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/syscalls.h> 121da177e4SLinus Torvalds #include <linux/slab.h> 131da177e4SLinus Torvalds #include <linux/sched.h> 141da177e4SLinus Torvalds #include <linux/smp_lock.h> 151da177e4SLinus Torvalds #include <linux/init.h> 1615a67dd8SRandy Dunlap #include <linux/kernel.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> 21f20a9eadSAndrew Morton #include <linux/sysfs.h> 221da177e4SLinus Torvalds #include <linux/seq_file.h> 236b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 241da177e4SLinus Torvalds #include <linux/namei.h> 251da177e4SLinus Torvalds #include <linux/security.h> 261da177e4SLinus Torvalds #include <linux/mount.h> 2707f3f05cSDavid Howells #include <linux/ramfs.h> 281da177e4SLinus Torvalds #include <asm/uaccess.h> 291da177e4SLinus Torvalds #include <asm/unistd.h> 3007b20889SRam Pai #include "pnode.h" 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 331da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 341da177e4SLinus Torvalds 355addc5ddSAl Viro static int event; 365addc5ddSAl Viro 37fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 386c231b7bSRavikiran G Thirumalai static int hash_mask __read_mostly, hash_bits __read_mostly; 39e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 40390c6843SRam Pai static struct rw_semaphore namespace_sem; 411da177e4SLinus Torvalds 42f87fd4c2SMiklos Szeredi /* /sys/fs */ 43f87fd4c2SMiklos Szeredi decl_subsys(fs, NULL, NULL); 44f87fd4c2SMiklos Szeredi EXPORT_SYMBOL_GPL(fs_subsys); 45f87fd4c2SMiklos Szeredi 461da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 471da177e4SLinus Torvalds { 481da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 491da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 501da177e4SLinus Torvalds tmp = tmp + (tmp >> hash_bits); 511da177e4SLinus Torvalds return tmp & hash_mask; 521da177e4SLinus Torvalds } 531da177e4SLinus Torvalds 541da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 551da177e4SLinus Torvalds { 56*c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 571da177e4SLinus Torvalds if (mnt) { 581da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 591da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 601da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 611da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 621da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 6355e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 6403e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 65a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 66a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 671da177e4SLinus Torvalds if (name) { 681da177e4SLinus Torvalds int size = strlen(name) + 1; 691da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 701da177e4SLinus Torvalds if (newname) { 711da177e4SLinus Torvalds memcpy(newname, name, size); 721da177e4SLinus Torvalds mnt->mnt_devname = newname; 731da177e4SLinus Torvalds } 741da177e4SLinus Torvalds } 751da177e4SLinus Torvalds } 761da177e4SLinus Torvalds return mnt; 771da177e4SLinus Torvalds } 781da177e4SLinus Torvalds 79454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 80454e2398SDavid Howells { 81454e2398SDavid Howells mnt->mnt_sb = sb; 82454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 83454e2398SDavid Howells return 0; 84454e2398SDavid Howells } 85454e2398SDavid Howells 86454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 87454e2398SDavid Howells 881da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 891da177e4SLinus Torvalds { 901da177e4SLinus Torvalds kfree(mnt->mnt_devname); 911da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 921da177e4SLinus Torvalds } 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* 95a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 96a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 971da177e4SLinus Torvalds */ 98a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 99a05964f3SRam Pai int dir) 1001da177e4SLinus Torvalds { 1011da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 1021da177e4SLinus Torvalds struct list_head *tmp = head; 1031da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 1041da177e4SLinus Torvalds 1051da177e4SLinus Torvalds for (;;) { 106a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 1071da177e4SLinus Torvalds p = NULL; 1081da177e4SLinus Torvalds if (tmp == head) 1091da177e4SLinus Torvalds break; 1101da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 1111da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 112a05964f3SRam Pai found = p; 1131da177e4SLinus Torvalds break; 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds return found; 1171da177e4SLinus Torvalds } 1181da177e4SLinus Torvalds 119a05964f3SRam Pai /* 120a05964f3SRam Pai * lookup_mnt increments the ref count before returning 121a05964f3SRam Pai * the vfsmount struct. 122a05964f3SRam Pai */ 123a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 124a05964f3SRam Pai { 125a05964f3SRam Pai struct vfsmount *child_mnt; 126a05964f3SRam Pai spin_lock(&vfsmount_lock); 127a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 128a05964f3SRam Pai mntget(child_mnt); 129a05964f3SRam Pai spin_unlock(&vfsmount_lock); 130a05964f3SRam Pai return child_mnt; 131a05964f3SRam Pai } 132a05964f3SRam Pai 1331da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 1341da177e4SLinus Torvalds { 1356b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds 1386b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 1395addc5ddSAl Viro { 1405addc5ddSAl Viro if (ns) { 1415addc5ddSAl Viro ns->event = ++event; 1425addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1435addc5ddSAl Viro } 1445addc5ddSAl Viro } 1455addc5ddSAl Viro 1466b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 1475addc5ddSAl Viro { 1485addc5ddSAl Viro if (ns && ns->event != event) { 1495addc5ddSAl Viro ns->event = event; 1505addc5ddSAl Viro wake_up_interruptible(&ns->poll); 1515addc5ddSAl Viro } 1525addc5ddSAl Viro } 1535addc5ddSAl Viro 1541da177e4SLinus Torvalds static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) 1551da177e4SLinus Torvalds { 1561da177e4SLinus Torvalds old_nd->dentry = mnt->mnt_mountpoint; 1571da177e4SLinus Torvalds old_nd->mnt = mnt->mnt_parent; 1581da177e4SLinus Torvalds mnt->mnt_parent = mnt; 1591da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 1601da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 1611da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 1621da177e4SLinus Torvalds old_nd->dentry->d_mounted--; 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 165b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 166b90fa9aeSRam Pai struct vfsmount *child_mnt) 167b90fa9aeSRam Pai { 168b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 169b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 170b90fa9aeSRam Pai dentry->d_mounted++; 171b90fa9aeSRam Pai } 172b90fa9aeSRam Pai 1731da177e4SLinus Torvalds static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) 1741da177e4SLinus Torvalds { 175b90fa9aeSRam Pai mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); 176b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 177b90fa9aeSRam Pai hash(nd->mnt, nd->dentry)); 1781da177e4SLinus Torvalds list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); 179b90fa9aeSRam Pai } 180b90fa9aeSRam Pai 181b90fa9aeSRam Pai /* 182b90fa9aeSRam Pai * the caller must hold vfsmount_lock 183b90fa9aeSRam Pai */ 184b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 185b90fa9aeSRam Pai { 186b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 187b90fa9aeSRam Pai struct vfsmount *m; 188b90fa9aeSRam Pai LIST_HEAD(head); 1896b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 190b90fa9aeSRam Pai 191b90fa9aeSRam Pai BUG_ON(parent == mnt); 192b90fa9aeSRam Pai 193b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 194b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 1956b3286edSKirill Korotaev m->mnt_ns = n; 196b90fa9aeSRam Pai list_splice(&head, n->list.prev); 197b90fa9aeSRam Pai 198b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 199b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 200b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 2016b3286edSKirill Korotaev touch_mnt_namespace(n); 2021da177e4SLinus Torvalds } 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 2051da177e4SLinus Torvalds { 2061da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 2071da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 2081da177e4SLinus Torvalds while (1) { 2091da177e4SLinus Torvalds if (p == root) 2101da177e4SLinus Torvalds return NULL; 2111da177e4SLinus Torvalds next = p->mnt_child.next; 2121da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 2131da177e4SLinus Torvalds break; 2141da177e4SLinus Torvalds p = p->mnt_parent; 2151da177e4SLinus Torvalds } 2161da177e4SLinus Torvalds } 2171da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds 2209676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 2219676f0c6SRam Pai { 2229676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 2239676f0c6SRam Pai while (prev != &p->mnt_mounts) { 2249676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 2259676f0c6SRam Pai prev = p->mnt_mounts.prev; 2269676f0c6SRam Pai } 2279676f0c6SRam Pai return p; 2289676f0c6SRam Pai } 2299676f0c6SRam Pai 23036341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 23136341f64SRam Pai int flag) 2321da177e4SLinus Torvalds { 2331da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 2341da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 2351da177e4SLinus Torvalds 2361da177e4SLinus Torvalds if (mnt) { 2371da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 2381da177e4SLinus Torvalds atomic_inc(&sb->s_active); 