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> 191da177e4SLinus Torvalds #include <linux/module.h> 201da177e4SLinus Torvalds #include <linux/seq_file.h> 211da177e4SLinus Torvalds #include <linux/namespace.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/security.h> 241da177e4SLinus Torvalds #include <linux/mount.h> 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds #include <asm/unistd.h> 2707b20889SRam Pai #include "pnode.h" 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds extern int __init init_rootfs(void); 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds #ifdef CONFIG_SYSFS 321da177e4SLinus Torvalds extern int __init sysfs_init(void); 331da177e4SLinus Torvalds #else 341da177e4SLinus Torvalds static inline int sysfs_init(void) 351da177e4SLinus Torvalds { 361da177e4SLinus Torvalds return 0; 371da177e4SLinus Torvalds } 381da177e4SLinus Torvalds #endif 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 411da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 421da177e4SLinus Torvalds 435addc5ddSAl Viro static int event; 445addc5ddSAl Viro 451da177e4SLinus Torvalds static struct list_head *mount_hashtable; 466c231b7bSRavikiran G Thirumalai static int hash_mask __read_mostly, hash_bits __read_mostly; 471da177e4SLinus Torvalds static kmem_cache_t *mnt_cache; 48390c6843SRam Pai static struct rw_semaphore namespace_sem; 491da177e4SLinus Torvalds 501da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 511da177e4SLinus Torvalds { 521da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 531da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 541da177e4SLinus Torvalds tmp = tmp + (tmp >> hash_bits); 551da177e4SLinus Torvalds return tmp & hash_mask; 561da177e4SLinus Torvalds } 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 591da177e4SLinus Torvalds { 601da177e4SLinus Torvalds struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); 611da177e4SLinus Torvalds if (mnt) { 621da177e4SLinus Torvalds memset(mnt, 0, sizeof(struct vfsmount)); 631da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 641da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 651da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 661da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 671da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 6855e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 6903e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 70a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 71a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 721da177e4SLinus Torvalds if (name) { 731da177e4SLinus Torvalds int size = strlen(name) + 1; 741da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 751da177e4SLinus Torvalds if (newname) { 761da177e4SLinus Torvalds memcpy(newname, name, size); 771da177e4SLinus Torvalds mnt->mnt_devname = newname; 781da177e4SLinus Torvalds } 791da177e4SLinus Torvalds } 801da177e4SLinus Torvalds } 811da177e4SLinus Torvalds return mnt; 821da177e4SLinus Torvalds } 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 851da177e4SLinus Torvalds { 861da177e4SLinus Torvalds kfree(mnt->mnt_devname); 871da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 881da177e4SLinus Torvalds } 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /* 91a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 92a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 931da177e4SLinus Torvalds */ 94a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 95a05964f3SRam Pai int dir) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 981da177e4SLinus Torvalds struct list_head *tmp = head; 991da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 1001da177e4SLinus Torvalds 1011da177e4SLinus Torvalds for (;;) { 102a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 1031da177e4SLinus Torvalds p = NULL; 1041da177e4SLinus Torvalds if (tmp == head) 1051da177e4SLinus Torvalds break; 1061da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 1071da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 108a05964f3SRam Pai found = p; 1091da177e4SLinus Torvalds break; 1101da177e4SLinus Torvalds } 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds return found; 1131da177e4SLinus Torvalds } 1141da177e4SLinus Torvalds 115a05964f3SRam Pai /* 116a05964f3SRam Pai * lookup_mnt increments the ref count before returning 117a05964f3SRam Pai * the vfsmount struct. 118a05964f3SRam Pai */ 119a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 120a05964f3SRam Pai { 121a05964f3SRam Pai struct vfsmount *child_mnt; 122a05964f3SRam Pai spin_lock(&vfsmount_lock); 123a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 124a05964f3SRam Pai mntget(child_mnt); 125a05964f3SRam Pai spin_unlock(&vfsmount_lock); 126a05964f3SRam Pai return child_mnt; 127a05964f3SRam Pai } 128a05964f3SRam Pai 1291da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 1301da177e4SLinus Torvalds { 1311da177e4SLinus Torvalds return mnt->mnt_namespace == current->namespace; 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds 1345addc5ddSAl Viro static void touch_namespace(struct namespace *ns) 1355addc5ddSAl Viro { 1365addc5ddSAl Viro if (ns) { 1375addc5ddSAl Viro ns->event = ++event; 1385addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1395addc5ddSAl Viro } 1405addc5ddSAl Viro } 1415addc5ddSAl Viro 1425addc5ddSAl Viro static void __touch_namespace(struct namespace *ns) 1435addc5ddSAl Viro { 1445addc5ddSAl Viro if (ns && ns->event != event) { 1455addc5ddSAl Viro ns->event = event; 1465addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1475addc5ddSAl Viro } 1485addc5ddSAl Viro } 1495addc5ddSAl Viro 1501da177e4SLinus Torvalds static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) 1511da177e4SLinus Torvalds { 1521da177e4SLinus Torvalds old_nd->dentry = mnt->mnt_mountpoint; 1531da177e4SLinus Torvalds old_nd->mnt = mnt->mnt_parent; 1541da177e4SLinus Torvalds mnt->mnt_parent = mnt; 1551da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 1561da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 1571da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 1581da177e4SLinus Torvalds old_nd->dentry->d_mounted--; 1591da177e4SLinus Torvalds } 1601da177e4SLinus Torvalds 161b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 162b90fa9aeSRam Pai struct vfsmount *child_mnt) 163b90fa9aeSRam Pai { 164b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 165b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 166b90fa9aeSRam Pai dentry->d_mounted++; 167b90fa9aeSRam Pai } 168b90fa9aeSRam Pai 1691da177e4SLinus Torvalds static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) 1701da177e4SLinus Torvalds { 171b90fa9aeSRam Pai mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); 172b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 173b90fa9aeSRam Pai hash(nd->mnt, nd->dentry)); 1741da177e4SLinus Torvalds list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); 175b90fa9aeSRam Pai } 176b90fa9aeSRam Pai 177b90fa9aeSRam Pai /* 178b90fa9aeSRam Pai * the caller must hold vfsmount_lock 179b90fa9aeSRam Pai */ 180b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 181b90fa9aeSRam Pai { 182b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 183b90fa9aeSRam Pai struct vfsmount *m; 184b90fa9aeSRam Pai LIST_HEAD(head); 185b90fa9aeSRam Pai struct namespace *n = parent->mnt_namespace; 186b90fa9aeSRam Pai 187b90fa9aeSRam Pai BUG_ON(parent == mnt); 188b90fa9aeSRam Pai 189b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 190b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 191b90fa9aeSRam Pai m->mnt_namespace = n; 192b90fa9aeSRam Pai list_splice(&head, n->list.prev); 193b90fa9aeSRam Pai 194b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 195b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 196b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 197b90fa9aeSRam Pai touch_namespace(n); 1981da177e4SLinus Torvalds } 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 2011da177e4SLinus Torvalds { 2021da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 2031da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 2041da177e4SLinus Torvalds while (1) { 2051da177e4SLinus Torvalds if (p == root) 2061da177e4SLinus Torvalds return NULL; 2071da177e4SLinus Torvalds next = p->mnt_child.next; 2081da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 2091da177e4SLinus Torvalds break; 2101da177e4SLinus Torvalds p = p->mnt_parent; 2111da177e4SLinus Torvalds } 2121da177e4SLinus Torvalds } 2131da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 2141da177e4SLinus Torvalds } 2151da177e4SLinus Torvalds 2169676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 2179676f0c6SRam Pai { 2189676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 2199676f0c6SRam Pai while (prev != &p->mnt_mounts) { 2209676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 2219676f0c6SRam Pai prev = p->mnt_mounts.prev; 2229676f0c6SRam Pai } 2239676f0c6SRam Pai return p; 2249676f0c6SRam Pai } 2259676f0c6SRam Pai 22636341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 22736341f64SRam Pai int flag) 2281da177e4SLinus Torvalds { 2291da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 2301da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 2311da177e4SLinus Torvalds 2321da177e4SLinus Torvalds if (mnt) { 2331da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 2341da177e4SLinus Torvalds atomic_inc(&sb->s_active); 2351da177e4SLinus Torvalds mnt->mnt_sb = sb; 2361da177e4SLinus Torvalds mnt->mnt_root = dget(root); 2371da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 2381da177e4SLinus