blob: 0953a3a6d45e6a8af96dc5a7c6a5a74719bc5ba2 [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>
Nick Piggin99b7db72010-08-18 04:37:39 +100014#include <linux/spinlock.h>
15#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070017#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#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>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040026#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/security.h>
28#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080030#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010031#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040032#include <linux/fs_struct.h>
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -050033#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
35#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050036#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazet13f14b42008-02-06 01:37:57 -080039#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
Al Viro5addc5d2005-11-07 17:15:49 -050042static 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);
Nick Piggin99b7db72010-08-18 04:37:39 +100045static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040046static int mnt_id_start = 0;
47static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Eric Dumazetfa3536c2006-03-26 01:37:24 -080049static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050051static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060054struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080056
Nick Piggin99b7db72010-08-18 04:37:39 +100057/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
Ram Paib58fed82005-11-07 17:16:09 -050069 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080071 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073}
74
Dave Hansen3d733632008-02-15 14:37:59 -080075#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
Nick Piggin99b7db72010-08-18 04:37:39 +100077/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010081static int mnt_alloc_id(struct vfsmount *mnt)
82{
83 int res;
84
85retry:
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100087 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040088 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
89 if (!res)
90 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100091 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010092 if (res == -EAGAIN)
93 goto retry;
94
95 return res;
96}
97
98static void mnt_free_id(struct vfsmount *mnt)
99{
Al Virof21f6222009-06-24 03:12:00 -0400100 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000101 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
104 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000105 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100106}
107
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100108/*
109 * Allocate a new peer group ID
110 *
111 * mnt_group_ida is protected by namespace_sem
112 */
113static int mnt_alloc_group_id(struct vfsmount *mnt)
114{
Al Virof21f6222009-06-24 03:12:00 -0400115 int res;
116
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118 return -ENOMEM;
119
Al Virof21f6222009-06-24 03:12:00 -0400120 res = ida_get_new_above(&mnt_group_ida,
121 mnt_group_start,
122 &mnt->mnt_group_id);
123 if (!res)
124 mnt_group_start = mnt->mnt_group_id + 1;
125
126 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127}
128
129/*
130 * Release a peer group ID
131 */
132void mnt_release_group_id(struct vfsmount *mnt)
133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100138 mnt->mnt_group_id = 0;
139}
140
Nick Pigginb3e19d92011-01-07 17:50:11 +1100141/*
142 * vfsmount lock must be held for read
143 */
144static inline void mnt_add_count(struct vfsmount *mnt, int n)
145{
146#ifdef CONFIG_SMP
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
148#else
149 preempt_disable();
150 mnt->mnt_count += n;
151 preempt_enable();
152#endif
153}
154
Nick Pigginb3e19d92011-01-07 17:50:11 +1100155/*
156 * vfsmount lock must be held for write
157 */
158unsigned int mnt_get_count(struct vfsmount *mnt)
159{
160#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500161 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100162 int cpu;
163
164 for_each_possible_cpu(cpu) {
165 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
166 }
167
168 return count;
169#else
170 return mnt->mnt_count;
171#endif
172}
173
Al Viro9d412a42011-03-17 22:08:28 -0400174static struct vfsmount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800176 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100178 int err;
179
180 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800181 if (err)
182 goto out_free_cache;
183
184 if (name) {
185 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
186 if (!mnt->mnt_devname)
187 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100188 }
189
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190#ifdef CONFIG_SMP
191 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
192 if (!mnt->mnt_pcp)
193 goto out_free_devname;
194
Al Virof03c6592011-01-14 22:30:21 -0500195 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100196#else
197 mnt->mnt_count = 1;
198 mnt->mnt_writers = 0;
199#endif
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 INIT_LIST_HEAD(&mnt->mnt_hash);
202 INIT_LIST_HEAD(&mnt->mnt_child);
203 INIT_LIST_HEAD(&mnt->mnt_mounts);
204 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700205 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500206 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500207 INIT_LIST_HEAD(&mnt->mnt_slave_list);
208 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500209#ifdef CONFIG_FSNOTIFY
210 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
211#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
213 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800214
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000215#ifdef CONFIG_SMP
216out_free_devname:
217 kfree(mnt->mnt_devname);
218#endif
Li Zefan88b38782008-07-21 18:06:36 +0800219out_free_id:
220 mnt_free_id(mnt);
221out_free_cache:
222 kmem_cache_free(mnt_cache, mnt);
223 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
Dave Hansen83660252008-02-15 14:37:30 -0800226/*
227 * Most r/o checks on a fs are for operations that take
228 * discrete amounts of time, like a write() or unlink().
229 * We must keep track of when those operations start
230 * (for permission checks) and when they end, so that
231 * we can determine when writes are able to occur to
232 * a filesystem.
233 */
Dave Hansen3d733632008-02-15 14:37:59 -0800234/*
235 * __mnt_is_readonly: check whether a mount is read-only
236 * @mnt: the mount to check for its write status
237 *
238 * This shouldn't be used directly ouside of the VFS.
239 * It does not guarantee that the filesystem will stay
240 * r/w, just that it is right *now*. This can not and
241 * should not be used in place of IS_RDONLY(inode).
242 * mnt_want/drop_write() will _keep_ the filesystem
243 * r/w.
244 */
245int __mnt_is_readonly(struct vfsmount *mnt)
246{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800247 if (mnt->mnt_flags & MNT_READONLY)
248 return 1;
249 if (mnt->mnt_sb->s_flags & MS_RDONLY)
250 return 1;
251 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800252}
253EXPORT_SYMBOL_GPL(__mnt_is_readonly);
254
Nick Pigginc6653a82011-01-07 17:50:10 +1100255static inline void mnt_inc_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800256{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000257#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100258 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000259#else
260 mnt->mnt_writers++;
261#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800262}
Dave Hansen3d733632008-02-15 14:37:59 -0800263
Nick Pigginc6653a82011-01-07 17:50:10 +1100264static inline void mnt_dec_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800265{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000266#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100267 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000268#else
269 mnt->mnt_writers--;
270#endif
271}
272
Nick Pigginc6653a82011-01-07 17:50:10 +1100273static unsigned int mnt_get_writers(struct vfsmount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000274{
275#ifdef CONFIG_SMP
276 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800277 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800278
279 for_each_possible_cpu(cpu) {
Nick Pigginb3e19d92011-01-07 17:50:11 +1100280 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800281 }
Dave Hansen3d733632008-02-15 14:37:59 -0800282
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000283 return count;
284#else
285 return mnt->mnt_writers;
286#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800287}
288
289/*
290 * Most r/o checks on a fs are for operations that take
291 * discrete amounts of time, like a write() or unlink().
292 * We must keep track of when those operations start
293 * (for permission checks) and when they end, so that
294 * we can determine when writes are able to occur to
295 * a filesystem.
296 */
Dave Hansen83660252008-02-15 14:37:30 -0800297/**
298 * mnt_want_write - get write access to a mount
299 * @mnt: the mount on which to take a write
300 *
301 * This tells the low-level filesystem that a write is
302 * about to be performed to it, and makes sure that
303 * writes are allowed before returning success. When
304 * the write operation is finished, mnt_drop_write()
305 * must be called. This is effectively a refcount.
306 */
307int mnt_want_write(struct vfsmount *mnt)
308{
Dave Hansen3d733632008-02-15 14:37:59 -0800309 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800310
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000311 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100312 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000313 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100314 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000315 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
316 * incremented count after it has set MNT_WRITE_HOLD.
317 */
318 smp_mb();
319 while (mnt->mnt_flags & MNT_WRITE_HOLD)
320 cpu_relax();
321 /*
322 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
323 * be set to match its requirements. So we must not load that until
324 * MNT_WRITE_HOLD is cleared.
325 */
326 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800327 if (__mnt_is_readonly(mnt)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100328 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800329 ret = -EROFS;
330 goto out;
331 }
Dave Hansen3d733632008-02-15 14:37:59 -0800332out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000333 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800334 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800335}
336EXPORT_SYMBOL_GPL(mnt_want_write);
337
338/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000339 * mnt_clone_write - get write access to a mount
340 * @mnt: the mount on which to take a write
341 *
342 * This is effectively like mnt_want_write, except
343 * it must only be used to take an extra write reference
344 * on a mountpoint that we already know has a write reference
345 * on it. This allows some optimisation.
346 *
347 * After finished, mnt_drop_write must be called as usual to
348 * drop the reference.
349 */
350int mnt_clone_write(struct vfsmount *mnt)
351{
352 /* superblock may be r/o */
353 if (__mnt_is_readonly(mnt))
354 return -EROFS;
355 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100356 mnt_inc_writers(mnt);
npiggin@suse.de96029c42009-04-26 20:25:55 +1000357 preempt_enable();
358 return 0;
359}
360EXPORT_SYMBOL_GPL(mnt_clone_write);
361
362/**
363 * mnt_want_write_file - get write access to a file's mount
364 * @file: the file who's mount on which to take a write
365 *
366 * This is like mnt_want_write, but it takes a file and can
367 * do some optimisations if the file is open for write already
368 */
369int mnt_want_write_file(struct file *file)
370{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700371 struct inode *inode = file->f_dentry->d_inode;
372 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000373 return mnt_want_write(file->f_path.mnt);
374 else
375 return mnt_clone_write(file->f_path.mnt);
376}
377EXPORT_SYMBOL_GPL(mnt_want_write_file);
378
379/**
Dave Hansen83660252008-02-15 14:37:30 -0800380 * mnt_drop_write - give up write access to a mount
381 * @mnt: the mount on which to give up write access
382 *
383 * Tells the low-level filesystem that we are done
384 * performing writes to it. Must be matched with
385 * mnt_want_write() call above.
