11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/config.h> 121da177e4SLinus Torvalds #include <linux/syscalls.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 141da177e4SLinus Torvalds #include <linux/sched.h> 151da177e4SLinus Torvalds #include <linux/smp_lock.h> 161da177e4SLinus Torvalds #include <linux/init.h> 171da177e4SLinus Torvalds #include <linux/quotaops.h> 181da177e4SLinus Torvalds #include <linux/acct.h> 1916f7e0feSRandy Dunlap #include <linux/capability.h> 201da177e4SLinus Torvalds #include <linux/module.h> 211da177e4SLinus Torvalds #include <linux/seq_file.h> 221da177e4SLinus Torvalds #include <linux/namespace.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/security.h> 251da177e4SLinus Torvalds #include <linux/mount.h> 261da177e4SLinus Torvalds #include <asm/uaccess.h> 271da177e4SLinus Torvalds #include <asm/unistd.h> 2807b20889SRam Pai #include "pnode.h" 291da177e4SLinus Torvalds 301da177e4SLinus Torvalds extern int __init init_rootfs(void); 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #ifdef CONFIG_SYSFS 331da177e4SLinus Torvalds extern int __init sysfs_init(void); 341da177e4SLinus Torvalds #else 351da177e4SLinus Torvalds static inline int sysfs_init(void) 361da177e4SLinus Torvalds { 371da177e4SLinus Torvalds return 0; 381da177e4SLinus Torvalds } 391da177e4SLinus Torvalds #endif 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 421da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 431da177e4SLinus Torvalds 445addc5ddSAl Viro static int event; 455addc5ddSAl Viro 46fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 476c231b7bSRavikiran G Thirumalai static int hash_mask __read_mostly, hash_bits __read_mostly; 48fa3536ccSEric Dumazet static kmem_cache_t *mnt_cache __read_mostly; 49390c6843SRam Pai static struct rw_semaphore namespace_sem; 501da177e4SLinus Torvalds 51f87fd4c2SMiklos Szeredi /* /sys/fs */ 52f87fd4c2SMiklos Szeredi decl_subsys(fs, NULL, NULL); 53f87fd4c2SMiklos Szeredi EXPORT_SYMBOL_GPL(fs_subsys); 54f87fd4c2SMiklos Szeredi 551da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 561da177e4SLinus Torvalds { 571da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 581da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 591da177e4SLinus Torvalds tmp = tmp + (tmp >> hash_bits); 601da177e4SLinus Torvalds return tmp & hash_mask; 611da177e4SLinus Torvalds } 621da177e4SLinus Torvalds 631da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 641da177e4SLinus Torvalds { 651da177e4SLinus Torvalds struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); 661da177e4SLinus Torvalds if (mnt) { 671da177e4SLinus Torvalds memset(mnt, 0, sizeof(struct vfsmount)); 681da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 691da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 701da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 711da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 721da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 7355e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 7403e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 75a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 76a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 771da177e4SLinus Torvalds if (name) { 781da177e4SLinus Torvalds int size = strlen(name) + 1; 791da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 801da177e4SLinus Torvalds if (newname) { 811da177e4SLinus Torvalds memcpy(newname, name, size); 821da177e4SLinus Torvalds mnt->mnt_devname = newname; 831da177e4SLinus Torvalds } 841da177e4SLinus Torvalds } 851da177e4SLinus Torvalds } 861da177e4SLinus Torvalds return mnt; 871da177e4SLinus Torvalds } 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 901da177e4SLinus Torvalds { 911da177e4SLinus Torvalds kfree(mnt->mnt_devname); 921da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 931da177e4SLinus Torvalds } 941da177e4SLinus Torvalds 951da177e4SLinus Torvalds /* 96a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 97a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 981da177e4SLinus Torvalds */ 99a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 100a05964f3SRam Pai int dir) 1011da177e4SLinus Torvalds { 1021da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 1031da177e4SLinus Torvalds struct list_head *tmp = head; 1041da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds for (;;) { 107a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 1081da177e4SLinus Torvalds p = NULL; 1091da177e4SLinus Torvalds if (tmp == head) 1101da177e4SLinus Torvalds break; 1111da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 1121da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 113a05964f3SRam Pai found = p; 1141da177e4SLinus Torvalds break; 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds } 1171da177e4SLinus Torvalds return found; 1181da177e4SLinus Torvalds } 1191da177e4SLinus Torvalds 120a05964f3SRam Pai /* 121a05964f3SRam Pai * lookup_mnt increments the ref count before returning 122a05964f3SRam Pai * the vfsmount struct. 123a05964f3SRam Pai */ 124a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 125a05964f3SRam Pai { 126a05964f3SRam Pai struct vfsmount *child_mnt; 127a05964f3SRam Pai spin_lock(&vfsmount_lock); 128a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 129a05964f3SRam Pai mntget(child_mnt); 130a05964f3SRam Pai spin_unlock(&vfsmount_lock); 131a05964f3SRam Pai return child_mnt; 132a05964f3SRam Pai } 133a05964f3SRam Pai 1341da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 1351da177e4SLinus Torvalds { 1361da177e4SLinus Torvalds return mnt->mnt_namespace == current->namespace; 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds 1395addc5ddSAl Viro static void touch_namespace(struct namespace *ns) 1405addc5ddSAl Viro { 1415addc5ddSAl Viro if (ns) { 1425addc5ddSAl Viro ns->event = ++event; 1435addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1445addc5ddSAl Viro } 1455addc5ddSAl Viro } 1465addc5ddSAl Viro 1475addc5ddSAl Viro static void __touch_namespace(struct namespace *ns) 1485addc5ddSAl Viro { 1495addc5ddSAl Viro if (ns && ns->event != event) { 1505addc5ddSAl Viro ns->event = event; 1515addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1525addc5ddSAl Viro } 1535addc5ddSAl Viro } 1545addc5ddSAl Viro 1551da177e4SLinus Torvalds static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) 1561da177e4SLinus Torvalds { 1571da177e4SLinus Torvalds old_nd->dentry = mnt->mnt_mountpoint; 1581da177e4SLinus Torvalds old_nd->mnt = mnt->mnt_parent; 1591da177e4SLinus Torvalds mnt->mnt_parent = mnt; 1601da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 1611da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 1621da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 1631da177e4SLinus Torvalds old_nd->dentry->d_mounted--; 1641da177e4SLinus Torvalds } 1651da177e4SLinus Torvalds 166b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 167b90fa9aeSRam Pai struct vfsmount *child_mnt) 168b90fa9aeSRam Pai { 169b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 170b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 171b90fa9aeSRam Pai dentry->d_mounted++; 172b90fa9aeSRam Pai } 173b90fa9aeSRam Pai 1741da177e4SLinus Torvalds static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) 1751da177e4SLinus Torvalds { 176b90fa9aeSRam Pai mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); 177b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 178b90fa9aeSRam Pai hash(nd->mnt, nd->dentry)); 1791da177e4SLinus Torvalds list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); 180b90fa9aeSRam Pai } 181b90fa9aeSRam Pai 182b90fa9aeSRam Pai /* 183b90fa9aeSRam Pai * the caller must hold vfsmount_lock 184b90fa9aeSRam Pai */ 185b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 186b90fa9aeSRam Pai { 187b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 188b90fa9aeSRam Pai struct vfsmount *m; 189b90fa9aeSRam Pai LIST_HEAD(head); 190b90fa9aeSRam Pai struct namespace *n = parent->mnt_namespace; 191b90fa9aeSRam Pai 192b90fa9aeSRam Pai BUG_ON(parent == mnt); 193b90fa9aeSRam Pai 194b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 195b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 196b90fa9aeSRam Pai m->mnt_namespace = n; 197b90fa9aeSRam Pai list_splice(&head, n->list.prev); 198b90fa9aeSRam Pai 199b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 200b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 201b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 202b90fa9aeSRam Pai touch_namespace(n); 2031da177e4SLinus Torvalds } 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 2061da177e4SLinus Torvalds { 2071da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 2081da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 2091da177e4SLinus Torvalds while (1) { 2101da177e4SLinus Torvalds if (p == root) 2111da177e4SLinus Torvalds return NULL; 2121da177e4SLinus Torvalds next = p->mnt_child.next; 2131da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 2141da177e4SLinus Torvalds break; 2151da177e4SLinus Torvalds p = p->mnt_parent; 2161da177e4SLinus Torvalds } 2171da177e4SLinus Torvalds } 2181da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds 2219676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 2229676f0c6SRam Pai { 2239676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 2249676f0c6SRam Pai while (prev != &p->mnt_mounts) { 2259676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 2269676f0c6SRam Pai prev = p->mnt_mounts.prev; 2279676f0c6SRam Pai } 2289676f0c6SRam Pai return p; 2299676f0c6SRam Pai } 2309676f0c6SRam Pai 23136341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 23236341f64SRam Pai int flag) 2331da177e4SLinus Torvalds { 2341da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 2351da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds if (mnt) { 2381da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 2391da177e4SLinus Torvalds atomic_inc(&sb->s_active); 2401da177e4SLinus Torvalds mnt->mnt_sb = sb; 