Torvalds mnt->mnt_parent = mnt; 239b90fa9aeSRam Pai 2405afe0022SRam Pai if (flag & CL_SLAVE) { 2415afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 2425afe0022SRam Pai mnt->mnt_master = old; 2435afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 2445afe0022SRam Pai } else { 245b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 246b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 2475afe0022SRam Pai if (IS_MNT_SLAVE(old)) 2485afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 2495afe0022SRam Pai mnt->mnt_master = old->mnt_master; 2505afe0022SRam Pai } 251b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 252b90fa9aeSRam Pai set_mnt_shared(mnt); 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 2551da177e4SLinus Torvalds * as the original if that was on one */ 25636341f64SRam Pai if (flag & CL_EXPIRE) { 2571da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 25855e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 25955e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 2601da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2611da177e4SLinus Torvalds } 26236341f64SRam Pai } 2631da177e4SLinus Torvalds return mnt; 2641da177e4SLinus Torvalds } 2651da177e4SLinus Torvalds 2667b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 2671da177e4SLinus Torvalds { 2681da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 2691da177e4SLinus Torvalds dput(mnt->mnt_root); 2701da177e4SLinus Torvalds free_vfsmnt(mnt); 2711da177e4SLinus Torvalds deactivate_super(sb); 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 2747b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 2757b7b1aceSAl Viro { 2767b7b1aceSAl Viro repeat: 2777b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 2787b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 2797b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2807b7b1aceSAl Viro __mntput(mnt); 2817b7b1aceSAl Viro return; 2827b7b1aceSAl Viro } 2837b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 2847b7b1aceSAl Viro mnt->mnt_pinned = 0; 2857b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2867b7b1aceSAl Viro acct_auto_close_mnt(mnt); 2877b7b1aceSAl Viro security_sb_umount_close(mnt); 2887b7b1aceSAl Viro goto repeat; 2897b7b1aceSAl Viro } 2907b7b1aceSAl Viro } 2917b7b1aceSAl Viro 2927b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 2937b7b1aceSAl Viro 2947b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 2957b7b1aceSAl Viro { 2967b7b1aceSAl Viro spin_lock(&vfsmount_lock); 2977b7b1aceSAl Viro mnt->mnt_pinned++; 2987b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2997b7b1aceSAl Viro } 3007b7b1aceSAl Viro 3017b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 3027b7b1aceSAl Viro 3037b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 3047b7b1aceSAl Viro { 3057b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3067b7b1aceSAl Viro if (mnt->mnt_pinned) { 3077b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 3087b7b1aceSAl Viro mnt->mnt_pinned--; 3097b7b1aceSAl Viro } 3107b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3117b7b1aceSAl Viro } 3127b7b1aceSAl Viro 3137b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 3141da177e4SLinus Torvalds 3151da177e4SLinus Torvalds /* iterator */ 3161da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 3171da177e4SLinus Torvalds { 3181da177e4SLinus Torvalds struct namespace *n = m->private; 3191da177e4SLinus Torvalds struct list_head *p; 3201da177e4SLinus Torvalds loff_t l = *pos; 3211da177e4SLinus Torvalds 322390c6843SRam Pai down_read(&namespace_sem); 3231da177e4SLinus Torvalds list_for_each(p, &n->list) 3241da177e4SLinus Torvalds if (!l--) 3251da177e4SLinus Torvalds return list_entry(p, struct vfsmount, mnt_list); 3261da177e4SLinus Torvalds return NULL; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 3301da177e4SLinus Torvalds { 3311da177e4SLinus Torvalds struct namespace *n = m->private; 3321da177e4SLinus Torvalds struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; 3331da177e4SLinus Torvalds (*pos)++; 3341da177e4SLinus Torvalds return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds 3371da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 3381da177e4SLinus Torvalds { 339390c6843SRam Pai up_read(&namespace_sem); 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds static inline void mangle(struct seq_file *m, const char *s) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds seq_escape(m, s, " \t\n\\"); 3451da177e4SLinus Torvalds } 3461da177e4SLinus Torvalds 3471da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 3481da177e4SLinus Torvalds { 3491da177e4SLinus Torvalds struct vfsmount *mnt = v; 3501da177e4SLinus Torvalds int err = 0; 3511da177e4SLinus Torvalds static struct proc_fs_info { 3521da177e4SLinus Torvalds int flag; 3531da177e4SLinus Torvalds char *str; 3541da177e4SLinus Torvalds } fs_info[] = { 3551da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 3561da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 3571da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 3581da177e4SLinus Torvalds { MS_NOATIME, ",noatime" }, 3591da177e4SLinus Torvalds { MS_NODIRATIME, ",nodiratime" }, 3601da177e4SLinus Torvalds { 0, NULL } 3611da177e4SLinus Torvalds }; 3621da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 3631da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 3641da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 3651da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 3661da177e4SLinus Torvalds { 0, NULL } 3671da177e4SLinus Torvalds }; 3681da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 3711da177e4SLinus Torvalds seq_putc(m, ' '); 3721da177e4SLinus Torvalds seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 3731da177e4SLinus Torvalds seq_putc(m, ' '); 3741da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 3751da177e4SLinus Torvalds seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 3761da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 3771da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 3781da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 3791da177e4SLinus Torvalds } 3801da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 3811da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 3821da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 3851da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 3861da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 3871da177e4SLinus Torvalds return err; 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds 3901da177e4SLinus Torvalds struct seq_operations mounts_op = { 3911da177e4SLinus Torvalds .start = m_start, 3921da177e4SLinus Torvalds .next = m_next, 3931da177e4SLinus Torvalds .stop = m_stop, 3941da177e4SLinus Torvalds .show = show_vfsmnt 3951da177e4SLinus Torvalds }; 3961da177e4SLinus Torvalds 3971da177e4SLinus Torvalds /** 3981da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 3991da177e4SLinus Torvalds * @mnt: root of mount tree 4001da177e4SLinus Torvalds * 4011da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 4021da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 4031da177e4SLinus Torvalds * busy. 4041da177e4SLinus Torvalds */ 4051da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 4061da177e4SLinus Torvalds { 40736341f64SRam Pai int actual_refs = 0; 40836341f64SRam Pai int minimum_refs = 0; 40936341f64SRam Pai struct vfsmount *p; 4101da177e4SLinus Torvalds 4111da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 41236341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 4131da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 4141da177e4SLinus Torvalds minimum_refs += 2; 4151da177e4SLinus Torvalds } 4161da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 4171da177e4SLinus Torvalds 4181da177e4SLinus Torvalds if (actual_refs > minimum_refs) 4191da177e4SLinus Torvalds return -EBUSY; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds return 0; 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds /** 4271da177e4SLinus Torvalds * may_umount - check if a mount point is busy 4281da177e4SLinus Torvalds * @mnt: root of mount 4291da177e4SLinus Torvalds * 4301da177e4SLinus Torvalds * This is called to check if a mount point has any 4311da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 4321da177e4SLinus Torvalds * mount has sub mounts this will return busy 4331da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 4341da177e4SLinus Torvalds * 4351da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 4361da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 4371da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 4381da177e4SLinus Torvalds */ 4391da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 4401da177e4SLinus Torvalds { 441a05964f3SRam Pai int ret = 0; 442a05964f3SRam Pai spin_lock(&vfsmount_lock); 443a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 444a05964f3SRam Pai ret = -EBUSY; 445a05964f3SRam Pai spin_unlock(&vfsmount_lock); 446a05964f3SRam Pai return ret; 4471da177e4SLinus Torvalds } 4481da177e4SLinus Torvalds 4491da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 4501da177e4SLinus Torvalds 451b90fa9aeSRam Pai void release_mounts(struct list_head *head) 4521da177e4SLinus Torvalds { 45370fbcdf4SRam Pai struct vfsmount *mnt; 45470fbcdf4SRam Pai while(!list_empty(head)) { 45570fbcdf4SRam Pai mnt = list_entry(head->next, struct vfsmount, mnt_hash); 45670fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 45770fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 45870fbcdf4SRam Pai struct dentry *dentry; 45970fbcdf4SRam Pai struct