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> 2813f14b4dSEric Dumazet #include <linux/log2.h> 291da177e4SLinus Torvalds #include <asm/uaccess.h> 301da177e4SLinus Torvalds #include <asm/unistd.h> 3107b20889SRam Pai #include "pnode.h" 32948730b0SAdrian Bunk #include "internal.h" 331da177e4SLinus Torvalds 3413f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3513f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3613f14b4dSEric Dumazet 371da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 381da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 391da177e4SLinus Torvalds 405addc5ddSAl Viro static int event; 415addc5ddSAl Viro 42fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 43e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 44390c6843SRam Pai static struct rw_semaphore namespace_sem; 451da177e4SLinus Torvalds 46f87fd4c2SMiklos Szeredi /* /sys/fs */ 4700d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4800d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 49f87fd4c2SMiklos Szeredi 501da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 511da177e4SLinus Torvalds { 521da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 531da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5413f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 5513f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 561da177e4SLinus Torvalds } 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 591da177e4SLinus Torvalds { 60c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 611da177e4SLinus Torvalds if (mnt) { 621da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 631da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 641da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 651da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 661da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 6755e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 6803e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 69a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 70a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 711da177e4SLinus Torvalds if (name) { 721da177e4SLinus Torvalds int size = strlen(name) + 1; 731da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 741da177e4SLinus Torvalds if (newname) { 751da177e4SLinus Torvalds memcpy(newname, name, size); 761da177e4SLinus Torvalds mnt->mnt_devname = newname; 771da177e4SLinus Torvalds } 781da177e4SLinus Torvalds } 791da177e4SLinus Torvalds } 801da177e4SLinus Torvalds return mnt; 811da177e4SLinus Torvalds } 821da177e4SLinus Torvalds 838366025eSDave Hansen /* 848366025eSDave Hansen * Most r/o checks on a fs are for operations that take 858366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 868366025eSDave Hansen * We must keep track of when those operations start 878366025eSDave Hansen * (for permission checks) and when they end, so that 888366025eSDave Hansen * we can determine when writes are able to occur to 898366025eSDave Hansen * a filesystem. 908366025eSDave Hansen */ 918366025eSDave Hansen /** 928366025eSDave Hansen * mnt_want_write - get write access to a mount 938366025eSDave Hansen * @mnt: the mount on which to take a write 948366025eSDave Hansen * 958366025eSDave Hansen * This tells the low-level filesystem that a write is 968366025eSDave Hansen * about to be performed to it, and makes sure that 978366025eSDave Hansen * writes are allowed before returning success. When 988366025eSDave Hansen * the write operation is finished, mnt_drop_write() 998366025eSDave Hansen * must be called. This is effectively a refcount. 1008366025eSDave Hansen */ 1018366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 1028366025eSDave Hansen { 1038366025eSDave Hansen if (__mnt_is_readonly(mnt)) 1048366025eSDave Hansen return -EROFS; 1058366025eSDave Hansen return 0; 1068366025eSDave Hansen } 1078366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 1088366025eSDave Hansen 1098366025eSDave Hansen /** 1108366025eSDave Hansen * mnt_drop_write - give up write access to a mount 1118366025eSDave Hansen * @mnt: the mount on which to give up write access 1128366025eSDave Hansen * 1138366025eSDave Hansen * Tells the low-level filesystem that we are done 1148366025eSDave Hansen * performing writes to it. Must be matched with 1158366025eSDave Hansen * mnt_want_write() call above. 1168366025eSDave Hansen */ 1178366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 1188366025eSDave Hansen { 1198366025eSDave Hansen } 1208366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 1218366025eSDave Hansen 1228366025eSDave Hansen /* 1238366025eSDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1248366025eSDave Hansen * @mnt: the mount to check for its write status 1258366025eSDave Hansen * 1268366025eSDave Hansen * This shouldn't be used directly ouside of the VFS. 1278366025eSDave Hansen * It does not guarantee that the filesystem will stay 1288366025eSDave Hansen * r/w, just that it is right *now*. This can not and 1298366025eSDave Hansen * should not be used in place of IS_RDONLY(inode). 1308366025eSDave Hansen */ 1318366025eSDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1328366025eSDave Hansen { 1338366025eSDave Hansen return (mnt->mnt_sb->s_flags & MS_RDONLY); 1348366025eSDave Hansen } 1358366025eSDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 1368366025eSDave Hansen 137454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 138454e2398SDavid Howells { 139454e2398SDavid Howells mnt->mnt_sb = sb; 140454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 141454e2398SDavid Howells return 0; 142454e2398SDavid Howells } 143454e2398SDavid Howells 144454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 145454e2398SDavid Howells 1461da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 1471da177e4SLinus Torvalds { 1481da177e4SLinus Torvalds kfree(mnt->mnt_devname); 1491da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 1501da177e4SLinus Torvalds } 1511da177e4SLinus Torvalds 1521da177e4SLinus Torvalds /* 153a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 154a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 1551da177e4SLinus Torvalds */ 156a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 157a05964f3SRam Pai int dir) 1581da177e4SLinus Torvalds { 1591da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 1601da177e4SLinus Torvalds struct list_head *tmp = head; 1611da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds for (;;) { 164a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 1651da177e4SLinus Torvalds p = NULL; 1661da177e4SLinus Torvalds if (tmp == head) 1671da177e4SLinus Torvalds break; 1681da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 1691da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 170a05964f3SRam Pai found = p; 1711da177e4SLinus Torvalds break; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds } 1741da177e4SLinus Torvalds return found; 1751da177e4SLinus Torvalds } 1761da177e4SLinus Torvalds 177a05964f3SRam Pai /* 178a05964f3SRam Pai * lookup_mnt increments the ref count before returning 179a05964f3SRam Pai * the vfsmount struct. 180a05964f3SRam Pai */ 181a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 182a05964f3SRam Pai { 183a05964f3SRam Pai struct vfsmount *child_mnt; 184a05964f3SRam Pai spin_lock(&vfsmount_lock); 185a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 186a05964f3SRam Pai mntget(child_mnt); 187a05964f3SRam Pai spin_unlock(&vfsmount_lock); 188a05964f3SRam Pai return child_mnt; 189a05964f3SRam Pai } 190a05964f3SRam Pai 1911da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 1921da177e4SLinus Torvalds { 1936b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1966b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 1975addc5ddSAl Viro { 1985addc5ddSAl Viro if (ns) { 1995addc5ddSAl Viro ns->event = ++event; 2005addc5ddSAl Viro wake_up_interruptible(&ns->poll); 2015addc5ddSAl Viro } 2025addc5ddSAl Viro } 2035addc5ddSAl Viro 2046b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 2055addc5ddSAl Viro { 2065addc5ddSAl Viro if (ns && ns->event != event) { 2075addc5ddSAl Viro ns->event = event; 2085addc5ddSAl Viro wake_up_interruptible(&ns->poll); 2095addc5ddSAl Viro } 2105addc5ddSAl Viro } 2115addc5ddSAl Viro 2121a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 2131da177e4SLinus Torvalds { 2141a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 2151a390689SAl Viro old_path->mnt = mnt->mnt_parent; 2161da177e4SLinus Torvalds mnt->mnt_parent = mnt; 2171da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 2181da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 2191da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 2201a390689SAl Viro old_path->dentry->d_mounted--; 2211da177e4SLinus Torvalds } 2221da177e4SLinus Torvalds 223b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 224b90fa9aeSRam Pai struct vfsmount *child_mnt) 225b90fa9aeSRam Pai { 226b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 227b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 228b90fa9aeSRam Pai dentry->d_mounted++; 229b90fa9aeSRam Pai } 230b90fa9aeSRam Pai 2311a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 2321da177e4SLinus Torvalds { 2331a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 234b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 2351a390689SAl Viro hash(path->mnt, path->dentry)); 2361a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 237b90fa9aeSRam Pai } 238b90fa9aeSRam Pai 239b90fa9aeSRam Pai /* 240b90fa9aeSRam Pai * the caller must hold vfsmount_lock 241b90fa9aeSRam Pai */ 242b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 243b90fa9aeSRam Pai { 244b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 245b90fa9aeSRam Pai struct vfsmount *m; 246b90fa9aeSRam Pai LIST_HEAD(head); 2476b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 248b90fa9aeSRam Pai 249b90fa9aeSRam Pai BUG_ON(parent == mnt); 250b90fa9aeSRam Pai 251b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 252b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 2536b3286edSKirill Korotaev m->mnt_ns = n; 254b90fa9aeSRam Pai list_splice(&head, n->list.prev); 255b90fa9aeSRam Pai 256b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 257b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 258b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 2596b3286edSKirill Korotaev touch_mnt_namespace(n); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 2631da177e4SLinus Torvalds { 2641da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 