fs/nsfs.c

   1 #include <linux/mount.h>
   2 #include <linux/file.h>
   3 #include <linux/fs.h>
   4 #include <linux/proc_ns.h>
   5 #include <linux/magic.h>
   6 #include <linux/ktime.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/user_namespace.h>
   9 #include <linux/nsfs.h>
  10 #include <linux/uaccess.h>
  11
  12 static struct vfsmount *nsfs_mnt;
  13
  14 static long ns_ioctl(struct file *filp, unsigned int ioctl,
  15                         unsigned long arg);
  16 static const struct file_operations ns_file_operations = {
  17         .llseek         = no_llseek,
  18         .unlocked_ioctl = ns_ioctl,
  19 };
  20
  21 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
  22 {
  23         struct inode *inode = d_inode(dentry);
  24         const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
  25
  26         return dynamic_dname(dentry, buffer, buflen, "%s:[%lu]",
  27                 ns_ops->name, inode->i_ino);
  28 }
  29
  30 static void ns_prune_dentry(struct dentry *dentry)
  31 {
  32         struct inode *inode = d_inode(dentry);
  33         if (inode) {
  34                 struct ns_common *ns = inode->i_private;
  35                 atomic_long_set(&ns->stashed, 0);
  36         }
  37 }
  38
  39 const struct dentry_operations ns_dentry_operations =
  40 {
  41         .d_prune        = ns_prune_dentry,
  42         .d_delete       = always_delete_dentry,
  43         .d_dname        = ns_dname,
  44 };
  45
  46 static void nsfs_evict(struct inode *inode)
  47 {
  48         struct ns_common *ns = inode->i_private;
  49         clear_inode(inode);
  50         ns->ops->put(ns);
  51 }
  52
  53 static void *__ns_get_path(struct path *path, struct ns_common *ns)
  54 {
  55         struct vfsmount *mnt = nsfs_mnt;
  56         struct qstr qname = { .name = "", };
  57         struct dentry *dentry;
  58         struct inode *inode;
  59         unsigned long d;
  60
  61         rcu_read_lock();
  62         d = atomic_long_read(&ns->stashed);
  63         if (!d)
  64                 goto slow;
  65         dentry = (struct dentry *)d;
  66         if (!lockref_get_not_dead(&dentry->d_lockref))
  67                 goto slow;
  68         rcu_read_unlock();
  69         ns->ops->put(ns);
  70 got_it:
  71         path->mnt = mntget(mnt);
  72         path->dentry = dentry;
  73         return NULL;
  74 slow:
  75         rcu_read_unlock();
  76         inode = new_inode_pseudo(mnt->mnt_sb);
  77         if (!inode) {
  78                 ns->ops->put(ns);
  79                 return ERR_PTR(-ENOMEM);
  80         }
  81         inode->i_ino = ns->inum;
  82         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
  83         inode->i_flags |= S_IMMUTABLE;
  84         inode->i_mode = S_IFREG | S_IRUGO;
  85         inode->i_fop = &ns_file_operations;
  86         inode->i_private = ns;
  87
  88         dentry = d_alloc_pseudo(mnt->mnt_sb, &qname);
  89         if (!dentry) {
  90                 iput(inode);
  91                 return ERR_PTR(-ENOMEM);
  92         }
  93         d_instantiate(dentry, inode);
  94         dentry->d_flags |= DCACHE_RCUACCESS;
  95         dentry->d_fsdata = (void *)ns->ops;
  96         d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry);
  97         if (d) {
  98                 d_delete(dentry);       /* make sure ->d_prune() does nothing */
  99                 dput(dentry);
 100                 cpu_relax();
 101                 return ERR_PTR(-EAGAIN);
 102         }
 103         goto got_it;
 104 }
 105
 106 void *ns_get_path(struct path *path, struct task_struct *task,
 107                         const struct proc_ns_operations *ns_ops)
 108 {
 109         struct ns_common *ns;
 110         void *ret;
 111
 112 again:
 113         ns = ns_ops->get(task);
 114         if (!ns)
 115                 return ERR_PTR(-ENOENT);
 116
 117         ret = __ns_get_path(path, ns);
 118         if (IS_ERR(ret) && PTR_ERR(ret) == -EAGAIN)
 119                 goto again;
 120         return ret;
 121 }
 122
 123 int open_related_ns(struct ns_common *ns,
 124                    struct ns_common *(*get_ns)(struct ns_common *ns))
 125 {
 126         struct path path = {};
 127         struct file *f;