2411da177e4SLinus Torvalds mnt->mnt_root = dget(root); 2421da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 2431da177e4SLinus Torvalds mnt->mnt_parent = mnt; 244b90fa9aeSRam Pai 2455afe0022SRam Pai if (flag & CL_SLAVE) { 2465afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 2475afe0022SRam Pai mnt->mnt_master = old; 2485afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 2495afe0022SRam Pai } else { 250b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 251b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 2525afe0022SRam Pai if (IS_MNT_SLAVE(old)) 2535afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 2545afe0022SRam Pai mnt->mnt_master = old->mnt_master; 2555afe0022SRam Pai } 256b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 257b90fa9aeSRam Pai set_mnt_shared(mnt); 2581da177e4SLinus Torvalds 2591da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 2601da177e4SLinus Torvalds * as the original if that was on one */ 26136341f64SRam Pai if (flag & CL_EXPIRE) { 2621da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 26355e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 26455e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 2651da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2661da177e4SLinus Torvalds } 26736341f64SRam Pai } 2681da177e4SLinus Torvalds return mnt; 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2717b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 2721da177e4SLinus Torvalds { 2731da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 2741da177e4SLinus Torvalds dput(mnt->mnt_root); 2751da177e4SLinus Torvalds free_vfsmnt(mnt); 2761da177e4SLinus Torvalds deactivate_super(sb); 2771da177e4SLinus Torvalds } 2781da177e4SLinus Torvalds 2797b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 2807b7b1aceSAl Viro { 2817b7b1aceSAl Viro repeat: 2827b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 2837b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 2847b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2857b7b1aceSAl Viro __mntput(mnt); 2867b7b1aceSAl Viro return; 2877b7b1aceSAl Viro } 2887b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 2897b7b1aceSAl Viro mnt->mnt_pinned = 0; 2907b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2917b7b1aceSAl Viro acct_auto_close_mnt(mnt); 2927b7b1aceSAl Viro security_sb_umount_close(mnt); 2937b7b1aceSAl Viro goto repeat; 2947b7b1aceSAl Viro } 2957b7b1aceSAl Viro } 2967b7b1aceSAl Viro 2977b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 2987b7b1aceSAl Viro 2997b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 3007b7b1aceSAl Viro { 3017b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3027b7b1aceSAl Viro mnt->mnt_pinned++; 3037b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3047b7b1aceSAl Viro } 3057b7b1aceSAl Viro 3067b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 3077b7b1aceSAl Viro 3087b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 3097b7b1aceSAl Viro { 3107b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3117b7b1aceSAl Viro if (mnt->mnt_pinned) { 3127b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 3137b7b1aceSAl Viro mnt->mnt_pinned--; 3147b7b1aceSAl Viro } 3157b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3167b7b1aceSAl Viro } 3177b7b1aceSAl Viro 3187b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 3191da177e4SLinus Torvalds 3201da177e4SLinus Torvalds /* iterator */ 3211da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 3221da177e4SLinus Torvalds { 3231da177e4SLinus Torvalds struct namespace *n = m->private; 3241da177e4SLinus Torvalds struct list_head *p; 3251da177e4SLinus Torvalds loff_t l = *pos; 3261da177e4SLinus Torvalds 327390c6843SRam Pai down_read(&namespace_sem); 3281da177e4SLinus Torvalds list_for_each(p, &n->list) 3291da177e4SLinus Torvalds if (!l--) 3301da177e4SLinus Torvalds return list_entry(p, struct vfsmount, mnt_list); 3311da177e4SLinus Torvalds return NULL; 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 3351da177e4SLinus Torvalds { 3361da177e4SLinus Torvalds struct namespace *n = m->private; 3371da177e4SLinus Torvalds struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; 3381da177e4SLinus Torvalds (*pos)++; 3391da177e4SLinus Torvalds return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 3431da177e4SLinus Torvalds { 344390c6843SRam Pai up_read(&namespace_sem); 3451da177e4SLinus Torvalds } 3461da177e4SLinus Torvalds 3471da177e4SLinus Torvalds static inline void mangle(struct seq_file *m, const char *s) 3481da177e4SLinus Torvalds { 3491da177e4SLinus Torvalds seq_escape(m, s, " \t\n\\"); 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 3531da177e4SLinus Torvalds { 3541da177e4SLinus Torvalds struct vfsmount *mnt = v; 3551da177e4SLinus Torvalds int err = 0; 3561da177e4SLinus Torvalds static struct proc_fs_info { 3571da177e4SLinus Torvalds int flag; 3581da177e4SLinus Torvalds char *str; 3591da177e4SLinus Torvalds } fs_info[] = { 3601da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 3611da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 3621da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 3631da177e4SLinus Torvalds { 0, NULL } 3641da177e4SLinus Torvalds }; 3651da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 3661da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 3671da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 3681da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 369fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 370fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 3711da177e4SLinus Torvalds { 0, NULL } 3721da177e4SLinus Torvalds }; 3731da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 3741da177e4SLinus Torvalds 3751da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 3761da177e4SLinus Torvalds seq_putc(m, ' '); 3771da177e4SLinus Torvalds seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 3781da177e4SLinus Torvalds seq_putc(m, ' '); 3791da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 3801da177e4SLinus Torvalds seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 3811da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 3821da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 3831da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 3861da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 3871da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 3901da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 3911da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 3921da177e4SLinus Torvalds return err; 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 3951da177e4SLinus Torvalds struct seq_operations mounts_op = { 3961da177e4SLinus Torvalds .start = m_start, 3971da177e4SLinus Torvalds .next = m_next, 3981da177e4SLinus Torvalds .stop = m_stop, 3991da177e4SLinus Torvalds .show = show_vfsmnt 4001da177e4SLinus Torvalds }; 4011da177e4SLinus Torvalds 402b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 403b4629fe2SChuck Lever { 404b4629fe2SChuck Lever struct vfsmount *mnt = v; 405b4629fe2SChuck Lever int err = 0; 406b4629fe2SChuck Lever 407b4629fe2SChuck Lever /* device */ 408b4629fe2SChuck Lever if (mnt->mnt_devname) { 409b4629fe2SChuck Lever seq_puts(m, "device "); 410b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 411b4629fe2SChuck Lever } else 412b4629fe2SChuck Lever seq_puts(m, "no device"); 413b4629fe2SChuck Lever 414b4629fe2SChuck Lever /* mount point */ 415b4629fe2SChuck Lever seq_puts(m, " mounted on "); 416b4629fe2SChuck Lever seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 417b4629fe2SChuck Lever seq_putc(m, ' '); 418b4629fe2SChuck Lever 419b4629fe2SChuck Lever /* file system type */ 420b4629fe2SChuck Lever seq_puts(m, "with fstype "); 421b4629fe2SChuck Lever mangle(m, mnt->mnt_sb->s_type->name); 422b4629fe2SChuck Lever 423b4629fe2SChuck Lever /* optional statistics */ 424b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 425b4629fe2SChuck Lever seq_putc(m, ' '); 426b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 427b4629fe2SChuck Lever } 428b4629fe2SChuck Lever 429b4629fe2SChuck Lever seq_putc(m, '\n'); 430b4629fe2SChuck Lever return err; 431b4629fe2SChuck Lever } 432b4629fe2SChuck Lever 433b4629fe2SChuck Lever struct seq_operations mountstats_op = { 434b4629fe2SChuck Lever .start = m_start, 435b4629fe2SChuck Lever .next = m_next, 436b4629fe2SChuck Lever .stop = m_stop, 437b4629fe2SChuck Lever .show = show_vfsstat, 438b4629fe2SChuck Lever }; 439b4629fe2SChuck Lever 4401da177e4SLinus Torvalds /** 4411da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 4421da177e4SLinus Torvalds * @mnt: root of mount tree 4431da177e4SLinus Torvalds * 4441da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 4451da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 4461da177e4SLinus Torvalds * busy. 4471da177e4SLinus Torvalds */ 4481da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 4491da177e4SLinus Torvalds { 45036341f64SRam Pai int actual_refs = 0; 45136341f64SRam Pai int minimum_refs = 0; 45236341f64SRam Pai struct vfsmount *p; 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 45536341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 4561da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 4571da177e4SLinus Torvalds minimum_refs += 2; 4581da177e4SLinus Torvalds } 4591da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds if (actual_refs > minimum_refs) 4621da177e4SLinus Torvalds return 0; 463*e3474a8eSIan Kent 464*e3474a8eSIan Kent return 1; 4651da177e4SLinus Torvalds } 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds /** 4701da177e4SLinus Torvalds * may_umount - check if a mount point is busy 4711da177e4SLinus Torvalds * @mnt: root of mount 4721da177e4SLinus Torvalds * 4731da177e4SLinus Torvalds * This is called to check if a mount point has any 4741da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 4751da177e4SLinus Torvalds * mount has sub mounts this will return busy 4761da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 