vfsmount *m; 46070fbcdf4SRam Pai spin_lock(&vfsmount_lock); 46170fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 46270fbcdf4SRam Pai m = mnt->mnt_parent; 46370fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 46470fbcdf4SRam Pai mnt->mnt_parent = mnt; 4651da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 46670fbcdf4SRam Pai dput(dentry); 46770fbcdf4SRam Pai mntput(m); 4681da177e4SLinus Torvalds } 4691da177e4SLinus Torvalds mntput(mnt); 47070fbcdf4SRam Pai } 47170fbcdf4SRam Pai } 47270fbcdf4SRam Pai 473a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 47470fbcdf4SRam Pai { 47570fbcdf4SRam Pai struct vfsmount *p; 47670fbcdf4SRam Pai 47770fbcdf4SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 47870fbcdf4SRam Pai list_del(&p->mnt_hash); 47970fbcdf4SRam Pai list_add(&p->mnt_hash, kill); 48070fbcdf4SRam Pai } 48170fbcdf4SRam Pai 482a05964f3SRam Pai if (propagate) 483a05964f3SRam Pai propagate_umount(kill); 484a05964f3SRam Pai 48570fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 48670fbcdf4SRam Pai list_del_init(&p->mnt_expire); 48770fbcdf4SRam Pai list_del_init(&p->mnt_list); 48870fbcdf4SRam Pai __touch_namespace(p->mnt_namespace); 48970fbcdf4SRam Pai p->mnt_namespace = NULL; 49070fbcdf4SRam Pai list_del_init(&p->mnt_child); 49170fbcdf4SRam Pai if (p->mnt_parent != p) 49270fbcdf4SRam Pai mnt->mnt_mountpoint->d_mounted--; 493a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 4981da177e4SLinus Torvalds { 4991da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 5001da177e4SLinus Torvalds int retval; 50170fbcdf4SRam Pai LIST_HEAD(umount_list); 5021da177e4SLinus Torvalds 5031da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 5041da177e4SLinus Torvalds if (retval) 5051da177e4SLinus Torvalds return retval; 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds /* 5081da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 5091da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 5101da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 5111da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 5121da177e4SLinus Torvalds */ 5131da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 5141da177e4SLinus Torvalds if (mnt == current->fs->rootmnt || 5151da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 5161da177e4SLinus Torvalds return -EINVAL; 5171da177e4SLinus Torvalds 5181da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 5191da177e4SLinus Torvalds return -EBUSY; 5201da177e4SLinus Torvalds 5211da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 5221da177e4SLinus Torvalds return -EAGAIN; 5231da177e4SLinus Torvalds } 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * If we may have to abort operations to get out of this 5271da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 5281da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 5291da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 5301da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 5311da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 5321da177e4SLinus Torvalds * about for the moment. 5331da177e4SLinus Torvalds */ 5341da177e4SLinus Torvalds 5351da177e4SLinus Torvalds lock_kernel(); 5361da177e4SLinus Torvalds if ((flags & MNT_FORCE) && sb->s_op->umount_begin) 5371da177e4SLinus Torvalds sb->s_op->umount_begin(sb); 5381da177e4SLinus Torvalds unlock_kernel(); 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds /* 5411da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 5421da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 5431da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 5441da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 5451da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 5461da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 5471da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 5481da177e4SLinus Torvalds */ 5491da177e4SLinus Torvalds if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { 5501da177e4SLinus Torvalds /* 5511da177e4SLinus Torvalds * Special case for "unmounting" root ... 5521da177e4SLinus Torvalds * we just try to remount it readonly. 5531da177e4SLinus Torvalds */ 5541da177e4SLinus Torvalds down_write(&sb->s_umount); 5551da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 5561da177e4SLinus Torvalds lock_kernel(); 5571da177e4SLinus Torvalds DQUOT_OFF(sb); 5581da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 5591da177e4SLinus Torvalds unlock_kernel(); 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds up_write(&sb->s_umount); 5621da177e4SLinus Torvalds return retval; 5631da177e4SLinus Torvalds } 5641da177e4SLinus Torvalds 565390c6843SRam Pai down_write(&namespace_sem); 5661da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 5675addc5ddSAl Viro event++; 5681da177e4SLinus Torvalds 5691da177e4SLinus Torvalds retval = -EBUSY; 570a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 5711da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 572a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 5731da177e4SLinus Torvalds retval = 0; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 5761da177e4SLinus Torvalds if (retval) 5771da177e4SLinus Torvalds security_sb_umount_busy(mnt); 578390c6843SRam Pai up_write(&namespace_sem); 57970fbcdf4SRam Pai release_mounts(&umount_list); 5801da177e4SLinus Torvalds return retval; 5811da177e4SLinus Torvalds } 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds /* 5841da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 5851da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 5861da177e4SLinus Torvalds * 5871da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 5881da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 5891da177e4SLinus Torvalds */ 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 5921da177e4SLinus Torvalds { 5931da177e4SLinus Torvalds struct nameidata nd; 5941da177e4SLinus Torvalds int retval; 5951da177e4SLinus Torvalds 5961da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 5971da177e4SLinus Torvalds if (retval) 5981da177e4SLinus Torvalds goto out; 5991da177e4SLinus Torvalds retval = -EINVAL; 6001da177e4SLinus Torvalds if (nd.dentry != nd.mnt->mnt_root) 6011da177e4SLinus Torvalds goto dput_and_out; 6021da177e4SLinus Torvalds if (!check_mnt(nd.mnt)) 6031da177e4SLinus Torvalds goto dput_and_out; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds retval = -EPERM; 6061da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 6071da177e4SLinus Torvalds goto dput_and_out; 6081da177e4SLinus Torvalds 6091da177e4SLinus Torvalds retval = do_umount(nd.mnt, flags); 6101da177e4SLinus Torvalds dput_and_out: 6111da177e4SLinus Torvalds path_release_on_umount(&nd); 6121da177e4SLinus Torvalds out: 6131da177e4SLinus Torvalds return retval; 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds 6161da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds /* 6191da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 6201da177e4SLinus Torvalds */ 6211da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 6221da177e4SLinus Torvalds { 6231da177e4SLinus Torvalds return sys_umount(name, 0); 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds 6261da177e4SLinus Torvalds #endif 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 6291da177e4SLinus Torvalds { 6301da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 6311da177e4SLinus Torvalds return 0; 6321da177e4SLinus Torvalds return -EPERM; 6331da177e4SLinus Torvalds #ifdef notyet 6341da177e4SLinus Torvalds if (S_ISLNK(nd->dentry->d_inode->i_mode)) 6351da177e4SLinus Torvalds return -EPERM; 6361da177e4SLinus Torvalds if (nd->dentry->d_inode->i_mode & S_ISVTX) { 6371da177e4SLinus Torvalds if (current->uid != nd->dentry->d_inode->i_uid) 6381da177e4SLinus Torvalds return -EPERM; 6391da177e4SLinus Torvalds } 640e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 6411da177e4SLinus Torvalds return -EPERM; 6421da177e4SLinus Torvalds return 0; 6431da177e4SLinus Torvalds #endif 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 646b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 6471da177e4SLinus Torvalds { 6481da177e4SLinus Torvalds while (1) { 6491da177e4SLinus Torvalds if (d == dentry) 6501da177e4SLinus Torvalds return 1; 6511da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 6521da177e4SLinus Torvalds return 0; 6531da177e4SLinus Torvalds d = d->d_parent; 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 657b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 65836341f64SRam Pai int flag) 6591da177e4SLinus Torvalds { 6601da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 6611da177e4SLinus Torvalds struct nameidata nd; 6621da177e4SLinus Torvalds 6639676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 6649676f0c6SRam Pai return NULL; 6659676f0c6SRam Pai 66636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 6671da177e4SLinus Torvalds if (!q) 6681da177e4SLinus Torvalds goto Enomem; 6691da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds p = mnt; 672fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 6731da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 6741da177e4SLinus Torvalds continue; 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 6779676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 6789676f0c6SRam Pai s = skip_mnt_tree(s); 6799676f0c6SRam Pai continue; 6809676f0c6SRam Pai } 6811da177e4SLinus Torvalds while (p != s->mnt_parent) { 6821da177e4SLinus Torvalds p = p->mnt_parent; 6831da177e4SLinus Torvalds q = q->mnt_parent; 6841da177e4SLinus Torvalds } 6851da177e4SLinus Torvalds p = s; 6861da177e4SLinus Torvalds nd.mnt = q; 6871da177e4SLinus Torvalds nd.dentry = p->mnt_mountpoint; 68836341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 6891da177e4SLinus Torvalds if (!q) 6901da177e4SLinus Torvalds goto Enomem; 6911da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6921da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 6931da177e4SLinus Torvalds attach_mnt(q, &nd); 6941da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds return res; 6981da177e4SLinus Torvalds Enomem: 6991da177e4SLinus Torvalds if (res) { 70070fbcdf4SRam Pai LIST_HEAD(umount_list); 7011da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 702a05964f3SRam Pai umount_tree(res, 0, &umount_list); 7031da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 70470fbcdf4SRam Pai release_mounts(&umount_list); 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds return NULL; 7071da177e4SLinus Torvalds } 7081da177e4SLinus Torvalds 709b90fa9aeSRam Pai /* 710b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 711b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 71221444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 71321444403SRam Pai * store the parent mount and mountpoint dentry. 