2391da177e4SLinus Torvalds mnt->mnt_sb = sb; 2401da177e4SLinus Torvalds mnt->mnt_root = dget(root); 2411da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 2421da177e4SLinus Torvalds mnt->mnt_parent = mnt; 243b90fa9aeSRam Pai 2445afe0022SRam Pai if (flag & CL_SLAVE) { 2455afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 2465afe0022SRam Pai mnt->mnt_master = old; 2475afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 2485afe0022SRam Pai } else { 249b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 250b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 2515afe0022SRam Pai if (IS_MNT_SLAVE(old)) 2525afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 2535afe0022SRam Pai mnt->mnt_master = old->mnt_master; 2545afe0022SRam Pai } 255b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 256b90fa9aeSRam Pai set_mnt_shared(mnt); 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 2591da177e4SLinus Torvalds * as the original if that was on one */ 26036341f64SRam Pai if (flag & CL_EXPIRE) { 2611da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 26255e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 26355e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 2641da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2651da177e4SLinus Torvalds } 26636341f64SRam Pai } 2671da177e4SLinus Torvalds return mnt; 2681da177e4SLinus Torvalds } 2691da177e4SLinus Torvalds 2707b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 2711da177e4SLinus Torvalds { 2721da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 2731da177e4SLinus Torvalds dput(mnt->mnt_root); 2741da177e4SLinus Torvalds free_vfsmnt(mnt); 2751da177e4SLinus Torvalds deactivate_super(sb); 2761da177e4SLinus Torvalds } 2771da177e4SLinus Torvalds 2787b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 2797b7b1aceSAl Viro { 2807b7b1aceSAl Viro repeat: 2817b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 2827b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 2837b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2847b7b1aceSAl Viro __mntput(mnt); 2857b7b1aceSAl Viro return; 2867b7b1aceSAl Viro } 2877b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 2887b7b1aceSAl Viro mnt->mnt_pinned = 0; 2897b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 2907b7b1aceSAl Viro acct_auto_close_mnt(mnt); 2917b7b1aceSAl Viro security_sb_umount_close(mnt); 2927b7b1aceSAl Viro goto repeat; 2937b7b1aceSAl Viro } 2947b7b1aceSAl Viro } 2957b7b1aceSAl Viro 2967b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 2977b7b1aceSAl Viro 2987b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 2997b7b1aceSAl Viro { 3007b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3017b7b1aceSAl Viro mnt->mnt_pinned++; 3027b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3037b7b1aceSAl Viro } 3047b7b1aceSAl Viro 3057b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 3067b7b1aceSAl Viro 3077b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 3087b7b1aceSAl Viro { 3097b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3107b7b1aceSAl Viro if (mnt->mnt_pinned) { 3117b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 3127b7b1aceSAl Viro mnt->mnt_pinned--; 3137b7b1aceSAl Viro } 3147b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3157b7b1aceSAl Viro } 3167b7b1aceSAl Viro 3177b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds /* iterator */ 3201da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 3211da177e4SLinus Torvalds { 3226b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 3231da177e4SLinus Torvalds struct list_head *p; 3241da177e4SLinus Torvalds loff_t l = *pos; 3251da177e4SLinus Torvalds 326390c6843SRam Pai down_read(&namespace_sem); 3271da177e4SLinus Torvalds list_for_each(p, &n->list) 3281da177e4SLinus Torvalds if (!l--) 3291da177e4SLinus Torvalds return list_entry(p, struct vfsmount, mnt_list); 3301da177e4SLinus Torvalds return NULL; 3311da177e4SLinus Torvalds } 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 3341da177e4SLinus Torvalds { 3356b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 3361da177e4SLinus Torvalds struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; 3371da177e4SLinus Torvalds (*pos)++; 3381da177e4SLinus Torvalds return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3411da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 3421da177e4SLinus Torvalds { 343390c6843SRam Pai up_read(&namespace_sem); 3441da177e4SLinus Torvalds } 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds static inline void mangle(struct seq_file *m, const char *s) 3471da177e4SLinus Torvalds { 3481da177e4SLinus Torvalds seq_escape(m, s, " \t\n\\"); 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 3521da177e4SLinus Torvalds { 3531da177e4SLinus Torvalds struct vfsmount *mnt = v; 3541da177e4SLinus Torvalds int err = 0; 3551da177e4SLinus Torvalds static struct proc_fs_info { 3561da177e4SLinus Torvalds int flag; 3571da177e4SLinus Torvalds char *str; 3581da177e4SLinus Torvalds } fs_info[] = { 3591da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 3601da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 3611da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 3621da177e4SLinus Torvalds { 0, NULL } 3631da177e4SLinus Torvalds }; 3641da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 3651da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 3661da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 3671da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 368fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 369fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 37047ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 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; 463e3474a8eSIan Kent 464e3474a8eSIan 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 { 484e3474a8eSIan Kent int ret = 1; 485a05964f3SRam Pai spin_lock(&vfsmount_lock); 486a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 487e3474a8eSIan 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 5201bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 5211bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 52270fbcdf4SRam Pai 523a05964f3SRam Pai if (propagate) 524a05964f3SRam Pai propagate_umount(kill); 525a05964f3SRam Pai 52670fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 52770fbcdf4SRam Pai list_del_init(&p->mnt_expire); 52870fbcdf4SRam Pai list_del_init(&p->mnt_list); 5296b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 5306b3286edSKirill Korotaev p->mnt_ns = NULL; 53170fbcdf4SRam Pai list_del_init(&p->mnt_child); 53270fbcdf4SRam Pai if (p->mnt_parent != p) 533f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 534a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 5351da177e4SLinus Torvalds } 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds 5381da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 5391da177e4SLinus Torvalds { 5401da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 5411da177e4SLinus Torvalds int retval; 54270fbcdf4SRam Pai LIST_HEAD(umount_list); 5431da177e4SLinus Torvalds 5441da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 5451da177e4SLinus Torvalds if (retval) 5461da177e4SLinus Torvalds return retval; 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds /* 5491da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 5501da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 5511da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 5521da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 5531da177e4SLinus Torvalds */ 5541da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 5551da177e4SLinus Torvalds if (mnt == current->fs->rootmnt || 5561da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 5571da177e4SLinus Torvalds return -EINVAL; 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 5601da177e4SLinus Torvalds return -EBUSY; 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 5631da177e4SLinus Torvalds return -EAGAIN; 5641da177e4SLinus Torvalds } 5651da177e4SLinus Torvalds 5661da177e4SLinus Torvalds /* 5671da177e4SLinus Torvalds * If we may have to abort operations to get out of this 5681da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 5691da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 5701da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 5711da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 5721da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 5731da177e4SLinus Torvalds * about for the moment. 5741da177e4SLinus Torvalds */ 5751da177e4SLinus Torvalds 5761da177e4SLinus Torvalds lock_kernel(); 5778b512d9aSTrond Myklebust if (sb->s_op->umount_begin) 5788b512d9aSTrond Myklebust sb->s_op->umount_begin(mnt, flags); 5791da177e4SLinus Torvalds unlock_kernel(); 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds /* 5821da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 5831da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 5841da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 5851da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 5861da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 5871da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 5881da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 5891da177e4SLinus Torvalds */ 5901da177e4SLinus Torvalds if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { 5911da177e4SLinus Torvalds /* 5921da177e4SLinus Torvalds * Special case for "unmounting" root ... 5931da177e4SLinus Torvalds * we just try to remount it readonly. 