1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2006 IBM Corporation 4 * 5 * Author: Serge Hallyn <serue@us.ibm.com> 6 * 7 * Jun 2006 - namespaces support 8 * OpenVZ, SWsoft Inc. 9 * Pavel Emelianov <xemul@openvz.org> 10 */ 11 12 #include <linux/slab.h> 13 #include <linux/export.h> 14 #include <linux/nsproxy.h> 15 #include <linux/init_task.h> 16 #include <linux/mnt_namespace.h> 17 #include <linux/utsname.h> 18 #include <linux/pid_namespace.h> 19 #include <net/net_namespace.h> 20 #include <linux/ipc_namespace.h> 21 #include <linux/time_namespace.h> 22 #include <linux/proc_ns.h> 23 #include <linux/file.h> 24 #include <linux/syscalls.h> 25 #include <linux/cgroup.h> 26 #include <linux/perf_event.h> 27 28 static struct kmem_cache *nsproxy_cachep; 29 30 struct nsproxy init_nsproxy = { 31 .count = ATOMIC_INIT(1), 32 .uts_ns = &init_uts_ns, 33 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC) 34 .ipc_ns = &init_ipc_ns, 35 #endif 36 .mnt_ns = NULL, 37 .pid_ns_for_children = &init_pid_ns, 38 #ifdef CONFIG_NET 39 .net_ns = &init_net, 40 #endif 41 #ifdef CONFIG_CGROUPS 42 .cgroup_ns = &init_cgroup_ns, 43 #endif 44 #ifdef CONFIG_TIME_NS 45 .time_ns = &init_time_ns, 46 .time_ns_for_children = &init_time_ns, 47 #endif 48 }; 49 50 static inline struct nsproxy *create_nsproxy(void) 51 { 52 struct nsproxy *nsproxy; 53 54 nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL); 55 if (nsproxy) 56 atomic_set(&nsproxy->count, 1); 57 return nsproxy; 58 } 59 60 /* 61 * Create new nsproxy and all of its the associated namespaces. 62 * Return the newly created nsproxy. Do not attach this to the task, 63 * leave it to the caller to do proper locking and attach it to task. 64 */ 65 static struct nsproxy *create_new_namespaces(unsigned long flags, 66 struct task_struct *tsk, struct user_namespace *user_ns, 67 struct fs_struct *new_fs) 68 { 69 struct nsproxy *new_nsp; 70 int err; 71 72 new_nsp = create_nsproxy(); 73 if (!new_nsp) 74 return ERR_PTR(-ENOMEM); 75 76 new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs); 77 if (IS_ERR(new_nsp->mnt_ns)) { 78 err = PTR_ERR(new_nsp->mnt_ns); 79 goto out_ns; 80 } 81 82 new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns); 83 if (IS_ERR(new_nsp->uts_ns)) { 84 err = PTR_ERR(new_nsp->uts_ns); 85 goto out_uts; 86 } 87 88 new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns); 89 if (IS_ERR(new_nsp->ipc_ns)) { 90 err = PTR_ERR(new_nsp->ipc_ns); 91 goto out_ipc; 92 } 93 94 new_nsp->pid_ns_for_children = 95 copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children); 96 if (IS_ERR(new_nsp->pid_ns_for_children)) { 97 err = PTR_ERR(new_nsp->pid_ns_for_children); 98 goto out_pid; 99 } 100 101 new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns, 102 tsk->nsproxy->cgroup_ns); 103 if (IS_ERR(new_nsp->cgroup_ns)) { 104 err = PTR_ERR(new_nsp->cgroup_ns); 105 goto out_cgroup; 106 } 107 108 new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns); 109 if (IS_ERR(new_nsp->net_ns)) { 110 err = PTR_ERR(new_nsp->net_ns); 111 goto out_net; 112 } 113 114 new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns, 115 tsk->nsproxy->time_ns_for_children); 116 if (IS_ERR(new_nsp->time_ns_for_children)) { 117 err = PTR_ERR(new_nsp->time_ns_for_children); 118 goto out_time; 119 } 120 new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns); 121 122 return new_nsp; 123 124 out_time: 125 put_net(new_nsp->net_ns); 126 out_net: 127 put_cgroup_ns(new_nsp->cgroup_ns); 128 out_cgroup: 129 if (new_nsp->pid_ns_for_children) 130 put_pid_ns(new_nsp->pid_ns_for_children); 131 out_pid: 132 if (new_nsp->ipc_ns) 133 put_ipc_ns(new_nsp->ipc_ns); 134 out_ipc: 135 if (new_nsp->uts_ns) 136 put_uts_ns(new_nsp->uts_ns); 137 out_uts: 138 if (new_nsp->mnt_ns) 139 put_mnt_ns(new_nsp->mnt_ns); 140 out_ns: 141 kmem_cache_free(nsproxy_cachep, new_nsp); 142 return ERR_PTR(err); 143 } 144 145 /* 146 * called from clone. This now handles copy for nsproxy and all 147 * namespaces therein. 