]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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/syscalls.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/sched.h> | |
15 | #include <linux/smp_lock.h> | |
16 | #include <linux/init.h> | |
17 | #include <linux/quotaops.h> | |
18 | #include <linux/acct.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/seq_file.h> | |
21 | #include <linux/namespace.h> | |
22 | #include <linux/namei.h> | |
23 | #include <linux/security.h> | |
24 | #include <linux/mount.h> | |
25 | #include <asm/uaccess.h> | |
26 | #include <asm/unistd.h> | |
07b20889 | 27 | #include "pnode.h" |
1da177e4 LT |
28 | |
29 | extern int __init init_rootfs(void); | |
30 | ||
36341f64 RP |
31 | #define CL_EXPIRE 0x01 |
32 | ||
1da177e4 LT |
33 | #ifdef CONFIG_SYSFS |
34 | extern int __init sysfs_init(void); | |
35 | #else | |
36 | static inline int sysfs_init(void) | |
37 | { | |
38 | return 0; | |
39 | } | |
40 | #endif | |
41 | ||
42 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ | |
5addc5dd AV |
43 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
44 | ||
45 | static int event; | |
1da177e4 LT |
46 | |
47 | static struct list_head *mount_hashtable; | |
6c231b7b | 48 | static int hash_mask __read_mostly, hash_bits __read_mostly; |
b58fed8b | 49 | static kmem_cache_t *mnt_cache; |
390c6843 | 50 | static struct rw_semaphore namespace_sem; |
1da177e4 LT |
51 | |
52 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) | |
53 | { | |
b58fed8b RP |
54 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
55 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
1da177e4 LT |
56 | tmp = tmp + (tmp >> hash_bits); |
57 | return tmp & hash_mask; | |
58 | } | |
59 | ||
60 | struct vfsmount *alloc_vfsmnt(const char *name) | |
61 | { | |
b58fed8b | 62 | struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); |
1da177e4 LT |
63 | if (mnt) { |
64 | memset(mnt, 0, sizeof(struct vfsmount)); | |
b58fed8b | 65 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
66 | INIT_LIST_HEAD(&mnt->mnt_hash); |
67 | INIT_LIST_HEAD(&mnt->mnt_child); | |
68 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
69 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 70 | INIT_LIST_HEAD(&mnt->mnt_expire); |
1da177e4 | 71 | if (name) { |
b58fed8b | 72 | int size = strlen(name) + 1; |
1da177e4 LT |
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 | void free_vfsmnt(struct vfsmount *mnt) | |
84 | { | |
85 | kfree(mnt->mnt_devname); | |
86 | kmem_cache_free(mnt_cache, mnt); | |
87 | } | |
88 | ||
89 | /* | |
90 | * Now, lookup_mnt increments the ref count before returning | |
91 | * the vfsmount struct. | |
92 | */ | |
93 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
94 | { | |
b58fed8b RP |
95 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
96 | struct list_head *tmp = head; | |
1da177e4 LT |
97 | struct vfsmount *p, *found = NULL; |
98 | ||
99 | spin_lock(&vfsmount_lock); | |
100 | for (;;) { | |
101 | tmp = tmp->next; | |
102 | p = NULL; | |
103 | if (tmp == head) | |
104 | break; | |
105 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
106 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
107 | found = mntget(p); | |
108 | break; | |
109 | } | |
110 | } | |
111 | spin_unlock(&vfsmount_lock); | |
112 | return found; | |
113 | } | |
114 | ||
115 | static inline int check_mnt(struct vfsmount *mnt) | |
116 | { | |
117 | return mnt->mnt_namespace == current->namespace; | |
118 | } | |
119 | ||
5addc5dd AV |
120 | static void touch_namespace(struct namespace *ns) |
121 | { | |
122 | if (ns) { | |
123 | ns->event = ++event; | |
124 | wake_up_interruptible(&ns->poll); | |
125 | } | |
126 | } | |
127 | ||
128 | static void __touch_namespace(struct namespace *ns) | |
129 | { | |
130 | if (ns && ns->event != event) { | |
131 | ns->event = event; | |
132 | wake_up_interruptible(&ns->poll); | |
133 | } | |
134 | } | |
135 | ||
1da177e4 LT |
136 | static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) |
137 | { | |
138 | old_nd->dentry = mnt->mnt_mountpoint; | |
139 | old_nd->mnt = mnt->mnt_parent; | |
140 | mnt->mnt_parent = mnt; | |
141 | mnt->mnt_mountpoint = mnt->mnt_root; | |
142 | list_del_init(&mnt->mnt_child); | |
143 | list_del_init(&mnt->mnt_hash); | |
144 | old_nd->dentry->d_mounted--; | |
145 | } | |
146 | ||
147 | static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) | |
148 | { | |
149 | mnt->mnt_parent = mntget(nd->mnt); | |
150 | mnt->mnt_mountpoint = dget(nd->dentry); | |
b58fed8b | 151 | list_add(&mnt->mnt_hash, mount_hashtable + hash(nd->mnt, nd->dentry)); |
1da177e4 LT |
152 | list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); |
153 | nd->dentry->d_mounted++; | |
154 | } | |
155 | ||
156 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
157 | { | |
158 | struct list_head *next = p->mnt_mounts.next; | |
159 | if (next == &p->mnt_mounts) { | |
160 | while (1) { | |
161 | if (p == root) | |
162 | return NULL; | |
163 | next = p->mnt_child.next; | |
164 | if (next != &p->mnt_parent->mnt_mounts) | |
165 | break; | |
166 | p = p->mnt_parent; | |
167 | } | |
168 | } | |
169 | return list_entry(next, struct vfsmount, mnt_child); | |
170 | } | |
171 | ||
36341f64 RP |
172 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
173 | int flag) | |
1da177e4 LT |
174 | { |
175 | struct super_block *sb = old->mnt_sb; | |
176 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
177 | ||
178 | if (mnt) { | |
179 | mnt->mnt_flags = old->mnt_flags; | |
180 | atomic_inc(&sb->s_active); | |
181 | mnt->mnt_sb = sb; | |
182 | mnt->mnt_root = dget(root); | |
183 | mnt->mnt_mountpoint = mnt->mnt_root; | |
184 | mnt->mnt_parent = mnt; | |
68b47139 | 185 | mnt->mnt_namespace = current->namespace; |
1da177e4 LT |
186 | |
187 | /* stick the duplicate mount on the same expiry list | |
188 | * as the original if that was on one */ | |
36341f64 RP |
189 | if (flag & CL_EXPIRE) { |
190 | spin_lock(&vfsmount_lock); | |
191 | if (!list_empty(&old->mnt_expire)) | |
192 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
193 | spin_unlock(&vfsmount_lock); | |
194 | } | |
1da177e4 LT |
195 | } |
196 | return mnt; | |
197 | } | |
198 | ||
7b7b1ace | 199 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
200 | { |
201 | struct super_block *sb = mnt->mnt_sb; | |
202 | dput(mnt->mnt_root); | |
203 | free_vfsmnt(mnt); | |
204 | deactivate_super(sb); | |
205 | } | |
206 | ||
7b7b1ace AV |
207 | void mntput_no_expire(struct vfsmount *mnt) |
