]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/namespace.c
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[mirror_ubuntu-bionic-kernel.git] / 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/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <linux/log2.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
31 #include "pnode.h"
32 #include "internal.h"
33
34 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35 #define HASH_SIZE (1UL << HASH_SHIFT)
36
37 /* spinlock for vfsmount related operations, inplace of dcache_lock */
38 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
39
40 static int event;
41
42 static struct list_head *mount_hashtable __read_mostly;
43 static struct kmem_cache *mnt_cache __read_mostly;
44 static struct rw_semaphore namespace_sem;
45
46 /* /sys/fs */
47 struct kobject *fs_kobj;
48 EXPORT_SYMBOL_GPL(fs_kobj);
49
50 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
51 {
52 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
53 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
54 tmp = tmp + (tmp >> HASH_SHIFT);
55 return tmp & (HASH_SIZE - 1);
56 }
57
58 struct vfsmount *alloc_vfsmnt(const char *name)
59 {
60 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
61 if (mnt) {
62 atomic_set(&mnt->mnt_count, 1);
63 INIT_LIST_HEAD(&mnt->mnt_hash);
64 INIT_LIST_HEAD(&mnt->mnt_child);
65 INIT_LIST_HEAD(&mnt->mnt_mounts);
66 INIT_LIST_HEAD(&mnt->mnt_list);
67 INIT_LIST_HEAD(&mnt->mnt_expire);
68 INIT_LIST_HEAD(&mnt->mnt_share);
69 INIT_LIST_HEAD(&mnt->mnt_slave_list);
70 INIT_LIST_HEAD(&mnt->mnt_slave);
71 if (name) {
72 int size = strlen(name) + 1;
73 char *newname = kmalloc(size, GFP_KERNEL);
74 if (newname) {
75 memcpy(newname, name, size);
76 mnt->mnt_devname = newname;
77 }
78 }
79 }
80 return mnt;
81 }
82
83 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
84 {
85 mnt->mnt_sb = sb;
86 mnt->mnt_root = dget(sb->s_root);
87 return 0;
88 }
89
90 EXPORT_SYMBOL(simple_set_mnt);
91
92 void free_vfsmnt(struct vfsmount *mnt)
93 {
94 kfree(mnt->mnt_devname);
95 kmem_cache_free(mnt_cache, mnt);
96 }
97
98 /*
99 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
101 */
102 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
103 int dir)
104 {
105 struct list_head *head = mount_hashtable + hash(mnt, dentry);
106 struct list_head *tmp = head;
107 struct vfsmount *p, *found = NULL;
108
109 for (;;) {
110 tmp = dir ? tmp->next : tmp->prev;
111 p = NULL;
112 if (tmp == head)
113 break;
114 p = list_entry(tmp, struct vfsmount, mnt_hash);
115 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
116 found = p;
117 break;
118 }
119 }
120 return found;
121 }
122
123 /*
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
126 */
127 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
128 {
129 struct vfsmount *child_mnt;
130 spin_lock(&vfsmount_lock);
131 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 mntget(child_mnt);
133 spin_unlock(&vfsmount_lock);
134 return child_mnt;
135 }
136
137 static inline int check_mnt(struct vfsmount *mnt)
138 {
139 return mnt->mnt_ns == current->nsproxy->mnt_ns;
140 }
141
142 static void touch_mnt_namespace(struct mnt_namespace *ns)
143 {
144 if (ns) {
145 ns->event = ++event;
146 wake_up_interruptible(&ns->poll);
147 }
148 }
149
150 static void __touch_mnt_namespace(struct mnt_namespace *ns)
151 {
152 if (ns && ns->event != event) {
153 ns->event = event;
154 wake_up_interruptible(&ns->poll);
155 }
156 }
157
158 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
159 {
160 old_nd->dentry = mnt->mnt_mountpoint;
161 old_nd->mnt = mnt->mnt_parent;
162 mnt->mnt_parent = mnt;
163 mnt->mnt_mountpoint = mnt->mnt_root;
164 list_del_init(&mnt->mnt_child);
165 list_del_init(&mnt->mnt_hash);
166 old_nd->dentry->d_mounted--;
167 }
168
169 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
170 struct vfsmount *child_mnt)
171 {
172 child_mnt->mnt_parent = mntget(mnt);
173 child_mnt->mnt_mountpoint = dget(dentry);
174 dentry->d_mounted++;
175 }
176
177 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
178 {
179 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
181 hash(nd->mnt, nd->dentry));
182 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
183 }
184
185 /*
186 * the caller must hold vfsmount_lock
187 */
188 static void commit_tree(struct vfsmount *mnt)
189 {
190 struct vfsmount *parent = mnt->mnt_parent;
191 struct vfsmount *m;
192 LIST_HEAD(head);
193 struct mnt_namespace *n = parent->mnt_ns;
194
195 BUG_ON(parent == mnt);
196
197 list_add_tail(&head, &mnt->mnt_list);
198 list_for_each_entry(m, &head, mnt_list)
199 m->mnt_ns = n;
200 list_splice(&head, n->list.prev);
201
202 list_add_tail(&mnt->mnt_hash, mount_hashtable +
203 hash(parent, mnt->mnt_mountpoint));
204 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
205 touch_mnt_namespace(n);
206 }
207
208 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
209 {
210 struct list_head *next = p->mnt_mounts.next;
211 if (next == &p->mnt_mounts) {
212 while (1) {
213 if (p == root)
214 return NULL;
215 next = p->mnt_child.next;
216 if (next != &p->mnt_parent->mnt_mounts)
217 break;
218 p = p->mnt_parent;
219 }
220 }
221 return list_entry(next, struct vfsmount, mnt_child);
222 }
223
224 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
225 {
226 struct list_head *prev = p->mnt_mounts.prev;
227 while (prev != &p->mnt_mounts) {
228 p = list_entry(prev, struct vfsmount, mnt_child);
229 prev = p->mnt_mounts.prev;
230 }
231 return p;
232 }
233
234 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
235 int flag)
236 {
237 struct super_block *sb = old->mnt_sb;
238 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
239
240 if (mnt) {
241 mnt->mnt_flags = old->mnt_flags;
242 atomic_inc(&sb->s_active);
243 mnt->mnt_sb = sb;
244 mnt->mnt_root = dget(root);
245 mnt->mnt_mountpoint = mnt->mnt_root;
246 mnt->mnt_parent = mnt;
247
248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
252 } else if (!(flag & CL_PRIVATE)) {
253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
258 }
259 if (flag & CL_MAKE_SHARED)
260 set_mnt_shared(mnt);
261
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
269 }
270 }
271 return mnt;
272 }
273
274 static inline void __mntput(struct vfsmount *mnt)
275 {
276 struct super_block *sb = mnt->mnt_sb;
277 dput(mnt->mnt_root);
278 free_vfsmnt(mnt);
279 deactivate_super(sb);
280 }
281
282 void mntput_no_expire(struct vfsmount *mnt)
283 {
284 repeat:
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
288 __mntput(mnt);
289 return;
290 }
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
292 mnt->mnt_pinned = 0;
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
296 goto repeat;
297 }
298 }
299
300 EXPORT_SYMBOL(mntput_no_expire);
301
302 void mnt_pin(struct vfsmount *mnt)
303 {
304 spin_lock(&vfsmount_lock);
305 mnt->mnt_pinned++;
306 spin_unlock(&vfsmount_lock);
307 }
308
309 EXPORT_SYMBOL(mnt_pin);
310
311 void mnt_unpin(struct vfsmount *mnt)
312 {
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
316 mnt->mnt_pinned--;
317 }
318 spin_unlock(&vfsmount_lock);
319 }
320
321 EXPORT_SYMBOL(mnt_unpin);
322
323 static inline void mangle(struct seq_file *m, const char *s)
324 {
325 seq_escape(m, s, " \t\n\\");
326 }
327
328 /*
329 * Simple .show_options callback for filesystems which don't want to
330 * implement more complex mount option showing.