4771da177e4SLinus Torvalds * 4781da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 4791da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 4801da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 4811da177e4SLinus Torvalds */ 4821da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 4831da177e4SLinus Torvalds { 484*e3474a8eSIan Kent int ret = 1; 485a05964f3SRam Pai spin_lock(&vfsmount_lock); 486a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 487*e3474a8eSIan Kent ret = 0; 488a05964f3SRam Pai spin_unlock(&vfsmount_lock); 489a05964f3SRam Pai return ret; 4901da177e4SLinus Torvalds } 4911da177e4SLinus Torvalds 4921da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 4931da177e4SLinus Torvalds 494b90fa9aeSRam Pai void release_mounts(struct list_head *head) 4951da177e4SLinus Torvalds { 49670fbcdf4SRam Pai struct vfsmount *mnt; 49770fbcdf4SRam Pai while (!list_empty(head)) { 49870fbcdf4SRam Pai mnt = list_entry(head->next, struct vfsmount, mnt_hash); 49970fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 50070fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 50170fbcdf4SRam Pai struct dentry *dentry; 50270fbcdf4SRam Pai struct vfsmount *m; 50370fbcdf4SRam Pai spin_lock(&vfsmount_lock); 50470fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 50570fbcdf4SRam Pai m = mnt->mnt_parent; 50670fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 50770fbcdf4SRam Pai mnt->mnt_parent = mnt; 5081da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 50970fbcdf4SRam Pai dput(dentry); 51070fbcdf4SRam Pai mntput(m); 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds mntput(mnt); 51370fbcdf4SRam Pai } 51470fbcdf4SRam Pai } 51570fbcdf4SRam Pai 516a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 51770fbcdf4SRam Pai { 51870fbcdf4SRam Pai struct vfsmount *p; 51970fbcdf4SRam Pai 52070fbcdf4SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 52170fbcdf4SRam Pai list_del(&p->mnt_hash); 52270fbcdf4SRam Pai list_add(&p->mnt_hash, kill); 52370fbcdf4SRam Pai } 52470fbcdf4SRam Pai 525a05964f3SRam Pai if (propagate) 526a05964f3SRam Pai propagate_umount(kill); 527a05964f3SRam Pai 52870fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 52970fbcdf4SRam Pai list_del_init(&p->mnt_expire); 53070fbcdf4SRam Pai list_del_init(&p->mnt_list); 53170fbcdf4SRam Pai __touch_namespace(p->mnt_namespace); 53270fbcdf4SRam Pai p->mnt_namespace = NULL; 53370fbcdf4SRam Pai list_del_init(&p->mnt_child); 53470fbcdf4SRam Pai if (p->mnt_parent != p) 535f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 536a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 5411da177e4SLinus Torvalds { 5421da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 5431da177e4SLinus Torvalds int retval; 54470fbcdf4SRam Pai LIST_HEAD(umount_list); 5451da177e4SLinus Torvalds 5461da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 5471da177e4SLinus Torvalds if (retval) 5481da177e4SLinus Torvalds return retval; 5491da177e4SLinus Torvalds 5501da177e4SLinus Torvalds /* 5511da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 5521da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 5531da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 5541da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 5551da177e4SLinus Torvalds */ 5561da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 5571da177e4SLinus Torvalds if (mnt == current->fs->rootmnt || 5581da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 5591da177e4SLinus Torvalds return -EINVAL; 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 5621da177e4SLinus Torvalds return -EBUSY; 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 5651da177e4SLinus Torvalds return -EAGAIN; 5661da177e4SLinus Torvalds } 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds /* 5691da177e4SLinus Torvalds * If we may have to abort operations to get out of this 5701da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 5711da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 5721da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 5731da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 5741da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 5751da177e4SLinus Torvalds * about for the moment. 5761da177e4SLinus Torvalds */ 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds lock_kernel(); 5791da177e4SLinus Torvalds if ((flags & MNT_FORCE) && sb->s_op->umount_begin) 5801da177e4SLinus Torvalds sb->s_op->umount_begin(sb); 5811da177e4SLinus Torvalds unlock_kernel(); 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds /* 5841da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 5851da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 5861da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 5871da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 5881da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 5891da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 5901da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 5911da177e4SLinus Torvalds */ 5921da177e4SLinus Torvalds if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { 5931da177e4SLinus Torvalds /* 5941da177e4SLinus Torvalds * Special case for "unmounting" root ... 5951da177e4SLinus Torvalds * we just try to remount it readonly. 5961da177e4SLinus Torvalds */ 5971da177e4SLinus Torvalds down_write(&sb->s_umount); 5981da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 5991da177e4SLinus Torvalds lock_kernel(); 6001da177e4SLinus Torvalds DQUOT_OFF(sb); 6011da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 6021da177e4SLinus Torvalds unlock_kernel(); 6031da177e4SLinus Torvalds } 6041da177e4SLinus Torvalds up_write(&sb->s_umount); 6051da177e4SLinus Torvalds return retval; 6061da177e4SLinus Torvalds } 6071da177e4SLinus Torvalds 608390c6843SRam Pai down_write(&namespace_sem); 6091da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6105addc5ddSAl Viro event++; 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds retval = -EBUSY; 613a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 6141da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 615a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 6161da177e4SLinus Torvalds retval = 0; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6191da177e4SLinus Torvalds if (retval) 6201da177e4SLinus Torvalds security_sb_umount_busy(mnt); 621390c6843SRam Pai up_write(&namespace_sem); 62270fbcdf4SRam Pai release_mounts(&umount_list); 6231da177e4SLinus Torvalds return retval; 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds 6261da177e4SLinus Torvalds /* 6271da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 6281da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 6291da177e4SLinus Torvalds * 6301da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 6311da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 6321da177e4SLinus Torvalds */ 6331da177e4SLinus Torvalds 6341da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 6351da177e4SLinus Torvalds { 6361da177e4SLinus Torvalds struct nameidata nd; 6371da177e4SLinus Torvalds int retval; 6381da177e4SLinus Torvalds 6391da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 6401da177e4SLinus Torvalds if (retval) 6411da177e4SLinus Torvalds goto out; 6421da177e4SLinus Torvalds retval = -EINVAL; 6431da177e4SLinus Torvalds if (nd.dentry != nd.mnt->mnt_root) 6441da177e4SLinus Torvalds goto dput_and_out; 6451da177e4SLinus Torvalds if (!check_mnt(nd.mnt)) 6461da177e4SLinus Torvalds goto dput_and_out; 6471da177e4SLinus Torvalds 6481da177e4SLinus Torvalds retval = -EPERM; 6491da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 6501da177e4SLinus Torvalds goto dput_and_out; 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds retval = do_umount(nd.mnt, flags); 6531da177e4SLinus Torvalds dput_and_out: 6541da177e4SLinus Torvalds path_release_on_umount(&nd); 6551da177e4SLinus Torvalds out: 6561da177e4SLinus Torvalds return retval; 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds /* 6621da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 6631da177e4SLinus Torvalds */ 6641da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 6651da177e4SLinus Torvalds { 6661da177e4SLinus Torvalds return sys_umount(name, 0); 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds #endif 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 6721da177e4SLinus Torvalds { 6731da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 6741da177e4SLinus Torvalds return 0; 6751da177e4SLinus Torvalds return -EPERM; 6761da177e4SLinus Torvalds #ifdef notyet 6771da177e4SLinus Torvalds if (S_ISLNK(nd->dentry->d_inode->i_mode)) 6781da177e4SLinus Torvalds return -EPERM; 6791da177e4SLinus Torvalds if (nd->dentry->d_inode->i_mode & S_ISVTX) { 6801da177e4SLinus Torvalds if (current->uid != nd->dentry->d_inode->i_uid) 6811da177e4SLinus Torvalds return -EPERM; 6821da177e4SLinus Torvalds } 683e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 6841da177e4SLinus Torvalds return -EPERM; 6851da177e4SLinus Torvalds return 0; 6861da177e4SLinus Torvalds #endif 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds 689b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 6901da177e4SLinus Torvalds { 6911da177e4SLinus Torvalds while (1) { 6921da177e4SLinus Torvalds if (d == dentry) 6931da177e4SLinus Torvalds return 1; 6941da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 6951da177e4SLinus Torvalds return 0; 6961da177e4SLinus Torvalds d = d->d_parent; 6971da177e4SLinus Torvalds } 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 700b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 70136341f64SRam Pai int flag) 7021da177e4SLinus Torvalds { 7031da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 7041da177e4SLinus Torvalds struct nameidata nd; 7051da177e4SLinus Torvalds 7069676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 7079676f0c6SRam Pai return NULL; 7089676f0c6SRam Pai 70936341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 7101da177e4SLinus Torvalds if (!q) 7111da177e4SLinus