blob: 06f8e63f6cb1dc929e0ad9c0627320066db57cb1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080019#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070021#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080023#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/namei.h>
25#include <linux/security.h>
26#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010029#include <linux/idr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
31#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050032#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070033#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Eric Dumazet13f14b42008-02-06 01:37:57 -080035#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
36#define HASH_SIZE (1UL << HASH_SHIFT)
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050039__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
40
41static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010042static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010043static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Eric Dumazetfa3536c2006-03-26 01:37:24 -080045static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080046static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050047static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080049/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060050struct kobject *fs_kobj;
51EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
54{
Ram Paib58fed82005-11-07 17:16:09 -050055 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
56 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080057 tmp = tmp + (tmp >> HASH_SHIFT);
58 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059}
60
Dave Hansen3d733632008-02-15 14:37:59 -080061#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
62
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010063/* allocation is serialized by namespace_sem */
64static int mnt_alloc_id(struct vfsmount *mnt)
65{
66 int res;
67
68retry:
69 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
70 spin_lock(&vfsmount_lock);
71 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
72 spin_unlock(&vfsmount_lock);
73 if (res == -EAGAIN)
74 goto retry;
75
76 return res;
77}
78
79static void mnt_free_id(struct vfsmount *mnt)
80{
81 spin_lock(&vfsmount_lock);
82 ida_remove(&mnt_id_ida, mnt->mnt_id);
83 spin_unlock(&vfsmount_lock);
84}
85
Miklos Szeredi719f5d72008-03-27 13:06:23 +010086/*
87 * Allocate a new peer group ID
88 *
89 * mnt_group_ida is protected by namespace_sem
90 */
91static int mnt_alloc_group_id(struct vfsmount *mnt)
92{
93 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
94 return -ENOMEM;
95
96 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
97}
98
99/*
100 * Release a peer group ID
101 */
102void mnt_release_group_id(struct vfsmount *mnt)
103{
104 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
105 mnt->mnt_group_id = 0;
106}
107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108struct vfsmount *alloc_vfsmnt(const char *name)
109{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800110 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100112 int err;
113
114 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800115 if (err)
116 goto out_free_cache;
117
118 if (name) {
119 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
120 if (!mnt->mnt_devname)
121 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100122 }
123
Ram Paib58fed82005-11-07 17:16:09 -0500124 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 INIT_LIST_HEAD(&mnt->mnt_hash);
126 INIT_LIST_HEAD(&mnt->mnt_child);
127 INIT_LIST_HEAD(&mnt->mnt_mounts);
128 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700129 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500130 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500131 INIT_LIST_HEAD(&mnt->mnt_slave_list);
132 INIT_LIST_HEAD(&mnt->mnt_slave);
Dave Hansen3d733632008-02-15 14:37:59 -0800133 atomic_set(&mnt->__mnt_writers, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 }
135 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800136
137out_free_id:
138 mnt_free_id(mnt);
139out_free_cache:
140 kmem_cache_free(mnt_cache, mnt);
141 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
Dave Hansen83660252008-02-15 14:37:30 -0800144/*
145 * Most r/o checks on a fs are for operations that take
146 * discrete amounts of time, like a write() or unlink().
147 * We must keep track of when those operations start
148 * (for permission checks) and when they end, so that
149 * we can determine when writes are able to occur to
150 * a filesystem.
151 */
Dave Hansen3d733632008-02-15 14:37:59 -0800152/*
153 * __mnt_is_readonly: check whether a mount is read-only
154 * @mnt: the mount to check for its write status
155 *
156 * This shouldn't be used directly ouside of the VFS.
157 * It does not guarantee that the filesystem will stay
158 * r/w, just that it is right *now*. This can not and
159 * should not be used in place of IS_RDONLY(inode).
160 * mnt_want/drop_write() will _keep_ the filesystem
161 * r/w.
162 */
163int __mnt_is_readonly(struct vfsmount *mnt)
164{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800165 if (mnt->mnt_flags & MNT_READONLY)
166 return 1;
167 if (mnt->mnt_sb->s_flags & MS_RDONLY)
168 return 1;
169 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800170}
171EXPORT_SYMBOL_GPL(__mnt_is_readonly);
172
173struct mnt_writer {
174 /*
175 * If holding multiple instances of this lock, they
176 * must be ordered by cpu number.
177 */
178 spinlock_t lock;
179 struct lock_class_key lock_class; /* compiles out with !lockdep */
180 unsigned long count;
181 struct vfsmount *mnt;
182} ____cacheline_aligned_in_smp;
183static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
184
185static int __init init_mnt_writers(void)
186{
187 int cpu;
188 for_each_possible_cpu(cpu) {
189 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
190 spin_lock_init(&writer->lock);
191 lockdep_set_class(&writer->lock, &writer->lock_class);
192 writer->count = 0;
193 }
194 return 0;
195}
196fs_initcall(init_mnt_writers);
197
198static void unlock_mnt_writers(void)
199{
200 int cpu;
201 struct mnt_writer *cpu_writer;
202
203 for_each_possible_cpu(cpu) {
204 cpu_writer = &per_cpu(mnt_writers, cpu);
205 spin_unlock(&cpu_writer->lock);
206 }
207}
208
209static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
210{
211 if (!cpu_writer->mnt)
212 return;
213 /*
214 * This is in case anyone ever leaves an invalid,
215 * old ->mnt and a count of 0.
216 */
217 if (!cpu_writer->count)
218 return;
219 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
220 cpu_writer->count = 0;
221}
222 /*
223 * must hold cpu_writer->lock
224 */
225static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
226 struct vfsmount *mnt)
227{
228 if (cpu_writer->mnt == mnt)
229 return;
230 __clear_mnt_count(cpu_writer);
231 cpu_writer->mnt = mnt;
232}
233
234/*
235 * Most r/o checks on a fs are for operations that take
236 * discrete amounts of time, like a write() or unlink().
237 * We must keep track of when those operations start
238 * (for permission checks) and when they end, so that
239 * we can determine when writes are able to occur to
240 * a filesystem.
241 */
Dave Hansen83660252008-02-15 14:37:30 -0800242/**
243 * mnt_want_write - get write access to a mount
244 * @mnt: the mount on which to take a write
245 *
246 * This tells the low-level filesystem that a write is
247 * about to be performed to it, and makes sure that
248 * writes are allowed before returning success. When
249 * the write operation is finished, mnt_drop_write()
250 * must be called. This is effectively a refcount.
251 */
252int mnt_want_write(struct vfsmount *mnt)
253{
Dave Hansen3d733632008-02-15 14:37:59 -0800254 int ret = 0;
255 struct mnt_writer *cpu_writer;
256
257 cpu_writer = &get_cpu_var(mnt_writers);
258 spin_lock(&cpu_writer->lock);
259 if (__mnt_is_readonly(mnt)) {
260 ret = -EROFS;
261 goto out;
262 }
263 use_cpu_writer_for_mount(cpu_writer, mnt);
264 cpu_writer->count++;
265out:
266 spin_unlock(&cpu_writer->lock);
267 put_cpu_var(mnt_writers);
268 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800269}
270EXPORT_SYMBOL_GPL(mnt_want_write);
271
Dave Hansen3d733632008-02-15 14:37:59 -0800272static void lock_mnt_writers(void)
273{
274 int cpu;
275 struct mnt_writer *cpu_writer;
276
277 for_each_possible_cpu(cpu) {
278 cpu_writer = &per_cpu(mnt_writers, cpu);
279 spin_lock(&cpu_writer->lock);
280 __clear_mnt_count(cpu_writer);
281 cpu_writer->mnt = NULL;
282 }
283}
284
285/*
286 * These per-cpu write counts are not guaranteed to have
287 * matched increments and decrements on any given cpu.
288 * A file open()ed for write on one cpu and close()d on
289 * another cpu will imbalance this count. Make sure it
290 * does not get too far out of whack.
