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/syscalls.h>
  14 #include <linux/err.h>
  15 #include <linux/acct.h>
  16 #include <linux/slab.h>
  17 #include <linux/proc_fs.h>
  18 #include <linux/reboot.h>
  19 #include <linux/export.h>
  20
  21 #define BITS_PER_PAGE           (PAGE_SIZE*8)
  22
  23 struct pid_cache {
  24         int nr_ids;
  25         char name[16];
  26         struct kmem_cache *cachep;
  27         struct list_head list;
  28 };
  29
  30 static LIST_HEAD(pid_caches_lh);
  31 static DEFINE_MUTEX(pid_caches_mutex);
  32 static struct kmem_cache *pid_ns_cachep;
  33
  34 /*
  35  * creates the kmem cache to allocate pids from.
  36  * @nr_ids: the number of numerical ids this pid will have to carry
  37  */
  38
  39 static struct kmem_cache *create_pid_cachep(int nr_ids)
  40 {
  41         struct pid_cache *pcache;
  42         struct kmem_cache *cachep;
  43
  44         mutex_lock(&pid_caches_mutex);
  45         list_for_each_entry(pcache, &pid_caches_lh, list)
  46                 if (pcache->nr_ids == nr_ids)
  47                         goto out;
  48
  49         pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
  50         if (pcache == NULL)
  51                 goto err_alloc;
  52
  53         snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
  54         cachep = kmem_cache_create(pcache->name,
  55                         sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
  56                         0, SLAB_HWCACHE_ALIGN, NULL);
  57         if (cachep == NULL)
  58                 goto err_cachep;
  59
  60         pcache->nr_ids = nr_ids;
  61         pcache->cachep = cachep;
  62         list_add(&pcache->list, &pid_caches_lh);
  63 out:
  64         mutex_unlock(&pid_caches_mutex);
  65         return pcache->cachep;
  66
  67 err_cachep:
  68         kfree(pcache);
  69 err_alloc:
  70         mutex_unlock(&pid_caches_mutex);
  71         return NULL;
  72 }
  73
  74 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
  75 #define MAX_PID_NS_LEVEL 32
  76
  77 static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
  78 {
  79         struct pid_namespace *ns;
  80         unsigned int level = parent_pid_ns->level + 1;
  81         int i;
  82         int err;
  83
  84         if (level > MAX_PID_NS_LEVEL) {
  85                 err = -EINVAL;
  86                 goto out;
  87         }
  88
  89         err = -ENOMEM;
  90         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
  91         if (ns == NULL)
  92                 goto out;
  93
  94         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
  95         if (!ns->pidmap[0].page)
  96                 goto out_free;
  97
  98         ns->pid_cachep = create_pid_cachep(level + 1);
  99         if (ns->pid_cachep == NULL)
 100                 goto out_free_map;
 101
 102         kref_init(&ns->kref);
 103         ns->level = level;
 104         ns->parent = get_pid_ns(parent_pid_ns);
 105
 106         set_bit(0, ns->pidmap[0].page);
 107         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 108
 109         for (i = 1; i < PIDMAP_ENTRIES; i++)
 110                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 111
 112         err = pid_ns_prepare_proc(ns);
 113         if (err)
 114                 goto out_put_parent_pid_ns;
 115
 116         return ns;
 117
 118 out_put_parent_pid_ns:
 119         put_pid_ns(parent_pid_ns);
 120 out_free_map:
 121         kfree(ns->pidmap[0].page);
 122 out_free:
 123         kmem_cache_free(pid_ns_cachep, ns);
 124 out:
 125         return ERR_PTR(err);
 126 }
 127
 128 static void destroy_pid_namespace(struct pid_namespace *ns)
 129 {
 130         int i;
 131
 132         for (i = 0; i < PIDMAP_ENTRIES; i++)
 133                 kfree(ns->pidmap[i].page);
 134         kmem_cache_free(pid_ns_cachep, ns);
 135 }
 136
 137 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
 138 {
 139         if (!(flags & CLONE_NEWPID))
 140                 return get_pid_ns(old_ns);
 141         if (flags & (CLONE_THREAD|CLONE_PARENT))
 142                 return ERR_PTR(-EINVAL);
 143         return create_pid_namespace(old_ns);
 144 }
 145
 146 static void free_pid_ns(struct kref *kref)
 147 {
 148         struct pid_namespace *ns;
 149
 150         ns = container_of(kref, struct pid_namespace, kref);
 151         destroy_pid_namespace(ns);
 152 }
 153
 154 void put_pid_ns(struct pid_namespace *ns)
 155 {
 156         struct pid_namespace *parent;
 157
 158         while (ns != &init_pid_ns) {
 159                 parent = ns->parent;
 160                 if (!kref_put(&ns->kref, free_pid_ns))
 161                         break;
 162                 ns = parent;
 163         }
 164 }
 165 EXPORT_SYMBOL_GPL(put_pid_ns);
 166
 167 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 168 {
 169         int nr;
 170         int rc;
 171         struct task_struct *task, *me = current;
 172
 173         /* Ignore SIGCHLD causing any terminated children to autoreap */
 174         spin_lock_irq(&me->sighand->siglock);
 175         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 176         spin_unlock_irq(&me->sighand->siglock);
 177
 178         /*
 179          * The last thread in the cgroup-init thread group is terminating.
