blob: f48f98110c30aefaa114ff0f15c0b7cece3f938d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/quotaops.h>
18#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080020#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070022#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080024#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/namei.h>
26#include <linux/security.h>
27#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020028#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080029#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010030#include <linux/idr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/uaccess.h>
32#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050033#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070034#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Eric Dumazet13f14b42008-02-06 01:37:57 -080036#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37#define HASH_SIZE (1UL << HASH_SHIFT)
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050040__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
41
42static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010043static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010044static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Eric Dumazetfa3536c2006-03-26 01:37:24 -080046static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080047static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050048static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080050/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060051struct kobject *fs_kobj;
52EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
55{
Ram Paib58fed82005-11-07 17:16:09 -050056 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080058 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060}
61
Dave Hansen3d733632008-02-15 14:37:59 -080062#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
63
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010064/* allocation is serialized by namespace_sem */
65static int mnt_alloc_id(struct vfsmount *mnt)
66{
67 int res;
68
69retry:
70 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
71 spin_lock(&vfsmount_lock);
72 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
73 spin_unlock(&vfsmount_lock);
74 if (res == -EAGAIN)
75 goto retry;
76
77 return res;
78}
79
80static void mnt_free_id(struct vfsmount *mnt)
81{
82 spin_lock(&vfsmount_lock);
83 ida_remove(&mnt_id_ida, mnt->mnt_id);
84 spin_unlock(&vfsmount_lock);
85}
86
Miklos Szeredi719f5d72008-03-27 13:06:23 +010087/*
88 * Allocate a new peer group ID
89 *
90 * mnt_group_ida is protected by namespace_sem
91 */
92static int mnt_alloc_group_id(struct vfsmount *mnt)
93{
94 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
95 return -ENOMEM;
96
97 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
98}
99
100/*
101 * Release a peer group ID
102 */
103void mnt_release_group_id(struct vfsmount *mnt)
104{
105 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
106 mnt->mnt_group_id = 0;
107}
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109struct vfsmount *alloc_vfsmnt(const char *name)
110{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800111 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100113 int err;
114
115 err = mnt_alloc_id(mnt);
116 if (err) {
117 kmem_cache_free(mnt_cache, mnt);
118 return NULL;
119 }
120
Ram Paib58fed82005-11-07 17:16:09 -0500121 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 INIT_LIST_HEAD(&mnt->mnt_hash);
123 INIT_LIST_HEAD(&mnt->mnt_child);
124 INIT_LIST_HEAD(&mnt->mnt_mounts);
125 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700126 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500127 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500128 INIT_LIST_HEAD(&mnt->mnt_slave_list);
129 INIT_LIST_HEAD(&mnt->mnt_slave);
Dave Hansen3d733632008-02-15 14:37:59 -0800130 atomic_set(&mnt->__mnt_writers, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -0500132 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 char *newname = kmalloc(size, GFP_KERNEL);
134 if (newname) {
135 memcpy(newname, name, size);
136 mnt->mnt_devname = newname;
137 }
138 }
139 }
140 return mnt;
141}
142
Dave Hansen83660252008-02-15 14:37:30 -0800143/*
144 * Most r/o checks on a fs are for operations that take
145 * discrete amounts of time, like a write() or unlink().
146 * We must keep track of when those operations start
147 * (for permission checks) and when they end, so that
148 * we can determine when writes are able to occur to
149 * a filesystem.
150 */
Dave Hansen3d733632008-02-15 14:37:59 -0800151/*
152 * __mnt_is_readonly: check whether a mount is read-only
153 * @mnt: the mount to check for its write status
154 *
155 * This shouldn't be used directly ouside of the VFS.
156 * It does not guarantee that the filesystem will stay
157 * r/w, just that it is right *now*. This can not and
158 * should not be used in place of IS_RDONLY(inode).
159 * mnt_want/drop_write() will _keep_ the filesystem
160 * r/w.
161 */
162int __mnt_is_readonly(struct vfsmount *mnt)
163{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800164 if (mnt->mnt_flags & MNT_READONLY)
165 return 1;
166 if (mnt->mnt_sb->s_flags & MS_RDONLY)
167 return 1;
168 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800169}
170EXPORT_SYMBOL_GPL(__mnt_is_readonly);
171
172struct mnt_writer {
173 /*
174 * If holding multiple instances of this lock, they
175 * must be ordered by cpu number.
176 */
177 spinlock_t lock;
178 struct lock_class_key lock_class; /* compiles out with !lockdep */
179 unsigned long count;
180 struct vfsmount *mnt;
181} ____cacheline_aligned_in_smp;
182static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
183
184static int __init init_mnt_writers(void)
185{
186 int cpu;
187 for_each_possible_cpu(cpu) {
188 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
189 spin_lock_init(&writer->lock);
190 lockdep_set_class(&writer->lock, &writer->lock_class);
191 writer->count = 0;
192 }
193 return 0;
194}
195fs_initcall(init_mnt_writers);
196
197static void unlock_mnt_writers(void)
198{
199 int cpu;
200 struct mnt_writer *cpu_writer;
201
202 for_each_possible_cpu(cpu) {
203 cpu_writer = &per_cpu(mnt_writers, cpu);
204 spin_unlock(&cpu_writer->lock);
205 }
206}
207
208static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
209{
210 if (!cpu_writer->mnt)
211 return;
212 /*
213 * This is in case anyone ever leaves an invalid,
214 * old ->mnt and a count of 0.
215 */
216 if (!cpu_writer->count)
217 return;
218 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
219 cpu_writer->count = 0;
220}
221 /*
222 * must hold cpu_writer->lock
223 */
224static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
225 struct vfsmount *mnt)
226{
227 if (cpu_writer->mnt == mnt)
228 return;
229 __clear_mnt_count(cpu_writer);
230 cpu_writer->mnt = mnt;
231}
232
233/*
234 * Most r/o checks on a fs are for operations that take
235 * discrete amounts of time, like a write() or unlink().
236 * We must keep track of when those operations start
237 * (for permission checks) and when they end, so that
238 * we can determine when writes are able to occur to
239 * a filesystem.
240 */
Dave Hansen83660252008-02-15 14:37:30 -0800241/**
242 * mnt_want_write - get write access to a mount
243 * @mnt: the mount on which to take a write
244 *
245 * This tells the low-level filesystem that a write is
246 * about to be performed to it, and makes sure that
247 * writes are allowed before returning success. When
248 * the write operation is finished, mnt_drop_write()
249 * must be called. This is effectively a refcount.
250 */
251int mnt_want_write(struct vfsmount *mnt)
252{
Dave Hansen3d733632008-02-15 14:37:59 -0800253 int ret = 0;
254 struct mnt_writer *cpu_writer;
255
256 cpu_writer = &get_cpu_var(mnt_writers);
257 spin_lock(&cpu_writer->lock);
258 if (__mnt_is_readonly(mnt)) {
259 ret = -EROFS;
260 goto out;
261 }
262 use_cpu_writer_for_mount(cpu_writer, mnt);
263 cpu_writer->count++;
264out:
265 spin_unlock(&cpu_writer->lock);
266 put_cpu_var(mnt_writers);
267 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800268}
269EXPORT_SYMBOL_GPL(mnt_want_write);
270
Dave Hansen3d733632008-02-15 14:37:59 -0800271static void lock_mnt_writers(void)
272{
273 int cpu;
274 struct mnt_writer *cpu_writer;
275
276 for_each_possible_cpu(cpu) {
277 cpu_writer = &per_cpu(mnt_writers, cpu);
278 spin_lock(&cpu_writer->lock);
279 __clear_mnt_count(cpu_writer);
280 cpu_writer->mnt = NULL;
281 }
282}
283
284/*
285 * These per-cpu write counts are not guaranteed to have
286 * matched increments and decrements on any given cpu.
287 * A file open()ed for write on one cpu and close()d on
288 * another cpu will imbalance this count. Make sure it
289 * does not get too far out of whack.
290 */
291static void handle_write_count_underflow(struct vfsmount *mnt)
292{
293 if (atomic_read(&mnt->__mnt_writers) >=
294 MNT_WRITER_UNDERFLOW_LIMIT)
295 return;
296 /*
297 * It isn't necessary to hold all of the locks
298 * at the same time, but doing it this way makes
299 * us share a lot more code.