386 */
387void mnt_drop_write(struct vfsmount *mnt)
388{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000389 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100390 mnt_dec_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000391 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800392}
393EXPORT_SYMBOL_GPL(mnt_drop_write);
394
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800395static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800396{
Dave Hansen3d733632008-02-15 14:37:59 -0800397 int ret = 0;
398
Nick Piggin99b7db72010-08-18 04:37:39 +1000399 br_write_lock(vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000400 mnt->mnt_flags |= MNT_WRITE_HOLD;
401 /*
402 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
403 * should be visible before we do.
404 */
405 smp_mb();
406
407 /*
408 * With writers on hold, if this value is zero, then there are
409 * definitely no active writers (although held writers may subsequently
410 * increment the count, they'll have to wait, and decrement it after
411 * seeing MNT_READONLY).
412 *
413 * It is OK to have counter incremented on one CPU and decremented on
414 * another: the sum will add up correctly. The danger would be when we
415 * sum up each counter, if we read a counter before it is incremented,
416 * but then read another CPU's count which it has been subsequently
417 * decremented from -- we would see more decrements than we should.
418 * MNT_WRITE_HOLD protects against this scenario, because
419 * mnt_want_write first increments count, then smp_mb, then spins on
420 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
421 * we're counting up here.
422 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100423 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000424 ret = -EBUSY;
425 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800426 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000427 /*
428 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
429 * that become unheld will see MNT_READONLY.
430 */
431 smp_wmb();
432 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Nick Piggin99b7db72010-08-18 04:37:39 +1000433 br_write_unlock(vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800434 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800435}
Dave Hansen83660252008-02-15 14:37:30 -0800436
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800437static void __mnt_unmake_readonly(struct vfsmount *mnt)
438{
Nick Piggin99b7db72010-08-18 04:37:39 +1000439 br_write_lock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800440 mnt->mnt_flags &= ~MNT_READONLY;
Nick Piggin99b7db72010-08-18 04:37:39 +1000441 br_write_unlock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800442}
443
Al Viro9d412a42011-03-17 22:08:28 -0400444static void free_vfsmnt(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
446 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100447 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000448#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100449 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000450#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 kmem_cache_free(mnt_cache, mnt);
452}
453
454/*
Ram Paia05964f2005-11-07 17:20:17 -0500455 * find the first or last mount at @dentry on vfsmount @mnt depending on
456 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000457 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
Ram Paia05964f2005-11-07 17:20:17 -0500459struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
460 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Ram Paib58fed82005-11-07 17:16:09 -0500462 struct list_head *head = mount_hashtable + hash(mnt, dentry);
463 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 struct vfsmount *p, *found = NULL;
465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500467 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 p = NULL;
469 if (tmp == head)
470 break;
471 p = list_entry(tmp, struct vfsmount, mnt_hash);
472 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500473 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 break;
475 }
476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 return found;
478}
479
Ram Paia05964f2005-11-07 17:20:17 -0500480/*
481 * lookup_mnt increments the ref count before returning
482 * the vfsmount struct.
483 */
Al Viro1c755af2009-04-18 14:06:57 -0400484struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500485{
486 struct vfsmount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000487
488 br_read_lock(vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400489 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500490 mntget(child_mnt);
Nick Piggin99b7db72010-08-18 04:37:39 +1000491 br_read_unlock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500492 return child_mnt;
493}
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495static inline int check_mnt(struct vfsmount *mnt)
496{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800497 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499
Nick Piggin99b7db72010-08-18 04:37:39 +1000500/*
501 * vfsmount lock must be held for write
502 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800503static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500504{
505 if (ns) {
506 ns->event = ++event;
507 wake_up_interruptible(&ns->poll);
508 }
509}
510
Nick Piggin99b7db72010-08-18 04:37:39 +1000511/*
512 * vfsmount lock must be held for write
513 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800514static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500515{
516 if (ns && ns->event != event) {
517 ns->event = event;
518 wake_up_interruptible(&ns->poll);
519 }
520}
521
Nick Piggin99b7db72010-08-18 04:37:39 +1000522/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100523 * Clear dentry's mounted state if it has no remaining mounts.
524 * vfsmount_lock must be held for write.
525 */
526static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
527{
528 unsigned u;
529
530 for (u = 0; u < HASH_SIZE; u++) {
531 struct vfsmount *p;
532
533 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
534 if (p->mnt_mountpoint == dentry)
535 return;
536 }
537 }
538 spin_lock(&dentry->d_lock);
539 dentry->d_flags &= ~DCACHE_MOUNTED;
540 spin_unlock(&dentry->d_lock);
541}
542
543/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000544 * vfsmount lock must be held for write
545 */
Al Viro1a390682008-03-21 20:48:19 -0400546static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Al Viro1a390682008-03-21 20:48:19 -0400548 old_path->dentry = mnt->mnt_mountpoint;
549 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 mnt->mnt_parent = mnt;
551 mnt->mnt_mountpoint = mnt->mnt_root;
552 list_del_init(&mnt->mnt_child);
553 list_del_init(&mnt->mnt_hash);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100554 dentry_reset_mounted(old_path->mnt, old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
556
Nick Piggin99b7db72010-08-18 04:37:39 +1000557/*
558 * vfsmount lock must be held for write
559 */
Ram Paib90fa9a2005-11-07 17:19:50 -0500560void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
561 struct vfsmount *child_mnt)
562{
563 child_mnt->mnt_parent = mntget(mnt);
564 child_mnt->mnt_mountpoint = dget(dentry);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100565 spin_lock(&dentry->d_lock);
566 dentry->d_flags |= DCACHE_MOUNTED;
567 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500568}
569
Nick Piggin99b7db72010-08-18 04:37:39 +1000570/*
571 * vfsmount lock must be held for write
572 */
Al Viro1a390682008-03-21 20:48:19 -0400573static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Al Viro1a390682008-03-21 20:48:19 -0400575 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500576 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400577 hash(path->mnt, path->dentry));
578 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500579}
580
Al Viro7e3d0eb02011-01-16 16:32:11 -0500581static inline void __mnt_make_longterm(struct vfsmount *mnt)
582{
583#ifdef CONFIG_SMP
584 atomic_inc(&mnt->mnt_longterm);
585#endif
586}
587
588/* needs vfsmount lock for write */
589static inline void __mnt_make_shortterm(struct vfsmount *mnt)
590{
591#ifdef CONFIG_SMP
592 atomic_dec(&mnt->mnt_longterm);
593#endif
594}
595
Ram Paib90fa9a2005-11-07 17:19:50 -0500596/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000597 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500598 */
599static void commit_tree(struct vfsmount *mnt)
600{
601 struct vfsmount *parent = mnt->mnt_parent;
602 struct vfsmount *m;
603 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800604 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500605
606 BUG_ON(parent == mnt);
607
608 list_add_tail(&head, &mnt->mnt_list);
Al Virof03c6592011-01-14 22:30:21 -0500609 list_for_each_entry(m, &head, mnt_list) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800610 m->mnt_ns = n;
Al Viro7e3d0eb02011-01-16 16:32:11 -0500611 __mnt_make_longterm(m);
Al Virof03c6592011-01-14 22:30:21 -0500612 }
613
Ram Paib90fa9a2005-11-07 17:19:50 -0500614 list_splice(&head, n->list.prev);
615
616 list_add_tail(&mnt->mnt_hash, mount_hashtable +
617 hash(parent, mnt->mnt_mountpoint));
618 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800619 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
623{
624 struct list_head *next = p->mnt_mounts.next;
625 if (next == &p->mnt_mounts) {
626 while (1) {
627 if (p == root)
628 return NULL;
629 next = p->mnt_child.next;
630 if (next != &p->mnt_parent->mnt_mounts)
631 break;
632 p = p->mnt_parent;
633 }
634 }
635 return list_entry(next, struct vfsmount, mnt_child);
636}
637
Ram Pai9676f0c2005-11-07 17:21:20 -0500638static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
639{
640 struct list_head *prev = p->mnt_mounts.prev;
641 while (prev != &p->mnt_mounts) {
642 p = list_entry(prev, struct vfsmount, mnt_child);
643 prev = p->mnt_mounts.prev;
644 }
645 return p;
646}
647
Al Viro9d412a42011-03-17 22:08:28 -0400648struct vfsmount *
649vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
650{
651 struct vfsmount *mnt;
652 struct dentry *root;
653
654 if (!type)
655 return ERR_PTR(-ENODEV);
656
657 mnt = alloc_vfsmnt(name);
658 if (!mnt)
659 return ERR_PTR(-ENOMEM);
660
661 if (flags & MS_KERNMOUNT)
662 mnt->mnt_flags = MNT_INTERNAL;
663
664 root = mount_fs(type, flags, name, data);
665 if (IS_ERR(root)) {
666 free_vfsmnt(mnt);
667 return ERR_CAST(root);
668 }
669
670 mnt->mnt_root = root;
671 mnt->mnt_sb = root->d_sb;
672 mnt->mnt_mountpoint = mnt->mnt_root;
673 mnt->mnt_parent = mnt;
674 return mnt;
675}
676EXPORT_SYMBOL_GPL(vfs_kern_mount);
677
Ram Pai36341f62005-11-07 17:17:22 -0500678static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
679 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 struct super_block *sb = old->mnt_sb;
682 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
683
684 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100685 if (flag & (CL_SLAVE | CL_PRIVATE))
686 mnt->mnt_group_id = 0; /* not a peer of original */
687 else
688 mnt->mnt_group_id = old->mnt_group_id;
689
690 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
691 int err = mnt_alloc_group_id(mnt);
692 if (err)
693 goto out_free;
694 }
695
Miklos Szeredibe1a16a2010-10-05 12:31:09 +0200696 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 atomic_inc(&sb->s_active);
698 mnt->mnt_sb = sb;
699 mnt->mnt_root = dget(root);
700 mnt->mnt_mountpoint = mnt->mnt_root;
701 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500702
Ram Pai5afe0022005-11-07 17:21:01 -0500703 if (flag & CL_SLAVE) {
704 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
705 mnt->mnt_master = old;
706 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400707 } else if (!(flag & CL_PRIVATE)) {
Al Viro796a6b52010-01-16 13:28:47 -0500708 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500709 list_add(&mnt->mnt_share, &old->mnt_share);
710 if (IS_MNT_SLAVE(old))
711 list_add(&mnt->mnt_slave, &old->mnt_slave);
712 mnt->mnt_master = old->mnt_master;
713 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500714 if (flag & CL_MAKE_SHARED)
715 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
717 /* stick the duplicate mount on the same expiry list
718 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500719 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500720 if (!list_empty(&old->mnt_expire))
721 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500722 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
724 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100725
726 out_free:
727 free_vfsmnt(mnt);
728 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
Nick Pigginb3e19d92011-01-07 17:50:11 +1100731static inline void mntfree(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 struct super_block *sb = mnt->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100734
Dave Hansen3d733632008-02-15 14:37:59 -0800735 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800736 * This probably indicates that somebody messed
737 * up a mnt_want/drop_write() pair. If this
738 * happens, the filesystem was probably unable
739 * to make r/w->r/o transitions.