148 */ 149 int copy_namespaces(unsigned long flags, struct task_struct *tsk) 150 { 151 struct nsproxy *old_ns = tsk->nsproxy; 152 struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns); 153 struct nsproxy *new_ns; 154 int ret; 155 156 if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | 157 CLONE_NEWPID | CLONE_NEWNET | 158 CLONE_NEWCGROUP | CLONE_NEWTIME)))) { 159 if (likely(old_ns->time_ns_for_children == old_ns->time_ns)) { 160 get_nsproxy(old_ns); 161 return 0; 162 } 163 } else if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 164 return -EPERM; 165 166 /* 167 * CLONE_NEWIPC must detach from the undolist: after switching 168 * to a new ipc namespace, the semaphore arrays from the old 169 * namespace are unreachable. In clone parlance, CLONE_SYSVSEM 170 * means share undolist with parent, so we must forbid using 171 * it along with CLONE_NEWIPC. 172 */ 173 if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) == 174 (CLONE_NEWIPC | CLONE_SYSVSEM)) 175 return -EINVAL; 176 177 new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs); 178 if (IS_ERR(new_ns)) 179 return PTR_ERR(new_ns); 180 181 ret = timens_on_fork(new_ns, tsk); 182 if (ret) { 183 free_nsproxy(new_ns); 184 return ret; 185 } 186 187 tsk->nsproxy = new_ns; 188 return 0; 189 } 190 191 void free_nsproxy(struct nsproxy *ns) 192 { 193 if (ns->mnt_ns) 194 put_mnt_ns(ns->mnt_ns); 195 if (ns->uts_ns) 196 put_uts_ns(ns->uts_ns); 197 if (ns->ipc_ns) 198 put_ipc_ns(ns->ipc_ns); 199 if (ns->pid_ns_for_children) 200 put_pid_ns(ns->pid_ns_for_children); 201 if (ns->time_ns) 202 put_time_ns(ns->time_ns); 203 if (ns->time_ns_for_children) 204 put_time_ns(ns->time_ns_for_children); 205 put_cgroup_ns(ns->cgroup_ns); 206 put_net(ns->net_ns); 207 kmem_cache_free(nsproxy_cachep, ns); 208 } 209 210 /* 211 * Called from unshare. Unshare all the namespaces part of nsproxy. 212 * On success, returns the new nsproxy. 213 */ 214 int unshare_nsproxy_namespaces(unsigned long unshare_flags, 215 struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs) 216 { 217 struct user_namespace *user_ns; 218 int err = 0; 219 220 if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | 221 CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP | 222 CLONE_NEWTIME))) 223 return 0; 224 225 user_ns = new_cred ? new_cred->user_ns : current_user_ns(); 226 if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 227 return -EPERM; 228 229 *new_nsp = create_new_namespaces(unshare_flags, current, user_ns, 230 new_fs ? new_fs : current->fs); 231 if (IS_ERR(*new_nsp)) { 232 err = PTR_ERR(*new_nsp); 233 goto out; 234 } 235 236 out: 237 return err; 238 } 239 240 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new) 241 { 242 struct nsproxy *ns; 243 244 might_sleep(); 245 246 task_lock(p); 247 ns = p->nsproxy; 248 p->nsproxy = new; 249 task_unlock(p); 250 251 if (ns && atomic_dec_and_test(&ns->count)) 252 free_nsproxy(ns); 253 } 254 255 void exit_task_namespaces(struct task_struct *p) 256 { 257 switch_task_namespaces(p, NULL); 258 } 259 260 SYSCALL_DEFINE2(setns, int, fd, int, nstype) 261 { 262 struct task_struct *tsk = current; 263 struct nsproxy *new_nsproxy; 264 struct file *file; 265 struct ns_common *ns; 266 int err; 267 268 file = proc_ns_fget(fd); 269 if (IS_ERR(file)) 270 return PTR_ERR(file); 271 272 err = -EINVAL; 273 ns = get_proc_ns(file_inode(file)); 274 if (nstype && (ns->ops->type != nstype)) 275 goto out; 276 277 new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs); 278 if (IS_ERR(new_nsproxy)) { 279 err = PTR_ERR(new_nsproxy); 280 goto out; 281 } 282 283 err = ns->ops->install(new_nsproxy, ns); 284 if (err) { 285 free_nsproxy(new_nsproxy); 286 goto out; 287 } 288 switch_task_namespaces(tsk, new_nsproxy); 289 290 perf_event_namespaces(tsk); 291 out: 292 fput(file); 293 return err; 294 } 295 296 int __init nsproxy_cache_init(void) 297 { 298 nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC); 299 return 0; 300 } 301