300 */
301 lock_mnt_writers();
302 /*
303 * vfsmount_lock is for mnt_flags.
304 */
305 spin_lock(&vfsmount_lock);
306 /*
307 * If coalescing the per-cpu writer counts did not
308 * get us back to a positive writer count, we have
309 * a bug.
310 */
311 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
312 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
313 printk(KERN_DEBUG "leak detected on mount(%p) writers "
314 "count: %d\n",
315 mnt, atomic_read(&mnt->__mnt_writers));
316 WARN_ON(1);
317 /* use the flag to keep the dmesg spam down */
318 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
319 }
320 spin_unlock(&vfsmount_lock);
321 unlock_mnt_writers();
322}
323
Dave Hansen83660252008-02-15 14:37:30 -0800324/**
325 * mnt_drop_write - give up write access to a mount
326 * @mnt: the mount on which to give up write access
327 *
328 * Tells the low-level filesystem that we are done
329 * performing writes to it. Must be matched with
330 * mnt_want_write() call above.
331 */
332void mnt_drop_write(struct vfsmount *mnt)
333{
Dave Hansen3d733632008-02-15 14:37:59 -0800334 int must_check_underflow = 0;
335 struct mnt_writer *cpu_writer;
336
337 cpu_writer = &get_cpu_var(mnt_writers);
338 spin_lock(&cpu_writer->lock);
339
340 use_cpu_writer_for_mount(cpu_writer, mnt);
341 if (cpu_writer->count > 0) {
342 cpu_writer->count--;
343 } else {
344 must_check_underflow = 1;
345 atomic_dec(&mnt->__mnt_writers);
346 }
347
348 spin_unlock(&cpu_writer->lock);
349 /*
350 * Logically, we could call this each time,
351 * but the __mnt_writers cacheline tends to
352 * be cold, and makes this expensive.
353 */
354 if (must_check_underflow)
355 handle_write_count_underflow(mnt);
356 /*
357 * This could be done right after the spinlock
358 * is taken because the spinlock keeps us on
359 * the cpu, and disables preemption. However,
360 * putting it here bounds the amount that
361 * __mnt_writers can underflow. Without it,
362 * we could theoretically wrap __mnt_writers.
363 */
364 put_cpu_var(mnt_writers);
Dave Hansen83660252008-02-15 14:37:30 -0800365}
366EXPORT_SYMBOL_GPL(mnt_drop_write);
367
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800368static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800369{
Dave Hansen3d733632008-02-15 14:37:59 -0800370 int ret = 0;
371
372 lock_mnt_writers();
373 /*
374 * With all the locks held, this value is stable
375 */
376 if (atomic_read(&mnt->__mnt_writers) > 0) {
377 ret = -EBUSY;
378 goto out;
379 }
380 /*
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800381 * nobody can do a successful mnt_want_write() with all
382 * of the counts in MNT_DENIED_WRITE and the locks held.
Dave Hansen3d733632008-02-15 14:37:59 -0800383 */
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800384 spin_lock(&vfsmount_lock);
385 if (!ret)
386 mnt->mnt_flags |= MNT_READONLY;
387 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800388out:
389 unlock_mnt_writers();
390 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800391}
Dave Hansen83660252008-02-15 14:37:30 -0800392
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800393static void __mnt_unmake_readonly(struct vfsmount *mnt)
394{
395 spin_lock(&vfsmount_lock);
396 mnt->mnt_flags &= ~MNT_READONLY;
397 spin_unlock(&vfsmount_lock);
398}
399
David Howells454e2392006-06-23 02:02:57 -0700400int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
401{
402 mnt->mnt_sb = sb;
403 mnt->mnt_root = dget(sb->s_root);
404 return 0;
405}
406
407EXPORT_SYMBOL(simple_set_mnt);
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409void free_vfsmnt(struct vfsmount *mnt)
410{
411 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100412 mnt_free_id(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 kmem_cache_free(mnt_cache, mnt);
414}
415
416/*
Ram Paia05964f2005-11-07 17:20:17 -0500417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 */
Ram Paia05964f2005-11-07 17:20:17 -0500420struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
421 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Ram Paib58fed82005-11-07 17:16:09 -0500423 struct list_head *head = mount_hashtable + hash(mnt, dentry);
424 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 struct vfsmount *p, *found = NULL;
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500428 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 p = NULL;
430 if (tmp == head)
431 break;
432 p = list_entry(tmp, struct vfsmount, mnt_hash);
433 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500434 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break;
436 }
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 return found;
439}
440
Ram Paia05964f2005-11-07 17:20:17 -0500441/*
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
444 */
445struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
446{
447 struct vfsmount *child_mnt;
448 spin_lock(&vfsmount_lock);
449 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
450 mntget(child_mnt);
451 spin_unlock(&vfsmount_lock);
452 return child_mnt;
453}
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455static inline int check_mnt(struct vfsmount *mnt)
456{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800457 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800460static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500461{
462 if (ns) {
463 ns->event = ++event;
464 wake_up_interruptible(&ns->poll);
465 }
466}
467
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800468static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500469{
470 if (ns && ns->event != event) {
471 ns->event = event;
472 wake_up_interruptible(&ns->poll);
473 }
474}
475
Al Viro1a390682008-03-21 20:48:19 -0400476static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Al Viro1a390682008-03-21 20:48:19 -0400478 old_path->dentry = mnt->mnt_mountpoint;
479 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 mnt->mnt_parent = mnt;
481 mnt->mnt_mountpoint = mnt->mnt_root;
482 list_del_init(&mnt->mnt_child);
483 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400484 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Ram Paib90fa9a2005-11-07 17:19:50 -0500487void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
488 struct vfsmount *child_mnt)
489{
490 child_mnt->mnt_parent = mntget(mnt);
491 child_mnt->mnt_mountpoint = dget(dentry);
492 dentry->d_mounted++;
493}
494
Al Viro1a390682008-03-21 20:48:19 -0400495static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Al Viro1a390682008-03-21 20:48:19 -0400497 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500498 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400499 hash(path->mnt, path->dentry));
500 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500501}
502
503/*
504 * the caller must hold vfsmount_lock
505 */
506static void commit_tree(struct vfsmount *mnt)
507{
508 struct vfsmount *parent = mnt->mnt_parent;
509 struct vfsmount *m;
510 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800511 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500512
513 BUG_ON(parent == mnt);
514
515 list_add_tail(&head, &mnt->mnt_list);
516 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800517 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500518 list_splice(&head, n->list.prev);
519
520 list_add_tail(&mnt->mnt_hash, mount_hashtable +
521 hash(parent, mnt->mnt_mountpoint));
522 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800523 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
527{
528 struct list_head *next = p->mnt_mounts.next;
529 if (next == &p->mnt_mounts) {
530 while (1) {
531 if (p == root)
532 return NULL;
533 next = p->mnt_child.next;
534 if (next != &p->mnt_parent->mnt_mounts)
535 break;
536 p = p->mnt_parent;
537 }
538 }
539 return list_entry(next, struct vfsmount, mnt_child);
540}
541
Ram Pai9676f0c2005-11-07 17:21:20 -0500542static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
543{
544 struct list_head *prev = p->mnt_mounts.prev;
545 while (prev != &p->mnt_mounts) {
546 p = list_entry(prev, struct vfsmount, mnt_child);
547 prev = p->mnt_mounts.prev;
548 }
549 return p;
550}
551
Ram Pai36341f62005-11-07 17:17:22 -0500552static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
553 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 struct super_block *sb = old->mnt_sb;
556 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
557
558 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100559 if (flag & (CL_SLAVE | CL_PRIVATE))
560 mnt->mnt_group_id = 0; /* not a peer of original */
561 else
562 mnt->mnt_group_id = old->mnt_group_id;
563
564 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
565 int err = mnt_alloc_group_id(mnt);
566 if (err)
567 goto out_free;
568 }
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 mnt->mnt_flags = old->mnt_flags;
571 atomic_inc(&sb->s_active);
572 mnt->mnt_sb = sb;
573 mnt->mnt_root = dget(root);
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500576
Ram Pai5afe0022005-11-07 17:21:01 -0500577 if (flag & CL_SLAVE) {
578 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
579 mnt->mnt_master = old;
580 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400581 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500582 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
583 list_add(&mnt->mnt_share, &old->mnt_share);
584 if (IS_MNT_SLAVE(old))
585 list_add(&mnt->mnt_slave, &old->mnt_slave);
586 mnt->mnt_master = old->mnt_master;
587 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500588 if (flag & CL_MAKE_SHARED)
589 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500593 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500594 if (!list_empty(&old->mnt_expire))
595 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100599
600 out_free:
601 free_vfsmnt(mnt);
602 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Al Viro7b7b1ac2005-11-07 17:13:39 -0500605static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Dave Hansen3d733632008-02-15 14:37:59 -0800607 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800609 /*
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
613 * can come in.