291 */
292static void handle_write_count_underflow(struct vfsmount *mnt)
293{
294 if (atomic_read(&mnt->__mnt_writers) >=
295 MNT_WRITER_UNDERFLOW_LIMIT)
296 return;
297 /*
298 * It isn't necessary to hold all of the locks
299 * at the same time, but doing it this way makes
300 * us share a lot more code.
301 */
302 lock_mnt_writers();
303 /*
304 * vfsmount_lock is for mnt_flags.
305 */
306 spin_lock(&vfsmount_lock);
307 /*
308 * If coalescing the per-cpu writer counts did not
309 * get us back to a positive writer count, we have
310 * a bug.
311 */
312 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
313 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
Arjan van de Ven5c752ad2008-07-25 19:45:40 -0700314 WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
Dave Hansen3d733632008-02-15 14:37:59 -0800315 "count: %d\n",
316 mnt, atomic_read(&mnt->__mnt_writers));
Dave Hansen3d733632008-02-15 14:37:59 -0800317 /* use the flag to keep the dmesg spam down */
318 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
319 }
320 spin_unlock(&vfsmount_lock);
321 unlock_mnt_writers();
322}
323
Dave Hansen83660252008-02-15 14:37:30 -0800324/**
325 * mnt_drop_write - give up write access to a mount
326 * @mnt: the mount on which to give up write access
327 *
328 * Tells the low-level filesystem that we are done
329 * performing writes to it. Must be matched with
330 * mnt_want_write() call above.
331 */
332void mnt_drop_write(struct vfsmount *mnt)
333{
Dave Hansen3d733632008-02-15 14:37:59 -0800334 int must_check_underflow = 0;
335 struct mnt_writer *cpu_writer;
336
337 cpu_writer = &get_cpu_var(mnt_writers);
338 spin_lock(&cpu_writer->lock);
339
340 use_cpu_writer_for_mount(cpu_writer, mnt);
341 if (cpu_writer->count > 0) {
342 cpu_writer->count--;
343 } else {
344 must_check_underflow = 1;
345 atomic_dec(&mnt->__mnt_writers);
346 }
347
348 spin_unlock(&cpu_writer->lock);
349 /*
350 * Logically, we could call this each time,
351 * but the __mnt_writers cacheline tends to
352 * be cold, and makes this expensive.
353 */
354 if (must_check_underflow)
355 handle_write_count_underflow(mnt);
356 /*
357 * This could be done right after the spinlock
358 * is taken because the spinlock keeps us on
359 * the cpu, and disables preemption. However,
360 * putting it here bounds the amount that
361 * __mnt_writers can underflow. Without it,
362 * we could theoretically wrap __mnt_writers.
363 */
364 put_cpu_var(mnt_writers);
Dave Hansen83660252008-02-15 14:37:30 -0800365}
366EXPORT_SYMBOL_GPL(mnt_drop_write);
367
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800368static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800369{
Dave Hansen3d733632008-02-15 14:37:59 -0800370 int ret = 0;
371
372 lock_mnt_writers();
373 /*
374 * With all the locks held, this value is stable
375 */
376 if (atomic_read(&mnt->__mnt_writers) > 0) {
377 ret = -EBUSY;
378 goto out;
379 }
380 /*
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800381 * nobody can do a successful mnt_want_write() with all
382 * of the counts in MNT_DENIED_WRITE and the locks held.
Dave Hansen3d733632008-02-15 14:37:59 -0800383 */
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800384 spin_lock(&vfsmount_lock);
385 if (!ret)
386 mnt->mnt_flags |= MNT_READONLY;
387 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800388out:
389 unlock_mnt_writers();
390 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800391}
Dave Hansen83660252008-02-15 14:37:30 -0800392
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800393static void __mnt_unmake_readonly(struct vfsmount *mnt)
394{
395 spin_lock(&vfsmount_lock);
396 mnt->mnt_flags &= ~MNT_READONLY;
397 spin_unlock(&vfsmount_lock);
398}
399
David Howells454e2392006-06-23 02:02:57 -0700400int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
401{
402 mnt->mnt_sb = sb;
403 mnt->mnt_root = dget(sb->s_root);
404 return 0;
405}
406
407EXPORT_SYMBOL(simple_set_mnt);
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409void free_vfsmnt(struct vfsmount *mnt)
410{
411 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100412 mnt_free_id(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 kmem_cache_free(mnt_cache, mnt);
414}
415
416/*
Ram Paia05964f2005-11-07 17:20:17 -0500417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 */
Ram Paia05964f2005-11-07 17:20:17 -0500420struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
421 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Ram Paib58fed82005-11-07 17:16:09 -0500423 struct list_head *head = mount_hashtable + hash(mnt, dentry);
424 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 struct vfsmount *p, *found = NULL;
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500428 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 p = NULL;
430 if (tmp == head)
431 break;
432 p = list_entry(tmp, struct vfsmount, mnt_hash);
433 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500434 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break;
436 }
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 return found;
439}
440
Ram Paia05964f2005-11-07 17:20:17 -0500441/*
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
444 */
445struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
446{
447 struct vfsmount *child_mnt;
448 spin_lock(&vfsmount_lock);
449 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
450 mntget(child_mnt);
451 spin_unlock(&vfsmount_lock);
452 return child_mnt;
453}
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455static inline int check_mnt(struct vfsmount *mnt)
456{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800457 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800460static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500461{
462 if (ns) {
463 ns->event = ++event;
464 wake_up_interruptible(&ns->poll);
465 }
466}
467
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800468static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500469{
470 if (ns && ns->event != event) {
471 ns->event = event;
472 wake_up_interruptible(&ns->poll);
473 }
474}
475
Al Viro1a390682008-03-21 20:48:19 -0400476static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Al Viro1a390682008-03-21 20:48:19 -0400478 old_path->dentry = mnt->mnt_mountpoint;
479 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 mnt->mnt_parent = mnt;
481 mnt->mnt_mountpoint = mnt->mnt_root;
482 list_del_init(&mnt->mnt_child);
483 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400484 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Ram Paib90fa9a2005-11-07 17:19:50 -0500487void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
488 struct vfsmount *child_mnt)
489{
490 child_mnt->mnt_parent = mntget(mnt);
491 child_mnt->mnt_mountpoint = dget(dentry);
492 dentry->d_mounted++;
493}
494
Al Viro1a390682008-03-21 20:48:19 -0400495static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Al Viro1a390682008-03-21 20:48:19 -0400497 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500498 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400499 hash(path->mnt, path->dentry));
500 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500501}
502
503/*
504 * the caller must hold vfsmount_lock
505 */
506static void commit_tree(struct vfsmount *mnt)
507{
508 struct vfsmount *parent = mnt->mnt_parent;
509 struct vfsmount *m;
510 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800511 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500512
513 BUG_ON(parent == mnt);
514
515 list_add_tail(&head, &mnt->mnt_list);
516 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800517 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500518 list_splice(&head, n->list.prev);
519
520 list_add_tail(&mnt->mnt_hash, mount_hashtable +
521 hash(parent, mnt->mnt_mountpoint));
522 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800523 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
527{
528 struct list_head *next = p->mnt_mounts.next;
529 if (next == &p->mnt_mounts) {
530 while (1) {
531 if (p == root)
532 return NULL;
533 next = p->mnt_child.next;
534 if (next != &p->mnt_parent->mnt_mounts)
535 break;
536 p = p->mnt_parent;
537 }
538 }
539 return list_entry(next, struct vfsmount, mnt_child);
540}
541
Ram Pai9676f0c2005-11-07 17:21:20 -0500542static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
543{
544 struct list_head *prev = p->mnt_mounts.prev;
545 while (prev != &p->mnt_mounts) {
546 p = list_entry(prev, struct vfsmount, mnt_child);
547 prev = p->mnt_mounts.prev;
548 }
549 return p;
550}
551
Ram Pai36341f62005-11-07 17:17:22 -0500552static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
553 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 struct super_block *sb = old->mnt_sb;
556 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
557
558 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100559 if (flag & (CL_SLAVE | CL_PRIVATE))
560 mnt->mnt_group_id = 0; /* not a peer of original */
561 else
562 mnt->mnt_group_id = old->mnt_group_id;
563
564 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
565 int err = mnt_alloc_group_id(mnt);
566 if (err)
567 goto out_free;
568 }
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 mnt->mnt_flags = old->mnt_flags;
571 atomic_inc(&sb->s_active);
572 mnt->mnt_sb = sb;
573 mnt->mnt_root = dget(root);
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500576
Ram Pai5afe0022005-11-07 17:21:01 -0500577 if (flag & CL_SLAVE) {
578 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
579 mnt->mnt_master = old;
580 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400581 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500582 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
583 list_add(&mnt->mnt_share, &old->mnt_share);
584 if (IS_MNT_SLAVE(old))
585 list_add(&mnt->mnt_slave, &old->mnt_slave);
586 mnt->mnt_master = old->mnt_master;
587 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500588 if (flag & CL_MAKE_SHARED)
589 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500593 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500594 if (!list_empty(&old->mnt_expire))
595 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100599
600 out_free:
601 free_vfsmnt(mnt);
602 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Al Viro7b7b1ac2005-11-07 17:13:39 -0500605static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Dave Hansen3d733632008-02-15 14:37:59 -0800607 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800609 /*
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
613 * can come in.