71421444403SRam Pai * (done when source_mnt is moved) 715b90fa9aeSRam Pai * 716b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 717b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 7189676f0c6SRam Pai * --------------------------------------------------------------------------- 719b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 7209676f0c6SRam Pai * |************************************************************************** 7219676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7229676f0c6SRam Pai * | dest | | | | | 7239676f0c6SRam Pai * | | | | | | | 7249676f0c6SRam Pai * | v | | | | | 7259676f0c6SRam Pai * |************************************************************************** 7269676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 7275afe0022SRam Pai * | | | | | | 7289676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 7299676f0c6SRam Pai * *************************************************************************** 730b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 731b90fa9aeSRam Pai * destination mount. 732b90fa9aeSRam Pai * 733b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 734b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 735b90fa9aeSRam Pai * the peer group of the source mount. 736b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 737b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 738b90fa9aeSRam Pai * mount. 7395afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 7405afe0022SRam Pai * of the destination mount and the cloned mount is made slave 7415afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 7425afe0022SRam Pai * is marked as 'shared and slave'. 7435afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 7445afe0022SRam Pai * source mount. 7455afe0022SRam Pai * 7469676f0c6SRam Pai * --------------------------------------------------------------------------- 74721444403SRam Pai * | MOVE MOUNT OPERATION | 7489676f0c6SRam Pai * |************************************************************************** 7499676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7509676f0c6SRam Pai * | dest | | | | | 7519676f0c6SRam Pai * | | | | | | | 7529676f0c6SRam Pai * | v | | | | | 7539676f0c6SRam Pai * |************************************************************************** 7549676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 7555afe0022SRam Pai * | | | | | | 7569676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 7579676f0c6SRam Pai * *************************************************************************** 7585afe0022SRam Pai * 7595afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 7605afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 76121444403SRam Pai * (+*) the mount is moved to the destination. 7625afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 7635afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 7645afe0022SRam Pai * the mount is marked as 'shared and slave'. 7655afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 766b90fa9aeSRam Pai * 767b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 768b90fa9aeSRam Pai * applied to each mount in the tree. 769b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 770b90fa9aeSRam Pai * in allocations. 771b90fa9aeSRam Pai */ 772b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 77321444403SRam Pai struct nameidata *nd, struct nameidata *parent_nd) 774b90fa9aeSRam Pai { 775b90fa9aeSRam Pai LIST_HEAD(tree_list); 776b90fa9aeSRam Pai struct vfsmount *dest_mnt = nd->mnt; 777b90fa9aeSRam Pai struct dentry *dest_dentry = nd->dentry; 778b90fa9aeSRam Pai struct vfsmount *child, *p; 779b90fa9aeSRam Pai 780b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 781b90fa9aeSRam Pai return -EINVAL; 782b90fa9aeSRam Pai 783b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 784b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 785b90fa9aeSRam Pai set_mnt_shared(p); 786b90fa9aeSRam Pai } 787b90fa9aeSRam Pai 788b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 78921444403SRam Pai if (parent_nd) { 79021444403SRam Pai detach_mnt(source_mnt, parent_nd); 79121444403SRam Pai attach_mnt(source_mnt, nd); 79221444403SRam Pai touch_namespace(current->namespace); 79321444403SRam Pai } else { 794b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 795b90fa9aeSRam Pai commit_tree(source_mnt); 79621444403SRam Pai } 797b90fa9aeSRam Pai 798b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 799b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 800b90fa9aeSRam Pai commit_tree(child); 801b90fa9aeSRam Pai } 802b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 803b90fa9aeSRam Pai return 0; 804b90fa9aeSRam Pai } 805b90fa9aeSRam Pai 8061da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 8071da177e4SLinus Torvalds { 8081da177e4SLinus Torvalds int err; 8091da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 8101da177e4SLinus Torvalds return -EINVAL; 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 8131da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 8141da177e4SLinus Torvalds return -ENOTDIR; 8151da177e4SLinus Torvalds 8161da177e4SLinus Torvalds err = -ENOENT; 8171da177e4SLinus Torvalds down(&nd->dentry->d_inode->i_sem); 8181da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 8191da177e4SLinus Torvalds goto out_unlock; 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 8221da177e4SLinus Torvalds if (err) 8231da177e4SLinus Torvalds goto out_unlock; 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds err = -ENOENT; 826b90fa9aeSRam Pai if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) 82721444403SRam Pai err = attach_recursive_mnt(mnt, nd, NULL); 8281da177e4SLinus Torvalds out_unlock: 8291da177e4SLinus Torvalds up(&nd->dentry->d_inode->i_sem); 8301da177e4SLinus Torvalds if (!err) 8311da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 8321da177e4SLinus Torvalds return err; 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds 8351da177e4SLinus Torvalds /* 83607b20889SRam Pai * recursively change the type of the mountpoint. 83707b20889SRam Pai */ 83807b20889SRam Pai static int do_change_type(struct nameidata *nd, int flag) 83907b20889SRam Pai { 84007b20889SRam Pai struct vfsmount *m, *mnt = nd->mnt; 84107b20889SRam Pai int recurse = flag & MS_REC; 84207b20889SRam Pai int type = flag & ~MS_REC; 84307b20889SRam Pai 84407b20889SRam Pai if (nd->dentry != nd->mnt->mnt_root) 84507b20889SRam Pai return -EINVAL; 84607b20889SRam Pai 84707b20889SRam Pai down_write(&namespace_sem); 84807b20889SRam Pai spin_lock(&vfsmount_lock); 84907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 85007b20889SRam Pai change_mnt_propagation(m, type); 85107b20889SRam Pai spin_unlock(&vfsmount_lock); 85207b20889SRam Pai up_write(&namespace_sem); 85307b20889SRam Pai return 0; 85407b20889SRam Pai } 85507b20889SRam Pai 85607b20889SRam Pai /* 8571da177e4SLinus Torvalds * do loopback mount. 8581da177e4SLinus Torvalds */ 8591da177e4SLinus Torvalds static int do_loopback(struct nameidata *nd, char *old_name, int recurse) 8601da177e4SLinus Torvalds { 8611da177e4SLinus Torvalds struct nameidata old_nd; 8621da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 8631da177e4SLinus Torvalds int err = mount_is_safe(nd); 8641da177e4SLinus Torvalds if (err) 8651da177e4SLinus Torvalds return err; 8661da177e4SLinus Torvalds if (!old_name || !