331 *
332 * See also save_mount_options().
333 */
334 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
335 {
336 const char *options = mnt->mnt_sb->s_options;
337
338 if (options != NULL && options[0]) {
339 seq_putc(m, ',');
340 mangle(m, options);
341 }
342
343 return 0;
344 }
345 EXPORT_SYMBOL(generic_show_options);
346
347 /*
348 * If filesystem uses generic_show_options(), this function should be
349 * called from the fill_super() callback.
350 *
351 * The .remount_fs callback usually needs to be handled in a special
352 * way, to make sure, that previous options are not overwritten if the
353 * remount fails.
354 *
355 * Also note, that if the filesystem's .remount_fs function doesn't
356 * reset all options to their default value, but changes only newly
357 * given options, then the displayed options will not reflect reality
358 * any more.
359 */
360 void save_mount_options(struct super_block *sb, char *options)
361 {
362 kfree(sb->s_options);
363 sb->s_options = kstrdup(options, GFP_KERNEL);
364 }
365 EXPORT_SYMBOL(save_mount_options);
366
367 /* iterator */
368 static void *m_start(struct seq_file *m, loff_t *pos)
369 {
370 struct mnt_namespace *n = m->private;
371
372 down_read(&namespace_sem);
373 return seq_list_start(&n->list, *pos);
374 }
375
376 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
377 {
378 struct mnt_namespace *n = m->private;
379
380 return seq_list_next(v, &n->list, pos);
381 }
382
383 static void m_stop(struct seq_file *m, void *v)
384 {
385 up_read(&namespace_sem);
386 }
387
388 static int show_vfsmnt(struct seq_file *m, void *v)
389 {
390 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
391 int err = 0;
392 static struct proc_fs_info {
393 int flag;
394 char *str;
395 } fs_info[] = {
396 { MS_SYNCHRONOUS, ",sync" },
397 { MS_DIRSYNC, ",dirsync" },
398 { MS_MANDLOCK, ",mand" },
399 { 0, NULL }
400 };
401 static struct proc_fs_info mnt_info[] = {
402 { MNT_NOSUID, ",nosuid" },
403 { MNT_NODEV, ",nodev" },
404 { MNT_NOEXEC, ",noexec" },
405 { MNT_NOATIME, ",noatime" },
406 { MNT_NODIRATIME, ",nodiratime" },
407 { MNT_RELATIME, ",relatime" },
408 { 0, NULL }
409 };
410 struct proc_fs_info *fs_infop;
411
412 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
413 seq_putc(m, ' ');
414 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
415 seq_putc(m, ' ');
416 mangle(m, mnt->mnt_sb->s_type->name);
417 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
418 seq_putc(m, '.');
419 mangle(m, mnt->mnt_sb->s_subtype);
420 }
421 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
422 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
423 if (mnt->mnt_sb->s_flags & fs_infop->flag)
424 seq_puts(m, fs_infop->str);
425 }
426 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
427 if (mnt->mnt_flags & fs_infop->flag)
428 seq_puts(m, fs_infop->str);
429 }
430 if (mnt->mnt_sb->s_op->show_options)
431 err = mnt->mnt_sb->s_op->show_options(m, mnt);
432 seq_puts(m, " 0 0\n");
433 return err;
434 }
435
436 struct seq_operations mounts_op = {
437 .start = m_start,
438 .next = m_next,
439 .stop = m_stop,
440 .show = show_vfsmnt
441 };
442
443 static int show_vfsstat(struct seq_file *m, void *v)
444 {
445 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
446 int err = 0;
447
448 /* device */
449 if (mnt->mnt_devname) {
450 seq_puts(m, "device ");
451 mangle(m, mnt->mnt_devname);
452 } else
453 seq_puts(m, "no device");
454
455 /* mount point */
456 seq_puts(m, " mounted on ");
457 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
458 seq_putc(m, ' ');
459
460 /* file system type */
461 seq_puts(m, "with fstype ");
462 mangle(m, mnt->mnt_sb->s_type->name);
463
464 /* optional statistics */
465 if (mnt->mnt_sb->s_op->show_stats) {
466 seq_putc(m, ' ');
467 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
468 }
469
470 seq_putc(m, '\n');
471 return err;
472 }
473
474 struct seq_operations mountstats_op = {
475 .start = m_start,
476 .next = m_next,
477 .stop = m_stop,
478 .show = show_vfsstat,
479 };
480
481 /**
482 * may_umount_tree - check if a mount tree is busy
483 * @mnt: root of mount tree
484 *
485 * This is called to check if a tree of mounts has any
486 * open files, pwds, chroots or sub mounts that are
487 * busy.
488 */
489 int may_umount_tree(struct vfsmount *mnt)
490 {
491 int actual_refs = 0;
492 int minimum_refs = 0;
493 struct vfsmount *p;
494
495 spin_lock(&vfsmount_lock);
496 for (p = mnt; p; p = next_mnt(p, mnt)) {
497 actual_refs += atomic_read(&p->mnt_count);
498 minimum_refs += 2;
499 }
500 spin_unlock(&vfsmount_lock);
501
502 if (actual_refs > minimum_refs)
503 return 0;
504
505 return 1;
506 }
507
508 EXPORT_SYMBOL(may_umount_tree);
509
510 /**
511 * may_umount - check if a mount point is busy
512 * @mnt: root of mount
513 *
514 * This is called to check if a mount point has any
515 * open files, pwds, chroots or sub mounts. If the
516 * mount has sub mounts this will return busy
517 * regardless of whether the sub mounts are busy.
518 *
519 * Doesn't take quota and stuff into account. IOW, in some cases it will
520 * give false negatives. The main reason why it's here is that we need
521 * a non-destructive way to look for easily umountable filesystems.
