1 /* 2 * Pid namespaces 3 * 4 * Authors: 5 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc. 6 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM 7 * Many thanks to Oleg Nesterov for comments and help 8 * 9 */ 10 11 #include <linux/pid.h> 12 #include <linux/pid_namespace.h> 13 #include <linux/user_namespace.h> 14 #include <linux/syscalls.h> 15 #include <linux/err.h> 16 #include <linux/acct.h> 17 #include <linux/slab.h> 18 #include <linux/proc_fs.h> 19 #include <linux/reboot.h> 20 #include <linux/export.h> 21 22 #define BITS_PER_PAGE (PAGE_SIZE*8) 23 24 struct pid_cache { 25 int nr_ids; 26 char name[16]; 27 struct kmem_cache *cachep; 28 struct list_head list; 29 }; 30 31 static LIST_HEAD(pid_caches_lh); 32 static DEFINE_MUTEX(pid_caches_mutex); 33 static struct kmem_cache *pid_ns_cachep; 34 35 /* 36 * creates the kmem cache to allocate pids from. 37 * @nr_ids: the number of numerical ids this pid will have to carry 38 */ 39 40 static struct kmem_cache *create_pid_cachep(int nr_ids) 41 { 42 struct pid_cache *pcache; 43 struct kmem_cache *cachep; 44 45 mutex_lock(&pid_caches_mutex); 46 list_for_each_entry(pcache, &pid_caches_lh, list) 47 if (pcache->nr_ids == nr_ids) 48 goto out; 49 50 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); 51 if (pcache == NULL) 52 goto err_alloc; 53 54 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); 55 cachep = kmem_cache_create(pcache->name, 56 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid), 57 0, SLAB_HWCACHE_ALIGN, NULL); 58 if (cachep == NULL) 59 goto err_cachep; 60 61 pcache->nr_ids = nr_ids; 62 pcache->cachep = cachep; 63 list_add(&pcache->list, &pid_caches_lh); 64 out: 65 mutex_unlock(&pid_caches_mutex); 66 return pcache->cachep; 67 68 err_cachep: 69 kfree(pcache); 70 err_alloc: 71 mutex_unlock(&pid_caches_mutex); 72 return NULL; 73 } 74 75 static void proc_cleanup_work(struct work_struct *work) 76 { 77 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); 78 pid_ns_release_proc(ns); 79 } 80 81 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ 82 #define MAX_PID_NS_LEVEL 32 83 84 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, 85 struct pid_namespace *parent_pid_ns) 86 { 87 struct pid_namespace *ns; 88 unsigned int level = parent_pid_ns->level + 1; 89 int i; 90 int err; 91 92 if (level > MAX_PID_NS_LEVEL) { 93 err = -EINVAL; 94 goto out; 95 } 96 97 err = -ENOMEM; 98 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); 99 if (ns == NULL) 100 goto out; 101 102 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); 103 if (!ns->pidmap[0].page) 104 goto out_free; 105 106 ns->pid_cachep = create_pid_cachep(level + 1); 107 if (ns->pid_cachep == NULL) 108 goto out_free_map; 109 110 err = proc_alloc_inum(&ns->proc_inum); 111 if (err) 112 goto out_free_map; 113 114 kref_init(&ns->kref); 115 ns->level = level; 116 ns->parent = get_pid_ns(parent_pid_ns); 117 ns->user_ns = get_user_ns(user_ns); 118 INIT_WORK(&ns->proc_work, proc_cleanup_work); 119 120 set_bit(0, ns->pidmap[0].page); 121 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); 122 123 for (i = 1; i < PIDMAP_ENTRIES; i++) 124 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); 125 126 return ns; 127 128 out_free_map: 129 kfree(ns->pidmap[0].page); 130 out_free: 131 kmem_cache_free(pid_ns_cachep, ns); 132 out: 133 return ERR_PTR(err); 134 } 135 136 static void destroy_pid_namespace(struct pid_namespace *ns) 137 { 138 int i; 139 140 proc_free_inum(ns->proc_inum); 141 for (i = 0; i < PIDMAP_ENTRIES; i++) 142 kfree(ns->pidmap[i].page); 143 put_user_ns(ns->user_ns); 144 kmem_cache_free(pid_ns_cachep, ns); 145 } 146 147 struct pid_namespace *copy_pid_ns(unsigned long flags, 148 struct user_namespace *user_ns, struct pid_namespace *old_ns) 149 { 150 if (!