Merge tag 'v3.10.55' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.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/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/ramfs.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 //#define UMOUNT_LOG //enable kernel layer unmount log when unmount fail
25 #ifdef UMOUNT_LOG
26 #include <linux/types.h>
27 #endif
28 #include <linux/proc_ns.h>
29 #include <linux/magic.h>
30 #include "pnode.h"
31 #include "internal.h"
32
33 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
34 #define HASH_SIZE (1UL << HASH_SHIFT)
35
36 static int event;
37 static DEFINE_IDA(mnt_id_ida);
38 static DEFINE_IDA(mnt_group_ida);
39 static DEFINE_SPINLOCK(mnt_id_lock);
40 static int mnt_id_start = 0;
41 static int mnt_group_start = 1;
42
43 static struct list_head *mount_hashtable __read_mostly;
44 static struct list_head *mountpoint_hashtable __read_mostly;
45 static struct kmem_cache *mnt_cache __read_mostly;
46 static struct rw_semaphore namespace_sem;
47
48 /* /sys/fs */
49 struct kobject *fs_kobj;
50 EXPORT_SYMBOL_GPL(fs_kobj);
51
52 /*
53 * vfsmount lock may be taken for read to prevent changes to the
54 * vfsmount hash, ie. during mountpoint lookups or walking back
55 * up the tree.
56 *
57 * It should be taken for write in all cases where the vfsmount
58 * tree or hash is modified or when a vfsmount structure is modified.
59 */
60 DEFINE_BRLOCK(vfsmount_lock);
61
62 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
63 {
64 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
65 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
66 tmp = tmp + (tmp >> HASH_SHIFT);
67 return tmp & (HASH_SIZE - 1);
68 }
69
70 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
71
72 /*
73 * allocation is serialized by namespace_sem, but we need the spinlock to
74 * serialize with freeing.
75 */
76 static int mnt_alloc_id(struct mount *mnt)
77 {
78 int res;
79
80 retry:
81 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
82 spin_lock(&mnt_id_lock);
83 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
84 if (!res)
85 mnt_id_start = mnt->mnt_id + 1;
86 spin_unlock(&mnt_id_lock);
87 if (res == -EAGAIN)
88 goto retry;
89
90 return res;
91 }
92
93 static void mnt_free_id(struct mount *mnt)
94 {
95 int id = mnt->mnt_id;
96 spin_lock(&mnt_id_lock);
97 ida_remove(&mnt_id_ida, id);
98 if (mnt_id_start > id)
99 mnt_id_start = id;
100 spin_unlock(&mnt_id_lock);
101 }
102
103 /*
104 * Allocate a new peer group ID
105 *
106 * mnt_group_ida is protected by namespace_sem
107 */
108 static int mnt_alloc_group_id(struct mount *mnt)
109 {
110 int res;
111
112 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
113 return -ENOMEM;
114
115 res = ida_get_new_above(&mnt_group_ida,
116 mnt_group_start,
117 &mnt->mnt_group_id);
118 if (!res)
119 mnt_group_start = mnt->mnt_group_id + 1;
120
121 return res;
122 }
123
124 /*
125 * Release a peer group ID
126 */
127 void mnt_release_group_id(struct mount *mnt)
128 {
129 int id = mnt->mnt_group_id;
130 ida_remove(&mnt_group_ida, id);
131 if (mnt_group_start > id)
132 mnt_group_start = id;
133 mnt->mnt_group_id = 0;
134 }
135
136 /*
137 * vfsmount lock must be held for read
138 */
139 #ifdef UMOUNT_LOG
140 #define UMOUNT_Partition "/emmc@usrdata"
141 struct record_ref_count{
142 pid_t pid;
143 char name[TASK_COMM_LEN];
144 int count;
145 struct record_ref_count *next;
146 };
147
148 struct record_ref_count *ref_head = NULL;
149 struct record_ref_count *ref_current = NULL;
150 struct record_ref_count *ref_prev = NULL;
151 int s_total_count = 0;
152 #endif
153 static inline void mnt_add_count(struct mount *mnt, int n)
154 {
155 #ifdef UMOUNT_LOG
156 int print_link_list=0;
157 #ifndef CONFIG_SMP
158 preempt_disable();
159 #endif
160
161 if (strcmp(UMOUNT_Partition,mnt->mnt_devname)==0)
162 {
163 //if (strcmp("mobile_log_d",current->comm)!=0)
164 {
165 //if (current->pid < 100) //((current->pid < 70) && (current->pid > 60))
166 {
167 //printk("Ahsin n=%d current->pid=%d name=%s \n",n,current->pid,current->comm);
168 spin_lock(&mnt_id_lock);
169 if (ref_head == NULL) //linked list head (start)
170 {
171 printk("Ahsin link list init mnt_get_count=%d \n",mnt_get_count(mnt));
172
173 ref_current = kmalloc(sizeof(struct record_ref_count), GFP_KERNEL);
174 if (ref_current == NULL)
175 printk("Ahsin can't allocate memory for ref_current /n");
176
177 ref_current->next = NULL;
178 ref_current->pid = current->pid;
179 strncpy(ref_current->name, current->comm, TASK_COMM_LEN -1);
180 ref_current->name[TASK_COMM_LEN -1] = '\0';
181 ref_current->count = n;
182 s_total_count = s_total_count + n;
183 ref_head = ref_current;
184
185 printk("Ahsin ref_head == NULL pid=%d name=%s counter=%d n=%d \n",ref_current->pid,ref_current->name,ref_current->count,n);
186 }
187 else //check exist first and then add linked list or modify counter
188 {
189 ref_prev = ref_head;
190 while(ref_prev != NULL)
191 {
192 //printk("Ahsin PID= %d, Name= %s, Count= %d n=%d current->pid=%d \n", ref_prev->pid, ref_prev->name, ref_prev->count,n,current->pid);
193 if (strcmp(ref_prev->name,current->comm)==0) //(ref_prev->pid==current->pid)//exist and find, modify counter
194 {
195 ref_prev->count = ref_prev->count + n;
196 s_total_count = s_total_count + n;
197 //printk("Ahsin (ref_prev->name,current->comm) pid=%d name=%s counter=%d n=%d \n",ref_prev->pid,ref_prev->name,ref_prev->count,n);
198 break;
199 }
200 else
201 {
202 if (ref_prev->next != NULL)
203 ref_prev = ref_prev->next;
204 else
205 { // end of link list
206 ref_current = kmalloc(sizeof(struct record_ref_count), GFP_KERNEL);
207 if (ref_current == NULL)
208 printk("Ahsin can't allocate memory for ref_prev /n");
209
210 ref_current->next = NULL;
211 ref_current->pid = current->pid;
212 strncpy(ref_current->name, current->comm, TASK_COMM_LEN -1);
213 ref_current->name[TASK_COMM_LEN -1] = '\0';
214 ref_current->count = n;
215 s_total_count = s_total_count + n;
216 ref_prev->next = ref_current;
217 //printk("Ahsin new node(end of link list) pid=%d name=%s counter=%d n=%d \n",ref_current->pid,ref_current->name,ref_current->count,n);
218 break;
219
220 }
221 }
222 }
223 }
224 spin_unlock(&mnt_id_lock);
225 }
226 }
227 }
228
229 #ifndef CONFIG_SMP
230 preempt_enable();
231 #endif
232 #endif
233 #ifdef CONFIG_SMP
234 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
235 #ifdef UMOUNT_LOG
236 #if 0
237 if (strcmp(UMOUNT_Partition,mnt->mnt_devname)==0)
238 {
239 if (strcmp("mobile_log_d",current->comm)!=0)
240 {
241 printk("Ahsin s_total_count=%d mnt_get_count=%d n=%d current->pid=%d \n",s_total_count,mnt_get_count(mnt),n,current->pid);
242 //if (current->pid < 100)
243 {
244 // print linked list
245 spin_lock(&mnt_id_lock);
246 ref_current = ref_head;
247 while(ref_current != NULL)
248 {
249 if (ref_current->count)
250 {
251 print_link_list = print_link_list + ref_current->count;
252 //printk("Ahsin PID= %d, Name = %s, Count= %d \n", ref_current->pid, ref_current->name, ref_current->count);
253 }
254 ref_current = ref_current->next;
255 }
256 spin_unlock(&mnt_id_lock);
257 printk("Ahsin print_link_list=%d \n",print_link_list);
258 }
259 }
260 }
261 #endif
262 #endif
263 #else
264 preempt_disable();
265 mnt->mnt_count += n;
266 preempt_enable();
267 #endif
268 }
269
270 /*
271 * vfsmount lock must be held for write
272 */
273 unsigned int mnt_get_count(struct mount *mnt)
274 {
275 #ifdef CONFIG_SMP
276 unsigned int count = 0;
277 int cpu;
278
279 for_each_possible_cpu(cpu) {
280 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
281 }
282
283 return count;
284 #else
285 return mnt->mnt_count;
286 #endif
287 }
288
289 static struct mount *alloc_vfsmnt(const char *name)
290 {
291 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
292 if (mnt) {
293 int err;
294
295 err = mnt_alloc_id(mnt);
296 if (err)
297 goto out_free_cache;
298
299 if (name) {
300 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
301 if (!mnt->mnt_devname)
302 goto out_free_id;
303 }
304
305 #ifdef CONFIG_SMP
306 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
307 if (!mnt->mnt_pcp)
308 goto out_free_devname;
309
310 #ifdef UMOUNT_LOG
311 if (strcmp(UMOUNT_Partition,mnt->mnt_devname)==0)
312 {
313 printk("Ahsin alloc_vfsmnt current->pid=%d name=%s \n",current->pid,current->comm);
314 }
315 #endif
316 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
317 #else
318 mnt->mnt_count = 1;
319 mnt->mnt_writers = 0;
320 #endif
321
322 INIT_LIST_HEAD(&mnt->mnt_hash);
323 INIT_LIST_HEAD(&mnt->mnt_child);
324 INIT_LIST_HEAD(&mnt->mnt_mounts);
325 INIT_LIST_HEAD(&mnt->mnt_list);
326 INIT_LIST_HEAD(&mnt->mnt_expire);
327 INIT_LIST_HEAD(&mnt->mnt_share);
328 INIT_LIST_HEAD(&mnt->mnt_slave_list);
329 INIT_LIST_HEAD(&mnt->mnt_slave);
330 #ifdef CONFIG_FSNOTIFY
331 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
332 #endif
333 }
334 return mnt;
335
336 #ifdef CONFIG_SMP
337 out_free_devname:
338 kfree(mnt->mnt_devname);
339 #endif
340 out_free_id:
341 mnt_free_id(mnt);
342 out_free_cache:
343 kmem_cache_free(mnt_cache, mnt);
344 return NULL;
345 }
346
347 /*
348 * Most r/o checks on a fs are for operations that take
349 * discrete amounts of time, like a write() or unlink().
350 * We must keep track of when those operations start
351 * (for permission checks) and when they end, so that
352 * we can determine when writes are able to occur to
353 * a filesystem.
354 */
355 /*
356 * __mnt_is_readonly: check whether a mount is read-only
357 * @mnt: the mount to check for its write status
358 *
359 * This shouldn't be used directly ouside of the VFS.