208 | { | |
209 | repeat: | |
210 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
211 | if (likely(!mnt->mnt_pinned)) { | |
212 | spin_unlock(&vfsmount_lock); | |
213 | __mntput(mnt); | |
214 | return; | |
215 | } | |
216 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
217 | mnt->mnt_pinned = 0; | |
218 | spin_unlock(&vfsmount_lock); | |
219 | acct_auto_close_mnt(mnt); | |
220 | security_sb_umount_close(mnt); | |
221 | goto repeat; | |
222 | } | |
223 | } | |
224 | ||
225 | EXPORT_SYMBOL(mntput_no_expire); | |
226 | ||
227 | void mnt_pin(struct vfsmount *mnt) | |
228 | { | |
229 | spin_lock(&vfsmount_lock); | |
230 | mnt->mnt_pinned++; | |
231 | spin_unlock(&vfsmount_lock); | |
232 | } | |
233 | ||
234 | EXPORT_SYMBOL(mnt_pin); | |
235 | ||
236 | void mnt_unpin(struct vfsmount *mnt) | |
237 | { | |
238 | spin_lock(&vfsmount_lock); | |
239 | if (mnt->mnt_pinned) { | |
240 | atomic_inc(&mnt->mnt_count); | |
241 | mnt->mnt_pinned--; | |
242 | } | |
243 | spin_unlock(&vfsmount_lock); | |
244 | } | |
245 | ||
246 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 LT |
247 | |
248 | /* iterator */ | |
249 | static void *m_start(struct seq_file *m, loff_t *pos) | |
250 | { | |
251 | struct namespace *n = m->private; | |
252 | struct list_head *p; | |
253 | loff_t l = *pos; | |
254 | ||
390c6843 | 255 | down_read(&namespace_sem); |
1da177e4 LT |
256 | list_for_each(p, &n->list) |
257 | if (!l--) | |
258 | return list_entry(p, struct vfsmount, mnt_list); | |
259 | return NULL; | |
260 | } | |
261 | ||
262 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
263 | { | |
264 | struct namespace *n = m->private; | |
265 | struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; | |
266 | (*pos)++; | |
b58fed8b | 267 | return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); |
1da177e4 LT |
268 | } |
269 | ||
270 | static void m_stop(struct seq_file *m, void *v) | |
271 | { | |
390c6843 | 272 | up_read(&namespace_sem); |
1da177e4 LT |
273 | } |
274 | ||
275 | static inline void mangle(struct seq_file *m, const char *s) | |
276 | { | |
277 | seq_escape(m, s, " \t\n\\"); | |
278 | } | |
279 | ||
280 | static int show_vfsmnt(struct seq_file *m, void *v) | |
281 | { | |
282 | struct vfsmount *mnt = v; | |
283 | int err = 0; | |
284 | static struct proc_fs_info { | |
285 | int flag; | |
286 | char *str; | |
287 | } fs_info[] = { | |
288 | { MS_SYNCHRONOUS, ",sync" }, | |
289 | { MS_DIRSYNC, ",dirsync" }, | |
290 | { MS_MANDLOCK, ",mand" }, | |
291 | { MS_NOATIME, ",noatime" }, | |
292 | { MS_NODIRATIME, ",nodiratime" }, | |
293 | { 0, NULL } | |
294 | }; | |
295 | static struct proc_fs_info mnt_info[] = { | |
296 | { MNT_NOSUID, ",nosuid" }, | |
297 | { MNT_NODEV, ",nodev" }, | |
298 | { MNT_NOEXEC, ",noexec" }, | |
299 | { 0, NULL } | |
300 | }; | |
301 | struct proc_fs_info *fs_infop; | |
302 | ||
303 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
304 | seq_putc(m, ' '); | |
305 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
306 | seq_putc(m, ' '); | |
307 | mangle(m, mnt->mnt_sb->s_type->name); | |
308 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
309 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
310 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
311 | seq_puts(m, fs_infop->str); | |
312 | } | |
313 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
314 | if (mnt->mnt_flags & fs_infop->flag) | |
315 | seq_puts(m, fs_infop->str); | |
316 | } | |
317 | if (mnt->mnt_sb->s_op->show_options) | |
318 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
319 | seq_puts(m, " 0 0\n"); | |
320 | return err; | |
321 | } | |
322 | ||
323 | struct seq_operations mounts_op = { | |
324 | .start = m_start, | |
325 | .next = m_next, | |
326 | .stop = m_stop, | |
327 | .show = show_vfsmnt | |
328 | }; | |
329 | ||
330 | /** | |
331 | * may_umount_tree - check if a mount tree is busy | |
332 | * @mnt: root of mount tree | |
333 | * | |
334 | * This is called to check if a tree of mounts has any | |
335 | * open files, pwds, chroots or sub mounts that are | |
336 | * busy. | |
337 | */ | |
338 | int may_umount_tree(struct vfsmount *mnt) | |
339 | { | |
36341f64 RP |
340 | int actual_refs = 0; |
341 | int minimum_refs = 0; | |
342 | struct vfsmount *p; | |
1da177e4 LT |
343 | |
344 | spin_lock(&vfsmount_lock); | |
36341f64 | 345 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
346 | actual_refs += atomic_read(&p->mnt_count); |
347 | minimum_refs += 2; | |
1da177e4 LT |
348 | } |
349 | spin_unlock(&vfsmount_lock); | |
350 | ||
351 | if (actual_refs > minimum_refs) | |
352 | return -EBUSY; | |
353 | ||
354 | return 0; | |
355 | } | |
356 | ||
357 | EXPORT_SYMBOL(may_umount_tree); | |
358 | ||
359 | /** | |
360 | * may_umount - check if a mount point is busy | |
361 | * @mnt: root of mount | |
362 | * | |
363 | * This is called to check if a mount point has any | |
364 | * open files, pwds, chroots or sub mounts. If the | |
365 | * mount has sub mounts this will return busy | |
366 | * regardless of whether the sub mounts are busy. | |
367 | * | |
368 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
369 | * give false negatives. The main reason why it's here is that we need | |
370 | * a non-destructive way to look for easily umountable filesystems. | |
371 | */ | |
372 | int may_umount(struct vfsmount *mnt) | |
373 | { | |
374 | if (atomic_read(&mnt->mnt_count) > 2) | |
375 | return -EBUSY; | |
376 | return 0; | |
377 | } | |
378 | ||
379 | EXPORT_SYMBOL(may_umount); | |
380 | ||
70fbcdf4 RP |
381 | static void release_mounts(struct list_head *head) |
382 | { | |
383 | struct vfsmount *mnt; | |
384 | while(!list_empty(head)) { | |
385 | mnt = list_entry(head->next, struct vfsmount, mnt_hash); | |
386 | list_del_init(&mnt->mnt_hash); | |
387 | if (mnt->mnt_parent != mnt) { | |
388 | struct dentry *dentry; | |
389 | struct vfsmount *m; | |
390 | spin_lock(&vfsmount_lock); | |
391 | dentry = mnt->mnt_mountpoint; | |
392 | m = mnt->mnt_parent; | |
393 | mnt->mnt_mountpoint = mnt->mnt_root; | |
394 | mnt->mnt_parent = mnt; | |
395 | spin_unlock(&vfsmount_lock); | |
396 | dput(dentry); | |
397 | mntput(m); | |
398 | } | |
399 | mntput(mnt); | |
400 | } | |
401 | } | |
402 | ||
403 | static void umount_tree(struct vfsmount *mnt, struct list_head *kill) | |
1da177e4 LT |
404 | { |
405 | struct vfsmount *p; | |
1da177e4 LT |
406 | |
407 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
70fbcdf4 RP |
408 | list_del(&p->mnt_hash); |