614 */
615 for_each_possible_cpu(cpu) {
616 struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
617 if (cpu_writer->mnt != mnt)
618 continue;
619 spin_lock(&cpu_writer->lock);
620 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
621 cpu_writer->count = 0;
622 /*
623 * Might as well do this so that no one
624 * ever sees the pointer and expects
625 * it to be valid.
626 */
627 cpu_writer->mnt = NULL;
628 spin_unlock(&cpu_writer->lock);
629 }
630 /*
631 * This probably indicates that somebody messed
632 * up a mnt_want/drop_write() pair. If this
633 * happens, the filesystem was probably unable
634 * to make r/w->r/o transitions.
635 */
636 WARN_ON(atomic_read(&mnt->__mnt_writers));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 dput(mnt->mnt_root);
638 free_vfsmnt(mnt);
639 deactivate_super(sb);
640}
641
Al Viro7b7b1ac2005-11-07 17:13:39 -0500642void mntput_no_expire(struct vfsmount *mnt)
643{
644repeat:
645 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
646 if (likely(!mnt->mnt_pinned)) {
647 spin_unlock(&vfsmount_lock);
648 __mntput(mnt);
649 return;
650 }
651 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
652 mnt->mnt_pinned = 0;
653 spin_unlock(&vfsmount_lock);
654 acct_auto_close_mnt(mnt);
655 security_sb_umount_close(mnt);
656 goto repeat;
657 }
658}
659
660EXPORT_SYMBOL(mntput_no_expire);
661
662void mnt_pin(struct vfsmount *mnt)
663{
664 spin_lock(&vfsmount_lock);
665 mnt->mnt_pinned++;
666 spin_unlock(&vfsmount_lock);
667}
668
669EXPORT_SYMBOL(mnt_pin);
670
671void mnt_unpin(struct vfsmount *mnt)
672{
673 spin_lock(&vfsmount_lock);
674 if (mnt->mnt_pinned) {
675 atomic_inc(&mnt->mnt_count);
676 mnt->mnt_pinned--;
677 }
678 spin_unlock(&vfsmount_lock);
679}
680
681EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800683static inline void mangle(struct seq_file *m, const char *s)
684{
685 seq_escape(m, s, " \t\n\\");
686}
687
688/*
689 * Simple .show_options callback for filesystems which don't want to
690 * implement more complex mount option showing.
691 *
692 * See also save_mount_options().
693 */
694int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
695{
696 const char *options = mnt->mnt_sb->s_options;
697
698 if (options != NULL && options[0]) {
699 seq_putc(m, ',');
700 mangle(m, options);
701 }
702
703 return 0;
704}
705EXPORT_SYMBOL(generic_show_options);
706
707/*
708 * If filesystem uses generic_show_options(), this function should be
709 * called from the fill_super() callback.
710 *
711 * The .remount_fs callback usually needs to be handled in a special
712 * way, to make sure, that previous options are not overwritten if the
713 * remount fails.
714 *
715 * Also note, that if the filesystem's .remount_fs function doesn't
716 * reset all options to their default value, but changes only newly
717 * given options, then the displayed options will not reflect reality
718 * any more.
719 */
720void save_mount_options(struct super_block *sb, char *options)
721{
722 kfree(sb->s_options);
723 sb->s_options = kstrdup(options, GFP_KERNEL);
724}
725EXPORT_SYMBOL(save_mount_options);
726
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100727#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728/* iterator */
729static void *m_start(struct seq_file *m, loff_t *pos)
730{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100731 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Ram Pai390c6842005-11-07 17:17:51 -0500733 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100734 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
737static void *m_next(struct seq_file *m, void *v, loff_t *pos)
738{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100739 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700740
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100741 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
744static void m_stop(struct seq_file *m, void *v)
745{
Ram Pai390c6842005-11-07 17:17:51 -0500746 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
Ram Pai2d4d4862008-03-27 13:06:25 +0100749struct proc_fs_info {
750 int flag;
751 const char *str;
752};
753
754static void show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
Ram Pai2d4d4862008-03-27 13:06:25 +0100756 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 { MS_SYNCHRONOUS, ",sync" },
758 { MS_DIRSYNC, ",dirsync" },
759 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 { 0, NULL }
761 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100762 const struct proc_fs_info *fs_infop;
763
764 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
765 if (sb->s_flags & fs_infop->flag)
766 seq_puts(m, fs_infop->str);
767 }
768}
769
770static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
771{
772 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 { MNT_NOSUID, ",nosuid" },
774 { MNT_NODEV, ",nodev" },
775 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800776 { MNT_NOATIME, ",noatime" },
777 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800778 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 { 0, NULL }
780 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100781 const struct proc_fs_info *fs_infop;
782
783 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
784 if (mnt->mnt_flags & fs_infop->flag)
785 seq_puts(m, fs_infop->str);
786 }
787}
788
789static void show_type(struct seq_file *m, struct super_block *sb)
790{
791 mangle(m, sb->s_type->name);
792 if (sb->s_subtype && sb->s_subtype[0]) {
793 seq_putc(m, '.');
794 mangle(m, sb->s_subtype);
795 }
796}
797
798static int show_vfsmnt(struct seq_file *m, void *v)
799{
800 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
801 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800802 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
805 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800806 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100808 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800809 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Ram Pai2d4d4862008-03-27 13:06:25 +0100810 show_sb_opts(m, mnt->mnt_sb);
811 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (mnt->mnt_sb->s_op->show_options)
813 err = mnt->mnt_sb->s_op->show_options(m, mnt);
814 seq_puts(m, " 0 0\n");
815 return err;
816}
817
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100818const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 .start = m_start,
820 .next = m_next,
821 .stop = m_stop,
822 .show = show_vfsmnt
823};
824
Ram Pai2d4d4862008-03-27 13:06:25 +0100825static int show_mountinfo(struct seq_file *m, void *v)
826{
827 struct proc_mounts *p = m->private;
828 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
829 struct super_block *sb = mnt->mnt_sb;
830 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
831 struct path root = p->root;
832 int err = 0;
833
834 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
835 MAJOR(sb->s_dev), MINOR(sb->s_dev));
836 seq_dentry(m, mnt->mnt_root, " \t\n\\");
837 seq_putc(m, ' ');
838 seq_path_root(m, &mnt_path, &root, " \t\n\\");
839 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
840 /*
841 * Mountpoint is outside root, discard that one. Ugly,
842 * but less so than trying to do that in iterator in a
843 * race-free way (due to renames).