740 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000741 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100742 * The locking used to deal with mnt_count decrement provides barriers,
743 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000744 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100745 WARN_ON(mnt_get_writers(mnt));
Andreas Gruenbacherca9c7262009-12-17 21:24:27 -0500746 fsnotify_vfsmount_delete(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 dput(mnt->mnt_root);
748 free_vfsmnt(mnt);
749 deactivate_super(sb);
750}
751
Al Virof03c6592011-01-14 22:30:21 -0500752static void mntput_no_expire(struct vfsmount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500753{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100754put_again:
Al Virof03c6592011-01-14 22:30:21 -0500755#ifdef CONFIG_SMP
756 br_read_lock(vfsmount_lock);
757 if (likely(atomic_read(&mnt->mnt_longterm))) {
Al Viroaa9c0e02011-11-23 19:04:52 -0500758 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100759 br_read_unlock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500760 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100761 }
Al Virof03c6592011-01-14 22:30:21 -0500762 br_read_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100763
764 br_write_lock(vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500765 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100766 if (mnt_get_count(mnt)) {
767 br_write_unlock(vfsmount_lock);
768 return;
769 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100770#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500771 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100772 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000773 return;
774 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500775#endif
Nick Pigginb3e19d92011-01-07 17:50:11 +1100776 if (unlikely(mnt->mnt_pinned)) {
777 mnt_add_count(mnt, mnt->mnt_pinned + 1);
778 mnt->mnt_pinned = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +1000779 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100780 acct_auto_close_mnt(mnt);
781 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500782 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000783 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100784 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500785}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100786
787void mntput(struct vfsmount *mnt)
788{
789 if (mnt) {
790 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
791 if (unlikely(mnt->mnt_expiry_mark))
792 mnt->mnt_expiry_mark = 0;
Al Virof03c6592011-01-14 22:30:21 -0500793 mntput_no_expire(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100794 }
795}
796EXPORT_SYMBOL(mntput);
797
798struct vfsmount *mntget(struct vfsmount *mnt)
799{
800 if (mnt)
Al Viroaa9c0e02011-11-23 19:04:52 -0500801 mnt_add_count(mnt, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100802 return mnt;
803}
804EXPORT_SYMBOL(mntget);
805
Al Viro7b7b1ac2005-11-07 17:13:39 -0500806void mnt_pin(struct vfsmount *mnt)
807{
Nick Piggin99b7db72010-08-18 04:37:39 +1000808 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500809 mnt->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000810 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500811}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500812EXPORT_SYMBOL(mnt_pin);
813
814void mnt_unpin(struct vfsmount *mnt)
815{
Nick Piggin99b7db72010-08-18 04:37:39 +1000816 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500817 if (mnt->mnt_pinned) {
Al Viroaa9c0e02011-11-23 19:04:52 -0500818 mnt_add_count(mnt, 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500819 mnt->mnt_pinned--;
820 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000821 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500822}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500823EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800825static inline void mangle(struct seq_file *m, const char *s)
826{
827 seq_escape(m, s, " \t\n\\");
828}
829
830/*
831 * Simple .show_options callback for filesystems which don't want to
832 * implement more complex mount option showing.
833 *
834 * See also save_mount_options().
835 */
836int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
837{
Al Viro2a32ceb2009-05-08 16:05:57 -0400838 const char *options;
839
840 rcu_read_lock();
841 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800842
843 if (options != NULL && options[0]) {
844 seq_putc(m, ',');
845 mangle(m, options);
846 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400847 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800848
849 return 0;
850}
851EXPORT_SYMBOL(generic_show_options);
852
853/*
854 * If filesystem uses generic_show_options(), this function should be
855 * called from the fill_super() callback.
856 *
857 * The .remount_fs callback usually needs to be handled in a special
858 * way, to make sure, that previous options are not overwritten if the
859 * remount fails.
860 *
861 * Also note, that if the filesystem's .remount_fs function doesn't
862 * reset all options to their default value, but changes only newly
863 * given options, then the displayed options will not reflect reality
864 * any more.
865 */
866void save_mount_options(struct super_block *sb, char *options)
867{
Al Viro2a32ceb2009-05-08 16:05:57 -0400868 BUG_ON(sb->s_options);
869 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800870}
871EXPORT_SYMBOL(save_mount_options);
872
Al Viro2a32ceb2009-05-08 16:05:57 -0400873void replace_mount_options(struct super_block *sb, char *options)
874{
875 char *old = sb->s_options;
876 rcu_assign_pointer(sb->s_options, options);
877 if (old) {
878 synchronize_rcu();
879 kfree(old);
880 }
881}
882EXPORT_SYMBOL(replace_mount_options);
883
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100884#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/* iterator */
886static void *m_start(struct seq_file *m, loff_t *pos)
887{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100888 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Ram Pai390c6842005-11-07 17:17:51 -0500890 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100891 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
894static void *m_next(struct seq_file *m, void *v, loff_t *pos)
895{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100896 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700897
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100898 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
901static void m_stop(struct seq_file *m, void *v)
902{
Ram Pai390c6842005-11-07 17:17:51 -0500903 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
Al Viro9f5596a2010-02-05 00:40:25 -0500906int mnt_had_events(struct proc_mounts *p)
907{
908 struct mnt_namespace *ns = p->ns;
909 int res = 0;
910
Nick Piggin99b7db72010-08-18 04:37:39 +1000911 br_read_lock(vfsmount_lock);
Kay Sieversf1514632011-07-12 20:48:39 +0200912 if (p->m.poll_event != ns->event) {
913 p->m.poll_event = ns->event;
Al Viro9f5596a2010-02-05 00:40:25 -0500914 res = 1;
915 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000916 br_read_unlock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500917
918 return res;
919}
920
Ram Pai2d4d4862008-03-27 13:06:25 +0100921struct proc_fs_info {
922 int flag;
923 const char *str;
924};
925
Eric Paris2069f452008-07-04 09:47:13 +1000926static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
Ram Pai2d4d4862008-03-27 13:06:25 +0100928 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 { MS_SYNCHRONOUS, ",sync" },
930 { MS_DIRSYNC, ",dirsync" },
931 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 { 0, NULL }
933 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100934 const struct proc_fs_info *fs_infop;
935
936 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
937 if (sb->s_flags & fs_infop->flag)
938 seq_puts(m, fs_infop->str);
939 }
Eric Paris2069f452008-07-04 09:47:13 +1000940
941 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100942}
943
944static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
945{
946 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 { MNT_NOSUID, ",nosuid" },
948 { MNT_NODEV, ",nodev" },
949 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800950 { MNT_NOATIME, ",noatime" },
951 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800952 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 { 0, NULL }
954 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100955 const struct proc_fs_info *fs_infop;
956
957 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
958 if (mnt->mnt_flags & fs_infop->flag)
959 seq_puts(m, fs_infop->str);
960 }
961}
962
963static void show_type(struct seq_file *m, struct super_block *sb)
964{
965 mangle(m, sb->s_type->name);
966 if (sb->s_subtype && sb->s_subtype[0]) {
967 seq_putc(m, '.');
968 mangle(m, sb->s_subtype);
969 }
970}
971
972static int show_vfsmnt(struct seq_file *m, void *v)
973{
974 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
975 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800976 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Al Viroc7f404b2011-03-16 06:59:40 -0400978 if (mnt->mnt_sb->s_op->show_devname) {
979 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
980 if (err)
981 goto out;
982 } else {
983 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800986 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100988 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800989 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000990 err = show_sb_opts(m, mnt->mnt_sb);
991 if (err)
992 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100993 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (mnt->mnt_sb->s_op->show_options)
995 err = mnt->mnt_sb->s_op->show_options(m, mnt);
996 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000997out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 return err;
999}
1000
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001001const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 .start = m_start,
1003 .next = m_next,
1004 .stop = m_stop,
1005 .show = show_vfsmnt
1006};
1007
Ram Pai2d4d4862008-03-27 13:06:25 +01001008static int show_mountinfo(struct seq_file *m, void *v)
1009{
1010 struct proc_mounts *p = m->private;
1011 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1012 struct super_block *sb = mnt->mnt_sb;
1013 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1014 struct path root = p->root;
1015 int err = 0;
1016
1017 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1018 MAJOR(sb->s_dev), MINOR(sb->s_dev));
Al Viroc7f404b2011-03-16 06:59:40 -04001019 if (sb->s_op->show_path)
1020 err = sb->s_op->show_path(m, mnt);
1021 else
1022 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1023 if (err)
1024 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001025 seq_putc(m, ' ');
Al Viro02125a82011-12-05 08:43:34 -05001026
