kernel/pid_namespace.c

   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_ns.h>
  19 #include <linux/reboot.h>
  20 #include <linux/export.h>
  21
  22 struct pid_cache {
  23         int nr_ids;
  24         char name[16];
  25         struct kmem_cache *cachep;
  26         struct list_head list;
  27 };
  28
  29 static LIST_HEAD(pid_caches_lh);
  30 static DEFINE_MUTEX(pid_caches_mutex);
  31 static struct kmem_cache *pid_ns_cachep;
  32
  33 /*
  34  * creates the kmem cache to allocate pids from.
  35  * @nr_ids: the number of numerical ids this pid will have to carry
  36  */
  37
  38 static struct kmem_cache *create_pid_cachep(int nr_ids)
  39 {
  40         struct pid_cache *pcache;
  41         struct kmem_cache *cachep;
  42
  43         mutex_lock(&pid_caches_mutex);
  44         list_for_each_entry(pcache, &pid_caches_lh, list)
  45                 if (pcache->nr_ids == nr_ids)
  46                         goto out;
  47
  48         pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
  49         if (pcache == NULL)
  50                 goto err_alloc;
  51
  52         snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
  53         cachep = kmem_cache_create(pcache->name,
  54                         sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
  55                         0, SLAB_HWCACHE_ALIGN, NULL);
  56         if (cachep == NULL)
  57                 goto err_cachep;
  58
  59         pcache->nr_ids = nr_ids;
  60         pcache->cachep = cachep;
  61         list_add(&pcache->list, &pid_caches_lh);
  62 out:
  63         mutex_unlock(&pid_caches_mutex);
  64         return pcache->cachep;
  65
  66 err_cachep:
  67         kfree(pcache);
  68 err_alloc:
  69         mutex_unlock(&pid_caches_mutex);
  70         return NULL;
  71 }
  72
  73 static void proc_cleanup_work(struct work_struct *work)
  74 {
  75         struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
  76         pid_ns_release_proc(ns);
  77 }
  78
  79 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
  80 #define MAX_PID_NS_LEVEL 32
  81
  82 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
  83         struct pid_namespace *parent_pid_ns)
  84 {
  85         struct pid_namespace *ns;
  86         unsigned int level = parent_pid_ns->level + 1;
  87         int i;
  88         int err;
  89
  90         if (level > MAX_PID_NS_LEVEL) {
  91                 err = -EINVAL;
  92                 goto out;
  93         }
  94
  95         err = -ENOMEM;
  96         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
  97         if (ns == NULL)
  98                 goto out;
  99
 100         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 101         if (!ns->pidmap[0].page)
 102                 goto out_free;
 103
 104         ns->pid_cachep = create_pid_cachep(level + 1);
 105         if (ns->pid_cachep == NULL)
 106                 goto out_free_map;
 107
 108         err = proc_alloc_inum(&ns->proc_inum);
 109         if (err)
 110                 goto out_free_map;
 111
 112         kref_init(&ns->kref);
 113         ns->level = level;
 114         ns->parent = get_pid_ns(parent_pid_ns);
 115         ns->user_ns = get_user_ns(user_ns);
 116         ns->nr_hashed = PIDNS_HASH_ADDING;
 117         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 118
 119         set_bit(0, ns->pidmap[0].page);
 120         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 121
 122         for (i = 1; i < PIDMAP_ENTRIES; i++)
 123                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 124
 125         return ns;
 126
 127 out_free_map:
 128         kfree(ns->pidmap[0].page);
 129 out_free:
 130         kmem_cache_free(pid_ns_cachep, ns);
 131 out:
 132         return ERR_PTR(err);
 133 }
 134
 135 static void destroy_pid_namespace(struct pid_namespace *ns)
 136 {
 137         int i;
 138
 139         proc_free_inum(ns->proc_inum);
 140         for (i = 0; i < PIDMAP_ENTRIES; i++)
 141                 kfree(ns->pidmap[i].page);
 142         put_user_ns(ns->user_ns);
 143         kmem_cache_free(pid_ns_cachep, ns);
 144 }
 145
 146 struct pid_namespace *copy_pid_ns(unsigned long flags,
 147         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 148 {