844 */
845 return SEQ_SKIP;
846 }
847 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
848 show_mnt_opts(m, mnt);
849
850 /* Tagged fields ("foo:X" or "bar") */
851 if (IS_MNT_SHARED(mnt))
852 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100853 if (IS_MNT_SLAVE(mnt)) {
854 int master = mnt->mnt_master->mnt_group_id;
855 int dom = get_dominating_id(mnt, &p->root);
856 seq_printf(m, " master:%i", master);
857 if (dom && dom != master)
858 seq_printf(m, " propagate_from:%i", dom);
859 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100860 if (IS_MNT_UNBINDABLE(mnt))
861 seq_puts(m, " unbindable");
862
863 /* Filesystem specific data */
864 seq_puts(m, " - ");
865 show_type(m, sb);
866 seq_putc(m, ' ');
867 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
868 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
869 show_sb_opts(m, sb);
870 if (sb->s_op->show_options)
871 err = sb->s_op->show_options(m, mnt);
872 seq_putc(m, '\n');
873 return err;
874}
875
876const struct seq_operations mountinfo_op = {
877 .start = m_start,
878 .next = m_next,
879 .stop = m_stop,
880 .show = show_mountinfo,
881};
882
Chuck Leverb4629fe2006-03-20 13:44:12 -0500883static int show_vfsstat(struct seq_file *m, void *v)
884{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700885 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800886 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500887 int err = 0;
888
889 /* device */
890 if (mnt->mnt_devname) {
891 seq_puts(m, "device ");
892 mangle(m, mnt->mnt_devname);
893 } else
894 seq_puts(m, "no device");
895
896 /* mount point */
897 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800898 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500899 seq_putc(m, ' ');
900
901 /* file system type */
902 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100903 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500904
905 /* optional statistics */
906 if (mnt->mnt_sb->s_op->show_stats) {
907 seq_putc(m, ' ');
908 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
909 }
910
911 seq_putc(m, '\n');
912 return err;
913}
914
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100915const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500916 .start = m_start,
917 .next = m_next,
918 .stop = m_stop,
919 .show = show_vfsstat,
920};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100921#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923/**
924 * may_umount_tree - check if a mount tree is busy
925 * @mnt: root of mount tree
926 *
927 * This is called to check if a tree of mounts has any
928 * open files, pwds, chroots or sub mounts that are
929 * busy.
930 */
931int may_umount_tree(struct vfsmount *mnt)
932{
Ram Pai36341f62005-11-07 17:17:22 -0500933 int actual_refs = 0;
934 int minimum_refs = 0;
935 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500938 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 actual_refs += atomic_read(&p->mnt_count);
940 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 }
942 spin_unlock(&vfsmount_lock);
943
944 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800945 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Ian Kente3474a82006-03-27 01:14:51 -0800947 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948}
949
950EXPORT_SYMBOL(may_umount_tree);
951
952/**
953 * may_umount - check if a mount point is busy
954 * @mnt: root of mount
955 *
956 * This is called to check if a mount point has any
957 * open files, pwds, chroots or sub mounts. If the
958 * mount has sub mounts this will return busy
959 * regardless of whether the sub mounts are busy.
960 *
961 * Doesn't take quota and stuff into account. IOW, in some cases it will
962 * give false negatives. The main reason why it's here is that we need
963 * a non-destructive way to look for easily umountable filesystems.
964 */
965int may_umount(struct vfsmount *mnt)
966{
Ian Kente3474a82006-03-27 01:14:51 -0800967 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500968 spin_lock(&vfsmount_lock);
969 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800970 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500971 spin_unlock(&vfsmount_lock);
972 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
974
975EXPORT_SYMBOL(may_umount);
976
Ram Paib90fa9a2005-11-07 17:19:50 -0500977void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500979 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800980 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700981 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500982 list_del_init(&mnt->mnt_hash);
983 if (mnt->mnt_parent != mnt) {
984 struct dentry *dentry;
985 struct vfsmount *m;
986 spin_lock(&vfsmount_lock);
987 dentry = mnt->mnt_mountpoint;
988 m = mnt->mnt_parent;
989 mnt->mnt_mountpoint = mnt->mnt_root;
990 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -0400991 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500993 dput(dentry);
994 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 }
996 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500997 }
998}
999
Ram Paia05964f2005-11-07 17:20:17 -05001000void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001001{
1002 struct vfsmount *p;
1003
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001004 for (p = mnt; p; p = next_mnt(p, mnt))
1005 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001006
Ram Paia05964f2005-11-07 17:20:17 -05001007 if (propagate)
1008 propagate_umount(kill);
1009
Ram Pai70fbcdf2005-11-07 17:17:04 -05001010 list_for_each_entry(p, kill, mnt_hash) {
1011 list_del_init(&p->mnt_expire);
1012 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001013 __touch_mnt_namespace(p->mnt_ns);
1014 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001015 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001016 if (p->mnt_parent != p) {
1017 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001018 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001019 }
Ram Paia05964f2005-11-07 17:20:17 -05001020 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
1022}
1023
Al Viroc35038b2008-03-22 00:46:23 -04001024static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026static int do_umount(struct vfsmount *mnt, int flags)
1027{
Ram Paib58fed82005-11-07 17:16:09 -05001028 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001030 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 retval = security_sb_umount(mnt, flags);
1033 if (retval)
1034 return retval;
1035
1036 /*
1037 * Allow userspace to request a mountpoint be expired rather than
1038 * unmounting unconditionally. Unmount only happens if:
1039 * (1) the mark is already set (the mark is cleared by mntput())
1040 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1041 */
1042 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001043 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 flags & (MNT_FORCE | MNT_DETACH))
1045 return -EINVAL;
1046
1047 if (atomic_read(&mnt->mnt_count) != 2)
1048 return -EBUSY;
1049
1050 if (!xchg(&mnt->mnt_expiry_mark, 1))
1051 return -EAGAIN;
1052 }
1053
1054 /*
1055 * If we may have to abort operations to get out of this
1056 * mount, and they will themselves hold resources we must
1057 * allow the fs to do things. In the Unix tradition of
1058 * 'Gee thats tricky lets do it in userspace' the umount_begin
1059 * might fail to complete on the first run through as other tasks
1060 * must return, and the like. Thats for the mount program to worry
1061 * about for the moment.
1062 */
1063
Al Viro42faad92008-04-24 07:21:56 -04001064 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1065 lock_kernel();
1066 sb->s_op->umount_begin(sb);
1067 unlock_kernel();
1068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 /*
1071 * No sense to grab the lock for this test, but test itself looks
1072 * somewhat bogus. Suggestions for better replacement?
1073 * Ho-hum... In principle, we might treat that as umount + switch
1074 * to rootfs. GC would eventually take care of the old vfsmount.
1075 * Actually it makes sense, especially if rootfs would contain a
1076 * /reboot - static binary that would close all descriptors and
1077 * call reboot(9). Then init(8) could umount root and exec /reboot.
1078 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001079 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 /*
1081 * Special case for "unmounting" root ...
1082 * we just try to remount it readonly.
1083 */
1084 down_write(&sb->s_umount);
1085 if (!(sb->s_flags & MS_RDONLY)) {
1086 lock_kernel();
1087 DQUOT_OFF(sb);
1088 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1089 unlock_kernel();
1090 }
1091 up_write(&sb->s_umount);
1092 return retval;
1093 }
1094
Ram Pai390c6842005-11-07 17:17:51 -05001095 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001097 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Al Viroc35038b2008-03-22 00:46:23 -04001099 if (!(flags & MNT_DETACH))
1100 shrink_submounts(mnt, &umount_list);
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001103 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001105 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 retval = 0;
1107 }
1108 spin_unlock(&vfsmount_lock);
1109 if (retval)
1110 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001111 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001112 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 return retval;
1114}
1115
1116/*
1117 * Now umount can handle mount points as well as block devices.
1118 * This is important for filesystems which use unnamed block devices.