1027 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1028 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1029 if (err)
1030 goto out;
1031
Ram Pai2d4d4862008-03-27 13:06:25 +01001032 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1033 show_mnt_opts(m, mnt);
1034
1035 /* Tagged fields ("foo:X" or "bar") */
1036 if (IS_MNT_SHARED(mnt))
1037 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +01001038 if (IS_MNT_SLAVE(mnt)) {
1039 int master = mnt->mnt_master->mnt_group_id;
1040 int dom = get_dominating_id(mnt, &p->root);
1041 seq_printf(m, " master:%i", master);
1042 if (dom && dom != master)
1043 seq_printf(m, " propagate_from:%i", dom);
1044 }
Ram Pai2d4d4862008-03-27 13:06:25 +01001045 if (IS_MNT_UNBINDABLE(mnt))
1046 seq_puts(m, " unbindable");
1047
1048 /* Filesystem specific data */
1049 seq_puts(m, " - ");
1050 show_type(m, sb);
1051 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001052 if (sb->s_op->show_devname)
1053 err = sb->s_op->show_devname(m, mnt);
1054 else
1055 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1056 if (err)
1057 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001058 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001059 err = show_sb_opts(m, sb);
1060 if (err)
1061 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001062 if (sb->s_op->show_options)
1063 err = sb->s_op->show_options(m, mnt);
1064 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +10001065out:
Ram Pai2d4d4862008-03-27 13:06:25 +01001066 return err;
1067}
1068
1069const struct seq_operations mountinfo_op = {
1070 .start = m_start,
1071 .next = m_next,
1072 .stop = m_stop,
1073 .show = show_mountinfo,
1074};
1075
Chuck Leverb4629fe2006-03-20 13:44:12 -05001076static int show_vfsstat(struct seq_file *m, void *v)
1077{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001078 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -08001079 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -05001080 int err = 0;
1081
1082 /* device */
Al Viroc7f404b2011-03-16 06:59:40 -04001083 if (mnt->mnt_sb->s_op->show_devname) {
Bryan Schumakera877ee02011-10-07 13:41:15 -04001084 seq_puts(m, "device ");
Al Viroc7f404b2011-03-16 06:59:40 -04001085 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1086 } else {
1087 if (mnt->mnt_devname) {
1088 seq_puts(m, "device ");
1089 mangle(m, mnt->mnt_devname);
1090 } else
1091 seq_puts(m, "no device");
1092 }
Chuck Leverb4629fe2006-03-20 13:44:12 -05001093
1094 /* mount point */
1095 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08001096 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -05001097 seq_putc(m, ' ');
1098
1099 /* file system type */
1100 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +01001101 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001102
1103 /* optional statistics */
1104 if (mnt->mnt_sb->s_op->show_stats) {
1105 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001106 if (!err)
1107 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001108 }
1109
1110 seq_putc(m, '\n');
1111 return err;
1112}
1113
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001114const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -05001115 .start = m_start,
1116 .next = m_next,
1117 .stop = m_stop,
1118 .show = show_vfsstat,
1119};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001120#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122/**
1123 * may_umount_tree - check if a mount tree is busy
1124 * @mnt: root of mount tree
1125 *
1126 * This is called to check if a tree of mounts has any
1127 * open files, pwds, chroots or sub mounts that are
1128 * busy.
1129 */
1130int may_umount_tree(struct vfsmount *mnt)
1131{
Ram Pai36341f62005-11-07 17:17:22 -05001132 int actual_refs = 0;
1133 int minimum_refs = 0;
1134 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Nick Pigginb3e19d92011-01-07 17:50:11 +11001136 /* write lock needed for mnt_get_count */
1137 br_write_lock(vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -05001138 for (p = mnt; p; p = next_mnt(p, mnt)) {
Nick Pigginb3e19d92011-01-07 17:50:11 +11001139 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
Nick Pigginb3e19d92011-01-07 17:50:11 +11001142 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001145 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Ian Kente3474a82006-03-27 01:14:51 -08001147 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148}
1149
1150EXPORT_SYMBOL(may_umount_tree);
1151
1152/**
1153 * may_umount - check if a mount point is busy
1154 * @mnt: root of mount
1155 *
1156 * This is called to check if a mount point has any
1157 * open files, pwds, chroots or sub mounts. If the
1158 * mount has sub mounts this will return busy
1159 * regardless of whether the sub mounts are busy.
1160 *
1161 * Doesn't take quota and stuff into account. IOW, in some cases it will
1162 * give false negatives. The main reason why it's here is that we need
1163 * a non-destructive way to look for easily umountable filesystems.
1164 */
1165int may_umount(struct vfsmount *mnt)
1166{
Ian Kente3474a82006-03-27 01:14:51 -08001167 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001168 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001169 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001170 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -08001171 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001172 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001173 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001174 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
1176
1177EXPORT_SYMBOL(may_umount);
1178
Ram Paib90fa9a2005-11-07 17:19:50 -05001179void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
Ram Pai70fbcdf2005-11-07 17:17:04 -05001181 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001182 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001183 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001184 list_del_init(&mnt->mnt_hash);
Al Virob2dba1a2011-11-23 19:26:23 -05001185 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001186 struct dentry *dentry;
1187 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001188
1189 br_write_lock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001190 dentry = mnt->mnt_mountpoint;
1191 m = mnt->mnt_parent;
1192 mnt->mnt_mountpoint = mnt->mnt_root;
1193 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001194 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001195 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001196 dput(dentry);
1197 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
Al Virof03c6592011-01-14 22:30:21 -05001199 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001200 }
1201}
1202
Nick Piggin99b7db72010-08-18 04:37:39 +10001203/*
1204 * vfsmount lock must be held for write
1205 * namespace_sem must be held for write
1206 */
Ram Paia05964f2005-11-07 17:20:17 -05001207void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001208{
Al Viro7b8a53f2011-01-15 20:08:44 -05001209 LIST_HEAD(tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001210 struct vfsmount *p;
1211
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001212 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro7b8a53f2011-01-15 20:08:44 -05001213 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001214
Ram Paia05964f2005-11-07 17:20:17 -05001215 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001216 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001217
Al Viro7b8a53f2011-01-15 20:08:44 -05001218 list_for_each_entry(p, &tmp_list, mnt_hash) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001219 list_del_init(&p->mnt_expire);
1220 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001221 __touch_mnt_namespace(p->mnt_ns);
1222 p->mnt_ns = NULL;
Al Viro7e3d0eb02011-01-16 16:32:11 -05001223 __mnt_make_shortterm(p);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001224 list_del_init(&p->mnt_child);
Al Virob2dba1a2011-11-23 19:26:23 -05001225 if (mnt_has_parent(p)) {
Al Viro7c4b93d2008-03-21 23:59:49 -04001226 p->mnt_parent->mnt_ghosts++;
Nick Piggin5f57cbc2011-01-07 17:49:54 +11001227 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001228 }
Ram Paia05964f2005-11-07 17:20:17 -05001229 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001231 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
1233
Al Viroc35038b2008-03-22 00:46:23 -04001234static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236static int do_umount(struct vfsmount *mnt, int flags)
1237{
Ram Paib58fed82005-11-07 17:16:09 -05001238 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001240 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 retval = security_sb_umount(mnt, flags);
1243 if (retval)
1244 return retval;
1245
1246 /*
1247 * Allow userspace to request a mountpoint be expired rather than
1248 * unmounting unconditionally. Unmount only happens if:
1249 * (1) the mark is already set (the mark is cleared by mntput())
1250 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1251 */
1252 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001253 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 flags & (MNT_FORCE | MNT_DETACH))
1255 return -EINVAL;
1256
Nick Pigginb3e19d92011-01-07 17:50:11 +11001257 /*
1258 * probably don't strictly need the lock here if we examined
1259 * all race cases, but it's a slowpath.
1260 */
1261 br_write_lock(vfsmount_lock);
1262 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001263 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001265 }
1266 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
1268 if (!xchg(&mnt->mnt_expiry_mark, 1))
1269 return -EAGAIN;
1270 }
1271
1272 /*
1273 * If we may have to abort operations to get out of this
1274 * mount, and they will themselves hold resources we must
1275 * allow the fs to do things. In the Unix tradition of
1276 * 'Gee thats tricky lets do it in userspace' the umount_begin
1277 * might fail to complete on the first run through as other tasks
1278 * must return, and the like. Thats for the mount program to worry
1279 * about for the moment.
1280 */
1281
Al Viro42faad92008-04-24 07:21:56 -04001282 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001283 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 /*
1287 * No sense to grab the lock for this test, but test itself looks
1288 * somewhat bogus. Suggestions for better replacement?
1289 * Ho-hum... In principle, we might treat that as umount + switch
1290 * to rootfs. GC would eventually take care of the old vfsmount.
1291 * Actually it makes sense, especially if rootfs would contain a
1292 * /reboot - static binary that would close all descriptors and
1293 * call reboot(9). Then init(8) could umount root and exec /reboot.
1294 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001295 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 /*
1297 * Special case for "unmounting" root ...
1298 * we just try to remount it readonly.