522 */
523 int may_umount(struct vfsmount *mnt)
524 {
525 int ret = 1;
526 spin_lock(&vfsmount_lock);
527 if (propagate_mount_busy(mnt, 2))
528 ret = 0;
529 spin_unlock(&vfsmount_lock);
530 return ret;
531 }
532
533 EXPORT_SYMBOL(may_umount);
534
535 void release_mounts(struct list_head *head)
536 {
537 struct vfsmount *mnt;
538 while (!list_empty(head)) {
539 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
540 list_del_init(&mnt->mnt_hash);
541 if (mnt->mnt_parent != mnt) {
542 struct dentry *dentry;
543 struct vfsmount *m;
544 spin_lock(&vfsmount_lock);
545 dentry = mnt->mnt_mountpoint;
546 m = mnt->mnt_parent;
547 mnt->mnt_mountpoint = mnt->mnt_root;
548 mnt->mnt_parent = mnt;
549 spin_unlock(&vfsmount_lock);
550 dput(dentry);
551 mntput(m);
552 }
553 mntput(mnt);
554 }
555 }
556
557 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
558 {
559 struct vfsmount *p;
560
561 for (p = mnt; p; p = next_mnt(p, mnt))
562 list_move(&p->mnt_hash, kill);
563
564 if (propagate)
565 propagate_umount(kill);
566
567 list_for_each_entry(p, kill, mnt_hash) {
568 list_del_init(&p->mnt_expire);
569 list_del_init(&p->mnt_list);
570 __touch_mnt_namespace(p->mnt_ns);
571 p->mnt_ns = NULL;
572 list_del_init(&p->mnt_child);
573 if (p->mnt_parent != p)
574 p->mnt_mountpoint->d_mounted--;
575 change_mnt_propagation(p, MS_PRIVATE);
576 }
577 }
578
579 static int do_umount(struct vfsmount *mnt, int flags)
580 {
581 struct super_block *sb = mnt->mnt_sb;
582 int retval;
583 LIST_HEAD(umount_list);
584
585 retval = security_sb_umount(mnt, flags);
586 if (retval)
587 return retval;
588
589 /*
590 * Allow userspace to request a mountpoint be expired rather than
591 * unmounting unconditionally. Unmount only happens if:
592 * (1) the mark is already set (the mark is cleared by mntput())
593 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
594 */
595 if (flags & MNT_EXPIRE) {
596 if (mnt == current->fs->rootmnt ||
597 flags & (MNT_FORCE | MNT_DETACH))
598 return -EINVAL;
599
600 if (atomic_read(&mnt->mnt_count) != 2)
601 return -EBUSY;
602
603 if (!xchg(&mnt->mnt_expiry_mark, 1))
604 return -EAGAIN;
605 }
606
607 /*
608 * If we may have to abort operations to get out of this
609 * mount, and they will themselves hold resources we must
610 * allow the fs to do things. In the Unix tradition of
611 * 'Gee thats tricky lets do it in userspace' the umount_begin
612 * might fail to complete on the first run through as other tasks
613 * must return, and the like. Thats for the mount program to worry
614 * about for the moment.
615 */
616
617 lock_kernel();
618 if (sb->s_op->umount_begin)
619 sb->s_op->umount_begin(mnt, flags);
620 unlock_kernel();
621
622 /*
623 * No sense to grab the lock for this test, but test itself looks
624 * somewhat bogus. Suggestions for better replacement?
625 * Ho-hum... In principle, we might treat that as umount + switch
626 * to rootfs. GC would eventually take care of the old vfsmount.
627 * Actually it makes sense, especially if rootfs would contain a
628 * /reboot - static binary that would close all descriptors and
629 * call reboot(9). Then init(8) could umount root and exec /reboot.
630 */
631 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
632 /*
633 * Special case for "unmounting" root ...
634 * we just try to remount it readonly.
635 */
636 down_write(&sb->s_umount);
637 if (!(sb->s_flags & MS_RDONLY)) {
638 lock_kernel();
639 DQUOT_OFF(sb);
640 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
641 unlock_kernel();
642 }
643 up_write(&sb->s_umount);
644 return retval;
645 }
646
647 down_write(&namespace_sem);
648 spin_lock(&vfsmount_lock);
649 event++;
650
651 retval = -EBUSY;
652 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
653 if (!list_empty(&mnt->mnt_list))
654 umount_tree(mnt, 1, &umount_list);
655 retval = 0;
656 }
657 spin_unlock(&vfsmount_lock);
658 if (retval)
659 security_sb_umount_busy(mnt);
660 up_write(&namespace_sem);
661 release_mounts(&umount_list);
662 return retval;
663 }
664
665 /*
666 * Now umount can handle mount points as well as block devices.
667 * This is important for filesystems which use unnamed block devices.
668 *
669 * We now support a flag for forced unmount like the other 'big iron'
670 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
671 */
672
673 asmlinkage long sys_umount(char __user * name, int flags)
674 {
675 struct nameidata nd;
676 int retval;
677
678 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
679 if (retval)
680 goto out;
681 retval = -EINVAL;
682 if (nd.dentry != nd.mnt->mnt_root)
683 goto dput_and_out;
684 if (!check_mnt(nd.mnt))
685 goto dput_and_out;
686
687 retval = -EPERM;
688 if (!capable(CAP_SYS_ADMIN))
689 goto dput_and_out;
690
691 retval = do_umount(nd.mnt, flags);
692 dput_and_out:
693 path_release_on_umount(&nd);
694 out:
695 return retval;
696 }
697
698 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
699
700 /*
701 * The 2.0 compatible umount. No flags.