(flags & CLONE_NEWPID)) 151 return get_pid_ns(old_ns); 152 if (task_active_pid_ns(current) != old_ns) 153 return ERR_PTR(-EINVAL); 154 return create_pid_namespace(user_ns, old_ns); 155 } 156 157 static void free_pid_ns(struct kref *kref) 158 { 159 struct pid_namespace *ns; 160 161 ns = container_of(kref, struct pid_namespace, kref); 162 destroy_pid_namespace(ns); 163 } 164 165 void put_pid_ns(struct pid_namespace *ns) 166 { 167 struct pid_namespace *parent; 168 169 while (ns != &init_pid_ns) { 170 parent = ns->parent; 171 if (!kref_put(&ns->kref, free_pid_ns)) 172 break; 173 ns = parent; 174 } 175 } 176 EXPORT_SYMBOL_GPL(put_pid_ns); 177 178 void zap_pid_ns_processes(struct pid_namespace *pid_ns) 179 { 180 int nr; 181 int rc; 182 struct task_struct *task, *me = current; 183 184 /* Ignore SIGCHLD causing any terminated children to autoreap */ 185 spin_lock_irq(&me->sighand->siglock); 186 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; 187 spin_unlock_irq(&me->sighand->siglock); 188 189 /* 190 * The last thread in the cgroup-init thread group is terminating. 191 * Find remaining pid_ts in the namespace, signal and wait for them 192 * to exit. 193 * 194 * Note: This signals each threads in the namespace - even those that 195 * belong to the same thread group, To avoid this, we would have 196 * to walk the entire tasklist looking a processes in this 197 * namespace, but that could be unnecessarily expensive if the 198 * pid namespace has just a few processes. Or we need to 199 * maintain a tasklist for each pid namespace. 200 * 201 */ 202 read_lock(&tasklist_lock); 203 nr = next_pidmap(pid_ns, 1); 204 while (nr > 0) { 205 rcu_read_lock(); 206 207 task = pid_task(find_vpid(nr), PIDTYPE_PID); 208 if (task && !__fatal_signal_pending(task)) 209 send_sig_info(SIGKILL, SEND_SIG_FORCED, task); 210 211 rcu_read_unlock(); 212 213 nr = next_pidmap(pid_ns, nr); 214 } 215 read_unlock(&tasklist_lock); 216 217 /* Firstly reap the EXIT_ZOMBIE children we may have. */ 218 do { 219 clear_thread_flag(TIF_SIGPENDING); 220 rc = sys_wait4(-1, NULL, __WALL, NULL); 221 } while (rc != -ECHILD); 222 223 /* 224 * sys_wait4() above can't reap the TASK_DEAD children. 225 * Make sure they all go away, see free_pid(). 226 */ 227 for (;;) { 228 set_current_state(TASK_UNINTERRUPTIBLE); 229 if (pid_ns->nr_hashed == 1) 230 break; 231 schedule(); 232 } 233 __set_current_state(TASK_RUNNING); 234 235 if (pid_ns->reboot) 236 current->signal->group_exit_code = pid_ns->reboot; 237 238 acct_exit_ns(pid_ns); 239 return; 240 } 241 242 #ifdef CONFIG_CHECKPOINT_RESTORE 243 static int pid_ns_ctl_handler(struct ctl_table *table, int write, 244 void __user *buffer, size_t *lenp, loff_t *ppos) 245 { 246 struct pid_namespace *pid_ns = task_active_pid_ns(current); 247 struct ctl_table tmp = *table; 248 249 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) 250 return -EPERM; 251 252 /* 253 * Writing directly to ns' last_pid field is OK, since this field 254 * is volatile in a living namespace anyway and a code writing to 255 * it should synchronize its usage with external means. 