1299 */
1300 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001301 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 up_write(&sb->s_umount);
1304 return retval;
1305 }
1306
Ram Pai390c6842005-11-07 17:17:51 -05001307 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001308 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001309 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Al Viroc35038b2008-03-22 00:46:23 -04001311 if (!(flags & MNT_DETACH))
1312 shrink_submounts(mnt, &umount_list);
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001315 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001317 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 retval = 0;
1319 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001320 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001321 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001322 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return retval;
1324}
1325
1326/*
1327 * Now umount can handle mount points as well as block devices.
1328 * This is important for filesystems which use unnamed block devices.
1329 *
1330 * We now support a flag for forced unmount like the other 'big iron'
1331 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1332 */
1333
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001334SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
Al Viro2d8f3032008-07-22 09:59:21 -04001336 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001338 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001340 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1341 return -EINVAL;
1342
1343 if (!(flags & UMOUNT_NOFOLLOW))
1344 lookup_flags |= LOOKUP_FOLLOW;
1345
1346 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (retval)
1348 goto out;
1349 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001350 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001352 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 goto dput_and_out;
1354
1355 retval = -EPERM;
1356 if (!capable(CAP_SYS_ADMIN))
1357 goto dput_and_out;
1358
Al Viro2d8f3032008-07-22 09:59:21 -04001359 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001361 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001362 dput(path.dentry);
1363 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364out:
1365 return retval;
1366}
1367
1368#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1369
1370/*
Ram Paib58fed82005-11-07 17:16:09 -05001371 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001373SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
Ram Paib58fed82005-11-07 17:16:09 -05001375 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
1378#endif
1379
Al Viro2d92ab32008-08-02 00:51:11 -04001380static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381{
1382 if (capable(CAP_SYS_ADMIN))
1383 return 0;
1384 return -EPERM;
1385#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001386 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001388 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001389 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 return -EPERM;
1391 }
Al Viro2d92ab32008-08-02 00:51:11 -04001392 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return -EPERM;
1394 return 0;
1395#endif
1396}
1397
Ram Paib90fa9a2005-11-07 17:19:50 -05001398struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001399 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001402 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Ram Pai9676f0c2005-11-07 17:21:20 -05001404 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1405 return NULL;
1406
Ram Pai36341f62005-11-07 17:17:22 -05001407 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (!q)
1409 goto Enomem;
1410 q->mnt_mountpoint = mnt->mnt_mountpoint;
1411
1412 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001413 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001414 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 continue;
1416
1417 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001418 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1419 s = skip_mnt_tree(s);
1420 continue;
1421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 while (p != s->mnt_parent) {
1423 p = p->mnt_parent;
1424 q = q->mnt_parent;
1425 }
1426 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001427 path.mnt = q;
1428 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001429 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (!q)
1431 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001432 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001434 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001435 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
1437 }
1438 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001439Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001441 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001442 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001443 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001444 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001445 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447 return NULL;
1448}
1449
Al Viro589ff872009-04-18 03:28:19 -04001450struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001451{
1452 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001453 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001454 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001455 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001456 return tree;
1457}
1458
1459void drop_collected_mounts(struct vfsmount *mnt)
1460{
1461 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001462 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001463 br_write_lock(vfsmount_lock);
Al Viro8aec0802007-06-07 12:20:32 -04001464 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001465 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001466 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001467 release_mounts(&umount_list);
1468}
1469
Al Viro1f707132010-01-30 22:51:25 -05001470int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1471 struct vfsmount *root)
1472{
1473 struct vfsmount *mnt;
1474 int res = f(root, arg);
1475 if (res)
1476 return res;
1477 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1478 res = f(mnt, arg);
1479 if (res)
1480 return res;
1481 }
1482 return 0;
1483}
1484
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001485static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1486{
1487 struct vfsmount *p;
1488
1489 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1490 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1491 mnt_release_group_id(p);
1492 }
1493}
1494
1495static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1496{
1497 struct vfsmount *p;
1498
1499 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1500 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1501 int err = mnt_alloc_group_id(p);
1502 if (err) {
1503 cleanup_group_ids(mnt, p);
1504 return err;
1505 }
1506 }
1507 }
1508
1509 return 0;
1510}
1511
Ram Paib90fa9a2005-11-07 17:19:50 -05001512/*
1513 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001514 * @nd : place the mount tree @source_mnt is attached
1515 * @parent_nd : if non-null, detach the source_mnt from its parent and
1516 * store the parent mount and mountpoint dentry.
1517 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001518 *
1519 * NOTE: in the table below explains the semantics when a source mount
1520 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001521 * ---------------------------------------------------------------------------
1522 * | BIND MOUNT OPERATION |
1523 * |**************************************************************************
1524 * | source-->| shared | private | slave | unbindable |
1525 * | dest | | | | |
1526 * | | | | | | |
1527 * | v | | | | |
1528 * |**************************************************************************
1529 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1530 * | | | | | |
1531 * |non-shared| shared (+) | private | slave (*) | invalid |
1532 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001533 * A bind operation clones the source mount and mounts the clone on the
1534 * destination mount.
1535 *
1536 * (++) the cloned mount is propagated to all the mounts in the propagation
1537 * tree of the destination mount and the cloned mount is added to
1538 * the peer group of the source mount.
1539 * (+) the cloned mount is created under the destination mount and is marked
1540 * as shared. The cloned mount is added to the peer group of the source
1541 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001542 * (+++) the mount is propagated to all the mounts in the propagation tree
1543 * of the destination mount and the cloned mount is made slave
1544 * of the same master as that of the source mount. The cloned mount
1545 * is marked as 'shared and slave'.
1546 * (*) the cloned mount is made a slave of the same master as that of the
1547 * source mount.
1548 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001549 * ---------------------------------------------------------------------------
1550 * | MOVE MOUNT OPERATION |
1551 * |**************************************************************************
1552 * | source-->| shared | private | slave | unbindable |
1553 * | dest | | | | |
1554 * | | | | | | |
1555 * | v | | | | |
1556 * |**************************************************************************
1557 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1558 * | | | | | |
1559 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1560 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001561 *
1562 * (+) the mount is moved to the destination. And is then propagated to
1563 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001564 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001565 * (+++) the mount is moved to the destination and is then propagated to
1566 * all the mounts belonging to the destination mount's propagation tree.
1567 * the mount is marked as 'shared and slave'.
1568 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001569 *
1570 * if the source mount is a tree, the operations explained above is
1571 * applied to each mount in the tree.
1572 * Must be called without spinlocks held, since this function can sleep
1573 * in allocations.
1574 */
1575static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001576 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001577{
1578 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001579 struct vfsmount *dest_mnt = path->mnt;
1580 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001581 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001582 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001583
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001584 if (IS_MNT_SHARED(dest_mnt)) {
1585 err = invent_group_ids(source_mnt, true);
1586 if (err)
1587 goto out;
1588 }
1589 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1590 if (err)
1591 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001592
Nick Piggin99b7db72010-08-18 04:37:39 +10001593 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001594
Ram Paib90fa9a2005-11-07 17:19:50 -05001595 if (IS_MNT_SHARED(dest_mnt)) {
1596 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1597 set_mnt_shared(p);
1598 }
Al Viro1a390682008-03-21 20:48:19 -04001599 if (parent_path) {
1600 detach_mnt(source_mnt, parent_path);
1601 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001602 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001603 } else {
1604 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1605 commit_tree(source_mnt);
1606 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001607
1608 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1609 list_del_init(&child->mnt_hash);
1610 commit_tree(child);
1611 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001612 br_write_unlock(vfsmount_lock);
1613
Ram Paib90fa9a2005-11-07 17:19:50 -05001614 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001615
1616 out_cleanup_ids:
1617 if (IS_MNT_SHARED(dest_mnt))
1618 cleanup_group_ids(source_mnt, NULL);
1619 out:
1620 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001621}
1622
Al Virob12cea92011-03-18 08:55:38 -04001623static int lock_mount(struct path *path)
1624{
1625 struct vfsmount *mnt;
1626retry:
1627 mutex_lock(&path->dentry->d_inode->i_mutex);
1628 if (unlikely(cant_mount(path->dentry))) {
1629 mutex_unlock(&path->dentry->d_inode->i_mutex);
1630 return -ENOENT;
1631 }
1632 down_write(&namespace_sem);
1633 mnt = lookup_mnt(path);
1634 if (likely(!mnt))
1635 return 0;
1636 up_write(&namespace_sem);
1637 mutex_unlock(&path->dentry->d_inode->i_mutex);
1638 path_put(path);
1639 path->mnt = mnt;
1640 path->dentry = dget(mnt->mnt_root);
1641 goto retry;
1642}
1643
1644static void unlock_mount(struct path *path)
1645{
1646 up_write(&namespace_sem);
1647 mutex_unlock(&path->dentry->d_inode->i_mutex);
1648}
1649
Al Viro8c3ee422008-03-22 18:00:39 -04001650static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1653 return -EINVAL;
1654
Al Viro8c3ee422008-03-22 18:00:39 -04001655 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1657 return -ENOTDIR;
1658
Al Virob12cea92011-03-18 08:55:38 -04001659 if (d_unlinked(path->dentry))
1660 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Al Virob12cea92011-03-18 08:55:38 -04001662 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
1664
1665/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001666 * Sanity check the flags to change_mnt_propagation.
1667 */
1668
1669static int flags_to_propagation_type(int flags)
1670{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001671 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001672
1673 /* Fail if any non-propagation flags are set */
1674 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1675 return 0;
1676 /* Only one propagation flag should be set */
1677 if (!is_power_of_2(type))
1678 return 0;
1679 return type;
1680}
1681
1682/*
Ram Pai07b20882005-11-07 17:19:07 -05001683 * recursively change the type of the mountpoint.