1119 *
1120 * We now support a flag for forced unmount like the other 'big iron'
1121 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1122 */
1123
1124asmlinkage long sys_umount(char __user * name, int flags)
1125{
1126 struct nameidata nd;
1127 int retval;
1128
1129 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
1130 if (retval)
1131 goto out;
1132 retval = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001133 if (nd.path.dentry != nd.path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 goto dput_and_out;
Jan Blunck4ac91372008-02-14 19:34:32 -08001135 if (!check_mnt(nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 goto dput_and_out;
1137
1138 retval = -EPERM;
1139 if (!capable(CAP_SYS_ADMIN))
1140 goto dput_and_out;
1141
Jan Blunck4ac91372008-02-14 19:34:32 -08001142 retval = do_umount(nd.path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001144 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Jan Blunck4ac91372008-02-14 19:34:32 -08001145 dput(nd.path.dentry);
1146 mntput_no_expire(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147out:
1148 return retval;
1149}
1150
1151#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1152
1153/*
Ram Paib58fed82005-11-07 17:16:09 -05001154 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156asmlinkage long sys_oldumount(char __user * name)
1157{
Ram Paib58fed82005-11-07 17:16:09 -05001158 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159}
1160
1161#endif
1162
1163static int mount_is_safe(struct nameidata *nd)
1164{
1165 if (capable(CAP_SYS_ADMIN))
1166 return 0;
1167 return -EPERM;
1168#ifdef notyet
Jan Blunck4ac91372008-02-14 19:34:32 -08001169 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 return -EPERM;
Jan Blunck4ac91372008-02-14 19:34:32 -08001171 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
1172 if (current->uid != nd->path.dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 return -EPERM;
1174 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001175 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 return -EPERM;
1177 return 0;
1178#endif
1179}
1180
Ram Paib58fed82005-11-07 17:16:09 -05001181static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
1183 while (1) {
1184 if (d == dentry)
1185 return 1;
1186 if (d == NULL || d == d->d_parent)
1187 return 0;
1188 d = d->d_parent;
1189 }
1190}
1191
Ram Paib90fa9a2005-11-07 17:19:50 -05001192struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001193 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194{
1195 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001196 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Ram Pai9676f0c2005-11-07 17:21:20 -05001198 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1199 return NULL;
1200
Ram Pai36341f62005-11-07 17:17:22 -05001201 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (!q)
1203 goto Enomem;
1204 q->mnt_mountpoint = mnt->mnt_mountpoint;
1205
1206 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001207 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
1209 continue;
1210
1211 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001212 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1213 s = skip_mnt_tree(s);
1214 continue;
1215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 while (p != s->mnt_parent) {
1217 p = p->mnt_parent;
1218 q = q->mnt_parent;
1219 }
1220 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001221 path.mnt = q;
1222 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001223 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 if (!q)
1225 goto Enomem;
1226 spin_lock(&vfsmount_lock);
1227 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001228 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 spin_unlock(&vfsmount_lock);
1230 }
1231 }
1232 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001233Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001235 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001237 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001239 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
1241 return NULL;
1242}
1243
Al Viro8aec0802007-06-07 12:20:32 -04001244struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1245{
1246 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001247 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001248 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001249 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001250 return tree;
1251}
1252
1253void drop_collected_mounts(struct vfsmount *mnt)
1254{
1255 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001256 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001257 spin_lock(&vfsmount_lock);
1258 umount_tree(mnt, 0, &umount_list);
1259 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001260 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001261 release_mounts(&umount_list);
1262}
1263
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001264static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1265{
1266 struct vfsmount *p;
1267
1268 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1269 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1270 mnt_release_group_id(p);
1271 }
1272}
1273
1274static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1275{
1276 struct vfsmount *p;
1277
1278 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1279 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1280 int err = mnt_alloc_group_id(p);
1281 if (err) {
1282 cleanup_group_ids(mnt, p);
1283 return err;
1284 }
1285 }
1286 }
1287
1288 return 0;
1289}
1290
Ram Paib90fa9a2005-11-07 17:19:50 -05001291/*
1292 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001293 * @nd : place the mount tree @source_mnt is attached
1294 * @parent_nd : if non-null, detach the source_mnt from its parent and
1295 * store the parent mount and mountpoint dentry.
1296 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001297 *
1298 * NOTE: in the table below explains the semantics when a source mount
1299 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001300 * ---------------------------------------------------------------------------
1301 * | BIND MOUNT OPERATION |
1302 * |**************************************************************************
1303 * | source-->| shared | private | slave | unbindable |
1304 * | dest | | | | |
1305 * | | | | | | |
1306 * | v | | | | |
1307 * |**************************************************************************
1308 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1309 * | | | | | |
1310 * |non-shared| shared (+) | private | slave (*) | invalid |
1311 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001312 * A bind operation clones the source mount and mounts the clone on the
1313 * destination mount.
1314 *
1315 * (++) the cloned mount is propagated to all the mounts in the propagation
1316 * tree of the destination mount and the cloned mount is added to
1317 * the peer group of the source mount.
1318 * (+) the cloned mount is created under the destination mount and is marked
1319 * as shared. The cloned mount is added to the peer group of the source
1320 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001321 * (+++) the mount is propagated to all the mounts in the propagation tree
1322 * of the destination mount and the cloned mount is made slave
1323 * of the same master as that of the source mount. The cloned mount
1324 * is marked as 'shared and slave'.
1325 * (*) the cloned mount is made a slave of the same master as that of the
1326 * source mount.
1327 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001328 * ---------------------------------------------------------------------------
1329 * | MOVE MOUNT OPERATION |
1330 * |**************************************************************************
1331 * | source-->| shared | private | slave | unbindable |
1332 * | dest | | | | |
1333 * | | | | | | |
1334 * | v | | | | |
1335 * |**************************************************************************
1336 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1337 * | | | | | |
1338 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1339 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001340 *
1341 * (+) the mount is moved to the destination. And is then propagated to
1342 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001343 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001344 * (+++) the mount is moved to the destination and is then propagated to
1345 * all the mounts belonging to the destination mount's propagation tree.
1346 * the mount is marked as 'shared and slave'.
1347 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001348 *
1349 * if the source mount is a tree, the operations explained above is
1350 * applied to each mount in the tree.
1351 * Must be called without spinlocks held, since this function can sleep
1352 * in allocations.
1353 */
1354static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001355 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001356{
1357 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001358 struct vfsmount *dest_mnt = path->mnt;
1359 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001360 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001361 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001362
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001363 if (IS_MNT_SHARED(dest_mnt)) {
1364 err = invent_group_ids(source_mnt, true);
1365 if (err)
1366 goto out;
1367 }
1368 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1369 if (err)
1370 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001371
1372 if (IS_MNT_SHARED(dest_mnt)) {
1373 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1374 set_mnt_shared(p);
1375 }
1376
1377 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001378 if (parent_path) {
1379 detach_mnt(source_mnt, parent_path);
1380 attach_mnt(source_mnt, path);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001381 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001382 } else {
1383 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1384 commit_tree(source_mnt);
1385 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001386
1387 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1388 list_del_init(&child->mnt_hash);
1389 commit_tree(child);
1390 }
1391 spin_unlock(&vfsmount_lock);
1392 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001393
1394 out_cleanup_ids:
1395 if (IS_MNT_SHARED(dest_mnt))
1396 cleanup_group_ids(source_mnt, NULL);
1397 out:
1398 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001399}
1400
Al Viro8c3ee422008-03-22 18:00:39 -04001401static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402{
1403 int err;
1404 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1405 return -EINVAL;
1406
Al Viro8c3ee422008-03-22 18:00:39 -04001407 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1409 return -ENOTDIR;
1410
1411 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001412 mutex_lock(&path->dentry->d_inode->i_mutex);
1413 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto out_unlock;
1415
Al Viro8c3ee422008-03-22 18:00:39 -04001416 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 if (err)
1418 goto out_unlock;
1419
1420 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001421 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1422 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001424 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001426 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return err;
1428}
1429
1430/*
Ram Pai07b20882005-11-07 17:19:07 -05001431 * recursively change the type of the mountpoint.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001432 * noinline this do_mount helper to save do_mount stack space.