360 * It does not guarantee that the filesystem will stay
361 * r/w, just that it is right *now*. This can not and
362 * should not be used in place of IS_RDONLY(inode).
363 * mnt_want/drop_write() will _keep_ the filesystem
364 * r/w.
365 */
366 int __mnt_is_readonly(struct vfsmount *mnt)
367 {
368 if (mnt->mnt_flags & MNT_READONLY)
369 return 1;
370 if (mnt->mnt_sb->s_flags & MS_RDONLY)
371 return 1;
372 return 0;
373 }
374 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
375
376 static inline void mnt_inc_writers(struct mount *mnt)
377 {
378 #ifdef CONFIG_SMP
379 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
380 #else
381 mnt->mnt_writers++;
382 #endif
383 }
384
385 static inline void mnt_dec_writers(struct mount *mnt)
386 {
387 #ifdef CONFIG_SMP
388 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
389 #else
390 mnt->mnt_writers--;
391 #endif
392 }
393
394 static unsigned int mnt_get_writers(struct mount *mnt)
395 {
396 #ifdef CONFIG_SMP
397 unsigned int count = 0;
398 int cpu;
399
400 for_each_possible_cpu(cpu) {
401 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
402 }
403
404 return count;
405 #else
406 return mnt->mnt_writers;
407 #endif
408 }
409
410 static int mnt_is_readonly(struct vfsmount *mnt)
411 {
412 if (mnt->mnt_sb->s_readonly_remount)
413 return 1;
414 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
415 smp_rmb();
416 return __mnt_is_readonly(mnt);
417 }
418
419 /*
420 * Most r/o & frozen checks on a fs are for operations that take discrete
421 * amounts of time, like a write() or unlink(). We must keep track of when
422 * those operations start (for permission checks) and when they end, so that we
423 * can determine when writes are able to occur to a filesystem.
424 */
425 /**
426 * __mnt_want_write - get write access to a mount without freeze protection
427 * @m: the mount on which to take a write
428 *
429 * This tells the low-level filesystem that a write is about to be performed to
430 * it, and makes sure that writes are allowed (mnt it read-write) before
431 * returning success. This operation does not protect against filesystem being
432 * frozen. When the write operation is finished, __mnt_drop_write() must be
433 * called. This is effectively a refcount.
434 */
435 int __mnt_want_write(struct vfsmount *m)
436 {
437 struct mount *mnt = real_mount(m);
438 int ret = 0;
439
440 preempt_disable();
441 mnt_inc_writers(mnt);
442 /*
443 * The store to mnt_inc_writers must be visible before we pass
444 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
445 * incremented count after it has set MNT_WRITE_HOLD.
446 */
447 smp_mb();
448 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
449 cpu_relax();
450 /*
451 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
452 * be set to match its requirements. So we must not load that until
453 * MNT_WRITE_HOLD is cleared.
454 */
455 smp_rmb();
456 if (mnt_is_readonly(m)) {
457 mnt_dec_writers(mnt);
458 ret = -EROFS;
459 }
460 preempt_enable();
461
462 return ret;
463 }
464
465 /**
466 * mnt_want_write - get write access to a mount
467 * @m: the mount on which to take a write
468 *
469 * This tells the low-level filesystem that a write is about to be performed to
470 * it, and makes sure that writes are allowed (mount is read-write, filesystem
471 * is not frozen) before returning success. When the write operation is
472 * finished, mnt_drop_write() must be called. This is effectively a refcount.
473 */
474 int mnt_want_write(struct vfsmount *m)
475 {
476 int ret;
477
478 sb_start_write(m->mnt_sb);
479 ret = __mnt_want_write(m);
480 if (ret)
481 sb_end_write(m->mnt_sb);
482 return ret;
483 }
484 EXPORT_SYMBOL_GPL(mnt_want_write);
485
486 /**
487 * mnt_clone_write - get write access to a mount
488 * @mnt: the mount on which to take a write
489 *
490 * This is effectively like mnt_want_write, except
491 * it must only be used to take an extra write reference
492 * on a mountpoint that we already know has a write reference
493 * on it. This allows some optimisation.
494 *
495 * After finished, mnt_drop_write must be called as usual to
496 * drop the reference.
497 */
498 int mnt_clone_write(struct vfsmount *mnt)
499 {
500 /* superblock may be r/o */
501 if (__mnt_is_readonly(mnt))
502 return -EROFS;
503 preempt_disable();
504 mnt_inc_writers(real_mount(mnt));
505 preempt_enable();
506 return 0;
507 }
508 EXPORT_SYMBOL_GPL(mnt_clone_write);
509
510 /**
511 * __mnt_want_write_file - get write access to a file's mount
512 * @file: the file who's mount on which to take a write
513 *
514 * This is like __mnt_want_write, but it takes a file and can
515 * do some optimisations if the file is open for write already
516 */
517 int __mnt_want_write_file(struct file *file)
518 {
519 struct inode *inode = file_inode(file);
520
521 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
522 return __mnt_want_write(file->f_path.mnt);
523 else
524 return mnt_clone_write(file->f_path.mnt);
525 }
526
527 /**
528 * mnt_want_write_file - get write access to a file's mount
529 * @file: the file who's mount on which to take a write
530 *
531 * This is like mnt_want_write, but it takes a file and can
532 * do some optimisations if the file is open for write already
533 */
534 int mnt_want_write_file(struct file *file)
535 {
536 int ret;
537
538 sb_start_write(file->f_path.mnt->mnt_sb);
539 ret = __mnt_want_write_file(file);
540 if (ret)
541 sb_end_write(file->f_path.mnt->mnt_sb);
542 return ret;
543 }
544 EXPORT_SYMBOL_GPL(mnt_want_write_file);
545
546 /**
547 * __mnt_drop_write - give up write access to a mount
548 * @mnt: the mount on which to give up write access
549 *
550 * Tells the low-level filesystem that we are done
551 * performing writes to it. Must be matched with
552 * __mnt_want_write() call above.
553 */
554 void __mnt_drop_write(struct vfsmount *mnt)
555 {
556 preempt_disable();
557 mnt_dec_writers(real_mount(mnt));
558 preempt_enable();
559 }
560
561 /**
562 * mnt_drop_write - give up write access to a mount
563 * @mnt: the mount on which to give up write access
564 *
565 * Tells the low-level filesystem that we are done performing writes to it and
566 * also allows filesystem to be frozen again. Must be matched with
567 * mnt_want_write() call above.
568 */
569 void mnt_drop_write(struct vfsmount *mnt)
570 {
571 __mnt_drop_write(mnt);
572 sb_end_write(mnt->mnt_sb);
573 }
574 EXPORT_SYMBOL_GPL(mnt_drop_write);
575
576 void __mnt_drop_write_file(struct file *file)
577 {
578 __mnt_drop_write(file->f_path.mnt);
579 }
580
581 void mnt_drop_write_file(struct file *file)
582 {
583 mnt_drop_write(file->f_path.mnt);
584 }
585 EXPORT_SYMBOL(mnt_drop_write_file);
586
587 static int mnt_make_readonly(struct mount *mnt)
588 {
589 int ret = 0;
590
591 br_write_lock(&vfsmount_lock);
592 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
593 /*
594 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
595 * should be visible before we do.
596 */
597 smp_mb();
598
599 /*
600 * With writers on hold, if this value is zero, then there are
601 * definitely no active writers (although held writers may subsequently
602 * increment the count, they'll have to wait, and decrement it after
603 * seeing MNT_READONLY).
604 *
605 * It is OK to have counter incremented on one CPU and decremented on
606 * another: the sum will add up correctly. The danger would be when we
607 * sum up each counter, if we read a counter before it is incremented,
608 * but then read another CPU's count which it has been subsequently
609 * decremented from -- we would see more decrements than we should.
610 * MNT_WRITE_HOLD protects against this scenario, because
611 * mnt_want_write first increments count, then smp_mb, then spins on
612 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
613 * we're counting up here.
614 */
615 if (mnt_get_writers(mnt) > 0)
616 ret = -EBUSY;
617 else
618 mnt->mnt.mnt_flags |= MNT_READONLY;
619 /*
620 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
621 * that become unheld will see MNT_READONLY.
622 */
623 smp_wmb();
624 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
625 br_write_unlock(&vfsmount_lock);
626 return ret;
627 }
628
629 static void __mnt_unmake_readonly(struct mount *mnt)
630 {
631 br_write_lock(&vfsmount_lock);
632 mnt->mnt.mnt_flags &= ~MNT_READONLY;
633 br_write_unlock(&vfsmount_lock);
634 }
635
636 int sb_prepare_remount_readonly(struct super_block *sb)
637 {
638 struct mount *mnt;
639 int err = 0;
640
641 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
642 if (atomic_long_read(&sb->s_remove_count))
643 return -EBUSY;
644
645 br_write_lock(&vfsmount_lock);
646 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
647 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
648 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
649 smp_mb();
650 if (mnt_get_writers(mnt) > 0) {
651 err = -EBUSY;
652 break;
653 }
654 }
655 }
656 if (!err && atomic_long_read(&sb->s_remove_count))
657 err = -EBUSY;
658
659 if (!err) {
660 sb->s_readonly_remount = 1;
661 smp_wmb();
662 }
663 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
664 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
665 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
666 }
667 br_write_unlock(&vfsmount_lock);
668
669 return err;
670 }
671
672 static void free_vfsmnt(struct mount *mnt)
673 {
674 kfree(mnt->mnt_devname);
675 mnt_free_id(mnt);
676 #ifdef CONFIG_SMP
677 free_percpu(mnt->mnt_pcp);
678 #endif
679 kmem_cache_free(mnt_cache, mnt);
680 }
681
682 /*
683 * find the first or last mount at @dentry on vfsmount @mnt depending on
684 * @dir. If @dir is set return the first mount else return the last mount.
685 * vfsmount_lock must be held for read or write.
686 */
687 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
688 int dir)
689 {
690 struct list_head *head = mount_hashtable + hash(mnt, dentry);
691 struct list_head *tmp = head;
692 struct mount *p, *found = NULL;
693
694 for (;;) {
695 tmp = dir ? tmp->next : tmp->prev;
696 p = NULL;
697 if (tmp == head)
698 break;
699 p = list_entry(tmp, struct mount, mnt_hash);
700 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
701 found = p;
702 break;
703 }
704 }
705 return found;
706 }
707
708 /*
709 * lookup_mnt - Return the first child mount mounted at path
710 *
711 * "First" means first mounted chronologically. If you create the
712 * following mounts:
713 *
714 * mount /dev/sda1 /mnt
715 * mount /dev/sda2 /mnt
716 * mount /dev/sda3 /mnt
717 *
718 * Then lookup_mnt() on the base /mnt dentry in the root mount will
719 * return successively the root dentry and vfsmount of /dev/sda1, then
720 * /dev/sda2, then /dev/sda3, then NULL.
721 *
722 * lookup_mnt takes a reference to the found vfsmount.