5941da177e4SLinus Torvalds */ 5951da177e4SLinus Torvalds down_write(&sb->s_umount); 5961da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 5971da177e4SLinus Torvalds lock_kernel(); 5981da177e4SLinus Torvalds DQUOT_OFF(sb); 5991da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 6001da177e4SLinus Torvalds unlock_kernel(); 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds up_write(&sb->s_umount); 6031da177e4SLinus Torvalds return retval; 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds 606390c6843SRam Pai down_write(&namespace_sem); 6071da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6085addc5ddSAl Viro event++; 6091da177e4SLinus Torvalds 6101da177e4SLinus Torvalds retval = -EBUSY; 611a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 6121da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 613a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 6141da177e4SLinus Torvalds retval = 0; 6151da177e4SLinus Torvalds } 6161da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6171da177e4SLinus Torvalds if (retval) 6181da177e4SLinus Torvalds security_sb_umount_busy(mnt); 619390c6843SRam Pai up_write(&namespace_sem); 62070fbcdf4SRam Pai release_mounts(&umount_list); 6211da177e4SLinus Torvalds return retval; 6221da177e4SLinus Torvalds } 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 6261da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 6271da177e4SLinus Torvalds * 6281da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 6291da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 6301da177e4SLinus Torvalds */ 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 6331da177e4SLinus Torvalds { 6341da177e4SLinus Torvalds struct nameidata nd; 6351da177e4SLinus Torvalds int retval; 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 6381da177e4SLinus Torvalds if (retval) 6391da177e4SLinus Torvalds goto out; 6401da177e4SLinus Torvalds retval = -EINVAL; 6411da177e4SLinus Torvalds if (nd.dentry != nd.mnt->mnt_root) 6421da177e4SLinus Torvalds goto dput_and_out; 6431da177e4SLinus Torvalds if (!check_mnt(nd.mnt)) 6441da177e4SLinus Torvalds goto dput_and_out; 6451da177e4SLinus Torvalds 6461da177e4SLinus Torvalds retval = -EPERM; 6471da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 6481da177e4SLinus Torvalds goto dput_and_out; 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds retval = do_umount(nd.mnt, flags); 6511da177e4SLinus Torvalds dput_and_out: 6521da177e4SLinus Torvalds path_release_on_umount(&nd); 6531da177e4SLinus Torvalds out: 6541da177e4SLinus Torvalds return retval; 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds /* 6601da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 6611da177e4SLinus Torvalds */ 6621da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 6631da177e4SLinus Torvalds { 6641da177e4SLinus Torvalds return sys_umount(name, 0); 6651da177e4SLinus Torvalds } 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds #endif 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 6701da177e4SLinus Torvalds { 6711da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 6721da177e4SLinus Torvalds return 0; 6731da177e4SLinus Torvalds return -EPERM; 6741da177e4SLinus Torvalds #ifdef notyet 6751da177e4SLinus Torvalds if (S_ISLNK(nd->dentry->d_inode->i_mode)) 6761da177e4SLinus Torvalds return -EPERM; 6771da177e4SLinus Torvalds if (nd->dentry->d_inode->i_mode & S_ISVTX) { 6781da177e4SLinus Torvalds if (current->uid != nd->dentry->d_inode->i_uid) 6791da177e4SLinus Torvalds return -EPERM; 6801da177e4SLinus Torvalds } 681e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 6821da177e4SLinus Torvalds return -EPERM; 6831da177e4SLinus Torvalds return 0; 6841da177e4SLinus Torvalds #endif 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 687b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 6881da177e4SLinus Torvalds { 6891da177e4SLinus Torvalds while (1) { 6901da177e4SLinus Torvalds if (d == dentry) 6911da177e4SLinus Torvalds return 1; 6921da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 6931da177e4SLinus Torvalds return 0; 6941da177e4SLinus Torvalds d = d->d_parent; 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 698b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 69936341f64SRam Pai int flag) 7001da177e4SLinus Torvalds { 7011da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 7021da177e4SLinus Torvalds struct nameidata nd; 7031da177e4SLinus Torvalds 7049676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 7059676f0c6SRam Pai return NULL; 7069676f0c6SRam Pai 70736341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 7081da177e4SLinus Torvalds if (!q) 7091da177e4SLinus Torvalds goto Enomem; 7101da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 7111da177e4SLinus Torvalds 7121da177e4SLinus Torvalds p = mnt; 713fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 7141da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 7151da177e4SLinus Torvalds continue; 7161da177e4SLinus Torvalds 7171da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 7189676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 7199676f0c6SRam Pai s = skip_mnt_tree(s); 7209676f0c6SRam Pai continue; 7219676f0c6SRam Pai } 7221da177e4SLinus Torvalds while (p != s->mnt_parent) { 7231da177e4SLinus Torvalds p = p->mnt_parent; 7241da177e4SLinus Torvalds q = q->mnt_parent; 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds p = s; 7271da177e4SLinus Torvalds nd.mnt = q; 7281da177e4SLinus Torvalds nd.dentry = p->mnt_mountpoint; 72936341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 7301da177e4SLinus Torvalds if (!q) 7311da177e4SLinus Torvalds goto Enomem; 7321da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 7331da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 7341da177e4SLinus Torvalds attach_mnt(q, &nd); 7351da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds return res; 7391da177e4SLinus Torvalds Enomem: 7401da177e4SLinus Torvalds if (res) { 74170fbcdf4SRam Pai LIST_HEAD(umount_list); 7421da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 743a05964f3SRam Pai umount_tree(res, 0, &umount_list); 7441da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 74570fbcdf4SRam Pai release_mounts(&umount_list); 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds return NULL; 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds 750b90fa9aeSRam Pai /* 751b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 752b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 75321444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 75421444403SRam Pai * store the parent mount and mountpoint dentry. 75521444403SRam Pai * (done when source_mnt is moved) 756b90fa9aeSRam Pai * 757b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 758b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 7599676f0c6SRam Pai * --------------------------------------------------------------------------- 760b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 7619676f0c6SRam Pai * |************************************************************************** 7629676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7639676f0c6SRam Pai * | dest | | | | | 7649676f0c6SRam Pai * | | | | | | | 7659676f0c6SRam Pai * | v | | | | | 7669676f0c6SRam Pai * |************************************************************************** 7679676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 7685afe0022SRam Pai * | | | | | | 7699676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 7709676f0c6SRam Pai * *************************************************************************** 771b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 772b90fa9aeSRam Pai * destination mount. 773b90fa9aeSRam Pai * 774b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 775b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 776b90fa9aeSRam Pai * the peer group of the source mount. 777b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 778b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 779b90fa9aeSRam Pai * mount. 7805afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 7815afe0022SRam Pai * of the destination mount and the cloned mount is made slave 7825afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 7835afe0022SRam Pai * is marked as 'shared and slave'. 7845afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 7855afe0022SRam Pai * source mount. 7865afe0022SRam Pai * 7879676f0c6SRam Pai * --------------------------------------------------------------------------- 78821444403SRam Pai * | MOVE MOUNT OPERATION | 7899676f0c6SRam Pai * |************************************************************************** 7909676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 7919676f0c6SRam Pai * | dest | | | | | 7929676f0c6SRam Pai * | | | | | | | 7939676f0c6SRam Pai * | v | | | | | 7949676f0c6SRam Pai * |************************************************************************** 7959676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 7965afe0022SRam Pai * | | | | | | 7979676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 7989676f0c6SRam Pai * *************************************************************************** 7995afe0022SRam Pai * 8005afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 8015afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 80221444403SRam Pai * (+*) the mount is moved to the destination. 8035afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 8045afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 8055afe0022SRam Pai * the mount is marked as 'shared and slave'. 8065afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 807b90fa9aeSRam Pai * 808b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 809b90fa9aeSRam Pai * applied to each mount in the tree. 810b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 811b90fa9aeSRam Pai * in allocations. 