2651da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 2661da177e4SLinus Torvalds while (1) { 2671da177e4SLinus Torvalds if (p == root) 2681da177e4SLinus Torvalds return NULL; 2691da177e4SLinus Torvalds next = p->mnt_child.next; 2701da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 2711da177e4SLinus Torvalds break; 2721da177e4SLinus Torvalds p = p->mnt_parent; 2731da177e4SLinus Torvalds } 2741da177e4SLinus Torvalds } 2751da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 2761da177e4SLinus Torvalds } 2771da177e4SLinus Torvalds 2789676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 2799676f0c6SRam Pai { 2809676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 2819676f0c6SRam Pai while (prev != &p->mnt_mounts) { 2829676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 2839676f0c6SRam Pai prev = p->mnt_mounts.prev; 2849676f0c6SRam Pai } 2859676f0c6SRam Pai return p; 2869676f0c6SRam Pai } 2879676f0c6SRam Pai 28836341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 28936341f64SRam Pai int flag) 2901da177e4SLinus Torvalds { 2911da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 2921da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds if (mnt) { 2951da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 2961da177e4SLinus Torvalds atomic_inc(&sb->s_active); 2971da177e4SLinus Torvalds mnt->mnt_sb = sb; 2981da177e4SLinus Torvalds mnt->mnt_root = dget(root); 2991da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 3001da177e4SLinus Torvalds mnt->mnt_parent = mnt; 301b90fa9aeSRam Pai 3025afe0022SRam Pai if (flag & CL_SLAVE) { 3035afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 3045afe0022SRam Pai mnt->mnt_master = old; 3055afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 3068aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 307b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 308b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 3095afe0022SRam Pai if (IS_MNT_SLAVE(old)) 3105afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 3115afe0022SRam Pai mnt->mnt_master = old->mnt_master; 3125afe0022SRam Pai } 313b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 314b90fa9aeSRam Pai set_mnt_shared(mnt); 3151da177e4SLinus Torvalds 3161da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 3171da177e4SLinus Torvalds * as the original if that was on one */ 31836341f64SRam Pai if (flag & CL_EXPIRE) { 31955e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 32055e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 3211da177e4SLinus Torvalds } 32236341f64SRam Pai } 3231da177e4SLinus Torvalds return mnt; 3241da177e4SLinus Torvalds } 3251da177e4SLinus Torvalds 3267b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 3271da177e4SLinus Torvalds { 3281da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 3291da177e4SLinus Torvalds dput(mnt->mnt_root); 3301da177e4SLinus Torvalds free_vfsmnt(mnt); 3311da177e4SLinus Torvalds deactivate_super(sb); 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 3347b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 3357b7b1aceSAl Viro { 3367b7b1aceSAl Viro repeat: 3377b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 3387b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 3397b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3407b7b1aceSAl Viro __mntput(mnt); 3417b7b1aceSAl Viro return; 3427b7b1aceSAl Viro } 3437b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 3447b7b1aceSAl Viro mnt->mnt_pinned = 0; 3457b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3467b7b1aceSAl Viro acct_auto_close_mnt(mnt); 3477b7b1aceSAl Viro security_sb_umount_close(mnt); 3487b7b1aceSAl Viro goto repeat; 3497b7b1aceSAl Viro } 3507b7b1aceSAl Viro } 3517b7b1aceSAl Viro 3527b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 3537b7b1aceSAl Viro 3547b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 3557b7b1aceSAl Viro { 3567b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3577b7b1aceSAl Viro mnt->mnt_pinned++; 3587b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3597b7b1aceSAl Viro } 3607b7b1aceSAl Viro 3617b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 3627b7b1aceSAl Viro 3637b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 3647b7b1aceSAl Viro { 3657b7b1aceSAl Viro spin_lock(&vfsmount_lock); 3667b7b1aceSAl Viro if (mnt->mnt_pinned) { 3677b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 3687b7b1aceSAl Viro mnt->mnt_pinned--; 3697b7b1aceSAl Viro } 3707b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 3717b7b1aceSAl Viro } 3727b7b1aceSAl Viro 3737b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 3741da177e4SLinus Torvalds 375b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 376b3b304a2SMiklos Szeredi { 377b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 378b3b304a2SMiklos Szeredi } 379b3b304a2SMiklos Szeredi 380b3b304a2SMiklos Szeredi /* 381b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 382b3b304a2SMiklos Szeredi * implement more complex mount option showing. 383b3b304a2SMiklos Szeredi * 384b3b304a2SMiklos Szeredi * See also save_mount_options(). 385b3b304a2SMiklos Szeredi */ 386b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 387b3b304a2SMiklos Szeredi { 388b3b304a2SMiklos Szeredi const char *options = mnt->mnt_sb->s_options; 389b3b304a2SMiklos Szeredi 390b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 391b3b304a2SMiklos Szeredi seq_putc(m, ','); 392b3b304a2SMiklos Szeredi mangle(m, options); 393b3b304a2SMiklos Szeredi } 394b3b304a2SMiklos Szeredi 395b3b304a2SMiklos Szeredi return 0; 396b3b304a2SMiklos Szeredi } 397b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 398b3b304a2SMiklos Szeredi 399b3b304a2SMiklos Szeredi /* 400b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 401b3b304a2SMiklos Szeredi * called from the fill_super() callback. 402b3b304a2SMiklos Szeredi * 403b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 404b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 405b3b304a2SMiklos Szeredi * remount fails. 406b3b304a2SMiklos Szeredi * 407b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 408b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 409b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 410b3b304a2SMiklos Szeredi * any more. 411b3b304a2SMiklos Szeredi */ 412b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 413b3b304a2SMiklos Szeredi { 414b3b304a2SMiklos Szeredi kfree(sb->s_options); 415b3b304a2SMiklos Szeredi sb->s_options = kstrdup(options, GFP_KERNEL); 416b3b304a2SMiklos Szeredi } 417b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 418b3b304a2SMiklos Szeredi 4191da177e4SLinus Torvalds /* iterator */ 4201da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 4211da177e4SLinus Torvalds { 4226b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 4231da177e4SLinus Torvalds 424390c6843SRam Pai down_read(&namespace_sem); 425b0765fb8SPavel Emelianov return seq_list_start(&n->list, *pos); 4261da177e4SLinus Torvalds } 4271da177e4SLinus Torvalds 4281da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 4291da177e4SLinus Torvalds { 4306b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 431b0765fb8SPavel Emelianov 432b0765fb8SPavel Emelianov return seq_list_next(v, &n->list, pos); 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 4361da177e4SLinus Torvalds { 437390c6843SRam Pai up_read(&namespace_sem); 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 4401da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 4411da177e4SLinus Torvalds { 442b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 4431da177e4SLinus Torvalds int err = 0; 4441da177e4SLinus Torvalds static struct proc_fs_info { 4451da177e4SLinus Torvalds int flag; 4461da177e4SLinus Torvalds char *str; 4471da177e4SLinus Torvalds } fs_info[] = { 4481da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 4491da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 4501da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 4511da177e4SLinus Torvalds { 0, NULL } 4521da177e4SLinus Torvalds }; 4531da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 4541da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 4551da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 4561da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 457fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 458fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 45947ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 4601da177e4SLinus Torvalds { 0, NULL } 4611da177e4SLinus Torvalds }; 4621da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 463c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 4641da177e4SLinus Torvalds 4651da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 4661da177e4SLinus Torvalds seq_putc(m, ' '); 467c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 4681da177e4SLinus Torvalds seq_putc(m, ' '); 4691da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 47079c0b2dfSMiklos Szeredi if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { 47179c0b2dfSMiklos Szeredi seq_putc(m, '.'); 47279c0b2dfSMiklos Szeredi mangle(m, mnt->mnt_sb->s_subtype); 47379c0b2dfSMiklos Szeredi } 4741da177e4SLinus Torvalds seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 4751da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 4761da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 4771da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 4781da177e4SLinus Torvalds } 4791da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 4801da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 4811da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 4841da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 4851da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 4861da177e4SLinus Torvalds return err; 4871da177e4SLinus Torvalds } 4881da177e4SLinus Torvalds 4891da177e4SLinus Torvalds struct seq_operations mounts_op = { 4901da177e4SLinus Torvalds .start = m_start, 4911da177e4SLinus Torvalds .next = m_next, 4921da177e4SLinus Torvalds .stop = m_stop, 4931da177e4SLinus Torvalds .show = show_vfsmnt 4941da177e4SLinus Torvalds }; 4951da177e4SLinus Torvalds 496b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 497b4629fe2SChuck Lever { 498b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 499c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 500b4629fe2SChuck Lever int err = 0; 501b4629fe2SChuck Lever 502b4629fe2SChuck Lever /* device */ 503b4629fe2SChuck Lever if (mnt->mnt_devname) { 504b4629fe2SChuck Lever seq_puts(m, "device "); 