723 */
724 struct vfsmount *lookup_mnt(struct path *path)
725 {
726 struct mount *child_mnt;
727
728 br_read_lock(&vfsmount_lock);
729 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
730 if (child_mnt) {
731 mnt_add_count(child_mnt, 1);
732 br_read_unlock(&vfsmount_lock);
733 return &child_mnt->mnt;
734 } else {
735 br_read_unlock(&vfsmount_lock);
736 return NULL;
737 }
738 }
739
740 static struct mountpoint *new_mountpoint(struct dentry *dentry)
741 {
742 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
743 struct mountpoint *mp;
744
745 list_for_each_entry(mp, chain, m_hash) {
746 if (mp->m_dentry == dentry) {
747 /* might be worth a WARN_ON() */
748 if (d_unlinked(dentry))
749 return ERR_PTR(-ENOENT);
750 mp->m_count++;
751 return mp;
752 }
753 }
754
755 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
756 if (!mp)
757 return ERR_PTR(-ENOMEM);
758
759 spin_lock(&dentry->d_lock);
760 if (d_unlinked(dentry)) {
761 spin_unlock(&dentry->d_lock);
762 kfree(mp);
763 return ERR_PTR(-ENOENT);
764 }
765 dentry->d_flags |= DCACHE_MOUNTED;
766 spin_unlock(&dentry->d_lock);
767 mp->m_dentry = dentry;
768 mp->m_count = 1;
769 list_add(&mp->m_hash, chain);
770 return mp;
771 }
772
773 static void put_mountpoint(struct mountpoint *mp)
774 {
775 if (!--mp->m_count) {
776 struct dentry *dentry = mp->m_dentry;
777 spin_lock(&dentry->d_lock);
778 dentry->d_flags &= ~DCACHE_MOUNTED;
779 spin_unlock(&dentry->d_lock);
780 list_del(&mp->m_hash);
781 kfree(mp);
782 }
783 }
784
785 static inline int check_mnt(struct mount *mnt)
786 {
787 return mnt->mnt_ns == current->nsproxy->mnt_ns;
788 }
789
790 /*
791 * vfsmount lock must be held for write
792 */
793 static void touch_mnt_namespace(struct mnt_namespace *ns)
794 {
795 if (ns) {
796 ns->event = ++event;
797 wake_up_interruptible(&ns->poll);
798 }
799 }
800
801 /*
802 * vfsmount lock must be held for write
803 */
804 static void __touch_mnt_namespace(struct mnt_namespace *ns)
805 {
806 if (ns && ns->event != event) {
807 ns->event = event;
808 wake_up_interruptible(&ns->poll);
809 }
810 }
811
812 /*
813 * vfsmount lock must be held for write
814 */
815 static void detach_mnt(struct mount *mnt, struct path *old_path)
816 {
817 old_path->dentry = mnt->mnt_mountpoint;
818 old_path->mnt = &mnt->mnt_parent->mnt;
819 mnt->mnt_parent = mnt;
820 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
821 list_del_init(&mnt->mnt_child);
822 list_del_init(&mnt->mnt_hash);
823 put_mountpoint(mnt->mnt_mp);
824 mnt->mnt_mp = NULL;
825 }
826
827 /*
828 * vfsmount lock must be held for write
829 */
830 void mnt_set_mountpoint(struct mount *mnt,
831 struct mountpoint *mp,
832 struct mount *child_mnt)
833 {
834 mp->m_count++;
835 mnt_add_count(mnt, 1); /* essentially, that's mntget */
836 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
837 child_mnt->mnt_parent = mnt;
838 child_mnt->mnt_mp = mp;
839 }
840
841 /*
842 * vfsmount lock must be held for write
843 */
844 static void attach_mnt(struct mount *mnt,
845 struct mount *parent,
846 struct mountpoint *mp)
847 {
848 mnt_set_mountpoint(parent, mp, mnt);
849 list_add_tail(&mnt->mnt_hash, mount_hashtable +
850 hash(&parent->mnt, mp->m_dentry));
851 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
852 }
853
854 /*
855 * vfsmount lock must be held for write
856 */
857 static void commit_tree(struct mount *mnt)
858 {
859 struct mount *parent = mnt->mnt_parent;
860 struct mount *m;
861 LIST_HEAD(head);
862 struct mnt_namespace *n = parent->mnt_ns;
863
864 BUG_ON(parent == mnt);
865
866 list_add_tail(&head, &mnt->mnt_list);
867 list_for_each_entry(m, &head, mnt_list)
868 m->mnt_ns = n;
869
870 list_splice(&head, n->list.prev);
871
872 list_add_tail(&mnt->mnt_hash, mount_hashtable +
873 hash(&parent->mnt, mnt->mnt_mountpoint));
874 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
875 touch_mnt_namespace(n);
876 }
877
878 static struct mount *next_mnt(struct mount *p, struct mount *root)
879 {
880 struct list_head *next = p->mnt_mounts.next;
881 if (next == &p->mnt_mounts) {
882 while (1) {
883 if (p == root)
884 return NULL;
885 next = p->mnt_child.next;
886 if (next != &p->mnt_parent->mnt_mounts)
887 break;
888 p = p->mnt_parent;
889 }
890 }
891 return list_entry(next, struct mount, mnt_child);
892 }
893
894 static struct mount *skip_mnt_tree(struct mount *p)
895 {
896 struct list_head *prev = p->mnt_mounts.prev;
897 while (prev != &p->mnt_mounts) {
898 p = list_entry(prev, struct mount, mnt_child);
899 prev = p->mnt_mounts.prev;
900 }
901 return p;
902 }
903
904 struct vfsmount *
905 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
906 {
907 struct mount *mnt;
908 struct dentry *root;
909
910 if (!type)
911 return ERR_PTR(-ENODEV);
912
913 mnt = alloc_vfsmnt(name);
914 if (!mnt)
915 return ERR_PTR(-ENOMEM);
916
917 if (flags & MS_KERNMOUNT)
918 mnt->mnt.mnt_flags = MNT_INTERNAL;
919
920 root = mount_fs(type, flags, name, data);
921 if (IS_ERR(root)) {
922 free_vfsmnt(mnt);
923 return ERR_CAST(root);
924 }
925
926 mnt->mnt.mnt_root = root;
927 mnt->mnt.mnt_sb = root->d_sb;
928 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
929 mnt->mnt_parent = mnt;
930 br_write_lock(&vfsmount_lock);
931 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
932 br_write_unlock(&vfsmount_lock);
933 return &mnt->mnt;
934 }
935 EXPORT_SYMBOL_GPL(vfs_kern_mount);
936
937 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
938 int flag)
939 {
940 struct super_block *sb = old->mnt.mnt_sb;
941 struct mount *mnt;
942 int err;
943
944 mnt = alloc_vfsmnt(old->mnt_devname);
945 if (!mnt)
946 return ERR_PTR(-ENOMEM);
947
948 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
949 mnt->mnt_group_id = 0; /* not a peer of original */
950 else
951 mnt->mnt_group_id = old->mnt_group_id;
952
953 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
954 err = mnt_alloc_group_id(mnt);
955 if (err)
956 goto out_free;
957 }
958
959 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
960 /* Don't allow unprivileged users to change mount flags */
961 if (flag & CL_UNPRIVILEGED) {
962 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
963
964 if (mnt->mnt.mnt_flags & MNT_READONLY)
965 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
966
967 if (mnt->mnt.mnt_flags & MNT_NODEV)
968 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
969
970 if (mnt->mnt.mnt_flags & MNT_NOSUID)
971 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
972
973 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
974 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
975 }
976
977 atomic_inc(&sb->s_active);
978 mnt->mnt.mnt_sb = sb;
979 mnt->mnt.mnt_root = dget(root);
980 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
981 mnt->mnt_parent = mnt;
982 br_write_lock(&vfsmount_lock);
983 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
984 br_write_unlock(&vfsmount_lock);
985
986 if ((flag & CL_SLAVE) ||
987 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
988 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
989 mnt->mnt_master = old;
990 CLEAR_MNT_SHARED(mnt);
991 } else if (!(flag & CL_PRIVATE)) {
992 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
993 list_add(&mnt->mnt_share, &old->mnt_share);
994 if (IS_MNT_SLAVE(old))
995 list_add(&mnt->mnt_slave, &old->mnt_slave);
996 mnt->mnt_master = old->mnt_master;
997 }
998 if (flag & CL_MAKE_SHARED)
999 set_mnt_shared(mnt);
1000
1001 /* stick the duplicate mount on the same expiry list
1002 * as the original if that was on one */
1003 if (flag & CL_EXPIRE) {
1004 if (!list_empty(&old->mnt_expire))
1005 list_add(&mnt->mnt_expire, &old->mnt_expire);
1006 }
1007
1008 return mnt;
1009
1010 out_free:
1011 free_vfsmnt(mnt);
1012 return ERR_PTR(err);
1013 }
1014
1015 static inline void mntfree(struct mount *mnt)
1016 {
1017 struct vfsmount *m = &mnt->mnt;
1018 struct super_block *sb = m->mnt_sb;
1019
1020 /*
1021 * This probably indicates that somebody messed
1022 * up a mnt_want/drop_write() pair. If this
1023 * happens, the filesystem was probably unable
1024 * to make r/w->r/o transitions.
1025 */
1026 /*
1027 * The locking used to deal with mnt_count decrement provides barriers,
1028 * so mnt_get_writers() below is safe.
1029 */
1030 WARN_ON(mnt_get_writers(mnt));
1031 fsnotify_vfsmount_delete(m);
1032 dput(m->mnt_root);
1033 free_vfsmnt(mnt);
1034 deactivate_super(sb);
1035 }
1036
1037 static void mntput_no_expire(struct mount *mnt)
1038 {
1039 put_again:
1040 #ifdef CONFIG_SMP
1041 br_read_lock(&vfsmount_lock);
1042 if (likely(mnt->mnt_ns)) {
1043 /* shouldn't be the last one */
1044 mnt_add_count(mnt, -1);
1045 br_read_unlock(&vfsmount_lock);
1046 return;
1047 }
1048 br_read_unlock(&vfsmount_lock);
1049
1050 br_write_lock(&vfsmount_lock);
1051 mnt_add_count(mnt, -1);
1052 if (mnt_get_count(mnt)) {
1053 br_write_unlock(&vfsmount_lock);
1054 return;
1055 }
1056 #else
1057 mnt_add_count(mnt, -1);
1058 if (likely(mnt_get_count(mnt)))
1059 return;
1060 br_write_lock(&vfsmount_lock);
1061 #endif
1062 if (unlikely(mnt->mnt_pinned)) {
1063 mnt_add_count(mnt, mnt->mnt_pinned + 1);
1064 mnt->mnt_pinned = 0;
1065 br_write_unlock(&vfsmount_lock);
1066 acct_auto_close_mnt(&mnt->mnt);
1067 goto put_again;
1068 }
1069
1070 list_del(&mnt->mnt_instance);
1071 br_write_unlock(&vfsmount_lock);
1072 mntfree(mnt);
1073 }
1074
1075 void mntput(struct vfsmount *mnt)
1076 {
1077 if (mnt) {
1078 struct mount *m = real_mount(mnt);
1079 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1080 if (unlikely(m->mnt_expiry_mark))
1081 m->mnt_expiry_mark = 0;
1082 mntput_no_expire(m);
1083 }
1084 }
1085 EXPORT_SYMBOL(mntput);
1086
1087 struct vfsmount *mntget(struct vfsmount *mnt)
1088 {
1089 if (mnt)
1090 mnt_add_count(real_mount(mnt), 1);
1091 return mnt;
1092 }
1093 EXPORT_SYMBOL(mntget);
1094
1095 void mnt_pin(struct vfsmount *mnt)
1096 {
1097 br_write_lock(&vfsmount_lock);
1098 real_mount(mnt)->mnt_pinned++;
1099 br_write_unlock(&vfsmount_lock);
1100 }
1101 EXPORT_SYMBOL(mnt_pin);
1102
1103 void mnt_unpin(struct vfsmount *m)
1104 {
1105 struct mount *mnt = real_mount(m);
1106 br_write_lock(&vfsmount_lock);
1107 if (mnt->mnt_pinned) {
1108 mnt_add_count(mnt, 1);
1109 mnt->mnt_pinned--;
1110 }
1111 br_write_unlock(&vfsmount_lock);
1112 }
1113 EXPORT_SYMBOL(mnt_unpin);
1114
1115 static inline void mangle(struct seq_file *m, const char *s)
1116 {
1117 seq_escape(m, s, " \t\n\\");
1118 }
1119
1120 /*
1121 * Simple .show_options callback for filesystems which don't want to
1122 * implement more complex mount option showing.