812b90fa9aeSRam Pai */ 813b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 81421444403SRam Pai struct nameidata *nd, struct nameidata *parent_nd) 815b90fa9aeSRam Pai { 816b90fa9aeSRam Pai LIST_HEAD(tree_list); 817b90fa9aeSRam Pai struct vfsmount *dest_mnt = nd->mnt; 818b90fa9aeSRam Pai struct dentry *dest_dentry = nd->dentry; 819b90fa9aeSRam Pai struct vfsmount *child, *p; 820b90fa9aeSRam Pai 821b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 822b90fa9aeSRam Pai return -EINVAL; 823b90fa9aeSRam Pai 824b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 825b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 826b90fa9aeSRam Pai set_mnt_shared(p); 827b90fa9aeSRam Pai } 828b90fa9aeSRam Pai 829b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 83021444403SRam Pai if (parent_nd) { 83121444403SRam Pai detach_mnt(source_mnt, parent_nd); 83221444403SRam Pai attach_mnt(source_mnt, nd); 8336b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 83421444403SRam Pai } else { 835b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 836b90fa9aeSRam Pai commit_tree(source_mnt); 83721444403SRam Pai } 838b90fa9aeSRam Pai 839b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 840b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 841b90fa9aeSRam Pai commit_tree(child); 842b90fa9aeSRam Pai } 843b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 844b90fa9aeSRam Pai return 0; 845b90fa9aeSRam Pai } 846b90fa9aeSRam Pai 8471da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 8481da177e4SLinus Torvalds { 8491da177e4SLinus Torvalds int err; 8501da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 8511da177e4SLinus Torvalds return -EINVAL; 8521da177e4SLinus Torvalds 8531da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 8541da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 8551da177e4SLinus Torvalds return -ENOTDIR; 8561da177e4SLinus Torvalds 8571da177e4SLinus Torvalds err = -ENOENT; 8581b1dcc1bSJes Sorensen mutex_lock(&nd->dentry->d_inode->i_mutex); 8591da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 8601da177e4SLinus Torvalds goto out_unlock; 8611da177e4SLinus Torvalds 8621da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 8631da177e4SLinus Torvalds if (err) 8641da177e4SLinus Torvalds goto out_unlock; 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds err = -ENOENT; 867b90fa9aeSRam Pai if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) 86821444403SRam Pai err = attach_recursive_mnt(mnt, nd, NULL); 8691da177e4SLinus Torvalds out_unlock: 8701b1dcc1bSJes Sorensen mutex_unlock(&nd->dentry->d_inode->i_mutex); 8711da177e4SLinus Torvalds if (!err) 8721da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 8731da177e4SLinus Torvalds return err; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds /* 87707b20889SRam Pai * recursively change the type of the mountpoint. 87807b20889SRam Pai */ 87907b20889SRam Pai static int do_change_type(struct nameidata *nd, int flag) 88007b20889SRam Pai { 88107b20889SRam Pai struct vfsmount *m, *mnt = nd->mnt; 88207b20889SRam Pai int recurse = flag & MS_REC; 88307b20889SRam Pai int type = flag & ~MS_REC; 88407b20889SRam Pai 88507b20889SRam Pai if (nd->dentry != nd->mnt->mnt_root) 88607b20889SRam Pai return -EINVAL; 88707b20889SRam Pai 88807b20889SRam Pai down_write(&namespace_sem); 88907b20889SRam Pai spin_lock(&vfsmount_lock); 89007b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 89107b20889SRam Pai change_mnt_propagation(m, type); 89207b20889SRam Pai spin_unlock(&vfsmount_lock); 89307b20889SRam Pai up_write(&namespace_sem); 89407b20889SRam Pai return 0; 89507b20889SRam Pai } 89607b20889SRam Pai 89707b20889SRam Pai /* 8981da177e4SLinus Torvalds * do loopback mount. 8991da177e4SLinus Torvalds */ 900eee391a6SAndrew Morton static int do_loopback(struct nameidata *nd, char *old_name, int recurse) 9011da177e4SLinus Torvalds { 9021da177e4SLinus Torvalds struct nameidata old_nd; 9031da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 9041da177e4SLinus Torvalds int err = mount_is_safe(nd); 9051da177e4SLinus Torvalds if (err) 9061da177e4SLinus Torvalds return err; 9071da177e4SLinus Torvalds if (!old_name || !*old_name) 9081da177e4SLinus Torvalds return -EINVAL; 9091da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 9101da177e4SLinus Torvalds if (err) 9111da177e4SLinus Torvalds return err; 9121da177e4SLinus Torvalds 913390c6843SRam Pai down_write(&namespace_sem); 9141da177e4SLinus Torvalds err = -EINVAL; 9159676f0c6SRam Pai if (IS_MNT_UNBINDABLE(old_nd.mnt)) 9169676f0c6SRam Pai goto out; 9179676f0c6SRam Pai 918ccd48bc7SAl Viro if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 919ccd48bc7SAl Viro goto out; 920ccd48bc7SAl Viro 9211da177e4SLinus Torvalds err = -ENOMEM; 9221da177e4SLinus Torvalds if (recurse) 92336341f64SRam Pai mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); 9241da177e4SLinus Torvalds else 92536341f64SRam Pai mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); 9261da177e4SLinus Torvalds 927ccd48bc7SAl Viro if (!mnt) 928ccd48bc7SAl Viro goto out; 929ccd48bc7SAl Viro 9301da177e4SLinus Torvalds err = graft_tree(mnt, nd); 9311da177e4SLinus Torvalds if (err) { 93270fbcdf4SRam Pai LIST_HEAD(umount_list); 9331da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 934a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 9351da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 93670fbcdf4SRam Pai release_mounts(&umount_list); 9375b83d2c5SRam Pai } 9381da177e4SLinus Torvalds 939ccd48bc7SAl Viro out: 940390c6843SRam Pai up_write(&namespace_sem); 9411da177e4SLinus Torvalds path_release(&old_nd); 9421da177e4SLinus Torvalds return err; 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds 9451da177e4SLinus Torvalds /* 9461da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 9471da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 9481da177e4SLinus Torvalds * on it - tough luck. 9491da177e4SLinus Torvalds */ 9501da177e4SLinus Torvalds static int do_remount(struct nameidata *nd, int flags, int mnt_flags, 9511da177e4SLinus Torvalds void *data) 9521da177e4SLinus Torvalds { 9531da177e4SLinus Torvalds int err; 9541da177e4SLinus Torvalds struct super_block *sb = nd->mnt->mnt_sb; 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9571da177e4SLinus Torvalds return -EPERM; 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 9601da177e4SLinus Torvalds return -EINVAL; 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds if (nd->dentry != nd->mnt->mnt_root) 9631da177e4SLinus Torvalds return -EINVAL; 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds down_write(&sb->s_umount); 9661da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 9671da177e4SLinus Torvalds if (!err) 9681da177e4SLinus Torvalds nd->mnt->mnt_flags = mnt_flags; 9691da177e4SLinus Torvalds up_write(&sb->s_umount); 9701da177e4SLinus Torvalds if (!err) 9711da177e4SLinus Torvalds security_sb_post_remount(nd->mnt, flags, data); 9721da177e4SLinus Torvalds return err; 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds 9759676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 9769676f0c6SRam Pai { 9779676f0c6SRam Pai struct vfsmount *p; 9789676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9799676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 9809676f0c6SRam Pai return 1; 9819676f0c6SRam Pai } 9829676f0c6SRam Pai return 0; 9839676f0c6SRam Pai } 9849676f0c6SRam Pai 9851da177e4SLinus Torvalds static int do_move_mount(struct nameidata *nd, char *old_name) 9861da177e4SLinus Torvalds { 9871da177e4SLinus Torvalds struct nameidata old_nd, parent_nd; 9881da177e4SLinus Torvalds struct vfsmount *p; 9891da177e4SLinus Torvalds int err = 0; 9901da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9911da177e4SLinus Torvalds return -EPERM; 9921da177e4SLinus Torvalds if (!old_name || !*old_name) 9931da177e4SLinus Torvalds return -EINVAL; 9941da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 9951da177e4SLinus Torvalds if (err) 9961da177e4SLinus Torvalds return err; 9971da177e4SLinus Torvalds 998390c6843SRam Pai down_write(&namespace_sem); 9991da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 10001da177e4SLinus Torvalds ; 10011da177e4SLinus Torvalds err = -EINVAL; 10021da177e4SLinus Torvalds if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 10031da177e4SLinus Torvalds goto out; 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds err = -ENOENT; 10061b1dcc1bSJes Sorensen mutex_lock(&nd->dentry->d_inode->i_mutex); 10071da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 10081da177e4SLinus Torvalds goto out1; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) 101121444403SRam Pai goto out1; 10121da177e4SLinus Torvalds 10131da177e4SLinus Torvalds err = -EINVAL; 10141da177e4SLinus Torvalds if (old_nd.dentry != old_nd.mnt->mnt_root) 101521444403SRam Pai goto out1; 10161da177e4SLinus Torvalds 10171da177e4SLinus Torvalds if (old_nd.mnt == old_nd.mnt->mnt_parent) 101821444403SRam Pai goto out1; 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 10211da177e4SLinus Torvalds S_ISDIR(old_nd.dentry->d_inode->i_mode)) 102221444403SRam Pai goto out1; 102321444403SRam Pai /* 102421444403SRam Pai * Don't move a mount residing in a shared parent. 102521444403SRam Pai */ 102621444403SRam Pai if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) 102721444403SRam Pai goto out1; 10289676f0c6SRam Pai /* 10299676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 10309676f0c6SRam Pai * mount which is shared. 