614 */
615 for_each_possible_cpu(cpu) {
616 struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800617 spin_lock(&cpu_writer->lock);
Al Viro1a88b532009-02-16 02:38:12 +0000618 if (cpu_writer->mnt != mnt) {
619 spin_unlock(&cpu_writer->lock);
620 continue;
621 }
Dave Hansen3d733632008-02-15 14:37:59 -0800622 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
623 cpu_writer->count = 0;
624 /*
625 * Might as well do this so that no one
626 * ever sees the pointer and expects
627 * it to be valid.
628 */
629 cpu_writer->mnt = NULL;
630 spin_unlock(&cpu_writer->lock);
631 }
632 /*
633 * This probably indicates that somebody messed
634 * up a mnt_want/drop_write() pair. If this
635 * happens, the filesystem was probably unable
636 * to make r/w->r/o transitions.
637 */
638 WARN_ON(atomic_read(&mnt->__mnt_writers));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 dput(mnt->mnt_root);
640 free_vfsmnt(mnt);
641 deactivate_super(sb);
642}
643
Al Viro7b7b1ac2005-11-07 17:13:39 -0500644void mntput_no_expire(struct vfsmount *mnt)
645{
646repeat:
647 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
648 if (likely(!mnt->mnt_pinned)) {
649 spin_unlock(&vfsmount_lock);
650 __mntput(mnt);
651 return;
652 }
653 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
654 mnt->mnt_pinned = 0;
655 spin_unlock(&vfsmount_lock);
656 acct_auto_close_mnt(mnt);
657 security_sb_umount_close(mnt);
658 goto repeat;
659 }
660}
661
662EXPORT_SYMBOL(mntput_no_expire);
663
664void mnt_pin(struct vfsmount *mnt)
665{
666 spin_lock(&vfsmount_lock);
667 mnt->mnt_pinned++;
668 spin_unlock(&vfsmount_lock);
669}
670
671EXPORT_SYMBOL(mnt_pin);
672
673void mnt_unpin(struct vfsmount *mnt)
674{
675 spin_lock(&vfsmount_lock);
676 if (mnt->mnt_pinned) {
677 atomic_inc(&mnt->mnt_count);
678 mnt->mnt_pinned--;
679 }
680 spin_unlock(&vfsmount_lock);
681}
682
683EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800685static inline void mangle(struct seq_file *m, const char *s)
686{
687 seq_escape(m, s, " \t\n\\");
688}
689
690/*
691 * Simple .show_options callback for filesystems which don't want to
692 * implement more complex mount option showing.
693 *
694 * See also save_mount_options().
695 */
696int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
697{
698 const char *options = mnt->mnt_sb->s_options;
699
700 if (options != NULL && options[0]) {
701 seq_putc(m, ',');
702 mangle(m, options);
703 }
704
705 return 0;
706}
707EXPORT_SYMBOL(generic_show_options);
708
709/*
710 * If filesystem uses generic_show_options(), this function should be
711 * called from the fill_super() callback.
712 *
713 * The .remount_fs callback usually needs to be handled in a special
714 * way, to make sure, that previous options are not overwritten if the
715 * remount fails.
716 *
717 * Also note, that if the filesystem's .remount_fs function doesn't
718 * reset all options to their default value, but changes only newly
719 * given options, then the displayed options will not reflect reality
720 * any more.
721 */
722void save_mount_options(struct super_block *sb, char *options)
723{
724 kfree(sb->s_options);
725 sb->s_options = kstrdup(options, GFP_KERNEL);
726}
727EXPORT_SYMBOL(save_mount_options);
728
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100729#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/* iterator */
731static void *m_start(struct seq_file *m, loff_t *pos)
732{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100733 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Ram Pai390c6842005-11-07 17:17:51 -0500735 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100736 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739static void *m_next(struct seq_file *m, void *v, loff_t *pos)
740{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100741 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700742
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100743 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
746static void m_stop(struct seq_file *m, void *v)
747{
Ram Pai390c6842005-11-07 17:17:51 -0500748 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
Ram Pai2d4d4862008-03-27 13:06:25 +0100751struct proc_fs_info {
752 int flag;
753 const char *str;
754};
755
Eric Paris2069f452008-07-04 09:47:13 +1000756static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
Ram Pai2d4d4862008-03-27 13:06:25 +0100758 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 { MS_SYNCHRONOUS, ",sync" },
760 { MS_DIRSYNC, ",dirsync" },
761 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 { 0, NULL }
763 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100764 const struct proc_fs_info *fs_infop;
765
766 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
767 if (sb->s_flags & fs_infop->flag)
768 seq_puts(m, fs_infop->str);
769 }
Eric Paris2069f452008-07-04 09:47:13 +1000770
771 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100772}
773
774static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
775{
776 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 { MNT_NOSUID, ",nosuid" },
778 { MNT_NODEV, ",nodev" },
779 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800780 { MNT_NOATIME, ",noatime" },
781 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800782 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 { 0, NULL }
784 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100785 const struct proc_fs_info *fs_infop;
786
787 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
788 if (mnt->mnt_flags & fs_infop->flag)
789 seq_puts(m, fs_infop->str);
790 }
791}
792
793static void show_type(struct seq_file *m, struct super_block *sb)
794{
795 mangle(m, sb->s_type->name);
796 if (sb->s_subtype && sb->s_subtype[0]) {
797 seq_putc(m, '.');
798 mangle(m, sb->s_subtype);
799 }
800}
801
802static int show_vfsmnt(struct seq_file *m, void *v)
803{
804 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
805 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800806 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
809 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800810 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100812 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800813 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000814 err = show_sb_opts(m, mnt->mnt_sb);
815 if (err)
816 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100817 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (mnt->mnt_sb->s_op->show_options)
819 err = mnt->mnt_sb->s_op->show_options(m, mnt);
820 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000821out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 return err;
823}
824
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100825const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 .start = m_start,
827 .next = m_next,
828 .stop = m_stop,
829 .show = show_vfsmnt
830};
831
Ram Pai2d4d4862008-03-27 13:06:25 +0100832static int show_mountinfo(struct seq_file *m, void *v)
833{
834 struct proc_mounts *p = m->private;
835 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
836 struct super_block *sb = mnt->mnt_sb;
837 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
838 struct path root = p->root;
839 int err = 0;
840
841 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
842 MAJOR(sb->s_dev), MINOR(sb->s_dev));
843 seq_dentry(m, mnt->mnt_root, " \t\n\\");
844 seq_putc(m, ' ');
845 seq_path_root(m, &mnt_path, &root, " \t\n\\");
846 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
847 /*
848 * Mountpoint is outside root, discard that one. Ugly,
849 * but less so than trying to do that in iterator in a
850 * race-free way (due to renames).