1684 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001685static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001686{
Al Viro2d92ab32008-08-02 00:51:11 -04001687 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001688 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001689 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001690 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001691
Miklos Szerediee6f9582007-05-08 00:30:40 -07001692 if (!capable(CAP_SYS_ADMIN))
1693 return -EPERM;
1694
Al Viro2d92ab32008-08-02 00:51:11 -04001695 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001696 return -EINVAL;
1697
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001698 type = flags_to_propagation_type(flag);
1699 if (!type)
1700 return -EINVAL;
1701
Ram Pai07b20882005-11-07 17:19:07 -05001702 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001703 if (type == MS_SHARED) {
1704 err = invent_group_ids(mnt, recurse);
1705 if (err)
1706 goto out_unlock;
1707 }
1708
Nick Piggin99b7db72010-08-18 04:37:39 +10001709 br_write_lock(vfsmount_lock);
Ram Pai07b20882005-11-07 17:19:07 -05001710 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1711 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001712 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001713
1714 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001715 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001716 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001717}
1718
1719/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * do loopback mount.
1721 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001722static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001723 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
Al Virob12cea92011-03-18 08:55:38 -04001725 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001726 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001728 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (err)
1730 return err;
1731 if (!old_name || !*old_name)
1732 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001733 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 if (err)
1735 return err;
1736
Al Virob12cea92011-03-18 08:55:38 -04001737 err = lock_mount(path);
1738 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001739 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001740
Al Virob12cea92011-03-18 08:55:38 -04001741 err = -EINVAL;
1742 if (IS_MNT_UNBINDABLE(old_path.mnt))
1743 goto out2;
1744
Al Viro2d92ab32008-08-02 00:51:11 -04001745 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Virob12cea92011-03-18 08:55:38 -04001746 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Al Viroccd48bc2005-11-07 17:15:04 -05001748 err = -ENOMEM;
1749 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001750 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001751 else
Al Viro2d92ab32008-08-02 00:51:11 -04001752 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001753
1754 if (!mnt)
Al Virob12cea92011-03-18 08:55:38 -04001755 goto out2;
Al Viroccd48bc2005-11-07 17:15:04 -05001756
Al Viro2d92ab32008-08-02 00:51:11 -04001757 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001758 if (err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001759 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001760 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001761 br_write_unlock(vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001762 }
Al Virob12cea92011-03-18 08:55:38 -04001763out2:
1764 unlock_mount(path);
1765 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001766out:
Al Viro2d92ab32008-08-02 00:51:11 -04001767 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 return err;
1769}
1770
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001771static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1772{
1773 int error = 0;
1774 int readonly_request = 0;
1775
1776 if (ms_flags & MS_RDONLY)
1777 readonly_request = 1;
1778 if (readonly_request == __mnt_is_readonly(mnt))
1779 return 0;
1780
1781 if (readonly_request)
1782 error = mnt_make_readonly(mnt);
1783 else
1784 __mnt_unmake_readonly(mnt);
1785 return error;
1786}
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788/*
1789 * change filesystem flags. dir should be a physical root of filesystem.
1790 * If you've mounted a non-root directory somewhere and want to do remount
1791 * on it - tough luck.
1792 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001793static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 void *data)
1795{
1796 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001797 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
1799 if (!capable(CAP_SYS_ADMIN))
1800 return -EPERM;
1801
Al Viro2d92ab32008-08-02 00:51:11 -04001802 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return -EINVAL;
1804
Al Viro2d92ab32008-08-02 00:51:11 -04001805 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 return -EINVAL;
1807
Eric Parisff36fe22011-03-03 16:09:14 -05001808 err = security_sb_remount(sb, data);
1809 if (err)
1810 return err;
1811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001813 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001814 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001815 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001816 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001817 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001818 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001819 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001820 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001821 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001824 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001825 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001826 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001827 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 return err;
1830}
1831
Ram Pai9676f0c2005-11-07 17:21:20 -05001832static inline int tree_contains_unbindable(struct vfsmount *mnt)
1833{
1834 struct vfsmount *p;
1835 for (p = mnt; p; p = next_mnt(p, mnt)) {
1836 if (IS_MNT_UNBINDABLE(p))
1837 return 1;
1838 }
1839 return 0;
1840}
1841
Al Viro0a0d8a42008-08-02 00:55:27 -04001842static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843{
Al Viro2d92ab32008-08-02 00:51:11 -04001844 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 struct vfsmount *p;
1846 int err = 0;
1847 if (!capable(CAP_SYS_ADMIN))
1848 return -EPERM;
1849 if (!old_name || !*old_name)
1850 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001851 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (err)
1853 return err;
1854
Al Virob12cea92011-03-18 08:55:38 -04001855 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001856 if (err < 0)
1857 goto out;
1858
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001860 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 goto out1;
1862
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001863 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001864 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
1866 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001867 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001868 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
Al Virob2dba1a2011-11-23 19:26:23 -05001870 if (!mnt_has_parent(old_path.mnt))
Ram Pai21444402005-11-07 17:20:03 -05001871 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Al Viro2d92ab32008-08-02 00:51:11 -04001873 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1874 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001875 goto out1;
1876 /*
1877 * Don't move a mount residing in a shared parent.
1878 */
Al Viroafac7cb2011-11-23 19:34:49 -05001879 if (IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001880 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001881 /*
1882 * Don't move a mount tree containing unbindable mounts to a destination
1883 * mount which is shared.
1884 */
Al Viro2d92ab32008-08-02 00:51:11 -04001885 if (IS_MNT_SHARED(path->mnt) &&
1886 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001887 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 err = -ELOOP;
Al Virob2dba1a2011-11-23 19:26:23 -05001889 for (p = path->mnt; mnt_has_parent(p); p = p->mnt_parent)
Al Viro2d92ab32008-08-02 00:51:11 -04001890 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001891 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Al Viro2d92ab32008-08-02 00:51:11 -04001893 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001894 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001895 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 /* if the mount is moved, it should no longer be expire
1898 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001899 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900out1:
Al Virob12cea92011-03-18 08:55:38 -04001901 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001904 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001905 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 return err;
1907}
1908
Al Viro9d412a42011-03-17 22:08:28 -04001909static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1910{
1911 int err;
1912 const char *subtype = strchr(fstype, '.');
1913 if (subtype) {
1914 subtype++;
1915 err = -EINVAL;
1916 if (!subtype[0])
1917 goto err;
1918 } else
1919 subtype = "";
1920
1921 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1922 err = -ENOMEM;
1923 if (!mnt->mnt_sb->s_subtype)
1924 goto err;
1925 return mnt;
1926
1927 err:
1928 mntput(mnt);
1929 return ERR_PTR(err);
1930}
1931
1932struct vfsmount *
1933do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1934{
1935 struct file_system_type *type = get_fs_type(fstype);
1936 struct vfsmount *mnt;
1937 if (!type)
1938 return ERR_PTR(-ENODEV);
1939 mnt = vfs_kern_mount(type, flags, name, data);
1940 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1941 !mnt->mnt_sb->s_subtype)
1942 mnt = fs_set_subtype(mnt, fstype);
1943 put_filesystem(type);
1944 return mnt;
1945}
1946EXPORT_SYMBOL_GPL(do_kern_mount);
1947
1948/*
1949 * add a mount into a namespace's mount tree
1950 */
1951static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1952{
1953 int err;
1954
1955 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1956
Al Virob12cea92011-03-18 08:55:38 -04001957 err = lock_mount(path);
1958 if (err)
1959 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001960
1961 err = -EINVAL;
1962 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1963 goto unlock;
1964
1965 /* Refuse the same filesystem on the same mount point */
1966 err = -EBUSY;
1967 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1968 path->mnt->mnt_root == path->dentry)
1969 goto unlock;
1970
1971 err = -EINVAL;
1972 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1973 goto unlock;
1974
1975 newmnt->mnt_flags = mnt_flags;
1976 err = graft_tree(newmnt, path);
1977
1978unlock:
Al Virob12cea92011-03-18 08:55:38 -04001979 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04001980 return err;
1981}
Al Virob1e75df2011-01-17 01:47:59 -05001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983/*
1984 * create a new mount for userspace and request it to be added into the
1985 * namespace's tree
1986 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001987static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 int mnt_flags, char *name, void *data)
1989{
1990 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001991 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Vegard Nossumeca6f532009-09-18 13:05:45 -07001993 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return -EINVAL;
1995
1996 /* we need capabilities... */
1997 if (!capable(CAP_SYS_ADMIN))
1998 return -EPERM;
1999
2000 mnt = do_kern_mount(type, flags, name, data);
2001 if (IS_ERR(mnt))
2002 return PTR_ERR(mnt);
2003
Al Viro15f9a3f2011-01-17 01:41:58 -05002004 err = do_add_mount(mnt, path, mnt_flags);
2005 if (err)
2006 mntput(mnt);
2007 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008}
2009
Al Viro19a167a2011-01-17 01:35:23 -05002010int finish_automount(struct vfsmount *m, struct path *path)
2011{
2012 int err;
2013 /* The new mount record should have at least 2 refs to prevent it being
2014 * expired before we get a chance to add it
2015 */
2016 BUG_ON(mnt_get_count(m) < 2);
2017
2018 if (m->mnt_sb == path->mnt->mnt_sb &&
2019 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002020 err = -ELOOP;
2021 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002022 }
2023
Al Viro19a167a2011-01-17 01:35:23 -05002024 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002025 if (!err)
2026 return 0;
2027fail:
2028 /* remove m from any expiration list it may be on */
2029 if (!list_empty(&m->mnt_expire)) {
2030 down_write(&namespace_sem);
2031 br_write_lock(vfsmount_lock);
2032 list_del_init(&m->mnt_expire);
2033 br_write_unlock(vfsmount_lock);
2034 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05002035 }
Al Virob1e75df2011-01-17 01:47:59 -05002036 mntput(m);
2037 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002038 return err;
2039}
2040
David Howellsea5b7782011-01-14 19:10:03 +00002041/**
2042 * mnt_set_expiry - Put a mount on an expiration list
2043 * @mnt: The mount to list.