505b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 506b4629fe2SChuck Lever } else 507b4629fe2SChuck Lever seq_puts(m, "no device"); 508b4629fe2SChuck Lever 509b4629fe2SChuck Lever /* mount point */ 510b4629fe2SChuck Lever seq_puts(m, " mounted on "); 511c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 512b4629fe2SChuck Lever seq_putc(m, ' '); 513b4629fe2SChuck Lever 514b4629fe2SChuck Lever /* file system type */ 515b4629fe2SChuck Lever seq_puts(m, "with fstype "); 516b4629fe2SChuck Lever mangle(m, mnt->mnt_sb->s_type->name); 517b4629fe2SChuck Lever 518b4629fe2SChuck Lever /* optional statistics */ 519b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 520b4629fe2SChuck Lever seq_putc(m, ' '); 521b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 522b4629fe2SChuck Lever } 523b4629fe2SChuck Lever 524b4629fe2SChuck Lever seq_putc(m, '\n'); 525b4629fe2SChuck Lever return err; 526b4629fe2SChuck Lever } 527b4629fe2SChuck Lever 528b4629fe2SChuck Lever struct seq_operations mountstats_op = { 529b4629fe2SChuck Lever .start = m_start, 530b4629fe2SChuck Lever .next = m_next, 531b4629fe2SChuck Lever .stop = m_stop, 532b4629fe2SChuck Lever .show = show_vfsstat, 533b4629fe2SChuck Lever }; 534b4629fe2SChuck Lever 5351da177e4SLinus Torvalds /** 5361da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 5371da177e4SLinus Torvalds * @mnt: root of mount tree 5381da177e4SLinus Torvalds * 5391da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 5401da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 5411da177e4SLinus Torvalds * busy. 5421da177e4SLinus Torvalds */ 5431da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 5441da177e4SLinus Torvalds { 54536341f64SRam Pai int actual_refs = 0; 54636341f64SRam Pai int minimum_refs = 0; 54736341f64SRam Pai struct vfsmount *p; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 55036341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 5511da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 5521da177e4SLinus Torvalds minimum_refs += 2; 5531da177e4SLinus Torvalds } 5541da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 5551da177e4SLinus Torvalds 5561da177e4SLinus Torvalds if (actual_refs > minimum_refs) 5571da177e4SLinus Torvalds return 0; 558e3474a8eSIan Kent 559e3474a8eSIan Kent return 1; 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds /** 5651da177e4SLinus Torvalds * may_umount - check if a mount point is busy 5661da177e4SLinus Torvalds * @mnt: root of mount 5671da177e4SLinus Torvalds * 5681da177e4SLinus Torvalds * This is called to check if a mount point has any 5691da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 5701da177e4SLinus Torvalds * mount has sub mounts this will return busy 5711da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 5721da177e4SLinus Torvalds * 5731da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 5741da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 5751da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 5761da177e4SLinus Torvalds */ 5771da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 5781da177e4SLinus Torvalds { 579e3474a8eSIan Kent int ret = 1; 580a05964f3SRam Pai spin_lock(&vfsmount_lock); 581a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 582e3474a8eSIan Kent ret = 0; 583a05964f3SRam Pai spin_unlock(&vfsmount_lock); 584a05964f3SRam Pai return ret; 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 5871da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 5881da177e4SLinus Torvalds 589b90fa9aeSRam Pai void release_mounts(struct list_head *head) 5901da177e4SLinus Torvalds { 59170fbcdf4SRam Pai struct vfsmount *mnt; 59270fbcdf4SRam Pai while (!list_empty(head)) { 593b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 59470fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 59570fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 59670fbcdf4SRam Pai struct dentry *dentry; 59770fbcdf4SRam Pai struct vfsmount *m; 59870fbcdf4SRam Pai spin_lock(&vfsmount_lock); 59970fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 60070fbcdf4SRam Pai m = mnt->mnt_parent; 60170fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 60270fbcdf4SRam Pai mnt->mnt_parent = mnt; 6037c4b93d8SAl Viro m->mnt_ghosts--; 6041da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 60570fbcdf4SRam Pai dput(dentry); 60670fbcdf4SRam Pai mntput(m); 6071da177e4SLinus Torvalds } 6081da177e4SLinus Torvalds mntput(mnt); 60970fbcdf4SRam Pai } 61070fbcdf4SRam Pai } 61170fbcdf4SRam Pai 612a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 61370fbcdf4SRam Pai { 61470fbcdf4SRam Pai struct vfsmount *p; 61570fbcdf4SRam Pai 6161bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 6171bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 61870fbcdf4SRam Pai 619a05964f3SRam Pai if (propagate) 620a05964f3SRam Pai propagate_umount(kill); 621a05964f3SRam Pai 62270fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 62370fbcdf4SRam Pai list_del_init(&p->mnt_expire); 62470fbcdf4SRam Pai list_del_init(&p->mnt_list); 6256b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 6266b3286edSKirill Korotaev p->mnt_ns = NULL; 62770fbcdf4SRam Pai list_del_init(&p->mnt_child); 6287c4b93d8SAl Viro if (p->mnt_parent != p) { 6297c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 630f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 6317c4b93d8SAl Viro } 632a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds } 6351da177e4SLinus Torvalds 636c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 637c35038beSAl Viro 6381da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 6391da177e4SLinus Torvalds { 6401da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 6411da177e4SLinus Torvalds int retval; 64270fbcdf4SRam Pai LIST_HEAD(umount_list); 6431da177e4SLinus Torvalds 6441da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 6451da177e4SLinus Torvalds if (retval) 6461da177e4SLinus Torvalds return retval; 6471da177e4SLinus Torvalds 6481da177e4SLinus Torvalds /* 6491da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 6501da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 6511da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 6521da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 6531da177e4SLinus Torvalds */ 6541da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 6556ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 6561da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 6571da177e4SLinus Torvalds return -EINVAL; 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 6601da177e4SLinus Torvalds return -EBUSY; 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 6631da177e4SLinus Torvalds return -EAGAIN; 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds 6661da177e4SLinus Torvalds /* 6671da177e4SLinus Torvalds * If we may have to abort operations to get out of this 6681da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 6691da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 6701da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 6711da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 6721da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 6731da177e4SLinus Torvalds * about for the moment. 6741da177e4SLinus Torvalds */ 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds lock_kernel(); 6778b512d9aSTrond Myklebust if (sb->s_op->umount_begin) 6788b512d9aSTrond Myklebust sb->s_op->umount_begin(mnt, flags); 6791da177e4SLinus Torvalds unlock_kernel(); 6801da177e4SLinus Torvalds 6811da177e4SLinus Torvalds /* 6821da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 6831da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 6841da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 6851da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 6861da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 6871da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 6881da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 6891da177e4SLinus Torvalds */ 6906ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Special case for "unmounting" root ... 6931da177e4SLinus Torvalds * we just try to remount it readonly. 6941da177e4SLinus Torvalds */ 6951da177e4SLinus Torvalds down_write(&sb->s_umount); 6961da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 6971da177e4SLinus Torvalds lock_kernel(); 6981da177e4SLinus Torvalds DQUOT_OFF(sb); 6991da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 7001da177e4SLinus Torvalds unlock_kernel(); 7011da177e4SLinus Torvalds } 7021da177e4SLinus Torvalds up_write(&sb->s_umount); 7031da177e4SLinus Torvalds return retval; 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds 706390c6843SRam Pai down_write(&namespace_sem); 7071da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 7085addc5ddSAl Viro event++; 7091da177e4SLinus Torvalds 710c35038beSAl Viro if (!(flags & MNT_DETACH)) 711c35038beSAl Viro shrink_submounts(mnt, &umount_list); 712c35038beSAl Viro 7131da177e4SLinus Torvalds retval = -EBUSY; 714a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 7151da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 716a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 7171da177e4SLinus Torvalds retval = 0; 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7201da177e4SLinus Torvalds if (retval) 7211da177e4SLinus Torvalds security_sb_umount_busy(mnt); 722390c6843SRam Pai up_write(&namespace_sem); 72370fbcdf4SRam Pai release_mounts(&umount_list); 7241da177e4SLinus Torvalds return retval; 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds /* 7281da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 7291da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 7301da177e4SLinus Torvalds * 7311da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 7321da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 7331da177e4SLinus Torvalds */ 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 7361da177e4SLinus Torvalds { 7371da177e4SLinus Torvalds struct nameidata nd; 7381da177e4SLinus Torvalds int retval; 7391da177e4SLinus Torvalds 7401da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 7411da177e4SLinus Torvalds if (retval) 7421da177e4SLinus Torvalds goto out; 7431da177e4SLinus Torvalds retval = -EINVAL; 7444ac91378SJan Blunck if (nd.path.dentry != nd.path.mnt->mnt_root) 7451da177e4SLinus Torvalds goto dput_and_out; 7464ac91378SJan Blunck if (!check_mnt(nd.path.mnt)) 7471da177e4SLinus Torvalds goto dput_and_out; 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds retval = -EPERM; 7501da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 