851 */
852 return SEQ_SKIP;
853 }
854 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
855 show_mnt_opts(m, mnt);
856
857 /* Tagged fields ("foo:X" or "bar") */
858 if (IS_MNT_SHARED(mnt))
859 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100860 if (IS_MNT_SLAVE(mnt)) {
861 int master = mnt->mnt_master->mnt_group_id;
862 int dom = get_dominating_id(mnt, &p->root);
863 seq_printf(m, " master:%i", master);
864 if (dom && dom != master)
865 seq_printf(m, " propagate_from:%i", dom);
866 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100867 if (IS_MNT_UNBINDABLE(mnt))
868 seq_puts(m, " unbindable");
869
870 /* Filesystem specific data */
871 seq_puts(m, " - ");
872 show_type(m, sb);
873 seq_putc(m, ' ');
874 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
875 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000876 err = show_sb_opts(m, sb);
877 if (err)
878 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100879 if (sb->s_op->show_options)
880 err = sb->s_op->show_options(m, mnt);
881 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000882out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100883 return err;
884}
885
886const struct seq_operations mountinfo_op = {
887 .start = m_start,
888 .next = m_next,
889 .stop = m_stop,
890 .show = show_mountinfo,
891};
892
Chuck Leverb4629fe2006-03-20 13:44:12 -0500893static int show_vfsstat(struct seq_file *m, void *v)
894{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700895 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800896 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500897 int err = 0;
898
899 /* device */
900 if (mnt->mnt_devname) {
901 seq_puts(m, "device ");
902 mangle(m, mnt->mnt_devname);
903 } else
904 seq_puts(m, "no device");
905
906 /* mount point */
907 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800908 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500909 seq_putc(m, ' ');
910
911 /* file system type */
912 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100913 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500914
915 /* optional statistics */
916 if (mnt->mnt_sb->s_op->show_stats) {
917 seq_putc(m, ' ');
918 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
919 }
920
921 seq_putc(m, '\n');
922 return err;
923}
924
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100925const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500926 .start = m_start,
927 .next = m_next,
928 .stop = m_stop,
929 .show = show_vfsstat,
930};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100931#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933/**
934 * may_umount_tree - check if a mount tree is busy
935 * @mnt: root of mount tree
936 *
937 * This is called to check if a tree of mounts has any
938 * open files, pwds, chroots or sub mounts that are
939 * busy.
940 */
941int may_umount_tree(struct vfsmount *mnt)
942{
Ram Pai36341f62005-11-07 17:17:22 -0500943 int actual_refs = 0;
944 int minimum_refs = 0;
945 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
947 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500948 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 actual_refs += atomic_read(&p->mnt_count);
950 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 }
952 spin_unlock(&vfsmount_lock);
953
954 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800955 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Ian Kente3474a82006-03-27 01:14:51 -0800957 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958}
959
960EXPORT_SYMBOL(may_umount_tree);
961
962/**
963 * may_umount - check if a mount point is busy
964 * @mnt: root of mount
965 *
966 * This is called to check if a mount point has any
967 * open files, pwds, chroots or sub mounts. If the
968 * mount has sub mounts this will return busy
969 * regardless of whether the sub mounts are busy.
970 *
971 * Doesn't take quota and stuff into account. IOW, in some cases it will
972 * give false negatives. The main reason why it's here is that we need
973 * a non-destructive way to look for easily umountable filesystems.
974 */
975int may_umount(struct vfsmount *mnt)
976{
Ian Kente3474a82006-03-27 01:14:51 -0800977 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500978 spin_lock(&vfsmount_lock);
979 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800980 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500981 spin_unlock(&vfsmount_lock);
982 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983}
984
985EXPORT_SYMBOL(may_umount);
986
Ram Paib90fa9a2005-11-07 17:19:50 -0500987void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500989 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800990 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700991 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500992 list_del_init(&mnt->mnt_hash);
993 if (mnt->mnt_parent != mnt) {
994 struct dentry *dentry;
995 struct vfsmount *m;
996 spin_lock(&vfsmount_lock);
997 dentry = mnt->mnt_mountpoint;
998 m = mnt->mnt_parent;
999 mnt->mnt_mountpoint = mnt->mnt_root;
1000 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001001 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001003 dput(dentry);
1004 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 }
1006 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001007 }
1008}
1009
Ram Paia05964f2005-11-07 17:20:17 -05001010void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001011{
1012 struct vfsmount *p;
1013
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001014 for (p = mnt; p; p = next_mnt(p, mnt))
1015 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001016
Ram Paia05964f2005-11-07 17:20:17 -05001017 if (propagate)
1018 propagate_umount(kill);
1019
Ram Pai70fbcdf2005-11-07 17:17:04 -05001020 list_for_each_entry(p, kill, mnt_hash) {
1021 list_del_init(&p->mnt_expire);
1022 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001023 __touch_mnt_namespace(p->mnt_ns);
1024 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001025 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001026 if (p->mnt_parent != p) {
1027 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001028 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001029 }
Ram Paia05964f2005-11-07 17:20:17 -05001030 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
1032}
1033
Al Viroc35038b2008-03-22 00:46:23 -04001034static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036static int do_umount(struct vfsmount *mnt, int flags)
1037{
Ram Paib58fed82005-11-07 17:16:09 -05001038 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001040 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 retval = security_sb_umount(mnt, flags);
1043 if (retval)
1044 return retval;
1045
1046 /*
1047 * Allow userspace to request a mountpoint be expired rather than
1048 * unmounting unconditionally. Unmount only happens if:
1049 * (1) the mark is already set (the mark is cleared by mntput())
1050 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1051 */
1052 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001053 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 flags & (MNT_FORCE | MNT_DETACH))
1055 return -EINVAL;
1056
1057 if (atomic_read(&mnt->mnt_count) != 2)
1058 return -EBUSY;
1059
1060 if (!xchg(&mnt->mnt_expiry_mark, 1))
1061 return -EAGAIN;
1062 }
1063
1064 /*
1065 * If we may have to abort operations to get out of this
1066 * mount, and they will themselves hold resources we must
1067 * allow the fs to do things. In the Unix tradition of
1068 * 'Gee thats tricky lets do it in userspace' the umount_begin
1069 * might fail to complete on the first run through as other tasks
1070 * must return, and the like. Thats for the mount program to worry
1071 * about for the moment.
1072 */
1073
Al Viro42faad92008-04-24 07:21:56 -04001074 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1075 lock_kernel();
1076 sb->s_op->umount_begin(sb);
1077 unlock_kernel();
1078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
1080 /*
1081 * No sense to grab the lock for this test, but test itself looks
1082 * somewhat bogus. Suggestions for better replacement?
1083 * Ho-hum... In principle, we might treat that as umount + switch
1084 * to rootfs. GC would eventually take care of the old vfsmount.
1085 * Actually it makes sense, especially if rootfs would contain a
1086 * /reboot - static binary that would close all descriptors and
1087 * call reboot(9). Then init(8) could umount root and exec /reboot.
1088 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001089 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 /*
1091 * Special case for "unmounting" root ...
1092 * we just try to remount it readonly.
