11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/config.h> 121da177e4SLinus Torvalds #include <linux/syscalls.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 141da177e4SLinus Torvalds #include <linux/sched.h> 151da177e4SLinus Torvalds #include <linux/smp_lock.h> 161da177e4SLinus Torvalds #include <linux/init.h> 171da177e4SLinus Torvalds #include <linux/quotaops.h> 181da177e4SLinus Torvalds #include <linux/acct.h> 191da177e4SLinus Torvalds #include <linux/module.h> 201da177e4SLinus Torvalds #include <linux/seq_file.h> 211da177e4SLinus Torvalds #include <linux/namespace.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/security.h> 241da177e4SLinus Torvalds #include <linux/mount.h> 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds #include <asm/unistd.h> 271da177e4SLinus Torvalds 281da177e4SLinus Torvalds extern int __init init_rootfs(void); 291da177e4SLinus Torvalds 301da177e4SLinus Torvalds #ifdef CONFIG_SYSFS 311da177e4SLinus Torvalds extern int __init sysfs_init(void); 321da177e4SLinus Torvalds #else 331da177e4SLinus Torvalds static inline int sysfs_init(void) 341da177e4SLinus Torvalds { 351da177e4SLinus Torvalds return 0; 361da177e4SLinus Torvalds } 371da177e4SLinus Torvalds #endif 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 401da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 411da177e4SLinus Torvalds 421da177e4SLinus Torvalds static struct list_head *mount_hashtable; 431da177e4SLinus Torvalds static int hash_mask, hash_bits; 441da177e4SLinus Torvalds static kmem_cache_t *mnt_cache; 451da177e4SLinus Torvalds 461da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 471da177e4SLinus Torvalds { 481da177e4SLinus Torvalds unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES); 491da177e4SLinus Torvalds tmp += ((unsigned long) dentry / L1_CACHE_BYTES); 501da177e4SLinus Torvalds tmp = tmp + (tmp >> hash_bits); 511da177e4SLinus Torvalds return tmp & hash_mask; 521da177e4SLinus Torvalds } 531da177e4SLinus Torvalds 541da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 551da177e4SLinus Torvalds { 561da177e4SLinus Torvalds struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); 571da177e4SLinus Torvalds if (mnt) { 581da177e4SLinus Torvalds memset(mnt, 0, sizeof(struct vfsmount)); 591da177e4SLinus Torvalds atomic_set(&mnt->mnt_count,1); 601da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 611da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 621da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 631da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 641da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_fslink); 651da177e4SLinus Torvalds if (name) { 661da177e4SLinus Torvalds int size = strlen(name)+1; 671da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 681da177e4SLinus Torvalds if (newname) { 691da177e4SLinus Torvalds memcpy(newname, name, size); 701da177e4SLinus Torvalds mnt->mnt_devname = newname; 711da177e4SLinus Torvalds } 721da177e4SLinus Torvalds } 731da177e4SLinus Torvalds } 741da177e4SLinus Torvalds return mnt; 751da177e4SLinus Torvalds } 761da177e4SLinus Torvalds 771da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 781da177e4SLinus Torvalds { 791da177e4SLinus Torvalds kfree(mnt->mnt_devname); 801da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 811da177e4SLinus Torvalds } 821da177e4SLinus Torvalds 831da177e4SLinus Torvalds /* 841da177e4SLinus Torvalds * Now, lookup_mnt increments the ref count before returning 851da177e4SLinus Torvalds * the vfsmount struct. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 881da177e4SLinus Torvalds { 891da177e4SLinus Torvalds struct list_head * head = mount_hashtable + hash(mnt, dentry); 901da177e4SLinus Torvalds struct list_head * tmp = head; 911da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 941da177e4SLinus Torvalds for (;;) { 951da177e4SLinus Torvalds tmp = tmp->next; 961da177e4SLinus Torvalds p = NULL; 971da177e4SLinus Torvalds if (tmp == head) 981da177e4SLinus Torvalds break; 991da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 1001da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 1011da177e4SLinus Torvalds found = mntget(p); 1021da177e4SLinus Torvalds break; 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds } 1051da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1061da177e4SLinus Torvalds return found; 1071da177e4SLinus Torvalds } 1081da177e4SLinus Torvalds 1091da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 1101da177e4SLinus Torvalds { 1111da177e4SLinus Torvalds return mnt->mnt_namespace == current->namespace; 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) 1151da177e4SLinus Torvalds { 1161da177e4SLinus Torvalds old_nd->dentry = mnt->mnt_mountpoint; 1171da177e4SLinus Torvalds old_nd->mnt = mnt->mnt_parent; 1181da177e4SLinus Torvalds mnt->mnt_parent = mnt; 1191da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 1201da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 1211da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 1221da177e4SLinus Torvalds old_nd->dentry->d_mounted--; 1231da177e4SLinus Torvalds } 1241da177e4SLinus Torvalds 1251da177e4SLinus Torvalds static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) 1261da177e4SLinus Torvalds { 1271da177e4SLinus Torvalds mnt->mnt_parent = mntget(nd->mnt); 1281da177e4SLinus Torvalds mnt->mnt_mountpoint = dget(nd->dentry); 1291da177e4SLinus Torvalds list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry)); 1301da177e4SLinus Torvalds list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); 1311da177e4SLinus Torvalds nd->dentry->d_mounted++; 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 1351da177e4SLinus Torvalds { 1361da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 1371da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 1381da177e4SLinus Torvalds while (1) { 1391da177e4SLinus Torvalds if (p == root) 1401da177e4SLinus Torvalds return NULL; 1411da177e4SLinus Torvalds next = p->mnt_child.next; 1421da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 1431da177e4SLinus Torvalds break; 1441da177e4SLinus Torvalds p = p->mnt_parent; 1451da177e4SLinus Torvalds } 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 1481da177e4SLinus Torvalds } 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds static struct vfsmount * 1511da177e4SLinus Torvalds clone_mnt(struct vfsmount *old, struct dentry *root) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 1541da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 1551da177e4SLinus Torvalds 1561da177e4SLinus Torvalds if (mnt) { 1571da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 1581da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1591da177e4SLinus Torvalds mnt->mnt_sb = sb; 1601da177e4SLinus Torvalds mnt->mnt_root = dget(root); 1611da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 1621da177e4SLinus Torvalds mnt->mnt_parent = mnt; 1631da177e4SLinus Torvalds mnt->mnt_namespace = old->mnt_namespace; 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 1661da177e4SLinus Torvalds * as the original if that was on one */ 1671da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1681da177e4SLinus Torvalds if (!list_empty(&old->mnt_fslink)) 1691da177e4SLinus Torvalds list_add(&mnt->mnt_fslink, &old->mnt_fslink); 1701da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1711da177e4SLinus Torvalds } 1721da177e4SLinus Torvalds return mnt; 1731da177e4SLinus Torvalds } 1741da177e4SLinus Torvalds 1751da177e4SLinus Torvalds void __mntput(struct vfsmount *mnt) 1761da177e4SLinus Torvalds { 1771da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 1781da177e4SLinus Torvalds dput(mnt->mnt_root); 1791da177e4SLinus Torvalds free_vfsmnt(mnt); 1801da177e4SLinus Torvalds deactivate_super(sb); 1811da177e4SLinus Torvalds } 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds EXPORT_SYMBOL(__mntput); 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds /* iterator */ 1861da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 1871da177e4SLinus Torvalds { 1881da177e4SLinus Torvalds struct namespace *n = m->private; 1891da177e4SLinus Torvalds struct list_head *p; 1901da177e4SLinus Torvalds loff_t l = *pos; 1911da177e4SLinus Torvalds 1921da177e4SLinus Torvalds