702 */
703 asmlinkage long sys_oldumount(char __user * name)
704 {
705 return sys_umount(name, 0);
706 }
707
708 #endif
709
710 static int mount_is_safe(struct nameidata *nd)
711 {
712 if (capable(CAP_SYS_ADMIN))
713 return 0;
714 return -EPERM;
715 #ifdef notyet
716 if (S_ISLNK(nd->dentry->d_inode->i_mode))
717 return -EPERM;
718 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
719 if (current->uid != nd->dentry->d_inode->i_uid)
720 return -EPERM;
721 }
722 if (vfs_permission(nd, MAY_WRITE))
723 return -EPERM;
724 return 0;
725 #endif
726 }
727
728 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
729 {
730 while (1) {
731 if (d == dentry)
732 return 1;
733 if (d == NULL || d == d->d_parent)
734 return 0;
735 d = d->d_parent;
736 }
737 }
738
739 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
740 int flag)
741 {
742 struct vfsmount *res, *p, *q, *r, *s;
743 struct nameidata nd;
744
745 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
746 return NULL;
747
748 res = q = clone_mnt(mnt, dentry, flag);
749 if (!q)
750 goto Enomem;
751 q->mnt_mountpoint = mnt->mnt_mountpoint;
752
753 p = mnt;
754 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
755 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
756 continue;
757
758 for (s = r; s; s = next_mnt(s, r)) {
759 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
760 s = skip_mnt_tree(s);
761 continue;
762 }
763 while (p != s->mnt_parent) {
764 p = p->mnt_parent;
765 q = q->mnt_parent;
766 }
767 p = s;
768 nd.mnt = q;
769 nd.dentry = p->mnt_mountpoint;
770 q = clone_mnt(p, p->mnt_root, flag);
771 if (!q)
772 goto Enomem;
773 spin_lock(&vfsmount_lock);
774 list_add_tail(&q->mnt_list, &res->mnt_list);
775 attach_mnt(q, &nd);
776 spin_unlock(&vfsmount_lock);
777 }
778 }
779 return res;
780 Enomem:
781 if (res) {
782 LIST_HEAD(umount_list);
783 spin_lock(&vfsmount_lock);
784 umount_tree(res, 0, &umount_list);
785 spin_unlock(&vfsmount_lock);
786 release_mounts(&umount_list);
787 }
788 return NULL;
789 }
790
791 struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
792 {
793 struct vfsmount *tree;
794 down_read(&namespace_sem);
795 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
796 up_read(&namespace_sem);
797 return tree;
798 }
799
800 void drop_collected_mounts(struct vfsmount *mnt)
801 {
802 LIST_HEAD(umount_list);
803 down_read(&namespace_sem);
804 spin_lock(&vfsmount_lock);
805 umount_tree(mnt, 0, &umount_list);
806 spin_unlock(&vfsmount_lock);
807 up_read(&namespace_sem);
808 release_mounts(&umount_list);
809 }
810
811 /*
812 * @source_mnt : mount tree to be attached
813 * @nd : place the mount tree @source_mnt is attached
814 * @parent_nd : if non-null, detach the source_mnt from its parent and
815 * store the parent mount and mountpoint dentry.
816 * (done when source_mnt is moved)
817 *
818 * NOTE: in the table below explains the semantics when a source mount
819 * of a given type is attached to a destination mount of a given type.
820 * ---------------------------------------------------------------------------
821 * | BIND MOUNT OPERATION |
822 * |**************************************************************************
823 * | source-->| shared | private | slave | unbindable |
824 * | dest | | | | |
825 * | | | | | | |
826 * | v | | | | |
827 * |**************************************************************************
828 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
829 * | | | | | |
830 * |non-shared| shared (+) | private | slave (*) | invalid |
831 * ***************************************************************************
832 * A bind operation clones the source mount and mounts the clone on the
833 * destination mount.
834 *
835 * (++) the cloned mount is propagated to all the mounts in the propagation
836 * tree of the destination mount and the cloned mount is added to
837 * the peer group of the source mount.
838 * (+) the cloned mount is created under the destination mount and is marked
839 * as shared. The cloned mount is added to the peer group of the source
840 * mount.
841 * (+++) the mount is propagated to all the mounts in the propagation tree
842 * of the destination mount and the cloned mount is made slave
843 * of the same master as that of the source mount. The cloned mount
844 * is marked as 'shared and slave'.
845 * (*) the cloned mount is made a slave of the same master as that of the
846 * source mount.
847 *
848 * ---------------------------------------------------------------------------
849 * | MOVE MOUNT OPERATION |
850 * |**************************************************************************
851 * | source-->| shared | private | slave | unbindable |
852 * | dest | | | | |
853 * | | | | | | |
854 * | v | | | | |
855 * |**************************************************************************
856 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
857 * | | | | | |
858 * |non-shared| shared (+*) | private | slave (*) | unbindable |
859 * ***************************************************************************
860 *
861 * (+) the mount is moved to the destination. And is then propagated to
862 * all the mounts in the propagation tree of the destination mount.
863 * (+*) the mount is moved to the destination.
864 * (+++) the mount is moved to the destination and is then propagated to
865 * all the mounts belonging to the destination mount's propagation tree.
866 * the mount is marked as 'shared and slave'.
867 * (*) the mount continues to be a slave at the new location.
868 *
869 * if the source mount is a tree, the operations explained above is
870 * applied to each mount in the tree.
871 * Must be called without spinlocks held, since this function can sleep
872 * in allocations.
873 */
874 static int attach_recursive_mnt(struct vfsmount *source_mnt,
875 struct nameidata *nd, struct nameidata *parent_nd)
876 {
877 LIST_HEAD(tree_list);
878 struct vfsmount *dest_mnt = nd->mnt;
879 struct dentry *dest_dentry = nd->dentry;
880 struct vfsmount *child, *p;
881
882 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
883 return -EINVAL;
884
885 if (IS_MNT_SHARED(dest_mnt)) {
886 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
887 set_mnt_shared(p);
888 }
889
890 spin_lock(&vfsmount_lock);
891 if (parent_nd) {
892 detach_mnt(source_mnt, parent_nd);
893 attach_mnt(source_mnt, nd);
894 touch_mnt_namespace(current->nsproxy->mnt_ns);
895 } else {
896 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
897 commit_tree(source_mnt);
898 }
899
900 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
901 list_del_init(&child->mnt_hash);
902 commit_tree(child);
903 }
904 spin_unlock(&vfsmount_lock);
905 return 0;
906 }
907
908 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
909 {
910 int err;
911 if (mnt->mnt_sb->s_flags & MS_NOUSER)
912 return -EINVAL;
913
914 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
915 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
916 return -ENOTDIR;
917
918 err = -ENOENT;
919 mutex_lock(&nd->dentry->d_inode->i_mutex);
920 if (IS_DEADDIR(nd->dentry->d_inode))
921 goto out_unlock;
922
923 err = security_sb_check_sb(mnt, nd);
924 if (err)
925 goto out_unlock;
926
927 err = -ENOENT;
928 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
929 err = attach_recursive_mnt(mnt, nd, NULL);
930 out_unlock:
931 mutex_unlock(&nd->dentry->d_inode->i_mutex);
932 if (!err)
933 security_sb_post_addmount(mnt, nd);
934 return err;
935 }
936
937 /*
938 * recursively change the type of the mountpoint.
939 * noinline this do_mount helper to save do_mount stack space.
940 */
941 static noinline int do_change_type(struct nameidata *nd, int flag)
942 {
943 struct vfsmount *m, *mnt = nd->mnt;
944 int recurse = flag & MS_REC;
945 int type = flag & ~MS_REC;
946
947 if (!capable(CAP_SYS_ADMIN))
948 return -EPERM;
949
950 if (nd->dentry != nd->mnt->mnt_root)
951 return -EINVAL;
952
953 down_write(&namespace_sem);
954 spin_lock(&vfsmount_lock);
955 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
956 change_mnt_propagation(m, type);
957 spin_unlock(&vfsmount_lock);
958 up_write(&namespace_sem);
959 return 0;
960 }
961
962 /*
963 * do loopback mount.
964 * noinline this do_mount helper to save do_mount stack space.