Torvalds goto Enomem; 7121da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 7131da177e4SLinus Torvalds 7141da177e4SLinus Torvalds p = mnt; 715fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 7161da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 7171da177e4SLinus Torvalds continue; 7181da177e4SLinus Torvalds 7191da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 7209676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 7219676f0c6SRam Pai s = skip_mnt_tree(s); 7229676f0c6SRam Pai continue; 7239676f0c6SRam Pai } 7241da177e4SLinus Torvalds while (p != s->mnt_parent) { 7251da177e4SLinus Torvalds p = p->mnt_parent; 7261da177e4SLinus Torvalds q = q->mnt_parent; 7271da177e4SLinus Torvalds } 7281da177e4SLinus Torvalds p = s; 7291da177e4SLinus Torvalds nd.mnt = q; 7301da177e4SLinus Torvalds nd.dentry = p->mnt_mountpoint; 73136341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 7321da177e4SLinus Torvalds if (!q) 7331da177e4SLinus Torvalds goto Enomem; 7341da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 7351da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 7361da177e4SLinus Torvalds attach_mnt(q, &nd); 7371da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds return res; 7411da177e4SLinus Torvalds Enomem: 7421da177e4SLinus Torvalds if (res) { 74370fbcdf4SRam Pai LIST_HEAD(umount_list); 7441da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 745a05964f3SRam Pai umount_tree(res, 0, &umount_list); 7461da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 74770fbcdf4SRam Pai release_mounts(&umount_list); 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds return NULL; 7501da177e4SLinus Torvalds } 7511da177e4SLinus Torvalds 752b90fa9aeSRam Pai /* 753b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 754b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 75521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 75621444403SRam Pai * store the parent mount and mountpoint dentry. 75721444403SRam Pai * (done when source_mnt is moved) 758b90fa9aeSRam Pai * 759b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 760b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 7619676f0c6SRam Pai * --------------------------------------------------------------------------- 762b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 7639676f0c6SRam Pai * |************************************************************************** 7649676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7659676f0c6SRam Pai * | dest | | | | | 7669676f0c6SRam Pai * | | | | | | | 7679676f0c6SRam Pai * | v | | | | | 7689676f0c6SRam Pai * |************************************************************************** 7699676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 7705afe0022SRam Pai * | | | | | | 7719676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 7729676f0c6SRam Pai * *************************************************************************** 773b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 774b90fa9aeSRam Pai * destination mount. 775b90fa9aeSRam Pai * 776b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 777b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 778b90fa9aeSRam Pai * the peer group of the source mount. 779b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 780b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 781b90fa9aeSRam Pai * mount. 7825afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 7835afe0022SRam Pai * of the destination mount and the cloned mount is made slave 7845afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 7855afe0022SRam Pai * is marked as 'shared and slave'. 7865afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 7875afe0022SRam Pai * source mount. 7885afe0022SRam Pai * 7899676f0c6SRam Pai * --------------------------------------------------------------------------- 79021444403SRam Pai * | MOVE MOUNT OPERATION | 7919676f0c6SRam Pai * |************************************************************************** 7929676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7939676f0c6SRam Pai * | dest | | | | | 7949676f0c6SRam Pai * | | | | | | | 7959676f0c6SRam Pai * | v | | | | | 7969676f0c6SRam Pai * |************************************************************************** 7979676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 7985afe0022SRam Pai * | | | | | | 7999676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 8009676f0c6SRam Pai * *************************************************************************** 8015afe0022SRam Pai * 8025afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 8035afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 80421444403SRam Pai * (+*) the mount is moved to the destination. 8055afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 8065afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 8075afe0022SRam Pai * the mount is marked as 'shared and slave'. 8085afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 809b90fa9aeSRam Pai * 810b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 811b90fa9aeSRam Pai * applied to each mount in the tree. 812b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 813b90fa9aeSRam Pai * in allocations. 814b90fa9aeSRam Pai */ 815b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 81621444403SRam Pai struct nameidata *nd, struct nameidata *parent_nd) 817b90fa9aeSRam Pai { 818b90fa9aeSRam Pai LIST_HEAD(tree_list); 819b90fa9aeSRam Pai struct vfsmount *dest_mnt = nd->mnt; 820b90fa9aeSRam Pai struct dentry *dest_dentry = nd->dentry; 821b90fa9aeSRam Pai struct vfsmount *child, *p; 822b90fa9aeSRam Pai 823b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 824b90fa9aeSRam Pai return -EINVAL; 825b90fa9aeSRam Pai 826b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 827b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 828b90fa9aeSRam Pai set_mnt_shared(p); 829b90fa9aeSRam Pai } 830b90fa9aeSRam Pai 831b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 83221444403SRam Pai if (parent_nd) { 83321444403SRam Pai detach_mnt(source_mnt, parent_nd); 83421444403SRam Pai attach_mnt(source_mnt, nd); 83521444403SRam Pai touch_namespace(current->namespace); 83621444403SRam Pai } else { 837b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 838b90fa9aeSRam Pai commit_tree(source_mnt); 83921444403SRam Pai } 840b90fa9aeSRam Pai 841b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 842b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 843b90fa9aeSRam Pai commit_tree(child); 844b90fa9aeSRam Pai } 845b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 846b90fa9aeSRam Pai return 0; 847b90fa9aeSRam Pai } 848b90fa9aeSRam Pai 8491da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 8501da177e4SLinus Torvalds { 8511da177e4SLinus Torvalds int err; 8521da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 8531da177e4SLinus Torvalds return -EINVAL; 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 8561da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 8571da177e4SLinus Torvalds return -ENOTDIR; 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds err = -ENOENT; 8601b1dcc1bSJes Sorensen mutex_lock(&nd->dentry->d_inode->i_mutex); 8611da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 8621da177e4SLinus Torvalds goto out_unlock; 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 8651da177e4SLinus Torvalds if (err) 8661da177e4SLinus Torvalds goto out_unlock; 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds err = -ENOENT; 869b90fa9aeSRam Pai if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) 87021444403SRam Pai err = attach_recursive_mnt(mnt, nd, NULL); 8711da177e4SLinus Torvalds out_unlock: 8721b1dcc1bSJes Sorensen mutex_unlock(&nd->dentry->d_inode->i_mutex); 8731da177e4SLinus Torvalds if (!err) 8741da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 8751da177e4SLinus Torvalds return err; 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds /* 87907b20889SRam Pai * recursively change the type of the mountpoint. 88007b20889SRam Pai */ 88107b20889SRam Pai static int do_change_type(struct nameidata *nd, int flag) 88207b20889SRam Pai { 88307b20889SRam Pai struct vfsmount *m, *mnt = nd->mnt; 88407b20889SRam Pai int recurse = flag & MS_REC; 88507b20889SRam Pai int type = flag & ~MS_REC; 88607b20889SRam Pai 88707b20889SRam Pai if (nd->dentry != nd->mnt->mnt_root) 88807b20889SRam Pai return -EINVAL; 88907b20889SRam Pai 89007b20889SRam Pai down_write(&namespace_sem); 89107b20889SRam Pai spin_lock(&vfsmount_lock); 89207b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 89307b20889SRam Pai change_mnt_propagation(m, type); 89407b20889SRam Pai spin_unlock(&vfsmount_lock); 89507b20889SRam Pai up_write(&namespace_sem); 89607b20889SRam Pai return 0; 89707b20889SRam Pai } 89807b20889SRam Pai 89907b20889SRam Pai /* 9001da177e4SLinus Torvalds * do loopback mount. 9011da177e4SLinus Torvalds */ 9021da177e4SLinus Torvalds static int do_loopback(struct nameidata *nd, char *old_name, int recurse) 9031da177e4SLinus Torvalds { 9041da177e4SLinus Torvalds struct nameidata old_nd; 9051da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 9061da177e4SLinus Torvalds int err = mount_is_safe(nd); 9071da177e4SLinus Torvalds if (err) 9081da177e4SLinus Torvalds return err; 9091da177e4SLinus Torvalds if (!old_name || !