1093 */
1094 down_write(&sb->s_umount);
1095 if (!(sb->s_flags & MS_RDONLY)) {
1096 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1098 unlock_kernel();
1099 }
1100 up_write(&sb->s_umount);
1101 return retval;
1102 }
1103
Ram Pai390c6842005-11-07 17:17:51 -05001104 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001106 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Al Viroc35038b2008-03-22 00:46:23 -04001108 if (!(flags & MNT_DETACH))
1109 shrink_submounts(mnt, &umount_list);
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001112 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001114 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 retval = 0;
1116 }
1117 spin_unlock(&vfsmount_lock);
1118 if (retval)
1119 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001120 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001121 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 return retval;
1123}
1124
1125/*
1126 * Now umount can handle mount points as well as block devices.
1127 * This is important for filesystems which use unnamed block devices.
1128 *
1129 * We now support a flag for forced unmount like the other 'big iron'
1130 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1131 */
1132
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001133SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
Al Viro2d8f3032008-07-22 09:59:21 -04001135 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 int retval;
1137
Al Viro2d8f3032008-07-22 09:59:21 -04001138 retval = user_path(name, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 if (retval)
1140 goto out;
1141 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001142 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001144 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 goto dput_and_out;
1146
1147 retval = -EPERM;
1148 if (!capable(CAP_SYS_ADMIN))
1149 goto dput_and_out;
1150
Al Viro2d8f3032008-07-22 09:59:21 -04001151 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001153 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001154 dput(path.dentry);
1155 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156out:
1157 return retval;
1158}
1159
1160#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1161
1162/*
Ram Paib58fed82005-11-07 17:16:09 -05001163 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001165SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
Ram Paib58fed82005-11-07 17:16:09 -05001167 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168}
1169
1170#endif
1171
Al Viro2d92ab32008-08-02 00:51:11 -04001172static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173{
1174 if (capable(CAP_SYS_ADMIN))
1175 return 0;
1176 return -EPERM;
1177#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001178 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001180 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001181 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return -EPERM;
1183 }
Al Viro2d92ab32008-08-02 00:51:11 -04001184 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return -EPERM;
1186 return 0;
1187#endif
1188}
1189
Ram Paib90fa9a2005-11-07 17:19:50 -05001190struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001191 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
1193 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001194 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Ram Pai9676f0c2005-11-07 17:21:20 -05001196 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1197 return NULL;
1198
Ram Pai36341f62005-11-07 17:17:22 -05001199 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 if (!q)
1201 goto Enomem;
1202 q->mnt_mountpoint = mnt->mnt_mountpoint;
1203
1204 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001205 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001206 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 continue;
1208
1209 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001210 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1211 s = skip_mnt_tree(s);
1212 continue;
1213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 while (p != s->mnt_parent) {
1215 p = p->mnt_parent;
1216 q = q->mnt_parent;
1217 }
1218 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001219 path.mnt = q;
1220 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001221 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 if (!q)
1223 goto Enomem;
1224 spin_lock(&vfsmount_lock);
1225 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001226 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 spin_unlock(&vfsmount_lock);
1228 }
1229 }
1230 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001231Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001233 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001235 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001237 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 }
1239 return NULL;
1240}
1241
Al Viro8aec0802007-06-07 12:20:32 -04001242struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1243{
1244 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001245 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001246 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001247 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001248 return tree;
1249}
1250
1251void drop_collected_mounts(struct vfsmount *mnt)
1252{
1253 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001254 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001255 spin_lock(&vfsmount_lock);
1256 umount_tree(mnt, 0, &umount_list);
1257 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001258 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001259 release_mounts(&umount_list);
1260}
1261
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001262static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1263{
1264 struct vfsmount *p;
1265
1266 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1267 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1268 mnt_release_group_id(p);
1269 }
1270}
1271
1272static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1273{
1274 struct vfsmount *p;
1275
1276 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1277 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1278 int err = mnt_alloc_group_id(p);
1279 if (err) {
1280 cleanup_group_ids(mnt, p);
1281 return err;
1282 }
1283 }
1284 }
1285
1286 return 0;
1287}
1288
Ram Paib90fa9a2005-11-07 17:19:50 -05001289/*
1290 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001291 * @nd : place the mount tree @source_mnt is attached
1292 * @parent_nd : if non-null, detach the source_mnt from its parent and
1293 * store the parent mount and mountpoint dentry.
1294 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001295 *
1296 * NOTE: in the table below explains the semantics when a source mount
1297 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001298 * ---------------------------------------------------------------------------
1299 * | BIND MOUNT OPERATION |
1300 * |**************************************************************************
1301 * | source-->| shared | private | slave | unbindable |
1302 * | dest | | | | |
1303 * | | | | | | |
1304 * | v | | | | |
1305 * |**************************************************************************
1306 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1307 * | | | | | |
1308 * |non-shared| shared (+) | private | slave (*) | invalid |
1309 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001310 * A bind operation clones the source mount and mounts the clone on the
1311 * destination mount.
1312 *
1313 * (++) the cloned mount is propagated to all the mounts in the propagation
1314 * tree of the destination mount and the cloned mount is added to
1315 * the peer group of the source mount.
1316 * (+) the cloned mount is created under the destination mount and is marked
1317 * as shared. The cloned mount is added to the peer group of the source
1318 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001319 * (+++) the mount is propagated to all the mounts in the propagation tree
1320 * of the destination mount and the cloned mount is made slave
1321 * of the same master as that of the source mount. The cloned mount
1322 * is marked as 'shared and slave'.
1323 * (*) the cloned mount is made a slave of the same master as that of the
1324 * source mount.
1325 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001326 * ---------------------------------------------------------------------------
1327 * | MOVE MOUNT OPERATION |
1328 * |**************************************************************************
1329 * | source-->| shared | private | slave | unbindable |
1330 * | dest | | | | |
1331 * | | | | | | |
1332 * | v | | | | |
1333 * |**************************************************************************
1334 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1335 * | | | | | |
1336 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1337 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001338 *
1339 * (+) the mount is moved to the destination. And is then propagated to
1340 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001341 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001342 * (+++) the mount is moved to the destination and is then propagated to
1343 * all the mounts belonging to the destination mount's propagation tree.
1344 * the mount is marked as 'shared and slave'.
1345 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001346 *
1347 * if the source mount is a tree, the operations explained above is
1348 * applied to each mount in the tree.
1349 * Must be called without spinlocks held, since this function can sleep
1350 * in allocations.
1351 */
1352static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001353 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001354{
1355 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001356 struct vfsmount *dest_mnt = path->mnt;
1357 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001358 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001359 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001360
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001361 if (IS_MNT_SHARED(dest_mnt)) {
1362 err = invent_group_ids(source_mnt, true);
1363 if (err)
1364 goto out;
1365 }
1366 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1367 if (err)
1368 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001369
1370 if (IS_MNT_SHARED(dest_mnt)) {
1371 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1372 set_mnt_shared(p);
1373 }
1374
1375 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001376 if (parent_path) {
1377 detach_mnt(source_mnt, parent_path);
1378 attach_mnt(source_mnt, path);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001379 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001380 } else {
1381 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1382 commit_tree(source_mnt);
1383 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001384
1385 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1386 list_del_init(&child->mnt_hash);
1387 commit_tree(child);
1388 }
1389 spin_unlock(&vfsmount_lock);
1390 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001391
1392 out_cleanup_ids:
1393 if (IS_MNT_SHARED(dest_mnt))
1394 cleanup_group_ids(source_mnt, NULL);
1395 out:
1396 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001397}
1398
Al Viro8c3ee422008-03-22 18:00:39 -04001399static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 int err;
1402 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1403 return -EINVAL;
1404
Al Viro8c3ee422008-03-22 18:00:39 -04001405 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1407 return -ENOTDIR;
1408
1409 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001410 mutex_lock(&path->dentry->d_inode->i_mutex);
1411 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 goto out_unlock;
1413
Al Viro8c3ee422008-03-22 18:00:39 -04001414 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 if (err)
1416 goto out_unlock;
1417
1418 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001419 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1420 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001422 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001424 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 return err;
1426}
1427
1428/*
Ram Pai07b20882005-11-07 17:19:07 -05001429 * recursively change the type of the mountpoint.