*old_name) 8671da177e4SLinus Torvalds return -EINVAL; 8681da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 8691da177e4SLinus Torvalds if (err) 8701da177e4SLinus Torvalds return err; 8711da177e4SLinus Torvalds 872390c6843SRam Pai down_write(&namespace_sem); 8731da177e4SLinus Torvalds err = -EINVAL; 8749676f0c6SRam Pai if (IS_MNT_UNBINDABLE(old_nd.mnt)) 8759676f0c6SRam Pai goto out; 8769676f0c6SRam Pai 877ccd48bc7SAl Viro if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 878ccd48bc7SAl Viro goto out; 879ccd48bc7SAl Viro 8801da177e4SLinus Torvalds err = -ENOMEM; 8811da177e4SLinus Torvalds if (recurse) 88236341f64SRam Pai mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); 8831da177e4SLinus Torvalds else 88436341f64SRam Pai mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); 8851da177e4SLinus Torvalds 886ccd48bc7SAl Viro if (!mnt) 887ccd48bc7SAl Viro goto out; 888ccd48bc7SAl Viro 8891da177e4SLinus Torvalds err = graft_tree(mnt, nd); 8901da177e4SLinus Torvalds if (err) { 89170fbcdf4SRam Pai LIST_HEAD(umount_list); 8921da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 893a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 8941da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 89570fbcdf4SRam Pai release_mounts(&umount_list); 8965b83d2c5SRam Pai } 8971da177e4SLinus Torvalds 898ccd48bc7SAl Viro out: 899390c6843SRam Pai up_write(&namespace_sem); 9001da177e4SLinus Torvalds path_release(&old_nd); 9011da177e4SLinus Torvalds return err; 9021da177e4SLinus Torvalds } 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds /* 9051da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 9061da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 9071da177e4SLinus Torvalds * on it - tough luck. 9081da177e4SLinus Torvalds */ 9091da177e4SLinus Torvalds static int do_remount(struct nameidata *nd, int flags, int mnt_flags, 9101da177e4SLinus Torvalds void *data) 9111da177e4SLinus Torvalds { 9121da177e4SLinus Torvalds int err; 9131da177e4SLinus Torvalds struct super_block *sb = nd->mnt->mnt_sb; 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9161da177e4SLinus Torvalds return -EPERM; 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 9191da177e4SLinus Torvalds return -EINVAL; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds if (nd->dentry != nd->mnt->mnt_root) 9221da177e4SLinus Torvalds return -EINVAL; 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds down_write(&sb->s_umount); 9251da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 9261da177e4SLinus Torvalds if (!err) 9271da177e4SLinus Torvalds nd->mnt->mnt_flags = mnt_flags; 9281da177e4SLinus Torvalds up_write(&sb->s_umount); 9291da177e4SLinus Torvalds if (!err) 9301da177e4SLinus Torvalds security_sb_post_remount(nd->mnt, flags, data); 9311da177e4SLinus Torvalds return err; 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds 9349676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 9359676f0c6SRam Pai { 9369676f0c6SRam Pai struct vfsmount *p; 9379676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9389676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 9399676f0c6SRam Pai return 1; 9409676f0c6SRam Pai } 9419676f0c6SRam Pai return 0; 9429676f0c6SRam Pai } 9439676f0c6SRam Pai 9441da177e4SLinus Torvalds static int do_move_mount(struct nameidata *nd, char *old_name) 9451da177e4SLinus Torvalds { 9461da177e4SLinus Torvalds struct nameidata old_nd, parent_nd; 9471da177e4SLinus Torvalds struct vfsmount *p; 9481da177e4SLinus Torvalds int err = 0; 9491da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9501da177e4SLinus Torvalds return -EPERM; 9511da177e4SLinus Torvalds if (!old_name || !*old_name) 9521da177e4SLinus Torvalds return -EINVAL; 9531da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 9541da177e4SLinus Torvalds if (err) 9551da177e4SLinus Torvalds return err; 9561da177e4SLinus Torvalds 957390c6843SRam Pai down_write(&namespace_sem); 9581da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 9591da177e4SLinus Torvalds ; 9601da177e4SLinus Torvalds err = -EINVAL; 9611da177e4SLinus Torvalds if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 9621da177e4SLinus Torvalds goto out; 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds err = -ENOENT; 9651da177e4SLinus Torvalds down(&nd->dentry->d_inode->i_sem); 9661da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 9671da177e4SLinus Torvalds goto out1; 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) 97021444403SRam Pai goto out1; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds err = -EINVAL; 9731da177e4SLinus Torvalds if (old_nd.dentry != old_nd.mnt->mnt_root) 97421444403SRam Pai goto out1; 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds if (old_nd.mnt == old_nd.mnt->mnt_parent) 97721444403SRam Pai goto out1; 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 9801da177e4SLinus Torvalds S_ISDIR(old_nd.dentry->d_inode->i_mode)) 98121444403SRam Pai goto out1; 98221444403SRam Pai /* 98321444403SRam Pai * Don't move a mount residing in a shared parent. 98421444403SRam Pai */ 98521444403SRam Pai if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) 98621444403SRam Pai goto out1; 9879676f0c6SRam Pai /* 9889676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 9899676f0c6SRam Pai * mount which is shared. 9909676f0c6SRam Pai */ 9919676f0c6SRam Pai if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt)) 9929676f0c6SRam Pai goto out1; 9931da177e4SLinus Torvalds err = -ELOOP; 9941da177e4SLinus Torvalds for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) 9951da177e4SLinus Torvalds if (p == old_nd.mnt) 99621444403SRam Pai goto out1; 9971da177e4SLinus Torvalds 99821444403SRam Pai if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) 99921444403SRam Pai goto out1; 10001da177e4SLinus Torvalds 100121444403SRam Pai spin_lock(&vfsmount_lock); 10021da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 10031da177e4SLinus Torvalds * automatically */ 100455e700b9SMiklos Szeredi list_del_init(&old_nd.mnt->mnt_expire); 10051da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10061da177e4SLinus Torvalds out1: 10071da177e4SLinus Torvalds up(&nd->dentry->d_inode->i_sem); 10081da177e4SLinus Torvalds out: 1009390c6843SRam Pai up_write(&namespace_sem); 10101da177e4SLinus Torvalds if (!err) 10111da177e4SLinus Torvalds path_release(&parent_nd); 10121da177e4SLinus Torvalds path_release(&old_nd); 10131da177e4SLinus Torvalds return err; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds /* 10171da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 10181da177e4SLinus Torvalds * namespace's tree 10191da177e4SLinus Torvalds */ 10201da177e4SLinus Torvalds static int do_new_mount(struct nameidata *nd, char *type, int flags, 10211da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 10221da177e4SLinus Torvalds { 10231da177e4SLinus Torvalds struct vfsmount *mnt; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 10261da177e4SLinus Torvalds return -EINVAL; 10271da177e4SLinus Torvalds 10281da177e4SLinus Torvalds /* we need capabilities... */ 10291da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10301da177e4SLinus Torvalds return -EPERM; 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 10331da177e4SLinus Torvalds if (IS_ERR(mnt)) 10341da177e4SLinus Torvalds return PTR_ERR(mnt); 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds /* 10401da177e4SLinus Torvalds * add a mount into a namespace's mount tree 10411da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 10421da177e4SLinus Torvalds */ 10431da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 10441da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 10451da177e4SLinus Torvalds { 10461da177e4SLinus Torvalds int err; 10471da177e4SLinus Torvalds 1048390c6843SRam Pai down_write(&namespace_sem); 10491da177e4SLinus Torvalds /* Something was mounted here while we slept */ 10501da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 10511da177e4SLinus Torvalds ; 10521da177e4SLinus Torvalds err = -EINVAL; 10531da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 10541da177e4SLinus Torvalds goto unlock; 10551da177e4SLinus Torvalds 10561da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 10571da177e4SLinus Torvalds err = -EBUSY; 10581da177e4SLinus Torvalds if (nd->mnt->mnt_sb == newmnt->mnt_sb && 10591da177e4SLinus Torvalds nd->mnt->mnt_root == nd->dentry) 10601da177e4SLinus Torvalds goto unlock; 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds err = -EINVAL; 10631da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 10641da177e4SLinus Torvalds goto unlock; 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 10675b83d2c5SRam Pai if ((err = graft_tree(newmnt, nd))) 10685b83d2c5SRam Pai goto unlock; 10691da177e4SLinus Torvalds 10705b83d2c5SRam Pai if (fslist) { 10711da177e4SLinus Torvalds /* add to the specified expiration list */ 10721da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 107355e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 10741da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10751da177e4SLinus Torvalds } 1076390c6843SRam Pai up_write(&namespace_sem); 10775b83d2c5SRam Pai return 0; 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds unlock: 1080390c6843SRam Pai up_write(&namespace_sem); 10811da177e4SLinus Torvalds mntput(newmnt); 10821da177e4SLinus Torvalds return err; 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 10851da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 10861da177e4SLinus Torvalds 108770fbcdf4SRam Pai static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, 108870fbcdf4SRam Pai struct list_head *umounts) 108924ca2af1SMiklos Szeredi { 109024ca2af1SMiklos Szeredi spin_lock(&vfsmount_lock); 109124ca2af1SMiklos Szeredi 109224ca2af1SMiklos Szeredi /* 1093ed42c879SMiklos Szeredi * Check if mount is still attached, if not, let whoever holds it deal 1094ed42c879SMiklos Szeredi * with the sucker 1095ed42c879SMiklos Szeredi */ 1096ed42c879SMiklos Szeredi if (mnt->mnt_parent == mnt) { 1097ed42c879SMiklos Szeredi spin_unlock(&vfsmount_lock); 1098ed42c879SMiklos Szeredi return; 1099ed42c879SMiklos Szeredi } 1100ed42c879SMiklos Szeredi 1101ed42c879SMiklos Szeredi /* 110224ca2af1SMiklos Szeredi * Check that it is still dead: the count should now be 2 - as 110324ca2af1SMiklos Szeredi * contributed by the vfsmount parent and the mntget above 110424ca2af1SMiklos Szeredi */ 1105a05964f3SRam Pai if (!propagate_mount_busy(mnt, 2)) { 110624ca2af1SMiklos Szeredi /* delete from the namespace */ 11075addc5ddSAl Viro touch_namespace(mnt->mnt_namespace); 110824ca2af1SMiklos Szeredi list_del_init(&mnt->mnt_list); 1109ac081153SMiklos Szeredi mnt->mnt_namespace = NULL; 1110a05964f3SRam