10319676f0c6SRam Pai */ 10329676f0c6SRam Pai if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt)) 10339676f0c6SRam Pai goto out1; 10341da177e4SLinus Torvalds err = -ELOOP; 10351da177e4SLinus Torvalds for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) 10361da177e4SLinus Torvalds if (p == old_nd.mnt) 103721444403SRam Pai goto out1; 10381da177e4SLinus Torvalds 103921444403SRam Pai if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) 104021444403SRam Pai goto out1; 10411da177e4SLinus Torvalds 104221444403SRam Pai spin_lock(&vfsmount_lock); 10431da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 10441da177e4SLinus Torvalds * automatically */ 104555e700b9SMiklos Szeredi list_del_init(&old_nd.mnt->mnt_expire); 10461da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10471da177e4SLinus Torvalds out1: 10481b1dcc1bSJes Sorensen mutex_unlock(&nd->dentry->d_inode->i_mutex); 10491da177e4SLinus Torvalds out: 1050390c6843SRam Pai up_write(&namespace_sem); 10511da177e4SLinus Torvalds if (!err) 10521da177e4SLinus Torvalds path_release(&parent_nd); 10531da177e4SLinus Torvalds path_release(&old_nd); 10541da177e4SLinus Torvalds return err; 10551da177e4SLinus Torvalds } 10561da177e4SLinus Torvalds 10571da177e4SLinus Torvalds /* 10581da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 10591da177e4SLinus Torvalds * namespace's tree 10601da177e4SLinus Torvalds */ 10611da177e4SLinus Torvalds static int do_new_mount(struct nameidata *nd, char *type, int flags, 10621da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 10631da177e4SLinus Torvalds { 10641da177e4SLinus Torvalds struct vfsmount *mnt; 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 10671da177e4SLinus Torvalds return -EINVAL; 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds /* we need capabilities... */ 10701da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10711da177e4SLinus Torvalds return -EPERM; 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 10741da177e4SLinus Torvalds if (IS_ERR(mnt)) 10751da177e4SLinus Torvalds return PTR_ERR(mnt); 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds /* 10811da177e4SLinus Torvalds * add a mount into a namespace's mount tree 10821da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 10831da177e4SLinus Torvalds */ 10841da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 10851da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 10861da177e4SLinus Torvalds { 10871da177e4SLinus Torvalds int err; 10881da177e4SLinus Torvalds 1089390c6843SRam Pai down_write(&namespace_sem); 10901da177e4SLinus Torvalds /* Something was mounted here while we slept */ 10911da177e4SLinus Torvalds while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 10921da177e4SLinus Torvalds ; 10931da177e4SLinus Torvalds err = -EINVAL; 10941da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 10951da177e4SLinus Torvalds goto unlock; 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 10981da177e4SLinus Torvalds err = -EBUSY; 10991da177e4SLinus Torvalds if (nd->mnt->mnt_sb == newmnt->mnt_sb && 11001da177e4SLinus Torvalds nd->mnt->mnt_root == nd->dentry) 11011da177e4SLinus Torvalds goto unlock; 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds err = -EINVAL; 11041da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 11051da177e4SLinus Torvalds goto unlock; 11061da177e4SLinus Torvalds 11071da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 11085b83d2c5SRam Pai if ((err = graft_tree(newmnt, nd))) 11095b83d2c5SRam Pai goto unlock; 11101da177e4SLinus Torvalds 11115b83d2c5SRam Pai if (fslist) { 11121da177e4SLinus Torvalds /* add to the specified expiration list */ 11131da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 111455e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 11151da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11161da177e4SLinus Torvalds } 1117390c6843SRam Pai up_write(&namespace_sem); 11185b83d2c5SRam Pai return 0; 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds unlock: 1121390c6843SRam Pai up_write(&namespace_sem); 11221da177e4SLinus Torvalds mntput(newmnt); 11231da177e4SLinus Torvalds return err; 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 11271da177e4SLinus Torvalds 112870fbcdf4SRam Pai static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, 112970fbcdf4SRam Pai struct list_head *umounts) 113024ca2af1SMiklos Szeredi { 113124ca2af1SMiklos Szeredi spin_lock(&vfsmount_lock); 113224ca2af1SMiklos Szeredi 113324ca2af1SMiklos Szeredi /* 1134ed42c879SMiklos Szeredi * Check if mount is still attached, if not, let whoever holds it deal 1135ed42c879SMiklos Szeredi * with the sucker 1136ed42c879SMiklos Szeredi */ 1137ed42c879SMiklos Szeredi if (mnt->mnt_parent == mnt) { 1138ed42c879SMiklos Szeredi spin_unlock(&vfsmount_lock); 1139ed42c879SMiklos Szeredi return; 1140ed42c879SMiklos Szeredi } 1141ed42c879SMiklos Szeredi 1142ed42c879SMiklos Szeredi /* 114324ca2af1SMiklos Szeredi * Check that it is still dead: the count should now be 2 - as 114424ca2af1SMiklos Szeredi * contributed by the vfsmount parent and the mntget above 114524ca2af1SMiklos Szeredi */ 1146a05964f3SRam Pai if (!propagate_mount_busy(mnt, 2)) { 114724ca2af1SMiklos Szeredi /* delete from the namespace */ 11486b3286edSKirill Korotaev touch_mnt_namespace(mnt->mnt_ns); 114924ca2af1SMiklos Szeredi list_del_init(&mnt->mnt_list); 11506b3286edSKirill Korotaev mnt->mnt_ns = NULL; 1151a05964f3SRam Pai umount_tree(mnt, 1, umounts); 115224ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 115324ca2af1SMiklos Szeredi } else { 115424ca2af1SMiklos Szeredi /* 115524ca2af1SMiklos Szeredi * Someone brought it back to life whilst we didn't have any 115624ca2af1SMiklos Szeredi * locks held so return it to the expiration list 115724ca2af1SMiklos Szeredi */ 115855e700b9SMiklos Szeredi list_add_tail(&mnt->mnt_expire, mounts); 115924ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 116024ca2af1SMiklos Szeredi } 116124ca2af1SMiklos Szeredi } 116224ca2af1SMiklos Szeredi 11631da177e4SLinus Torvalds /* 11645528f911STrond Myklebust * go through the vfsmounts we've just consigned to the graveyard to 11655528f911STrond Myklebust * - check that they're still dead 11665528f911STrond Myklebust * - delete the vfsmount from the appropriate namespace under lock 11675528f911STrond Myklebust * - dispose of the corpse 11685528f911STrond Myklebust */ 11695528f911STrond Myklebust static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts) 11705528f911STrond Myklebust { 11716b3286edSKirill Korotaev struct mnt_namespace *ns; 11725528f911STrond Myklebust struct vfsmount *mnt; 11735528f911STrond Myklebust 11745528f911STrond Myklebust while (!list_empty(graveyard)) { 11755528f911STrond Myklebust LIST_HEAD(umounts); 11765528f911STrond Myklebust mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire); 11775528f911STrond Myklebust list_del_init(&mnt->mnt_expire); 11785528f911STrond Myklebust 11795528f911STrond Myklebust /* don't do anything if the namespace is dead - all the 11805528f911STrond Myklebust * vfsmounts from it are going away anyway */ 11816b3286edSKirill Korotaev ns = mnt->mnt_ns; 11826b3286edSKirill Korotaev if (!ns || !ns->root) 11835528f911STrond Myklebust continue; 11846b3286edSKirill Korotaev get_mnt_ns(ns); 11855528f911STrond Myklebust 11865528f911STrond Myklebust spin_unlock(&vfsmount_lock); 11875528f911STrond Myklebust down_write(&namespace_sem); 11885528f911STrond Myklebust expire_mount(mnt, mounts, &umounts); 11895528f911STrond Myklebust up_write(&namespace_sem); 11905528f911STrond Myklebust release_mounts(&umounts); 11915528f911STrond Myklebust mntput(mnt); 11926b3286edSKirill Korotaev put_mnt_ns(ns); 11935528f911STrond Myklebust spin_lock(&vfsmount_lock); 11945528f911STrond Myklebust } 11955528f911STrond Myklebust } 11965528f911STrond Myklebust 11975528f911STrond Myklebust /* 11981da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 11991da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 12001da177e4SLinus Torvalds * here 12011da177e4SLinus Torvalds */ 12021da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 12031da177e4SLinus Torvalds { 12041da177e4SLinus Torvalds struct vfsmount *mnt, *next; 12051da177e4SLinus Torvalds LIST_HEAD(graveyard); 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds if (list_empty(mounts)) 12081da177e4SLinus Torvalds return; 12091da177e4SLinus Torvalds 12101da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 12131da177e4SLinus Torvalds * following criteria: 12141da177e4SLinus Torvalds * - only referenced by its parent vfsmount 12151da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 12161da177e4SLinus Torvalds * cleared by mntput()) 12171da177e4SLinus Torvalds */ 121855e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 12191da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 12201da177e4SLinus Torvalds atomic_read(&mnt->mnt_count) != 1) 12211da177e4SLinus Torvalds continue; 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds mntget(mnt); 122455e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12275528f911STrond Myklebust expire_mount_list(&graveyard, mounts); 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds /* 12355528f911STrond Myklebust * Ripoff of 'select_parent()' 12365528f911STrond Myklebust * 12375528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 12385528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 12395528f911STrond Myklebust */ 12405528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 12415528f911STrond Myklebust { 12425528f911STrond Myklebust struct vfsmount *this_parent = parent; 12435528f911STrond Myklebust struct list_head *next; 12445528f911STrond Myklebust int found = 0; 12455528f911STrond Myklebust 12465528f911STrond Myklebust repeat: 12475528f911STrond Myklebust next = this_parent->mnt_mounts.next; 12485528f911STrond Myklebust resume: 12495528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 12505528f911STrond Myklebust struct list_head *tmp = next; 12515528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 12525528f911STrond Myklebust 12535528f911STrond Myklebust