down_read(&n->sem); 1931da177e4SLinus Torvalds list_for_each(p, &n->list) 1941da177e4SLinus Torvalds if (!l--) 1951da177e4SLinus Torvalds return list_entry(p, struct vfsmount, mnt_list); 1961da177e4SLinus Torvalds return NULL; 1971da177e4SLinus Torvalds } 1981da177e4SLinus Torvalds 1991da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 2001da177e4SLinus Torvalds { 2011da177e4SLinus Torvalds struct namespace *n = m->private; 2021da177e4SLinus Torvalds struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; 2031da177e4SLinus Torvalds (*pos)++; 2041da177e4SLinus Torvalds return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 2081da177e4SLinus Torvalds { 2091da177e4SLinus Torvalds struct namespace *n = m->private; 2101da177e4SLinus Torvalds up_read(&n->sem); 2111da177e4SLinus Torvalds } 2121da177e4SLinus Torvalds 2131da177e4SLinus Torvalds static inline void mangle(struct seq_file *m, const char *s) 2141da177e4SLinus Torvalds { 2151da177e4SLinus Torvalds seq_escape(m, s, " \t\n\\"); 2161da177e4SLinus Torvalds } 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 2191da177e4SLinus Torvalds { 2201da177e4SLinus Torvalds struct vfsmount *mnt = v; 2211da177e4SLinus Torvalds int err = 0; 2221da177e4SLinus Torvalds static struct proc_fs_info { 2231da177e4SLinus Torvalds int flag; 2241da177e4SLinus Torvalds char *str; 2251da177e4SLinus Torvalds } fs_info[] = { 2261da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 2271da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 2281da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 2291da177e4SLinus Torvalds { MS_NOATIME, ",noatime" }, 2301da177e4SLinus Torvalds { MS_NODIRATIME, ",nodiratime" }, 2311da177e4SLinus Torvalds { 0, NULL } 2321da177e4SLinus Torvalds }; 2331da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 2341da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 2351da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 2361da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 2371da177e4SLinus Torvalds { 0, NULL } 2381da177e4SLinus Torvalds }; 2391da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 2401da177e4SLinus Torvalds 2411da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 2421da177e4SLinus Torvalds seq_putc(m, ' '); 2431da177e4SLinus Torvalds seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 2441da177e4SLinus Torvalds seq_putc(m, ' '); 2451da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 2461da177e4SLinus Torvalds seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 2471da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 2481da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 2491da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 2501da177e4SLinus Torvalds } 2511da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 2521da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 2531da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 2561da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 2571da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 2581da177e4SLinus Torvalds return err; 2591da177e4SLinus Torvalds } 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds struct seq_operations mounts_op = { 2621da177e4SLinus Torvalds .start = m_start, 2631da177e4SLinus Torvalds .next = m_next, 2641da177e4SLinus Torvalds .stop = m_stop, 2651da177e4SLinus Torvalds .show = show_vfsmnt 2661da177e4SLinus Torvalds }; 2671da177e4SLinus Torvalds 2681da177e4SLinus Torvalds /** 2691da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 2701da177e4SLinus Torvalds * @mnt: root of mount tree 2711da177e4SLinus Torvalds * 2721da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 2731da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 2741da177e4SLinus Torvalds * busy. 2751da177e4SLinus Torvalds */ 2761da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 2771da177e4SLinus Torvalds { 2781da177e4SLinus Torvalds struct list_head *next; 2791da177e4SLinus Torvalds struct vfsmount *this_parent = mnt; 2801da177e4SLinus Torvalds int actual_refs; 2811da177e4SLinus Torvalds int minimum_refs; 2821da177e4SLinus Torvalds 2831da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2841da177e4SLinus Torvalds actual_refs = atomic_read(&mnt->mnt_count); 2851da177e4SLinus Torvalds minimum_refs = 2; 2861da177e4SLinus Torvalds repeat: 2871da177e4SLinus Torvalds next = this_parent->mnt_mounts.next; 2881da177e4SLinus Torvalds resume: 2891da177e4SLinus Torvalds while (next != &this_parent->mnt_mounts) { 2901da177e4SLinus Torvalds struct vfsmount *p = list_entry(next, struct vfsmount, mnt_child); 2911da177e4SLinus Torvalds 2921da177e4SLinus Torvalds next = next->next; 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 2951da177e4SLinus Torvalds minimum_refs += 2; 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds if (!list_empty(&p->mnt_mounts)) { 2981da177e4SLinus Torvalds this_parent = p; 2991da177e4SLinus Torvalds goto repeat; 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds if (this_parent != mnt) { 3041da177e4SLinus Torvalds next = this_parent->mnt_child.next; 3051da177e4SLinus Torvalds this_parent = this_parent->mnt_parent; 3061da177e4SLinus Torvalds goto resume; 3071da177e4SLinus Torvalds } 3081da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds if (actual_refs > minimum_refs) 3111da177e4SLinus Torvalds return -EBUSY; 3121da177e4SLinus Torvalds 3131da177e4SLinus Torvalds return 0; 3141da177e4SLinus Torvalds } 3151da177e4SLinus Torvalds 3161da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds /** 3191da177e4SLinus Torvalds * may_umount - check if a mount point is busy 3201da177e4SLinus Torvalds * @mnt: root of mount 3211da177e4SLinus Torvalds * 3221da177e4SLinus Torvalds * This is called to check if a mount point has any 3231da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 3241da177e4SLinus Torvalds * mount has sub mounts this will return busy 3251da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 3261da177e4SLinus Torvalds * 3271da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 3281da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 3291da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 3301da177e4SLinus Torvalds */ 3311da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 3321da177e4SLinus Torvalds { 3331da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) > 2) 3341da177e4SLinus Torvalds return -EBUSY; 3351da177e4SLinus Torvalds return 0; 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 3391da177e4SLinus Torvalds 34052c1da39SAdrian Bunk static void umount_tree(struct vfsmount *mnt) 3411da177e4SLinus Torvalds { 3421da177e4SLinus Torvalds struct vfsmount *p; 3431da177e4SLinus Torvalds LIST_HEAD(kill); 3441da177e4SLinus Torvalds 3451da177e4SLinus Torvalds for (p = mnt; p; p = next_mnt(p, mnt)) { 3461da177e4SLinus Torvalds list_del(&p->mnt_list); 3471da177e4SLinus Torvalds list_add(&p->mnt_list, &kill); 348202322e6SMiklos Szeredi p->mnt_namespace = NULL; 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds while (!list_empty(&kill)) { 3521da177e4SLinus Torvalds mnt = list_entry(kill.next, struct vfsmount, mnt_list); 3531da177e4SLinus Torvalds list_del_init(&mnt->mnt_list); 3541da177e4SLinus Torvalds list_del_init(&mnt->mnt_fslink); 3551da177e4SLinus Torvalds if (mnt->mnt_parent == mnt) { 3561da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 3571da177e4SLinus Torvalds } else { 3581da177e4SLinus Torvalds struct nameidata old_nd; 3591da177e4SLinus Torvalds detach_mnt(mnt, &old_nd); 3601da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 3611da177e4SLinus Torvalds path_release(&old_nd); 3621da177e4SLinus Torvalds } 3631da177e4SLinus Torvalds mntput(mnt); 3641da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 3651da177e4SLinus Torvalds } 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 3691da177e4SLinus Torvalds { 3701da177e4SLinus Torvalds struct super_block * sb = mnt->mnt_sb; 3711da177e4SLinus Torvalds int retval; 3721da177e4SLinus Torvalds 3731da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 3741da177e4SLinus Torvalds if (retval) 3751da177e4SLinus Torvalds return retval; 3761da177e4SLinus Torvalds 3771da177e4SLinus Torvalds /* 3781da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 3791da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 3801da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 3811da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 3821da177e4SLinus Torvalds */ 3831da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 3841da177e4SLinus Torvalds if (mnt == current->fs->rootmnt || 3851da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 3861da177e4SLinus Torvalds return -EINVAL; 3871da177e4SLinus Torvalds 3881da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 3891da177e4SLinus Torvalds return -EBUSY; 3901da177e4SLinus Torvalds 3911da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 3921da177e4SLinus Torvalds return -EAGAIN; 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 3951da177e4SLinus Torvalds /* 3961da177e4SLinus Torvalds * If we may have to abort operations to get out of this 3971da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 3981da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 3991da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 4001da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 4011da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 4021da177e4SLinus Torvalds * about for the moment. 