Ram Pai07b20882005-11-07 17:19:07 -05001433 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001434static noinline int do_change_type(struct nameidata *nd, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001435{
Jan Blunck4ac91372008-02-14 19:34:32 -08001436 struct vfsmount *m, *mnt = nd->path.mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001437 int recurse = flag & MS_REC;
1438 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001439 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001440
Miklos Szerediee6f9582007-05-08 00:30:40 -07001441 if (!capable(CAP_SYS_ADMIN))
1442 return -EPERM;
1443
Jan Blunck4ac91372008-02-14 19:34:32 -08001444 if (nd->path.dentry != nd->path.mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001445 return -EINVAL;
1446
1447 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001448 if (type == MS_SHARED) {
1449 err = invent_group_ids(mnt, recurse);
1450 if (err)
1451 goto out_unlock;
1452 }
1453
Ram Pai07b20882005-11-07 17:19:07 -05001454 spin_lock(&vfsmount_lock);
1455 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1456 change_mnt_propagation(m, type);
1457 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001458
1459 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001460 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001461 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001462}
1463
1464/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * do loopback mount.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001466 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001468static noinline int do_loopback(struct nameidata *nd, char *old_name,
1469 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
1471 struct nameidata old_nd;
1472 struct vfsmount *mnt = NULL;
1473 int err = mount_is_safe(nd);
1474 if (err)
1475 return err;
1476 if (!old_name || !*old_name)
1477 return -EINVAL;
1478 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1479 if (err)
1480 return err;
1481
Ram Pai390c6842005-11-07 17:17:51 -05001482 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001484 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
1485 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001486
Jan Blunck4ac91372008-02-14 19:34:32 -08001487 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001488 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Al Viroccd48bc2005-11-07 17:15:04 -05001490 err = -ENOMEM;
1491 if (recurse)
Jan Blunck4ac91372008-02-14 19:34:32 -08001492 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001493 else
Jan Blunck4ac91372008-02-14 19:34:32 -08001494 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001495
1496 if (!mnt)
1497 goto out;
1498
Al Viro8c3ee422008-03-22 18:00:39 -04001499 err = graft_tree(mnt, &nd->path);
Al Viroccd48bc2005-11-07 17:15:04 -05001500 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001501 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001503 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001505 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Al Viroccd48bc2005-11-07 17:15:04 -05001508out:
Ram Pai390c6842005-11-07 17:17:51 -05001509 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001510 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return err;
1512}
1513
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001514static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1515{
1516 int error = 0;
1517 int readonly_request = 0;
1518
1519 if (ms_flags & MS_RDONLY)
1520 readonly_request = 1;
1521 if (readonly_request == __mnt_is_readonly(mnt))
1522 return 0;
1523
1524 if (readonly_request)
1525 error = mnt_make_readonly(mnt);
1526 else
1527 __mnt_unmake_readonly(mnt);
1528 return error;
1529}
1530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531/*
1532 * change filesystem flags. dir should be a physical root of filesystem.
1533 * If you've mounted a non-root directory somewhere and want to do remount
1534 * on it - tough luck.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001535 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001537static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 void *data)
1539{
1540 int err;
Jan Blunck4ac91372008-02-14 19:34:32 -08001541 struct super_block *sb = nd->path.mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543 if (!capable(CAP_SYS_ADMIN))
1544 return -EPERM;
1545
Jan Blunck4ac91372008-02-14 19:34:32 -08001546 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 return -EINVAL;
1548
Jan Blunck4ac91372008-02-14 19:34:32 -08001549 if (nd->path.dentry != nd->path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return -EINVAL;
1551
1552 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001553 if (flags & MS_BIND)
1554 err = change_mount_flags(nd->path.mnt, flags);
1555 else
1556 err = do_remount_sb(sb, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001558 nd->path.mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 up_write(&sb->s_umount);
1560 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001561 security_sb_post_remount(nd->path.mnt, flags, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return err;
1563}
1564
Ram Pai9676f0c2005-11-07 17:21:20 -05001565static inline int tree_contains_unbindable(struct vfsmount *mnt)
1566{
1567 struct vfsmount *p;
1568 for (p = mnt; p; p = next_mnt(p, mnt)) {
1569 if (IS_MNT_UNBINDABLE(p))
1570 return 1;
1571 }
1572 return 0;
1573}
1574
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001575/*
1576 * noinline this do_mount helper to save do_mount stack space.
1577 */
1578static noinline int do_move_mount(struct nameidata *nd, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579{
Al Viro1a390682008-03-21 20:48:19 -04001580 struct nameidata old_nd;
1581 struct path parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 struct vfsmount *p;
1583 int err = 0;
1584 if (!capable(CAP_SYS_ADMIN))
1585 return -EPERM;
1586 if (!old_name || !*old_name)
1587 return -EINVAL;
1588 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1589 if (err)
1590 return err;
1591
Ram Pai390c6842005-11-07 17:17:51 -05001592 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001593 while (d_mountpoint(nd->path.dentry) &&
1594 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 ;
1596 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001597 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 goto out;
1599
1600 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001601 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1602 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 goto out1;
1604
Jan Blunck4ac91372008-02-14 19:34:32 -08001605 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05001606 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001609 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001610 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Jan Blunck4ac91372008-02-14 19:34:32 -08001612 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001613 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
Jan Blunck4ac91372008-02-14 19:34:32 -08001615 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1616 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001617 goto out1;
1618 /*
1619 * Don't move a mount residing in a shared parent.
1620 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001621 if (old_nd.path.mnt->mnt_parent &&
1622 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001623 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001624 /*
1625 * Don't move a mount tree containing unbindable mounts to a destination
1626 * mount which is shared.
1627 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001628 if (IS_MNT_SHARED(nd->path.mnt) &&
1629 tree_contains_unbindable(old_nd.path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001630 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 err = -ELOOP;
Jan Blunck4ac91372008-02-14 19:34:32 -08001632 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1633 if (p == old_nd.path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001634 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Al Viro1a390682008-03-21 20:48:19 -04001636 err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001637 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001638 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
1640 /* if the mount is moved, it should no longer be expire
1641 * automatically */
Jan Blunck4ac91372008-02-14 19:34:32 -08001642 list_del_init(&old_nd.path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643out1:
Jan Blunck4ac91372008-02-14 19:34:32 -08001644 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645out:
Ram Pai390c6842005-11-07 17:17:51 -05001646 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001648 path_put(&parent_path);
Jan Blunck1d957f92008-02-14 19:34:35 -08001649 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 return err;
1651}
1652
1653/*
1654 * create a new mount for userspace and request it to be added into the
1655 * namespace's tree
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001656 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001658static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 int mnt_flags, char *name, void *data)
1660{
1661 struct vfsmount *mnt;
1662
1663 if (!type || !memchr(type, 0, PAGE_SIZE))
1664 return -EINVAL;
1665
1666 /* we need capabilities... */
1667 if (!capable(CAP_SYS_ADMIN))
1668 return -EPERM;
1669
1670 mnt = do_kern_mount(type, flags, name, data);
1671 if (IS_ERR(mnt))
1672 return PTR_ERR(mnt);
1673
1674 return do_add_mount(mnt, nd, mnt_flags, NULL);
1675}
1676
1677/*
1678 * add a mount into a namespace's mount tree
1679 * - provide the option of adding the new mount to an expiration list
1680 */
1681int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1682 int mnt_flags, struct list_head *fslist)
1683{
1684 int err;
1685
Ram Pai390c6842005-11-07 17:17:51 -05001686 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 /* Something was mounted here while we slept */
Jan Blunck4ac91372008-02-14 19:34:32 -08001688 while (d_mountpoint(nd->path.dentry) &&
1689 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 ;
1691 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001692 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 goto unlock;
1694
1695 /* Refuse the same filesystem on the same mount point */
1696 err = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001697 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1698 nd->path.mnt->mnt_root == nd->path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 goto unlock;
1700
1701 err = -EINVAL;
1702 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1703 goto unlock;
1704
1705 newmnt->mnt_flags = mnt_flags;
Al Viro8c3ee422008-03-22 18:00:39 -04001706 if ((err = graft_tree(newmnt, &nd->path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001707 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Al Viro6758f952008-03-22 16:14:30 -04001709 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001710 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001711
Ram Pai390c6842005-11-07 17:17:51 -05001712 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001713 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001716 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 mntput(newmnt);
1718 return err;
1719}
1720
1721EXPORT_SYMBOL_GPL(do_add_mount);
1722
Trond Myklebust5528f9112006-06-09 09:34:17 -04001723/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 * process a list of expirable mountpoints with the intent of discarding any
1725 * mountpoints that aren't in use and haven't been touched since last we came
1726 * here
1727 */
1728void mark_mounts_for_expiry(struct list_head *mounts)
1729{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 struct vfsmount *mnt, *next;
1731 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001732 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 if (list_empty(mounts))
1735 return;
1736
Al Virobcc5c7d2008-03-22 00:21:53 -04001737 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 spin_lock(&vfsmount_lock);
1739
1740 /* extract from the expiration list every vfsmount that matches the
1741 * following criteria:
1742 * - only referenced by its parent vfsmount
1743 * - still marked for expiry (marked on the last call here; marks are
1744 * cleared by mntput())
1745 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001746 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001748 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001750 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001752 while (!list_empty(&graveyard)) {
1753 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1754 touch_mnt_namespace(mnt->mnt_ns);
1755 umount_tree(mnt, 1, &umounts);
1756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001758 up_write(&namespace_sem);
1759
1760 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761}
1762
1763EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1764
1765/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001766 * Ripoff of 'select_parent()'
1767 *
1768 * search the list of submounts for a given mountpoint, and move any
1769 * shrinkable submounts to the 'graveyard' list.