409 | list_add(&p->mnt_hash, kill); | |
1da177e4 LT |
410 | } |
411 | ||
70fbcdf4 RP |
412 | list_for_each_entry(p, kill, mnt_hash) { |
413 | list_del_init(&p->mnt_expire); | |
414 | list_del_init(&p->mnt_list); | |
415 | __touch_namespace(p->mnt_namespace); | |
416 | p->mnt_namespace = NULL; | |
417 | list_del_init(&p->mnt_child); | |
418 | if (p->mnt_parent != p) | |
419 | mnt->mnt_mountpoint->d_mounted--; | |
1da177e4 LT |
420 | } |
421 | } | |
422 | ||
423 | static int do_umount(struct vfsmount *mnt, int flags) | |
424 | { | |
b58fed8b | 425 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 426 | int retval; |
70fbcdf4 | 427 | LIST_HEAD(umount_list); |
1da177e4 LT |
428 | |
429 | retval = security_sb_umount(mnt, flags); | |
430 | if (retval) | |
431 | return retval; | |
432 | ||
433 | /* | |
434 | * Allow userspace to request a mountpoint be expired rather than | |
435 | * unmounting unconditionally. Unmount only happens if: | |
436 | * (1) the mark is already set (the mark is cleared by mntput()) | |
437 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
438 | */ | |
439 | if (flags & MNT_EXPIRE) { | |
440 | if (mnt == current->fs->rootmnt || | |
441 | flags & (MNT_FORCE | MNT_DETACH)) | |
442 | return -EINVAL; | |
443 | ||
444 | if (atomic_read(&mnt->mnt_count) != 2) | |
445 | return -EBUSY; | |
446 | ||
447 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
448 | return -EAGAIN; | |
449 | } | |
450 | ||
451 | /* | |
452 | * If we may have to abort operations to get out of this | |
453 | * mount, and they will themselves hold resources we must | |
454 | * allow the fs to do things. In the Unix tradition of | |
455 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
456 | * might fail to complete on the first run through as other tasks | |
457 | * must return, and the like. Thats for the mount program to worry | |
458 | * about for the moment. | |
459 | */ | |
460 | ||
461 | lock_kernel(); | |
b58fed8b | 462 | if ((flags & MNT_FORCE) && sb->s_op->umount_begin) |
1da177e4 LT |
463 | sb->s_op->umount_begin(sb); |
464 | unlock_kernel(); | |
465 | ||
466 | /* | |
467 | * No sense to grab the lock for this test, but test itself looks | |
468 | * somewhat bogus. Suggestions for better replacement? | |
469 | * Ho-hum... In principle, we might treat that as umount + switch | |
470 | * to rootfs. GC would eventually take care of the old vfsmount. | |
471 | * Actually it makes sense, especially if rootfs would contain a | |
472 | * /reboot - static binary that would close all descriptors and | |
473 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
474 | */ | |
475 | if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { | |
476 | /* | |
477 | * Special case for "unmounting" root ... | |
478 | * we just try to remount it readonly. | |
479 | */ | |
480 | down_write(&sb->s_umount); | |
481 | if (!(sb->s_flags & MS_RDONLY)) { | |
482 | lock_kernel(); | |
483 | DQUOT_OFF(sb); | |
484 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
485 | unlock_kernel(); | |
486 | } | |
487 | up_write(&sb->s_umount); | |
488 | return retval; | |
489 | } | |
490 | ||
390c6843 | 491 | down_write(&namespace_sem); |
1da177e4 | 492 | spin_lock(&vfsmount_lock); |
5addc5dd | 493 | event++; |
1da177e4 | 494 | |
1da177e4 LT |
495 | retval = -EBUSY; |
496 | if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) { | |
497 | if (!list_empty(&mnt->mnt_list)) | |
70fbcdf4 | 498 | umount_tree(mnt, &umount_list); |
1da177e4 LT |
499 | retval = 0; |
500 | } | |
501 | spin_unlock(&vfsmount_lock); | |
502 | if (retval) | |
503 | security_sb_umount_busy(mnt); | |
390c6843 | 504 | up_write(&namespace_sem); |
70fbcdf4 | 505 | release_mounts(&umount_list); |
1da177e4 LT |
506 | return retval; |
507 | } | |
508 | ||
509 | /* | |
510 | * Now umount can handle mount points as well as block devices. | |
511 | * This is important for filesystems which use unnamed block devices. | |
512 | * | |
513 | * We now support a flag for forced unmount like the other 'big iron' | |
514 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
515 | */ | |
516 | ||
517 | asmlinkage long sys_umount(char __user * name, int flags) | |
518 | { | |
519 | struct nameidata nd; | |
520 | int retval; | |
521 | ||
522 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
523 | if (retval) | |
524 | goto out; | |
525 | retval = -EINVAL; | |
526 | if (nd.dentry != nd.mnt->mnt_root) | |
527 | goto dput_and_out; | |
528 | if (!check_mnt(nd.mnt)) | |
529 | goto dput_and_out; | |
530 | ||
531 | retval = -EPERM; | |
532 | if (!capable(CAP_SYS_ADMIN)) | |
533 | goto dput_and_out; | |
534 | ||
535 | retval = do_umount(nd.mnt, flags); | |
536 | dput_and_out: | |
537 | path_release_on_umount(&nd); | |
538 | out: | |
539 | return retval; | |
540 | } | |
541 | ||
542 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
543 | ||
544 | /* | |
b58fed8b | 545 | * The 2.0 compatible umount. No flags. |
1da177e4 | 546 | */ |
1da177e4 LT |
547 | asmlinkage long sys_oldumount(char __user * name) |
548 | { | |
b58fed8b | 549 | return sys_umount(name, 0); |
1da177e4 LT |
550 | } |
551 | ||
552 | #endif | |
553 | ||
554 | static int mount_is_safe(struct nameidata *nd) | |
555 | { | |
556 | if (capable(CAP_SYS_ADMIN)) | |
557 | return 0; | |
558 | return -EPERM; | |
559 | #ifdef notyet | |
560 | if (S_ISLNK(nd->dentry->d_inode->i_mode)) | |
561 | return -EPERM; | |
562 | if (nd->dentry->d_inode->i_mode & S_ISVTX) { | |
563 | if (current->uid != nd->dentry->d_inode->i_uid) | |
564 | return -EPERM; | |
565 | } | |
566 | if (permission(nd->dentry->d_inode, MAY_WRITE, nd)) | |
567 | return -EPERM; | |
568 | return 0; | |
569 | #endif | |
570 | } | |
571 | ||
b58fed8b | 572 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
573 | { |
574 | while (1) { | |
575 | if (d == dentry) | |
576 | return 1; | |
577 | if (d == NULL || d == d->d_parent) | |
578 | return 0; | |
579 | d = d->d_parent; | |
580 | } | |
581 | } | |
582 | ||
36341f64 RP |
583 | static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
584 | int flag) | |
1da177e4 LT |
585 | { |
586 | struct vfsmount *res, *p, *q, *r, *s; | |
1da177e4 LT |
587 | struct nameidata nd; |
588 | ||
36341f64 | 589 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
590 | if (!q) |
591 | goto Enomem; | |
592 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
593 | ||
594 | p = mnt; | |
fdadd65f | 595 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