4031da177e4SLinus Torvalds */ 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds lock_kernel(); 4061da177e4SLinus Torvalds if( (flags&MNT_FORCE) && sb->s_op->umount_begin) 4071da177e4SLinus Torvalds sb->s_op->umount_begin(sb); 4081da177e4SLinus Torvalds unlock_kernel(); 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds /* 4111da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 4121da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 4131da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 4141da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 4151da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 4161da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 4171da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 4181da177e4SLinus Torvalds */ 4191da177e4SLinus Torvalds if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Special case for "unmounting" root ... 4221da177e4SLinus Torvalds * we just try to remount it readonly. 4231da177e4SLinus Torvalds */ 4241da177e4SLinus Torvalds down_write(&sb->s_umount); 4251da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 4261da177e4SLinus Torvalds lock_kernel(); 4271da177e4SLinus Torvalds DQUOT_OFF(sb); 4281da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 4291da177e4SLinus Torvalds unlock_kernel(); 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds up_write(&sb->s_umount); 4321da177e4SLinus Torvalds return retval; 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds down_write(¤t->namespace->sem); 4361da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 4371da177e4SLinus Torvalds 4381da177e4SLinus Torvalds if (atomic_read(&sb->s_active) == 1) { 4391da177e4SLinus Torvalds /* last instance - try to be smart */ 4401da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 4411da177e4SLinus Torvalds lock_kernel(); 4421da177e4SLinus Torvalds DQUOT_OFF(sb); 4431da177e4SLinus Torvalds acct_auto_close(sb); 4441da177e4SLinus Torvalds unlock_kernel(); 4451da177e4SLinus Torvalds security_sb_umount_close(mnt); 4461da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 4471da177e4SLinus Torvalds } 4481da177e4SLinus Torvalds retval = -EBUSY; 4491da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) { 4501da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 4511da177e4SLinus Torvalds umount_tree(mnt); 4521da177e4SLinus Torvalds retval = 0; 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 4551da177e4SLinus Torvalds if (retval) 4561da177e4SLinus Torvalds security_sb_umount_busy(mnt); 4571da177e4SLinus Torvalds up_write(¤t->namespace->sem); 4581da177e4SLinus Torvalds return retval; 4591da177e4SLinus Torvalds } 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds /* 4621da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 4631da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 4641da177e4SLinus Torvalds * 4651da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 4661da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 4671da177e4SLinus Torvalds */ 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 4701da177e4SLinus Torvalds { 4711da177e4SLinus Torvalds struct nameidata nd; 4721da177e4SLinus Torvalds int retval; 4731da177e4SLinus Torvalds 4741da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 4751da177e4SLinus Torvalds if (retval) 4761da177e4SLinus Torvalds goto out; 4771da177e4SLinus Torvalds retval = -EINVAL; 4781da177e4SLinus Torvalds if (nd.dentry != nd.mnt->mnt_root) 4791da177e4SLinus Torvalds goto dput_and_out; 4801da177e4SLinus Torvalds if (!check_mnt(nd.mnt)) 4811da177e4SLinus Torvalds goto dput_and_out; 4821da177e4SLinus Torvalds 4831da177e4SLinus Torvalds retval = -EPERM; 4841da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 4851da177e4SLinus Torvalds goto dput_and_out; 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds retval = do_umount(nd.mnt, flags); 4881da177e4SLinus Torvalds dput_and_out: 4891da177e4SLinus Torvalds path_release_on_umount(&nd); 4901da177e4SLinus Torvalds out: 4911da177e4SLinus Torvalds return retval; 4921da177e4SLinus Torvalds } 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 4951da177e4SLinus Torvalds 4961da177e4SLinus Torvalds /* 4971da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 4981da177e4SLinus Torvalds */ 4991da177e4SLinus Torvalds 5001da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 5011da177e4SLinus Torvalds { 5021da177e4SLinus Torvalds return sys_umount(name,0); 5031da177e4SLinus Torvalds } 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds #endif 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 5081da177e4SLinus Torvalds { 5091da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 5101da177e4SLinus Torvalds return 0; 5111da177e4SLinus Torvalds return -EPERM; 5121da177e4SLinus Torvalds #ifdef notyet 5131da177e4SLinus Torvalds if (S_ISLNK(nd->dentry->d_inode->i_mode)) 5141da177e4SLinus Torvalds return -EPERM; 5151da177e4SLinus Torvalds if (nd->dentry->d_inode->i_mode & S_ISVTX) { 5161da177e4SLinus Torvalds if (current->uid != nd->dentry->d_inode->i_uid) 5171da177e4SLinus Torvalds return -EPERM; 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds if (permission(nd->dentry->d_inode, MAY_WRITE, nd)) 5201da177e4SLinus Torvalds return -EPERM; 5211da177e4SLinus Torvalds return 0; 5221da177e4SLinus Torvalds #endif 5231da177e4SLinus Torvalds } 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds static int 5261da177e4SLinus Torvalds lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 5271da177e4SLinus Torvalds { 5281da177e4SLinus Torvalds while (1) { 5291da177e4SLinus Torvalds if (d == dentry) 5301da177e4SLinus Torvalds return 1; 5311da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 5321da177e4SLinus Torvalds return 0; 5331da177e4SLinus Torvalds d = d->d_parent; 5341da177e4SLinus Torvalds } 5351da177e4SLinus Torvalds } 5361da177e4SLinus Torvalds 5371da177e4SLinus Torvalds static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry) 5381da177e4SLinus Torvalds { 5391da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 5401da177e4SLinus Torvalds struct list_head *h; 5411da177e4SLinus Torvalds struct nameidata nd; 5421da177e4SLinus Torvalds 5431da177e4SLinus Torvalds res = q = clone_mnt(mnt, dentry); 5441da177e4SLinus Torvalds if (!q) 5451da177e4SLinus Torvalds goto Enomem; 5461da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds p = mnt; 5491da177e4SLinus Torvalds for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) { 5501da177e4SLinus Torvalds r = list_entry(h, struct vfsmount, mnt_child); 5511da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 5521da177e4SLinus Torvalds continue; 5531da177e4SLinus Torvalds 5541da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 5551da177e4SLinus Torvalds while (p != s->mnt_parent) { 5561da177e4SLinus Torvalds p = p->mnt_parent; 5571da177e4SLinus Torvalds q = q->mnt_parent; 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds p = s; 5601da177e4SLinus Torvalds nd.mnt = q; 5611da177e4SLinus Torvalds nd.dentry = p->mnt_mountpoint; 