 149         if (!(flags & CLONE_NEWPID))
 150                 return get_pid_ns(old_ns);
 151         if (task_active_pid_ns(current) != old_ns)
 152                 return ERR_PTR(-EINVAL);
 153         return create_pid_namespace(user_ns, old_ns);
 154 }
 155
 156 static void free_pid_ns(struct kref *kref)
 157 {
 158         struct pid_namespace *ns;
 159
 160         ns = container_of(kref, struct pid_namespace, kref);
 161         destroy_pid_namespace(ns);
 162 }
 163
 164 void put_pid_ns(struct pid_namespace *ns)
 165 {
 166         struct pid_namespace *parent;
 167
 168         while (ns != &init_pid_ns) {
 169                 parent = ns->parent;
 170                 if (!kref_put(&ns->kref, free_pid_ns))
 171                         break;
 172                 ns = parent;
 173         }
 174 }
 175 EXPORT_SYMBOL_GPL(put_pid_ns);
 176
 177 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 178 {
 179         int nr;
 180         int rc;
 181         struct task_struct *task, *me = current;
 182         int init_pids = thread_group_leader(me) ? 1 : 2;
 183
 184         /* Don't allow any more processes into the pid namespace */
 185         disable_pid_allocation(pid_ns);
 186
 187         /* Ignore SIGCHLD causing any terminated children to autoreap */
 188         spin_lock_irq(&me->sighand->siglock);
 189         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 190         spin_unlock_irq(&me->sighand->siglock);
 191
 192         /*
 193          * The last thread in the cgroup-init thread group is terminating.
 194          * Find remaining pid_ts in the namespace, signal and wait for them
 195          * to exit.
 196          *
 197          * Note:  This signals each threads in the namespace - even those that
 198          *        belong to the same thread group, To avoid this, we would have
 199          *        to walk the entire tasklist looking a processes in this
 200          *        namespace, but that could be unnecessarily expensive if the
 201          *        pid namespace has just a few processes. Or we need to
 202          *        maintain a tasklist for each pid namespace.
 203          *
 204          */
 205         read_lock(&tasklist_lock);
 206         nr = next_pidmap(pid_ns, 1);
 207         while (nr > 0) {
 208                 rcu_read_lock();
 209
 210                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 211                 if (task && !__fatal_signal_pending(task))
 212                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 213
 214                 rcu_read_unlock();
 215
 216                 nr = next_pidmap(pid_ns, nr);
 217         }
 218         read_unlock(&tasklist_lock);
 219
 220         /* Firstly reap the EXIT_ZOMBIE children we may have. */
 221         do {
 222                 clear_thread_flag(TIF_SIGPENDING);
 223                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 224         } while (rc != -ECHILD);
 225
 226         /*
 227          * sys_wait4() above can't reap the TASK_DEAD children.
 228          * Make sure they all go away, see free_pid().
 229          */
 230         for (;;) {
 231                 set_current_state(TASK_UNINTERRUPTIBLE);
 232                 if (pid_ns->nr_hashed == init_pids)
 233                         break;
 234                 schedule();
 235         }
 236         __set_current_state(TASK_RUNNING);
 237
 238         if (pid_ns->reboot)
 239                 current->signal->group_exit_code = pid_ns->reboot;
 240
 241         acct_exit_ns(pid_ns);
 242         return;
 243 }
 244
 245 #ifdef CONFIG_CHECKPOINT_RESTORE
 246 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 247                 void __user *buffer, size_t *lenp, loff_t *ppos)
 248 {
 249         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 250         struct ctl_table tmp = *table;
 251
 252         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 253                 return -EPERM;
 254
 255         /*
 256          * Writing directly to ns' last_pid field is OK, since this field
 257          * is volatile in a living namespace anyway and a code writing to
 258          * it should synchronize its usage with external means.