 128         void *err;
 129         int fd;
 130
 131         fd = get_unused_fd_flags(O_CLOEXEC);
 132         if (fd < 0)
 133                 return fd;
 134
 135         while (1) {
 136                 struct ns_common *relative;
 137
 138                 relative = get_ns(ns);
 139                 if (IS_ERR(relative)) {
 140                         put_unused_fd(fd);
 141                         return PTR_ERR(relative);
 142                 }
 143
 144                 err = __ns_get_path(&path, relative);
 145                 if (IS_ERR(err) && PTR_ERR(err) == -EAGAIN)
 146                         continue;
 147                 break;
 148         }
 149         if (IS_ERR(err)) {
 150                 put_unused_fd(fd);
 151                 return PTR_ERR(err);
 152         }
 153
 154         f = dentry_open(&path, O_RDONLY, current_cred());
 155         path_put(&path);
 156         if (IS_ERR(f)) {
 157                 put_unused_fd(fd);
 158                 fd = PTR_ERR(f);
 159         } else
 160                 fd_install(fd, f);
 161
 162         return fd;
 163 }
 164
 165 static long ns_ioctl(struct file *filp, unsigned int ioctl,
 166                         unsigned long arg)
 167 {
 168         struct user_namespace *user_ns;
 169         struct ns_common *ns = get_proc_ns(file_inode(filp));
 170         uid_t __user *argp;
 171         uid_t uid;
 172
 173         switch (ioctl) {
 174         case NS_GET_USERNS:
 175                 return open_related_ns(ns, ns_get_owner);
 176         case NS_GET_PARENT:
 177                 if (!ns->ops->get_parent)
 178                         return -EINVAL;
 179                 return open_related_ns(ns, ns->ops->get_parent);
 180         case NS_GET_NSTYPE:
 181                 return ns->ops->type;
 182         case NS_GET_OWNER_UID:
 183                 if (ns->ops->type != CLONE_NEWUSER)
 184                         return -EINVAL;
 185                 user_ns = container_of(ns, struct user_namespace, ns);
 186                 argp = (uid_t __user *) arg;
 187                 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
 188                 return put_user(uid, argp);
 189         default:
 190                 return -ENOTTY;
 191         }
 192 }
 193
 194 int ns_get_name(char *buf, size_t size, struct task_struct *task,
 195                         const struct proc_ns_operations *ns_ops)
 196 {
 197         struct ns_common *ns;
 198         int res = -ENOENT;
 199         ns = ns_ops->get(task);
 200         if (ns) {
 201                 res = snprintf(buf, size, "%s:[%u]", ns_ops->name, ns->inum);
 202                 ns_ops->put(ns);
 203         }
 204         return res;
 205 }
 206
 207 struct file *proc_ns_fget(int fd)
 208 {
 209         struct file *file;
 210
 211         file = fget(fd);
 212         if (!file)
 213                 return ERR_PTR(-EBADF);
 214
 215         if (file->f_op != &ns_file_operations)
 216                 goto out_invalid;
 217
 218         return file;
 219
 220 out_invalid:
 221         fput(file);
 222         return ERR_PTR(-EINVAL);
 223 }
 224
 225 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
 226 {
 227         struct inode *inode = d_inode(dentry);
 228         const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
 229
 230         seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
 231         return 0;
 232 }
 233
 234 static const struct super_operations nsfs_ops = {
 235         .statfs = simple_statfs,
 236         .evict_inode = nsfs_evict,
 237         .show_path = nsfs_show_path,
 238 };
 239 static struct dentry *nsfs_mount(struct file_system_type *fs_type,
 240                         int flags, const char *dev_name, void *data)
 241 {
 242         return mount_pseudo(fs_type, "nsfs:", &nsfs_ops,
 243                         &ns_dentry_operations, NSFS_MAGIC);
 244 }
 245 static struct file_system_type nsfs = {
 246         .name = "nsfs",
 247         .mount = nsfs_mount,
 248         .kill_sb = kill_anon_super,
 249 };
 250
 251 void __init nsfs_init(void)
 252 {
 253         nsfs_mnt = kern_mount(&nsfs);
 254         if (IS_ERR(nsfs_mnt))
 255                 panic("can't set nsfs up\n");
 256         nsfs_mnt->mnt_sb->s_flags &= ~MS_NOUSER;
 257 }