5621da177e4SLinus Torvalds q = clone_mnt(p, p->mnt_root); 5631da177e4SLinus Torvalds if (!q) 5641da177e4SLinus Torvalds goto Enomem; 5651da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 5661da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 5671da177e4SLinus Torvalds attach_mnt(q, &nd); 5681da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds return res; 5721da177e4SLinus Torvalds Enomem: 5731da177e4SLinus Torvalds if (res) { 5741da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 5751da177e4SLinus Torvalds umount_tree(res); 5761da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 5771da177e4SLinus Torvalds } 5781da177e4SLinus Torvalds return NULL; 5791da177e4SLinus Torvalds } 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 5821da177e4SLinus Torvalds { 5831da177e4SLinus Torvalds int err; 5841da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 5851da177e4SLinus Torvalds return -EINVAL; 5861da177e4SLinus Torvalds 5871da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 5881da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 5891da177e4SLinus Torvalds return -ENOTDIR; 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds err = -ENOENT; 5921da177e4SLinus Torvalds down(&nd->dentry->d_inode->i_sem); 5931da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 5941da177e4SLinus Torvalds goto out_unlock; 5951da177e4SLinus Torvalds 5961da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 5971da177e4SLinus Torvalds if (err) 5981da177e4SLinus Torvalds goto out_unlock; 5991da177e4SLinus Torvalds 6001da177e4SLinus Torvalds err = -ENOENT; 6011da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6021da177e4SLinus Torvalds if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) { 6031da177e4SLinus Torvalds struct list_head head; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds attach_mnt(mnt, nd); 6061da177e4SLinus Torvalds list_add_tail(&head, &mnt->mnt_list); 6071da177e4SLinus Torvalds list_splice(&head, current->namespace->list.prev); 6081da177e4SLinus Torvalds mntget(mnt); 6091da177e4SLinus Torvalds err = 0; 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6121da177e4SLinus Torvalds out_unlock: 6131da177e4SLinus Torvalds up(&nd->dentry->d_inode->i_sem); 6141da177e4SLinus Torvalds if (!err) 6151da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 6161da177e4SLinus Torvalds return err; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds 6191da177e4SLinus Torvalds /* 6201da177e4SLinus Torvalds * do loopback mount. 6211da177e4SLinus Torvalds */ 6221da177e4SLinus Torvalds static int do_loopback(struct nameidata *nd, char *old_name, int recurse) 6231da177e4SLinus Torvalds { 6241da177e4SLinus Torvalds struct nameidata old_nd; 6251da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 6261da177e4SLinus Torvalds int err = mount_is_safe(nd); 6271da177e4SLinus Torvalds if (err) 6281da177e4SLinus Torvalds return err; 6291da177e4SLinus Torvalds if (!old_name || !*old_name) 6301da177e4SLinus Torvalds return -EINVAL; 6311da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 6321da177e4SLinus Torvalds if (err) 6331da177e4SLinus Torvalds return err; 6341da177e4SLinus Torvalds 6351da177e4SLinus Torvalds down_write(¤t->namespace->sem); 6361da177e4SLinus Torvalds err = -EINVAL; 6371da177e4SLinus Torvalds if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) { 6381da177e4SLinus Torvalds err = -ENOMEM; 6391da177e4SLinus Torvalds if (recurse) 6401da177e4SLinus Torvalds mnt = copy_tree(old_nd.mnt, old_nd.dentry); 6411da177e4SLinus Torvalds else 6421da177e4SLinus Torvalds mnt = clone_mnt(old_nd.mnt, old_nd.dentry); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds if (mnt) { 6461da177e4SLinus Torvalds /* stop bind mounts from expiring */ 6471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6481da177e4SLinus Torvalds list_del_init(&mnt->mnt_fslink); 6491da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6501da177e4SLinus Torvalds 6511da177e4SLinus Torvalds err = graft_tree(mnt, nd); 6521da177e4SLinus Torvalds if (err) { 6531da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6541da177e4SLinus Torvalds umount_tree(mnt); 6551da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6561da177e4SLinus Torvalds } else 6571da177e4SLinus Torvalds mntput(mnt); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 6601da177e4SLinus Torvalds up_write(¤t->namespace->sem); 6611da177e4SLinus Torvalds path_release(&old_nd); 6621da177e4SLinus Torvalds return err; 6631da177e4SLinus Torvalds } 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds /* 6661da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 6671da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 6681da177e4SLinus Torvalds * on it - tough luck. 6691da177e4SLinus Torvalds */ 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds static int do_remount(struct nameidata *nd, int flags, int mnt_flags, 6721da177e4SLinus Torvalds void *data) 6731da177e4SLinus Torvalds { 6741da177e4SLinus Torvalds int err; 6751da177e4SLinus Torvalds struct super_block * sb = nd->mnt->mnt_sb; 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 6781da177e4SLinus Torvalds return -EPERM; 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 6811da177e4SLinus Torvalds return -EINVAL; 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds if (nd->dentry != nd->mnt->mnt_root) 6841da177e4SLinus Torvalds return -EINVAL; 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds down_write(&sb->s_umount); 6871da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 6881da177e4SLinus Torvalds if (!err) 6891da177e4SLinus Torvalds nd->mnt->mnt_flags=mnt_flags; 6901da177e4SLinus Torvalds up_write(&sb->s_umount); 6911da177e4SLinus Torvalds if (!err) 6921da177e4SLinus Torvalds security_sb_post_remount(nd->mnt, flags, data); 6931da177e4SLinus Torvalds return err; 6941da177e4SLinus Torvalds } 6951da177e4SLinus Torvalds 6961da177e4SLinus Torvalds static int do_move_mount(struct nameidata *nd, char *old_name) 6971da177e4SLinus Torvalds { 6981da177e4SLinus Torvalds struct nameidata old_nd, parent_nd; 6991da177e4SLinus Torvalds struct vfsmount *p; 7001da177e4SLinus Torvalds int err = 0; 7011da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 7021da177e4SLinus Torvalds return -EPERM; 7031da177e4SLinus Torvalds if (!old_name || !*old_name) 7041da177e4SLinus Torvalds return -EINVAL; 7051da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 7061da177e4SLinus Torvalds if (err) 7071da177e4SLinus Torvalds return err; 7081da177e4SLinus Torvalds 7091da177e4SLinus Torvalds down_write(¤t->namespace->sem); 7101da177e4SLinus Torvalds while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 7111da177e4SLinus Torvalds ; 7121da177e4SLinus Torvalds err = -EINVAL; 7131da177e4SLinus Torvalds if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 7141da177e4SLinus Torvalds goto out; 7151da177e4SLinus Torvalds 7161da177e4SLinus Torvalds err = -ENOENT; 7171da177e4SLinus Torvalds down(&nd->dentry->d_inode->i_sem); 7181da177e4SLinus Torvalds if (IS_DEADDIR(nd->dentry->d_inode)) 7191da177e4SLinus Torvalds goto out1; 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 7221da177e4SLinus Torvalds if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) 7231da177e4SLinus Torvalds goto out2; 7241da177e4SLinus Torvalds 7251da177e4SLinus Torvalds err = -EINVAL; 7261da177e4SLinus Torvalds if (old_nd.dentry != old_nd.mnt->mnt_root) 7271da177e4SLinus Torvalds goto out2; 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds if (old_nd.mnt == old_nd.mnt->mnt_parent) 7301da177e4SLinus Torvalds goto out2; 7311da177e4SLinus Torvalds 7321da177e4SLinus Torvalds if (S_ISDIR(nd->dentry->d_inode->i_mode) != 7331da177e4SLinus Torvalds S_ISDIR(old_nd.dentry->d_inode->i_mode)) 7341da177e4SLinus Torvalds goto out2; 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds err = -ELOOP; 7371da177e4SLinus Torvalds for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent) 7381da177e4SLinus Torvalds if (p == old_nd.mnt) 7391da177e4SLinus Torvalds goto out2; 7401da177e4SLinus Torvalds err = 0; 7411da177e4SLinus Torvalds 7421da177e4SLinus Torvalds detach_mnt(old_nd.mnt, &parent_nd); 7431da177e4SLinus Torvalds attach_mnt(old_nd.mnt, nd); 7441da177e4SLinus