*old_name) 9101da177e4SLinus Torvalds return -EINVAL; 9111da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 9121da177e4SLinus Torvalds if (err) 9131da177e4SLinus Torvalds return err; 9141da177e4SLinus Torvalds 915390c6843SRam Pai down_write(&namespace_sem); 9161da177e4SLinus Torvalds err = -EINVAL; 9179676f0c6SRam Pai if (IS_MNT_UNBINDABLE(old_nd.mnt)) 9189676f0c6SRam Pai goto out; 9199676f0c6SRam Pai 920ccd48bc7SAl Viro if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 921ccd48bc7SAl Viro goto out; 922ccd48bc7SAl Viro 9231da177e4SLinus Torvalds err = -ENOMEM; 9241da177e4SLinus Torvalds if (recurse) 92536341f64SRam Pai mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); 9261da177e4SLinus Torvalds else 92736341f64SRam Pai mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); 9281da177e4SLinus Torvalds 929ccd48bc7SAl Viro if (!mnt) 930ccd48bc7SAl Viro goto out; 931ccd48bc7SAl Viro 9321da177e4SLinus Torvalds err = graft_tree(mnt, nd); 9331da177e4SLinus Torvalds if (err) { 93470fbcdf4SRam Pai LIST_HEAD(umount_list); 9351da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 936a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 9371da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 93870fbcdf4SRam Pai release_mounts(&umount_list); 9395b83d2c5SRam Pai } 9401da177e4SLinus Torvalds 941ccd48bc7SAl Viro out: 942390c6843SRam Pai up_write(&namespace_sem); 9431da177e4SLinus Torvalds path_release(&old_nd); 9441da177e4SLinus Torvalds return err; 9451da177e4SLinus Torvalds } 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds /* 9481da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 9491da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 9501da177e4SLinus Torvalds * on it - tough luck. 9511da177e4SLinus Torvalds */ 9521da177e4SLinus Torvalds static int do_remount(struct nameidata *nd, int flags, int mnt_flags, 9531da177e4SLinus Torvalds void *data) 9541da177e4SLinus Torvalds { 9551da177e4SLinus Torvalds int err; 9561da177e4SLinus Torvalds struct super_block *sb = nd->mnt->mnt_sb; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9591da177e4SLinus Torvalds return -EPERM; 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 9621da177e4SLinus Torvalds return -EINVAL; 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds if (nd->dentry != nd->mnt->mnt_root) 9651da177e4SLinus Torvalds return -EINVAL; 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds down_write(&sb->s_umount); 9681da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 9691da177e4SLinus Torvalds if (!err) 9701da177e4SLinus Torvalds nd->mnt->mnt_flags = mnt_flags; 9711da177e4SLinus Torvalds up_write(&sb->s_umount); 9721da177e4SLinus Torvalds if (!err) 9731da177e4SLinus Torvalds security_sb_post_remount(nd->mnt, flags, data); 9741da177e4SLinus Torvalds return err; 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds 9779676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 9789676f0c6SRam Pai { 9799676f0c6SRam Pai struct vfsmount *p; 9809676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9819676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 9829676f0c6SRam Pai return 1; 9839676f0c6SRam Pai } 9849676f0c6SRam Pai return 0; 9859676f0c6SRam Pai } 9869676f0c6SRam Pai 9871da177e4SLinus Torvalds static int do_move_mount(struct nameidata *nd, char *old_name) 9881da177e4SLinus Torvalds { 9891da177e4SLinus Torvalds struct nameidata old_nd, parent_nd; 9901da177e4SLinus Torvalds struct vfsmount *p; 9911da177e4SLinus Torvalds int err = 0; 9921da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9931da177e4SLinus Torvalds return -EPERM; 9941da177e4SLinus Torvalds if (!old_name || !*old_name) 9951da177e4SLinus Torvalds return -EINVAL; 9961da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 9971da177e4SLinus Torvalds if (err) 9981da177e4SLinus Torvalds return err; 9991da177e4SLinus Torvalds 1000390c6843SRam Pai down_write(&namespace_sem); 10011da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 10021da177e4SLinus Torvalds ; 10031da177e4SLinus Torvalds err = -EINVAL; 10041da177e4SLinus Torvalds if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 10051da177e4SLinus Torvalds goto out; 10061da177e4SLinus Torvalds 10071da177e4SLinus Torvalds err = -ENOENT; 10081b1dcc1bSJes Sorensen mutex_lock(&nd->dentry->d_inode->i_mutex); 10091da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 10101da177e4SLinus Torvalds goto out1; 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) 101321444403SRam Pai goto out1; 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds err = -EINVAL; 10161da177e4SLinus Torvalds if (old_nd.dentry != old_nd.mnt->mnt_root) 101721444403SRam Pai goto out1; 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds if (old_nd.mnt == old_nd.mnt->mnt_parent) 102021444403SRam Pai goto out1; 10211da177e4SLinus Torvalds 10221da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 10231da177e4SLinus Torvalds S_ISDIR(old_nd.dentry->d_inode->i_mode)) 102421444403SRam Pai goto out1; 102521444403SRam Pai /* 102621444403SRam Pai * Don't move a mount residing in a shared parent. 102721444403SRam Pai */ 102821444403SRam Pai if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) 102921444403SRam Pai goto out1; 10309676f0c6SRam Pai /* 10319676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 10329676f0c6SRam Pai * mount which is shared. 10339676f0c6SRam Pai */ 10349676f0c6SRam Pai if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt)) 10359676f0c6SRam Pai goto out1; 10361da177e4SLinus Torvalds err = -ELOOP; 10371da177e4SLinus Torvalds for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) 10381da177e4SLinus Torvalds if (p == old_nd.mnt) 103921444403SRam Pai goto out1; 10401da177e4SLinus Torvalds 104121444403SRam Pai if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) 104221444403SRam Pai goto out1; 10431da177e4SLinus Torvalds 104421444403SRam Pai spin_lock(&vfsmount_lock); 10451da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 10461da177e4SLinus Torvalds * automatically */ 104755e700b9SMiklos Szeredi list_del_init(&old_nd.mnt->mnt_expire); 10481da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10491da177e4SLinus Torvalds out1: 10501b1dcc1bSJes Sorensen mutex_unlock(&nd->dentry->d_inode->i_mutex); 10511da177e4SLinus Torvalds out: 1052390c6843SRam Pai up_write(&namespace_sem); 10531da177e4SLinus Torvalds if (!err) 10541da177e4SLinus Torvalds path_release(&parent_nd); 10551da177e4SLinus Torvalds path_release(&old_nd); 10561da177e4SLinus Torvalds return err; 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 10611da177e4SLinus Torvalds * namespace's tree 10621da177e4SLinus Torvalds */ 10631da177e4SLinus Torvalds static int do_new_mount(struct nameidata *nd, char *type, int flags, 10641da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 10651da177e4SLinus Torvalds { 10661da177e4SLinus Torvalds struct vfsmount *mnt; 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 10691da177e4SLinus Torvalds return -EINVAL; 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* we need capabilities... */ 10721da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10731da177e4SLinus Torvalds return -EPERM; 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 10761da177e4SLinus Torvalds if (IS_ERR(mnt)) 10771da177e4SLinus Torvalds return PTR_ERR(mnt); 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10821da177e4SLinus Torvalds /* 10831da177e4SLinus Torvalds * add a mount into a namespace's mount tree 10841da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 10851da177e4SLinus Torvalds */ 10861da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 10871da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 10881da177e4SLinus Torvalds { 10891da177e4SLinus Torvalds int err; 10901da177e4SLinus Torvalds 1091390c6843SRam Pai down_write(&namespace_sem); 10921da177e4SLinus Torvalds /* Something was mounted here while we slept */ 10931da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 10941da177e4SLinus Torvalds ; 10951da177e4SLinus Torvalds err = -EINVAL; 10961da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 10971da177e4SLinus Torvalds goto unlock; 10981da177e4SLinus Torvalds 10991da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 11001da177e4SLinus Torvalds err = -EBUSY; 11011da177e4SLinus Torvalds if (nd->mnt->mnt_sb == newmnt->mnt_sb && 11021da177e4SLinus Torvalds nd->mnt->mnt_root == nd->dentry) 11031da177e4SLinus Torvalds goto unlock; 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds err = -EINVAL; 11061da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 11071da177e4SLinus Torvalds goto unlock; 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 11105b83d2c5SRam Pai if ((err = graft_tree(newmnt, nd))) 11115b83d2c5SRam Pai goto unlock; 11121da177e4SLinus Torvalds 11135b83d2c5SRam Pai if (fslist) { 11141da177e4SLinus Torvalds /* add to the specified expiration list */ 11151da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 111655e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 11171da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11181da177e4SLinus Torvalds } 1119390c6843SRam Pai up_write(&namespace_sem); 11205b83d2c5SRam Pai return 0; 11211da177e4SLinus Torvalds 11221da177e4SLinus Torvalds unlock: 1123390c6843SRam Pai up_write(&namespace_sem); 11241da177e4SLinus Torvalds mntput(newmnt); 11251da177e4SLinus Torvalds return err; 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 11291da177e4SLinus Torvalds 113070fbcdf4SRam Pai static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, 113170fbcdf4SRam Pai struct list_head *umounts) 113224ca2af1SMiklos Szeredi { 113324ca2af1SMiklos Szeredi spin_lock(&vfsmount_lock); 113424ca2af1SMiklos Szeredi 113524ca2af1SMiklos Szeredi /* 1136ed42c879SMiklos Szeredi * Check if mount is still attached, if not, let whoever holds it deal 1137ed42c879SMiklos Szeredi * with the sucker 1138ed42c879SMiklos Szeredi */ 1139ed42c879SMiklos Szeredi if (mnt->mnt_parent == mnt) { 1140ed42c879SMiklos Szeredi spin_unlock(&vfsmount_lock); 1141ed42c879SMiklos Szeredi return; 1142ed42c879SMiklos Szeredi } 1143ed42c879SMiklos Szeredi 1144ed42c879SMiklos Szeredi /* 114524ca2af1SMiklos Szeredi * Check that it is still dead: the count should now be 2 - as 114624ca2af1SMiklos Szeredi * contributed by the vfsmount parent and the mntget above 114724ca2af1SMiklos Szeredi */ 1148a05964f3SRam Pai if (!propagate_mount_busy(mnt, 2)) { 114924ca2af1SMiklos Szeredi /* delete from the namespace */ 11505addc5ddSAl Viro touch_namespace(mnt->mnt_namespace); 115124ca2af1SMiklos Szeredi list_del_init(&mnt->mnt_list); 1152ac081153SMiklos Szeredi mnt->mnt_namespace = NULL; 1153a05964f3SRam Pai umount_tree(mnt, 1, umounts); 