1123 *
1124 * See also save_mount_options().
1125 */
1126 int generic_show_options(struct seq_file *m, struct dentry *root)
1127 {
1128 const char *options;
1129
1130 rcu_read_lock();
1131 options = rcu_dereference(root->d_sb->s_options);
1132
1133 if (options != NULL && options[0]) {
1134 seq_putc(m, ',');
1135 mangle(m, options);
1136 }
1137 rcu_read_unlock();
1138
1139 return 0;
1140 }
1141 EXPORT_SYMBOL(generic_show_options);
1142
1143 /*
1144 * If filesystem uses generic_show_options(), this function should be
1145 * called from the fill_super() callback.
1146 *
1147 * The .remount_fs callback usually needs to be handled in a special
1148 * way, to make sure, that previous options are not overwritten if the
1149 * remount fails.
1150 *
1151 * Also note, that if the filesystem's .remount_fs function doesn't
1152 * reset all options to their default value, but changes only newly
1153 * given options, then the displayed options will not reflect reality
1154 * any more.
1155 */
1156 void save_mount_options(struct super_block *sb, char *options)
1157 {
1158 BUG_ON(sb->s_options);
1159 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1160 }
1161 EXPORT_SYMBOL(save_mount_options);
1162
1163 void replace_mount_options(struct super_block *sb, char *options)
1164 {
1165 char *old = sb->s_options;
1166 rcu_assign_pointer(sb->s_options, options);
1167 if (old) {
1168 synchronize_rcu();
1169 kfree(old);
1170 }
1171 }
1172 EXPORT_SYMBOL(replace_mount_options);
1173
1174 #ifdef CONFIG_PROC_FS
1175 /* iterator; we want it to have access to namespace_sem, thus here... */
1176 static void *m_start(struct seq_file *m, loff_t *pos)
1177 {
1178 struct proc_mounts *p = proc_mounts(m);
1179
1180 down_read(&namespace_sem);
1181 return seq_list_start(&p->ns->list, *pos);
1182 }
1183
1184 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1185 {
1186 struct proc_mounts *p = proc_mounts(m);
1187
1188 return seq_list_next(v, &p->ns->list, pos);
1189 }
1190
1191 static void m_stop(struct seq_file *m, void *v)
1192 {
1193 up_read(&namespace_sem);
1194 }
1195
1196 static int m_show(struct seq_file *m, void *v)
1197 {
1198 struct proc_mounts *p = proc_mounts(m);
1199 struct mount *r = list_entry(v, struct mount, mnt_list);
1200 return p->show(m, &r->mnt);
1201 }
1202
1203 const struct seq_operations mounts_op = {
1204 .start = m_start,
1205 .next = m_next,
1206 .stop = m_stop,
1207 .show = m_show,
1208 };
1209 #endif /* CONFIG_PROC_FS */
1210
1211 /**
1212 * may_umount_tree - check if a mount tree is busy
1213 * @mnt: root of mount tree
1214 *
1215 * This is called to check if a tree of mounts has any
1216 * open files, pwds, chroots or sub mounts that are
1217 * busy.
1218 */
1219 int may_umount_tree(struct vfsmount *m)
1220 {
1221 struct mount *mnt = real_mount(m);
1222 int actual_refs = 0;
1223 int minimum_refs = 0;
1224 struct mount *p;
1225 BUG_ON(!m);
1226
1227 /* write lock needed for mnt_get_count */
1228 br_write_lock(&vfsmount_lock);
1229 for (p = mnt; p; p = next_mnt(p, mnt)) {
1230 actual_refs += mnt_get_count(p);
1231 minimum_refs += 2;
1232 }
1233 br_write_unlock(&vfsmount_lock);
1234
1235 if (actual_refs > minimum_refs)
1236 return 0;
1237
1238 return 1;
1239 }
1240
1241 EXPORT_SYMBOL(may_umount_tree);
1242
1243 /**
1244 * may_umount - check if a mount point is busy
1245 * @mnt: root of mount
1246 *
1247 * This is called to check if a mount point has any
1248 * open files, pwds, chroots or sub mounts. If the
1249 * mount has sub mounts this will return busy
1250 * regardless of whether the sub mounts are busy.
1251 *
1252 * Doesn't take quota and stuff into account. IOW, in some cases it will
1253 * give false negatives. The main reason why it's here is that we need
1254 * a non-destructive way to look for easily umountable filesystems.
1255 */
1256 int may_umount(struct vfsmount *mnt)
1257 {
1258 int ret = 1;
1259 down_read(&namespace_sem);
1260 br_write_lock(&vfsmount_lock);
1261 if (propagate_mount_busy(real_mount(mnt), 2))
1262 ret = 0;
1263 br_write_unlock(&vfsmount_lock);
1264 up_read(&namespace_sem);
1265 return ret;
1266 }
1267
1268 EXPORT_SYMBOL(may_umount);
1269
1270 static LIST_HEAD(unmounted); /* protected by namespace_sem */
1271
1272 static void namespace_unlock(void)
1273 {
1274 struct mount *mnt;
1275 LIST_HEAD(head);
1276
1277 if (likely(list_empty(&unmounted))) {
1278 up_write(&namespace_sem);
1279 return;
1280 }
1281
1282 list_splice_init(&unmounted, &head);
1283 up_write(&namespace_sem);
1284
1285 while (!list_empty(&head)) {
1286 mnt = list_first_entry(&head, struct mount, mnt_hash);
1287 list_del_init(&mnt->mnt_hash);
1288 if (mnt_has_parent(mnt)) {
1289 struct dentry *dentry;
1290 struct mount *m;
1291
1292 br_write_lock(&vfsmount_lock);
1293 dentry = mnt->mnt_mountpoint;
1294 m = mnt->mnt_parent;
1295 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1296 mnt->mnt_parent = mnt;
1297 m->mnt_ghosts--;
1298 br_write_unlock(&vfsmount_lock);
1299 dput(dentry);
1300 mntput(&m->mnt);
1301 }
1302 mntput(&mnt->mnt);
1303 }
1304 }
1305
1306 static inline void namespace_lock(void)
1307 {
1308 down_write(&namespace_sem);
1309 }
1310
1311 /*
1312 * vfsmount lock must be held for write
1313 * namespace_sem must be held for write
1314 */
1315 void umount_tree(struct mount *mnt, int propagate)
1316 {
1317 LIST_HEAD(tmp_list);
1318 struct mount *p;
1319
1320 for (p = mnt; p; p = next_mnt(p, mnt))
1321 list_move(&p->mnt_hash, &tmp_list);
1322
1323 if (propagate)
1324 propagate_umount(&tmp_list);
1325
1326 list_for_each_entry(p, &tmp_list, mnt_hash) {
1327 list_del_init(&p->mnt_expire);
1328 list_del_init(&p->mnt_list);
1329 __touch_mnt_namespace(p->mnt_ns);
1330 p->mnt_ns = NULL;
1331 list_del_init(&p->mnt_child);
1332 if (mnt_has_parent(p)) {
1333 p->mnt_parent->mnt_ghosts++;
1334 put_mountpoint(p->mnt_mp);
1335 p->mnt_mp = NULL;
1336 }
1337 change_mnt_propagation(p, MS_PRIVATE);
1338 }
1339 list_splice(&tmp_list, &unmounted);
1340 }
1341
1342 static void shrink_submounts(struct mount *mnt);
1343
1344 static int do_umount(struct mount *mnt, int flags)
1345 {
1346 struct super_block *sb = mnt->mnt.mnt_sb;
1347 int retval;
1348
1349 retval = security_sb_umount(&mnt->mnt, flags);
1350 if (retval)
1351 return retval;
1352
1353 /*
1354 * Allow userspace to request a mountpoint be expired rather than
1355 * unmounting unconditionally. Unmount only happens if:
1356 * (1) the mark is already set (the mark is cleared by mntput())
1357 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1358 */
1359 if (flags & MNT_EXPIRE) {
1360 if (&mnt->mnt == current->fs->root.mnt ||
1361 flags & (MNT_FORCE | MNT_DETACH))
1362 return -EINVAL;
1363
1364 /*
1365 * probably don't strictly need the lock here if we examined
1366 * all race cases, but it's a slowpath.
1367 */
1368 br_write_lock(&vfsmount_lock);
1369 if (mnt_get_count(mnt) != 2) {
1370 br_write_unlock(&vfsmount_lock);
1371 return -EBUSY;
1372 }
1373 br_write_unlock(&vfsmount_lock);
1374
1375 if (!xchg(&mnt->mnt_expiry_mark, 1))
1376 return -EAGAIN;
1377 }
1378
1379 /*
1380 * If we may have to abort operations to get out of this
1381 * mount, and they will themselves hold resources we must
1382 * allow the fs to do things. In the Unix tradition of
1383 * 'Gee thats tricky lets do it in userspace' the umount_begin
1384 * might fail to complete on the first run through as other tasks
1385 * must return, and the like. Thats for the mount program to worry
1386 * about for the moment.
1387 */
1388
1389 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1390 sb->s_op->umount_begin(sb);
1391 }
1392
1393 /*
1394 * No sense to grab the lock for this test, but test itself looks
1395 * somewhat bogus. Suggestions for better replacement?
1396 * Ho-hum... In principle, we might treat that as umount + switch
1397 * to rootfs. GC would eventually take care of the old vfsmount.
1398 * Actually it makes sense, especially if rootfs would contain a
1399 * /reboot - static binary that would close all descriptors and
1400 * call reboot(9). Then init(8) could umount root and exec /reboot.
1401 */
1402 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1403 /*
1404 * Special case for "unmounting" root ...