Pai umount_tree(mnt, 1, umounts); 111124ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 111224ca2af1SMiklos Szeredi } else { 111324ca2af1SMiklos Szeredi /* 111424ca2af1SMiklos Szeredi * Someone brought it back to life whilst we didn't have any 111524ca2af1SMiklos Szeredi * locks held so return it to the expiration list 111624ca2af1SMiklos Szeredi */ 111755e700b9SMiklos Szeredi list_add_tail(&mnt->mnt_expire, mounts); 111824ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 111924ca2af1SMiklos Szeredi } 112024ca2af1SMiklos Szeredi } 112124ca2af1SMiklos Szeredi 11221da177e4SLinus Torvalds /* 11231da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 11241da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 11251da177e4SLinus Torvalds * here 11261da177e4SLinus Torvalds */ 11271da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 11281da177e4SLinus Torvalds { 11291da177e4SLinus Torvalds struct namespace *namespace; 11301da177e4SLinus Torvalds struct vfsmount *mnt, *next; 11311da177e4SLinus Torvalds LIST_HEAD(graveyard); 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds if (list_empty(mounts)) 11341da177e4SLinus Torvalds return; 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 11391da177e4SLinus Torvalds * following criteria: 11401da177e4SLinus Torvalds * - only referenced by its parent vfsmount 11411da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 11421da177e4SLinus Torvalds * cleared by mntput()) 11431da177e4SLinus Torvalds */ 114455e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 11451da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 11461da177e4SLinus Torvalds atomic_read(&mnt->mnt_count) != 1) 11471da177e4SLinus Torvalds continue; 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds mntget(mnt); 115055e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds /* 11541da177e4SLinus Torvalds * go through the vfsmounts we've just consigned to the graveyard to 11551da177e4SLinus Torvalds * - check that they're still dead 11561da177e4SLinus Torvalds * - delete the vfsmount from the appropriate namespace under lock 11571da177e4SLinus Torvalds * - dispose of the corpse 11581da177e4SLinus Torvalds */ 11591da177e4SLinus Torvalds while (!list_empty(&graveyard)) { 116070fbcdf4SRam Pai LIST_HEAD(umounts); 116155e700b9SMiklos Szeredi mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire); 116255e700b9SMiklos Szeredi list_del_init(&mnt->mnt_expire); 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* don't do anything if the namespace is dead - all the 11651da177e4SLinus Torvalds * vfsmounts from it are going away anyway */ 11661da177e4SLinus Torvalds namespace = mnt->mnt_namespace; 11671ce88cf4SMiklos Szeredi if (!namespace || !namespace->root) 11681da177e4SLinus Torvalds continue; 11691da177e4SLinus Torvalds get_namespace(namespace); 11701da177e4SLinus Torvalds 11711da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1172390c6843SRam Pai down_write(&namespace_sem); 117370fbcdf4SRam Pai expire_mount(mnt, mounts, &umounts); 1174390c6843SRam Pai up_write(&namespace_sem); 117570fbcdf4SRam Pai release_mounts(&umounts); 11761da177e4SLinus Torvalds mntput(mnt); 11771da177e4SLinus Torvalds put_namespace(namespace); 11781da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11791da177e4SLinus Torvalds } 11801da177e4SLinus Torvalds 11811da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 11851da177e4SLinus Torvalds 11861da177e4SLinus Torvalds /* 11871da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 11881da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 11891da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 11901da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 11911da177e4SLinus Torvalds */ 1192b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1193b58fed8bSRam Pai unsigned long n) 11941da177e4SLinus Torvalds { 11951da177e4SLinus Torvalds char *t = to; 11961da177e4SLinus Torvalds const char __user *f = from; 11971da177e4SLinus Torvalds char c; 11981da177e4SLinus Torvalds 11991da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 12001da177e4SLinus Torvalds return n; 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds while (n) { 12031da177e4SLinus Torvalds if (__get_user(c, f)) { 12041da177e4SLinus Torvalds memset(t, 0, n); 12051da177e4SLinus Torvalds break; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds *t++ = c; 12081da177e4SLinus Torvalds f++; 12091da177e4SLinus Torvalds n--; 12101da177e4SLinus Torvalds } 12111da177e4SLinus Torvalds return n; 12121da177e4SLinus Torvalds } 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 12151da177e4SLinus Torvalds { 12161da177e4SLinus Torvalds int i; 12171da177e4SLinus Torvalds unsigned long page; 12181da177e4SLinus Torvalds unsigned long size; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds *where = 0; 12211da177e4SLinus Torvalds if (!data) 12221da177e4SLinus Torvalds return 0; 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 12251da177e4SLinus Torvalds return -ENOMEM; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* We only care that *some* data at the address the user 12281da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 12291da177e4SLinus Torvalds * the remainder of the page. 12301da177e4SLinus Torvalds */ 12311da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 12321da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 12331da177e4SLinus Torvalds if (size > PAGE_SIZE) 12341da177e4SLinus Torvalds size = PAGE_SIZE; 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 12371da177e4SLinus Torvalds if (!i) { 12381da177e4SLinus Torvalds free_page(page); 12391da177e4SLinus Torvalds return -EFAULT; 12401da177e4SLinus Torvalds } 12411da177e4SLinus Torvalds if (i != PAGE_SIZE) 12421da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 12431da177e4SLinus Torvalds *where = page; 12441da177e4SLinus Torvalds return 0; 12451da177e4SLinus Torvalds } 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds /* 12481da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 12491da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 12501da177e4SLinus Torvalds * 12511da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 12521da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 12531da177e4SLinus Torvalds * information (or be NULL). 12541da177e4SLinus Torvalds * 12551da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 12561da177e4SLinus Torvalds * When the flags word was introduced its top half was required 12571da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 12581da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 12591da177e4SLinus Torvalds * and must be discarded. 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 12621da177e4SLinus Torvalds unsigned long flags, void *data_page) 12631da177e4SLinus Torvalds { 12641da177e4SLinus Torvalds struct nameidata nd; 12651da177e4SLinus Torvalds int retval = 0; 12661da177e4SLinus Torvalds int mnt_flags = 0; 12671da177e4SLinus Torvalds 12681da177e4SLinus Torvalds /* Discard magic */ 12691da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 12701da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds /* Basic sanity checks */ 12731da177e4SLinus Torvalds 12741da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 12751da177e4SLinus Torvalds return -EINVAL; 12761da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 12771da177e4SLinus Torvalds return -EINVAL; 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds if (data_page) 12801da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 12831da177e4SLinus Torvalds if (flags & MS_NOSUID) 12841da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 12851da177e4SLinus Torvalds if (flags & MS_NODEV) 12861da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 12871da177e4SLinus Torvalds if (flags & MS_NOEXEC) 12881da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 12891da177e4SLinus Torvalds flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE); 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* ... and get the mountpoint */ 12921da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 12931da177e4SLinus Torvalds if (retval) 12941da177e4SLinus Torvalds return retval; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 12971da177e4SLinus Torvalds if (retval) 12981da177e4SLinus Torvalds goto dput_out; 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds if (flags & MS_REMOUNT) 13011da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 13021da177e4SLinus Torvalds data_page); 13031da177e4SLinus Torvalds else if (flags & MS_BIND) 13041da177e4SLinus Torvalds retval = do_loopback(&nd, dev_name, flags & MS_REC); 13059676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 130607b20889SRam Pai retval = do_change_type(&nd, flags); 13071da177e4SLinus Torvalds else if (flags & MS_MOVE) 13081da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 13091da177e4SLinus Torvalds else 13101da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 13111da177e4SLinus Torvalds dev_name, data_page); 13121da177e4SLinus Torvalds dput_out: 13131da177e4SLinus Torvalds path_release(&nd); 13141da177e4SLinus Torvalds return retval; 13151da177e4SLinus Torvalds } 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds int copy_namespace(int flags, struct task_struct *tsk) 13181da177e4SLinus Torvalds { 13191da177e4SLinus Torvalds struct namespace *namespace = tsk->namespace; 13201da177e4SLinus Torvalds struct namespace *new_ns; 13211da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 13221da177e4SLinus Torvalds struct fs_struct *fs = tsk->fs; 13231da177e4SLinus Torvalds struct vfsmount *p, *q; 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds if (!namespace) 13261da177e4SLinus Torvalds return 0; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds get_namespace(namespace); 13291da177e4SLinus Torvalds 13301da177e4SLinus Torvalds if (!