965 */
966 static noinline int do_loopback(struct nameidata *nd, char *old_name,
967 int recurse)
968 {
969 struct nameidata old_nd;
970 struct vfsmount *mnt = NULL;
971 int err = mount_is_safe(nd);
972 if (err)
973 return err;
974 if (!old_name || !*old_name)
975 return -EINVAL;
976 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
977 if (err)
978 return err;
979
980 down_write(&namespace_sem);
981 err = -EINVAL;
982 if (IS_MNT_UNBINDABLE(old_nd.mnt))
983 goto out;
984
985 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
986 goto out;
987
988 err = -ENOMEM;
989 if (recurse)
990 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
991 else
992 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
993
994 if (!mnt)
995 goto out;
996
997 err = graft_tree(mnt, nd);
998 if (err) {
999 LIST_HEAD(umount_list);
1000 spin_lock(&vfsmount_lock);
1001 umount_tree(mnt, 0, &umount_list);
1002 spin_unlock(&vfsmount_lock);
1003 release_mounts(&umount_list);
1004 }
1005
1006 out:
1007 up_write(&namespace_sem);
1008 path_release(&old_nd);
1009 return err;
1010 }
1011
1012 /*
1013 * change filesystem flags. dir should be a physical root of filesystem.
1014 * If you've mounted a non-root directory somewhere and want to do remount
1015 * on it - tough luck.
1016 * noinline this do_mount helper to save do_mount stack space.
1017 */
1018 static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1019 void *data)
1020 {
1021 int err;
1022 struct super_block *sb = nd->mnt->mnt_sb;
1023
1024 if (!capable(CAP_SYS_ADMIN))
1025 return -EPERM;
1026
1027 if (!check_mnt(nd->mnt))
1028 return -EINVAL;
1029
1030 if (nd->dentry != nd->mnt->mnt_root)
1031 return -EINVAL;
1032
1033 down_write(&sb->s_umount);
1034 err = do_remount_sb(sb, flags, data, 0);
1035 if (!err)
1036 nd->mnt->mnt_flags = mnt_flags;
1037 up_write(&sb->s_umount);
1038 if (!err)
1039 security_sb_post_remount(nd->mnt, flags, data);
1040 return err;
1041 }
1042
1043 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1044 {
1045 struct vfsmount *p;
1046 for (p = mnt; p; p = next_mnt(p, mnt)) {
1047 if (IS_MNT_UNBINDABLE(p))
1048 return 1;
1049 }
1050 return 0;
1051 }
1052
1053 /*
1054 * noinline this do_mount helper to save do_mount stack space.
1055 */
1056 static noinline int do_move_mount(struct nameidata *nd, char *old_name)
1057 {
1058 struct nameidata old_nd, parent_nd;
1059 struct vfsmount *p;
1060 int err = 0;
1061 if (!capable(CAP_SYS_ADMIN))
1062 return -EPERM;
1063 if (!old_name || !*old_name)
1064 return -EINVAL;
1065 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1066 if (err)
1067 return err;
1068
1069 down_write(&namespace_sem);
1070 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1071 ;
1072 err = -EINVAL;
1073 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1074 goto out;
1075
1076 err = -ENOENT;
1077 mutex_lock(&nd->dentry->d_inode->i_mutex);
1078 if (IS_DEADDIR(nd->dentry->d_inode))
1079 goto out1;
1080
1081 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1082 goto out1;
1083
1084 err = -EINVAL;
1085 if (old_nd.dentry != old_nd.mnt->mnt_root)
1086 goto out1;
1087
1088 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1089 goto out1;
1090
1091 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1092 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1093 goto out1;
1094 /*
1095 * Don't move a mount residing in a shared parent.
1096 */
1097 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1098 goto out1;
1099 /*
1100 * Don't move a mount tree containing unbindable mounts to a destination
1101 * mount which is shared.
1102 */
1103 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1104 goto out1;
1105 err = -ELOOP;
1106 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1107 if (p == old_nd.mnt)
1108 goto out1;
1109
1110 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1111 goto out1;
1112
1113 spin_lock(&vfsmount_lock);
1114 /* if the mount is moved, it should no longer be expire
1115 * automatically */
1116 list_del_init(&old_nd.mnt->mnt_expire);
1117 spin_unlock(&vfsmount_lock);
1118 out1:
1119 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1120 out:
1121 up_write(&namespace_sem);
1122 if (!err)
1123 path_release(&parent_nd);
1124 path_release(&old_nd);
1125 return err;
1126 }
1127
1128 /*
1129 * create a new mount for userspace and request it to be added into the
1130 * namespace's tree
1131 * noinline this do_mount helper to save do_mount stack space.
1132 */
1133 static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
1134 int mnt_flags, char *name, void *data)
1135 {
1136 struct vfsmount *mnt;
1137
1138 if (!type || !memchr(type, 0, PAGE_SIZE))
1139 return -EINVAL;
1140
1141 /* we need capabilities... */
1142 if (!capable(CAP_SYS_ADMIN))
1143 return -EPERM;
1144
1145 mnt = do_kern_mount(type, flags, name, data);
1146 if (IS_ERR(mnt))
1147 return PTR_ERR(mnt);
1148
1149 return do_add_mount(mnt, nd, mnt_flags, NULL);
1150 }
1151
1152 /*
1153 * add a mount into a namespace's mount tree
1154 * - provide the option of adding the new mount to an expiration list
1155 */
1156 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1157 int mnt_flags, struct list_head *fslist)
1158 {
1159 int err;
1160
1161 down_write(&namespace_sem);
1162 /* Something was mounted here while we slept */
1163 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1164 ;
1165 err = -EINVAL;
1166 if (!check_mnt(nd->mnt))
1167 goto unlock;
1168
1169 /* Refuse the same filesystem on the same mount point */
1170 err = -EBUSY;
1171 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1172 nd->mnt->mnt_root == nd->dentry)
1173 goto unlock;
1174
1175 err = -EINVAL;
1176 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1177 goto unlock;
1178
1179 newmnt->mnt_flags = mnt_flags;
1180 if ((err = graft_tree(newmnt, nd)))
1181 goto unlock;
1182
1183 if (fslist) {
1184 /* add to the specified expiration list */
1185 spin_lock(&vfsmount_lock);
1186 list_add_tail(&newmnt->mnt_expire, fslist);
1187 spin_unlock(&vfsmount_lock);
1188 }
1189 up_write(&namespace_sem);
1190 return 0;
1191
1192 unlock:
1193 up_write(&namespace_sem);
1194 mntput(newmnt);
1195 return err;
1196 }
1197
1198 EXPORT_SYMBOL_GPL(do_add_mount);
1199
1200 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1201 struct list_head *umounts)
1202 {
1203 spin_lock(&vfsmount_lock);
1204
1205 /*
1206 * Check if mount is still attached, if not, let whoever holds it deal
1207 * with the sucker
1208 */
1209 if (mnt->mnt_parent == mnt) {
1210 spin_unlock(&vfsmount_lock);
1211 return;
1212 }
1213
1214 /*