7511da177e4SLinus Torvalds goto dput_and_out; 7521da177e4SLinus Torvalds 7534ac91378SJan Blunck retval = do_umount(nd.path.mnt, flags); 7541da177e4SLinus Torvalds dput_and_out: 755429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 7564ac91378SJan Blunck dput(nd.path.dentry); 7574ac91378SJan Blunck mntput_no_expire(nd.path.mnt); 7581da177e4SLinus Torvalds out: 7591da177e4SLinus Torvalds return retval; 7601da177e4SLinus Torvalds } 7611da177e4SLinus Torvalds 7621da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 7631da177e4SLinus Torvalds 7641da177e4SLinus Torvalds /* 7651da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 7661da177e4SLinus Torvalds */ 7671da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 7681da177e4SLinus Torvalds { 7691da177e4SLinus Torvalds return sys_umount(name, 0); 7701da177e4SLinus Torvalds } 7711da177e4SLinus Torvalds 7721da177e4SLinus Torvalds #endif 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 7751da177e4SLinus Torvalds { 7761da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 7771da177e4SLinus Torvalds return 0; 7781da177e4SLinus Torvalds return -EPERM; 7791da177e4SLinus Torvalds #ifdef notyet 7804ac91378SJan Blunck if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) 7811da177e4SLinus Torvalds return -EPERM; 7824ac91378SJan Blunck if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { 7834ac91378SJan Blunck if (current->uid != nd->path.dentry->d_inode->i_uid) 7841da177e4SLinus Torvalds return -EPERM; 7851da177e4SLinus Torvalds } 786e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 7871da177e4SLinus Torvalds return -EPERM; 7881da177e4SLinus Torvalds return 0; 7891da177e4SLinus Torvalds #endif 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds 792b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 7931da177e4SLinus Torvalds { 7941da177e4SLinus Torvalds while (1) { 7951da177e4SLinus Torvalds if (d == dentry) 7961da177e4SLinus Torvalds return 1; 7971da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 7981da177e4SLinus Torvalds return 0; 7991da177e4SLinus Torvalds d = d->d_parent; 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds } 8021da177e4SLinus Torvalds 803b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 80436341f64SRam Pai int flag) 8051da177e4SLinus Torvalds { 8061da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 8071a390689SAl Viro struct path path; 8081da177e4SLinus Torvalds 8099676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 8109676f0c6SRam Pai return NULL; 8119676f0c6SRam Pai 81236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 8131da177e4SLinus Torvalds if (!q) 8141da177e4SLinus Torvalds goto Enomem; 8151da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 8161da177e4SLinus Torvalds 8171da177e4SLinus Torvalds p = mnt; 818fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 8191da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 8201da177e4SLinus Torvalds continue; 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 8239676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 8249676f0c6SRam Pai s = skip_mnt_tree(s); 8259676f0c6SRam Pai continue; 8269676f0c6SRam Pai } 8271da177e4SLinus Torvalds while (p != s->mnt_parent) { 8281da177e4SLinus Torvalds p = p->mnt_parent; 8291da177e4SLinus Torvalds q = q->mnt_parent; 8301da177e4SLinus Torvalds } 8311da177e4SLinus Torvalds p = s; 8321a390689SAl Viro path.mnt = q; 8331a390689SAl Viro path.dentry = p->mnt_mountpoint; 83436341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 8351da177e4SLinus Torvalds if (!q) 8361da177e4SLinus Torvalds goto Enomem; 8371da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 8381da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 8391a390689SAl Viro attach_mnt(q, &path); 8401da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds } 8431da177e4SLinus Torvalds return res; 8441da177e4SLinus Torvalds Enomem: 8451da177e4SLinus Torvalds if (res) { 84670fbcdf4SRam Pai LIST_HEAD(umount_list); 8471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 848a05964f3SRam Pai umount_tree(res, 0, &umount_list); 8491da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 85070fbcdf4SRam Pai release_mounts(&umount_list); 8511da177e4SLinus Torvalds } 8521da177e4SLinus Torvalds return NULL; 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 8558aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 8568aec0809SAl Viro { 8578aec0809SAl Viro struct vfsmount *tree; 8588aec0809SAl Viro down_read(&namespace_sem); 8598aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 8608aec0809SAl Viro up_read(&namespace_sem); 8618aec0809SAl Viro return tree; 8628aec0809SAl Viro } 8638aec0809SAl Viro 8648aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 8658aec0809SAl Viro { 8668aec0809SAl Viro LIST_HEAD(umount_list); 8678aec0809SAl Viro down_read(&namespace_sem); 8688aec0809SAl Viro spin_lock(&vfsmount_lock); 8698aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 8708aec0809SAl Viro spin_unlock(&vfsmount_lock); 8718aec0809SAl Viro up_read(&namespace_sem); 8728aec0809SAl Viro release_mounts(&umount_list); 8738aec0809SAl Viro } 8748aec0809SAl Viro 875b90fa9aeSRam Pai /* 876b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 877b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 87821444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 87921444403SRam Pai * store the parent mount and mountpoint dentry. 88021444403SRam Pai * (done when source_mnt is moved) 881b90fa9aeSRam Pai * 882b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 883b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 8849676f0c6SRam Pai * --------------------------------------------------------------------------- 885b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 8869676f0c6SRam Pai * |************************************************************************** 8879676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 8889676f0c6SRam Pai * | dest | | | | | 8899676f0c6SRam Pai * | | | | | | | 8909676f0c6SRam Pai * | v | | | | | 8919676f0c6SRam Pai * |************************************************************************** 8929676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 8935afe0022SRam Pai * | | | | | | 8949676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 8959676f0c6SRam Pai * *************************************************************************** 896b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 897b90fa9aeSRam Pai * destination mount. 898b90fa9aeSRam Pai * 899b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 900b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 901b90fa9aeSRam Pai * the peer group of the source mount. 902b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 903b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 904b90fa9aeSRam Pai * mount. 9055afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 9065afe0022SRam Pai * of the destination mount and the cloned mount is made slave 9075afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 9085afe0022SRam Pai * is marked as 'shared and slave'. 9095afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 9105afe0022SRam Pai * source mount. 9115afe0022SRam Pai * 9129676f0c6SRam Pai * --------------------------------------------------------------------------- 91321444403SRam Pai * | MOVE MOUNT OPERATION | 9149676f0c6SRam Pai * |************************************************************************** 9159676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 9169676f0c6SRam Pai * | dest | | | | | 9179676f0c6SRam Pai * | | | | | | | 9189676f0c6SRam Pai * | v | | | | | 9199676f0c6SRam Pai * |************************************************************************** 9209676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 9215afe0022SRam Pai * | | | | | | 9229676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 9239676f0c6SRam Pai * *************************************************************************** 9245afe0022SRam Pai * 9255afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 9265afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 92721444403SRam Pai * (+*) the mount is moved to the destination. 9285afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 9295afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 9305afe0022SRam Pai * the mount is marked as 'shared and slave'. 9315afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 932b90fa9aeSRam Pai * 933b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 934b90fa9aeSRam Pai * applied to each mount in the tree. 935b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 936b90fa9aeSRam Pai * in allocations. 937b90fa9aeSRam Pai */ 938b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 9391a390689SAl Viro struct path *path, struct path *parent_path) 940b90fa9aeSRam Pai { 941b90fa9aeSRam Pai LIST_HEAD(tree_list); 9421a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 9431a390689SAl Viro struct dentry *dest_dentry = path->dentry; 944b90fa9aeSRam Pai struct vfsmount *child, *p; 945b90fa9aeSRam Pai 946b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 947b90fa9aeSRam Pai return -EINVAL; 948b90fa9aeSRam Pai 949b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 950b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 951b90fa9aeSRam Pai set_mnt_shared(p); 952b90fa9aeSRam Pai } 953b90fa9aeSRam Pai 954b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 9551a390689SAl Viro if (parent_path) { 9561a390689SAl Viro detach_mnt(source_mnt, parent_path); 9571a390689SAl Viro attach_mnt(source_mnt, path); 9586b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 95921444403SRam Pai } else { 960b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 961b90fa9aeSRam Pai commit_tree(source_mnt); 96221444403SRam Pai } 963b90fa9aeSRam Pai 964b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 965b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 966b90fa9aeSRam Pai commit_tree(child); 967b90fa9aeSRam Pai } 968b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 969b90fa9aeSRam Pai return 0; 970b90fa9aeSRam Pai } 971b90fa9aeSRam Pai 9721da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 9731da177e4SLinus Torvalds { 9741da177e4SLinus Torvalds int err; 9751da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 9761da177e4SLinus Torvalds return -EINVAL; 9771da177e4SLinus Torvalds 9784ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 9791da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 9801da177e4SLinus Torvalds return -ENOTDIR; 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds err = -ENOENT; 9834ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 9844ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 9851da177e4SLinus Torvalds goto out_unlock; 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 9881da177e4SLinus Torvalds if (err) 9891da177e4SLinus Torvalds goto out_unlock; 9901da177e4SLinus Torvalds 9911da177e4SLinus Torvalds err = -ENOENT; 9924ac91378SJan Blunck if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) 9931a390689SAl Viro err = attach_recursive_mnt(mnt, &nd->path, NULL); 9941da177e4SLinus Torvalds out_unlock: 9954ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 9961da177e4SLinus Torvalds if (!err) 9971da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 9981da177e4SLinus Torvalds return err; 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds /* 100207b20889SRam Pai * recursively change the type of the mountpoint. 