596 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
597 | continue; | |
598 | ||
599 | for (s = r; s; s = next_mnt(s, r)) { | |
600 | while (p != s->mnt_parent) { | |
601 | p = p->mnt_parent; | |
602 | q = q->mnt_parent; | |
603 | } | |
604 | p = s; | |
605 | nd.mnt = q; | |
606 | nd.dentry = p->mnt_mountpoint; | |
36341f64 | 607 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
608 | if (!q) |
609 | goto Enomem; | |
610 | spin_lock(&vfsmount_lock); | |
611 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
612 | attach_mnt(q, &nd); | |
613 | spin_unlock(&vfsmount_lock); | |
614 | } | |
615 | } | |
616 | return res; | |
b58fed8b | 617 | Enomem: |
1da177e4 | 618 | if (res) { |
70fbcdf4 | 619 | LIST_HEAD(umount_list); |
1da177e4 | 620 | spin_lock(&vfsmount_lock); |
70fbcdf4 | 621 | umount_tree(res, &umount_list); |
1da177e4 | 622 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 623 | release_mounts(&umount_list); |
1da177e4 LT |
624 | } |
625 | return NULL; | |
626 | } | |
627 | ||
628 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) | |
629 | { | |
630 | int err; | |
631 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
632 | return -EINVAL; | |
633 | ||
634 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
635 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) | |
636 | return -ENOTDIR; | |
637 | ||
638 | err = -ENOENT; | |
639 | down(&nd->dentry->d_inode->i_sem); | |
640 | if (IS_DEADDIR(nd->dentry->d_inode)) | |
641 | goto out_unlock; | |
642 | ||
643 | err = security_sb_check_sb(mnt, nd); | |
644 | if (err) | |
645 | goto out_unlock; | |
646 | ||
647 | err = -ENOENT; | |
648 | spin_lock(&vfsmount_lock); | |
649 | if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) { | |
650 | struct list_head head; | |
651 | ||
652 | attach_mnt(mnt, nd); | |
653 | list_add_tail(&head, &mnt->mnt_list); | |
654 | list_splice(&head, current->namespace->list.prev); | |
1da177e4 | 655 | err = 0; |
5addc5dd | 656 | touch_namespace(current->namespace); |
1da177e4 LT |
657 | } |
658 | spin_unlock(&vfsmount_lock); | |
659 | out_unlock: | |
660 | up(&nd->dentry->d_inode->i_sem); | |
661 | if (!err) | |
662 | security_sb_post_addmount(mnt, nd); | |
663 | return err; | |
664 | } | |
665 | ||
07b20889 RP |
666 | /* |
667 | * recursively change the type of the mountpoint. | |
668 | */ | |
669 | static int do_change_type(struct nameidata *nd, int flag) | |
670 | { | |
671 | struct vfsmount *m, *mnt = nd->mnt; | |
672 | int recurse = flag & MS_REC; | |
673 | int type = flag & ~MS_REC; | |
674 | ||
675 | if (nd->dentry != nd->mnt->mnt_root) | |
676 | return -EINVAL; | |
677 | ||
678 | down_write(&namespace_sem); | |
679 | spin_lock(&vfsmount_lock); | |
680 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
681 | change_mnt_propagation(m, type); | |
682 | spin_unlock(&vfsmount_lock); | |
683 | up_write(&namespace_sem); | |
684 | return 0; | |
685 | } | |
686 | ||
1da177e4 LT |
687 | /* |
688 | * do loopback mount. | |
689 | */ | |
690 | static int do_loopback(struct nameidata *nd, char *old_name, int recurse) | |
691 | { | |
692 | struct nameidata old_nd; | |
693 | struct vfsmount *mnt = NULL; | |
694 | int err = mount_is_safe(nd); | |
695 | if (err) | |
696 | return err; | |
697 | if (!old_name || !*old_name) | |
698 | return -EINVAL; | |
699 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
700 | if (err) | |
701 | return err; | |
702 | ||
390c6843 | 703 | down_write(&namespace_sem); |
1da177e4 | 704 | err = -EINVAL; |
ccd48bc7 AV |
705 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) |
706 | goto out; | |
1da177e4 | 707 | |
ccd48bc7 AV |
708 | err = -ENOMEM; |
709 | if (recurse) | |
36341f64 | 710 | mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 | 711 | else |
36341f64 | 712 | mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 AV |
713 | |
714 | if (!mnt) | |
715 | goto out; | |
716 | ||
ccd48bc7 AV |
717 | err = graft_tree(mnt, nd); |
718 | if (err) { | |
70fbcdf4 | 719 | LIST_HEAD(umount_list); |
1da177e4 | 720 | spin_lock(&vfsmount_lock); |
70fbcdf4 | 721 | umount_tree(mnt, &umount_list); |
1da177e4 | 722 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 723 | release_mounts(&umount_list); |
5b83d2c5 | 724 | } |
1da177e4 | 725 | |
ccd48bc7 | 726 | out: |
390c6843 | 727 | up_write(&namespace_sem); |
1da177e4 LT |
728 | path_release(&old_nd); |
729 | return err; | |
730 | } | |
731 | ||
732 | /* | |
733 | * change filesystem flags. dir should be a physical root of filesystem. | |
734 | * If you've mounted a non-root directory somewhere and want to do remount | |
735 | * on it - tough luck. | |
736 | */ | |
1da177e4 LT |
737 | static int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
738 | void *data) | |
739 | { | |
740 | int err; | |
b58fed8b | 741 | struct super_block *sb = nd->mnt->mnt_sb; |
1da177e4 LT |
742 | |
743 | if (!capable(CAP_SYS_ADMIN)) | |
744 | return -EPERM; | |
745 | ||
746 | if (!check_mnt(nd->mnt)) | |
747 | return -EINVAL; | |
748 | ||
749 | if (nd->dentry != nd->mnt->mnt_root) | |
750 | return -EINVAL; | |
751 | ||
752 | down_write(&sb->s_umount); | |
753 | err = do_remount_sb(sb, flags, data, 0); | |
754 | if (!err) | |
b58fed8b | 755 | nd->mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
756 | up_write(&sb->s_umount); |
757 | if (!err) | |
758 | security_sb_post_remount(nd->mnt, flags, data); | |
759 | return err; | |
760 | } | |
761 | ||
762 | static int do_move_mount(struct nameidata *nd, char *old_name) | |
763 | { | |
764 | struct nameidata old_nd, parent_nd; | |
765 | struct vfsmount *p; | |
766 | int err = 0; | |
767 | if (!capable(CAP_SYS_ADMIN)) | |
768 | return -EPERM; | |
769 | if (!old_name || !*old_name) | |
770 | return -EINVAL; | |
771 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
772 | if (err) | |
773 | return err; | |
774 | ||
390c6843 | 775 | down_write(&namespace_sem); |
b58fed8b | 776 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
777 | ; |
778 | err = -EINVAL; | |
779 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) | |
780 | goto out; | |
781 | ||
782 | err = -ENOENT; | |
783 | down(&nd->dentry->d_inode->i_sem); | |
784 | if (IS_DEADDIR(nd->dentry->d_inode)) | |
785 | goto out1; | |
786 | ||
787 | spin_lock(&vfsmount_lock); | |
788 | if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) | |
789 | goto out2; | |
790 | ||
791 | err = -EINVAL; | |
792 | if (old_nd.dentry != old_nd.mnt->mnt_root) | |
793 | goto out2; | |
794 | ||
795 | if (old_nd.mnt == old_nd.mnt->mnt_parent) | |