1430 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001431static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001432{
Al Viro2d92ab32008-08-02 00:51:11 -04001433 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001434 int recurse = flag & MS_REC;
1435 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001436 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001437
Miklos Szerediee6f9582007-05-08 00:30:40 -07001438 if (!capable(CAP_SYS_ADMIN))
1439 return -EPERM;
1440
Al Viro2d92ab32008-08-02 00:51:11 -04001441 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001442 return -EINVAL;
1443
1444 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001445 if (type == MS_SHARED) {
1446 err = invent_group_ids(mnt, recurse);
1447 if (err)
1448 goto out_unlock;
1449 }
1450
Ram Pai07b20882005-11-07 17:19:07 -05001451 spin_lock(&vfsmount_lock);
1452 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1453 change_mnt_propagation(m, type);
1454 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001455
1456 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001457 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001458 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001459}
1460
1461/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 * do loopback mount.
1463 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001464static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001465 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Al Viro2d92ab32008-08-02 00:51:11 -04001467 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001469 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 if (err)
1471 return err;
1472 if (!old_name || !*old_name)
1473 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001474 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (err)
1476 return err;
1477
Ram Pai390c6842005-11-07 17:17:51 -05001478 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001480 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001481 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001482
Al Viro2d92ab32008-08-02 00:51:11 -04001483 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001484 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Al Viroccd48bc2005-11-07 17:15:04 -05001486 err = -ENOMEM;
1487 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001488 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001489 else
Al Viro2d92ab32008-08-02 00:51:11 -04001490 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001491
1492 if (!mnt)
1493 goto out;
1494
Al Viro2d92ab32008-08-02 00:51:11 -04001495 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001496 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001497 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001499 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001501 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Al Viroccd48bc2005-11-07 17:15:04 -05001504out:
Ram Pai390c6842005-11-07 17:17:51 -05001505 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001506 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return err;
1508}
1509
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001510static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1511{
1512 int error = 0;
1513 int readonly_request = 0;
1514
1515 if (ms_flags & MS_RDONLY)
1516 readonly_request = 1;
1517 if (readonly_request == __mnt_is_readonly(mnt))
1518 return 0;
1519
1520 if (readonly_request)
1521 error = mnt_make_readonly(mnt);
1522 else
1523 __mnt_unmake_readonly(mnt);
1524 return error;
1525}
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527/*
1528 * change filesystem flags. dir should be a physical root of filesystem.
1529 * If you've mounted a non-root directory somewhere and want to do remount
1530 * on it - tough luck.
1531 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001532static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 void *data)
1534{
1535 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001536 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 if (!capable(CAP_SYS_ADMIN))
1539 return -EPERM;
1540
Al Viro2d92ab32008-08-02 00:51:11 -04001541 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 return -EINVAL;
1543
Al Viro2d92ab32008-08-02 00:51:11 -04001544 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return -EINVAL;
1546
1547 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001548 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001549 err = change_mount_flags(path->mnt, flags);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001550 else
1551 err = do_remount_sb(sb, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (!err)
Al Viro2d92ab32008-08-02 00:51:11 -04001553 path->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001555 if (!err) {
Al Viro2d92ab32008-08-02 00:51:11 -04001556 security_sb_post_remount(path->mnt, flags, data);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001557
1558 spin_lock(&vfsmount_lock);
1559 touch_mnt_namespace(path->mnt->mnt_ns);
1560 spin_unlock(&vfsmount_lock);
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return err;
1563}
1564
Ram Pai9676f0c2005-11-07 17:21:20 -05001565static inline int tree_contains_unbindable(struct vfsmount *mnt)
1566{
1567 struct vfsmount *p;
1568 for (p = mnt; p; p = next_mnt(p, mnt)) {
1569 if (IS_MNT_UNBINDABLE(p))
1570 return 1;
1571 }
1572 return 0;
1573}
1574
Al Viro0a0d8a42008-08-02 00:55:27 -04001575static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576{
Al Viro2d92ab32008-08-02 00:51:11 -04001577 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 struct vfsmount *p;
1579 int err = 0;
1580 if (!capable(CAP_SYS_ADMIN))
1581 return -EPERM;
1582 if (!old_name || !*old_name)
1583 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001584 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (err)
1586 return err;
1587
Ram Pai390c6842005-11-07 17:17:51 -05001588 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001589 while (d_mountpoint(path->dentry) &&
1590 follow_down(&path->mnt, &path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 ;
1592 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001593 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 goto out;
1595
1596 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001597 mutex_lock(&path->dentry->d_inode->i_mutex);
1598 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 goto out1;
1600
Al Viro2d92ab32008-08-02 00:51:11 -04001601 if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001602 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
1604 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001605 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001606 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Al Viro2d92ab32008-08-02 00:51:11 -04001608 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001609 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Al Viro2d92ab32008-08-02 00:51:11 -04001611 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1612 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001613 goto out1;
1614 /*
1615 * Don't move a mount residing in a shared parent.
1616 */
Al Viro2d92ab32008-08-02 00:51:11 -04001617 if (old_path.mnt->mnt_parent &&
1618 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001619 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001620 /*
1621 * Don't move a mount tree containing unbindable mounts to a destination
1622 * mount which is shared.
1623 */
Al Viro2d92ab32008-08-02 00:51:11 -04001624 if (IS_MNT_SHARED(path->mnt) &&
1625 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001626 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001628 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1629 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001630 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Al Viro2d92ab32008-08-02 00:51:11 -04001632 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001633 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001634 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
1636 /* if the mount is moved, it should no longer be expire
1637 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001638 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001640 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641out:
Ram Pai390c6842005-11-07 17:17:51 -05001642 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001644 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001645 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return err;
1647}
1648
1649/*
1650 * create a new mount for userspace and request it to be added into the
1651 * namespace's tree
1652 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001653static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 int mnt_flags, char *name, void *data)
1655{
1656 struct vfsmount *mnt;
1657
1658 if (!type || !memchr(type, 0, PAGE_SIZE))
1659 return -EINVAL;
1660
1661 /* we need capabilities... */
1662 if (!capable(CAP_SYS_ADMIN))
1663 return -EPERM;
1664
1665 mnt = do_kern_mount(type, flags, name, data);
1666 if (IS_ERR(mnt))
1667 return PTR_ERR(mnt);
1668
Al Viro2d92ab32008-08-02 00:51:11 -04001669 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670}
1671
1672/*
1673 * add a mount into a namespace's mount tree
1674 * - provide the option of adding the new mount to an expiration list
1675 */
Al Viro8d66bf52008-08-01 09:05:54 -04001676int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 int mnt_flags, struct list_head *fslist)
1678{
1679 int err;
1680
Ram Pai390c6842005-11-07 17:17:51 -05001681 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001683 while (d_mountpoint(path->dentry) &&
1684 follow_down(&path->mnt, &path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 ;
1686 err = -EINVAL;
Al Viro8d66bf52008-08-01 09:05:54 -04001687 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 goto unlock;
1689
1690 /* Refuse the same filesystem on the same mount point */
1691 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001692 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1693 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 goto unlock;
1695
1696 err = -EINVAL;
1697 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1698 goto unlock;
1699
1700 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001701 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001702 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
Al Viro6758f952008-03-22 16:14:30 -04001704 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001705 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001706
Ram Pai390c6842005-11-07 17:17:51 -05001707 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001708 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
1710unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001711 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 mntput(newmnt);
1713 return err;
1714}
1715
1716EXPORT_SYMBOL_GPL(do_add_mount);
1717
Trond Myklebust5528f9112006-06-09 09:34:17 -04001718/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 * process a list of expirable mountpoints with the intent of discarding any
1720 * mountpoints that aren't in use and haven't been touched since last we came
1721 * here
1722 */
1723void mark_mounts_for_expiry(struct list_head *mounts)
1724{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 struct vfsmount *mnt, *next;
1726 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001727 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729 if (list_empty(mounts))
1730 return;
1731
Al Virobcc5c7d2008-03-22 00:21:53 -04001732 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 spin_lock(&vfsmount_lock);
1734
1735 /* extract from the expiration list every vfsmount that matches the
1736 * following criteria:
1737 * - only referenced by its parent vfsmount
1738 * - still marked for expiry (marked on the last call here; marks are
1739 * cleared by mntput())
1740 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001741 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001743 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001745 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001747 while (!list_empty(&graveyard)) {
1748 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1749 touch_mnt_namespace(mnt->mnt_ns);
1750 umount_tree(mnt, 1, &umounts);
1751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001753 up_write(&namespace_sem);
1754
1755 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756}
1757
1758EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1759
1760/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001761 * Ripoff of 'select_parent()'
1762 *
1763 * search the list of submounts for a given mountpoint, and move any
1764 * shrinkable submounts to the 'graveyard' list.