1405 * we just try to remount it readonly.
1406 */
1407 down_write(&sb->s_umount);
1408 if (!(sb->s_flags & MS_RDONLY))
1409 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1410 up_write(&sb->s_umount);
1411 return retval;
1412 }
1413
1414 namespace_lock();
1415 br_write_lock(&vfsmount_lock);
1416 event++;
1417
1418 if (!(flags & MNT_DETACH))
1419 shrink_submounts(mnt);
1420
1421 retval = -EBUSY;
1422 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1423 if (!list_empty(&mnt->mnt_list))
1424 umount_tree(mnt, 1);
1425 retval = 0;
1426 }
1427 br_write_unlock(&vfsmount_lock);
1428 namespace_unlock();
1429 return retval;
1430 }
1431
1432 /*
1433 * Is the caller allowed to modify his namespace?
1434 */
1435 static inline bool may_mount(void)
1436 {
1437 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1438 }
1439
1440 /*
1441 * Now umount can handle mount points as well as block devices.
1442 * This is important for filesystems which use unnamed block devices.
1443 *
1444 * We now support a flag for forced unmount like the other 'big iron'
1445 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1446 */
1447
1448 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1449 {
1450 struct path path;
1451 struct mount *mnt;
1452 int retval;
1453 int lookup_flags = 0;
1454 #ifdef UMOUNT_LOG
1455 int total_value =0;
1456 #endif
1457 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1458 return -EINVAL;
1459
1460 if (!may_mount())
1461 return -EPERM;
1462
1463 if (!(flags & UMOUNT_NOFOLLOW))
1464 lookup_flags |= LOOKUP_FOLLOW;
1465
1466 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1467 if (retval)
1468 goto out;
1469 mnt = real_mount(path.mnt);
1470 retval = -EINVAL;
1471 if (path.dentry != path.mnt->mnt_root)
1472 goto dput_and_out;
1473 if (!check_mnt(mnt))
1474 goto dput_and_out;
1475
1476 retval = do_umount(mnt, flags);
1477 #ifdef UMOUNT_LOG
1478 {
1479 printk("Ahsin do_umount retval=%d \n",retval);
1480 //do_umount success: 0, do_umount busy: -16
1481 //if do_umount fail, need to dump the link list here
1482
1483 if(retval)
1484 printk("Ahsin do_umount fail; mnt_get_count=%d mnt->mnt_devname=%s\n",mnt_get_count(mnt),mnt->mnt_devname);
1485 else
1486 printk("Ahsin do_umount success; mnt_get_count=%d mnt->mnt_devname=%s\n",mnt_get_count(mnt),mnt->mnt_devname);
1487
1488 // print linked list
1489 spin_lock(&mnt_id_lock);
1490 ref_current = ref_head;
1491 while(ref_current != NULL)
1492 {
1493 total_value = total_value + ref_current->count;
1494
1495 if (ref_current->count)
1496 printk("Ahsin PID= %d, Name = %s, Count= %d \n", ref_current->pid, ref_current->name, ref_current->count);
1497 ref_current = ref_current->next;
1498 }
1499 spin_unlock(&mnt_id_lock);
1500 printk("Ahsin total_value=%d \n",total_value);
1501 }
1502 #endif
1503 dput_and_out:
1504 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1505 dput(path.dentry);
1506 mntput_no_expire(mnt);
1507 out:
1508 return retval;
1509 }
1510
1511 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1512
1513 /*
1514 * The 2.0 compatible umount. No flags.
1515 */
1516 SYSCALL_DEFINE1(oldumount, char __user *, name)
1517 {
1518 return sys_umount(name, 0);
1519 }
1520
1521 #endif
1522
1523 static bool mnt_ns_loop(struct path *path)
1524 {
1525 /* Could bind mounting the mount namespace inode cause a
1526 * mount namespace loop?
1527 */
1528 struct inode *inode = path->dentry->d_inode;
1529 struct proc_ns *ei;
1530 struct mnt_namespace *mnt_ns;
1531
1532 if (!proc_ns_inode(inode))
1533 return false;
1534
1535 ei = get_proc_ns(inode);
1536 if (ei->ns_ops != &mntns_operations)
1537 return false;
1538
1539 mnt_ns = ei->ns;
1540 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1541 }
1542
1543 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1544 int flag)
1545 {
1546 struct mount *res, *p, *q, *r, *parent;
1547
1548 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1549 return ERR_PTR(-EINVAL);
1550
1551 res = q = clone_mnt(mnt, dentry, flag);
1552 if (IS_ERR(q))
1553 return q;
1554
1555 q->mnt_mountpoint = mnt->mnt_mountpoint;
1556
1557 p = mnt;
1558 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1559 struct mount *s;
1560 if (!is_subdir(r->mnt_mountpoint, dentry))
1561 continue;
1562
1563 for (s = r; s; s = next_mnt(s, r)) {
1564 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1565 s = skip_mnt_tree(s);
1566 continue;
1567 }
1568 while (p != s->mnt_parent) {
1569 p = p->mnt_parent;
1570 q = q->mnt_parent;
1571 }
1572 p = s;
1573 parent = q;
1574 q = clone_mnt(p, p->mnt.mnt_root, flag);
1575 if (IS_ERR(q))
1576 goto out;
1577 br_write_lock(&vfsmount_lock);
1578 list_add_tail(&q->mnt_list, &res->mnt_list);
1579 attach_mnt(q, parent, p->mnt_mp);
1580 br_write_unlock(&vfsmount_lock);
1581 }
1582 }
1583 return res;
1584 out:
1585 if (res) {
1586 br_write_lock(&vfsmount_lock);
1587 umount_tree(res, 0);
1588 br_write_unlock(&vfsmount_lock);
1589 }
1590 return q;
1591 }
1592
1593 /* Caller should check returned pointer for errors */
1594
1595 struct vfsmount *collect_mounts(struct path *path)
1596 {
1597 struct mount *tree;
1598 namespace_lock();
1599 tree = copy_tree(real_mount(path->mnt), path->dentry,
1600 CL_COPY_ALL | CL_PRIVATE);
1601 namespace_unlock();
1602 if (IS_ERR(tree))
1603 return ERR_CAST(tree);
1604 return &tree->mnt;
1605 }
1606
1607 void drop_collected_mounts(struct vfsmount *mnt)
1608 {
1609 namespace_lock();
1610 br_write_lock(&vfsmount_lock);
1611 umount_tree(real_mount(mnt), 0);
1612 br_write_unlock(&vfsmount_lock);
1613 namespace_unlock();
1614 }
1615
1616 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1617 struct vfsmount *root)
1618 {
1619 struct mount *mnt;
1620 int res = f(root, arg);
1621 if (res)
1622 return res;
1623 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1624 res = f(&mnt->mnt, arg);
1625 if (res)
1626 return res;
1627 }
1628 return 0;
1629 }
1630
1631 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1632 {
1633 struct mount *p;
1634
1635 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1636 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1637 mnt_release_group_id(p);
1638 }
1639 }
1640
1641 static int invent_group_ids(struct mount *mnt, bool recurse)
1642 {
1643 struct mount *p;
1644
1645 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1646 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1647 int err = mnt_alloc_group_id(p);
1648 if (err) {
1649 cleanup_group_ids(mnt, p);
1650 return err;
1651 }
1652 }
1653 }
1654
1655 return 0;
1656 }
1657
1658 /*
1659 * @source_mnt : mount tree to be attached
1660 * @nd : place the mount tree @source_mnt is attached
1661 * @parent_nd : if non-null, detach the source_mnt from its parent and
1662 * store the parent mount and mountpoint dentry.
1663 * (done when source_mnt is moved)
1664 *
1665 * NOTE: in the table below explains the semantics when a source mount
1666 * of a given type is attached to a destination mount of a given type.
1667 * ---------------------------------------------------------------------------
1668 * | BIND MOUNT OPERATION |
1669 * |**************************************************************************
1670 * | source-->| shared | private | slave | unbindable |
1671 * | dest | | | | |
1672 * | | | | | | |
1673 * | v | | | | |
1674 * |**************************************************************************
1675 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1676 * | | | | | |
1677 * |non-shared| shared (+) | private | slave (*) | invalid |
1678 * ***************************************************************************
1679 * A bind operation clones the source mount and mounts the clone on the
1680 * destination mount.
1681 *
1682 * (++) the cloned mount is propagated to all the mounts in the propagation
1683 * tree of the destination mount and the cloned mount is added to
1684 * the peer group of the source mount.
1685 * (+) the cloned mount is created under the destination mount and is marked
1686 * as shared. The cloned mount is added to the peer group of the source
1687 * mount.
1688 * (+++) the mount is propagated to all the mounts in the propagation tree
1689 * of the destination mount and the cloned mount is made slave
1690 * of the same master as that of the source mount. The cloned mount
1691 * is marked as 'shared and slave'.
1692 * (*) the cloned mount is made a slave of the same master as that of the
1693 * source mount.
1694 *
1695 * ---------------------------------------------------------------------------
1696 * | MOVE MOUNT OPERATION |
1697 * |**************************************************************************
1698 * | source-->| shared | private | slave | unbindable |
1699 * | dest | | | | |
1700 * | | | | | | |
1701 * | v | | | | |
1702 * |**************************************************************************
1703 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1704 * | | | | | |
1705 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1706 * ***************************************************************************
1707 *
1708 * (+) the mount is moved to the destination. And is then propagated to
1709 * all the mounts in the propagation tree of the destination mount.
1710 * (+*) the mount is moved to the destination.
1711 * (+++) the mount is moved to the destination and is then propagated to
1712 * all the mounts belonging to the destination mount's propagation tree.
1713 * the mount is marked as 'shared and slave'.
1714 * (*) the mount continues to be a slave at the new location.
1715 *
1716 * if the source mount is a tree, the operations explained above is
1717 * applied to each mount in the tree.
1718 * Must be called without spinlocks held, since this function can sleep
1719 * in allocations.