115424ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 115524ca2af1SMiklos Szeredi } else { 115624ca2af1SMiklos Szeredi /* 115724ca2af1SMiklos Szeredi * Someone brought it back to life whilst we didn't have any 115824ca2af1SMiklos Szeredi * locks held so return it to the expiration list 115924ca2af1SMiklos Szeredi */ 116055e700b9SMiklos Szeredi list_add_tail(&mnt->mnt_expire, mounts); 116124ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 116224ca2af1SMiklos Szeredi } 116324ca2af1SMiklos Szeredi } 116424ca2af1SMiklos Szeredi 11651da177e4SLinus Torvalds /* 11661da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 11671da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 11681da177e4SLinus Torvalds * here 11691da177e4SLinus Torvalds */ 11701da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 11711da177e4SLinus Torvalds { 11721da177e4SLinus Torvalds struct namespace *namespace; 11731da177e4SLinus Torvalds struct vfsmount *mnt, *next; 11741da177e4SLinus Torvalds LIST_HEAD(graveyard); 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds if (list_empty(mounts)) 11771da177e4SLinus Torvalds return; 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11801da177e4SLinus Torvalds 11811da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 11821da177e4SLinus Torvalds * following criteria: 11831da177e4SLinus Torvalds * - only referenced by its parent vfsmount 11841da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 11851da177e4SLinus Torvalds * cleared by mntput()) 11861da177e4SLinus Torvalds */ 118755e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 11881da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 11891da177e4SLinus Torvalds atomic_read(&mnt->mnt_count) != 1) 11901da177e4SLinus Torvalds continue; 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds mntget(mnt); 119355e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 11941da177e4SLinus Torvalds } 11951da177e4SLinus Torvalds 11961da177e4SLinus Torvalds /* 11971da177e4SLinus Torvalds * go through the vfsmounts we've just consigned to the graveyard to 11981da177e4SLinus Torvalds * - check that they're still dead 11991da177e4SLinus Torvalds * - delete the vfsmount from the appropriate namespace under lock 12001da177e4SLinus Torvalds * - dispose of the corpse 12011da177e4SLinus Torvalds */ 12021da177e4SLinus Torvalds while (!list_empty(&graveyard)) { 120370fbcdf4SRam Pai LIST_HEAD(umounts); 120455e700b9SMiklos Szeredi mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire); 120555e700b9SMiklos Szeredi list_del_init(&mnt->mnt_expire); 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds /* don't do anything if the namespace is dead - all the 12081da177e4SLinus Torvalds * vfsmounts from it are going away anyway */ 12091da177e4SLinus Torvalds namespace = mnt->mnt_namespace; 12101ce88cf4SMiklos Szeredi if (!namespace || !namespace->root) 12111da177e4SLinus Torvalds continue; 12121da177e4SLinus Torvalds get_namespace(namespace); 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1215390c6843SRam Pai down_write(&namespace_sem); 121670fbcdf4SRam Pai expire_mount(mnt, mounts, &umounts); 1217390c6843SRam Pai up_write(&namespace_sem); 121870fbcdf4SRam Pai release_mounts(&umounts); 12191da177e4SLinus Torvalds mntput(mnt); 12201da177e4SLinus Torvalds put_namespace(namespace); 12211da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* 12301da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 12311da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 12321da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 12331da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 12341da177e4SLinus Torvalds */ 1235b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1236b58fed8bSRam Pai unsigned long n) 12371da177e4SLinus Torvalds { 12381da177e4SLinus Torvalds char *t = to; 12391da177e4SLinus Torvalds const char __user *f = from; 12401da177e4SLinus Torvalds char c; 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 12431da177e4SLinus Torvalds return n; 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds while (n) { 12461da177e4SLinus Torvalds if (__get_user(c, f)) { 12471da177e4SLinus Torvalds memset(t, 0, n); 12481da177e4SLinus Torvalds break; 12491da177e4SLinus Torvalds } 12501da177e4SLinus Torvalds *t++ = c; 12511da177e4SLinus Torvalds f++; 12521da177e4SLinus Torvalds n--; 12531da177e4SLinus Torvalds } 12541da177e4SLinus Torvalds return n; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 12581da177e4SLinus Torvalds { 12591da177e4SLinus Torvalds int i; 12601da177e4SLinus Torvalds unsigned long page; 12611da177e4SLinus Torvalds unsigned long size; 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds *where = 0; 12641da177e4SLinus Torvalds if (!data) 12651da177e4SLinus Torvalds return 0; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 12681da177e4SLinus Torvalds return -ENOMEM; 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds /* We only care that *some* data at the address the user 12711da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 12721da177e4SLinus Torvalds * the remainder of the page. 12731da177e4SLinus Torvalds */ 12741da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 12751da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 12761da177e4SLinus Torvalds if (size > PAGE_SIZE) 12771da177e4SLinus Torvalds size = PAGE_SIZE; 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 12801da177e4SLinus Torvalds if (!i) { 12811da177e4SLinus Torvalds free_page(page); 12821da177e4SLinus Torvalds return -EFAULT; 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds if (i != PAGE_SIZE) 12851da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 12861da177e4SLinus Torvalds *where = page; 12871da177e4SLinus Torvalds return 0; 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds /* 12911da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 12921da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 12931da177e4SLinus Torvalds * 12941da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 12951da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 12961da177e4SLinus Torvalds * information (or be NULL). 12971da177e4SLinus Torvalds * 12981da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 12991da177e4SLinus Torvalds * When the flags word was introduced its top half was required 13001da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 13011da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 13021da177e4SLinus Torvalds * and must be discarded. 13031da177e4SLinus Torvalds */ 13041da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 13051da177e4SLinus Torvalds unsigned long flags, void *data_page) 13061da177e4SLinus Torvalds { 13071da177e4SLinus Torvalds struct nameidata nd; 13081da177e4SLinus Torvalds int retval = 0; 13091da177e4SLinus Torvalds int mnt_flags = 0; 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds /* Discard magic */ 13121da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 13131da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds /* Basic sanity checks */ 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 13181da177e4SLinus Torvalds return -EINVAL; 13191da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 13201da177e4SLinus Torvalds return -EINVAL; 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds if (data_page) 13231da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 13261da177e4SLinus Torvalds if (flags & MS_NOSUID) 13271da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 13281da177e4SLinus Torvalds if (flags & MS_NODEV) 13291da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 13301da177e4SLinus Torvalds if (flags & MS_NOEXEC) 13311da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1332fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1333fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1334fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1335fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 1336fc33a7bbSChristoph Hellwig 1337fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 1338fc33a7bbSChristoph Hellwig MS_NOATIME | MS_NODIRATIME); 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds /* ... and get the mountpoint */ 13411da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 13421da177e4SLinus Torvalds if (retval) 13431da177e4SLinus Torvalds return retval; 13441da177e4SLinus Torvalds 13451da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 13461da177e4SLinus Torvalds if (retval) 13471da177e4SLinus Torvalds goto dput_out; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds if (flags & MS_REMOUNT) 13501da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 13511da177e4SLinus Torvalds data_page); 13521da177e4SLinus Torvalds else if (flags & MS_BIND) 13531da177e4SLinus Torvalds retval = do_loopback(&nd, dev_name, flags & MS_REC); 13549676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 135507b20889SRam Pai retval = do_change_type(&nd, flags); 13561da177e4SLinus Torvalds else if (flags & MS_MOVE) 13571da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 13581da177e4SLinus Torvalds else 13591da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 13601da177e4SLinus Torvalds dev_name, data_page); 13611da177e4SLinus Torvalds dput_out: 13621da177e4SLinus Torvalds path_release(&nd); 13631da177e4SLinus Torvalds return retval; 13641da177e4SLinus Torvalds } 13651da177e4SLinus Torvalds 1366741a2951SJANAK DESAI /* 1367741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1368741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1369741a2951SJANAK DESAI */ 1370741a2951SJANAK DESAI struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs) 13711da177e4SLinus Torvalds { 13721da177e4SLinus Torvalds struct namespace *namespace = tsk->namespace; 13731da177e4SLinus Torvalds struct namespace *new_ns; 13741da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 13751da177e4SLinus Torvalds struct vfsmount *p, *q; 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); 13781da177e4SLinus Torvalds