Back to home page

LXR

 
 

    


0001 /*
0002  * Pid namespaces
0003  *
0004  * Authors:
0005  *    (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
0006  *    (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
0007  *     Many thanks to Oleg Nesterov for comments and help
0008  *
0009  */
0010 
0011 #include <linux/pid.h>
0012 #include <linux/pid_namespace.h>
0013 #include <linux/user_namespace.h>
0014 #include <linux/syscalls.h>
0015 #include <linux/err.h>
0016 #include <linux/acct.h>
0017 #include <linux/slab.h>
0018 #include <linux/proc_ns.h>
0019 #include <linux/reboot.h>
0020 #include <linux/export.h>
0021 
0022 struct pid_cache {
0023     int nr_ids;
0024     char name[16];
0025     struct kmem_cache *cachep;
0026     struct list_head list;
0027 };
0028 
0029 static LIST_HEAD(pid_caches_lh);
0030 static DEFINE_MUTEX(pid_caches_mutex);
0031 static struct kmem_cache *pid_ns_cachep;
0032 
0033 /*
0034  * creates the kmem cache to allocate pids from.
0035  * @nr_ids: the number of numerical ids this pid will have to carry
0036  */
0037 
0038 static struct kmem_cache *create_pid_cachep(int nr_ids)
0039 {
0040     struct pid_cache *pcache;
0041     struct kmem_cache *cachep;
0042 
0043     mutex_lock(&pid_caches_mutex);
0044     list_for_each_entry(pcache, &pid_caches_lh, list)
0045         if (pcache->nr_ids == nr_ids)
0046             goto out;
0047 
0048     pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
0049     if (pcache == NULL)
0050         goto err_alloc;
0051 
0052     snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
0053     cachep = kmem_cache_create(pcache->name,
0054             sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
0055             0, SLAB_HWCACHE_ALIGN, NULL);
0056     if (cachep == NULL)
0057         goto err_cachep;
0058 
0059     pcache->nr_ids = nr_ids;
0060     pcache->cachep = cachep;
0061     list_add(&pcache->list, &pid_caches_lh);
0062 out:
0063     mutex_unlock(&pid_caches_mutex);
0064     return pcache->cachep;
0065 
0066 err_cachep:
0067     kfree(pcache);
0068 err_alloc:
0069     mutex_unlock(&pid_caches_mutex);
0070     return NULL;
0071 }
0072 
0073 static void proc_cleanup_work(struct work_struct *work)
0074 {
0075     struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
0076     pid_ns_release_proc(ns);
0077 }
0078 
0079 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
0080 #define MAX_PID_NS_LEVEL 32
0081 
0082 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
0083 {
0084     return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
0085 }
0086 
0087 static void dec_pid_namespaces(struct ucounts *ucounts)
0088 {
0089     dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
0090 }
0091 
0092 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
0093     struct pid_namespace *parent_pid_ns)
0094 {
0095     struct pid_namespace *ns;
0096     unsigned int level = parent_pid_ns->level + 1;
0097     struct ucounts *ucounts;
0098     int i;
0099     int err;
0100 
0101     err = -ENOSPC;
0102     if (level > MAX_PID_NS_LEVEL)
0103         goto out;
0104     ucounts = inc_pid_namespaces(user_ns);
0105     if (!ucounts)
0106         goto out;
0107 
0108     err = -ENOMEM;
0109     ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
0110     if (ns == NULL)
0111         goto out_dec;
0112 
0113     ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
0114     if (!ns->pidmap[0].page)
0115         goto out_free;
0116 
0117     ns->pid_cachep = create_pid_cachep(level + 1);
0118     if (ns->pid_cachep == NULL)
0119         goto out_free_map;
0120 
0121     err = ns_alloc_inum(&ns->ns);
0122     if (err)
0123         goto out_free_map;
0124     ns->ns.ops = &pidns_operations;
0125 
0126     kref_init(&ns->kref);
0127     ns->level = level;
0128     ns->parent = get_pid_ns(parent_pid_ns);
0129     ns->user_ns = get_user_ns(user_ns);
0130     ns->ucounts = ucounts;
0131     ns->nr_hashed = PIDNS_HASH_ADDING;
0132     INIT_WORK(&ns->proc_work, proc_cleanup_work);
0133 
0134     set_bit(0, ns->pidmap[0].page);
0135     atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
0136 
0137     for (i = 1; i < PIDMAP_ENTRIES; i++)
0138         atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
0139 
0140     return ns;
0141 
0142 out_free_map:
0143     kfree(ns->pidmap[0].page);
0144 out_free:
0145     kmem_cache_free(pid_ns_cachep, ns);
0146 out_dec:
0147     dec_pid_namespaces(ucounts);
0148 out:
0149     return ERR_PTR(err);
0150 }
0151 
0152 static void delayed_free_pidns(struct rcu_head *p)
0153 {
0154     struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
0155 
0156     dec_pid_namespaces(ns->ucounts);
0157     put_user_ns(ns->user_ns);
0158 
0159     kmem_cache_free(pid_ns_cachep, ns);
0160 }
0161 
0162 static void destroy_pid_namespace(struct pid_namespace *ns)
0163 {
0164     int i;
0165 
0166     ns_free_inum(&ns->ns);
0167     for (i = 0; i < PIDMAP_ENTRIES; i++)
0168         kfree(ns->pidmap[i].page);
0169     call_rcu(&ns->rcu, delayed_free_pidns);