796 | goto out2; | |
797 | ||
798 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
799 | S_ISDIR(old_nd.dentry->d_inode->i_mode)) | |
800 | goto out2; | |
801 | ||
802 | err = -ELOOP; | |
b58fed8b | 803 | for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1da177e4 LT |
804 | if (p == old_nd.mnt) |
805 | goto out2; | |
806 | err = 0; | |
807 | ||
808 | detach_mnt(old_nd.mnt, &parent_nd); | |
809 | attach_mnt(old_nd.mnt, nd); | |
5addc5dd | 810 | touch_namespace(current->namespace); |
1da177e4 LT |
811 | |
812 | /* if the mount is moved, it should no longer be expire | |
813 | * automatically */ | |
55e700b9 | 814 | list_del_init(&old_nd.mnt->mnt_expire); |
1da177e4 LT |
815 | out2: |
816 | spin_unlock(&vfsmount_lock); | |
817 | out1: | |
818 | up(&nd->dentry->d_inode->i_sem); | |
819 | out: | |
390c6843 | 820 | up_write(&namespace_sem); |
1da177e4 LT |
821 | if (!err) |
822 | path_release(&parent_nd); | |
823 | path_release(&old_nd); | |
824 | return err; | |
825 | } | |
826 | ||
827 | /* | |
828 | * create a new mount for userspace and request it to be added into the | |
829 | * namespace's tree | |
830 | */ | |
831 | static int do_new_mount(struct nameidata *nd, char *type, int flags, | |
832 | int mnt_flags, char *name, void *data) | |
833 | { | |
834 | struct vfsmount *mnt; | |
835 | ||
836 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
837 | return -EINVAL; | |
838 | ||
839 | /* we need capabilities... */ | |
840 | if (!capable(CAP_SYS_ADMIN)) | |
841 | return -EPERM; | |
842 | ||
843 | mnt = do_kern_mount(type, flags, name, data); | |
844 | if (IS_ERR(mnt)) | |
845 | return PTR_ERR(mnt); | |
846 | ||
847 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
848 | } | |
849 | ||
850 | /* | |
851 | * add a mount into a namespace's mount tree | |
852 | * - provide the option of adding the new mount to an expiration list | |
853 | */ | |
854 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
855 | int mnt_flags, struct list_head *fslist) | |
856 | { | |
857 | int err; | |
858 | ||
390c6843 | 859 | down_write(&namespace_sem); |
1da177e4 | 860 | /* Something was mounted here while we slept */ |
b58fed8b | 861 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
862 | ; |
863 | err = -EINVAL; | |
864 | if (!check_mnt(nd->mnt)) | |
865 | goto unlock; | |
866 | ||
867 | /* Refuse the same filesystem on the same mount point */ | |
868 | err = -EBUSY; | |
869 | if (nd->mnt->mnt_sb == newmnt->mnt_sb && | |
870 | nd->mnt->mnt_root == nd->dentry) | |
871 | goto unlock; | |
872 | ||
873 | err = -EINVAL; | |
874 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
875 | goto unlock; | |
876 | ||
877 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
878 | if ((err = graft_tree(newmnt, nd))) |
879 | goto unlock; | |
1da177e4 | 880 | |
5b83d2c5 | 881 | if (fslist) { |
1da177e4 LT |
882 | /* add to the specified expiration list */ |
883 | spin_lock(&vfsmount_lock); | |
55e700b9 | 884 | list_add_tail(&newmnt->mnt_expire, fslist); |
1da177e4 LT |
885 | spin_unlock(&vfsmount_lock); |
886 | } | |
390c6843 | 887 | up_write(&namespace_sem); |
5b83d2c5 | 888 | return 0; |
1da177e4 LT |
889 | |
890 | unlock: | |
390c6843 | 891 | up_write(&namespace_sem); |
1da177e4 LT |
892 | mntput(newmnt); |
893 | return err; | |
894 | } | |
895 | ||
896 | EXPORT_SYMBOL_GPL(do_add_mount); | |
897 | ||
70fbcdf4 RP |
898 | static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, |
899 | struct list_head *umounts) | |
24ca2af1 MS |
900 | { |
901 | spin_lock(&vfsmount_lock); | |
902 | ||
ed42c879 MS |
903 | /* |
904 | * Check if mount is still attached, if not, let whoever holds it deal | |
905 | * with the sucker | |
906 | */ | |
907 | if (mnt->mnt_parent == mnt) { | |
908 | spin_unlock(&vfsmount_lock); | |
909 | return; | |
910 | } | |
911 | ||
24ca2af1 MS |
912 | /* |
913 | * Check that it is still dead: the count should now be 2 - as | |
914 | * contributed by the vfsmount parent and the mntget above | |
915 | */ | |
916 | if (atomic_read(&mnt->mnt_count) == 2) { | |
24ca2af1 | 917 | /* delete from the namespace */ |
5addc5dd | 918 | touch_namespace(mnt->mnt_namespace); |
24ca2af1 | 919 | list_del_init(&mnt->mnt_list); |
ac081153 | 920 | mnt->mnt_namespace = NULL; |
70fbcdf4 | 921 | umount_tree(mnt, umounts); |
24ca2af1 | 922 | spin_unlock(&vfsmount_lock); |
24ca2af1 MS |
923 | } else { |
924 | /* | |
925 | * Someone brought it back to life whilst we didn't have any | |
926 | * locks held so return it to the expiration list | |
927 | */ | |
55e700b9 | 928 | list_add_tail(&mnt->mnt_expire, mounts); |
24ca2af1 MS |
929 | spin_unlock(&vfsmount_lock); |
930 | } | |
931 | } | |
932 | ||
1da177e4 LT |
933 | /* |
934 | * process a list of expirable mountpoints with the intent of discarding any | |
935 | * mountpoints that aren't in use and haven't been touched since last we came | |
936 | * here | |
937 | */ | |
938 | void mark_mounts_for_expiry(struct list_head *mounts) | |
939 | { | |
940 | struct namespace *namespace; | |
941 | struct vfsmount *mnt, *next; | |
942 | LIST_HEAD(graveyard); | |
943 | ||
944 | if (list_empty(mounts)) | |
945 | return; | |
946 | ||
947 | spin_lock(&vfsmount_lock); | |
948 | ||
949 | /* extract from the expiration list every vfsmount that matches the | |
950 | * following criteria: | |
951 | * - only referenced by its parent vfsmount | |
952 | * - still marked for expiry (marked on the last call here; marks are | |
953 | * cleared by mntput()) | |
954 | */ | |
55e700b9 | 955 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 LT |
956 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
957 | atomic_read(&mnt->mnt_count) != 1) | |
958 | continue; | |
959 | ||
960 | mntget(mnt); | |
55e700b9 | 961 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 LT |
962 | } |
963 | ||
964 | /* | |
965 | * go through the vfsmounts we've just consigned to the graveyard to | |
966 | * - check that they're still dead | |
967 | * - delete the vfsmount from the appropriate namespace under lock | |
968 | * - dispose of the corpse | |
969 | */ | |
970 | while (!list_empty(&graveyard)) { | |
70fbcdf4 | 971 | LIST_HEAD(umounts); |
55e700b9 MS |
972 | mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire); |
973 | list_del_init(&mnt->mnt_expire); | |
1da177e4 LT |
974 | |
975 | /* don't do anything if the namespace is dead - all the | |
976 | * vfsmounts from it are going away anyway */ | |
977 | namespace = mnt->mnt_namespace; | |