1720 */
1721 static int attach_recursive_mnt(struct mount *source_mnt,
1722 struct mount *dest_mnt,
1723 struct mountpoint *dest_mp,
1724 struct path *parent_path)
1725 {
1726 LIST_HEAD(tree_list);
1727 struct mount *child, *p;
1728 int err;
1729
1730 if (IS_MNT_SHARED(dest_mnt)) {
1731 err = invent_group_ids(source_mnt, true);
1732 if (err)
1733 goto out;
1734 }
1735 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1736 if (err)
1737 goto out_cleanup_ids;
1738
1739 br_write_lock(&vfsmount_lock);
1740
1741 if (IS_MNT_SHARED(dest_mnt)) {
1742 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1743 set_mnt_shared(p);
1744 }
1745 if (parent_path) {
1746 detach_mnt(source_mnt, parent_path);
1747 attach_mnt(source_mnt, dest_mnt, dest_mp);
1748 touch_mnt_namespace(source_mnt->mnt_ns);
1749 } else {
1750 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1751 commit_tree(source_mnt);
1752 }
1753
1754 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1755 list_del_init(&child->mnt_hash);
1756 commit_tree(child);
1757 }
1758 br_write_unlock(&vfsmount_lock);
1759
1760 return 0;
1761
1762 out_cleanup_ids:
1763 if (IS_MNT_SHARED(dest_mnt))
1764 cleanup_group_ids(source_mnt, NULL);
1765 out:
1766 return err;
1767 }
1768
1769 static struct mountpoint *lock_mount(struct path *path)
1770 {
1771 struct vfsmount *mnt;
1772 struct dentry *dentry = path->dentry;
1773 retry:
1774 mutex_lock(&dentry->d_inode->i_mutex);
1775 if (unlikely(cant_mount(dentry))) {
1776 mutex_unlock(&dentry->d_inode->i_mutex);
1777 return ERR_PTR(-ENOENT);
1778 }
1779 namespace_lock();
1780 mnt = lookup_mnt(path);
1781 if (likely(!mnt)) {
1782 struct mountpoint *mp = new_mountpoint(dentry);
1783 if (IS_ERR(mp)) {
1784 namespace_unlock();
1785 mutex_unlock(&dentry->d_inode->i_mutex);
1786 return mp;
1787 }
1788 return mp;
1789 }
1790 namespace_unlock();
1791 mutex_unlock(&path->dentry->d_inode->i_mutex);
1792 path_put(path);
1793 path->mnt = mnt;
1794 dentry = path->dentry = dget(mnt->mnt_root);
1795 goto retry;
1796 }
1797
1798 static void unlock_mount(struct mountpoint *where)
1799 {
1800 struct dentry *dentry = where->m_dentry;
1801 put_mountpoint(where);
1802 namespace_unlock();
1803 mutex_unlock(&dentry->d_inode->i_mutex);
1804 }
1805
1806 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1807 {
1808 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1809 return -EINVAL;
1810
1811 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1812 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1813 return -ENOTDIR;
1814
1815 return attach_recursive_mnt(mnt, p, mp, NULL);
1816 }
1817
1818 /*
1819 * Sanity check the flags to change_mnt_propagation.
1820 */
1821
1822 static int flags_to_propagation_type(int flags)
1823 {
1824 int type = flags & ~(MS_REC | MS_SILENT);
1825
1826 /* Fail if any non-propagation flags are set */
1827 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1828 return 0;
1829 /* Only one propagation flag should be set */
1830 if (!is_power_of_2(type))
1831 return 0;
1832 return type;
1833 }
1834
1835 /*
1836 * recursively change the type of the mountpoint.
1837 */
1838 static int do_change_type(struct path *path, int flag)
1839 {
1840 struct mount *m;
1841 struct mount *mnt = real_mount(path->mnt);
1842 int recurse = flag & MS_REC;
1843 int type;
1844 int err = 0;
1845
1846 if (path->dentry != path->mnt->mnt_root)
1847 return -EINVAL;
1848
1849 type = flags_to_propagation_type(flag);
1850 if (!type)
1851 return -EINVAL;
1852
1853 namespace_lock();
1854 if (type == MS_SHARED) {
1855 err = invent_group_ids(mnt, recurse);
1856 if (err)
1857 goto out_unlock;
1858 }
1859
1860 br_write_lock(&vfsmount_lock);
1861 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1862 change_mnt_propagation(m, type);
1863 br_write_unlock(&vfsmount_lock);
1864
1865 out_unlock:
1866 namespace_unlock();
1867 return err;
1868 }
1869
1870 /*
1871 * do loopback mount.
1872 */
1873 static int do_loopback(struct path *path, const char *old_name,
1874 int recurse)
1875 {
1876 struct path old_path;
1877 struct mount *mnt = NULL, *old, *parent;
1878 struct mountpoint *mp;
1879 int err;
1880 if (!old_name || !*old_name)
1881 return -EINVAL;
1882 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1883 if (err)
1884 return err;
1885
1886 err = -EINVAL;
1887 if (mnt_ns_loop(&old_path))
1888 goto out;
1889
1890 mp = lock_mount(path);
1891 err = PTR_ERR(mp);
1892 if (IS_ERR(mp))
1893 goto out;
1894
1895 old = real_mount(old_path.mnt);
1896 parent = real_mount(path->mnt);
1897
1898 err = -EINVAL;
1899 if (IS_MNT_UNBINDABLE(old))
1900 goto out2;
1901
1902 if (!check_mnt(parent) || !check_mnt(old))
1903 goto out2;
1904
1905 if (recurse)
1906 mnt = copy_tree(old, old_path.dentry, 0);
1907 else
1908 mnt = clone_mnt(old, old_path.dentry, 0);
1909
1910 if (IS_ERR(mnt)) {
1911 err = PTR_ERR(mnt);
1912 goto out2;
1913 }
1914
1915 err = graft_tree(mnt, parent, mp);
1916 if (err) {
1917 br_write_lock(&vfsmount_lock);
1918 umount_tree(mnt, 0);
1919 br_write_unlock(&vfsmount_lock);
1920 }
1921 out2:
1922 unlock_mount(mp);
1923 out:
1924 path_put(&old_path);
1925 return err;
1926 }
1927
1928 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1929 {
1930 int error = 0;
1931 int readonly_request = 0;
1932
1933 if (ms_flags & MS_RDONLY)
1934 readonly_request = 1;
1935 if (readonly_request == __mnt_is_readonly(mnt))
1936 return 0;
1937
1938 if (readonly_request)
1939 error = mnt_make_readonly(real_mount(mnt));
1940 else
1941 __mnt_unmake_readonly(real_mount(mnt));
1942 return error;
1943 }
1944
1945 /*
1946 * change filesystem flags. dir should be a physical root of filesystem.
1947 * If you've mounted a non-root directory somewhere and want to do remount
1948 * on it - tough luck.
1949 */
1950 static int do_remount(struct path *path, int flags, int mnt_flags,
1951 void *data)
1952 {
1953 int err;
1954 struct super_block *sb = path->mnt->mnt_sb;
1955 struct mount *mnt = real_mount(path->mnt);
1956
1957 if (!check_mnt(mnt))
1958 return -EINVAL;
1959
1960 if (path->dentry != path->mnt->mnt_root)
1961 return -EINVAL;
1962
1963 /* Don't allow changing of locked mnt flags.
1964 *
1965 * No locks need to be held here while testing the various
1966 * MNT_LOCK flags because those flags can never be cleared
1967 * once they are set.
1968 */
1969 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
1970 !(mnt_flags & MNT_READONLY)) {
1971 return -EPERM;
1972 }
1973 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
1974 !(mnt_flags & MNT_NODEV)) {
1975 return -EPERM;
1976 }
1977 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
1978 !(mnt_flags & MNT_NOSUID)) {
1979 return -EPERM;
1980 }
1981 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
1982 !(mnt_flags & MNT_NOEXEC)) {
1983 return -EPERM;
1984 }
1985 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
1986 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
1987 return -EPERM;
1988 }
1989
1990 err = security_sb_remount(sb, data);
1991 if (err)
1992 return err;
1993
1994 down_write(&sb->s_umount);
1995 if (flags & MS_BIND)
1996 err = change_mount_flags(path->mnt, flags);
1997 else if (!capable(CAP_SYS_ADMIN))
1998 err = -EPERM;
1999 else
2000 err = do_remount_sb(sb, flags, data, 0);
2001 if (!err) {
2002 br_write_lock(&vfsmount_lock);
2003 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2004 mnt->mnt.mnt_flags = mnt_flags;
2005 br_write_unlock(&vfsmount_lock);
2006 }
2007 up_write(&sb->s_umount);
2008 if (!err) {
2009 br_write_lock(&vfsmount_lock);
2010 touch_mnt_namespace(mnt->mnt_ns);
2011 br_write_unlock(&vfsmount_lock);
2012 }
2013 return err;
2014 }
2015
2016 static inline int tree_contains_unbindable(struct mount *mnt)
2017 {
2018 struct mount *p;
2019 for (p = mnt; p; p = next_mnt(p, mnt)) {
2020 if (IS_MNT_UNBINDABLE(p))
2021 return 1;
2022 }
2023 return 0;
2024 }
2025
2026 static int do_move_mount(struct path *path, const char *old_name)
2027 {
2028 struct path old_path, parent_path;
2029 struct mount *p;
2030 struct mount *old;
2031 struct mountpoint *mp;
2032 int err;
2033 if (!old_name || !*old_name)
2034 return -EINVAL;
2035 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2036 if (err)
2037 return err;
2038
2039 mp = lock_mount(path);
2040 err = PTR_ERR(mp);
2041 if (IS_ERR(mp))
2042 goto out;
2043
2044 old = real_mount(old_path.mnt);
2045 p = real_mount(path->mnt);
2046
2047 err = -EINVAL;
2048 if (!check_mnt(p) || !check_mnt(old))
2049 goto out1;
2050
2051 err = -EINVAL;
2052 if (old_path.dentry != old_path.mnt->mnt_root)
2053 goto out1;
2054
2055 if (!mnt_has_parent(old))
2056 goto out1;
2057
2058 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2059 S_ISDIR(old_path.dentry->d_inode->i_mode))
2060 goto out1;
2061 /*
2062 * Don't move a mount residing in a shared parent.
2063 */
2064 if (IS_MNT_SHARED(old->mnt_parent))
2065 goto out1;
2066 /*
2067 * Don't move a mount tree containing unbindable mounts to a destination
2068 * mount which is shared.
2069 */
2070 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2071 goto out1;
2072 err = -ELOOP;
2073 for (; mnt_has_parent(p); p = p->mnt_parent)
2074 if (p == old)
2075 goto out1;
2076
2077 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2078 if (err)
2079 goto out1;
2080
2081 /* if the mount is moved, it should no longer be expire
2082 * automatically */
2083 list_del_init(&old->mnt_expire);
2084 out1:
2085 unlock_mount(mp);
2086 out:
2087 if (!err)
2088 path_put(&parent_path);
2089 path_put(&old_path);
2090 return err;
2091 }
2092
2093 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2094 {
2095 int err;
2096 const char *subtype = strchr(fstype, '.');
2097 if (subtype) {
2098 subtype++;
2099 err = -EINVAL;
2100 if (!subtype[0])
2101 goto err;
2102 } else
2103 subtype = "";
2104
2105 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2106 err = -ENOMEM;
2107 if (!mnt->mnt_sb->s_subtype)
2108 goto err;
2109 return mnt;
2110
2111 err:
2112 mntput(mnt);
2113 return ERR_PTR(err);
2114 }
2115
2116 /*
2117 * add a mount into a namespace's mount tree
2118 */
2119 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2120 {
2121 struct mountpoint *mp;
2122 struct mount *parent;
2123 int err;
2124
2125 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
2126
2127 mp = lock_mount(path);
2128 if (IS_ERR(mp))
2129 return PTR_ERR(mp);
2130
2131 parent = real_mount(path->mnt);
2132 err = -EINVAL;
2133 if (unlikely(!check_mnt(parent))) {
2134 /* that's acceptable only for automounts done in private ns */
2135 if (!(mnt_flags & MNT_SHRINKABLE))
2136 goto unlock;
2137 /* ... and for those we'd better have mountpoint still alive */
2138 if (!parent->mnt_ns)
2139 goto unlock;
2140 }
2141
2142 /* Refuse the same filesystem on the same mount point */
2143 err = -EBUSY;
2144 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2145 path->mnt->mnt_root == path->dentry)
2146 goto unlock;
2147
2148 err = -EINVAL;
2149 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2150 goto unlock;
2151
2152 newmnt->mnt.mnt_flags = mnt_flags;
2153 err = graft_tree(newmnt, parent, mp);
2154
2155 unlock:
2156 unlock_mount(mp);
2157 return err;
2158 }
2159
2160 /*
2161 * create a new mount for userspace and request it to be added into the
2162 * namespace's tree
2163 */
2164 static int do_new_mount(struct path *path, const char *fstype, int flags,
2165 int mnt_flags, const char *name, void *data)
2166 {
2167 struct file_system_type *type;
2168 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2169 struct vfsmount *mnt;
2170 int err;
2171
2172 if (!fstype)
2173 return -EINVAL;
2174
2175 type = get_fs_type(fstype);
2176 if (!type)
2177 return -ENODEV;
2178
2179 if (user_ns != &init_user_ns) {
2180 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2181 put_filesystem(type);
2182 return -EPERM;
2183 }
2184 /* Only in special cases allow devices from mounts
2185 * created outside the initial user namespace.
