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_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         kref_init(&ns->kref);
 111         ns->level = level;
 112         ns->parent = get_pid_ns(parent_pid_ns);
 113         ns->user_ns = get_user_ns(user_ns);
 114         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 115
 116         set_bit(0, ns->pidmap[0].page);
 117         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 118
 119         for (i = 1; i < PIDMAP_ENTRIES; i++)
 120                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 121
 122         return ns;
 123
 124 out_free_map:
 125         kfree(ns->pidmap[0].page);
 126 out_free:
 127         kmem_cache_free(pid_ns_cachep, ns);
 128 out:
 129         return ERR_PTR(err);
 130 }
 131
 132 static void destroy_pid_namespace(struct pid_namespace *ns)
 133 {
 134         int i;
 135
 136         for (i = 0; i < PIDMAP_ENTRIES; i++)
 137                 kfree(ns->pidmap[i].page);
 138         put_user_ns(ns->user_ns);
 139         kmem_cache_free(pid_ns_cachep, ns);
 140 }
 141
 142 struct pid_namespace *copy_pid_ns(unsigned long flags,
 143         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 144 {
 145         if (!(flags & CLONE_NEWPID))
 146                 return get_pid_ns(old_ns);
 147         if (task_active_pid_ns(current) != old_ns)
 148                 return ERR_PTR(-EINVAL);
 149         return create_pid_namespace(user_ns, old_ns);
 150 }
 151
 152 static void free_pid_ns(struct kref *kref)
 153 {
 154         struct pid_namespace *ns;
 155
 156         ns = container_of(kref, struct pid_namespace, kref);
 157         destroy_pid_namespace(ns);
 158 }
 159
 160 void put_pid_ns(struct pid_namespace *ns)
 161 {
 162         struct pid_namespace *parent;
 163
 164         while (ns != &init_pid_ns) {
 165                 parent = ns->parent;
 166                 if (!kref_put(&ns->kref, free_pid_ns))
 167                         break;
 168                 ns = parent;
 169         }
 170 }
 171 EXPORT_SYMBOL_GPL(put_pid_ns);
 172
 173 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 174 {
 175         int nr;
 176         int rc;
 177         struct task_struct *task, *me = current;
 178
 179         /* Ignore SIGCHLD causing any terminated children to autoreap */
 180         spin_lock_irq(&me->sighand->siglock);
 181         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 182         spin_unlock_irq(&me->sighand->siglock);
 183
 184         /*
 185          * The last thread in the cgroup-init thread group is terminating.
 186          * Find remaining pid_ts in the namespace, signal and wait for them
 187          * to exit.
 188          *
 189          * Note:  This signals each threads in the namespace - even those that
 190          *        belong to the same thread group, To avoid this, we would have
 191          *        to walk the entire tasklist looking a processes in this
 192          *        namespace, but that could be unnecessarily expensive if the
 193          *        pid namespace has just a few processes. Or we need to
 194          *        maintain a tasklist for each pid namespace.
 195          *
 196          */
 197         read_lock(&tasklist_lock);
 198         nr = next_pidmap(pid_ns, 1);
 199         while (nr > 0) {
 200                 rcu_read_lock();
 201
 202                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 203                 if (task && !__fatal_signal_pending(task))
 204                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 205
 206                 rcu_read_unlock();
 207
 208                 nr = next_pidmap(pid_ns, nr);
 209         }
 210         read_unlock(&tasklist_lock);
 211
 212         /* Firstly reap the EXIT_ZOMBIE children we may have. */
 213         do {
 214                 clear_thread_flag(TIF_SIGPENDING);
 215                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 216         } while (rc != -ECHILD);
 217
 218         /*
 219          * sys_wait4() above can't reap the TASK_DEAD children.
 220          * Make sure they all go away, see free_pid().
 221          */
 222         for (;;) {
 223                 set_current_state(TASK_UNINTERRUPTIBLE);
 224                 if (pid_ns->nr_hashed == 1)
 225                         break;
 226                 schedule();
 227         }
 228         __set_current_state(TASK_RUNNING);
 229
 230         if (pid_ns->reboot)
 231                 current->signal->group_exit_code = pid_ns->reboot;
 232
 233         acct_exit_ns(pid_ns);
 234         return;
 235 }
 236
 237 #ifdef CONFIG_CHECKPOINT_RESTORE
 238 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 239                 void __user *buffer, size_t *lenp, loff_t *ppos)
 240 {
 241         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 242         struct ctl_table tmp = *table;
 243
 244         if (write && !ns_capable(pid_ns->user_ns, 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 = &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 void *pidns_get(struct task_struct *task)
 303 {
 304         struct pid_namespace *ns;
 305
 306         rcu_read_lock();
 307         ns = get_pid_ns(task_active_pid_ns(task));
 308         rcu_read_unlock();
 309
 310         return ns;
 311 }
 312
 313 static void pidns_put(void *ns)
 314 {
 315         put_pid_ns(ns);
 316 }
 317
 318 static int pidns_install(struct nsproxy *nsproxy, void *ns)
 319 {
 320         struct pid_namespace *active = task_active_pid_ns(current);
 321         struct pid_namespace *ancestor, *new = ns;
 322
 323         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN))
 324                 return -EPERM;
 325
 326         /*
 327          * Only allow entering the current active pid namespace
 328          * or a child of the current active pid namespace.
 329          *
 330          * This is required for fork to return a usable pid value and
 331          * this maintains the property that processes and their
 332          * children can not escape their current pid namespace.
 333          */
 334         if (new->level < active->level)
 335                 return -EINVAL;
 336
 337         ancestor = new;
 338         while (ancestor->level > active->level)
 339                 ancestor = ancestor->parent;
 340         if (ancestor != active)
 341                 return -EINVAL;
 342
 343         put_pid_ns(nsproxy->pid_ns);
 344         nsproxy->pid_ns = get_pid_ns(new);
 345         return 0;
 346 }
 347
 348 const struct proc_ns_operations pidns_operations = {
 349         .name           = "pid",
 350         .type           = CLONE_NEWPID,
 351         .get            = pidns_get,
 352         .put            = pidns_put,
 353         .install        = pidns_install,
 354 };
 355
 356 static __init int pid_namespaces_init(void)
 357 {
 358         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 359
 360 #ifdef CONFIG_CHECKPOINT_RESTORE
 361         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 362 #endif
 363         return 0;
 364 }
 365
 366 __initcall(pid_namespaces_init);