespace(namespace);
823 return -ENOMEM;
824 }
825
826 asmlinkage long sys_mount(char * dev_name, char * dir_name, char * type,
827 unsigned long flags, void * data)
828 {
829 int retval;
830 unsigned long data_page;
831 unsigned long type_page;
832 unsigned long dev_page;
833 char *dir_page;
834
835 retval = copy_mount_options (type, &type_page);
836 if (retval < 0)
837 return retval;
838
839 dir_page = getname(dir_name);
840 retval = PTR_ERR(dir_page);
841 if (IS_ERR(dir_page))
842 goto out1;
843
844 retval = copy_mount_options (dev_name, &dev_page);
845 if (retval < 0)
846 goto out2;
847
848 retval = copy_mount_options (data, &data_page);
849 if (retval < 0)
850 goto out3;
851
852 lock_kernel();
853 retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
854 flags, (void*)data_page);
855 unlock_kernel();
856 free_page(data_page);
857
858 out3:
859 free_page(dev_page);
860 out2:
861 putname(dir_page);
862 out1:
863 free_page(type_page);
864 return retval;
865 }
866
867 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
868 {
869 struct task_struct *p;
870 struct fs_struct *fs;
871
872 read_lock(&tasklist_lock);
873 for_each_task(p) {
874 task_lock(p);
875 fs = p->fs;
876 if (fs) {
877 atomic_inc(&fs->count);
878 task_unlock(p);
879 if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
880 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
881 if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
882 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
883 put_fs_struct(fs);
884 } else
885 task_unlock(p);
886 }
887 read_unlock(&tasklist_lock);
888 }
889
890 /*
891 * Moves the current root to put_root, and sets root/cwd of all processes
892 * which had them on the old root to new_root.
893 *
894 * Note:
895 * - we don't move root/cwd if they are not at the root (reason: if something
896 * cared enough to change them, it's probably wrong to force them elsewhere)
897 * - it's okay to pick a root that isn't the root of a file system, e.g.
898 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
899 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
900 * first.
901 */
902
903 asmlinkage long sys_pivot_root(const char *new_root, const char *put_old)
904 {
905 struct vfsmount *tmp;
906 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
907 int error;
908
909 if (!capable(CAP_SYS_ADMIN))
910 return -EPERM;
911
912 lock_kernel();
913
914 error = __user_walk(new_root, LOOKUP_POSITIVE|LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
915 if (error)
916 goto out0;
917 error = -EINVAL;
918 if (!check_mnt(new_nd.mnt))
919 goto out1;
920
921 error = __user_walk(put_old, LOOKUP_POSITIVE|LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
922 if (error)
923 goto out1;
924
925 read_lock(¤t->fs->lock);
926 user_nd.mnt = mntget(current->fs->rootmnt);
927 user_nd.dentry = dget(current->fs->root);
928 read_unlock(¤t->fs->lock);
929 down_write(¤t->namespace->sem);
930 down(&old_nd.dentry->d_inode->i_zombie);
931 error = -EINVAL;
932 if (!check_mnt(user_nd.mnt))
933 goto out2;
934 error = -ENOENT;
935 if (IS_DEADDIR(new_nd.dentry->d_inode))
936 goto out2;
937 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
938 goto out2;
939 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
940 goto out2;
941 error = -EBUSY;
942 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
943 goto out2; /* loop */
944 error = -EINVAL;
945 if (user_nd.mnt->mnt_root != user_nd.dentry)
946 goto out2;
947 if (new_nd.mnt->mnt_root != new_nd.dentry)
948 goto out2; /* not a mountpoint */
949 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
950 spin_lock(&dcache_lock);
951 if (tmp != new_nd.mnt) {
952 for (;;) {
953 if (tmp->mnt_parent == tmp)
954 goto out3;
955 if (tmp->mnt_parent == new_nd.mnt)
956 break;
957 tmp = tmp->mnt_parent;
958 }
959 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
960 goto out3;
961 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
962 goto out3;
963 detach_mnt(new_nd.mnt, &parent_nd);
964 detach_mnt(user_nd.mnt, &root_parent);
965 attach_mnt(user_nd.mnt, &old_nd);
966 attach_mnt(new_nd.mnt, &root_parent);
967 spin_unlock(&dcache_lock);
968 chroot_fs_refs(&user_nd, &new_nd);
969 error = 0;
970 path_release(&root_parent);
971 path_release(&parent_nd);
972 out2:
973 up(&old_nd.dentry->d_inode->i_zombie);
974 up_write(¤t->namespace->sem);
975 path_release(&user_nd);
976 path_release(&old_nd);
977 out1:
978 path_release(&new_nd);
979 out0:
980 unlock_kernel();
981 return error;
982 out3:
983 spin_unlock(&dcache_lock);
984 goto out2;
985 }
986
987 static void __init init_mount_tree(void)
988 {
989 struct vfsmount *mnt;
990 struct namespace *namespace;
991 struct task_struct *p;
992
993 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
994 if (IS_ERR(mnt))
995 panic("Can't create rootfs");
996 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
997 if (!namespace)
998 panic("Can't allocate initial namespace");
999 atomic_set(&namespace->count, 1);
1000 INIT_LIST_HEAD(&namespace->list);
1001 init_rwsem(&namespace->sem);
1002 list_add(&mnt->mnt_list, &namespace->list);
1003 namespace->root = mnt;
1004
1005 init_task.namespace = namespace;
1006 read_lock(&tasklist_lock);
1007 for_each_task(p) {
1008 get_namespace(namespace);
1009 p->namespace = namespace;
1010 }
1011 read_unlock(&tasklist_lock);
1012
1013 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1014 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1015 }
1016
1017 void __init mnt_init(unsigned long mempages)
1018 {
1019 struct list_head *d;
1020 unsigned long order;
1021 unsigned int nr_hash;
1022 int i;
1023
1024 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1025 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
1026 if (!mnt_cache)
1027 panic("Cannot create vfsmount cache");
1028
1029 /* using single pointer list heads would save half of the hash table. */
1030 order = 0;
1031 mount_hashtable = (struct list_head *)
1032 __get_free_pages(GFP_ATOMIC, order);
1033
1034 if (!mount_hashtable)
1035 panic("Failed to allocate mount hash table\n");
1036
1037 /*
1038 * Find the power-of-two list-heads that can fit into the allocation..
1039 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1040 * a power-of-two.
1041 */
1042 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct list_head);
1043 hash_bits = 0;
1044 do {
1045 hash_bits++;
1046 } while ((nr_hash >> hash_bits) != 0);
1047 hash_bits--;
1048
1049 /*
1050 * Re-calculate the actual number of entries and the mask
1051 * from the number of bits we can fit.
1052 */
1053 nr_hash = 1UL << hash_bits;
1054 hash_mask = nr_hash-1;
1055
1056 printk(KERN_INFO "Mount cache hash table entries: %d"
1057 " (order: %ld, %ld bytes)\n",
1058 nr_hash, order, (PAGE_SIZE << order));
1059
1060 /* And initialize the newly allocated array */
1061 d = mount_hashtable;
1062 i = nr_hash;
1063 do {
1064 INIT_LIST_HEAD(d);
1065 d++;
1066 i--;
1067 } while (i);
1068 init_rootfs();
1069 init_mount_tree();
1070 }
Cache object: 03947f7fae24dcbee05f0c5e4697b5c0
|