1215 * Check that it is still dead: the count should now be 2 - as
1216 * contributed by the vfsmount parent and the mntget above
1217 */
1218 if (!propagate_mount_busy(mnt, 2)) {
1219 /* delete from the namespace */
1220 touch_mnt_namespace(mnt->mnt_ns);
1221 list_del_init(&mnt->mnt_list);
1222 mnt->mnt_ns = NULL;
1223 umount_tree(mnt, 1, umounts);
1224 spin_unlock(&vfsmount_lock);
1225 } else {
1226 /*
1227 * Someone brought it back to life whilst we didn't have any
1228 * locks held so return it to the expiration list
1229 */
1230 list_add_tail(&mnt->mnt_expire, mounts);
1231 spin_unlock(&vfsmount_lock);
1232 }
1233 }
1234
1235 /*
1236 * go through the vfsmounts we've just consigned to the graveyard to
1237 * - check that they're still dead
1238 * - delete the vfsmount from the appropriate namespace under lock
1239 * - dispose of the corpse
1240 */
1241 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1242 {
1243 struct mnt_namespace *ns;
1244 struct vfsmount *mnt;
1245
1246 while (!list_empty(graveyard)) {
1247 LIST_HEAD(umounts);
1248 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
1249 list_del_init(&mnt->mnt_expire);
1250
1251 /* don't do anything if the namespace is dead - all the
1252 * vfsmounts from it are going away anyway */
1253 ns = mnt->mnt_ns;
1254 if (!ns || !ns->root)
1255 continue;
1256 get_mnt_ns(ns);
1257
1258 spin_unlock(&vfsmount_lock);
1259 down_write(&namespace_sem);
1260 expire_mount(mnt, mounts, &umounts);
1261 up_write(&namespace_sem);
1262 release_mounts(&umounts);
1263 mntput(mnt);
1264 put_mnt_ns(ns);
1265 spin_lock(&vfsmount_lock);
1266 }
1267 }
1268
1269 /*
1270 * process a list of expirable mountpoints with the intent of discarding any
1271 * mountpoints that aren't in use and haven't been touched since last we came
1272 * here
1273 */
1274 void mark_mounts_for_expiry(struct list_head *mounts)
1275 {
1276 struct vfsmount *mnt, *next;
1277 LIST_HEAD(graveyard);
1278
1279 if (list_empty(mounts))
1280 return;
1281
1282 spin_lock(&vfsmount_lock);
1283
1284 /* extract from the expiration list every vfsmount that matches the
1285 * following criteria:
1286 * - only referenced by its parent vfsmount
1287 * - still marked for expiry (marked on the last call here; marks are
1288 * cleared by mntput())
1289 */
1290 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1291 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1292 atomic_read(&mnt->mnt_count) != 1)
1293 continue;
1294
1295 mntget(mnt);
1296 list_move(&mnt->mnt_expire, &graveyard);
1297 }
1298
1299 expire_mount_list(&graveyard, mounts);
1300
1301 spin_unlock(&vfsmount_lock);
1302 }
1303
1304 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1305
1306 /*
1307 * Ripoff of 'select_parent()'
1308 *
1309 * search the list of submounts for a given mountpoint, and move any
1310 * shrinkable submounts to the 'graveyard' list.
1311 */
1312 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1313 {
1314 struct vfsmount *this_parent = parent;
1315 struct list_head *next;
1316 int found = 0;
1317
1318 repeat:
1319 next = this_parent->mnt_mounts.next;
1320 resume:
1321 while (next != &this_parent->mnt_mounts) {
1322 struct list_head *tmp = next;
1323 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1324
1325 next = tmp->next;
1326 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1327 continue;
1328 /*
1329 * Descend a level if the d_mounts list is non-empty.
1330 */
1331 if (!list_empty(&mnt->mnt_mounts)) {
1332 this_parent = mnt;
1333 goto repeat;
1334 }
1335
1336 if (!propagate_mount_busy(mnt, 1)) {
1337 mntget(mnt);
1338 list_move_tail(&mnt->mnt_expire, graveyard);
1339 found++;
1340 }
1341 }
1342 /*
1343 * All done at this level ... ascend and resume the search
1344 */
1345 if (this_parent != parent) {
1346 next = this_parent->mnt_child.next;
1347 this_parent = this_parent->mnt_parent;
1348 goto resume;
1349 }
1350 return found;
1351 }
1352
1353 /*
1354 * process a list of expirable mountpoints with the intent of discarding any
1355 * submounts of a specific parent mountpoint
1356 */
1357 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1358 {
1359 LIST_HEAD(graveyard);
1360 int found;
1361
1362 spin_lock(&vfsmount_lock);
1363
1364 /* extract submounts of 'mountpoint' from the expiration list */
1365 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1366 expire_mount_list(&graveyard, mounts);
1367
1368 spin_unlock(&vfsmount_lock);
1369 }
1370
1371 EXPORT_SYMBOL_GPL(shrink_submounts);
1372
1373 /*
1374 * Some copy_from_user() implementations do not return the exact number of
1375 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1376 * Note that this function differs from copy_from_user() in that it will oops
1377 * on bad values of `to', rather than returning a short copy.
1378 */
1379 static long exact_copy_from_user(void *to, const void __user * from,
1380 unsigned long n)
1381 {
1382 char *t = to;
1383 const char __user *f = from;
1384 char c;
1385
1386 if (!access_ok(VERIFY_READ, from, n))
1387 return n;
1388
1389 while (n) {
1390 if (__get_user(c, f)) {
1391 memset(t, 0, n);
1392 break;
1393 }
1394 *t++ = c;
1395 f++;
1396 n--;
1397 }
1398 return n;
1399 }
1400
1401 int copy_mount_options(const void __user * data, unsigned long *where)
1402 {
1403 int i;
1404 unsigned long page;
1405 unsigned long size;
1406
1407 *where = 0;
1408 if (!data)
1409 return 0;
1410
1411 if (!(page = __get_free_page(GFP_KERNEL)))
1412 return -ENOMEM;
1413
1414 /* We only care that *some* data at the address the user
1415 * gave us is valid. Just in case, we'll zero
1416 * the remainder of the page.
1417 */
1418 /* copy_from_user cannot cross TASK_SIZE ! */
1419 size = TASK_SIZE - (unsigned long)data;
1420 if (size > PAGE_SIZE)
1421 size = PAGE_SIZE;
1422
1423 i = size - exact_copy_from_user((void *)page, data, size);
1424 if (!i) {
1425 free_page(page);
1426 return -EFAULT;
1427 }
1428 if (i != PAGE_SIZE)
1429 memset((char *)page + i, 0, PAGE_SIZE - i);
1430 *where = page;
1431 return 0;
1432 }
1433
1434 /*
1435 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1436 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1437 *
1438 * data is a (void *) that can point to any structure up to
1439 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1440 * information (or be NULL).
1441 *
1442 * Pre-0.97 versions of mount() didn't have a flags word.
1443 * When the flags word was introduced its top half was required
1444 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1445 * Therefore, if this magic number is present, it carries no information
1446 * and must be discarded.