 259          */
 260
 261         tmp.data = &pid_ns->last_pid;
 262         return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 263 }
 264
 265 extern int pid_max;
 266 static int zero = 0;
 267 static struct ctl_table pid_ns_ctl_table[] = {
 268         {
 269                 .procname = "ns_last_pid",
 270                 .maxlen = sizeof(int),
 271                 .mode = 0666, /* permissions are checked in the handler */
 272                 .proc_handler = pid_ns_ctl_handler,
 273                 .extra1 = &zero,
 274                 .extra2 = &pid_max,
 275         },
 276         { }
 277 };
 278 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 279 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 280
 281 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 282 {
 283         if (pid_ns == &init_pid_ns)
 284                 return 0;
 285
 286         switch (cmd) {
 287         case LINUX_REBOOT_CMD_RESTART2:
 288         case LINUX_REBOOT_CMD_RESTART:
 289                 pid_ns->reboot = SIGHUP;
 290                 break;
 291
 292         case LINUX_REBOOT_CMD_POWER_OFF:
 293         case LINUX_REBOOT_CMD_HALT:
 294                 pid_ns->reboot = SIGINT;
 295                 break;
 296         default:
 297                 return -EINVAL;
 298         }
 299
 300         read_lock(&tasklist_lock);
 301         force_sig(SIGKILL, pid_ns->child_reaper);
 302         read_unlock(&tasklist_lock);
 303
 304         do_exit(0);
 305
 306         /* Not reached */
 307         return 0;
 308 }
 309
 310 static void *pidns_get(struct task_struct *task)
 311 {
 312         struct pid_namespace *ns;
 313
 314         rcu_read_lock();
 315         ns = get_pid_ns(task_active_pid_ns(task));
 316         rcu_read_unlock();
 317
 318         return ns;
 319 }
 320
 321 static void pidns_put(void *ns)
 322 {
 323         put_pid_ns(ns);
 324 }
 325
 326 static int pidns_install(struct nsproxy *nsproxy, void *ns)
 327 {
 328         struct pid_namespace *active = task_active_pid_ns(current);
 329         struct pid_namespace *ancestor, *new = ns;
 330
 331         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 332             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 333                 return -EPERM;
 334
 335         /*
 336          * Only allow entering the current active pid namespace
 337          * or a child of the current active pid namespace.
 338          *
 339          * This is required for fork to return a usable pid value and
 340          * this maintains the property that processes and their
 341          * children can not escape their current pid namespace.
 342          */
 343         if (new->level < active->level)
 344                 return -EINVAL;
 345
 346         ancestor = new;
 347         while (ancestor->level > active->level)
 348                 ancestor = ancestor->parent;
 349         if (ancestor != active)
 350                 return -EINVAL;
 351
 352         put_pid_ns(nsproxy->pid_ns_for_children);
 353         nsproxy->pid_ns_for_children = get_pid_ns(new);
 354         return 0;
 355 }
 356
 357 static unsigned int pidns_inum(void *ns)
 358 {
 359         struct pid_namespace *pid_ns = ns;
 360         return pid_ns->proc_inum;
 361 }
 362
 363 const struct proc_ns_operations pidns_operations = {
 364         .name           = "pid",
 365         .type           = CLONE_NEWPID,
 366         .get            = pidns_get,
 367         .put            = pidns_put,
 368         .install        = pidns_install,
 369         .inum           = pidns_inum,
 370 };
 371
 372 static __init int pid_namespaces_init(void)
 373 {
 374         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 375
 376 #ifdef CONFIG_CHECKPOINT_RESTORE
 377         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 378 #endif
 379         return 0;
 380 }
 381
 382 __initcall(pid_namespaces_init);