1770 */
1771static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1772{
1773 struct vfsmount *this_parent = parent;
1774 struct list_head *next;
1775 int found = 0;
1776
1777repeat:
1778 next = this_parent->mnt_mounts.next;
1779resume:
1780 while (next != &this_parent->mnt_mounts) {
1781 struct list_head *tmp = next;
1782 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1783
1784 next = tmp->next;
1785 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1786 continue;
1787 /*
1788 * Descend a level if the d_mounts list is non-empty.
1789 */
1790 if (!list_empty(&mnt->mnt_mounts)) {
1791 this_parent = mnt;
1792 goto repeat;
1793 }
1794
1795 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001796 list_move_tail(&mnt->mnt_expire, graveyard);
1797 found++;
1798 }
1799 }
1800 /*
1801 * All done at this level ... ascend and resume the search
1802 */
1803 if (this_parent != parent) {
1804 next = this_parent->mnt_child.next;
1805 this_parent = this_parent->mnt_parent;
1806 goto resume;
1807 }
1808 return found;
1809}
1810
1811/*
1812 * process a list of expirable mountpoints with the intent of discarding any
1813 * submounts of a specific parent mountpoint
1814 */
Al Viroc35038b2008-03-22 00:46:23 -04001815static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001816{
1817 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001818 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001819
Trond Myklebust5528f9112006-06-09 09:34:17 -04001820 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001821 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001822 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001823 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001824 mnt_expire);
1825 touch_mnt_namespace(mnt->mnt_ns);
Al Viroc35038b2008-03-22 00:46:23 -04001826 umount_tree(mnt, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001827 }
1828 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001829}
1830
Trond Myklebust5528f9112006-06-09 09:34:17 -04001831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * Some copy_from_user() implementations do not return the exact number of
1833 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1834 * Note that this function differs from copy_from_user() in that it will oops
1835 * on bad values of `to', rather than returning a short copy.
1836 */
Ram Paib58fed82005-11-07 17:16:09 -05001837static long exact_copy_from_user(void *to, const void __user * from,
1838 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
1840 char *t = to;
1841 const char __user *f = from;
1842 char c;
1843
1844 if (!access_ok(VERIFY_READ, from, n))
1845 return n;
1846
1847 while (n) {
1848 if (__get_user(c, f)) {
1849 memset(t, 0, n);
1850 break;
1851 }
1852 *t++ = c;
1853 f++;
1854 n--;
1855 }
1856 return n;
1857}
1858
Ram Paib58fed82005-11-07 17:16:09 -05001859int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860{
1861 int i;
1862 unsigned long page;
1863 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 *where = 0;
1866 if (!data)
1867 return 0;
1868
1869 if (!(page = __get_free_page(GFP_KERNEL)))
1870 return -ENOMEM;
1871
1872 /* We only care that *some* data at the address the user
1873 * gave us is valid. Just in case, we'll zero
1874 * the remainder of the page.
1875 */
1876 /* copy_from_user cannot cross TASK_SIZE ! */
1877 size = TASK_SIZE - (unsigned long)data;
1878 if (size > PAGE_SIZE)
1879 size = PAGE_SIZE;
1880
1881 i = size - exact_copy_from_user((void *)page, data, size);
1882 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001883 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 return -EFAULT;
1885 }
1886 if (i != PAGE_SIZE)
1887 memset((char *)page + i, 0, PAGE_SIZE - i);
1888 *where = page;
1889 return 0;
1890}
1891
1892/*
1893 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1894 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1895 *
1896 * data is a (void *) that can point to any structure up to
1897 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1898 * information (or be NULL).
1899 *
1900 * Pre-0.97 versions of mount() didn't have a flags word.
1901 * When the flags word was introduced its top half was required
1902 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1903 * Therefore, if this magic number is present, it carries no information
1904 * and must be discarded.
1905 */
Ram Paib58fed82005-11-07 17:16:09 -05001906long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 unsigned long flags, void *data_page)
1908{
1909 struct nameidata nd;
1910 int retval = 0;
1911 int mnt_flags = 0;
1912
1913 /* Discard magic */
1914 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1915 flags &= ~MS_MGC_MSK;
1916
1917 /* Basic sanity checks */
1918
1919 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1920 return -EINVAL;
1921 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1922 return -EINVAL;
1923
1924 if (data_page)
1925 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1926
1927 /* Separate the per-mountpoint flags */
1928 if (flags & MS_NOSUID)
1929 mnt_flags |= MNT_NOSUID;
1930 if (flags & MS_NODEV)
1931 mnt_flags |= MNT_NODEV;
1932 if (flags & MS_NOEXEC)
1933 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001934 if (flags & MS_NOATIME)
1935 mnt_flags |= MNT_NOATIME;
1936 if (flags & MS_NODIRATIME)
1937 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001938 if (flags & MS_RELATIME)
1939 mnt_flags |= MNT_RELATIME;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001940 if (flags & MS_RDONLY)
1941 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001942
1943 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001944 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 /* ... and get the mountpoint */
1947 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1948 if (retval)
1949 return retval;
1950
Al Virob5266eb2008-03-22 17:48:24 -04001951 retval = security_sb_mount(dev_name, &nd.path,
1952 type_page, flags, data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 if (retval)
1954 goto dput_out;
1955
1956 if (flags & MS_REMOUNT)
1957 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1958 data_page);
1959 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001960 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001961 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001962 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 else if (flags & MS_MOVE)
1964 retval = do_move_mount(&nd, dev_name);
1965 else
1966 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1967 dev_name, data_page);
1968dput_out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001969 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 return retval;
1971}
1972
JANAK DESAI741a2952006-02-07 12:59:00 -08001973/*
1974 * Allocate a new namespace structure and populate it with contents
1975 * copied from the namespace of the passed in task structure.
1976 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001977static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001978 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001980 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 struct vfsmount *p, *q;
1983
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001984 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001986 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
1988 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001990 init_waitqueue_head(&new_ns->poll);
1991 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Ram Pai390c6842005-11-07 17:17:51 -05001993 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001995 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001996 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001998 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07002000 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 }
2002 spin_lock(&vfsmount_lock);
2003 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2004 spin_unlock(&vfsmount_lock);
2005
2006 /*
2007 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2008 * as belonging to new namespace. We have already acquired a private
2009 * fs_struct, so tsk->fs->lock is not needed.
2010 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002011 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 q = new_ns->root;
2013 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002014 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002016 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002018 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002020 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002022 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002024 if (p == fs->altroot.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 altrootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002026 fs->altroot.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 }
2028 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002029 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 q = next_mnt(q, new_ns->root);
2031 }
Ram Pai390c6842005-11-07 17:17:51 -05002032 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 if (rootmnt)
2035 mntput(rootmnt);
2036 if (pwdmnt)
2037 mntput(pwdmnt);
2038 if (altrootmnt)
2039 mntput(altrootmnt);
2040
JANAK DESAI741a2952006-02-07 12:59:00 -08002041 return new_ns;
2042}
2043
Eric W. Biederman213dd262007-07-15 23:41:15 -07002044struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002045 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002046{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002047 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002048
Badari Pulavartye3222c42007-05-08 00:25:21 -07002049 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002050 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002051
2052 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002053 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002054
Badari Pulavartye3222c42007-05-08 00:25:21 -07002055 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002056
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002057 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002058 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059}
2060
2061asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
2062 char __user * type, unsigned long flags,
2063 void __user * data)
2064{
2065 int retval;
2066 unsigned long data_page;
2067 unsigned long type_page;
2068 unsigned long dev_page;
2069 char *dir_page;
2070
Ram Paib58fed82005-11-07 17:16:09 -05002071 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 if (retval < 0)
2073 return retval;
2074
2075 dir_page = getname(dir_name);
2076 retval = PTR_ERR(dir_page);
2077 if (IS_ERR(dir_page))
2078 goto out1;
2079
Ram Paib58fed82005-11-07 17:16:09 -05002080 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 if (retval < 0)
2082 goto out2;
2083
Ram Paib58fed82005-11-07 17:16:09 -05002084 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 if (retval < 0)
2086 goto out3;
2087
2088 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05002089 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2090 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 unlock_kernel();
2092 free_page(data_page);
2093
2094out3:
2095 free_page(dev_page);
2096out2:
2097 putname(dir_page);
2098out1:
2099 free_page(type_page);
2100 return retval;
2101}
2102
2103/*
2104 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2105 * It can block. Requires the big lock held.