Torvalds 7451da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 7461da177e4SLinus Torvalds * automatically */ 7471da177e4SLinus Torvalds list_del_init(&old_nd.mnt->mnt_fslink); 7481da177e4SLinus Torvalds out2: 7491da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7501da177e4SLinus Torvalds out1: 7511da177e4SLinus Torvalds up(&nd->dentry->d_inode->i_sem); 7521da177e4SLinus Torvalds out: 7531da177e4SLinus Torvalds up_write(¤t->namespace->sem); 7541da177e4SLinus Torvalds if (!err) 7551da177e4SLinus Torvalds path_release(&parent_nd); 7561da177e4SLinus Torvalds path_release(&old_nd); 7571da177e4SLinus Torvalds return err; 7581da177e4SLinus Torvalds } 7591da177e4SLinus Torvalds 7601da177e4SLinus Torvalds /* 7611da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 7621da177e4SLinus Torvalds * namespace's tree 7631da177e4SLinus Torvalds */ 7641da177e4SLinus Torvalds static int do_new_mount(struct nameidata *nd, char *type, int flags, 7651da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 7661da177e4SLinus Torvalds { 7671da177e4SLinus Torvalds struct vfsmount *mnt; 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 7701da177e4SLinus Torvalds return -EINVAL; 7711da177e4SLinus Torvalds 7721da177e4SLinus Torvalds /* we need capabilities... */ 7731da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 7741da177e4SLinus Torvalds return -EPERM; 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 7771da177e4SLinus Torvalds if (IS_ERR(mnt)) 7781da177e4SLinus Torvalds return PTR_ERR(mnt); 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds /* 7841da177e4SLinus Torvalds * add a mount into a namespace's mount tree 7851da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 7861da177e4SLinus Torvalds */ 7871da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 7881da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 7891da177e4SLinus Torvalds { 7901da177e4SLinus Torvalds int err; 7911da177e4SLinus Torvalds 7921da177e4SLinus Torvalds down_write(¤t->namespace->sem); 7931da177e4SLinus Torvalds /* Something was mounted here while we slept */ 7941da177e4SLinus Torvalds while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 7951da177e4SLinus Torvalds ; 7961da177e4SLinus Torvalds err = -EINVAL; 7971da177e4SLinus Torvalds if (!check_mnt(nd->mnt)) 7981da177e4SLinus Torvalds goto unlock; 7991da177e4SLinus Torvalds 8001da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 8011da177e4SLinus Torvalds err = -EBUSY; 8021da177e4SLinus Torvalds if (nd->mnt->mnt_sb == newmnt->mnt_sb && 8031da177e4SLinus Torvalds nd->mnt->mnt_root == nd->dentry) 8041da177e4SLinus Torvalds goto unlock; 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds err = -EINVAL; 8071da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 8081da177e4SLinus Torvalds goto unlock; 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 8111da177e4SLinus Torvalds err = graft_tree(newmnt, nd); 8121da177e4SLinus Torvalds 8131da177e4SLinus Torvalds if (err == 0 && fslist) { 8141da177e4SLinus Torvalds /* add to the specified expiration list */ 8151da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 8161da177e4SLinus Torvalds list_add_tail(&newmnt->mnt_fslink, fslist); 8171da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 8181da177e4SLinus Torvalds } 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds unlock: 8211da177e4SLinus Torvalds up_write(¤t->namespace->sem); 8221da177e4SLinus Torvalds mntput(newmnt); 8231da177e4SLinus Torvalds return err; 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 8271da177e4SLinus Torvalds 82824ca2af1SMiklos Szeredi static void expire_mount(struct vfsmount *mnt, struct list_head *mounts) 82924ca2af1SMiklos Szeredi { 83024ca2af1SMiklos Szeredi spin_lock(&vfsmount_lock); 83124ca2af1SMiklos Szeredi 83224ca2af1SMiklos Szeredi /* 833ed42c879SMiklos Szeredi * Check if mount is still attached, if not, let whoever holds it deal 834ed42c879SMiklos Szeredi * with the sucker 835ed42c879SMiklos Szeredi */ 836ed42c879SMiklos Szeredi if (mnt->mnt_parent == mnt) { 837ed42c879SMiklos Szeredi spin_unlock(&vfsmount_lock); 838ed42c879SMiklos Szeredi return; 839ed42c879SMiklos Szeredi } 840ed42c879SMiklos Szeredi 841ed42c879SMiklos Szeredi /* 84224ca2af1SMiklos Szeredi * Check that it is still dead: the count should now be 2 - as 84324ca2af1SMiklos Szeredi * contributed by the vfsmount parent and the mntget above 84424ca2af1SMiklos Szeredi */ 84524ca2af1SMiklos Szeredi if (atomic_read(&mnt->mnt_count) == 2) { 84624ca2af1SMiklos Szeredi struct nameidata old_nd; 84724ca2af1SMiklos Szeredi 84824ca2af1SMiklos Szeredi /* delete from the namespace */ 84924ca2af1SMiklos Szeredi list_del_init(&mnt->mnt_list); 85024ca2af1SMiklos Szeredi detach_mnt(mnt, &old_nd); 85124ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 85224ca2af1SMiklos Szeredi path_release(&old_nd); 85324ca2af1SMiklos Szeredi 85424ca2af1SMiklos Szeredi /* 85524ca2af1SMiklos Szeredi * Now lay it to rest if this was the last ref on the superblock 85624ca2af1SMiklos Szeredi */ 85724ca2af1SMiklos Szeredi if (atomic_read(&mnt->mnt_sb->s_active) == 1) { 85824ca2af1SMiklos Szeredi /* last instance - try to be smart */ 85924ca2af1SMiklos Szeredi lock_kernel(); 86024ca2af1SMiklos Szeredi DQUOT_OFF(mnt->mnt_sb); 86124ca2af1SMiklos Szeredi acct_auto_close(mnt->mnt_sb); 86224ca2af1SMiklos Szeredi unlock_kernel(); 86324ca2af1SMiklos Szeredi } 86424ca2af1SMiklos Szeredi mntput(mnt); 86524ca2af1SMiklos Szeredi } else { 86624ca2af1SMiklos Szeredi /* 86724ca2af1SMiklos Szeredi * Someone brought it back to life whilst we didn't have any 86824ca2af1SMiklos Szeredi * locks held so return it to the expiration list 86924ca2af1SMiklos Szeredi */ 87024ca2af1SMiklos Szeredi list_add_tail(&mnt->mnt_fslink, mounts); 87124ca2af1SMiklos Szeredi spin_unlock(&vfsmount_lock); 87224ca2af1SMiklos Szeredi } 87324ca2af1SMiklos Szeredi } 87424ca2af1SMiklos Szeredi 8751da177e4SLinus Torvalds /* 8761da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 8771da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 8781da177e4SLinus Torvalds * here 8791da177e4SLinus Torvalds */ 8801da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 8811da177e4SLinus Torvalds { 8821da177e4SLinus Torvalds struct namespace *namespace; 8831da177e4SLinus Torvalds struct vfsmount *mnt, *next; 8841da177e4SLinus Torvalds LIST_HEAD(graveyard); 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds if (list_empty(mounts)) 8871da177e4SLinus Torvalds return; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 8921da177e4SLinus Torvalds * following criteria: 8931da177e4SLinus Torvalds * - only referenced by its parent vfsmount 8941da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 8951da177e4SLinus Torvalds * cleared by mntput()) 8961da177e4SLinus Torvalds */ 8971da177e4SLinus Torvalds list_for_each_entry_safe(mnt, next, mounts, mnt_fslink) { 8981da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 8991da177e4SLinus Torvalds atomic_read(&mnt->mnt_count) != 1) 9001da177e4SLinus Torvalds continue; 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds mntget(mnt); 9031da177e4SLinus Torvalds list_move(&mnt->mnt_fslink, &graveyard); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * go through the vfsmounts we've just consigned to the graveyard to 9081da177e4SLinus Torvalds * - check that they're still dead 9091da177e4SLinus Torvalds * - delete the vfsmount from the appropriate namespace under lock 9101da177e4SLinus Torvalds * - dispose of the corpse 9111da177e4SLinus Torvalds */ 9121da177e4SLinus Torvalds while (!list_empty(&graveyard)) { 9131da177e4SLinus Torvalds mnt = list_entry(graveyard.next, struct vfsmount, mnt_fslink); 9141da177e4SLinus Torvalds list_del_init(&mnt->mnt_fslink); 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds /* don't do anything if the namespace is dead - all the 9171da177e4SLinus Torvalds * vfsmounts from it are going away anyway */ 9181da177e4SLinus Torvalds namespace = mnt->mnt_namespace; 9191ce88cf4SMiklos Szeredi if (!namespace || !namespace->root) 9201da177e4SLinus Torvalds continue; 9211da177e4SLinus Torvalds get_namespace(namespace); 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9241da177e4SLinus Torvalds down_write(&namespace->sem); 