2186 */
2187 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2188 flags |= MS_NODEV;
2189 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2190 }
2191 }
2192
2193 mnt = vfs_kern_mount(type, flags, name, data);
2194 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2195 !mnt->mnt_sb->s_subtype)
2196 mnt = fs_set_subtype(mnt, fstype);
2197
2198 put_filesystem(type);
2199 if (IS_ERR(mnt))
2200 return PTR_ERR(mnt);
2201
2202 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2203 if (err)
2204 mntput(mnt);
2205 return err;
2206 }
2207
2208 int finish_automount(struct vfsmount *m, struct path *path)
2209 {
2210 struct mount *mnt = real_mount(m);
2211 int err;
2212 /* The new mount record should have at least 2 refs to prevent it being
2213 * expired before we get a chance to add it
2214 */
2215 BUG_ON(mnt_get_count(mnt) < 2);
2216
2217 if (m->mnt_sb == path->mnt->mnt_sb &&
2218 m->mnt_root == path->dentry) {
2219 err = -ELOOP;
2220 goto fail;
2221 }
2222
2223 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2224 if (!err)
2225 return 0;
2226 fail:
2227 /* remove m from any expiration list it may be on */
2228 if (!list_empty(&mnt->mnt_expire)) {
2229 namespace_lock();
2230 br_write_lock(&vfsmount_lock);
2231 list_del_init(&mnt->mnt_expire);
2232 br_write_unlock(&vfsmount_lock);
2233 namespace_unlock();
2234 }
2235 mntput(m);
2236 mntput(m);
2237 return err;
2238 }
2239
2240 /**
2241 * mnt_set_expiry - Put a mount on an expiration list
2242 * @mnt: The mount to list.
2243 * @expiry_list: The list to add the mount to.
2244 */
2245 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2246 {
2247 namespace_lock();
2248 br_write_lock(&vfsmount_lock);
2249
2250 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2251
2252 br_write_unlock(&vfsmount_lock);
2253 namespace_unlock();
2254 }
2255 EXPORT_SYMBOL(mnt_set_expiry);
2256
2257 /*
2258 * process a list of expirable mountpoints with the intent of discarding any
2259 * mountpoints that aren't in use and haven't been touched since last we came
2260 * here
2261 */
2262 void mark_mounts_for_expiry(struct list_head *mounts)
2263 {
2264 struct mount *mnt, *next;
2265 LIST_HEAD(graveyard);
2266
2267 if (list_empty(mounts))
2268 return;
2269
2270 namespace_lock();
2271 br_write_lock(&vfsmount_lock);
2272
2273 /* extract from the expiration list every vfsmount that matches the
2274 * following criteria:
2275 * - only referenced by its parent vfsmount
2276 * - still marked for expiry (marked on the last call here; marks are
2277 * cleared by mntput())
2278 */
2279 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2280 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2281 propagate_mount_busy(mnt, 1))
2282 continue;
2283 list_move(&mnt->mnt_expire, &graveyard);
2284 }
2285 while (!list_empty(&graveyard)) {
2286 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2287 touch_mnt_namespace(mnt->mnt_ns);
2288 umount_tree(mnt, 1);
2289 }
2290 br_write_unlock(&vfsmount_lock);
2291 namespace_unlock();
2292 }
2293
2294 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2295
2296 /*
2297 * Ripoff of 'select_parent()'
2298 *
2299 * search the list of submounts for a given mountpoint, and move any
2300 * shrinkable submounts to the 'graveyard' list.
2301 */
2302 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2303 {
2304 struct mount *this_parent = parent;
2305 struct list_head *next;
2306 int found = 0;
2307
2308 repeat:
2309 next = this_parent->mnt_mounts.next;
2310 resume:
2311 while (next != &this_parent->mnt_mounts) {
2312 struct list_head *tmp = next;
2313 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2314
2315 next = tmp->next;
2316 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2317 continue;
2318 /*
2319 * Descend a level if the d_mounts list is non-empty.
2320 */
2321 if (!list_empty(&mnt->mnt_mounts)) {
2322 this_parent = mnt;
2323 goto repeat;
2324 }
2325
2326 if (!propagate_mount_busy(mnt, 1)) {
2327 list_move_tail(&mnt->mnt_expire, graveyard);
2328 found++;
2329 }
2330 }
2331 /*
2332 * All done at this level ... ascend and resume the search
2333 */
2334 if (this_parent != parent) {
2335 next = this_parent->mnt_child.next;
2336 this_parent = this_parent->mnt_parent;
2337 goto resume;
2338 }
2339 return found;
2340 }
2341
2342 /*
2343 * process a list of expirable mountpoints with the intent of discarding any
2344 * submounts of a specific parent mountpoint
2345 *
2346 * vfsmount_lock must be held for write
2347 */
2348 static void shrink_submounts(struct mount *mnt)
2349 {
2350 LIST_HEAD(graveyard);
2351 struct mount *m;
2352
2353 /* extract submounts of 'mountpoint' from the expiration list */
2354 while (select_submounts(mnt, &graveyard)) {
2355 while (!list_empty(&graveyard)) {
2356 m = list_first_entry(&graveyard, struct mount,
2357 mnt_expire);
2358 touch_mnt_namespace(m->mnt_ns);
2359 umount_tree(m, 1);
2360 }
2361 }
2362 }
2363
2364 /*
2365 * Some copy_from_user() implementations do not return the exact number of
2366 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2367 * Note that this function differs from copy_from_user() in that it will oops
2368 * on bad values of `to', rather than returning a short copy.
2369 */
2370 static long exact_copy_from_user(void *to, const void __user * from,
2371 unsigned long n)
2372 {
2373 char *t = to;
2374 const char __user *f = from;
2375 char c;
2376
2377 if (!access_ok(VERIFY_READ, from, n))
2378 return n;
2379
2380 while (n) {
2381 if (__get_user(c, f)) {
2382 memset(t, 0, n);
2383 break;
2384 }
2385 *t++ = c;
2386 f++;
2387 n--;
2388 }
2389 return n;
2390 }
2391
2392 int copy_mount_options(const void __user * data, unsigned long *where)
2393 {
2394 int i;
2395 unsigned long page;
2396 unsigned long size;
2397
2398 *where = 0;
2399 if (!data)
2400 return 0;
2401
2402 if (!(page = __get_free_page(GFP_KERNEL)))
2403 return -ENOMEM;
2404
2405 /* We only care that *some* data at the address the user
2406 * gave us is valid. Just in case, we'll zero
2407 * the remainder of the page.
2408 */
2409 /* copy_from_user cannot cross TASK_SIZE ! */
2410 size = TASK_SIZE - (unsigned long)data;
2411 if (size > PAGE_SIZE)
2412 size = PAGE_SIZE;
2413
2414 i = size - exact_copy_from_user((void *)page, data, size);
2415 if (!i) {
2416 free_page(page);
2417 return -EFAULT;
2418 }
2419 if (i != PAGE_SIZE)
2420 memset((char *)page + i, 0, PAGE_SIZE - i);
2421 *where = page;
2422 return 0;
2423 }
2424
2425 int copy_mount_string(const void __user *data, char **where)
2426 {
2427 char *tmp;
2428
2429 if (!data) {
2430 *where = NULL;
2431 return 0;
2432 }
2433
2434 tmp = strndup_user(data, PAGE_SIZE);
2435 if (IS_ERR(tmp))
2436 return PTR_ERR(tmp);
2437
2438 *where = tmp;
2439 return 0;
2440 }
2441
2442 /*
2443 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2444 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2445 *
2446 * data is a (void *) that can point to any structure up to
2447 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2448 * information (or be NULL).
2449 *
2450 * Pre-0.97 versions of mount() didn't have a flags word.
2451 * When the flags word was introduced its top half was required
2452 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2453 * Therefore, if this magic number is present, it carries no information
2454 * and must be discarded.
2455 */
2456 long do_mount(const char *dev_name, const char *dir_name,
2457 const char *type_page, unsigned long flags, void *data_page)
2458 {
2459 struct path path;
2460 int retval = 0;
2461 int mnt_flags = 0;
2462
2463 /* Discard magic */
2464 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2465 flags &= ~MS_MGC_MSK;
2466
2467 /* Basic sanity checks */
2468
2469 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2470 return -EINVAL;
2471
2472 if (data_page)
2473 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2474
2475 /* ... and get the mountpoint */
2476 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2477 if (retval)
2478 return retval;
2479
2480 retval = security_sb_mount(dev_name, &path,
2481 type_page, flags, data_page);
2482 if (!retval && !may_mount())
2483 retval = -EPERM;
2484 if (retval)
2485 goto dput_out;
2486
2487 /* Default to relatime unless overriden */
2488 if (!(flags & MS_NOATIME))
2489 mnt_flags |= MNT_RELATIME;
2490
2491 /* Separate the per-mountpoint flags */
2492 if (flags & MS_NOSUID)
2493 mnt_flags |= MNT_NOSUID;
2494 if (flags & MS_NODEV)
2495 mnt_flags |= MNT_NODEV;
2496 if (flags & MS_NOEXEC)
2497 mnt_flags |= MNT_NOEXEC;
2498 if (flags & MS_NOATIME)
2499 mnt_flags |= MNT_NOATIME;
2500 if (flags & MS_NODIRATIME)
2501 mnt_flags |= MNT_NODIRATIME;
2502 if (flags & MS_STRICTATIME)
2503 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2504 if (flags & MS_RDONLY)
2505 mnt_flags |= MNT_READONLY;
2506
2507 /* The default atime for remount is preservation */
2508 if ((flags & MS_REMOUNT) &&
2509 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2510 MS_STRICTATIME)) == 0)) {
2511 mnt_flags &= ~MNT_ATIME_MASK;
2512 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2513 }
2514
2515 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2516 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2517 MS_STRICTATIME);
2518
2519 if (flags & MS_REMOUNT)
2520 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2521 data_page);
2522 else if (flags & MS_BIND)
2523 retval = do_loopback(&path, dev_name, flags & MS_REC);
2524 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2525 retval = do_change_type(&path, flags);
2526 else if (flags & MS_MOVE)
2527 retval = do_move_mount(&path, dev_name);
2528 else
2529 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2530 dev_name, data_page);
2531 dput_out:
2532 path_put(&path);
2533 return retval;
2534 }
2535
2536 static void free_mnt_ns(struct mnt_namespace *ns)
2537 {
2538 proc_free_inum(ns->proc_inum);
2539 put_user_ns(ns->user_ns);
2540 kfree(ns);
2541 }
2542
2543 /*
2544 * Assign a sequence number so we can detect when we attempt to bind
2545 * mount a reference to an older mount namespace into the current
2546 * mount namespace, preventing reference counting loops. A 64bit
2547 * number incrementing at 10Ghz will take 12,427 years to wrap which
2548 * is effectively never, so we can ignore the possibility.