1ce88cf4 | 978 | if (!namespace || !namespace->root) |
1da177e4 LT |
979 | continue; |
980 | get_namespace(namespace); | |
981 | ||
982 | spin_unlock(&vfsmount_lock); | |
390c6843 | 983 | down_write(&namespace_sem); |
70fbcdf4 | 984 | expire_mount(mnt, mounts, &umounts); |
390c6843 | 985 | up_write(&namespace_sem); |
70fbcdf4 | 986 | release_mounts(&umounts); |
1da177e4 LT |
987 | mntput(mnt); |
988 | put_namespace(namespace); | |
1da177e4 LT |
989 | spin_lock(&vfsmount_lock); |
990 | } | |
991 | ||
992 | spin_unlock(&vfsmount_lock); | |
993 | } | |
994 | ||
995 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
996 | ||
997 | /* | |
998 | * Some copy_from_user() implementations do not return the exact number of | |
999 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1000 | * Note that this function differs from copy_from_user() in that it will oops | |
1001 | * on bad values of `to', rather than returning a short copy. | |
1002 | */ | |
b58fed8b RP |
1003 | static long exact_copy_from_user(void *to, const void __user * from, |
1004 | unsigned long n) | |
1da177e4 LT |
1005 | { |
1006 | char *t = to; | |
1007 | const char __user *f = from; | |
1008 | char c; | |
1009 | ||
1010 | if (!access_ok(VERIFY_READ, from, n)) | |
1011 | return n; | |
1012 | ||
1013 | while (n) { | |
1014 | if (__get_user(c, f)) { | |
1015 | memset(t, 0, n); | |
1016 | break; | |
1017 | } | |
1018 | *t++ = c; | |
1019 | f++; | |
1020 | n--; | |
1021 | } | |
1022 | return n; | |
1023 | } | |
1024 | ||
b58fed8b | 1025 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1026 | { |
1027 | int i; | |
1028 | unsigned long page; | |
1029 | unsigned long size; | |
b58fed8b | 1030 | |
1da177e4 LT |
1031 | *where = 0; |
1032 | if (!data) | |
1033 | return 0; | |
1034 | ||
1035 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1036 | return -ENOMEM; | |
1037 | ||
1038 | /* We only care that *some* data at the address the user | |
1039 | * gave us is valid. Just in case, we'll zero | |
1040 | * the remainder of the page. | |
1041 | */ | |
1042 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1043 | size = TASK_SIZE - (unsigned long)data; | |
1044 | if (size > PAGE_SIZE) | |
1045 | size = PAGE_SIZE; | |
1046 | ||
1047 | i = size - exact_copy_from_user((void *)page, data, size); | |
1048 | if (!i) { | |
b58fed8b | 1049 | free_page(page); |
1da177e4 LT |
1050 | return -EFAULT; |
1051 | } | |
1052 | if (i != PAGE_SIZE) | |
1053 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1054 | *where = page; | |
1055 | return 0; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1060 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1061 | * | |
1062 | * data is a (void *) that can point to any structure up to | |
1063 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1064 | * information (or be NULL). | |
1065 | * | |
1066 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1067 | * When the flags word was introduced its top half was required | |
1068 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1069 | * Therefore, if this magic number is present, it carries no information | |
1070 | * and must be discarded. | |
1071 | */ | |
b58fed8b | 1072 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1073 | unsigned long flags, void *data_page) |
1074 | { | |
1075 | struct nameidata nd; | |
1076 | int retval = 0; | |
1077 | int mnt_flags = 0; | |
1078 | ||
1079 | /* Discard magic */ | |
1080 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1081 | flags &= ~MS_MGC_MSK; | |
1082 | ||
1083 | /* Basic sanity checks */ | |
1084 | ||
1085 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1086 | return -EINVAL; | |
1087 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1088 | return -EINVAL; | |
1089 | ||
1090 | if (data_page) | |
1091 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1092 | ||
1093 | /* Separate the per-mountpoint flags */ | |
1094 | if (flags & MS_NOSUID) | |
1095 | mnt_flags |= MNT_NOSUID; | |
1096 | if (flags & MS_NODEV) | |
1097 | mnt_flags |= MNT_NODEV; | |
1098 | if (flags & MS_NOEXEC) | |
1099 | mnt_flags |= MNT_NOEXEC; | |
b58fed8b | 1100 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE); |
1da177e4 LT |
1101 | |
1102 | /* ... and get the mountpoint */ | |
1103 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1104 | if (retval) | |
1105 | return retval; | |
1106 | ||
1107 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1108 | if (retval) | |
1109 | goto dput_out; | |
1110 | ||
1111 | if (flags & MS_REMOUNT) | |
1112 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1113 | data_page); | |
1114 | else if (flags & MS_BIND) | |
1115 | retval = do_loopback(&nd, dev_name, flags & MS_REC); | |
07b20889 RP |
1116 | else if (flags & MS_PRIVATE) |
1117 | retval = do_change_type(&nd, flags); | |
1da177e4 LT |
1118 | else if (flags & MS_MOVE) |
1119 | retval = do_move_mount(&nd, dev_name); | |
1120 | else | |
1121 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1122 | dev_name, data_page); | |
1123 | dput_out: | |
1124 | path_release(&nd); | |
1125 | return retval; | |
1126 | } | |
1127 | ||
1128 | int copy_namespace(int flags, struct task_struct *tsk) | |
1129 | { | |
1130 | struct namespace *namespace = tsk->namespace; | |
1131 | struct namespace *new_ns; | |
1132 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; | |
1133 | struct fs_struct *fs = tsk->fs; | |
1134 | struct vfsmount *p, *q; | |
1135 | ||
1136 | if (!namespace) | |
1137 | return 0; | |
1138 | ||
1139 | get_namespace(namespace); | |
1140 | ||
1141 | if (!(flags & CLONE_NEWNS)) | |
1142 | return 0; | |
1143 | ||
1144 | if (!capable(CAP_SYS_ADMIN)) { | |
1145 | put_namespace(namespace); | |
1146 | return -EPERM; | |
1147 | } | |
1148 | ||
1149 | new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); | |
1150 | if (!new_ns) | |
1151 | goto out; | |
1152 | ||
1153 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1154 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1155 | init_waitqueue_head(&new_ns->poll); |
1156 | new_ns->event = 0; | |
1da177e4 | 1157 | |
390c6843 | 1158 | down_write(&namespace_sem); |
1da177e4 | 1159 | /* First pass: copy the tree topology */ |
36341f64 RP |
1160 | new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, |
1161 | CL_EXPIRE); | |
1da177e4 | 1162 | if (!new_ns->root) { |
390c6843 | 1163 | up_write(&namespace_sem); |
1da177e4 LT |
1164 | kfree(new_ns); |
1165 | goto out; | |
1166 | } | |
1167 | spin_lock(&vfsmount_lock); | |