92524ca2af1SMiklos Szeredi expire_mount(mnt, mounts); 9261da177e4SLinus Torvalds up_write(&namespace->sem); 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds mntput(mnt); 9291da177e4SLinus Torvalds put_namespace(namespace); 9301da177e4SLinus Torvalds 9311da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds 9341da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 9381da177e4SLinus Torvalds 9391da177e4SLinus Torvalds /* 9401da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 9411da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 9421da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 9431da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 9441da177e4SLinus Torvalds */ 9451da177e4SLinus Torvalds static long 9461da177e4SLinus Torvalds exact_copy_from_user(void *to, const void __user *from, unsigned long n) 9471da177e4SLinus Torvalds { 9481da177e4SLinus Torvalds char *t = to; 9491da177e4SLinus Torvalds const char __user *f = from; 9501da177e4SLinus Torvalds char c; 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 9531da177e4SLinus Torvalds return n; 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds while (n) { 9561da177e4SLinus Torvalds if (__get_user(c, f)) { 9571da177e4SLinus Torvalds memset(t, 0, n); 9581da177e4SLinus Torvalds break; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds *t++ = c; 9611da177e4SLinus Torvalds f++; 9621da177e4SLinus Torvalds n--; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds return n; 9651da177e4SLinus Torvalds } 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds int copy_mount_options(const void __user *data, unsigned long *where) 9681da177e4SLinus Torvalds { 9691da177e4SLinus Torvalds int i; 9701da177e4SLinus Torvalds unsigned long page; 9711da177e4SLinus Torvalds unsigned long size; 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds *where = 0; 9741da177e4SLinus Torvalds if (!data) 9751da177e4SLinus Torvalds return 0; 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 9781da177e4SLinus Torvalds return -ENOMEM; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds /* We only care that *some* data at the address the user 9811da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 9821da177e4SLinus Torvalds * the remainder of the page. 9831da177e4SLinus Torvalds */ 9841da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 9851da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 9861da177e4SLinus Torvalds if (size > PAGE_SIZE) 9871da177e4SLinus Torvalds size = PAGE_SIZE; 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 9901da177e4SLinus Torvalds if (!i) { 9911da177e4SLinus Torvalds free_page(page); 9921da177e4SLinus Torvalds return -EFAULT; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds if (i != PAGE_SIZE) 9951da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 9961da177e4SLinus Torvalds *where = page; 9971da177e4SLinus Torvalds return 0; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds /* 10011da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 10021da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 10031da177e4SLinus Torvalds * 10041da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 10051da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 10061da177e4SLinus Torvalds * information (or be NULL). 10071da177e4SLinus Torvalds * 10081da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 10091da177e4SLinus Torvalds * When the flags word was introduced its top half was required 10101da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 10111da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 10121da177e4SLinus Torvalds * and must be discarded. 10131da177e4SLinus Torvalds */ 10141da177e4SLinus Torvalds long do_mount(char * dev_name, char * dir_name, char *type_page, 10151da177e4SLinus Torvalds unsigned long flags, void *data_page) 10161da177e4SLinus Torvalds { 10171da177e4SLinus Torvalds struct nameidata nd; 10181da177e4SLinus Torvalds int retval = 0; 10191da177e4SLinus Torvalds int mnt_flags = 0; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds /* Discard magic */ 10221da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 10231da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds /* Basic sanity checks */ 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 10281da177e4SLinus Torvalds return -EINVAL; 10291da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 10301da177e4SLinus Torvalds return -EINVAL; 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds if (data_page) 10331da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 10361da177e4SLinus Torvalds if (flags & MS_NOSUID) 10371da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 10381da177e4SLinus Torvalds if (flags & MS_NODEV) 10391da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 10401da177e4SLinus Torvalds if (flags & MS_NOEXEC) 10411da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 10421da177e4SLinus Torvalds flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_ACTIVE); 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds /* ... and get the mountpoint */ 10451da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 10461da177e4SLinus Torvalds if (retval) 10471da177e4SLinus Torvalds return retval; 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 10501da177e4SLinus Torvalds if (retval) 10511da177e4SLinus Torvalds goto dput_out; 10521da177e4SLinus Torvalds 10531da177e4SLinus Torvalds if (flags & MS_REMOUNT) 10541da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 10551da177e4SLinus Torvalds data_page); 10561da177e4SLinus Torvalds else if (flags & MS_BIND) 10571da177e4SLinus Torvalds retval = do_loopback(&nd, dev_name, flags & MS_REC); 10581da177e4SLinus Torvalds else if (flags & MS_MOVE) 10591da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 10601da177e4SLinus Torvalds else 10611da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 10621da177e4SLinus Torvalds dev_name, data_page); 10631da177e4SLinus Torvalds dput_out: 10641da177e4SLinus Torvalds path_release(&nd); 10651da177e4SLinus Torvalds return retval; 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds int copy_namespace(int flags, struct task_struct *tsk) 10691da177e4SLinus Torvalds { 10701da177e4SLinus Torvalds struct namespace *namespace = tsk->namespace; 10711da177e4SLinus Torvalds struct namespace *new_ns; 10721da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 10731da177e4SLinus Torvalds struct fs_struct *fs = tsk->fs; 10741da177e4SLinus Torvalds struct vfsmount *p, *q; 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds if (!namespace) 10771da177e4SLinus Torvalds return 0; 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds get_namespace(namespace); 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds if (!(flags & CLONE_NEWNS)) 10821da177e4SLinus Torvalds return 0; 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) { 10851da177e4SLinus Torvalds put_namespace(namespace); 10861da177e4SLinus Torvalds return -EPERM; 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); 10901da177e4SLinus Torvalds if (!new_ns) 10911da177e4SLinus Torvalds goto out; 10921da177e4SLinus Torvalds 10931da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 10941da177e4SLinus Torvalds init_rwsem(&new_ns->sem); 10951da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds down_write(&tsk->namespace->sem); 10981da177e4SLinus Torvalds /* First pass: copy the tree topology */ 10991da177e4SLinus Torvalds new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root); 11001da177e4SLinus Torvalds if (!new_ns->root) { 11011da177e4SLinus Torvalds up_write(&tsk->namespace->sem); 11021da177e4SLinus Torvalds kfree(new_ns); 11031da177e4SLinus Torvalds goto out; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11061da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 11071da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds /* 11101da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 11111da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 11121da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 11131da177e4SLinus Torvalds */ 11141da177e4SLinus Torvalds p = namespace->root; 11151da177e4SLinus Torvalds q = new_ns->root; 11161da177e4SLinus Torvalds