10032dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 100407b20889SRam Pai */ 10052dafe1c4SEric Sandeen static noinline int do_change_type(struct nameidata *nd, int flag) 100607b20889SRam Pai { 10074ac91378SJan Blunck struct vfsmount *m, *mnt = nd->path.mnt; 100807b20889SRam Pai int recurse = flag & MS_REC; 100907b20889SRam Pai int type = flag & ~MS_REC; 101007b20889SRam Pai 1011ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1012ee6f9582SMiklos Szeredi return -EPERM; 1013ee6f9582SMiklos Szeredi 10144ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 101507b20889SRam Pai return -EINVAL; 101607b20889SRam Pai 101707b20889SRam Pai down_write(&namespace_sem); 101807b20889SRam Pai spin_lock(&vfsmount_lock); 101907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 102007b20889SRam Pai change_mnt_propagation(m, type); 102107b20889SRam Pai spin_unlock(&vfsmount_lock); 102207b20889SRam Pai up_write(&namespace_sem); 102307b20889SRam Pai return 0; 102407b20889SRam Pai } 102507b20889SRam Pai 102607b20889SRam Pai /* 10271da177e4SLinus Torvalds * do loopback mount. 10282dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 10291da177e4SLinus Torvalds */ 10302dafe1c4SEric Sandeen static noinline int do_loopback(struct nameidata *nd, char *old_name, 10312dafe1c4SEric Sandeen int recurse) 10321da177e4SLinus Torvalds { 10331da177e4SLinus Torvalds struct nameidata old_nd; 10341da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 10351da177e4SLinus Torvalds int err = mount_is_safe(nd); 10361da177e4SLinus Torvalds if (err) 10371da177e4SLinus Torvalds return err; 10381da177e4SLinus Torvalds if (!old_name || !*old_name) 10391da177e4SLinus Torvalds return -EINVAL; 10401da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 10411da177e4SLinus Torvalds if (err) 10421da177e4SLinus Torvalds return err; 10431da177e4SLinus Torvalds 1044390c6843SRam Pai down_write(&namespace_sem); 10451da177e4SLinus Torvalds err = -EINVAL; 10464ac91378SJan Blunck if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) 10479676f0c6SRam Pai goto out; 10489676f0c6SRam Pai 10494ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 1050ccd48bc7SAl Viro goto out; 1051ccd48bc7SAl Viro 10521da177e4SLinus Torvalds err = -ENOMEM; 10531da177e4SLinus Torvalds if (recurse) 10544ac91378SJan Blunck mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); 10551da177e4SLinus Torvalds else 10564ac91378SJan Blunck mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); 10571da177e4SLinus Torvalds 1058ccd48bc7SAl Viro if (!mnt) 1059ccd48bc7SAl Viro goto out; 1060ccd48bc7SAl Viro 10611da177e4SLinus Torvalds err = graft_tree(mnt, nd); 10621da177e4SLinus Torvalds if (err) { 106370fbcdf4SRam Pai LIST_HEAD(umount_list); 10641da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1065a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 10661da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 106770fbcdf4SRam Pai release_mounts(&umount_list); 10685b83d2c5SRam Pai } 10691da177e4SLinus Torvalds 1070ccd48bc7SAl Viro out: 1071390c6843SRam Pai up_write(&namespace_sem); 10721d957f9bSJan Blunck path_put(&old_nd.path); 10731da177e4SLinus Torvalds return err; 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* 10771da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 10781da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 10791da177e4SLinus Torvalds * on it - tough luck. 10802dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 10811da177e4SLinus Torvalds */ 10822dafe1c4SEric Sandeen static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, 10831da177e4SLinus Torvalds void *data) 10841da177e4SLinus Torvalds { 10851da177e4SLinus Torvalds int err; 10864ac91378SJan Blunck struct super_block *sb = nd->path.mnt->mnt_sb; 10871da177e4SLinus Torvalds 10881da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10891da177e4SLinus Torvalds return -EPERM; 10901da177e4SLinus Torvalds 10914ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 10921da177e4SLinus Torvalds return -EINVAL; 10931da177e4SLinus Torvalds 10944ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 10951da177e4SLinus Torvalds return -EINVAL; 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds down_write(&sb->s_umount); 10981da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 10991da177e4SLinus Torvalds if (!err) 11004ac91378SJan Blunck nd->path.mnt->mnt_flags = mnt_flags; 11011da177e4SLinus Torvalds up_write(&sb->s_umount); 11021da177e4SLinus Torvalds if (!err) 11034ac91378SJan Blunck security_sb_post_remount(nd->path.mnt, flags, data); 11041da177e4SLinus Torvalds return err; 11051da177e4SLinus Torvalds } 11061da177e4SLinus Torvalds 11079676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 11089676f0c6SRam Pai { 11099676f0c6SRam Pai struct vfsmount *p; 11109676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 11119676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 11129676f0c6SRam Pai return 1; 11139676f0c6SRam Pai } 11149676f0c6SRam Pai return 0; 11159676f0c6SRam Pai } 11169676f0c6SRam Pai 11172dafe1c4SEric Sandeen /* 11182dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 11192dafe1c4SEric Sandeen */ 11202dafe1c4SEric Sandeen static noinline int do_move_mount(struct nameidata *nd, char *old_name) 11211da177e4SLinus Torvalds { 11221a390689SAl Viro struct nameidata old_nd; 11231a390689SAl Viro struct path parent_path; 11241da177e4SLinus Torvalds struct vfsmount *p; 11251da177e4SLinus Torvalds int err = 0; 11261da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 11271da177e4SLinus Torvalds return -EPERM; 11281da177e4SLinus Torvalds if (!old_name || !*old_name) 11291da177e4SLinus Torvalds return -EINVAL; 11301da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 11311da177e4SLinus Torvalds if (err) 11321da177e4SLinus Torvalds return err; 11331da177e4SLinus Torvalds 1134390c6843SRam Pai down_write(&namespace_sem); 11354ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 11364ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 11371da177e4SLinus Torvalds ; 11381da177e4SLinus Torvalds err = -EINVAL; 11394ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 11401da177e4SLinus Torvalds goto out; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds err = -ENOENT; 11434ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 11444ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 11451da177e4SLinus Torvalds goto out1; 11461da177e4SLinus Torvalds 11474ac91378SJan Blunck if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) 114821444403SRam Pai goto out1; 11491da177e4SLinus Torvalds 11501da177e4SLinus Torvalds err = -EINVAL; 11514ac91378SJan Blunck if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) 115221444403SRam Pai goto out1; 11531da177e4SLinus Torvalds 11544ac91378SJan Blunck if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) 115521444403SRam Pai goto out1; 11561da177e4SLinus Torvalds 11574ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 11584ac91378SJan Blunck S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) 115921444403SRam Pai goto out1; 116021444403SRam Pai /* 116121444403SRam Pai * Don't move a mount residing in a shared parent. 116221444403SRam Pai */ 11634ac91378SJan Blunck if (old_nd.path.mnt->mnt_parent && 11644ac91378SJan Blunck IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) 116521444403SRam Pai goto out1; 11669676f0c6SRam Pai /* 11679676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 11689676f0c6SRam Pai * mount which is shared. 11699676f0c6SRam Pai */ 11704ac91378SJan Blunck if (IS_MNT_SHARED(nd->path.mnt) && 11714ac91378SJan Blunck tree_contains_unbindable(old_nd.path.mnt)) 11729676f0c6SRam Pai goto out1; 11731da177e4SLinus Torvalds err = -ELOOP; 11744ac91378SJan Blunck for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) 11754ac91378SJan Blunck if (p == old_nd.path.mnt) 117621444403SRam Pai goto out1; 11771da177e4SLinus Torvalds 11781a390689SAl Viro err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); 11794ac91378SJan Blunck if (err) 118021444403SRam Pai goto out1; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 11831da177e4SLinus Torvalds * automatically */ 11844ac91378SJan Blunck list_del_init(&old_nd.path.mnt->mnt_expire); 11851da177e4SLinus Torvalds out1: 11864ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 11871da177e4SLinus Torvalds out: 1188390c6843SRam Pai up_write(&namespace_sem); 11891da177e4SLinus Torvalds if (!err) 11901a390689SAl Viro path_put(&parent_path); 11911d957f9bSJan Blunck path_put(&old_nd.path); 11921da177e4SLinus Torvalds return err; 11931da177e4SLinus Torvalds } 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds /* 11961da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 11971da177e4SLinus Torvalds * namespace's tree 11982dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 11991da177e4SLinus Torvalds */ 12002dafe1c4SEric Sandeen static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, 12011da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 12021da177e4SLinus Torvalds { 12031da177e4SLinus Torvalds struct vfsmount *mnt; 12041da177e4SLinus Torvalds 12051da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 12061da177e4SLinus Torvalds return -EINVAL; 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* we need capabilities... */ 12091da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 12101da177e4SLinus Torvalds return -EPERM; 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 12131da177e4SLinus Torvalds if (IS_ERR(mnt)) 12141da177e4SLinus Torvalds return PTR_ERR(mnt); 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds /* 12201da177e4SLinus Torvalds * add a mount