0170 }
0171 
0172 struct pid_namespace *copy_pid_ns(unsigned long flags,
0173     struct user_namespace *user_ns, struct pid_namespace *old_ns)
0174 {
0175     if (!(flags & CLONE_NEWPID))
0176         return get_pid_ns(old_ns);
0177     if (task_active_pid_ns(current) != old_ns)
0178         return ERR_PTR(-EINVAL);
0179     return create_pid_namespace(user_ns, old_ns);
0180 }
0181 
0182 static void free_pid_ns(struct kref *kref)
0183 {
0184     struct pid_namespace *ns;
0185 
0186     ns = container_of(kref, struct pid_namespace, kref);
0187     destroy_pid_namespace(ns);
0188 }
0189 
0190 void put_pid_ns(struct pid_namespace *ns)
0191 {
0192     struct pid_namespace *parent;
0193 
0194     while (ns != &init_pid_ns) {
0195         parent = ns->parent;
0196         if (!kref_put(&ns->kref, free_pid_ns))
0197             break;
0198         ns = parent;
0199     }
0200 }
0201 EXPORT_SYMBOL_GPL(put_pid_ns);
0202 
0203 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
0204 {
0205     int nr;
0206     int rc;
0207     struct task_struct *task, *me = current;
0208     int init_pids = thread_group_leader(me) ? 1 : 2;
0209 
0210     /* Don't allow any more processes into the pid namespace */
0211     disable_pid_allocation(pid_ns);
0212 
0213     /*
0214      * Ignore SIGCHLD causing any terminated children to autoreap.
0215      * This speeds up the namespace shutdown, plus see the comment
0216      * below.
0217      */
0218     spin_lock_irq(&me->sighand->siglock);
0219     me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
0220     spin_unlock_irq(&me->sighand->siglock);
0221 
0222     /*
0223      * The last thread in the cgroup-init thread group is terminating.
0224      * Find remaining pid_ts in the namespace, signal and wait for them
0225      * to exit.
0226      *
0227      * Note:  This signals each threads in the namespace - even those that
0228      *    belong to the same thread group, To avoid this, we would have
0229      *    to walk the entire tasklist looking a processes in this
0230      *    namespace, but that could be unnecessarily expensive if the
0231      *    pid namespace has just a few processes. Or we need to
0232      *    maintain a tasklist for each pid namespace.
0233      *
0234      */
0235     read_lock(&tasklist_lock);
0236     nr = next_pidmap(pid_ns, 1);
0237     while (nr > 0) {
0238         rcu_read_lock();
0239 
0240         task = pid_task(find_vpid(nr), PIDTYPE_PID);
0241         if (task && !__fatal_signal_pending(task))
0242             send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
0243 
0244         rcu_read_unlock();
0245 
0246         nr = next_pidmap(pid_ns, nr);
0247     }
0248     read_unlock(&tasklist_lock);
0249 
0250     /*
0251      * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
0252      * sys_wait4() will also block until our children traced from the
0253      * parent namespace are detached and become EXIT_DEAD.
0254      */
0255     do {
0256         clear_thread_flag(TIF_SIGPENDING);
0257         rc = sys_wait4(-1, NULL, __WALL, NULL);
0258     } while (rc != -ECHILD);
0259 
0260     /*
0261      * sys_wait4() above can't reap the EXIT_DEAD children but we do not
0262      * really care, we could reparent them to the global init. We could
0263      * exit and reap ->child_reaper even if it is not the last thread in
0264      * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
0265      * pid_ns can not go away until proc_kill_sb() drops the reference.
0266      *
0267      * But this ns can also have other tasks injected by setns()+fork().
0268      * Again, ignoring the user visible semantics we do not really need
0269      * to wait until they are all reaped, but they can be reparented to
0270      * us and thus we need to ensure that pid->child_reaper stays valid
0271      * until they all go away. See free_pid()->wake_up_process().
0272      *
0273      * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
0274      * if reparented.
0275      */
0276     for (;;) {
0277         set_current_state(TASK_UNINTERRUPTIBLE);
0278         if (pid_ns->nr_hashed == init_pids)
0279             break;
0280         schedule();
0281     }
0282     __set_current_state(TASK_RUNNING);
0283 
0284     if (pid_ns->reboot)
0285         current->signal->group_exit_code = pid_ns->reboot;
0286 
0287     acct_exit_ns(pid_ns);
0288     return;
0289 }
0290 
0291 #ifdef CONFIG_CHECKPOINT_RESTORE
0292 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
0293         void __user *buffer, size_t *lenp, loff_t *ppos)
0294 {
0295     struct pid_namespace *pid_ns = task_active_pid_ns(current);
0296     struct ctl_table tmp = *table;
0297 
0298     if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
0299         return -EPERM;
0300 
0301     /*
0302      * Writing directly to ns' last_pid field is OK, since this field
0303      * is volatile in a living namespace anyway and a code writing to
0304      * it should synchronize its usage with external means.