2549 */
2550 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2551
2552 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2553 {
2554 struct mnt_namespace *new_ns;
2555 int ret;
2556
2557 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2558 if (!new_ns)
2559 return ERR_PTR(-ENOMEM);
2560 ret = proc_alloc_inum(&new_ns->proc_inum);
2561 if (ret) {
2562 kfree(new_ns);
2563 return ERR_PTR(ret);
2564 }
2565 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2566 atomic_set(&new_ns->count, 1);
2567 new_ns->root = NULL;
2568 INIT_LIST_HEAD(&new_ns->list);
2569 init_waitqueue_head(&new_ns->poll);
2570 new_ns->event = 0;
2571 new_ns->user_ns = get_user_ns(user_ns);
2572 return new_ns;
2573 }
2574
2575 /*
2576 * Allocate a new namespace structure and populate it with contents
2577 * copied from the namespace of the passed in task structure.
2578 */
2579 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2580 struct user_namespace *user_ns, struct fs_struct *fs)
2581 {
2582 struct mnt_namespace *new_ns;
2583 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2584 struct mount *p, *q;
2585 struct mount *old = mnt_ns->root;
2586 struct mount *new;
2587 int copy_flags;
2588
2589 new_ns = alloc_mnt_ns(user_ns);
2590 if (IS_ERR(new_ns))
2591 return new_ns;
2592
2593 namespace_lock();
2594 /* First pass: copy the tree topology */
2595 copy_flags = CL_COPY_ALL | CL_EXPIRE;
2596 if (user_ns != mnt_ns->user_ns)
2597 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2598 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2599 if (IS_ERR(new)) {
2600 namespace_unlock();
2601 free_mnt_ns(new_ns);
2602 return ERR_CAST(new);
2603 }
2604 new_ns->root = new;
2605 br_write_lock(&vfsmount_lock);
2606 list_add_tail(&new_ns->list, &new->mnt_list);
2607 br_write_unlock(&vfsmount_lock);
2608
2609 /*
2610 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2611 * as belonging to new namespace. We have already acquired a private
2612 * fs_struct, so tsk->fs->lock is not needed.
2613 */
2614 p = old;
2615 q = new;
2616 while (p) {
2617 q->mnt_ns = new_ns;
2618 if (fs) {
2619 if (&p->mnt == fs->root.mnt) {
2620 fs->root.mnt = mntget(&q->mnt);
2621 rootmnt = &p->mnt;
2622 }
2623 if (&p->mnt == fs->pwd.mnt) {
2624 fs->pwd.mnt = mntget(&q->mnt);
2625 pwdmnt = &p->mnt;
2626 }
2627 }
2628 p = next_mnt(p, old);
2629 q = next_mnt(q, new);
2630 }
2631 namespace_unlock();
2632
2633 if (rootmnt)
2634 mntput(rootmnt);
2635 if (pwdmnt)
2636 mntput(pwdmnt);
2637
2638 return new_ns;
2639 }
2640
2641 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2642 struct user_namespace *user_ns, struct fs_struct *new_fs)
2643 {
2644 struct mnt_namespace *new_ns;
2645
2646 BUG_ON(!ns);
2647 get_mnt_ns(ns);
2648
2649 if (!(flags & CLONE_NEWNS))
2650 return ns;
2651
2652 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2653
2654 put_mnt_ns(ns);
2655 return new_ns;
2656 }
2657
2658 /**
2659 * create_mnt_ns - creates a private namespace and adds a root filesystem
2660 * @mnt: pointer to the new root filesystem mountpoint
2661 */
2662 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2663 {
2664 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2665 if (!IS_ERR(new_ns)) {
2666 struct mount *mnt = real_mount(m);
2667 mnt->mnt_ns = new_ns;
2668 new_ns->root = mnt;
2669 list_add(&mnt->mnt_list, &new_ns->list);
2670 } else {
2671 mntput(m);
2672 }
2673 return new_ns;
2674 }
2675
2676 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2677 {
2678 struct mnt_namespace *ns;
2679 struct super_block *s;
2680 struct path path;
2681 int err;
2682
2683 ns = create_mnt_ns(mnt);
2684 if (IS_ERR(ns))
2685 return ERR_CAST(ns);
2686
2687 err = vfs_path_lookup(mnt->mnt_root, mnt,
2688 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2689
2690 put_mnt_ns(ns);
2691
2692 if (err)
2693 return ERR_PTR(err);
2694
2695 /* trade a vfsmount reference for active sb one */
2696 s = path.mnt->mnt_sb;
2697 atomic_inc(&s->s_active);
2698 mntput(path.mnt);
2699 /* lock the sucker */
2700 down_write(&s->s_umount);
2701 /* ... and return the root of (sub)tree on it */
2702 return path.dentry;
2703 }
2704 EXPORT_SYMBOL(mount_subtree);
2705
2706 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2707 char __user *, type, unsigned long, flags, void __user *, data)
2708 {
2709 int ret;
2710 char *kernel_type;
2711 struct filename *kernel_dir;
2712 char *kernel_dev;
2713 unsigned long data_page;
2714
2715 ret = copy_mount_string(type, &kernel_type);
2716 if (ret < 0)
2717 goto out_type;
2718
2719 kernel_dir = getname(dir_name);
2720 if (IS_ERR(kernel_dir)) {
2721 ret = PTR_ERR(kernel_dir);
2722 goto out_dir;
2723 }
2724
2725 ret = copy_mount_string(dev_name, &kernel_dev);
2726 if (ret < 0)
2727 goto out_dev;
2728
2729 ret = copy_mount_options(data, &data_page);
2730 if (ret < 0)
2731 goto out_data;
2732
2733 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2734 (void *) data_page);
2735
2736 free_page(data_page);
2737 out_data:
2738 kfree(kernel_dev);
2739 out_dev:
2740 putname(kernel_dir);
2741 out_dir:
2742 kfree(kernel_type);
2743 out_type:
2744 return ret;
2745 }
2746
2747 /*
2748 * Return true if path is reachable from root
2749 *
2750 * namespace_sem or vfsmount_lock is held
2751 */
2752 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2753 const struct path *root)
2754 {
2755 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2756 dentry = mnt->mnt_mountpoint;
2757 mnt = mnt->mnt_parent;
2758 }
2759 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2760 }
2761
2762 int path_is_under(struct path *path1, struct path *path2)
2763 {
2764 int res;
2765 br_read_lock(&vfsmount_lock);
2766 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2767 br_read_unlock(&vfsmount_lock);
2768 return res;
2769 }
2770 EXPORT_SYMBOL(path_is_under);
2771
2772 /*
2773 * pivot_root Semantics:
2774 * Moves the root file system of the current process to the directory put_old,
2775 * makes new_root as the new root file system of the current process, and sets
2776 * root/cwd of all processes which had them on the current root to new_root.
2777 *
2778 * Restrictions:
2779 * The new_root and put_old must be directories, and must not be on the
2780 * same file system as the current process root. The put_old must be
2781 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2782 * pointed to by put_old must yield the same directory as new_root. No other
2783 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2784 *
2785 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2786 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2787 * in this situation.
2788 *
2789 * Notes:
2790 * - we don't move root/cwd if they are not at the root (reason: if something
2791 * cared enough to change them, it's probably wrong to force them elsewhere)
2792 * - it's okay to pick a root that isn't the root of a file system, e.g.
2793 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2794 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2795 * first.
2796 */
2797 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2798 const char __user *, put_old)
2799 {
2800 struct path new, old, parent_path, root_parent, root;
2801 struct mount *new_mnt, *root_mnt, *old_mnt;
2802 struct mountpoint *old_mp, *root_mp;
2803 int error;
2804
2805 if (!may_mount())
2806 return -EPERM;
2807
2808 error = user_path_dir(new_root, &new);
2809 if (error)
2810 goto out0;
2811
2812 error = user_path_dir(put_old, &old);
2813 if (error)
2814 goto out1;
2815
2816 error = security_sb_pivotroot(&old, &new);
2817 if (error)
2818 goto out2;
2819
2820 get_fs_root(current->fs, &root);
2821 old_mp = lock_mount(&old);
2822 error = PTR_ERR(old_mp);
2823 if (IS_ERR(old_mp))
2824 goto out3;
2825
2826 error = -EINVAL;
2827 new_mnt = real_mount(new.mnt);
2828 root_mnt = real_mount(root.mnt);
2829 old_mnt = real_mount(old.mnt);
2830 if (IS_MNT_SHARED(old_mnt) ||
2831 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2832 IS_MNT_SHARED(root_mnt->mnt_parent))
2833 goto out4;
2834 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2835 goto out4;
2836 error = -ENOENT;
2837 if (d_unlinked(new.dentry))
2838 goto out4;
2839 error = -EBUSY;
2840 if (new_mnt == root_mnt || old_mnt == root_mnt)
2841 goto out4; /* loop, on the same file system */
2842 error = -EINVAL;
2843 if (root.mnt->mnt_root != root.dentry)
2844 goto out4; /* not a mountpoint */
2845 if (!mnt_has_parent(root_mnt))
2846 goto out4; /* not attached */
2847 root_mp = root_mnt->mnt_mp;
2848 if (new.mnt->mnt_root != new.dentry)
2849 goto out4; /* not a mountpoint */
2850 if (!mnt_has_parent(new_mnt))
2851 goto out4; /* not attached */
2852 /* make sure we can reach put_old from new_root */
2853 if (!is_path_reachable(old_mnt, old.dentry, &new))
2854 goto out4;
2855 root_mp->m_count++; /* pin it so it won't go away */
2856 br_write_lock(&vfsmount_lock);
2857 detach_mnt(new_mnt, &parent_path);
2858 detach_mnt(root_mnt, &root_parent);
2859 /* mount old root on put_old */
2860 attach_mnt(root_mnt, old_mnt, old_mp);
2861 /* mount new_root on / */
2862 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2863 touch_mnt_namespace(current->nsproxy->mnt_ns);
2864 br_write_unlock(&vfsmount_lock);
2865 chroot_fs_refs(&root, &new);
2866 put_mountpoint(root_mp);
2867 error = 0;
2868 out4:
2869 unlock_mount(old_mp