1447 */
1448 long do_mount(char *dev_name, char *dir_name, char *type_page,
1449 unsigned long flags, void *data_page)
1450 {
1451 struct nameidata nd;
1452 int retval = 0;
1453 int mnt_flags = 0;
1454
1455 /* Discard magic */
1456 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1457 flags &= ~MS_MGC_MSK;
1458
1459 /* Basic sanity checks */
1460
1461 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1462 return -EINVAL;
1463 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1464 return -EINVAL;
1465
1466 if (data_page)
1467 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1468
1469 /* Separate the per-mountpoint flags */
1470 if (flags & MS_NOSUID)
1471 mnt_flags |= MNT_NOSUID;
1472 if (flags & MS_NODEV)
1473 mnt_flags |= MNT_NODEV;
1474 if (flags & MS_NOEXEC)
1475 mnt_flags |= MNT_NOEXEC;
1476 if (flags & MS_NOATIME)
1477 mnt_flags |= MNT_NOATIME;
1478 if (flags & MS_NODIRATIME)
1479 mnt_flags |= MNT_NODIRATIME;
1480 if (flags & MS_RELATIME)
1481 mnt_flags |= MNT_RELATIME;
1482
1483 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1484 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
1485
1486 /* ... and get the mountpoint */
1487 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1488 if (retval)
1489 return retval;
1490
1491 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1492 if (retval)
1493 goto dput_out;
1494
1495 if (flags & MS_REMOUNT)
1496 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1497 data_page);
1498 else if (flags & MS_BIND)
1499 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1500 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1501 retval = do_change_type(&nd, flags);
1502 else if (flags & MS_MOVE)
1503 retval = do_move_mount(&nd, dev_name);
1504 else
1505 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1506 dev_name, data_page);
1507 dput_out:
1508 path_release(&nd);
1509 return retval;
1510 }
1511
1512 /*
1513 * Allocate a new namespace structure and populate it with contents
1514 * copied from the namespace of the passed in task structure.
1515 */
1516 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1517 struct fs_struct *fs)
1518 {
1519 struct mnt_namespace *new_ns;
1520 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1521 struct vfsmount *p, *q;
1522
1523 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1524 if (!new_ns)
1525 return ERR_PTR(-ENOMEM);
1526
1527 atomic_set(&new_ns->count, 1);
1528 INIT_LIST_HEAD(&new_ns->list);
1529 init_waitqueue_head(&new_ns->poll);
1530 new_ns->event = 0;
1531
1532 down_write(&namespace_sem);
1533 /* First pass: copy the tree topology */
1534 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1535 CL_COPY_ALL | CL_EXPIRE);
1536 if (!new_ns->root) {
1537 up_write(&namespace_sem);
1538 kfree(new_ns);
1539 return ERR_PTR(-ENOMEM);;
1540 }
1541 spin_lock(&vfsmount_lock);
1542 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1543 spin_unlock(&vfsmount_lock);
1544
1545 /*
1546 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1547 * as belonging to new namespace. We have already acquired a private
1548 * fs_struct, so tsk->fs->lock is not needed.
1549 */
1550 p = mnt_ns->root;
1551 q = new_ns->root;
1552 while (p) {
1553 q->mnt_ns = new_ns;
1554 if (fs) {
1555 if (p == fs->rootmnt) {
1556 rootmnt = p;
1557 fs->rootmnt = mntget(q);
1558 }
1559 if (p == fs->pwdmnt) {
1560 pwdmnt = p;
1561 fs->pwdmnt = mntget(q);
1562 }
1563 if (p == fs->altrootmnt) {
1564 altrootmnt = p;
1565 fs->altrootmnt = mntget(q);
1566 }
1567 }
1568 p = next_mnt(p, mnt_ns->root);
1569 q = next_mnt(q, new_ns->root);
1570 }
1571 up_write(&namespace_sem);
1572
1573 if (rootmnt)
1574 mntput(rootmnt);
1575 if (pwdmnt)
1576 mntput(pwdmnt);
1577 if (altrootmnt)
1578 mntput(altrootmnt);
1579
1580 return new_ns;
1581 }
1582
1583 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
1584 struct fs_struct *new_fs)
1585 {
1586 struct mnt_namespace *new_ns;
1587
1588 BUG_ON(!ns);
1589 get_mnt_ns(ns);
1590
1591 if (!(flags & CLONE_NEWNS))
1592 return ns;
1593
1594 new_ns = dup_mnt_ns(ns, new_fs);
1595
1596 put_mnt_ns(ns);
1597 return new_ns;
1598 }
1599
1600 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1601 char __user * type, unsigned long flags,
1602 void __user * data)
1603 {
1604 int retval;
1605 unsigned long data_page;
1606 unsigned long type_page;
1607 unsigned long dev_page;
1608 char *dir_page;
1609
1610 retval = copy_mount_options(type, &type_page);
1611 if (retval < 0)
1612 return retval;
1613
1614 dir_page = getname(dir_name);
1615 retval = PTR_ERR(dir_page);
1616 if (IS_ERR(dir_page))
1617 goto out1;
1618
1619 retval = copy_mount_options(dev_name, &dev_page);
1620 if (retval < 0)
1621 goto out2;
1622
1623 retval = copy_mount_options(data, &data_page);
1624 if (retval < 0)
1625 goto out3;
1626
1627 lock_kernel();
1628 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1629 flags, (void *)data_page);
1630 unlock_kernel();
1631 free_page(data_page);
1632
1633 out3:
1634 free_page(dev_page);
1635 out2:
1636 putname(dir_page);
1637 out1:
1638 free_page(type_page);
1639 return retval;
1640 }
1641
1642 /*
1643 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1644 * It can block. Requires the big lock held.
1645 */
1646 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1647 struct dentry *dentry)
1648 {
1649 struct dentry *old_root;
1650 struct vfsmount *old_rootmnt;
1651 write_lock(&fs->lock);
1652 old_root = fs->root;
1653 old_rootmnt = fs->rootmnt;
1654 fs->rootmnt = mntget(mnt);
1655 fs->root = dget(dentry);
1656 write_unlock(&fs->lock);
1657 if (old_root) {
1658 dput(old_root);
1659 mntput(old_rootmnt);
1660 }
1661 }
1662
1663 /*
1664 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1665 * It can block. Requires the big lock held.