2044 * @expiry_list: The list to add the mount to.
2045 */
2046void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2047{
2048 down_write(&namespace_sem);
2049 br_write_lock(vfsmount_lock);
2050
2051 list_add_tail(&mnt->mnt_expire, expiry_list);
2052
2053 br_write_unlock(vfsmount_lock);
2054 up_write(&namespace_sem);
2055}
2056EXPORT_SYMBOL(mnt_set_expiry);
2057
2058/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 * process a list of expirable mountpoints with the intent of discarding any
2060 * mountpoints that aren't in use and haven't been touched since last we came
2061 * here
2062 */
2063void mark_mounts_for_expiry(struct list_head *mounts)
2064{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 struct vfsmount *mnt, *next;
2066 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002067 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
2069 if (list_empty(mounts))
2070 return;
2071
Al Virobcc5c7d2008-03-22 00:21:53 -04002072 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002073 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075 /* extract from the expiration list every vfsmount that matches the
2076 * following criteria:
2077 * - only referenced by its parent vfsmount
2078 * - still marked for expiry (marked on the last call here; marks are
2079 * cleared by mntput())
2080 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002081 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04002083 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002085 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002087 while (!list_empty(&graveyard)) {
2088 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2089 touch_mnt_namespace(mnt->mnt_ns);
2090 umount_tree(mnt, 1, &umounts);
2091 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002092 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002093 up_write(&namespace_sem);
2094
2095 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096}
2097
2098EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2099
2100/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002101 * Ripoff of 'select_parent()'
2102 *
2103 * search the list of submounts for a given mountpoint, and move any
2104 * shrinkable submounts to the 'graveyard' list.
2105 */
2106static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2107{
2108 struct vfsmount *this_parent = parent;
2109 struct list_head *next;
2110 int found = 0;
2111
2112repeat:
2113 next = this_parent->mnt_mounts.next;
2114resume:
2115 while (next != &this_parent->mnt_mounts) {
2116 struct list_head *tmp = next;
2117 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2118
2119 next = tmp->next;
2120 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2121 continue;
2122 /*
2123 * Descend a level if the d_mounts list is non-empty.
2124 */
2125 if (!list_empty(&mnt->mnt_mounts)) {
2126 this_parent = mnt;
2127 goto repeat;
2128 }
2129
2130 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002131 list_move_tail(&mnt->mnt_expire, graveyard);
2132 found++;
2133 }
2134 }
2135 /*
2136 * All done at this level ... ascend and resume the search
2137 */
2138 if (this_parent != parent) {
2139 next = this_parent->mnt_child.next;
2140 this_parent = this_parent->mnt_parent;
2141 goto resume;
2142 }
2143 return found;
2144}
2145
2146/*
2147 * process a list of expirable mountpoints with the intent of discarding any
2148 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002149 *
2150 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002151 */
Al Viroc35038b2008-03-22 00:46:23 -04002152static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002153{
2154 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04002155 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002156
Trond Myklebust5528f9112006-06-09 09:34:17 -04002157 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002158 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002159 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04002160 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04002161 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002162 touch_mnt_namespace(m->mnt_ns);
2163 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002164 }
2165 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002166}
2167
Trond Myklebust5528f9112006-06-09 09:34:17 -04002168/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 * Some copy_from_user() implementations do not return the exact number of
2170 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2171 * Note that this function differs from copy_from_user() in that it will oops
2172 * on bad values of `to', rather than returning a short copy.
2173 */
Ram Paib58fed82005-11-07 17:16:09 -05002174static long exact_copy_from_user(void *to, const void __user * from,
2175 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176{
2177 char *t = to;
2178 const char __user *f = from;
2179 char c;
2180
2181 if (!access_ok(VERIFY_READ, from, n))
2182 return n;
2183
2184 while (n) {
2185 if (__get_user(c, f)) {
2186 memset(t, 0, n);
2187 break;
2188 }
2189 *t++ = c;
2190 f++;
2191 n--;
2192 }
2193 return n;
2194}
2195
Ram Paib58fed82005-11-07 17:16:09 -05002196int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197{
2198 int i;
2199 unsigned long page;
2200 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 *where = 0;
2203 if (!data)
2204 return 0;
2205
2206 if (!(page = __get_free_page(GFP_KERNEL)))
2207 return -ENOMEM;
2208
2209 /* We only care that *some* data at the address the user
2210 * gave us is valid. Just in case, we'll zero
2211 * the remainder of the page.
2212 */
2213 /* copy_from_user cannot cross TASK_SIZE ! */
2214 size = TASK_SIZE - (unsigned long)data;
2215 if (size > PAGE_SIZE)
2216 size = PAGE_SIZE;
2217
2218 i = size - exact_copy_from_user((void *)page, data, size);
2219 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002220 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 return -EFAULT;
2222 }
2223 if (i != PAGE_SIZE)
2224 memset((char *)page + i, 0, PAGE_SIZE - i);
2225 *where = page;
2226 return 0;
2227}
2228
Vegard Nossumeca6f532009-09-18 13:05:45 -07002229int copy_mount_string(const void __user *data, char **where)
2230{
2231 char *tmp;
2232
2233 if (!data) {
2234 *where = NULL;
2235 return 0;
2236 }
2237
2238 tmp = strndup_user(data, PAGE_SIZE);
2239 if (IS_ERR(tmp))
2240 return PTR_ERR(tmp);
2241
2242 *where = tmp;
2243 return 0;
2244}
2245
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246/*
2247 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2248 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2249 *
2250 * data is a (void *) that can point to any structure up to
2251 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2252 * information (or be NULL).
2253 *
2254 * Pre-0.97 versions of mount() didn't have a flags word.
2255 * When the flags word was introduced its top half was required
2256 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2257 * Therefore, if this magic number is present, it carries no information
2258 * and must be discarded.
2259 */
Ram Paib58fed82005-11-07 17:16:09 -05002260long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 unsigned long flags, void *data_page)
2262{
Al Viro2d92ab32008-08-02 00:51:11 -04002263 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 int retval = 0;
2265 int mnt_flags = 0;
2266
2267 /* Discard magic */
2268 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2269 flags &= ~MS_MGC_MSK;
2270
2271 /* Basic sanity checks */
2272
2273 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2274 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 if (data_page)
2277 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2278
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002279 /* ... and get the mountpoint */
2280 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2281 if (retval)
2282 return retval;
2283
2284 retval = security_sb_mount(dev_name, &path,
2285 type_page, flags, data_page);
2286 if (retval)
2287 goto dput_out;
2288
Andi Kleen613cbe32009-04-19 18:40:43 +02002289 /* Default to relatime unless overriden */
2290 if (!(flags & MS_NOATIME))
2291 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 /* Separate the per-mountpoint flags */
2294 if (flags & MS_NOSUID)
2295 mnt_flags |= MNT_NOSUID;
2296 if (flags & MS_NODEV)
2297 mnt_flags |= MNT_NODEV;
2298 if (flags & MS_NOEXEC)
2299 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002300 if (flags & MS_NOATIME)
2301 mnt_flags |= MNT_NOATIME;
2302 if (flags & MS_NODIRATIME)
2303 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002304 if (flags & MS_STRICTATIME)
2305 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002306 if (flags & MS_RDONLY)
2307 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002308
Al Viro7a4dec52010-08-09 12:05:43 -04002309 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002310 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2311 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002314 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 data_page);
2316 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002317 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002318 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002319 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002321 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 else
Al Viro2d92ab32008-08-02 00:51:11 -04002323 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 dev_name, data_page);
2325dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002326 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 return retval;
2328}
2329
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002330static struct mnt_namespace *alloc_mnt_ns(void)
2331{
2332 struct mnt_namespace *new_ns;
2333
2334 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2335 if (!new_ns)
2336 return ERR_PTR(-ENOMEM);
2337 atomic_set(&new_ns->count, 1);
2338 new_ns->root = NULL;
2339 INIT_LIST_HEAD(&new_ns->list);
2340 init_waitqueue_head(&new_ns->poll);
2341 new_ns->event = 0;
2342 return new_ns;
2343}
2344
Al Virof03c6592011-01-14 22:30:21 -05002345void mnt_make_longterm(struct vfsmount *mnt)
2346{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002347 __mnt_make_longterm(mnt);
Al Virof03c6592011-01-14 22:30:21 -05002348}
2349
2350void mnt_make_shortterm(struct vfsmount *mnt)
2351{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002352#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -05002353 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2354 return;
2355 br_write_lock(vfsmount_lock);
2356 atomic_dec(&mnt->mnt_longterm);
2357 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002358#endif
Al Virof03c6592011-01-14 22:30:21 -05002359}
2360
JANAK DESAI741a2952006-02-07 12:59:00 -08002361/*
2362 * Allocate a new namespace structure and populate it with contents
2363 * copied from the namespace of the passed in task structure.
2364 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002365static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002366 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002368 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002369 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 struct vfsmount *p, *q;
2371
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002372 new_ns = alloc_mnt_ns();
2373 if (IS_ERR(new_ns))
2374 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375
Ram Pai390c6842005-11-07 17:17:51 -05002376 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002378 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002379 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002381 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002383 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002385 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002387 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
2389 /*
2390 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2391 * as belonging to new namespace. We have already acquired a private
2392 * fs_struct, so tsk->fs->lock is not needed.