1168 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1169 | spin_unlock(&vfsmount_lock); | |
1170 | ||
1171 | /* | |
1172 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1173 | * as belonging to new namespace. We have already acquired a private | |
1174 | * fs_struct, so tsk->fs->lock is not needed. | |
1175 | */ | |
1176 | p = namespace->root; | |
1177 | q = new_ns->root; | |
1178 | while (p) { | |
1179 | q->mnt_namespace = new_ns; | |
1180 | if (fs) { | |
1181 | if (p == fs->rootmnt) { | |
1182 | rootmnt = p; | |
1183 | fs->rootmnt = mntget(q); | |
1184 | } | |
1185 | if (p == fs->pwdmnt) { | |
1186 | pwdmnt = p; | |
1187 | fs->pwdmnt = mntget(q); | |
1188 | } | |
1189 | if (p == fs->altrootmnt) { | |
1190 | altrootmnt = p; | |
1191 | fs->altrootmnt = mntget(q); | |
1192 | } | |
1193 | } | |
1194 | p = next_mnt(p, namespace->root); | |
1195 | q = next_mnt(q, new_ns->root); | |
1196 | } | |
390c6843 | 1197 | up_write(&namespace_sem); |
1da177e4 LT |
1198 | |
1199 | tsk->namespace = new_ns; | |
1200 | ||
1201 | if (rootmnt) | |
1202 | mntput(rootmnt); | |
1203 | if (pwdmnt) | |
1204 | mntput(pwdmnt); | |
1205 | if (altrootmnt) | |
1206 | mntput(altrootmnt); | |
1207 | ||
1208 | put_namespace(namespace); | |
1209 | return 0; | |
1210 | ||
1211 | out: | |
1212 | put_namespace(namespace); | |
1213 | return -ENOMEM; | |
1214 | } | |
1215 | ||
1216 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1217 | char __user * type, unsigned long flags, | |
1218 | void __user * data) | |
1219 | { | |
1220 | int retval; | |
1221 | unsigned long data_page; | |
1222 | unsigned long type_page; | |
1223 | unsigned long dev_page; | |
1224 | char *dir_page; | |
1225 | ||
b58fed8b | 1226 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1227 | if (retval < 0) |
1228 | return retval; | |
1229 | ||
1230 | dir_page = getname(dir_name); | |
1231 | retval = PTR_ERR(dir_page); | |
1232 | if (IS_ERR(dir_page)) | |
1233 | goto out1; | |
1234 | ||
b58fed8b | 1235 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1236 | if (retval < 0) |
1237 | goto out2; | |
1238 | ||
b58fed8b | 1239 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1240 | if (retval < 0) |
1241 | goto out3; | |
1242 | ||
1243 | lock_kernel(); | |
b58fed8b RP |
1244 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1245 | flags, (void *)data_page); | |
1da177e4 LT |
1246 | unlock_kernel(); |
1247 | free_page(data_page); | |
1248 | ||
1249 | out3: | |
1250 | free_page(dev_page); | |
1251 | out2: | |
1252 | putname(dir_page); | |
1253 | out1: | |
1254 | free_page(type_page); | |
1255 | return retval; | |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1260 | * It can block. Requires the big lock held. | |
1261 | */ | |
1262 | void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, | |
1263 | struct dentry *dentry) | |
1264 | { | |
1265 | struct dentry *old_root; | |
1266 | struct vfsmount *old_rootmnt; | |
1267 | write_lock(&fs->lock); | |
1268 | old_root = fs->root; | |
1269 | old_rootmnt = fs->rootmnt; | |
1270 | fs->rootmnt = mntget(mnt); | |
1271 | fs->root = dget(dentry); | |
1272 | write_unlock(&fs->lock); | |
1273 | if (old_root) { | |
1274 | dput(old_root); | |
1275 | mntput(old_rootmnt); | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1281 | * It can block. Requires the big lock held. | |
1282 | */ | |
1283 | void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, | |
1284 | struct dentry *dentry) | |
1285 | { | |
1286 | struct dentry *old_pwd; | |
1287 | struct vfsmount *old_pwdmnt; | |
1288 | ||
1289 | write_lock(&fs->lock); | |
1290 | old_pwd = fs->pwd; | |
1291 | old_pwdmnt = fs->pwdmnt; | |
1292 | fs->pwdmnt = mntget(mnt); | |
1293 | fs->pwd = dget(dentry); | |
1294 | write_unlock(&fs->lock); | |
1295 | ||
1296 | if (old_pwd) { | |
1297 | dput(old_pwd); | |
1298 | mntput(old_pwdmnt); | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) | |
1303 | { | |
1304 | struct task_struct *g, *p; | |
1305 | struct fs_struct *fs; | |
1306 | ||
1307 | read_lock(&tasklist_lock); | |
1308 | do_each_thread(g, p) { | |
1309 | task_lock(p); | |
1310 | fs = p->fs; | |
1311 | if (fs) { | |
1312 | atomic_inc(&fs->count); | |
1313 | task_unlock(p); | |
b58fed8b RP |
1314 | if (fs->root == old_nd->dentry |
1315 | && fs->rootmnt == old_nd->mnt) | |
1da177e4 | 1316 | set_fs_root(fs, new_nd->mnt, new_nd->dentry); |
b58fed8b RP |
1317 | if (fs->pwd == old_nd->dentry |
1318 | && fs->pwdmnt == old_nd->mnt) | |
1da177e4 LT |
1319 | set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); |
1320 | put_fs_struct(fs); | |
1321 | } else | |
1322 | task_unlock(p); | |
1323 | } while_each_thread(g, p); | |
1324 | read_unlock(&tasklist_lock); | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * pivot_root Semantics: | |
1329 | * Moves the root file system of the current process to the directory put_old, | |
1330 | * makes new_root as the new root file system of the current process, and sets | |
1331 | * root/cwd of all processes which had them on the current root to new_root. | |
1332 | * | |
1333 | * Restrictions: | |
1334 | * The new_root and put_old must be directories, and must not be on the | |
1335 | * same file system as the current process root. The put_old must be | |
1336 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1337 | * pointed to by put_old must yield the same directory as new_root. No other | |
1338 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1339 | * | |
1340 | * Notes: | |
1341 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1342 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1343 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1344 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1345 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1346 | * first. | |
1347 | */ | |
b58fed8b RP |
1348 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1349 | const char __user * put_old) | |
1da177e4 LT |
1350 | { |
1351 | struct vfsmount *tmp; | |
1352 | struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; | |
1353 | int error; | |
1354 | ||
1355 | if (!capable(CAP_SYS_ADMIN)) | |
1356 | return -EPERM; | |
1357 | ||
1358 | lock_kernel(); | |
1359 | ||
b58fed8b RP |
1360 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1361 | &new_nd); | |
1da177e4 LT |
1362 | if (error) |
1363 | goto out0; | |
1364 | error = -EINVAL; | |
1365 | if (!check_mnt(new_nd.mnt)) | |
1366 | goto out1; | |
1367 | ||
b58fed8b | 1368 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1369 | if (error) |