256 */ 257 258 tmp.data = &pid_ns->last_pid; 259 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 260 } 261 262 extern int pid_max; 263 static int zero = 0; 264 static struct ctl_table pid_ns_ctl_table[] = { 265 { 266 .procname = "ns_last_pid", 267 .maxlen = sizeof(int), 268 .mode = 0666, /* permissions are checked in the handler */ 269 .proc_handler = pid_ns_ctl_handler, 270 .extra1 = &zero, 271 .extra2 = &pid_max, 272 }, 273 { } 274 }; 275 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; 276 #endif /* CONFIG_CHECKPOINT_RESTORE */ 277 278 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) 279 { 280 if (pid_ns == &init_pid_ns) 281 return 0; 282 283 switch (cmd) { 284 case LINUX_REBOOT_CMD_RESTART2: 285 case LINUX_REBOOT_CMD_RESTART: 286 pid_ns->reboot = SIGHUP; 287 break; 288 289 case LINUX_REBOOT_CMD_POWER_OFF: 290 case LINUX_REBOOT_CMD_HALT: 291 pid_ns->reboot = SIGINT; 292 break; 293 default: 294 return -EINVAL; 295 } 296 297 read_lock(&tasklist_lock); 298 force_sig(SIGKILL, pid_ns->child_reaper); 299 read_unlock(&tasklist_lock); 300 301 do_exit(0); 302 303 /* Not reached */ 304 return 0; 305 } 306 307 static void *pidns_get(struct task_struct *task) 308 { 309 struct pid_namespace *ns; 310 311 rcu_read_lock(); 312 ns = get_pid_ns(task_active_pid_ns(task)); 313 rcu_read_unlock(); 314 315 return ns; 316 } 317 318 static void pidns_put(void *ns) 319 { 320 put_pid_ns(ns); 321 } 322 323 static int pidns_install(struct nsproxy *nsproxy, void *ns) 324 { 325 struct pid_namespace *active = task_active_pid_ns(current); 326 struct pid_namespace *ancestor, *new = ns; 327 328 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || 329 !nsown_capable(CAP_SYS_ADMIN)) 330 return -EPERM; 331 332 /* 333 * Only allow entering the current active pid namespace 334 * or a child of the current active pid namespace. 335 * 336 * This is required for fork to return a usable pid value and 337 * this maintains the property that processes and their 338 * children can not escape their current pid namespace. 339 */ 340 if (new->level < active->level) 341 return -EINVAL; 342 343 ancestor = new; 344 while (ancestor->level > active->level) 345 ancestor = ancestor->parent; 346 if (ancestor != active) 347 return -EINVAL; 348 349 put_pid_ns(nsproxy->pid_ns); 350 nsproxy->pid_ns = get_pid_ns(new); 351 return 0; 352 } 353 354 static unsigned int pidns_inum(void *ns) 355 { 356 struct pid_namespace *pid_ns = ns; 357 return pid_ns->proc_inum; 358 } 359 360 const struct proc_ns_operations pidns_operations = { 361 .name = "pid", 362 .type = CLONE_NEWPID, 363 .get = pidns_get, 364 .put = pidns_put, 365 .install = pidns_install, 366 .inum = pidns_inum, 367 }; 368 369 static __init int pid_namespaces_init(void) 370 { 371 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); 372 373 #ifdef CONFIG_CHECKPOINT_RESTORE 374 register_sysctl_paths(kern_path, pid_ns_ctl_table); 375 #endif 376 return 0; 377 } 378 379 __initcall(pid_namespaces_init); 380