1666 */
1667 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1668 struct dentry *dentry)
1669 {
1670 struct dentry *old_pwd;
1671 struct vfsmount *old_pwdmnt;
1672
1673 write_lock(&fs->lock);
1674 old_pwd = fs->pwd;
1675 old_pwdmnt = fs->pwdmnt;
1676 fs->pwdmnt = mntget(mnt);
1677 fs->pwd = dget(dentry);
1678 write_unlock(&fs->lock);
1679
1680 if (old_pwd) {
1681 dput(old_pwd);
1682 mntput(old_pwdmnt);
1683 }
1684 }
1685
1686 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1687 {
1688 struct task_struct *g, *p;
1689 struct fs_struct *fs;
1690
1691 read_lock(&tasklist_lock);
1692 do_each_thread(g, p) {
1693 task_lock(p);
1694 fs = p->fs;
1695 if (fs) {
1696 atomic_inc(&fs->count);
1697 task_unlock(p);
1698 if (fs->root == old_nd->dentry
1699 && fs->rootmnt == old_nd->mnt)
1700 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1701 if (fs->pwd == old_nd->dentry
1702 && fs->pwdmnt == old_nd->mnt)
1703 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1704 put_fs_struct(fs);
1705 } else
1706 task_unlock(p);
1707 } while_each_thread(g, p);
1708 read_unlock(&tasklist_lock);
1709 }
1710
1711 /*
1712 * pivot_root Semantics:
1713 * Moves the root file system of the current process to the directory put_old,
1714 * makes new_root as the new root file system of the current process, and sets
1715 * root/cwd of all processes which had them on the current root to new_root.
1716 *
1717 * Restrictions:
1718 * The new_root and put_old must be directories, and must not be on the
1719 * same file system as the current process root. The put_old must be
1720 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1721 * pointed to by put_old must yield the same directory as new_root. No other
1722 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1723 *
1724 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1725 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1726 * in this situation.
1727 *
1728 * Notes:
1729 * - we don't move root/cwd if they are not at the root (reason: if something
1730 * cared enough to change them, it's probably wrong to force them elsewhere)
1731 * - it's okay to pick a root that isn't the root of a file system, e.g.
1732 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1733 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1734 * first.
1735 */
1736 asmlinkage long sys_pivot_root(const char __user * new_root,
1737 const char __user * put_old)
1738 {
1739 struct vfsmount *tmp;
1740 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1741 int error;
1742
1743 if (!capable(CAP_SYS_ADMIN))
1744 return -EPERM;
1745
1746 lock_kernel();
1747
1748 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1749 &new_nd);
1750 if (error)
1751 goto out0;
1752 error = -EINVAL;
1753 if (!check_mnt(new_nd.mnt))
1754 goto out1;
1755
1756 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1757 if (error)
1758 goto out1;
1759
1760 error = security_sb_pivotroot(&old_nd, &new_nd);
1761 if (error) {
1762 path_release(&old_nd);
1763 goto out1;
1764 }
1765
1766 read_lock(&current->fs->lock);
1767 user_nd.mnt = mntget(current->fs->rootmnt);
1768 user_nd.dentry = dget(current->fs->root);
1769 read_unlock(&current->fs->lock);
1770 down_write(&namespace_sem);
1771 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1772 error = -EINVAL;
1773 if (IS_MNT_SHARED(old_nd.mnt) ||
1774 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1775 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1776 goto out2;
1777 if (!check_mnt(user_nd.mnt))
1778 goto out2;
1779 error = -ENOENT;
1780 if (IS_DEADDIR(new_nd.dentry->d_inode))
1781 goto out2;
1782 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1783 goto out2;
1784 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1785 goto out2;
1786 error = -EBUSY;
1787 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1788 goto out2; /* loop, on the same file system */
1789 error = -EINVAL;
1790 if (user_nd.mnt->mnt_root != user_nd.dentry)
1791 goto out2; /* not a mountpoint */
1792 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1793 goto out2; /* not attached */
1794 if (new_nd.mnt->mnt_root != new_nd.dentry)
1795 goto out2; /* not a mountpoint */
1796 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1797 goto out2; /* not attached */
1798 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1799 spin_lock(&vfsmount_lock);
1800 if (tmp != new_nd.mnt) {
1801 for (;;) {
1802 if (tmp->mnt_parent == tmp)
1803 goto out3; /* already mounted on put_old */
1804 if (tmp->mnt_parent == new_nd.mnt)
1805 break;
1806 tmp = tmp->mnt_parent;
1807 }
1808 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1809 goto out3;
1810 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1811 goto out3;
1812 detach_mnt(new_nd.mnt, &parent_nd);
1813 detach_mnt(user_nd.mnt, &root_parent);
1814 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1815 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1816 touch_mnt_namespace(current->nsproxy->mnt_ns);
1817 spin_unlock(&vfsmount_lock);
1818 chroot_fs_refs(&user_nd, &new_nd);
1819 security_sb_post_pivotroot(&user_nd, &new_nd);
1820 error = 0;
1821 path_release(&root_parent);
1822 path_release(&parent_nd);
1823 out2:
1824 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1825 up_write(&namespace_sem);
1826 path_release(&user_nd);
1827 path_release(&old_nd);
1828 out1:
1829 path_release(&new_nd);
1830 out0:
1831 unlock_kernel();
1832 return error;
1833 out3:
1834 spin_unlock(&vfsmount_lock);
1835 goto out2;
1836 }
1837
1838 static void __init init_mount_tree(void)
1839 {
1840 struct vfsmount *mnt;
1841 struct mnt_namespace *ns;
1842
1843 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1844 if (IS_ERR(mnt))
1845 panic("Can't create rootfs");
1846 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1847 if (!ns)
1848 panic("Can't allocate initial namespace");
1849 atomic_set(&ns->count, 1);
1850 INIT_LIST_HEAD(&ns->list);
1851 init_waitqueue_head(&ns->poll);
1852 ns->event = 0;
1853 list_add(&mnt->mnt_list, &ns->list);
1854 ns->root = mnt;
1855 mnt->mnt_ns = ns;
1856
1857 init_task.nsproxy->mnt_ns = ns;
1858 get_mnt_ns(ns);
1859
1860 set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1861 set_fs_root(current->fs, ns->root, ns->root->mnt_root);
1862 }
1863
1864 void __init mnt_init(void)
1865 {
1866 unsigned u;
1867 int err;
1868
1869 init_rwsem(&namespace_sem);
1870
1871 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1872 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1873
1874 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1875
1876 if (!mount_hashtable)
1877 panic("Failed to allocate mount hash table\n");
1878
1879 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
1880
1881 for (u = 0; u < HASH_SIZE; u++)
1882 INIT_LIST_HEAD(&mount_hashtable[u]);
1883
1884 err = sysfs_init();
1885 if (err)
1886 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1887 __FUNCTION__, err);
1888 fs_kobj = kobject_create_and_add("fs", NULL);
1889 if (!fs_kobj)
1890 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
1891 init_rootfs();
1892 init_mount_tree();
1893 }
1894
1895 void __put_mnt_ns(struct mnt_namespace *ns)
1896 {
1897 struct vfsmount *root = ns->root;
1898 LIST_HEAD(umount_list);
1899 ns->root = NULL;
1900 spin_unlock(&vfsmount_lock);
1901 down_write(&namespace_sem);
1902 spin_lock(&vfsmount_lock);
1903 umount_tree(root, 0, &umount_list);
1904 spin_unlock(&vfsmount_lock);
1905 up_write(&namespace_sem);
1906 release_mounts(&umount_list);
1907 kfree(ns);
1908 }