1765 */
1766static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1767{
1768 struct vfsmount *this_parent = parent;
1769 struct list_head *next;
1770 int found = 0;
1771
1772repeat:
1773 next = this_parent->mnt_mounts.next;
1774resume:
1775 while (next != &this_parent->mnt_mounts) {
1776 struct list_head *tmp = next;
1777 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1778
1779 next = tmp->next;
1780 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1781 continue;
1782 /*
1783 * Descend a level if the d_mounts list is non-empty.
1784 */
1785 if (!list_empty(&mnt->mnt_mounts)) {
1786 this_parent = mnt;
1787 goto repeat;
1788 }
1789
1790 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001791 list_move_tail(&mnt->mnt_expire, graveyard);
1792 found++;
1793 }
1794 }
1795 /*
1796 * All done at this level ... ascend and resume the search
1797 */
1798 if (this_parent != parent) {
1799 next = this_parent->mnt_child.next;
1800 this_parent = this_parent->mnt_parent;
1801 goto resume;
1802 }
1803 return found;
1804}
1805
1806/*
1807 * process a list of expirable mountpoints with the intent of discarding any
1808 * submounts of a specific parent mountpoint
1809 */
Al Viroc35038b2008-03-22 00:46:23 -04001810static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001811{
1812 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001813 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001814
Trond Myklebust5528f9112006-06-09 09:34:17 -04001815 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001816 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001817 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001818 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001819 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001820 touch_mnt_namespace(m->mnt_ns);
1821 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001822 }
1823 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001824}
1825
Trond Myklebust5528f9112006-06-09 09:34:17 -04001826/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 * Some copy_from_user() implementations do not return the exact number of
1828 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1829 * Note that this function differs from copy_from_user() in that it will oops
1830 * on bad values of `to', rather than returning a short copy.
1831 */
Ram Paib58fed82005-11-07 17:16:09 -05001832static long exact_copy_from_user(void *to, const void __user * from,
1833 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834{
1835 char *t = to;
1836 const char __user *f = from;
1837 char c;
1838
1839 if (!access_ok(VERIFY_READ, from, n))
1840 return n;
1841
1842 while (n) {
1843 if (__get_user(c, f)) {
1844 memset(t, 0, n);
1845 break;
1846 }
1847 *t++ = c;
1848 f++;
1849 n--;
1850 }
1851 return n;
1852}
1853
Ram Paib58fed82005-11-07 17:16:09 -05001854int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855{
1856 int i;
1857 unsigned long page;
1858 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 *where = 0;
1861 if (!data)
1862 return 0;
1863
1864 if (!(page = __get_free_page(GFP_KERNEL)))
1865 return -ENOMEM;
1866
1867 /* We only care that *some* data at the address the user
1868 * gave us is valid. Just in case, we'll zero
1869 * the remainder of the page.
1870 */
1871 /* copy_from_user cannot cross TASK_SIZE ! */
1872 size = TASK_SIZE - (unsigned long)data;
1873 if (size > PAGE_SIZE)
1874 size = PAGE_SIZE;
1875
1876 i = size - exact_copy_from_user((void *)page, data, size);
1877 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001878 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 return -EFAULT;
1880 }
1881 if (i != PAGE_SIZE)
1882 memset((char *)page + i, 0, PAGE_SIZE - i);
1883 *where = page;
1884 return 0;
1885}
1886
1887/*
1888 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1889 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1890 *
1891 * data is a (void *) that can point to any structure up to
1892 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1893 * information (or be NULL).
1894 *
1895 * Pre-0.97 versions of mount() didn't have a flags word.
1896 * When the flags word was introduced its top half was required
1897 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1898 * Therefore, if this magic number is present, it carries no information
1899 * and must be discarded.
1900 */
Ram Paib58fed82005-11-07 17:16:09 -05001901long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 unsigned long flags, void *data_page)
1903{
Al Viro2d92ab32008-08-02 00:51:11 -04001904 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 int retval = 0;
1906 int mnt_flags = 0;
1907
1908 /* Discard magic */
1909 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1910 flags &= ~MS_MGC_MSK;
1911
1912 /* Basic sanity checks */
1913
1914 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1915 return -EINVAL;
1916 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1917 return -EINVAL;
1918
1919 if (data_page)
1920 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1921
1922 /* Separate the per-mountpoint flags */
1923 if (flags & MS_NOSUID)
1924 mnt_flags |= MNT_NOSUID;
1925 if (flags & MS_NODEV)
1926 mnt_flags |= MNT_NODEV;
1927 if (flags & MS_NOEXEC)
1928 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001929 if (flags & MS_NOATIME)
1930 mnt_flags |= MNT_NOATIME;
1931 if (flags & MS_NODIRATIME)
1932 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001933 if (flags & MS_RELATIME)
1934 mnt_flags |= MNT_RELATIME;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001935 if (flags & MS_RDONLY)
1936 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001937
1938 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001939 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 /* ... and get the mountpoint */
Al Viro2d92ab32008-08-02 00:51:11 -04001942 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 if (retval)
1944 return retval;
1945
Al Viro2d92ab32008-08-02 00:51:11 -04001946 retval = security_sb_mount(dev_name, &path,
Al Virob5266eb2008-03-22 17:48:24 -04001947 type_page, flags, data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if (retval)
1949 goto dput_out;
1950
1951 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001952 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 data_page);
1954 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001955 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001956 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04001957 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04001959 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 else
Al Viro2d92ab32008-08-02 00:51:11 -04001961 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 dev_name, data_page);
1963dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04001964 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return retval;
1966}
1967
JANAK DESAI741a2952006-02-07 12:59:00 -08001968/*
1969 * Allocate a new namespace structure and populate it with contents
1970 * copied from the namespace of the passed in task structure.
1971 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001972static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001973 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001975 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04001976 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 struct vfsmount *p, *q;
1978
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001979 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001981 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001985 init_waitqueue_head(&new_ns->poll);
1986 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Ram Pai390c6842005-11-07 17:17:51 -05001988 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001990 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001991 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001993 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08001995 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 }
1997 spin_lock(&vfsmount_lock);
1998 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1999 spin_unlock(&vfsmount_lock);
2000
2001 /*
2002 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2003 * as belonging to new namespace. We have already acquired a private
2004 * fs_struct, so tsk->fs->lock is not needed.