 180          * Find remaining pid_ts in the namespace, signal and wait for them
 181          * to exit.
 182          *
 183          * Note:  This signals each threads in the namespace - even those that
 184          *        belong to the same thread group, To avoid this, we would have
 185          *        to walk the entire tasklist looking a processes in this
 186          *        namespace, but that could be unnecessarily expensive if the
 187          *        pid namespace has just a few processes. Or we need to
 188          *        maintain a tasklist for each pid namespace.
 189          *
 190          */
 191         read_lock(&tasklist_lock);
 192         nr = next_pidmap(pid_ns, 1);
 193         while (nr > 0) {
 194                 rcu_read_lock();
 195
 196                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 197                 if (task && !__fatal_signal_pending(task))
 198                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 199
 200                 rcu_read_unlock();
 201
 202                 nr = next_pidmap(pid_ns, nr);
 203         }
 204         read_unlock(&tasklist_lock);
 205
 206         /* Firstly reap the EXIT_ZOMBIE children we may have. */
 207         do {
 208                 clear_thread_flag(TIF_SIGPENDING);
 209                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 210         } while (rc != -ECHILD);
 211
 212         /*
 213          * sys_wait4() above can't reap the TASK_DEAD children.
 214          * Make sure they all go away, see __unhash_process().
 215          */
 216         for (;;) {
 217                 bool need_wait = false;
 218
 219                 read_lock(&tasklist_lock);
 220                 if (!list_empty(&current->children)) {
 221                         __set_current_state(TASK_UNINTERRUPTIBLE);
 222                         need_wait = true;
 223                 }
 224                 read_unlock(&tasklist_lock);
 225
 226                 if (!need_wait)
 227                         break;
 228                 schedule();
 229         }
 230
 231         if (pid_ns->reboot)
 232                 current->signal->group_exit_code = pid_ns->reboot;
 233
 234         acct_exit_ns(pid_ns);
 235         return;
 236 }
 237
 238 #ifdef CONFIG_CHECKPOINT_RESTORE
 239 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 240                 void __user *buffer, size_t *lenp, loff_t *ppos)
 241 {
 242         struct ctl_table tmp = *table;
 243
 244         if (write && !capable(CAP_SYS_ADMIN))
 245                 return -EPERM;
 246
 247         /*
 248          * Writing directly to ns' last_pid field is OK, since this field
 249          * is volatile in a living namespace anyway and a code writing to
 250          * it should synchronize its usage with external means.
 251          */
 252
 253         tmp.data = &current->nsproxy->pid_ns->last_pid;
 254         return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 255 }
 256
 257 extern int pid_max;
 258 static int zero = 0;
 259 static struct ctl_table pid_ns_ctl_table[] = {
 260         {
 261                 .procname = "ns_last_pid",
 262                 .maxlen = sizeof(int),
 263                 .mode = 0666, /* permissions are checked in the handler */
 264                 .proc_handler = pid_ns_ctl_handler,
 265                 .extra1 = &zero,
 266                 .extra2 = &pid_max,
 267         },
 268         { }
 269 };
 270 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 271 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 272
 273 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 274 {
 275         if (pid_ns == &init_pid_ns)
 276                 return 0;
 277
 278         switch (cmd) {
 279         case LINUX_REBOOT_CMD_RESTART2:
 280         case LINUX_REBOOT_CMD_RESTART:
 281                 pid_ns->reboot = SIGHUP;
 282                 break;
 283
 284         case LINUX_REBOOT_CMD_POWER_OFF:
 285         case LINUX_REBOOT_CMD_HALT:
 286                 pid_ns->reboot = SIGINT;
 287                 break;
 288         default:
 289                 return -EINVAL;
 290         }
 291
 292         read_lock(&tasklist_lock);
 293         force_sig(SIGKILL, pid_ns->child_reaper);
 294         read_unlock(&tasklist_lock);
 295
 296         do_exit(0);
 297
 298         /* Not reached */
 299         return 0;
 300 }
 301
 302 static __init int pid_namespaces_init(void)
 303 {
 304         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 305
 306 #ifdef CONFIG_CHECKPOINT_RESTORE
 307         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 308 #endif
 309         return 0;
 310 }
 311
 312 __initcall(pid_namespaces_init);