1370 | goto out1; | |
1371 | ||
1372 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1373 | if (error) { | |
1374 | path_release(&old_nd); | |
1375 | goto out1; | |
1376 | } | |
1377 | ||
1378 | read_lock(¤t->fs->lock); | |
1379 | user_nd.mnt = mntget(current->fs->rootmnt); | |
1380 | user_nd.dentry = dget(current->fs->root); | |
1381 | read_unlock(¤t->fs->lock); | |
390c6843 | 1382 | down_write(&namespace_sem); |
1da177e4 LT |
1383 | down(&old_nd.dentry->d_inode->i_sem); |
1384 | error = -EINVAL; | |
1385 | if (!check_mnt(user_nd.mnt)) | |
1386 | goto out2; | |
1387 | error = -ENOENT; | |
1388 | if (IS_DEADDIR(new_nd.dentry->d_inode)) | |
1389 | goto out2; | |
1390 | if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) | |
1391 | goto out2; | |
1392 | if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) | |
1393 | goto out2; | |
1394 | error = -EBUSY; | |
1395 | if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) | |
1396 | goto out2; /* loop, on the same file system */ | |
1397 | error = -EINVAL; | |
1398 | if (user_nd.mnt->mnt_root != user_nd.dentry) | |
1399 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1400 | if (user_nd.mnt->mnt_parent == user_nd.mnt) |
1401 | goto out2; /* not attached */ | |
1da177e4 LT |
1402 | if (new_nd.mnt->mnt_root != new_nd.dentry) |
1403 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1404 | if (new_nd.mnt->mnt_parent == new_nd.mnt) |
1405 | goto out2; /* not attached */ | |
1da177e4 LT |
1406 | tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ |
1407 | spin_lock(&vfsmount_lock); | |
1408 | if (tmp != new_nd.mnt) { | |
1409 | for (;;) { | |
1410 | if (tmp->mnt_parent == tmp) | |
1411 | goto out3; /* already mounted on put_old */ | |
1412 | if (tmp->mnt_parent == new_nd.mnt) | |
1413 | break; | |
1414 | tmp = tmp->mnt_parent; | |
1415 | } | |
1416 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) | |
1417 | goto out3; | |
1418 | } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) | |
1419 | goto out3; | |
1420 | detach_mnt(new_nd.mnt, &parent_nd); | |
1421 | detach_mnt(user_nd.mnt, &root_parent); | |
1422 | attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ | |
1423 | attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ | |
5addc5dd | 1424 | touch_namespace(current->namespace); |
1da177e4 LT |
1425 | spin_unlock(&vfsmount_lock); |
1426 | chroot_fs_refs(&user_nd, &new_nd); | |
1427 | security_sb_post_pivotroot(&user_nd, &new_nd); | |
1428 | error = 0; | |
1429 | path_release(&root_parent); | |
1430 | path_release(&parent_nd); | |
1431 | out2: | |
1432 | up(&old_nd.dentry->d_inode->i_sem); | |
390c6843 | 1433 | up_write(&namespace_sem); |
1da177e4 LT |
1434 | path_release(&user_nd); |
1435 | path_release(&old_nd); | |
1436 | out1: | |
1437 | path_release(&new_nd); | |
1438 | out0: | |
1439 | unlock_kernel(); | |
1440 | return error; | |
1441 | out3: | |
1442 | spin_unlock(&vfsmount_lock); | |
1443 | goto out2; | |
1444 | } | |
1445 | ||
1446 | static void __init init_mount_tree(void) | |
1447 | { | |
1448 | struct vfsmount *mnt; | |
1449 | struct namespace *namespace; | |
1450 | struct task_struct *g, *p; | |
1451 | ||
1452 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1453 | if (IS_ERR(mnt)) | |
1454 | panic("Can't create rootfs"); | |
1455 | namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); | |
1456 | if (!namespace) | |
1457 | panic("Can't allocate initial namespace"); | |
1458 | atomic_set(&namespace->count, 1); | |
1459 | INIT_LIST_HEAD(&namespace->list); | |
5addc5dd AV |
1460 | init_waitqueue_head(&namespace->poll); |
1461 | namespace->event = 0; | |
1da177e4 LT |
1462 | list_add(&mnt->mnt_list, &namespace->list); |
1463 | namespace->root = mnt; | |
1464 | mnt->mnt_namespace = namespace; | |
1465 | ||
1466 | init_task.namespace = namespace; | |
1467 | read_lock(&tasklist_lock); | |
1468 | do_each_thread(g, p) { | |
1469 | get_namespace(namespace); | |
1470 | p->namespace = namespace; | |
1471 | } while_each_thread(g, p); | |
1472 | read_unlock(&tasklist_lock); | |
1473 | ||
1474 | set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); | |
1475 | set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); | |
1476 | } | |
1477 | ||
1478 | void __init mnt_init(unsigned long mempages) | |
1479 | { | |
1480 | struct list_head *d; | |
1481 | unsigned int nr_hash; | |
1482 | int i; | |
1483 | ||
390c6843 RP |
1484 | init_rwsem(&namespace_sem); |
1485 | ||
1da177e4 | 1486 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
b58fed8b | 1487 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); |
1da177e4 | 1488 | |
b58fed8b | 1489 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1490 | |
1491 | if (!mount_hashtable) | |
1492 | panic("Failed to allocate mount hash table\n"); | |
1493 | ||
1494 | /* | |
1495 | * Find the power-of-two list-heads that can fit into the allocation.. | |
1496 | * We don't guarantee that "sizeof(struct list_head)" is necessarily | |
1497 | * a power-of-two. | |
1498 | */ | |
1499 | nr_hash = PAGE_SIZE / sizeof(struct list_head); | |
1500 | hash_bits = 0; | |
1501 | do { | |
1502 | hash_bits++; | |
1503 | } while ((nr_hash >> hash_bits) != 0); | |
1504 | hash_bits--; | |
1505 | ||
1506 | /* | |
1507 | * Re-calculate the actual number of entries and the mask | |
1508 | * from the number of bits we can fit. | |
1509 | */ | |
1510 | nr_hash = 1UL << hash_bits; | |
b58fed8b | 1511 | hash_mask = nr_hash - 1; |
1da177e4 LT |
1512 | |
1513 | printk("Mount-cache hash table entries: %d\n", nr_hash); | |
1514 | ||
1515 | /* And initialize the newly allocated array */ | |
1516 | d = mount_hashtable; | |
1517 | i = nr_hash; | |
1518 | do { | |
1519 | INIT_LIST_HEAD(d); | |
1520 | d++; | |
1521 | i--; | |
1522 | } while (i); | |
1523 | sysfs_init(); | |
1524 | init_rootfs(); | |
1525 | init_mount_tree(); | |
1526 | } | |
1527 | ||
1528 | void __put_namespace(struct namespace *namespace) | |
1529 | { | |
1ce88cf4 | 1530 | struct vfsmount *root = namespace->root; |
70fbcdf4 | 1531 | LIST_HEAD(umount_list); |
1ce88cf4 MS |
1532 | namespace->root = NULL; |
1533 | spin_unlock(&vfsmount_lock); | |
390c6843 | 1534 | down_write(&namespace_sem); |
1da177e4 | 1535 | spin_lock(&vfsmount_lock); |
70fbcdf4 | 1536 | umount_tree(root, &umount_list); |
1da177e4 | 1537 | spin_unlock(&vfsmount_lock); |
390c6843 | 1538 | up_write(&namespace_sem); |
70fbcdf4 | 1539 | release_mounts(&umount_list); |
1da177e4 LT |
1540 | kfree(namespace); |
1541 | } |