2393 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002394 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 q = new_ns->root;
2396 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002397 q->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002398 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002400 if (p == fs->root.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002401 fs->root.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002402 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002403 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 rootmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002406 if (p == fs->pwd.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002407 fs->pwd.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002408 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002409 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 pwdmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002413 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 q = next_mnt(q, new_ns->root);
2415 }
Ram Pai390c6842005-11-07 17:17:51 -05002416 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002419 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002421 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
JANAK DESAI741a2952006-02-07 12:59:00 -08002423 return new_ns;
2424}
2425
Eric W. Biederman213dd262007-07-15 23:41:15 -07002426struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002427 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002428{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002429 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002430
Badari Pulavartye3222c42007-05-08 00:25:21 -07002431 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002432 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002433
2434 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002435 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002436
Badari Pulavartye3222c42007-05-08 00:25:21 -07002437 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002438
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002439 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002440 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
2442
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002443/**
2444 * create_mnt_ns - creates a private namespace and adds a root filesystem
2445 * @mnt: pointer to the new root filesystem mountpoint
2446 */
Al Viro6c449c82011-11-16 22:01:36 -05002447static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002448{
2449 struct mnt_namespace *new_ns;
2450
2451 new_ns = alloc_mnt_ns();
2452 if (!IS_ERR(new_ns)) {
2453 mnt->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002454 __mnt_make_longterm(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002455 new_ns->root = mnt;
2456 list_add(&new_ns->list, &new_ns->root->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002457 } else {
2458 mntput(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002459 }
2460 return new_ns;
2461}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002462
Al Viroea441d12011-11-16 21:43:59 -05002463struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2464{
2465 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002466 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002467 struct path path;
2468 int err;
2469
2470 ns = create_mnt_ns(mnt);
2471 if (IS_ERR(ns))
2472 return ERR_CAST(ns);
2473
2474 err = vfs_path_lookup(mnt->mnt_root, mnt,
2475 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2476
2477 put_mnt_ns(ns);
2478
2479 if (err)
2480 return ERR_PTR(err);
2481
2482 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002483 s = path.mnt->mnt_sb;
2484 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002485 mntput(path.mnt);
2486 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002487 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002488 /* ... and return the root of (sub)tree on it */
2489 return path.dentry;
2490}
2491EXPORT_SYMBOL(mount_subtree);
2492
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002493SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2494 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002496 int ret;
2497 char *kernel_type;
2498 char *kernel_dir;
2499 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501
Vegard Nossumeca6f532009-09-18 13:05:45 -07002502 ret = copy_mount_string(type, &kernel_type);
2503 if (ret < 0)
2504 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Vegard Nossumeca6f532009-09-18 13:05:45 -07002506 kernel_dir = getname(dir_name);
2507 if (IS_ERR(kernel_dir)) {
2508 ret = PTR_ERR(kernel_dir);
2509 goto out_dir;
2510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Vegard Nossumeca6f532009-09-18 13:05:45 -07002512 ret = copy_mount_string(dev_name, &kernel_dev);
2513 if (ret < 0)
2514 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
Vegard Nossumeca6f532009-09-18 13:05:45 -07002516 ret = copy_mount_options(data, &data_page);
2517 if (ret < 0)
2518 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Vegard Nossumeca6f532009-09-18 13:05:45 -07002520 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2521 (void *) data_page);
2522
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002524out_data:
2525 kfree(kernel_dev);
2526out_dev:
2527 putname(kernel_dir);
2528out_dir:
2529 kfree(kernel_type);
2530out_type:
2531 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532}
2533
2534/*
Al Viroafac7cb2011-11-23 19:34:49 -05002535 * Return true if path is reachable from root
2536 *
2537 * namespace_sem or vfsmount_lock is held
2538 */
2539bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
2540 const struct path *root)
2541{
2542 while (mnt != root->mnt && mnt_has_parent(mnt)) {
2543 dentry = mnt->mnt_mountpoint;
2544 mnt = mnt->mnt_parent;
2545 }
2546 return mnt == root->mnt && is_subdir(dentry, root->dentry);
2547}
2548
2549int path_is_under(struct path *path1, struct path *path2)
2550{
2551 int res;
2552 br_read_lock(vfsmount_lock);
2553 res = is_path_reachable(path1->mnt, path1->dentry, path2);
2554 br_read_unlock(vfsmount_lock);
2555 return res;
2556}
2557EXPORT_SYMBOL(path_is_under);
2558
2559/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 * pivot_root Semantics:
2561 * Moves the root file system of the current process to the directory put_old,
2562 * makes new_root as the new root file system of the current process, and sets
2563 * root/cwd of all processes which had them on the current root to new_root.
2564 *
2565 * Restrictions:
2566 * The new_root and put_old must be directories, and must not be on the
2567 * same file system as the current process root. The put_old must be
2568 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2569 * pointed to by put_old must yield the same directory as new_root. No other
2570 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2571 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002572 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2573 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2574 * in this situation.
2575 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 * Notes:
2577 * - we don't move root/cwd if they are not at the root (reason: if something
2578 * cared enough to change them, it's probably wrong to force them elsewhere)
2579 * - it's okay to pick a root that isn't the root of a file system, e.g.
2580 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2581 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2582 * first.
2583 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002584SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2585 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586{
Al Viro2d8f3032008-07-22 09:59:21 -04002587 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 int error;
2589
2590 if (!capable(CAP_SYS_ADMIN))
2591 return -EPERM;
2592
Al Viro2d8f3032008-07-22 09:59:21 -04002593 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 if (error)
2595 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
Al Viro2d8f3032008-07-22 09:59:21 -04002597 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 if (error)
2599 goto out1;
2600
Al Viro2d8f3032008-07-22 09:59:21 -04002601 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002602 if (error)
2603 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002605 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002606 error = lock_mount(&old);
2607 if (error)
2608 goto out3;
2609
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002611 if (IS_MNT_SHARED(old.mnt) ||
2612 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002613 IS_MNT_SHARED(root.mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002614 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002615 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002616 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002618 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002619 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002620 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002621 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002623 if (new.mnt == root.mnt ||
2624 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002625 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002627 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002628 goto out4; /* not a mountpoint */
Al Virob2dba1a2011-11-23 19:26:23 -05002629 if (!mnt_has_parent(root.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002630 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002631 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002632 goto out4; /* not a mountpoint */
Al Virob2dba1a2011-11-23 19:26:23 -05002633 if (!mnt_has_parent(new.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002634 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002635 /* make sure we can reach put_old from new_root */
Al Viroafac7cb2011-11-23 19:34:49 -05002636 if (!is_path_reachable(old.mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002637 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002638 br_write_lock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002639 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002640 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002641 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002642 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002643 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002644 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002645 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002646 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002647 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002649out4:
2650 unlock_mount(&old);
2651 if (!error) {
2652 path_put(&root_parent);
2653 path_put(&parent_path);
2654 }
2655out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002656 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002657out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002658 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002660 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663}
2664
2665static void __init init_mount_tree(void)
2666{
2667 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002668 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002669 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
2671 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2672 if (IS_ERR(mnt))
2673 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002674
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002675 ns = create_mnt_ns(mnt);
2676 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002679 init_task.nsproxy->mnt_ns = ns;
2680 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Jan Blunckac748a02008-02-14 19:34:39 -08002682 root.mnt = ns->root;
2683 root.dentry = ns->root->mnt_root;
2684
2685 set_fs_pwd(current->fs, &root);
2686 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687}
2688
Denis Cheng74bf17c2007-10-16 23:26:30 -07002689void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002691 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002692 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
Ram Pai390c6842005-11-07 17:17:51 -05002694 init_rwsem(&namespace_sem);
2695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002697 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
Ram Paib58fed82005-11-07 17:16:09 -05002699 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
2701 if (!mount_hashtable)
2702 panic("Failed to allocate mount hash table\n");
2703
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002704 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705
Eric Dumazet13f14b42008-02-06 01:37:57 -08002706 for (u = 0; u < HASH_SIZE; u++)
2707 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
Nick Piggin99b7db72010-08-18 04:37:39 +10002709 br_lock_init(vfsmount_lock);
2710
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002711 err = sysfs_init();
2712 if (err)
2713 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002714 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002715 fs_kobj = kobject_create_and_add("fs", NULL);
2716 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002717 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 init_rootfs();
2719 init_mount_tree();
2720}
2721
Trond Myklebust616511d2009-06-22 15:09:13 -04002722void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002724 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002725
Al Virod498b252010-02-05 02:21:06 -05002726 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002727 return;
Ram Pai390c6842005-11-07 17:17:51 -05002728 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002729 br_write_lock(vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002730 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002731 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002732 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002733 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002734 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735}
Al Viro9d412a42011-03-17 22:08:28 -04002736
2737struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2738{
Tim Chen423e0ab2011-07-19 09:32:38 -07002739 struct vfsmount *mnt;
2740 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2741 if (!IS_ERR(mnt)) {
2742 /*
2743 * it is a longterm mount, don't release mnt until
2744 * we unmount before file sys is unregistered
2745 */
2746 mnt_make_longterm(mnt);
2747 }
2748 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002749}
2750EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002751
2752void kern_unmount(struct vfsmount *mnt)
2753{
2754 /* release long term mount so mount point can be released */
2755 if (!IS_ERR_OR_NULL(mnt)) {
2756 mnt_make_shortterm(mnt);
2757 mntput(mnt);
2758 }
2759}
2760EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002761
2762bool our_mnt(struct vfsmount *mnt)
2763{
2764 return check_mnt(mnt);
2765}