(flags & CLONE_NEWNS)) 13311da177e4SLinus Torvalds return 0; 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) { 13341da177e4SLinus Torvalds put_namespace(namespace); 13351da177e4SLinus Torvalds return -EPERM; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); 13391da177e4SLinus Torvalds if (!new_ns) 13401da177e4SLinus Torvalds goto out; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 13431da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 13445addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 13455addc5ddSAl Viro new_ns->event = 0; 13461da177e4SLinus Torvalds 1347390c6843SRam Pai down_write(&namespace_sem); 13481da177e4SLinus Torvalds /* First pass: copy the tree topology */ 134936341f64SRam Pai new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, 13509676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 13511da177e4SLinus Torvalds if (!new_ns->root) { 1352390c6843SRam Pai up_write(&namespace_sem); 13531da177e4SLinus Torvalds kfree(new_ns); 13541da177e4SLinus Torvalds goto out; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 13571da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 13581da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds /* 13611da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 13621da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 13631da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 13641da177e4SLinus Torvalds */ 13651da177e4SLinus Torvalds p = namespace->root; 13661da177e4SLinus Torvalds q = new_ns->root; 13671da177e4SLinus Torvalds while (p) { 13681da177e4SLinus Torvalds q->mnt_namespace = new_ns; 13691da177e4SLinus Torvalds if (fs) { 13701da177e4SLinus Torvalds if (p == fs->rootmnt) { 13711da177e4SLinus Torvalds rootmnt = p; 13721da177e4SLinus Torvalds fs->rootmnt = mntget(q); 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds if (p == fs->pwdmnt) { 13751da177e4SLinus Torvalds pwdmnt = p; 13761da177e4SLinus Torvalds fs->pwdmnt = mntget(q); 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds if (p == fs->altrootmnt) { 13791da177e4SLinus Torvalds altrootmnt = p; 13801da177e4SLinus Torvalds fs->altrootmnt = mntget(q); 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds } 13831da177e4SLinus Torvalds p = next_mnt(p, namespace->root); 13841da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 13851da177e4SLinus Torvalds } 1386390c6843SRam Pai up_write(&namespace_sem); 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds tsk->namespace = new_ns; 13891da177e4SLinus Torvalds 13901da177e4SLinus Torvalds if (rootmnt) 13911da177e4SLinus Torvalds mntput(rootmnt); 13921da177e4SLinus Torvalds if (pwdmnt) 13931da177e4SLinus Torvalds mntput(pwdmnt); 13941da177e4SLinus Torvalds if (altrootmnt) 13951da177e4SLinus Torvalds mntput(altrootmnt); 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds put_namespace(namespace); 13981da177e4SLinus Torvalds return 0; 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds out: 14011da177e4SLinus Torvalds put_namespace(namespace); 14021da177e4SLinus Torvalds return -ENOMEM; 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 14061da177e4SLinus Torvalds char __user * type, unsigned long flags, 14071da177e4SLinus Torvalds void __user * data) 14081da177e4SLinus Torvalds { 14091da177e4SLinus Torvalds int retval; 14101da177e4SLinus Torvalds unsigned long data_page; 14111da177e4SLinus Torvalds unsigned long type_page; 14121da177e4SLinus Torvalds unsigned long dev_page; 14131da177e4SLinus Torvalds char *dir_page; 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 14161da177e4SLinus Torvalds if (retval < 0) 14171da177e4SLinus Torvalds return retval; 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds dir_page = getname(dir_name); 14201da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 14211da177e4SLinus Torvalds if (IS_ERR(dir_page)) 14221da177e4SLinus Torvalds goto out1; 14231da177e4SLinus Torvalds 14241da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 14251da177e4SLinus Torvalds if (retval < 0) 14261da177e4SLinus Torvalds goto out2; 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 14291da177e4SLinus Torvalds if (retval < 0) 14301da177e4SLinus Torvalds goto out3; 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds lock_kernel(); 14331da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 14341da177e4SLinus Torvalds flags, (void *)data_page); 14351da177e4SLinus Torvalds unlock_kernel(); 14361da177e4SLinus Torvalds free_page(data_page); 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds out3: 14391da177e4SLinus Torvalds free_page(dev_page); 14401da177e4SLinus Torvalds out2: 14411da177e4SLinus Torvalds putname(dir_page); 14421da177e4SLinus Torvalds out1: 14431da177e4SLinus Torvalds free_page(type_page); 14441da177e4SLinus Torvalds return retval; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds /* 14481da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 14491da177e4SLinus Torvalds * It can block. Requires the big lock held. 14501da177e4SLinus Torvalds */ 14511da177e4SLinus Torvalds void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, 14521da177e4SLinus Torvalds struct dentry *dentry) 14531da177e4SLinus Torvalds { 14541da177e4SLinus Torvalds struct dentry *old_root; 14551da177e4SLinus Torvalds struct vfsmount *old_rootmnt; 14561da177e4SLinus Torvalds write_lock(&fs->lock); 14571da177e4SLinus Torvalds old_root = fs->root; 14581da177e4SLinus Torvalds old_rootmnt = fs->rootmnt; 14591da177e4SLinus Torvalds fs->rootmnt = mntget(mnt); 14601da177e4SLinus Torvalds fs->root = dget(dentry); 14611da177e4SLinus Torvalds write_unlock(&fs->lock); 14621da177e4SLinus Torvalds if (old_root) { 14631da177e4SLinus Torvalds dput(old_root); 14641da177e4SLinus Torvalds mntput(old_rootmnt); 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds } 14671da177e4SLinus Torvalds 14681da177e4SLinus Torvalds /* 14691da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 14701da177e4SLinus Torvalds * It can block. Requires the big lock held. 14711da177e4SLinus Torvalds */ 14721da177e4SLinus Torvalds void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, 14731da177e4SLinus Torvalds struct dentry *dentry) 14741da177e4SLinus Torvalds { 14751da177e4SLinus Torvalds struct dentry *old_pwd; 14761da177e4SLinus Torvalds struct vfsmount *old_pwdmnt; 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds write_lock(&fs->lock); 14791da177e4SLinus Torvalds old_pwd = fs->pwd; 14801da177e4SLinus Torvalds old_pwdmnt = fs->pwdmnt; 14811da177e4SLinus Torvalds fs->pwdmnt = mntget(mnt); 14821da177e4SLinus Torvalds fs->pwd = dget(dentry); 14831da177e4SLinus Torvalds write_unlock(&fs->lock); 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds if (old_pwd) { 14861da177e4SLinus Torvalds dput(old_pwd); 14871da177e4SLinus Torvalds mntput(old_pwdmnt); 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds } 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) 14921da177e4SLinus Torvalds { 14931da177e4SLinus Torvalds struct task_struct *g, *p; 14941da177e4SLinus Torvalds struct fs_struct *fs; 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds read_lock(&tasklist_lock); 14971da177e4SLinus Torvalds do_each_thread(g, p) { 14981da177e4SLinus Torvalds task_lock(p); 14991da177e4SLinus Torvalds fs = p->fs; 15001da177e4SLinus Torvalds if (fs) { 15011da177e4SLinus Torvalds atomic_inc(&fs->count); 15021da177e4SLinus Torvalds task_unlock(p); 1503b58fed8bSRam Pai if (fs->root == old_nd->dentry 1504b58fed8bSRam Pai && fs->rootmnt == old_nd->mnt) 15051da177e4SLinus Torvalds set_fs_root(fs, new_nd->mnt, new_nd->dentry); 1506b58fed8bSRam Pai if (fs->pwd == old_nd->dentry 1507b58fed8bSRam Pai && fs->pwdmnt == old_nd->mnt) 15081da177e4SLinus Torvalds set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); 15091da177e4SLinus Torvalds put_fs_struct(fs); 15101da177e4SLinus Torvalds } else 15111da177e4SLinus Torvalds task_unlock(p); 15121da177e4SLinus Torvalds } while_each_thread(g, p); 15131da177e4SLinus Torvalds read_unlock(&tasklist_lock); 15141da177e4SLinus Torvalds } 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds /* 15171da177e4SLinus Torvalds * pivot_root Semantics: 15181da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 15191da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 15201da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 15211da177e4SLinus Torvalds * 15221da177e4SLinus Torvalds * Restrictions: 15231da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 15241da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 15251da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 15261da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 15271da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 15281da177e4SLinus Torvalds * 15291da177e4SLinus Torvalds * Notes: 15301da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 15311da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 15321da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 15331da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 15341da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 15351da177e4SLinus Torvalds * first. 