while (p) { 11171da177e4SLinus Torvalds q->mnt_namespace = new_ns; 11181da177e4SLinus Torvalds if (fs) { 11191da177e4SLinus Torvalds if (p == fs->rootmnt) { 11201da177e4SLinus Torvalds rootmnt = p; 11211da177e4SLinus Torvalds fs->rootmnt = mntget(q); 11221da177e4SLinus Torvalds } 11231da177e4SLinus Torvalds if (p == fs->pwdmnt) { 11241da177e4SLinus Torvalds pwdmnt = p; 11251da177e4SLinus Torvalds fs->pwdmnt = mntget(q); 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds if (p == fs->altrootmnt) { 11281da177e4SLinus Torvalds altrootmnt = p; 11291da177e4SLinus Torvalds fs->altrootmnt = mntget(q); 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds p = next_mnt(p, namespace->root); 11331da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds up_write(&tsk->namespace->sem); 11361da177e4SLinus Torvalds 11371da177e4SLinus Torvalds tsk->namespace = new_ns; 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds if (rootmnt) 11401da177e4SLinus Torvalds mntput(rootmnt); 11411da177e4SLinus Torvalds if (pwdmnt) 11421da177e4SLinus Torvalds mntput(pwdmnt); 11431da177e4SLinus Torvalds if (altrootmnt) 11441da177e4SLinus Torvalds mntput(altrootmnt); 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds put_namespace(namespace); 11471da177e4SLinus Torvalds return 0; 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds out: 11501da177e4SLinus Torvalds put_namespace(namespace); 11511da177e4SLinus Torvalds return -ENOMEM; 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 11551da177e4SLinus Torvalds char __user * type, unsigned long flags, 11561da177e4SLinus Torvalds void __user * data) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds int retval; 11591da177e4SLinus Torvalds unsigned long data_page; 11601da177e4SLinus Torvalds unsigned long type_page; 11611da177e4SLinus Torvalds unsigned long dev_page; 11621da177e4SLinus Torvalds char *dir_page; 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds retval = copy_mount_options (type, &type_page); 11651da177e4SLinus Torvalds if (retval < 0) 11661da177e4SLinus Torvalds return retval; 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds dir_page = getname(dir_name); 11691da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 11701da177e4SLinus Torvalds if (IS_ERR(dir_page)) 11711da177e4SLinus Torvalds goto out1; 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds retval = copy_mount_options (dev_name, &dev_page); 11741da177e4SLinus Torvalds if (retval < 0) 11751da177e4SLinus Torvalds goto out2; 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds retval = copy_mount_options (data, &data_page); 11781da177e4SLinus Torvalds if (retval < 0) 11791da177e4SLinus Torvalds goto out3; 11801da177e4SLinus Torvalds 11811da177e4SLinus Torvalds lock_kernel(); 11821da177e4SLinus Torvalds retval = do_mount((char*)dev_page, dir_page, (char*)type_page, 11831da177e4SLinus Torvalds flags, (void*)data_page); 11841da177e4SLinus Torvalds unlock_kernel(); 11851da177e4SLinus Torvalds free_page(data_page); 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds out3: 11881da177e4SLinus Torvalds free_page(dev_page); 11891da177e4SLinus Torvalds out2: 11901da177e4SLinus Torvalds putname(dir_page); 11911da177e4SLinus Torvalds out1: 11921da177e4SLinus Torvalds free_page(type_page); 11931da177e4SLinus Torvalds return retval; 11941da177e4SLinus Torvalds } 11951da177e4SLinus Torvalds 11961da177e4SLinus Torvalds /* 11971da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 11981da177e4SLinus Torvalds * It can block. Requires the big lock held. 11991da177e4SLinus Torvalds */ 12001da177e4SLinus Torvalds void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, 12011da177e4SLinus Torvalds struct dentry *dentry) 12021da177e4SLinus Torvalds { 12031da177e4SLinus Torvalds struct dentry *old_root; 12041da177e4SLinus Torvalds struct vfsmount *old_rootmnt; 12051da177e4SLinus Torvalds write_lock(&fs->lock); 12061da177e4SLinus Torvalds old_root = fs->root; 12071da177e4SLinus Torvalds old_rootmnt = fs->rootmnt; 12081da177e4SLinus Torvalds fs->rootmnt = mntget(mnt); 12091da177e4SLinus Torvalds fs->root = dget(dentry); 12101da177e4SLinus Torvalds write_unlock(&fs->lock); 12111da177e4SLinus Torvalds if (old_root) { 12121da177e4SLinus Torvalds dput(old_root); 12131da177e4SLinus Torvalds mntput(old_rootmnt); 12141da177e4SLinus Torvalds } 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds /* 12181da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 12191da177e4SLinus Torvalds * It can block. Requires the big lock held. 12201da177e4SLinus Torvalds */ 12211da177e4SLinus Torvalds void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, 12221da177e4SLinus Torvalds struct dentry *dentry) 12231da177e4SLinus Torvalds { 12241da177e4SLinus Torvalds struct dentry *old_pwd; 12251da177e4SLinus Torvalds struct vfsmount *old_pwdmnt; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds write_lock(&fs->lock); 12281da177e4SLinus Torvalds old_pwd = fs->pwd; 12291da177e4SLinus Torvalds old_pwdmnt = fs->pwdmnt; 12301da177e4SLinus Torvalds fs->pwdmnt = mntget(mnt); 12311da177e4SLinus Torvalds fs->pwd = dget(dentry); 12321da177e4SLinus Torvalds write_unlock(&fs->lock); 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds if (old_pwd) { 12351da177e4SLinus Torvalds dput(old_pwd); 12361da177e4SLinus Torvalds mntput(old_pwdmnt); 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) 12411da177e4SLinus Torvalds { 12421da177e4SLinus Torvalds struct task_struct *g, *p; 12431da177e4SLinus Torvalds struct fs_struct *fs; 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds read_lock(&tasklist_lock); 12461da177e4SLinus Torvalds do_each_thread(g, p) { 12471da177e4SLinus Torvalds task_lock(p); 12481da177e4SLinus Torvalds fs = p->fs; 12491da177e4SLinus Torvalds if (fs) { 12501da177e4SLinus Torvalds atomic_inc(&fs->count); 12511da177e4SLinus Torvalds task_unlock(p); 12521da177e4SLinus Torvalds if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt) 12531da177e4SLinus Torvalds set_fs_root(fs, new_nd->mnt, new_nd->dentry); 12541da177e4SLinus Torvalds if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt) 12551da177e4SLinus Torvalds set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); 12561da177e4SLinus Torvalds put_fs_struct(fs); 12571da177e4SLinus Torvalds } else 12581da177e4SLinus Torvalds task_unlock(p); 12591da177e4SLinus Torvalds } while_each_thread(g, p); 12601da177e4SLinus Torvalds read_unlock(&tasklist_lock); 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds /* 12641da177e4SLinus Torvalds * pivot_root Semantics: 12651da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 12661da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 12671da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 12681da177e4SLinus Torvalds * 12691da177e4SLinus Torvalds * Restrictions: 12701da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 12711da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 12721da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 12731da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 12741da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 12751da177e4SLinus Torvalds * 12761da177e4SLinus Torvalds * Notes: 12771da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 12781da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 12791da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 12801da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 12811da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 12821da177e4SLinus Torvalds * first. 