next = tmp->next; 12545528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 12555528f911STrond Myklebust continue; 12565528f911STrond Myklebust /* 12575528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 12585528f911STrond Myklebust */ 12595528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 12605528f911STrond Myklebust this_parent = mnt; 12615528f911STrond Myklebust goto repeat; 12625528f911STrond Myklebust } 12635528f911STrond Myklebust 12645528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 12655528f911STrond Myklebust mntget(mnt); 12665528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 12675528f911STrond Myklebust found++; 12685528f911STrond Myklebust } 12695528f911STrond Myklebust } 12705528f911STrond Myklebust /* 12715528f911STrond Myklebust * All done at this level ... ascend and resume the search 12725528f911STrond Myklebust */ 12735528f911STrond Myklebust if (this_parent != parent) { 12745528f911STrond Myklebust next = this_parent->mnt_child.next; 12755528f911STrond Myklebust this_parent = this_parent->mnt_parent; 12765528f911STrond Myklebust goto resume; 12775528f911STrond Myklebust } 12785528f911STrond Myklebust return found; 12795528f911STrond Myklebust } 12805528f911STrond Myklebust 12815528f911STrond Myklebust /* 12825528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 12835528f911STrond Myklebust * submounts of a specific parent mountpoint 12845528f911STrond Myklebust */ 12855528f911STrond Myklebust void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts) 12865528f911STrond Myklebust { 12875528f911STrond Myklebust LIST_HEAD(graveyard); 12885528f911STrond Myklebust int found; 12895528f911STrond Myklebust 12905528f911STrond Myklebust spin_lock(&vfsmount_lock); 12915528f911STrond Myklebust 12925528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 12935528f911STrond Myklebust while ((found = select_submounts(mountpoint, &graveyard)) != 0) 12945528f911STrond Myklebust expire_mount_list(&graveyard, mounts); 12955528f911STrond Myklebust 12965528f911STrond Myklebust spin_unlock(&vfsmount_lock); 12975528f911STrond Myklebust } 12985528f911STrond Myklebust 12995528f911STrond Myklebust EXPORT_SYMBOL_GPL(shrink_submounts); 13005528f911STrond Myklebust 13015528f911STrond Myklebust /* 13021da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 13031da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 13041da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 13051da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 13061da177e4SLinus Torvalds */ 1307b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1308b58fed8bSRam Pai unsigned long n) 13091da177e4SLinus Torvalds { 13101da177e4SLinus Torvalds char *t = to; 13111da177e4SLinus Torvalds const char __user *f = from; 13121da177e4SLinus Torvalds char c; 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 13151da177e4SLinus Torvalds return n; 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds while (n) { 13181da177e4SLinus Torvalds if (__get_user(c, f)) { 13191da177e4SLinus Torvalds memset(t, 0, n); 13201da177e4SLinus Torvalds break; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds *t++ = c; 13231da177e4SLinus Torvalds f++; 13241da177e4SLinus Torvalds n--; 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds return n; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds 13291da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 13301da177e4SLinus Torvalds { 13311da177e4SLinus Torvalds int i; 13321da177e4SLinus Torvalds unsigned long page; 13331da177e4SLinus Torvalds unsigned long size; 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds *where = 0; 13361da177e4SLinus Torvalds if (!data) 13371da177e4SLinus Torvalds return 0; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 13401da177e4SLinus Torvalds return -ENOMEM; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds /* We only care that *some* data at the address the user 13431da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 13441da177e4SLinus Torvalds * the remainder of the page. 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 13471da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 13481da177e4SLinus Torvalds if (size > PAGE_SIZE) 13491da177e4SLinus Torvalds size = PAGE_SIZE; 13501da177e4SLinus Torvalds 13511da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 13521da177e4SLinus Torvalds if (!i) { 13531da177e4SLinus Torvalds free_page(page); 13541da177e4SLinus Torvalds return -EFAULT; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds if (i != PAGE_SIZE) 13571da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 13581da177e4SLinus Torvalds *where = page; 13591da177e4SLinus Torvalds return 0; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds /* 13631da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 13641da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 13651da177e4SLinus Torvalds * 13661da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 13671da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 13681da177e4SLinus Torvalds * information (or be NULL). 13691da177e4SLinus Torvalds * 13701da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 13711da177e4SLinus Torvalds * When the flags word was introduced its top half was required 13721da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 13731da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 13741da177e4SLinus Torvalds * and must be discarded. 13751da177e4SLinus Torvalds */ 13761da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 13771da177e4SLinus Torvalds unsigned long flags, void *data_page) 13781da177e4SLinus Torvalds { 13791da177e4SLinus Torvalds struct nameidata nd; 13801da177e4SLinus Torvalds int retval = 0; 13811da177e4SLinus Torvalds int mnt_flags = 0; 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds /* Discard magic */ 13841da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 13851da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 13861da177e4SLinus Torvalds 13871da177e4SLinus Torvalds /* Basic sanity checks */ 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 13901da177e4SLinus Torvalds return -EINVAL; 13911da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 13921da177e4SLinus Torvalds return -EINVAL; 13931da177e4SLinus Torvalds 13941da177e4SLinus Torvalds if (data_page) 13951da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 13981da177e4SLinus Torvalds if (flags & MS_NOSUID) 13991da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 14001da177e4SLinus Torvalds if (flags & MS_NODEV) 14011da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 14021da177e4SLinus Torvalds if (flags & MS_NOEXEC) 14031da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1404fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1405fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1406fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1407fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 140847ae32d6SValerie Henson if (flags & MS_RELATIME) 140947ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 1410fc33a7bbSChristoph Hellwig 1411fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 141247ae32d6SValerie Henson MS_NOATIME | MS_NODIRATIME | MS_RELATIME); 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds /* ... and get the mountpoint */ 14151da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 14161da177e4SLinus Torvalds if (retval) 14171da177e4SLinus Torvalds return retval; 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 14201da177e4SLinus Torvalds if (retval) 14211da177e4SLinus Torvalds goto dput_out; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds if (flags & MS_REMOUNT) 14241da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 14251da177e4SLinus Torvalds data_page); 14261da177e4SLinus Torvalds else if (flags & MS_BIND) 1427eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 14289676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 142907b20889SRam Pai retval = do_change_type(&nd, flags); 14301da177e4SLinus Torvalds else if (flags & MS_MOVE) 14311da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 14321da177e4SLinus Torvalds else 14331da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 14341da177e4SLinus Torvalds dev_name, data_page); 14351da177e4SLinus Torvalds dput_out: 14361da177e4SLinus Torvalds path_release(&nd); 14371da177e4SLinus Torvalds return retval; 14381da177e4SLinus Torvalds } 14391da177e4SLinus Torvalds 1440741a2951SJANAK DESAI /* 1441741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1442741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1443741a2951SJANAK DESAI */ 14446b3286edSKirill Korotaev struct mnt_namespace *dup_mnt_ns(struct task_struct *tsk, 14456b3286edSKirill Korotaev struct fs_struct *fs) 14461da177e4SLinus Torvalds { 14476b3286edSKirill Korotaev struct mnt_namespace *mnt_ns = tsk->nsproxy->mnt_ns; 14486b3286edSKirill Korotaev struct mnt_namespace *new_ns; 14491da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 14501da177e4SLinus Torvalds struct vfsmount *p, *q; 14511da177e4SLinus Torvalds 14526b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 14531da177e4SLinus Torvalds if (!new_ns) 1454f13b8358SAdrian Bunk return NULL; 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 14571da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 14585addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 14595addc5ddSAl Viro new_ns->event = 0; 14601da177e4SLinus Torvalds 1461390c6843SRam Pai down_write(&namespace_sem); 14621da177e4SLinus Torvalds /* First pass: copy the tree topology */ 14636b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 14649676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 14651da177e4SLinus Torvalds if (!new_ns->root) { 