if (!new_ns) 1379f13b8358SAdrian Bunk return NULL; 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 13821da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 13835addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 13845addc5ddSAl Viro new_ns->event = 0; 13851da177e4SLinus Torvalds 1386390c6843SRam Pai down_write(&namespace_sem); 13871da177e4SLinus Torvalds /* First pass: copy the tree topology */ 138836341f64SRam Pai new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, 13899676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 13901da177e4SLinus Torvalds if (!new_ns->root) { 1391390c6843SRam Pai up_write(&namespace_sem); 13921da177e4SLinus Torvalds kfree(new_ns); 1393f13b8358SAdrian Bunk return NULL; 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 13961da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 13971da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds /* 14001da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 14011da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 14021da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 14031da177e4SLinus Torvalds */ 14041da177e4SLinus Torvalds p = namespace->root; 14051da177e4SLinus Torvalds q = new_ns->root; 14061da177e4SLinus Torvalds while (p) { 14071da177e4SLinus Torvalds q->mnt_namespace = new_ns; 14081da177e4SLinus Torvalds if (fs) { 14091da177e4SLinus Torvalds if (p == fs->rootmnt) { 14101da177e4SLinus Torvalds rootmnt = p; 14111da177e4SLinus Torvalds fs->rootmnt = mntget(q); 14121da177e4SLinus Torvalds } 14131da177e4SLinus Torvalds if (p == fs->pwdmnt) { 14141da177e4SLinus Torvalds pwdmnt = p; 14151da177e4SLinus Torvalds fs->pwdmnt = mntget(q); 14161da177e4SLinus Torvalds } 14171da177e4SLinus Torvalds if (p == fs->altrootmnt) { 14181da177e4SLinus Torvalds altrootmnt = p; 14191da177e4SLinus Torvalds fs->altrootmnt = mntget(q); 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds } 14221da177e4SLinus Torvalds p = next_mnt(p, namespace->root); 14231da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 14241da177e4SLinus Torvalds } 1425390c6843SRam Pai up_write(&namespace_sem); 14261da177e4SLinus Torvalds 14271da177e4SLinus Torvalds if (rootmnt) 14281da177e4SLinus Torvalds mntput(rootmnt); 14291da177e4SLinus Torvalds if (pwdmnt) 14301da177e4SLinus Torvalds mntput(pwdmnt); 14311da177e4SLinus Torvalds if (altrootmnt) 14321da177e4SLinus Torvalds mntput(altrootmnt); 14331da177e4SLinus Torvalds 1434741a2951SJANAK DESAI return new_ns; 1435741a2951SJANAK DESAI } 1436741a2951SJANAK DESAI 1437741a2951SJANAK DESAI int copy_namespace(int flags, struct task_struct *tsk) 1438741a2951SJANAK DESAI { 1439741a2951SJANAK DESAI struct namespace *namespace = tsk->namespace; 1440741a2951SJANAK DESAI struct namespace *new_ns; 1441741a2951SJANAK DESAI int err = 0; 1442741a2951SJANAK DESAI 1443741a2951SJANAK DESAI if (!namespace) 14441da177e4SLinus Torvalds return 0; 14451da177e4SLinus Torvalds 1446741a2951SJANAK DESAI get_namespace(namespace); 1447741a2951SJANAK DESAI 1448741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1449741a2951SJANAK DESAI return 0; 1450741a2951SJANAK DESAI 1451741a2951SJANAK DESAI if (!capable(CAP_SYS_ADMIN)) { 1452741a2951SJANAK DESAI err = -EPERM; 1453741a2951SJANAK DESAI goto out; 1454741a2951SJANAK DESAI } 1455741a2951SJANAK DESAI 1456741a2951SJANAK DESAI new_ns = dup_namespace(tsk, tsk->fs); 1457741a2951SJANAK DESAI if (!new_ns) { 1458741a2951SJANAK DESAI err = -ENOMEM; 1459741a2951SJANAK DESAI goto out; 1460741a2951SJANAK DESAI } 1461741a2951SJANAK DESAI 1462741a2951SJANAK DESAI tsk->namespace = new_ns; 1463741a2951SJANAK DESAI 14641da177e4SLinus Torvalds out: 14651da177e4SLinus Torvalds put_namespace(namespace); 1466741a2951SJANAK DESAI return err; 14671da177e4SLinus Torvalds } 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 14701da177e4SLinus Torvalds char __user * type, unsigned long flags, 14711da177e4SLinus Torvalds void __user * data) 14721da177e4SLinus Torvalds { 14731da177e4SLinus Torvalds int retval; 14741da177e4SLinus Torvalds unsigned long data_page; 14751da177e4SLinus Torvalds unsigned long type_page; 14761da177e4SLinus Torvalds unsigned long dev_page; 14771da177e4SLinus Torvalds char *dir_page; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 14801da177e4SLinus Torvalds if (retval < 0) 14811da177e4SLinus Torvalds return retval; 14821da177e4SLinus Torvalds 14831da177e4SLinus Torvalds dir_page = getname(dir_name); 14841da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 14851da177e4SLinus Torvalds if (IS_ERR(dir_page)) 14861da177e4SLinus Torvalds goto out1; 14871da177e4SLinus Torvalds 14881da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 14891da177e4SLinus Torvalds if (retval < 0) 14901da177e4SLinus Torvalds goto out2; 14911da177e4SLinus Torvalds 14921da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 14931da177e4SLinus Torvalds if (retval < 0) 14941da177e4SLinus Torvalds goto out3; 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds lock_kernel(); 14971da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 14981da177e4SLinus Torvalds flags, (void *)data_page); 14991da177e4SLinus Torvalds unlock_kernel(); 15001da177e4SLinus Torvalds free_page(data_page); 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds out3: 15031da177e4SLinus Torvalds free_page(dev_page); 15041da177e4SLinus Torvalds out2: 15051da177e4SLinus Torvalds putname(dir_page); 15061da177e4SLinus Torvalds out1: 15071da177e4SLinus Torvalds free_page(type_page); 15081da177e4SLinus Torvalds return retval; 15091da177e4SLinus Torvalds } 15101da177e4SLinus Torvalds 15111da177e4SLinus Torvalds /* 15121da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 15131da177e4SLinus Torvalds * It can block. Requires the big lock held. 15141da177e4SLinus Torvalds */ 15151da177e4SLinus Torvalds void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, 15161da177e4SLinus Torvalds struct dentry *dentry) 15171da177e4SLinus Torvalds { 15181da177e4SLinus Torvalds struct dentry *old_root; 15191da177e4SLinus Torvalds struct vfsmount *old_rootmnt; 15201da177e4SLinus Torvalds write_lock(&fs->lock); 15211da177e4SLinus Torvalds old_root = fs->root; 15221da177e4SLinus Torvalds old_rootmnt = fs->rootmnt; 15231da177e4SLinus Torvalds fs->rootmnt = mntget(mnt); 15241da177e4SLinus Torvalds fs->root = dget(dentry); 15251da177e4SLinus Torvalds write_unlock(&fs->lock); 15261da177e4SLinus Torvalds if (old_root) { 15271da177e4SLinus Torvalds dput(old_root); 15281da177e4SLinus Torvalds mntput(old_rootmnt); 15291da177e4SLinus Torvalds } 15301da177e4SLinus Torvalds } 15311da177e4SLinus Torvalds 15321da177e4SLinus Torvalds /* 15331da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 15341da177e4SLinus Torvalds * It can block. Requires the big lock held. 15351da177e4SLinus Torvalds */ 15361da177e4SLinus Torvalds void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, 15371da177e4SLinus Torvalds struct dentry *dentry) 15381da177e4SLinus Torvalds { 15391da177e4SLinus Torvalds struct dentry *old_pwd; 15401da177e4SLinus Torvalds struct vfsmount *old_pwdmnt; 15411da177e4SLinus Torvalds 15421da177e4SLinus Torvalds write_lock(&fs->lock); 15431da177e4SLinus Torvalds old_pwd = fs->pwd; 15441da177e4SLinus Torvalds old_pwdmnt = fs->pwdmnt; 15451da177e4SLinus Torvalds fs->pwdmnt = mntget(mnt); 15461da177e4SLinus Torvalds fs->pwd = dget(dentry); 15471da177e4SLinus Torvalds write_unlock(&fs->lock); 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds if (old_pwd) { 15501da177e4SLinus Torvalds dput(old_pwd); 15511da177e4SLinus Torvalds mntput(old_pwdmnt); 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds } 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) 15561da177e4SLinus Torvalds { 15571da177e4SLinus Torvalds struct task_struct *g, *p; 15581da177e4SLinus Torvalds struct fs_struct *fs; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds read_lock(&tasklist_lock); 15611da177e4SLinus Torvalds do_each_thread(g, p) { 15621da177e4SLinus Torvalds task_lock(p); 15631da177e4SLinus Torvalds fs = p->fs; 15641da177e4SLinus Torvalds if (fs) { 15651da177e4SLinus Torvalds atomic_inc(&fs->count); 15661da177e4SLinus Torvalds task_unlock(p); 1567b58fed8bSRam Pai if (fs->root == old_nd->dentry 1568b58fed8bSRam Pai && fs->rootmnt == old_nd->mnt) 15691da177e4SLinus Torvalds set_fs_root(fs, new_nd->mnt, new_nd->dentry); 1570b58fed8bSRam Pai if (fs->pwd == old_nd->dentry 1571b58fed8bSRam Pai && fs->pwdmnt == old_nd->mnt) 15721da177e4SLinus Torvalds set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); 15731da177e4SLinus Torvalds put_fs_struct(fs); 15741da177e4SLinus Torvalds } else 15751da177e4SLinus Torvalds task_unlock(p); 15761da177e4SLinus Torvalds } while_each_thread(g, p); 15771da177e4SLinus Torvalds read_unlock(&tasklist_lock); 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds 15801da177e4SLinus Torvalds /* 15811da177e4SLinus Torvalds * pivot_root Semantics: 15821da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 15831da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 15841da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 15851da177e4SLinus Torvalds * 15861da177e4SLinus Torvalds * Restrictions: 15871da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 15881da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 15891da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 15901da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 15911da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 15921da177e4SLinus Torvalds * 15934a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 15944a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 15954a0d11faSNeil Brown * in this situation. 15964a0d11faSNeil Brown * 15971da177e4SLinus Torvalds * Notes: 15981da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 15991da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 16001da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 16011da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 16021da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 16031da177e4SLinus Torvalds * first. 