2106 */
Jan Blunckac748a02008-02-14 19:34:39 -08002107void set_fs_root(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
Jan Blunck6ac08c32008-02-14 19:34:38 -08002109 struct path old_root;
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 write_lock(&fs->lock);
2112 old_root = fs->root;
Jan Blunckac748a02008-02-14 19:34:39 -08002113 fs->root = *path;
2114 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 write_unlock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002116 if (old_root.dentry)
2117 path_put(&old_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118}
2119
2120/*
2121 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2122 * It can block. Requires the big lock held.
2123 */
Jan Blunckac748a02008-02-14 19:34:39 -08002124void set_fs_pwd(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Jan Blunck6ac08c32008-02-14 19:34:38 -08002126 struct path old_pwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
2128 write_lock(&fs->lock);
2129 old_pwd = fs->pwd;
Jan Blunckac748a02008-02-14 19:34:39 -08002130 fs->pwd = *path;
2131 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 write_unlock(&fs->lock);
2133
Jan Blunck6ac08c32008-02-14 19:34:38 -08002134 if (old_pwd.dentry)
2135 path_put(&old_pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136}
2137
Al Viro1a390682008-03-21 20:48:19 -04002138static void chroot_fs_refs(struct path *old_root, struct path *new_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139{
2140 struct task_struct *g, *p;
2141 struct fs_struct *fs;
2142
2143 read_lock(&tasklist_lock);
2144 do_each_thread(g, p) {
2145 task_lock(p);
2146 fs = p->fs;
2147 if (fs) {
2148 atomic_inc(&fs->count);
2149 task_unlock(p);
Al Viro1a390682008-03-21 20:48:19 -04002150 if (fs->root.dentry == old_root->dentry
2151 && fs->root.mnt == old_root->mnt)
2152 set_fs_root(fs, new_root);
2153 if (fs->pwd.dentry == old_root->dentry
2154 && fs->pwd.mnt == old_root->mnt)
2155 set_fs_pwd(fs, new_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 put_fs_struct(fs);
2157 } else
2158 task_unlock(p);
2159 } while_each_thread(g, p);
2160 read_unlock(&tasklist_lock);
2161}
2162
2163/*
2164 * pivot_root Semantics:
2165 * Moves the root file system of the current process to the directory put_old,
2166 * makes new_root as the new root file system of the current process, and sets
2167 * root/cwd of all processes which had them on the current root to new_root.
2168 *
2169 * Restrictions:
2170 * The new_root and put_old must be directories, and must not be on the
2171 * same file system as the current process root. The put_old must be
2172 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2173 * pointed to by put_old must yield the same directory as new_root. No other
2174 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2175 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002176 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2177 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2178 * in this situation.
2179 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 * Notes:
2181 * - we don't move root/cwd if they are not at the root (reason: if something
2182 * cared enough to change them, it's probably wrong to force them elsewhere)
2183 * - it's okay to pick a root that isn't the root of a file system, e.g.
2184 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2185 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2186 * first.
2187 */
Ram Paib58fed82005-11-07 17:16:09 -05002188asmlinkage long sys_pivot_root(const char __user * new_root,
2189 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
2191 struct vfsmount *tmp;
Al Viro8c3ee422008-03-22 18:00:39 -04002192 struct nameidata new_nd, old_nd;
2193 struct path parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 int error;
2195
2196 if (!capable(CAP_SYS_ADMIN))
2197 return -EPERM;
2198
Ram Paib58fed82005-11-07 17:16:09 -05002199 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
2200 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 if (error)
2202 goto out0;
2203 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08002204 if (!check_mnt(new_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 goto out1;
2206
Ram Paib58fed82005-11-07 17:16:09 -05002207 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (error)
2209 goto out1;
2210
Al Virob5266eb2008-03-22 17:48:24 -04002211 error = security_sb_pivotroot(&old_nd.path, &new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 if (error) {
Jan Blunck1d957f92008-02-14 19:34:35 -08002213 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 goto out1;
2215 }
2216
2217 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002218 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002219 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002221 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08002222 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08002224 if (IS_MNT_SHARED(old_nd.path.mnt) ||
2225 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002226 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002227 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002228 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 goto out2;
2230 error = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08002231 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08002233 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08002235 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 goto out2;
2237 error = -EBUSY;
Al Viro8c3ee422008-03-22 18:00:39 -04002238 if (new_nd.path.mnt == root.mnt ||
2239 old_nd.path.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 goto out2; /* loop, on the same file system */
2241 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002242 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002244 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002245 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002246 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08002248 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002249 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002250 /* make sure we can reach put_old from new_root */
2251 tmp = old_nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -08002253 if (tmp != new_nd.path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 for (;;) {
2255 if (tmp->mnt_parent == tmp)
2256 goto out3; /* already mounted on put_old */
Jan Blunck4ac91372008-02-14 19:34:32 -08002257 if (tmp->mnt_parent == new_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 break;
2259 tmp = tmp->mnt_parent;
2260 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002261 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002263 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 goto out3;
Al Viro1a390682008-03-21 20:48:19 -04002265 detach_mnt(new_nd.path.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002266 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002267 /* mount old root on put_old */
Al Viro8c3ee422008-03-22 18:00:39 -04002268 attach_mnt(root.mnt, &old_nd.path);
Jan Blunck4ac91372008-02-14 19:34:32 -08002269 /* mount new_root on / */
2270 attach_mnt(new_nd.path.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002271 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 spin_unlock(&vfsmount_lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002273 chroot_fs_refs(&root, &new_nd.path);
2274 security_sb_post_pivotroot(&root, &new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002276 path_put(&root_parent);
2277 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278out2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002279 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002280 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002281 path_put(&root);
Jan Blunck1d957f92008-02-14 19:34:35 -08002282 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283out1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002284 path_put(&new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return error;
2287out3:
2288 spin_unlock(&vfsmount_lock);
2289 goto out2;
2290}
2291
2292static void __init init_mount_tree(void)
2293{
2294 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002295 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002296 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
2298 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2299 if (IS_ERR(mnt))
2300 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002301 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2302 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002304 atomic_set(&ns->count, 1);
2305 INIT_LIST_HEAD(&ns->list);
2306 init_waitqueue_head(&ns->poll);
2307 ns->event = 0;
2308 list_add(&mnt->mnt_list, &ns->list);
2309 ns->root = mnt;
2310 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002312 init_task.nsproxy->mnt_ns = ns;
2313 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
Jan Blunckac748a02008-02-14 19:34:39 -08002315 root.mnt = ns->root;
2316 root.dentry = ns->root->mnt_root;
2317
2318 set_fs_pwd(current->fs, &root);
2319 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320}
2321
Denis Cheng74bf17c2007-10-16 23:26:30 -07002322void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002324 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002325 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Ram Pai390c6842005-11-07 17:17:51 -05002327 init_rwsem(&namespace_sem);
2328
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002330 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Ram Paib58fed82005-11-07 17:16:09 -05002332 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
2334 if (!mount_hashtable)
2335 panic("Failed to allocate mount hash table\n");
2336
Eric Dumazet13f14b42008-02-06 01:37:57 -08002337 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
Eric Dumazet13f14b42008-02-06 01:37:57 -08002339 for (u = 0; u < HASH_SIZE; u++)
2340 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002342 err = sysfs_init();
2343 if (err)
2344 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2345 __FUNCTION__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002346 fs_kobj = kobject_create_and_add("fs", NULL);
2347 if (!fs_kobj)
2348 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 init_rootfs();
2350 init_mount_tree();
2351}
2352
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002353void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002355 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05002356 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002357 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07002358 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002359 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05002361 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002363 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002364 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002365 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366}