1466390c6843SRam Pai up_write(&namespace_sem); 14671da177e4SLinus Torvalds kfree(new_ns); 1468f13b8358SAdrian Bunk return NULL; 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 14711da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 14721da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 14731da177e4SLinus Torvalds 14741da177e4SLinus Torvalds /* 14751da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 14761da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 14771da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 14781da177e4SLinus Torvalds */ 14796b3286edSKirill Korotaev p = mnt_ns->root; 14801da177e4SLinus Torvalds q = new_ns->root; 14811da177e4SLinus Torvalds while (p) { 14826b3286edSKirill Korotaev q->mnt_ns = new_ns; 14831da177e4SLinus Torvalds if (fs) { 14841da177e4SLinus Torvalds if (p == fs->rootmnt) { 14851da177e4SLinus Torvalds rootmnt = p; 14861da177e4SLinus Torvalds fs->rootmnt = mntget(q); 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds if (p == fs->pwdmnt) { 14891da177e4SLinus Torvalds pwdmnt = p; 14901da177e4SLinus Torvalds fs->pwdmnt = mntget(q); 14911da177e4SLinus Torvalds } 14921da177e4SLinus Torvalds if (p == fs->altrootmnt) { 14931da177e4SLinus Torvalds altrootmnt = p; 14941da177e4SLinus Torvalds fs->altrootmnt = mntget(q); 14951da177e4SLinus Torvalds } 14961da177e4SLinus Torvalds } 14976b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 14981da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 14991da177e4SLinus Torvalds } 1500390c6843SRam Pai up_write(&namespace_sem); 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds if (rootmnt) 15031da177e4SLinus Torvalds mntput(rootmnt); 15041da177e4SLinus Torvalds if (pwdmnt) 15051da177e4SLinus Torvalds mntput(pwdmnt); 15061da177e4SLinus Torvalds if (altrootmnt) 15071da177e4SLinus Torvalds mntput(altrootmnt); 15081da177e4SLinus Torvalds 1509741a2951SJANAK DESAI return new_ns; 1510741a2951SJANAK DESAI } 1511741a2951SJANAK DESAI 15126b3286edSKirill Korotaev int copy_mnt_ns(int flags, struct task_struct *tsk) 1513741a2951SJANAK DESAI { 15146b3286edSKirill Korotaev struct mnt_namespace *ns = tsk->nsproxy->mnt_ns; 15156b3286edSKirill Korotaev struct mnt_namespace *new_ns; 1516741a2951SJANAK DESAI int err = 0; 1517741a2951SJANAK DESAI 15186b3286edSKirill Korotaev if (!ns) 15191da177e4SLinus Torvalds return 0; 15201da177e4SLinus Torvalds 15216b3286edSKirill Korotaev get_mnt_ns(ns); 1522741a2951SJANAK DESAI 1523741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1524741a2951SJANAK DESAI return 0; 1525741a2951SJANAK DESAI 1526741a2951SJANAK DESAI if (!capable(CAP_SYS_ADMIN)) { 1527741a2951SJANAK DESAI err = -EPERM; 1528741a2951SJANAK DESAI goto out; 1529741a2951SJANAK DESAI } 1530741a2951SJANAK DESAI 15316b3286edSKirill Korotaev new_ns = dup_mnt_ns(tsk, tsk->fs); 1532741a2951SJANAK DESAI if (!new_ns) { 1533741a2951SJANAK DESAI err = -ENOMEM; 1534741a2951SJANAK DESAI goto out; 1535741a2951SJANAK DESAI } 1536741a2951SJANAK DESAI 15376b3286edSKirill Korotaev tsk->nsproxy->mnt_ns = new_ns; 1538741a2951SJANAK DESAI 15391da177e4SLinus Torvalds out: 15406b3286edSKirill Korotaev put_mnt_ns(ns); 1541741a2951SJANAK DESAI return err; 15421da177e4SLinus Torvalds } 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 15451da177e4SLinus Torvalds char __user * type, unsigned long flags, 15461da177e4SLinus Torvalds void __user * data) 15471da177e4SLinus Torvalds { 15481da177e4SLinus Torvalds int retval; 15491da177e4SLinus Torvalds unsigned long data_page; 15501da177e4SLinus Torvalds unsigned long type_page; 15511da177e4SLinus Torvalds unsigned long dev_page; 15521da177e4SLinus Torvalds char *dir_page; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 15551da177e4SLinus Torvalds if (retval < 0) 15561da177e4SLinus Torvalds return retval; 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds dir_page = getname(dir_name); 15591da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 15601da177e4SLinus Torvalds if (IS_ERR(dir_page)) 15611da177e4SLinus Torvalds goto out1; 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 15641da177e4SLinus Torvalds if (retval < 0) 15651da177e4SLinus Torvalds goto out2; 15661da177e4SLinus Torvalds 15671da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 15681da177e4SLinus Torvalds if (retval < 0) 15691da177e4SLinus Torvalds goto out3; 15701da177e4SLinus Torvalds 15711da177e4SLinus Torvalds lock_kernel(); 15721da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 15731da177e4SLinus Torvalds flags, (void *)data_page); 15741da177e4SLinus Torvalds unlock_kernel(); 15751da177e4SLinus Torvalds free_page(data_page); 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds out3: 15781da177e4SLinus Torvalds free_page(dev_page); 15791da177e4SLinus Torvalds out2: 15801da177e4SLinus Torvalds putname(dir_page); 15811da177e4SLinus Torvalds out1: 15821da177e4SLinus Torvalds free_page(type_page); 15831da177e4SLinus Torvalds return retval; 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds 15861da177e4SLinus Torvalds /* 15871da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 15881da177e4SLinus Torvalds * It can block. Requires the big lock held. 15891da177e4SLinus Torvalds */ 15901da177e4SLinus Torvalds void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, 15911da177e4SLinus Torvalds struct dentry *dentry) 15921da177e4SLinus Torvalds { 15931da177e4SLinus Torvalds struct dentry *old_root; 15941da177e4SLinus Torvalds struct vfsmount *old_rootmnt; 15951da177e4SLinus Torvalds write_lock(&fs->lock); 15961da177e4SLinus Torvalds old_root = fs->root; 15971da177e4SLinus Torvalds old_rootmnt = fs->rootmnt; 15981da177e4SLinus Torvalds fs->rootmnt = mntget(mnt); 15991da177e4SLinus Torvalds fs->root = dget(dentry); 16001da177e4SLinus Torvalds write_unlock(&fs->lock); 16011da177e4SLinus Torvalds if (old_root) { 16021da177e4SLinus Torvalds dput(old_root); 16031da177e4SLinus Torvalds mntput(old_rootmnt); 16041da177e4SLinus Torvalds } 16051da177e4SLinus Torvalds } 16061da177e4SLinus Torvalds 16071da177e4SLinus Torvalds /* 16081da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 16091da177e4SLinus Torvalds * It can block. Requires the big lock held. 16101da177e4SLinus Torvalds */ 16111da177e4SLinus Torvalds void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, 16121da177e4SLinus Torvalds struct dentry *dentry) 16131da177e4SLinus Torvalds { 16141da177e4SLinus Torvalds struct dentry *old_pwd; 16151da177e4SLinus Torvalds struct vfsmount *old_pwdmnt; 16161da177e4SLinus Torvalds 16171da177e4SLinus Torvalds write_lock(&fs->lock); 16181da177e4SLinus Torvalds old_pwd = fs->pwd; 16191da177e4SLinus Torvalds old_pwdmnt = fs->pwdmnt; 16201da177e4SLinus Torvalds fs->pwdmnt = mntget(mnt); 16211da177e4SLinus Torvalds fs->pwd = dget(dentry); 16221da177e4SLinus Torvalds write_unlock(&fs->lock); 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds if (old_pwd) { 16251da177e4SLinus Torvalds dput(old_pwd); 16261da177e4SLinus Torvalds mntput(old_pwdmnt); 16271da177e4SLinus Torvalds } 16281da177e4SLinus Torvalds } 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) 16311da177e4SLinus Torvalds { 16321da177e4SLinus Torvalds struct task_struct *g, *p; 16331da177e4SLinus Torvalds struct fs_struct *fs; 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds read_lock(&tasklist_lock); 16361da177e4SLinus Torvalds do_each_thread(g, p) { 16371da177e4SLinus Torvalds task_lock(p); 16381da177e4SLinus Torvalds fs = p->fs; 16391da177e4SLinus Torvalds if (fs) { 16401da177e4SLinus Torvalds atomic_inc(&fs->count); 16411da177e4SLinus Torvalds task_unlock(p); 1642b58fed8bSRam Pai if (fs->root == old_nd->dentry 1643b58fed8bSRam Pai && fs->rootmnt == old_nd->mnt) 16441da177e4SLinus Torvalds set_fs_root(fs, new_nd->mnt, new_nd->dentry); 1645b58fed8bSRam Pai if (fs->pwd == old_nd->dentry 1646b58fed8bSRam Pai && fs->pwdmnt == old_nd->mnt) 16471da177e4SLinus Torvalds set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); 16481da177e4SLinus Torvalds put_fs_struct(fs); 16491da177e4SLinus Torvalds } else 16501da177e4SLinus Torvalds task_unlock(p); 16511da177e4SLinus Torvalds } while_each_thread(g, p); 16521da177e4SLinus Torvalds read_unlock(&tasklist_lock); 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /* 16561da177e4SLinus Torvalds * pivot_root Semantics: 16571da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 16581da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 16591da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 16601da177e4SLinus Torvalds * 16611da177e4SLinus Torvalds * Restrictions: 16621da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 16631da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 16641da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 16651da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 16661da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 16671da177e4SLinus Torvalds * 16684a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 16694a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 16704a0d11faSNeil Brown * in this situation. 16714a0d11faSNeil Brown * 16721da177e4SLinus Torvalds * Notes: 16731da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 16741da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 16751da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 16761da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 16771da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 16781da177e4SLinus Torvalds * first. 