15361da177e4SLinus Torvalds */ 1537b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 1538b58fed8bSRam Pai const char __user * put_old) 15391da177e4SLinus Torvalds { 15401da177e4SLinus Torvalds struct vfsmount *tmp; 15411da177e4SLinus Torvalds struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; 15421da177e4SLinus Torvalds int error; 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15451da177e4SLinus Torvalds return -EPERM; 15461da177e4SLinus Torvalds 15471da177e4SLinus Torvalds lock_kernel(); 15481da177e4SLinus Torvalds 1549b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1550b58fed8bSRam Pai &new_nd); 15511da177e4SLinus Torvalds if (error) 15521da177e4SLinus Torvalds goto out0; 15531da177e4SLinus Torvalds error = -EINVAL; 15541da177e4SLinus Torvalds if (!check_mnt(new_nd.mnt)) 15551da177e4SLinus Torvalds goto out1; 15561da177e4SLinus Torvalds 15571da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 15581da177e4SLinus Torvalds if (error) 15591da177e4SLinus Torvalds goto out1; 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 15621da177e4SLinus Torvalds if (error) { 15631da177e4SLinus Torvalds path_release(&old_nd); 15641da177e4SLinus Torvalds goto out1; 15651da177e4SLinus Torvalds } 15661da177e4SLinus Torvalds 15671da177e4SLinus Torvalds read_lock(¤t->fs->lock); 15681da177e4SLinus Torvalds user_nd.mnt = mntget(current->fs->rootmnt); 15691da177e4SLinus Torvalds user_nd.dentry = dget(current->fs->root); 15701da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 1571390c6843SRam Pai down_write(&namespace_sem); 15721da177e4SLinus Torvalds down(&old_nd.dentry->d_inode->i_sem); 15731da177e4SLinus Torvalds error = -EINVAL; 157421444403SRam Pai if (IS_MNT_SHARED(old_nd.mnt) || 157521444403SRam Pai IS_MNT_SHARED(new_nd.mnt->mnt_parent) || 157621444403SRam Pai IS_MNT_SHARED(user_nd.mnt->mnt_parent)) 157721444403SRam Pai goto out2; 15781da177e4SLinus Torvalds if (!check_mnt(user_nd.mnt)) 15791da177e4SLinus Torvalds goto out2; 15801da177e4SLinus Torvalds error = -ENOENT; 15811da177e4SLinus Torvalds if (IS_DEADDIR(new_nd.dentry->d_inode)) 15821da177e4SLinus Torvalds goto out2; 15831da177e4SLinus Torvalds if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) 15841da177e4SLinus Torvalds goto out2; 15851da177e4SLinus Torvalds if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) 15861da177e4SLinus Torvalds goto out2; 15871da177e4SLinus Torvalds error = -EBUSY; 15881da177e4SLinus Torvalds if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) 15891da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 15901da177e4SLinus Torvalds error = -EINVAL; 15911da177e4SLinus Torvalds if (user_nd.mnt->mnt_root != user_nd.dentry) 15921da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 15930bb6fcc1SMiklos Szeredi if (user_nd.mnt->mnt_parent == user_nd.mnt) 15940bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 15951da177e4SLinus Torvalds if (new_nd.mnt->mnt_root != new_nd.dentry) 15961da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 15970bb6fcc1SMiklos Szeredi if (new_nd.mnt->mnt_parent == new_nd.mnt) 15980bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 15991da177e4SLinus Torvalds tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ 16001da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 16011da177e4SLinus Torvalds if (tmp != new_nd.mnt) { 16021da177e4SLinus Torvalds for (;;) { 16031da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 16041da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 16051da177e4SLinus Torvalds if (tmp->mnt_parent == new_nd.mnt) 16061da177e4SLinus Torvalds break; 16071da177e4SLinus Torvalds tmp = tmp->mnt_parent; 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) 16101da177e4SLinus Torvalds goto out3; 16111da177e4SLinus Torvalds } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) 16121da177e4SLinus Torvalds goto out3; 16131da177e4SLinus Torvalds detach_mnt(new_nd.mnt, &parent_nd); 16141da177e4SLinus Torvalds detach_mnt(user_nd.mnt, &root_parent); 16151da177e4SLinus Torvalds attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ 16161da177e4SLinus Torvalds attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ 16175addc5ddSAl Viro touch_namespace(current->namespace); 16181da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 16191da177e4SLinus Torvalds chroot_fs_refs(&user_nd, &new_nd); 16201da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 16211da177e4SLinus Torvalds error = 0; 16221da177e4SLinus Torvalds path_release(&root_parent); 16231da177e4SLinus Torvalds path_release(&parent_nd); 16241da177e4SLinus Torvalds out2: 16251da177e4SLinus Torvalds up(&old_nd.dentry->d_inode->i_sem); 1626390c6843SRam Pai up_write(&namespace_sem); 16271da177e4SLinus Torvalds path_release(&user_nd); 16281da177e4SLinus Torvalds path_release(&old_nd); 16291da177e4SLinus Torvalds out1: 16301da177e4SLinus Torvalds path_release(&new_nd); 16311da177e4SLinus Torvalds out0: 16321da177e4SLinus Torvalds unlock_kernel(); 16331da177e4SLinus Torvalds return error; 16341da177e4SLinus Torvalds out3: 16351da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 16361da177e4SLinus Torvalds goto out2; 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds static void __init init_mount_tree(void) 16401da177e4SLinus Torvalds { 16411da177e4SLinus Torvalds struct vfsmount *mnt; 16421da177e4SLinus Torvalds struct namespace *namespace; 16431da177e4SLinus Torvalds struct task_struct *g, *p; 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 16461da177e4SLinus Torvalds if (IS_ERR(mnt)) 16471da177e4SLinus Torvalds panic("Can't create rootfs"); 16481da177e4SLinus Torvalds namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); 16491da177e4SLinus Torvalds if (!namespace) 16501da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 16511da177e4SLinus Torvalds atomic_set(&namespace->count, 1); 16521da177e4SLinus Torvalds INIT_LIST_HEAD(&namespace->list); 16535addc5ddSAl Viro init_waitqueue_head(&namespace->poll); 16545addc5ddSAl Viro namespace->event = 0; 16551da177e4SLinus Torvalds list_add(&mnt->mnt_list, &namespace->list); 16561da177e4SLinus Torvalds namespace->root = mnt; 16571da177e4SLinus Torvalds mnt->mnt_namespace = namespace; 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds init_task.namespace = namespace; 16601da177e4SLinus Torvalds read_lock(&tasklist_lock); 16611da177e4SLinus Torvalds do_each_thread(g, p) { 16621da177e4SLinus Torvalds get_namespace(namespace); 16631da177e4SLinus Torvalds p->namespace = namespace; 16641da177e4SLinus Torvalds } while_each_thread(g, p); 16651da177e4SLinus Torvalds read_unlock(&tasklist_lock); 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); 16681da177e4SLinus Torvalds set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds 16711da177e4SLinus Torvalds void __init mnt_init(unsigned long mempages) 16721da177e4SLinus Torvalds { 16731da177e4SLinus Torvalds struct list_head *d; 16741da177e4SLinus Torvalds unsigned int nr_hash; 16751da177e4SLinus Torvalds int i; 16761da177e4SLinus Torvalds 1677390c6843SRam Pai init_rwsem(&namespace_sem); 1678390c6843SRam Pai 16791da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 16801da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); 16811da177e4SLinus Torvalds 1682b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds if (!mount_hashtable) 16851da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 16861da177e4SLinus Torvalds 16871da177e4SLinus Torvalds /* 16881da177e4SLinus Torvalds * Find the power-of-two list-heads that can fit into the allocation.. 16891da177e4SLinus Torvalds * We don't guarantee that "sizeof(struct list_head)" is necessarily 16901da177e4SLinus Torvalds * a power-of-two. 16911da177e4SLinus Torvalds */ 16921da177e4SLinus Torvalds nr_hash = PAGE_SIZE / sizeof(struct list_head); 16931da177e4SLinus Torvalds hash_bits = 0; 16941da177e4SLinus Torvalds do { 16951da177e4SLinus Torvalds hash_bits++; 16961da177e4SLinus Torvalds } while ((nr_hash >> hash_bits) != 0); 16971da177e4SLinus Torvalds hash_bits--; 16981da177e4SLinus Torvalds 16991da177e4SLinus Torvalds /* 17001da177e4SLinus Torvalds * Re-calculate the actual number of entries and the mask 17011da177e4SLinus Torvalds * from the number of bits we can fit. 17021da177e4SLinus Torvalds */ 17031da177e4SLinus Torvalds nr_hash = 1UL << hash_bits; 17041da177e4SLinus Torvalds hash_mask = nr_hash - 1; 17051da177e4SLinus Torvalds 17061da177e4SLinus Torvalds printk("Mount-cache hash table entries: %d\n", nr_hash); 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds /* And initialize the newly allocated array */ 17091da177e4SLinus Torvalds d = mount_hashtable; 17101da177e4SLinus Torvalds i = nr_hash; 17111da177e4SLinus Torvalds do { 17121da177e4SLinus Torvalds INIT_LIST_HEAD(d); 17131da177e4SLinus Torvalds d++; 17141da177e4SLinus Torvalds i--; 17151da177e4SLinus Torvalds } while (i); 17161da177e4SLinus Torvalds sysfs_init(); 17171da177e4SLinus Torvalds init_rootfs(); 17181da177e4SLinus Torvalds init_mount_tree(); 17191da177e4SLinus Torvalds } 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds void __put_namespace(struct namespace *namespace) 17221da177e4SLinus Torvalds { 17231ce88cf4SMiklos Szeredi struct vfsmount *root = namespace->root; 172470fbcdf4SRam Pai LIST_HEAD(umount_list); 17251ce88cf4SMiklos Szeredi namespace->root = NULL; 17261ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 1727390c6843SRam Pai down_write(&namespace_sem); 17281da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1729a05964f3SRam Pai umount_tree(root, 0, &umount_list); 17301da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1731390c6843SRam Pai up_write(&namespace_sem); 173270fbcdf4SRam Pai release_mounts(&umount_list); 17331da177e4SLinus Torvalds kfree(namespace); 17341da177e4SLinus Torvalds } 1735