16041da177e4SLinus Torvalds */ 1605b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 1606b58fed8bSRam Pai const char __user * put_old) 16071da177e4SLinus Torvalds { 16081da177e4SLinus Torvalds struct vfsmount *tmp; 16091da177e4SLinus Torvalds struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; 16101da177e4SLinus Torvalds int error; 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16131da177e4SLinus Torvalds return -EPERM; 16141da177e4SLinus Torvalds 16151da177e4SLinus Torvalds lock_kernel(); 16161da177e4SLinus Torvalds 1617b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1618b58fed8bSRam Pai &new_nd); 16191da177e4SLinus Torvalds if (error) 16201da177e4SLinus Torvalds goto out0; 16211da177e4SLinus Torvalds error = -EINVAL; 16221da177e4SLinus Torvalds if (!check_mnt(new_nd.mnt)) 16231da177e4SLinus Torvalds goto out1; 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 16261da177e4SLinus Torvalds if (error) 16271da177e4SLinus Torvalds goto out1; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 16301da177e4SLinus Torvalds if (error) { 16311da177e4SLinus Torvalds path_release(&old_nd); 16321da177e4SLinus Torvalds goto out1; 16331da177e4SLinus Torvalds } 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds read_lock(¤t->fs->lock); 16361da177e4SLinus Torvalds user_nd.mnt = mntget(current->fs->rootmnt); 16371da177e4SLinus Torvalds user_nd.dentry = dget(current->fs->root); 16381da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 1639390c6843SRam Pai down_write(&namespace_sem); 16401b1dcc1bSJes Sorensen mutex_lock(&old_nd.dentry->d_inode->i_mutex); 16411da177e4SLinus Torvalds error = -EINVAL; 164221444403SRam Pai if (IS_MNT_SHARED(old_nd.mnt) || 164321444403SRam Pai IS_MNT_SHARED(new_nd.mnt->mnt_parent) || 164421444403SRam Pai IS_MNT_SHARED(user_nd.mnt->mnt_parent)) 164521444403SRam Pai goto out2; 16461da177e4SLinus Torvalds if (!check_mnt(user_nd.mnt)) 16471da177e4SLinus Torvalds goto out2; 16481da177e4SLinus Torvalds error = -ENOENT; 16491da177e4SLinus Torvalds if (IS_DEADDIR(new_nd.dentry->d_inode)) 16501da177e4SLinus Torvalds goto out2; 16511da177e4SLinus Torvalds if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) 16521da177e4SLinus Torvalds goto out2; 16531da177e4SLinus Torvalds if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) 16541da177e4SLinus Torvalds goto out2; 16551da177e4SLinus Torvalds error = -EBUSY; 16561da177e4SLinus Torvalds if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) 16571da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 16581da177e4SLinus Torvalds error = -EINVAL; 16591da177e4SLinus Torvalds if (user_nd.mnt->mnt_root != user_nd.dentry) 16601da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 16610bb6fcc1SMiklos Szeredi if (user_nd.mnt->mnt_parent == user_nd.mnt) 16620bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 16631da177e4SLinus Torvalds if (new_nd.mnt->mnt_root != new_nd.dentry) 16641da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 16650bb6fcc1SMiklos Szeredi if (new_nd.mnt->mnt_parent == new_nd.mnt) 16660bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 16671da177e4SLinus Torvalds tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ 16681da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 16691da177e4SLinus Torvalds if (tmp != new_nd.mnt) { 16701da177e4SLinus Torvalds for (;;) { 16711da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 16721da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 16731da177e4SLinus Torvalds if (tmp->mnt_parent == new_nd.mnt) 16741da177e4SLinus Torvalds break; 16751da177e4SLinus Torvalds tmp = tmp->mnt_parent; 16761da177e4SLinus Torvalds } 16771da177e4SLinus Torvalds if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) 16781da177e4SLinus Torvalds goto out3; 16791da177e4SLinus Torvalds } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) 16801da177e4SLinus Torvalds goto out3; 16811da177e4SLinus Torvalds detach_mnt(new_nd.mnt, &parent_nd); 16821da177e4SLinus Torvalds detach_mnt(user_nd.mnt, &root_parent); 16831da177e4SLinus Torvalds attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ 16841da177e4SLinus Torvalds attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ 16855addc5ddSAl Viro touch_namespace(current->namespace); 16861da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 16871da177e4SLinus Torvalds chroot_fs_refs(&user_nd, &new_nd); 16881da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 16891da177e4SLinus Torvalds error = 0; 16901da177e4SLinus Torvalds path_release(&root_parent); 16911da177e4SLinus Torvalds path_release(&parent_nd); 16921da177e4SLinus Torvalds out2: 16931b1dcc1bSJes Sorensen mutex_unlock(&old_nd.dentry->d_inode->i_mutex); 1694390c6843SRam Pai up_write(&namespace_sem); 16951da177e4SLinus Torvalds path_release(&user_nd); 16961da177e4SLinus Torvalds path_release(&old_nd); 16971da177e4SLinus Torvalds out1: 16981da177e4SLinus Torvalds path_release(&new_nd); 16991da177e4SLinus Torvalds out0: 17001da177e4SLinus Torvalds unlock_kernel(); 17011da177e4SLinus Torvalds return error; 17021da177e4SLinus Torvalds out3: 17031da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 17041da177e4SLinus Torvalds goto out2; 17051da177e4SLinus Torvalds } 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds static void __init init_mount_tree(void) 17081da177e4SLinus Torvalds { 17091da177e4SLinus Torvalds struct vfsmount *mnt; 17101da177e4SLinus Torvalds struct namespace *namespace; 17111da177e4SLinus Torvalds struct task_struct *g, *p; 17121da177e4SLinus Torvalds 17131da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 17141da177e4SLinus Torvalds if (IS_ERR(mnt)) 17151da177e4SLinus Torvalds panic("Can't create rootfs"); 17161da177e4SLinus Torvalds namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); 17171da177e4SLinus Torvalds if (!namespace) 17181da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 17191da177e4SLinus Torvalds atomic_set(&namespace->count, 1); 17201da177e4SLinus Torvalds INIT_LIST_HEAD(&namespace->list); 17215addc5ddSAl Viro init_waitqueue_head(&namespace->poll); 17225addc5ddSAl Viro namespace->event = 0; 17231da177e4SLinus Torvalds list_add(&mnt->mnt_list, &namespace->list); 17241da177e4SLinus Torvalds namespace->root = mnt; 17251da177e4SLinus Torvalds mnt->mnt_namespace = namespace; 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds init_task.namespace = namespace; 17281da177e4SLinus Torvalds read_lock(&tasklist_lock); 17291da177e4SLinus Torvalds do_each_thread(g, p) { 17301da177e4SLinus Torvalds get_namespace(namespace); 17311da177e4SLinus Torvalds p->namespace = namespace; 17321da177e4SLinus Torvalds } while_each_thread(g, p); 17331da177e4SLinus Torvalds read_unlock(&tasklist_lock); 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); 17361da177e4SLinus Torvalds set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); 17371da177e4SLinus Torvalds } 17381da177e4SLinus Torvalds 17391da177e4SLinus Torvalds void __init mnt_init(unsigned long mempages) 17401da177e4SLinus Torvalds { 17411da177e4SLinus Torvalds struct list_head *d; 17421da177e4SLinus Torvalds unsigned int nr_hash; 17431da177e4SLinus Torvalds int i; 17441da177e4SLinus Torvalds 1745390c6843SRam Pai init_rwsem(&namespace_sem); 1746390c6843SRam Pai 17471da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 17481da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); 17491da177e4SLinus Torvalds 1750b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds if (!mount_hashtable) 17531da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds /* 17561da177e4SLinus Torvalds * Find the power-of-two list-heads that can fit into the allocation.. 17571da177e4SLinus Torvalds * We don't guarantee that "sizeof(struct list_head)" is necessarily 17581da177e4SLinus Torvalds * a power-of-two. 17591da177e4SLinus Torvalds */ 17601da177e4SLinus Torvalds nr_hash = PAGE_SIZE / sizeof(struct list_head); 17611da177e4SLinus Torvalds hash_bits = 0; 17621da177e4SLinus Torvalds do { 17631da177e4SLinus Torvalds hash_bits++; 17641da177e4SLinus Torvalds } while ((nr_hash >> hash_bits) != 0); 17651da177e4SLinus Torvalds hash_bits--; 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* 17681da177e4SLinus Torvalds * Re-calculate the actual number of entries and the mask 17691da177e4SLinus Torvalds * from the number of bits we can fit. 17701da177e4SLinus Torvalds */ 17711da177e4SLinus Torvalds nr_hash = 1UL << hash_bits; 17721da177e4SLinus Torvalds hash_mask = nr_hash - 1; 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds printk("Mount-cache hash table entries: %d\n", nr_hash); 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds /* And initialize the newly allocated array */ 17771da177e4SLinus Torvalds d = mount_hashtable; 17781da177e4SLinus Torvalds i = nr_hash; 17791da177e4SLinus Torvalds do { 17801da177e4SLinus Torvalds INIT_LIST_HEAD(d); 17811da177e4SLinus Torvalds d++; 17821da177e4SLinus Torvalds i--; 17831da177e4SLinus Torvalds } while (i); 17841da177e4SLinus Torvalds sysfs_init(); 1785f87fd4c2SMiklos Szeredi subsystem_register(&fs_subsys); 17861da177e4SLinus Torvalds init_rootfs(); 17871da177e4SLinus Torvalds init_mount_tree(); 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds void __put_namespace(struct namespace *namespace) 17911da177e4SLinus Torvalds { 17921ce88cf4SMiklos Szeredi struct vfsmount *root = namespace->root; 179370fbcdf4SRam Pai LIST_HEAD(umount_list); 17941ce88cf4SMiklos Szeredi namespace->root = NULL; 17951ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 1796390c6843SRam Pai down_write(&namespace_sem); 17971da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1798a05964f3SRam Pai umount_tree(root, 0, &umount_list); 17991da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1800390c6843SRam Pai up_write(&namespace_sem); 180170fbcdf4SRam Pai release_mounts(&umount_list); 18021da177e4SLinus Torvalds kfree(namespace); 18031da177e4SLinus Torvalds } 1804