2005 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002006 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 q = new_ns->root;
2008 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002009 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002011 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002013 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002015 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002017 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002020 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 q = next_mnt(q, new_ns->root);
2022 }
Ram Pai390c6842005-11-07 17:17:51 -05002023 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 if (rootmnt)
2026 mntput(rootmnt);
2027 if (pwdmnt)
2028 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
JANAK DESAI741a2952006-02-07 12:59:00 -08002030 return new_ns;
2031}
2032
Eric W. Biederman213dd262007-07-15 23:41:15 -07002033struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002034 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002035{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002036 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002037
Badari Pulavartye3222c42007-05-08 00:25:21 -07002038 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002039 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002040
2041 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002042 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002043
Badari Pulavartye3222c42007-05-08 00:25:21 -07002044 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002045
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002046 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002047 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048}
2049
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002050SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2051 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052{
2053 int retval;
2054 unsigned long data_page;
2055 unsigned long type_page;
2056 unsigned long dev_page;
2057 char *dir_page;
2058
Ram Paib58fed82005-11-07 17:16:09 -05002059 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 if (retval < 0)
2061 return retval;
2062
2063 dir_page = getname(dir_name);
2064 retval = PTR_ERR(dir_page);
2065 if (IS_ERR(dir_page))
2066 goto out1;
2067
Ram Paib58fed82005-11-07 17:16:09 -05002068 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (retval < 0)
2070 goto out2;
2071
Ram Paib58fed82005-11-07 17:16:09 -05002072 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 if (retval < 0)
2074 goto out3;
2075
2076 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05002077 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2078 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 unlock_kernel();
2080 free_page(data_page);
2081
2082out3:
2083 free_page(dev_page);
2084out2:
2085 putname(dir_page);
2086out1:
2087 free_page(type_page);
2088 return retval;
2089}
2090
2091/*
2092 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2093 * It can block. Requires the big lock held.
2094 */
Jan Blunckac748a02008-02-14 19:34:39 -08002095void set_fs_root(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096{
Jan Blunck6ac08c32008-02-14 19:34:38 -08002097 struct path old_root;
2098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 write_lock(&fs->lock);
2100 old_root = fs->root;
Jan Blunckac748a02008-02-14 19:34:39 -08002101 fs->root = *path;
2102 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 write_unlock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002104 if (old_root.dentry)
2105 path_put(&old_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
2108/*
2109 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2110 * It can block. Requires the big lock held.
2111 */
Jan Blunckac748a02008-02-14 19:34:39 -08002112void set_fs_pwd(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113{
Jan Blunck6ac08c32008-02-14 19:34:38 -08002114 struct path old_pwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
2116 write_lock(&fs->lock);
2117 old_pwd = fs->pwd;
Jan Blunckac748a02008-02-14 19:34:39 -08002118 fs->pwd = *path;
2119 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 write_unlock(&fs->lock);
2121
Jan Blunck6ac08c32008-02-14 19:34:38 -08002122 if (old_pwd.dentry)
2123 path_put(&old_pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124}
2125
Al Viro1a390682008-03-21 20:48:19 -04002126static void chroot_fs_refs(struct path *old_root, struct path *new_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127{
2128 struct task_struct *g, *p;
2129 struct fs_struct *fs;
2130
2131 read_lock(&tasklist_lock);
2132 do_each_thread(g, p) {
2133 task_lock(p);
2134 fs = p->fs;
2135 if (fs) {
2136 atomic_inc(&fs->count);
2137 task_unlock(p);
Al Viro1a390682008-03-21 20:48:19 -04002138 if (fs->root.dentry == old_root->dentry
2139 && fs->root.mnt == old_root->mnt)
2140 set_fs_root(fs, new_root);
2141 if (fs->pwd.dentry == old_root->dentry
2142 && fs->pwd.mnt == old_root->mnt)
2143 set_fs_pwd(fs, new_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 put_fs_struct(fs);
2145 } else
2146 task_unlock(p);
2147 } while_each_thread(g, p);
2148 read_unlock(&tasklist_lock);
2149}
2150
2151/*
2152 * pivot_root Semantics:
2153 * Moves the root file system of the current process to the directory put_old,
2154 * makes new_root as the new root file system of the current process, and sets
2155 * root/cwd of all processes which had them on the current root to new_root.
2156 *
2157 * Restrictions:
2158 * The new_root and put_old must be directories, and must not be on the
2159 * same file system as the current process root. The put_old must be
2160 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2161 * pointed to by put_old must yield the same directory as new_root. No other
2162 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2163 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002164 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2165 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2166 * in this situation.
2167 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 * Notes:
2169 * - we don't move root/cwd if they are not at the root (reason: if something
2170 * cared enough to change them, it's probably wrong to force them elsewhere)
2171 * - it's okay to pick a root that isn't the root of a file system, e.g.
2172 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2173 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2174 * first.
2175 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002176SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2177 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178{
2179 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002180 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 int error;
2182
2183 if (!capable(CAP_SYS_ADMIN))
2184 return -EPERM;
2185
Al Viro2d8f3032008-07-22 09:59:21 -04002186 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (error)
2188 goto out0;
2189 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002190 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 goto out1;
2192
Al Viro2d8f3032008-07-22 09:59:21 -04002193 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (error)
2195 goto out1;
2196
Al Viro2d8f3032008-07-22 09:59:21 -04002197 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002199 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 goto out1;
2201 }
2202
2203 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002204 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002205 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002207 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002208 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002210 if (IS_MNT_SHARED(old.mnt) ||
2211 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002212 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002213 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002214 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 goto out2;
2216 error = -ENOENT;
Al Viro2d8f3032008-07-22 09:59:21 -04002217 if (IS_DEADDIR(new.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 goto out2;
Al Viro2d8f3032008-07-22 09:59:21 -04002219 if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 goto out2;
Al Viro2d8f3032008-07-22 09:59:21 -04002221 if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 goto out2;
2223 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002224 if (new.mnt == root.mnt ||
2225 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 goto out2; /* loop, on the same file system */
2227 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002228 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002230 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002231 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002232 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002234 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002235 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002236 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002237 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002239 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 for (;;) {
2241 if (tmp->mnt_parent == tmp)
2242 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002243 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 break;
2245 tmp = tmp->mnt_parent;
2246 }
Al Viro2d8f3032008-07-22 09:59:21 -04002247 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002249 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002251 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002252 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002253 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002254 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002255 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002256 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002257 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002259 chroot_fs_refs(&root, &new);
2260 security_sb_post_pivotroot(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002262 path_put(&root_parent);
2263 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002265 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002266 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002267 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002268 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002270 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 return error;
2273out3:
2274 spin_unlock(&vfsmount_lock);
2275 goto out2;
2276}
2277
2278static void __init init_mount_tree(void)
2279{
2280 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002281 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002282 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2285 if (IS_ERR(mnt))
2286 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002287 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2288 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002290 atomic_set(&ns->count, 1);
2291 INIT_LIST_HEAD(&ns->list);
2292 init_waitqueue_head(&ns->poll);
2293 ns->event = 0;
2294 list_add(&mnt->mnt_list, &ns->list);
2295 ns->root = mnt;
2296 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002298 init_task.nsproxy->mnt_ns = ns;
2299 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
Jan Blunckac748a02008-02-14 19:34:39 -08002301 root.mnt = ns->root;
2302 root.dentry = ns->root->mnt_root;
2303
2304 set_fs_pwd(current->fs, &root);
2305 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306}
2307
Denis Cheng74bf17c2007-10-16 23:26:30 -07002308void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002310 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002311 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Ram Pai390c6842005-11-07 17:17:51 -05002313 init_rwsem(&namespace_sem);
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002316 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Ram Paib58fed82005-11-07 17:16:09 -05002318 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 if (!mount_hashtable)
2321 panic("Failed to allocate mount hash table\n");
2322
Eric Dumazet13f14b42008-02-06 01:37:57 -08002323 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
Eric Dumazet13f14b42008-02-06 01:37:57 -08002325 for (u = 0; u < HASH_SIZE; u++)
2326 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002328 err = sysfs_init();
2329 if (err)
2330 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002331 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002332 fs_kobj = kobject_create_and_add("fs", NULL);
2333 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002334 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 init_rootfs();
2336 init_mount_tree();
2337}
2338
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002339void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002341 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05002342 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002343 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07002344 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002345 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05002347 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002349 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002350 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002351 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352}