into a namespace's mount tree 12211da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 12221da177e4SLinus Torvalds */ 12231da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 12241da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 12251da177e4SLinus Torvalds { 12261da177e4SLinus Torvalds int err; 12271da177e4SLinus Torvalds 1228390c6843SRam Pai down_write(&namespace_sem); 12291da177e4SLinus Torvalds /* Something was mounted here while we slept */ 12304ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 12314ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 12321da177e4SLinus Torvalds ; 12331da177e4SLinus Torvalds err = -EINVAL; 12344ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 12351da177e4SLinus Torvalds goto unlock; 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 12381da177e4SLinus Torvalds err = -EBUSY; 12394ac91378SJan Blunck if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && 12404ac91378SJan Blunck nd->path.mnt->mnt_root == nd->path.dentry) 12411da177e4SLinus Torvalds goto unlock; 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds err = -EINVAL; 12441da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 12451da177e4SLinus Torvalds goto unlock; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 12485b83d2c5SRam Pai if ((err = graft_tree(newmnt, nd))) 12495b83d2c5SRam Pai goto unlock; 12501da177e4SLinus Torvalds 12516758f953SAl Viro if (fslist) /* add to the specified expiration list */ 125255e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 12536758f953SAl Viro 1254390c6843SRam Pai up_write(&namespace_sem); 12555b83d2c5SRam Pai return 0; 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds unlock: 1258390c6843SRam Pai up_write(&namespace_sem); 12591da177e4SLinus Torvalds mntput(newmnt); 12601da177e4SLinus Torvalds return err; 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 12641da177e4SLinus Torvalds 12655528f911STrond Myklebust /* 12661da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 12671da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 12681da177e4SLinus Torvalds * here 12691da177e4SLinus Torvalds */ 12701da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 12711da177e4SLinus Torvalds { 12721da177e4SLinus Torvalds struct vfsmount *mnt, *next; 12731da177e4SLinus Torvalds LIST_HEAD(graveyard); 1274bcc5c7d2SAl Viro LIST_HEAD(umounts); 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds if (list_empty(mounts)) 12771da177e4SLinus Torvalds return; 12781da177e4SLinus Torvalds 1279bcc5c7d2SAl Viro down_write(&namespace_sem); 12801da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 12831da177e4SLinus Torvalds * following criteria: 12841da177e4SLinus Torvalds * - only referenced by its parent vfsmount 12851da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 12861da177e4SLinus Torvalds * cleared by mntput()) 12871da177e4SLinus Torvalds */ 128855e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 12891da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1290bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 12911da177e4SLinus Torvalds continue; 129255e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 12931da177e4SLinus Torvalds } 1294bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1295bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1296bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1297bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1298bcc5c7d2SAl Viro } 12991da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1300bcc5c7d2SAl Viro up_write(&namespace_sem); 1301bcc5c7d2SAl Viro 1302bcc5c7d2SAl Viro release_mounts(&umounts); 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds /* 13085528f911STrond Myklebust * Ripoff of 'select_parent()' 13095528f911STrond Myklebust * 13105528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 13115528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 13125528f911STrond Myklebust */ 13135528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 13145528f911STrond Myklebust { 13155528f911STrond Myklebust struct vfsmount *this_parent = parent; 13165528f911STrond Myklebust struct list_head *next; 13175528f911STrond Myklebust int found = 0; 13185528f911STrond Myklebust 13195528f911STrond Myklebust repeat: 13205528f911STrond Myklebust next = this_parent->mnt_mounts.next; 13215528f911STrond Myklebust resume: 13225528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 13235528f911STrond Myklebust struct list_head *tmp = next; 13245528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 13255528f911STrond Myklebust 13265528f911STrond Myklebust next = tmp->next; 13275528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 13285528f911STrond Myklebust continue; 13295528f911STrond Myklebust /* 13305528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 13315528f911STrond Myklebust */ 13325528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 13335528f911STrond Myklebust this_parent = mnt; 13345528f911STrond Myklebust goto repeat; 13355528f911STrond Myklebust } 13365528f911STrond Myklebust 13375528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 13385528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 13395528f911STrond Myklebust found++; 13405528f911STrond Myklebust } 13415528f911STrond Myklebust } 13425528f911STrond Myklebust /* 13435528f911STrond Myklebust * All done at this level ... ascend and resume the search 13445528f911STrond Myklebust */ 13455528f911STrond Myklebust if (this_parent != parent) { 13465528f911STrond Myklebust next = this_parent->mnt_child.next; 13475528f911STrond Myklebust this_parent = this_parent->mnt_parent; 13485528f911STrond Myklebust goto resume; 13495528f911STrond Myklebust } 13505528f911STrond Myklebust return found; 13515528f911STrond Myklebust } 13525528f911STrond Myklebust 13535528f911STrond Myklebust /* 13545528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 13555528f911STrond Myklebust * submounts of a specific parent mountpoint 13565528f911STrond Myklebust */ 1357c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 13585528f911STrond Myklebust { 13595528f911STrond Myklebust LIST_HEAD(graveyard); 1360c35038beSAl Viro struct vfsmount *m; 13615528f911STrond Myklebust 13625528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1363c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1364bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1365c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1366bcc5c7d2SAl Viro mnt_expire); 1367bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1368c35038beSAl Viro umount_tree(mnt, 1, umounts); 1369bcc5c7d2SAl Viro } 1370bcc5c7d2SAl Viro } 13715528f911STrond Myklebust } 13725528f911STrond Myklebust 13735528f911STrond Myklebust /* 13741da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 13751da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 13761da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 13771da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 13781da177e4SLinus Torvalds */ 1379b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1380b58fed8bSRam Pai unsigned long n) 13811da177e4SLinus Torvalds { 13821da177e4SLinus Torvalds char *t = to; 13831da177e4SLinus Torvalds const char __user *f = from; 13841da177e4SLinus Torvalds char c; 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 13871da177e4SLinus Torvalds return n; 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds while (n) { 13901da177e4SLinus Torvalds if (__get_user(c, f)) { 13911da177e4SLinus Torvalds memset(t, 0, n); 13921da177e4SLinus Torvalds break; 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds *t++ = c; 13951da177e4SLinus Torvalds f++; 13961da177e4SLinus Torvalds n--; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds return n; 13991da177e4SLinus Torvalds } 14001da177e4SLinus Torvalds 14011da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 14021da177e4SLinus Torvalds { 14031da177e4SLinus Torvalds int i; 14041da177e4SLinus Torvalds unsigned long page; 14051da177e4SLinus Torvalds unsigned long size; 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds *where = 0; 14081da177e4SLinus Torvalds if (!data) 14091da177e4SLinus Torvalds return 0; 14101da177e4SLinus Torvalds 14111da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 14121da177e4SLinus Torvalds return -ENOMEM; 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds /* We only care that *some* data at the address the user 14151da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 14161da177e4SLinus Torvalds * the remainder of the page. 14171da177e4SLinus Torvalds */ 14181da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 14191da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 14201da177e4SLinus Torvalds if (size > PAGE_SIZE) 14211da177e4SLinus Torvalds size = PAGE_SIZE; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 14241da177e4SLinus Torvalds if (!i) { 14251da177e4SLinus Torvalds free_page(page); 14261da177e4SLinus Torvalds return -EFAULT; 14271da177e4SLinus Torvalds } 14281da177e4SLinus Torvalds if (i != PAGE_SIZE) 14291da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 14301da177e4SLinus Torvalds *where = page; 14311da177e4SLinus Torvalds return 0; 14321da177e4SLinus Torvalds } 14331da177e4SLinus Torvalds 14341da177e4SLinus Torvalds /* 14351da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 14361da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 14371da177e4SLinus Torvalds * 14381da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 14391da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 14401da177e4SLinus Torvalds * information (or be NULL). 14411da177e4SLinus Torvalds * 14421da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 14431da177e4SLinus Torvalds * When the flags word was introduced its top half was required 14441da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 14451da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 14461da177e4SLinus Torvalds * and must be discarded. 