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
  20 #define BITS_PER_PAGE           (PAGE_SIZE*8)
  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 struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
  74 {
  75         struct pid_namespace *ns;
  76         unsigned int level = parent_pid_ns->level + 1;
  77         int i, err = -ENOMEM;
  78
  79         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
  80         if (ns == NULL)
  81                 goto out;
  82
  83         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
  84         if (!ns->pidmap[0].page)
  85                 goto out_free;
  86
  87         ns->pid_cachep = create_pid_cachep(level + 1);
  88         if (ns->pid_cachep == NULL)
  89                 goto out_free_map;
  90
  91         kref_init(&ns->kref);
  92         ns->level = level;
  93         ns->parent = get_pid_ns(parent_pid_ns);
  94
  95         set_bit(0, ns->pidmap[0].page);
  96         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
  97
  98         for (i = 1; i < PIDMAP_ENTRIES; i++)
  99                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 100
 101         err = pid_ns_prepare_proc(ns);
 102         if (err)
 103                 goto out_put_parent_pid_ns;
 104
 105         return ns;
 106
 107 out_put_parent_pid_ns:
 108         put_pid_ns(parent_pid_ns);
 109 out_free_map:
 110         kfree(ns->pidmap[0].page);
 111 out_free:
 112         kmem_cache_free(pid_ns_cachep, ns);
 113 out:
 114         return ERR_PTR(err);
 115 }
 116
 117 static void destroy_pid_namespace(struct pid_namespace *ns)
 118 {
 119         int i;
 120
 121         for (i = 0; i < PIDMAP_ENTRIES; i++)
 122                 kfree(ns->pidmap[i].page);
 123         kmem_cache_free(pid_ns_cachep, ns);
 124 }
 125
 126 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
 127 {
 128         if (!(flags & CLONE_NEWPID))
 129                 return get_pid_ns(old_ns);
 130         if (flags & (CLONE_THREAD|CLONE_PARENT))
 131                 return ERR_PTR(-EINVAL);
 132         return create_pid_namespace(old_ns);
 133 }
 134
 135 void free_pid_ns(struct kref *kref)
 136 {
 137         struct pid_namespace *ns, *parent;
 138
 139         ns = container_of(kref, struct pid_namespace, kref);
 140
 141         parent = ns->parent;
 142         destroy_pid_namespace(ns);
 143
 144         if (parent != NULL)
 145                 put_pid_ns(parent);
 146 }
 147
 148 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 149 {
 150         int nr;
 151         int rc;
 152         struct task_struct *task, *me = current;
 153
 154         /* Ignore SIGCHLD causing any terminated children to autoreap */
 155         spin_lock_irq(&me->sighand->siglock);
 156         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 157         spin_unlock_irq(&me->sighand->siglock);
 158
 159         /*
 160          * The last thread in the cgroup-init thread group is terminating.
 161          * Find remaining pid_ts in the namespace, signal and wait for them
 162          * to exit.
 163          *
 164          * Note:  This signals each threads in the namespace - even those that
 165          *        belong to the same thread group, To avoid this, we would have
 166          *        to walk the entire tasklist looking a processes in this
 167          *        namespace, but that could be unnecessarily expensive if the
 168          *        pid namespace has just a few processes. Or we need to
 169          *        maintain a tasklist for each pid namespace.
 170          *
 171          */
 172         read_lock(&tasklist_lock);
 173         nr = next_pidmap(pid_ns, 1);
 174         while (nr > 0) {
 175                 rcu_read_lock();
 176
 177                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 178                 if (task && !__fatal_signal_pending(task))
 179                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 180
 181                 rcu_read_unlock();
 182
 183                 nr = next_pidmap(pid_ns, nr);
 184         }
 185         read_unlock(&tasklist_lock);
 186
 187         /* Firstly reap the EXIT_ZOMBIE children we may have. */
 188         do {
 189                 clear_thread_flag(TIF_SIGPENDING);
 190                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 191         } while (rc != -ECHILD);
 192
 193         /*
 194          * sys_wait4() above can't reap the TASK_DEAD children.
 195          * Make sure they all go away, see __unhash_process().
 196          */
 197         for (;;) {
 198                 bool need_wait = false;
 199
 200                 read_lock(&tasklist_lock);
 201                 if (!list_empty(&current->children)) {
 202                         __set_current_state(TASK_UNINTERRUPTIBLE);
 203                         need_wait = true;
 204                 }
 205                 read_unlock(&tasklist_lock);
 206
 207                 if (!need_wait)
 208                         break;
 209                 schedule();
 210         }
 211
 212         if (pid_ns->reboot)
 213                 current->signal->group_exit_code = pid_ns->reboot;
 214
 215         acct_exit_ns(pid_ns);
 216         return;
 217 }
 218
 219 #ifdef CONFIG_CHECKPOINT_RESTORE
 220 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 221                 void __user *buffer, size_t *lenp, loff_t *ppos)
 222 {
 223         struct ctl_table tmp = *table;
 224
 225         if (write && !capable(CAP_SYS_ADMIN))
 226                 return -EPERM;
 227
 228         /*
 229          * Writing directly to ns' last_pid field is OK, since this field
 230          * is volatile in a living namespace anyway and a code writing to
 231          * it should synchronize its usage with external means.
 232          */
 233
 234         tmp.data = &current->nsproxy->pid_ns->last_pid;
 235         return proc_dointvec(&tmp, write, buffer, lenp, ppos);
 236 }
 237
 238 static struct ctl_table pid_ns_ctl_table[] = {
 239         {
 240                 .procname = "ns_last_pid",
 241                 .maxlen = sizeof(int),
 242                 .mode = 0666, /* permissions are checked in the handler */
 243                 .proc_handler = pid_ns_ctl_handler,
 244         },
 245         { }
 246 };
 247 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 248 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 249
 250 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 251 {
 252         if (pid_ns == &init_pid_ns)
 253                 return 0;
 254
 255         switch (cmd) {
 256         case LINUX_REBOOT_CMD_RESTART2:
 257         case LINUX_REBOOT_CMD_RESTART:
 258                 pid_ns->reboot = SIGHUP;
 259                 break;
 260
 261         case LINUX_REBOOT_CMD_POWER_OFF:
 262         case LINUX_REBOOT_CMD_HALT:
 263                 pid_ns->reboot = SIGINT;
 264                 break;
 265         default:
 266                 return -EINVAL;
 267         }
 268
 269         read_lock(&tasklist_lock);
 270         force_sig(SIGKILL, pid_ns->child_reaper);
 271         read_unlock(&tasklist_lock);
 272
 273         do_exit(0);
 274
 275         /* Not reached */
 276         return 0;
 277 }
 278
 279 static __init int pid_namespaces_init(void)
 280 {
 281         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 282
 283 #ifdef CONFIG_CHECKPOINT_RESTORE
 284         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 285 #endif
 286         return 0;
 287 }
 288
 289 __initcall(pid_namespaces_init);