0305      */
0306 
0307     tmp.data = &pid_ns->last_pid;
0308     return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
0309 }
0310 
0311 extern int pid_max;
0312 static int zero = 0;
0313 static struct ctl_table pid_ns_ctl_table[] = {
0314     {
0315         .procname = "ns_last_pid",
0316         .maxlen = sizeof(int),
0317         .mode = 0666, /* permissions are checked in the handler */
0318         .proc_handler = pid_ns_ctl_handler,
0319         .extra1 = &zero,
0320         .extra2 = &pid_max,
0321     },
0322     { }
0323 };
0324 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
0325 #endif  /* CONFIG_CHECKPOINT_RESTORE */
0326 
0327 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
0328 {
0329     if (pid_ns == &init_pid_ns)
0330         return 0;
0331 
0332     switch (cmd) {
0333     case LINUX_REBOOT_CMD_RESTART2:
0334     case LINUX_REBOOT_CMD_RESTART:
0335         pid_ns->reboot = SIGHUP;
0336         break;
0337 
0338     case LINUX_REBOOT_CMD_POWER_OFF:
0339     case LINUX_REBOOT_CMD_HALT:
0340         pid_ns->reboot = SIGINT;
0341         break;
0342     default:
0343         return -EINVAL;
0344     }
0345 
0346     read_lock(&tasklist_lock);
0347     force_sig(SIGKILL, pid_ns->child_reaper);
0348     read_unlock(&tasklist_lock);
0349 
0350     do_exit(0);
0351 
0352     /* Not reached */
0353     return 0;
0354 }
0355 
0356 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
0357 {
0358     return container_of(ns, struct pid_namespace, ns);
0359 }
0360 
0361 static struct ns_common *pidns_get(struct task_struct *task)
0362 {
0363     struct pid_namespace *ns;
0364 
0365     rcu_read_lock();
0366     ns = task_active_pid_ns(task);
0367     if (ns)
0368         get_pid_ns(ns);
0369     rcu_read_unlock();
0370 
0371     return ns ? &ns->ns : NULL;
0372 }
0373 
0374 static void pidns_put(struct ns_common *ns)
0375 {
0376     put_pid_ns(to_pid_ns(ns));
0377 }
0378 
0379 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
0380 {
0381     struct pid_namespace *active = task_active_pid_ns(current);
0382     struct pid_namespace *ancestor, *new = to_pid_ns(ns);
0383 
0384     if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
0385         !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
0386         return -EPERM;
0387 
0388     /*
0389      * Only allow entering the current active pid namespace
0390      * or a child of the current active pid namespace.
0391      *
0392      * This is required for fork to return a usable pid value and
0393      * this maintains the property that processes and their
0394      * children can not escape their current pid namespace.
0395      */
0396     if (new->level < active->level)
0397         return -EINVAL;
0398 
0399     ancestor = new;
0400     while (ancestor->level > active->level)
0401         ancestor = ancestor->parent;
0402     if (ancestor != active)
0403         return -EINVAL;
0404 
0405     put_pid_ns(nsproxy->pid_ns_for_children);
0406     nsproxy->pid_ns_for_children = get_pid_ns(new);
0407     return 0;
0408 }
0409 
0410 static struct ns_common *pidns_get_parent(struct ns_common *ns)
0411 {
0412     struct pid_namespace *active = task_active_pid_ns(current);
0413     struct pid_namespace *pid_ns, *p;
0414 
0415     /* See if the parent is in the current namespace */
0416     pid_ns = p = to_pid_ns(ns)->parent;
0417     for (;;) {
0418         if (!p)
0419             return ERR_PTR(-EPERM);
0420         if (p == active)
0421             break;
0422         p = p->parent;
0423     }
0424 
0425     return &get_pid_ns(pid_ns)->ns;
0426 }
0427 
0428 static struct user_namespace *pidns_owner(struct ns_common *ns)
0429 {
0430     return to_pid_ns(ns)->user_ns;
0431 }
0432 
0433 const struct proc_ns_operations pidns_operations = {
0434     .name       = "pid",
0435     .type       = CLONE_NEWPID,
0436     .get        = pidns_get,
0437     .put        = pidns_put,
0438     .install    = pidns_install,
0439     .owner      = pidns_owner,
0440     .get_parent = pidns_get_parent,
0441 };
0442 
0443 static __init int pid_namespaces_init(void)
0444 {
0445     pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
0446 
0447 #ifdef CONFIG_CHECKPOINT_RESTORE
0448     register_sysctl_paths(kern_path, pid_ns_ctl_table);
0449 #endif
0450     return 0;
0451 }
0452 
0453 __initcall(pid_namespaces_init);