12831da177e4SLinus Torvalds */ 12841da177e4SLinus Torvalds 12851da177e4SLinus Torvalds asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old) 12861da177e4SLinus Torvalds { 12871da177e4SLinus Torvalds struct vfsmount *tmp; 12881da177e4SLinus Torvalds struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; 12891da177e4SLinus Torvalds int error; 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 12921da177e4SLinus Torvalds return -EPERM; 12931da177e4SLinus Torvalds 12941da177e4SLinus Torvalds lock_kernel(); 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd); 12971da177e4SLinus Torvalds if (error) 12981da177e4SLinus Torvalds goto out0; 12991da177e4SLinus Torvalds error = -EINVAL; 13001da177e4SLinus Torvalds if (!check_mnt(new_nd.mnt)) 13011da177e4SLinus Torvalds goto out1; 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd); 13041da177e4SLinus Torvalds if (error) 13051da177e4SLinus Torvalds goto out1; 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 13081da177e4SLinus Torvalds if (error) { 13091da177e4SLinus Torvalds path_release(&old_nd); 13101da177e4SLinus Torvalds goto out1; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds read_lock(¤t->fs->lock); 13141da177e4SLinus Torvalds user_nd.mnt = mntget(current->fs->rootmnt); 13151da177e4SLinus Torvalds user_nd.dentry = dget(current->fs->root); 13161da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 13171da177e4SLinus Torvalds down_write(¤t->namespace->sem); 13181da177e4SLinus Torvalds down(&old_nd.dentry->d_inode->i_sem); 13191da177e4SLinus Torvalds error = -EINVAL; 13201da177e4SLinus Torvalds if (!check_mnt(user_nd.mnt)) 13211da177e4SLinus Torvalds goto out2; 13221da177e4SLinus Torvalds error = -ENOENT; 13231da177e4SLinus Torvalds if (IS_DEADDIR(new_nd.dentry->d_inode)) 13241da177e4SLinus Torvalds goto out2; 13251da177e4SLinus Torvalds if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) 13261da177e4SLinus Torvalds goto out2; 13271da177e4SLinus Torvalds if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) 13281da177e4SLinus Torvalds goto out2; 13291da177e4SLinus Torvalds error = -EBUSY; 13301da177e4SLinus Torvalds if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) 13311da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 13321da177e4SLinus Torvalds error = -EINVAL; 13331da177e4SLinus Torvalds if (user_nd.mnt->mnt_root != user_nd.dentry) 13341da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 13351da177e4SLinus Torvalds if (new_nd.mnt->mnt_root != new_nd.dentry) 13361da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 13371da177e4SLinus Torvalds tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ 13381da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 13391da177e4SLinus Torvalds if (tmp != new_nd.mnt) { 13401da177e4SLinus Torvalds for (;;) { 13411da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 13421da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 13431da177e4SLinus Torvalds if (tmp->mnt_parent == new_nd.mnt) 13441da177e4SLinus Torvalds break; 13451da177e4SLinus Torvalds tmp = tmp->mnt_parent; 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) 13481da177e4SLinus Torvalds goto out3; 13491da177e4SLinus Torvalds } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) 13501da177e4SLinus Torvalds goto out3; 13511da177e4SLinus Torvalds detach_mnt(new_nd.mnt, &parent_nd); 13521da177e4SLinus Torvalds detach_mnt(user_nd.mnt, &root_parent); 13531da177e4SLinus Torvalds attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ 13541da177e4SLinus Torvalds attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ 13551da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 13561da177e4SLinus Torvalds chroot_fs_refs(&user_nd, &new_nd); 13571da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 13581da177e4SLinus Torvalds error = 0; 13591da177e4SLinus Torvalds path_release(&root_parent); 13601da177e4SLinus Torvalds path_release(&parent_nd); 13611da177e4SLinus Torvalds out2: 13621da177e4SLinus Torvalds up(&old_nd.dentry->d_inode->i_sem); 13631da177e4SLinus Torvalds up_write(¤t->namespace->sem); 13641da177e4SLinus Torvalds path_release(&user_nd); 13651da177e4SLinus Torvalds path_release(&old_nd); 13661da177e4SLinus Torvalds out1: 13671da177e4SLinus Torvalds path_release(&new_nd); 13681da177e4SLinus Torvalds out0: 13691da177e4SLinus Torvalds unlock_kernel(); 13701da177e4SLinus Torvalds return error; 13711da177e4SLinus Torvalds out3: 13721da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 13731da177e4SLinus Torvalds goto out2; 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds static void __init init_mount_tree(void) 13771da177e4SLinus Torvalds { 13781da177e4SLinus Torvalds struct vfsmount *mnt; 13791da177e4SLinus Torvalds struct namespace *namespace; 13801da177e4SLinus Torvalds struct task_struct *g, *p; 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 13831da177e4SLinus Torvalds if (IS_ERR(mnt)) 13841da177e4SLinus Torvalds panic("Can't create rootfs"); 13851da177e4SLinus Torvalds namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); 13861da177e4SLinus Torvalds if (!namespace) 13871da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 13881da177e4SLinus Torvalds atomic_set(&namespace->count, 1); 13891da177e4SLinus Torvalds INIT_LIST_HEAD(&namespace->list); 13901da177e4SLinus Torvalds init_rwsem(&namespace->sem); 13911da177e4SLinus Torvalds list_add(&mnt->mnt_list, &namespace->list); 13921da177e4SLinus Torvalds namespace->root = mnt; 13931da177e4SLinus Torvalds mnt->mnt_namespace = namespace; 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds init_task.namespace = namespace; 13961da177e4SLinus Torvalds read_lock(&tasklist_lock); 13971da177e4SLinus Torvalds do_each_thread(g, p) { 13981da177e4SLinus Torvalds get_namespace(namespace); 13991da177e4SLinus Torvalds p->namespace = namespace; 14001da177e4SLinus Torvalds } while_each_thread(g, p); 14011da177e4SLinus Torvalds read_unlock(&tasklist_lock); 14021da177e4SLinus Torvalds 14031da177e4SLinus Torvalds set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); 14041da177e4SLinus Torvalds set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds void __init mnt_init(unsigned long mempages) 14081da177e4SLinus Torvalds { 14091da177e4SLinus Torvalds struct list_head *d; 14101da177e4SLinus Torvalds unsigned int nr_hash; 14111da177e4SLinus Torvalds int i; 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 14141da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds mount_hashtable = (struct list_head *) 14171da177e4SLinus Torvalds __get_free_page(GFP_ATOMIC); 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds if (!mount_hashtable) 14201da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds /* 14231da177e4SLinus Torvalds * Find the power-of-two list-heads that can fit into the allocation.. 14241da177e4SLinus Torvalds * We don't guarantee that "sizeof(struct list_head)" is necessarily 14251da177e4SLinus Torvalds * a power-of-two. 14261da177e4SLinus Torvalds */ 14271da177e4SLinus Torvalds nr_hash = PAGE_SIZE / sizeof(struct list_head); 14281da177e4SLinus Torvalds hash_bits = 0; 14291da177e4SLinus Torvalds do { 14301da177e4SLinus Torvalds hash_bits++; 14311da177e4SLinus Torvalds } while ((nr_hash >> hash_bits) != 0); 14321da177e4SLinus Torvalds hash_bits--; 14331da177e4SLinus Torvalds 14341da177e4SLinus Torvalds /* 14351da177e4SLinus Torvalds * Re-calculate the actual number of entries and the mask 14361da177e4SLinus Torvalds * from the number of bits we can fit. 14371da177e4SLinus Torvalds */ 14381da177e4SLinus Torvalds nr_hash = 1UL << hash_bits; 14391da177e4SLinus Torvalds hash_mask = nr_hash-1; 14401da177e4SLinus Torvalds 14411da177e4SLinus Torvalds printk("Mount-cache hash table entries: %d\n", nr_hash); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds /* And initialize the newly allocated array */ 14441da177e4SLinus Torvalds d = mount_hashtable; 14451da177e4SLinus Torvalds i = nr_hash; 14461da177e4SLinus Torvalds do { 14471da177e4SLinus Torvalds INIT_LIST_HEAD(d); 14481da177e4SLinus Torvalds d++; 14491da177e4SLinus Torvalds i--; 14501da177e4SLinus Torvalds } while (i); 14511da177e4SLinus Torvalds sysfs_init(); 14521da177e4SLinus Torvalds init_rootfs(); 14531da177e4SLinus Torvalds init_mount_tree(); 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds void __put_namespace(struct namespace *namespace) 14571da177e4SLinus Torvalds { 14581ce88cf4SMiklos Szeredi struct vfsmount *root = namespace->root; 14591ce88cf4SMiklos Szeredi namespace->root = NULL; 14601ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 14611da177e4SLinus Torvalds down_write(&namespace->sem); 14621da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 14631ce88cf4SMiklos Szeredi umount_tree(root); 14641da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 14651da177e4SLinus Torvalds up_write(&namespace->sem); 14661da177e4SLinus Torvalds kfree(namespace); 14671da177e4SLinus Torvalds } 1468