16791da177e4SLinus Torvalds */ 1680b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 1681b58fed8bSRam Pai const char __user * put_old) 16821da177e4SLinus Torvalds { 16831da177e4SLinus Torvalds struct vfsmount *tmp; 16841da177e4SLinus Torvalds struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; 16851da177e4SLinus Torvalds int error; 16861da177e4SLinus Torvalds 16871da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16881da177e4SLinus Torvalds return -EPERM; 16891da177e4SLinus Torvalds 16901da177e4SLinus Torvalds lock_kernel(); 16911da177e4SLinus Torvalds 1692b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1693b58fed8bSRam Pai &new_nd); 16941da177e4SLinus Torvalds if (error) 16951da177e4SLinus Torvalds goto out0; 16961da177e4SLinus Torvalds error = -EINVAL; 16971da177e4SLinus Torvalds if (!check_mnt(new_nd.mnt)) 16981da177e4SLinus Torvalds goto out1; 16991da177e4SLinus Torvalds 17001da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 17011da177e4SLinus Torvalds if (error) 17021da177e4SLinus Torvalds goto out1; 17031da177e4SLinus Torvalds 17041da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 17051da177e4SLinus Torvalds if (error) { 17061da177e4SLinus Torvalds path_release(&old_nd); 17071da177e4SLinus Torvalds goto out1; 17081da177e4SLinus Torvalds } 17091da177e4SLinus Torvalds 17101da177e4SLinus Torvalds read_lock(¤t->fs->lock); 17111da177e4SLinus Torvalds user_nd.mnt = mntget(current->fs->rootmnt); 17121da177e4SLinus Torvalds user_nd.dentry = dget(current->fs->root); 17131da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 1714390c6843SRam Pai down_write(&namespace_sem); 17151b1dcc1bSJes Sorensen mutex_lock(&old_nd.dentry->d_inode->i_mutex); 17161da177e4SLinus Torvalds error = -EINVAL; 171721444403SRam Pai if (IS_MNT_SHARED(old_nd.mnt) || 171821444403SRam Pai IS_MNT_SHARED(new_nd.mnt->mnt_parent) || 171921444403SRam Pai IS_MNT_SHARED(user_nd.mnt->mnt_parent)) 172021444403SRam Pai goto out2; 17211da177e4SLinus Torvalds if (!check_mnt(user_nd.mnt)) 17221da177e4SLinus Torvalds goto out2; 17231da177e4SLinus Torvalds error = -ENOENT; 17241da177e4SLinus Torvalds if (IS_DEADDIR(new_nd.dentry->d_inode)) 17251da177e4SLinus Torvalds goto out2; 17261da177e4SLinus Torvalds if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) 17271da177e4SLinus Torvalds goto out2; 17281da177e4SLinus Torvalds if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) 17291da177e4SLinus Torvalds goto out2; 17301da177e4SLinus Torvalds error = -EBUSY; 17311da177e4SLinus Torvalds if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) 17321da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 17331da177e4SLinus Torvalds error = -EINVAL; 17341da177e4SLinus Torvalds if (user_nd.mnt->mnt_root != user_nd.dentry) 17351da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 17360bb6fcc1SMiklos Szeredi if (user_nd.mnt->mnt_parent == user_nd.mnt) 17370bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 17381da177e4SLinus Torvalds if (new_nd.mnt->mnt_root != new_nd.dentry) 17391da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 17400bb6fcc1SMiklos Szeredi if (new_nd.mnt->mnt_parent == new_nd.mnt) 17410bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 17421da177e4SLinus Torvalds tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ 17431da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17441da177e4SLinus Torvalds if (tmp != new_nd.mnt) { 17451da177e4SLinus Torvalds for (;;) { 17461da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 17471da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 17481da177e4SLinus Torvalds if (tmp->mnt_parent == new_nd.mnt) 17491da177e4SLinus Torvalds break; 17501da177e4SLinus Torvalds tmp = tmp->mnt_parent; 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) 17531da177e4SLinus Torvalds goto out3; 17541da177e4SLinus Torvalds } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) 17551da177e4SLinus Torvalds goto out3; 17561da177e4SLinus Torvalds detach_mnt(new_nd.mnt, &parent_nd); 17571da177e4SLinus Torvalds detach_mnt(user_nd.mnt, &root_parent); 17581da177e4SLinus Torvalds attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ 17591da177e4SLinus Torvalds attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ 17606b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 17611da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 17621da177e4SLinus Torvalds chroot_fs_refs(&user_nd, &new_nd); 17631da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 17641da177e4SLinus Torvalds error = 0; 17651da177e4SLinus Torvalds path_release(&root_parent); 17661da177e4SLinus Torvalds path_release(&parent_nd); 17671da177e4SLinus Torvalds out2: 17681b1dcc1bSJes Sorensen mutex_unlock(&old_nd.dentry->d_inode->i_mutex); 1769390c6843SRam Pai up_write(&namespace_sem); 17701da177e4SLinus Torvalds path_release(&user_nd); 17711da177e4SLinus Torvalds path_release(&old_nd); 17721da177e4SLinus Torvalds out1: 17731da177e4SLinus Torvalds path_release(&new_nd); 17741da177e4SLinus Torvalds out0: 17751da177e4SLinus Torvalds unlock_kernel(); 17761da177e4SLinus Torvalds return error; 17771da177e4SLinus Torvalds out3: 17781da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 17791da177e4SLinus Torvalds goto out2; 17801da177e4SLinus Torvalds } 17811da177e4SLinus Torvalds 17821da177e4SLinus Torvalds static void __init init_mount_tree(void) 17831da177e4SLinus Torvalds { 17841da177e4SLinus Torvalds struct vfsmount *mnt; 17856b3286edSKirill Korotaev struct mnt_namespace *ns; 17861da177e4SLinus Torvalds 17871da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 17881da177e4SLinus Torvalds if (IS_ERR(mnt)) 17891da177e4SLinus Torvalds panic("Can't create rootfs"); 17906b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 17916b3286edSKirill Korotaev if (!ns) 17921da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 17936b3286edSKirill Korotaev atomic_set(&ns->count, 1); 17946b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 17956b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 17966b3286edSKirill Korotaev ns->event = 0; 17976b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 17986b3286edSKirill Korotaev ns->root = mnt; 17996b3286edSKirill Korotaev mnt->mnt_ns = ns; 18001da177e4SLinus Torvalds 18016b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 18026b3286edSKirill Korotaev get_mnt_ns(ns); 18031da177e4SLinus Torvalds 18046b3286edSKirill Korotaev set_fs_pwd(current->fs, ns->root, ns->root->mnt_root); 18056b3286edSKirill Korotaev set_fs_root(current->fs, ns->root, ns->root->mnt_root); 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds 18081da177e4SLinus Torvalds void __init mnt_init(unsigned long mempages) 18091da177e4SLinus Torvalds { 18101da177e4SLinus Torvalds struct list_head *d; 18111da177e4SLinus Torvalds unsigned int nr_hash; 18121da177e4SLinus Torvalds int i; 181315a67dd8SRandy Dunlap int err; 18141da177e4SLinus Torvalds 1815390c6843SRam Pai init_rwsem(&namespace_sem); 1816390c6843SRam Pai 18171da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 18181da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); 18191da177e4SLinus Torvalds 1820b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds if (!mount_hashtable) 18231da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 18241da177e4SLinus Torvalds 18251da177e4SLinus Torvalds /* 18261da177e4SLinus Torvalds * Find the power-of-two list-heads that can fit into the allocation.. 18271da177e4SLinus Torvalds * We don't guarantee that "sizeof(struct list_head)" is necessarily 18281da177e4SLinus Torvalds * a power-of-two. 18291da177e4SLinus Torvalds */ 18301da177e4SLinus Torvalds nr_hash = PAGE_SIZE / sizeof(struct list_head); 18311da177e4SLinus Torvalds hash_bits = 0; 18321da177e4SLinus Torvalds do { 18331da177e4SLinus Torvalds hash_bits++; 18341da177e4SLinus Torvalds } while ((nr_hash >> hash_bits) != 0); 18351da177e4SLinus Torvalds hash_bits--; 18361da177e4SLinus Torvalds 18371da177e4SLinus Torvalds /* 18381da177e4SLinus Torvalds * Re-calculate the actual number of entries and the mask 18391da177e4SLinus Torvalds * from the number of bits we can fit. 18401da177e4SLinus Torvalds */ 18411da177e4SLinus Torvalds nr_hash = 1UL << hash_bits; 18421da177e4SLinus Torvalds hash_mask = nr_hash - 1; 18431da177e4SLinus Torvalds 18441da177e4SLinus Torvalds printk("Mount-cache hash table entries: %d\n", nr_hash); 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds /* And initialize the newly allocated array */ 18471da177e4SLinus Torvalds d = mount_hashtable; 18481da177e4SLinus Torvalds i = nr_hash; 18491da177e4SLinus Torvalds do { 18501da177e4SLinus Torvalds INIT_LIST_HEAD(d); 18511da177e4SLinus Torvalds d++; 18521da177e4SLinus Torvalds i--; 18531da177e4SLinus Torvalds } while (i); 185415a67dd8SRandy Dunlap err = sysfs_init(); 185515a67dd8SRandy Dunlap if (err) 185615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 185715a67dd8SRandy Dunlap __FUNCTION__, err); 185815a67dd8SRandy Dunlap err = subsystem_register(&fs_subsys); 185915a67dd8SRandy Dunlap if (err) 186015a67dd8SRandy Dunlap printk(KERN_WARNING "%s: subsystem_register error: %d\n", 186115a67dd8SRandy Dunlap __FUNCTION__, err); 18621da177e4SLinus Torvalds init_rootfs(); 18631da177e4SLinus Torvalds init_mount_tree(); 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds 18666b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 18671da177e4SLinus Torvalds { 18686b3286edSKirill Korotaev struct vfsmount *root = ns->root; 186970fbcdf4SRam Pai LIST_HEAD(umount_list); 18706b3286edSKirill Korotaev ns->root = NULL; 18711ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 1872390c6843SRam Pai down_write(&namespace_sem); 18731da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1874a05964f3SRam Pai umount_tree(root, 0, &umount_list); 18751da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1876390c6843SRam Pai up_write(&namespace_sem); 187770fbcdf4SRam Pai release_mounts(&umount_list); 18786b3286edSKirill Korotaev kfree(ns); 18791da177e4SLinus Torvalds } 1880