14471da177e4SLinus Torvalds */ 14481da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 14491da177e4SLinus Torvalds unsigned long flags, void *data_page) 14501da177e4SLinus Torvalds { 14511da177e4SLinus Torvalds struct nameidata nd; 14521da177e4SLinus Torvalds int retval = 0; 14531da177e4SLinus Torvalds int mnt_flags = 0; 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds /* Discard magic */ 14561da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 14571da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds /* Basic sanity checks */ 14601da177e4SLinus Torvalds 14611da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 14621da177e4SLinus Torvalds return -EINVAL; 14631da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 14641da177e4SLinus Torvalds return -EINVAL; 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds if (data_page) 14671da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 14701da177e4SLinus Torvalds if (flags & MS_NOSUID) 14711da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 14721da177e4SLinus Torvalds if (flags & MS_NODEV) 14731da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 14741da177e4SLinus Torvalds if (flags & MS_NOEXEC) 14751da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1476fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1477fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1478fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1479fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 148047ae32d6SValerie Henson if (flags & MS_RELATIME) 148147ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 1482fc33a7bbSChristoph Hellwig 1483fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 14848bf9725cSPavel Emelyanov MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); 14851da177e4SLinus Torvalds 14861da177e4SLinus Torvalds /* ... and get the mountpoint */ 14871da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 14881da177e4SLinus Torvalds if (retval) 14891da177e4SLinus Torvalds return retval; 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 14921da177e4SLinus Torvalds if (retval) 14931da177e4SLinus Torvalds goto dput_out; 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds if (flags & MS_REMOUNT) 14961da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 14971da177e4SLinus Torvalds data_page); 14981da177e4SLinus Torvalds else if (flags & MS_BIND) 1499eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 15009676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 150107b20889SRam Pai retval = do_change_type(&nd, flags); 15021da177e4SLinus Torvalds else if (flags & MS_MOVE) 15031da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 15041da177e4SLinus Torvalds else 15051da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 15061da177e4SLinus Torvalds dev_name, data_page); 15071da177e4SLinus Torvalds dput_out: 15081d957f9bSJan Blunck path_put(&nd.path); 15091da177e4SLinus Torvalds return retval; 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds 1512741a2951SJANAK DESAI /* 1513741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1514741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1515741a2951SJANAK DESAI */ 1516e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 15176b3286edSKirill Korotaev struct fs_struct *fs) 15181da177e4SLinus Torvalds { 15196b3286edSKirill Korotaev struct mnt_namespace *new_ns; 15201da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 15211da177e4SLinus Torvalds struct vfsmount *p, *q; 15221da177e4SLinus Torvalds 15236b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 15241da177e4SLinus Torvalds if (!new_ns) 1525467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 15261da177e4SLinus Torvalds 15271da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 15281da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 15295addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 15305addc5ddSAl Viro new_ns->event = 0; 15311da177e4SLinus Torvalds 1532390c6843SRam Pai down_write(&namespace_sem); 15331da177e4SLinus Torvalds /* First pass: copy the tree topology */ 15346b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 15359676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 15361da177e4SLinus Torvalds if (!new_ns->root) { 1537390c6843SRam Pai up_write(&namespace_sem); 15381da177e4SLinus Torvalds kfree(new_ns); 1539467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM);; 15401da177e4SLinus Torvalds } 15411da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 15421da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 15431da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 15441da177e4SLinus Torvalds 15451da177e4SLinus Torvalds /* 15461da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 15471da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 15481da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 15491da177e4SLinus Torvalds */ 15506b3286edSKirill Korotaev p = mnt_ns->root; 15511da177e4SLinus Torvalds q = new_ns->root; 15521da177e4SLinus Torvalds while (p) { 15536b3286edSKirill Korotaev q->mnt_ns = new_ns; 15541da177e4SLinus Torvalds if (fs) { 15556ac08c39SJan Blunck if (p == fs->root.mnt) { 15561da177e4SLinus Torvalds rootmnt = p; 15576ac08c39SJan Blunck fs->root.mnt = mntget(q); 15581da177e4SLinus Torvalds } 15596ac08c39SJan Blunck if (p == fs->pwd.mnt) { 15601da177e4SLinus Torvalds pwdmnt = p; 15616ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 15621da177e4SLinus Torvalds } 15636ac08c39SJan Blunck if (p == fs->altroot.mnt) { 15641da177e4SLinus Torvalds altrootmnt = p; 15656ac08c39SJan Blunck fs->altroot.mnt = mntget(q); 15661da177e4SLinus Torvalds } 15671da177e4SLinus Torvalds } 15686b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 15691da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 15701da177e4SLinus Torvalds } 1571390c6843SRam Pai up_write(&namespace_sem); 15721da177e4SLinus Torvalds 15731da177e4SLinus Torvalds if (rootmnt) 15741da177e4SLinus Torvalds mntput(rootmnt); 15751da177e4SLinus Torvalds if (pwdmnt) 15761da177e4SLinus Torvalds mntput(pwdmnt); 15771da177e4SLinus Torvalds if (altrootmnt) 15781da177e4SLinus Torvalds mntput(altrootmnt); 15791da177e4SLinus Torvalds 1580741a2951SJANAK DESAI return new_ns; 1581741a2951SJANAK DESAI } 1582741a2951SJANAK DESAI 1583213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 1584e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 1585741a2951SJANAK DESAI { 15866b3286edSKirill Korotaev struct mnt_namespace *new_ns; 1587741a2951SJANAK DESAI 1588e3222c4eSBadari Pulavarty BUG_ON(!ns); 15896b3286edSKirill Korotaev get_mnt_ns(ns); 1590741a2951SJANAK DESAI 1591741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1592e3222c4eSBadari Pulavarty return ns; 1593741a2951SJANAK DESAI 1594e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 1595741a2951SJANAK DESAI 15966b3286edSKirill Korotaev put_mnt_ns(ns); 1597e3222c4eSBadari Pulavarty return new_ns; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 16011da177e4SLinus Torvalds char __user * type, unsigned long flags, 16021da177e4SLinus Torvalds void __user * data) 16031da177e4SLinus Torvalds { 16041da177e4SLinus Torvalds int retval; 16051da177e4SLinus Torvalds unsigned long data_page; 16061da177e4SLinus Torvalds unsigned long type_page; 16071da177e4SLinus Torvalds unsigned long dev_page; 16081da177e4SLinus Torvalds char *dir_page; 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 16111da177e4SLinus Torvalds if (retval < 0) 16121da177e4SLinus Torvalds return retval; 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds dir_page = getname(dir_name); 16151da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 16161da177e4SLinus Torvalds if (IS_ERR(dir_page)) 16171da177e4SLinus Torvalds goto out1; 16181da177e4SLinus Torvalds 16191da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 16201da177e4SLinus Torvalds if (retval < 0) 16211da177e4SLinus Torvalds goto out2; 16221da177e4SLinus Torvalds 16231da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 16241da177e4SLinus Torvalds if (retval < 0) 16251da177e4SLinus Torvalds goto out3; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds lock_kernel(); 16281da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 16291da177e4SLinus Torvalds flags, (void *)data_page); 16301da177e4SLinus Torvalds unlock_kernel(); 16311da177e4SLinus Torvalds free_page(data_page); 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds out3: 16341da177e4SLinus Torvalds free_page(dev_page); 16351da177e4SLinus Torvalds out2: 16361da177e4SLinus Torvalds putname(dir_page); 16371da177e4SLinus Torvalds out1: 16381da177e4SLinus Torvalds free_page(type_page); 16391da177e4SLinus Torvalds return retval; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds /* 16431da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 16441da177e4SLinus Torvalds * It can block. Requires the big lock held. 16451da177e4SLinus Torvalds */ 1646ac748a09SJan Blunck void set_fs_root(struct fs_struct *fs, struct path *path) 16471da177e4SLinus Torvalds { 16486ac08c39SJan Blunck struct path old_root; 16496ac08c39SJan Blunck 16501da177e4SLinus Torvalds write_lock(&fs->lock); 16511da177e4SLinus Torvalds old_root = fs->root; 1652ac748a09SJan Blunck fs->root = *path; 1653ac748a09SJan Blunck path_get(path); 16541da177e4SLinus Torvalds write_unlock(&fs->lock); 16556ac08c39SJan Blunck if (old_root.dentry) 16566ac08c39SJan Blunck path_put(&old_root); 16571da177e4SLinus Torvalds } 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds /* 16601da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 16611da177e4SLinus Torvalds * It can block. Requires the big lock held. 16621da177e4SLinus Torvalds */ 1663ac748a09SJan Blunck void set_fs_pwd(struct fs_struct *fs, struct path *path) 16641da177e4SLinus Torvalds { 16656ac08c39SJan Blunck struct path old_pwd; 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds write_lock(&fs->lock); 16681da177e4SLinus Torvalds old_pwd = fs->pwd; 1669ac748a09SJan Blunck fs->pwd = *path; 1670ac748a09SJan Blunck path_get(path); 16711da177e4SLinus Torvalds write_unlock(&fs->lock); 16721da177e4SLinus Torvalds 16736ac08c39SJan Blunck if (old_pwd.dentry) 16746ac08c39SJan Blunck path_put(&old_pwd); 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds 16771a390689SAl Viro static void chroot_fs_refs(struct path *old_root, struct path *new_root) 16781da177e4SLinus Torvalds { 16791da177e4SLinus Torvalds struct task_struct *g, *p; 16801da177e4SLinus Torvalds struct fs_struct *fs; 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds read_lock(&tasklist_lock); 16831da177e4SLinus Torvalds do_each_thread(g, p) { 16841da177e4SLinus Torvalds task_lock(p); 16851da177e4SLinus Torvalds fs = p->fs; 16861da177e4SLinus Torvalds if (fs) { 16871da177e4SLinus Torvalds atomic_inc(&fs->count); 16881da177e4SLinus Torvalds task_unlock(p); 16891a390689SAl Viro if (fs->root.dentry == old_root->dentry 16901a390689SAl Viro && fs->root.mnt == old_root->mnt) 16911a390689SAl Viro set_fs_root(fs, new_root); 16921a390689SAl Viro if (fs->pwd.dentry == old_root->dentry 16931a390689SAl Viro && fs->pwd.mnt == old_root->mnt) 16941a390689SAl Viro set_fs_pwd(fs, new_root); 16951da177e4SLinus Torvalds put_fs_struct(fs); 16961da177e4SLinus Torvalds } else 16971da177e4SLinus Torvalds task_unlock(p); 16981da177e4SLinus Torvalds } while_each_thread(g, p); 16991da177e4SLinus Torvalds read_unlock(&tasklist_lock); 17001da177e4SLinus Torvalds } 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds /* 17031da177e4SLinus Torvalds * pivot_root Semantics: 17041da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 17051da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 17061da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 17071da177e4SLinus Torvalds * 17081da177e4SLinus Torvalds * Restrictions: 17091da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 17101da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 17111da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 17121da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 17131da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 17141da177e4SLinus Torvalds * 17154a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 17164a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 17174a0d11faSNeil Brown * in this situation. 17184a0d11faSNeil Brown * 17191da177e4SLinus Torvalds * Notes: 17201da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 17211da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 17221da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 17231da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 17241da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 17251da177e4SLinus Torvalds * first. 17261da177e4SLinus Torvalds */ 1727b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 1728b58fed8bSRam Pai const char __user * put_old) 17291da177e4SLinus Torvalds { 17301da177e4SLinus Torvalds struct vfsmount *tmp; 17311a390689SAl Viro struct nameidata new_nd, old_nd, user_nd; 17321a390689SAl Viro struct path parent_path, root_parent; 17331da177e4SLinus Torvalds int error; 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17361da177e4SLinus Torvalds return -EPERM; 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds lock_kernel(); 17391da177e4SLinus Torvalds 1740b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1741b58fed8bSRam Pai &new_nd); 17421da177e4SLinus Torvalds if (error) 17431da177e4SLinus Torvalds goto out0; 17441da177e4SLinus Torvalds error = -EINVAL; 17454ac91378SJan Blunck if (!check_mnt(new_nd.path.mnt)) 17461da177e4SLinus Torvalds goto out1; 17471da177e4SLinus Torvalds 17481da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 17491da177e4SLinus Torvalds if (error) 17501da177e4SLinus Torvalds goto out1; 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 17531da177e4SLinus Torvalds if (error) { 17541d957f9bSJan Blunck path_put(&old_nd.path); 17551da177e4SLinus Torvalds goto out1; 17561da177e4SLinus Torvalds } 17571da177e4SLinus Torvalds 17581da177e4SLinus Torvalds read_lock(¤t->fs->lock); 17596ac08c39SJan Blunck user_nd.path = current->fs->root; 17606ac08c39SJan Blunck path_get(¤t->fs->root); 17611da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 1762390c6843SRam Pai down_write(&namespace_sem); 17634ac91378SJan Blunck mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); 17641da177e4SLinus Torvalds error = -EINVAL; 17654ac91378SJan Blunck if (IS_MNT_SHARED(old_nd.path.mnt) || 17664ac91378SJan Blunck IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || 17674ac91378SJan Blunck IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) 176821444403SRam Pai goto out2; 17694ac91378SJan Blunck if (!check_mnt(user_nd.path.mnt)) 17701da177e4SLinus Torvalds goto out2; 17711da177e4SLinus Torvalds error = -ENOENT; 17724ac91378SJan Blunck if (IS_DEADDIR(new_nd.path.dentry->d_inode)) 17731da177e4SLinus Torvalds goto out2; 17744ac91378SJan Blunck if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) 17751da177e4SLinus Torvalds goto out2; 17764ac91378SJan Blunck if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) 17771da177e4SLinus Torvalds goto out2; 17781da177e4SLinus Torvalds error = -EBUSY; 17794ac91378SJan Blunck if (new_nd.path.mnt == user_nd.path.mnt || 17804ac91378SJan Blunck old_nd.path.mnt == user_nd.path.mnt) 17811da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 17821da177e4SLinus Torvalds error = -EINVAL; 17834ac91378SJan Blunck if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) 17841da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 17854ac91378SJan Blunck if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) 17860bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 17874ac91378SJan Blunck if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) 17881da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 17894ac91378SJan Blunck if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) 17900bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 17914ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 17924ac91378SJan Blunck tmp = old_nd.path.mnt; 17931da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17944ac91378SJan Blunck if (tmp != new_nd.path.mnt) { 17951da177e4SLinus Torvalds for (;;) { 17961da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 17971da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 17984ac91378SJan Blunck if (tmp->mnt_parent == new_nd.path.mnt) 17991da177e4SLinus Torvalds break; 18001da177e4SLinus Torvalds tmp = tmp->mnt_parent; 18011da177e4SLinus Torvalds } 18024ac91378SJan Blunck if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) 18031da177e4SLinus Torvalds goto out3; 18044ac91378SJan Blunck } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) 18051da177e4SLinus Torvalds goto out3; 18061a390689SAl Viro detach_mnt(new_nd.path.mnt, &parent_path); 18074ac91378SJan Blunck detach_mnt(user_nd.path.mnt, &root_parent); 18084ac91378SJan Blunck /* mount old root on put_old */ 18091a390689SAl Viro attach_mnt(user_nd.path.mnt, &old_nd.path); 18104ac91378SJan Blunck /* mount new_root on / */ 18114ac91378SJan Blunck attach_mnt(new_nd.path.mnt, &root_parent); 18126b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 18131da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 18141a390689SAl Viro chroot_fs_refs(&user_nd.path, &new_nd.path); 18151da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 18161da177e4SLinus Torvalds error = 0; 18171a390689SAl Viro path_put(&root_parent); 18181a390689SAl Viro path_put(&parent_path); 18191da177e4SLinus Torvalds out2: 18204ac91378SJan Blunck mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); 1821390c6843SRam Pai up_write(&namespace_sem); 18221d957f9bSJan Blunck path_put(&user_nd.path); 18231d957f9bSJan Blunck path_put(&old_nd.path); 18241da177e4SLinus Torvalds out1: 18251d957f9bSJan Blunck path_put(&new_nd.path); 18261da177e4SLinus Torvalds out0: 18271da177e4SLinus Torvalds unlock_kernel(); 18281da177e4SLinus Torvalds return error; 18291da177e4SLinus Torvalds out3: 18301da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 18311da177e4SLinus Torvalds goto out2; 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds static void __init init_mount_tree(void) 18351da177e4SLinus Torvalds { 18361da177e4SLinus Torvalds struct vfsmount *mnt; 18376b3286edSKirill Korotaev struct mnt_namespace *ns; 1838ac748a09SJan Blunck struct path root; 18391da177e4SLinus Torvalds 18401da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 18411da177e4SLinus Torvalds if (IS_ERR(mnt)) 18421da177e4SLinus Torvalds panic("Can't create rootfs"); 18436b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 18446b3286edSKirill Korotaev if (!ns) 18451da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 18466b3286edSKirill Korotaev atomic_set(&ns->count, 1); 18476b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 18486b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 18496b3286edSKirill Korotaev ns->event = 0; 18506b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 18516b3286edSKirill Korotaev ns->root = mnt; 18526b3286edSKirill Korotaev mnt->mnt_ns = ns; 18531da177e4SLinus Torvalds 18546b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 18556b3286edSKirill Korotaev get_mnt_ns(ns); 18561da177e4SLinus Torvalds 1857ac748a09SJan Blunck root.mnt = ns->root; 1858ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 1859ac748a09SJan Blunck 1860ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 1861ac748a09SJan Blunck set_fs_root(current->fs, &root); 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds 186474bf17cfSDenis Cheng void __init mnt_init(void) 18651da177e4SLinus Torvalds { 186613f14b4dSEric Dumazet unsigned u; 186715a67dd8SRandy Dunlap int err; 18681da177e4SLinus Torvalds 1869390c6843SRam Pai init_rwsem(&namespace_sem); 1870390c6843SRam Pai 18711da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 187220c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 18731da177e4SLinus Torvalds 1874b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 18751da177e4SLinus Torvalds 18761da177e4SLinus Torvalds if (!mount_hashtable) 18771da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 18781da177e4SLinus Torvalds 187913f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 18801da177e4SLinus Torvalds 188113f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 188213f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 18831da177e4SLinus Torvalds 188415a67dd8SRandy Dunlap err = sysfs_init(); 188515a67dd8SRandy Dunlap if (err) 188615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 188715a67dd8SRandy Dunlap __FUNCTION__, err); 188800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 188900d26666SGreg Kroah-Hartman if (!fs_kobj) 189000d26666SGreg Kroah-Hartman printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); 18911da177e4SLinus Torvalds init_rootfs(); 18921da177e4SLinus Torvalds init_mount_tree(); 18931da177e4SLinus Torvalds } 18941da177e4SLinus Torvalds 18956b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 18961da177e4SLinus Torvalds { 18976b3286edSKirill Korotaev struct vfsmount *root = ns->root; 189870fbcdf4SRam Pai LIST_HEAD(umount_list); 18996b3286edSKirill Korotaev ns->root = NULL; 19001ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 1901390c6843SRam Pai down_write(&namespace_sem); 19021da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1903a05964f3SRam Pai umount_tree(root, 0, &umount_list); 19041da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1905390c6843SRam Pai up_write(&namespace_sem); 190670fbcdf4SRam Pai release_mounts(&umount_list); 19076b3286edSKirill Korotaev kfree(ns); 19081da177e4SLinus Torvalds } 1909