Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Pid namespaces
4 *
5 * Authors:
6 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
7 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
8 * Many thanks to Oleg Nesterov for comments and help
9 *
10 */
11
12#include <linux/pid.h>
13#include <linux/pid_namespace.h>
14#include <linux/user_namespace.h>
15#include <linux/syscalls.h>
16#include <linux/cred.h>
17#include <linux/err.h>
18#include <linux/acct.h>
19#include <linux/slab.h>
20#include <linux/proc_ns.h>
21#include <linux/reboot.h>
22#include <linux/export.h>
23#include <linux/sched/task.h>
24#include <linux/sched/signal.h>
25#include <linux/idr.h>
26
27static DEFINE_MUTEX(pid_caches_mutex);
28static struct kmem_cache *pid_ns_cachep;
29/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
30#define MAX_PID_NS_LEVEL 32
31/* Write once array, filled from the beginning. */
32static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
33
34/*
35 * creates the kmem cache to allocate pids from.
36 * @level: pid namespace level
37 */
38
39static struct kmem_cache *create_pid_cachep(unsigned int level)
40{
41 /* Level 0 is init_pid_ns.pid_cachep */
42 struct kmem_cache **pkc = &pid_cache[level - 1];
43 struct kmem_cache *kc;
44 char name[4 + 10 + 1];
45 unsigned int len;
46
47 kc = READ_ONCE(*pkc);
48 if (kc)
49 return kc;
50
51 snprintf(name, sizeof(name), "pid_%u", level + 1);
52 len = sizeof(struct pid) + level * sizeof(struct upid);
53 mutex_lock(&pid_caches_mutex);
54 /* Name collision forces to do allocation under mutex. */
55 if (!*pkc)
56 *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
57 mutex_unlock(&pid_caches_mutex);
58 /* current can fail, but someone else can succeed. */
59 return READ_ONCE(*pkc);
60}
61
62static void proc_cleanup_work(struct work_struct *work)
63{
64 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
65 pid_ns_release_proc(ns);
66}
67
68static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
69{
70 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
71}
72
73static void dec_pid_namespaces(struct ucounts *ucounts)
74{
75 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
76}
77
78static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
79 struct pid_namespace *parent_pid_ns)
80{
81 struct pid_namespace *ns;
82 unsigned int level = parent_pid_ns->level + 1;
83 struct ucounts *ucounts;
84 int err;
85
86 err = -EINVAL;
87 if (!in_userns(parent_pid_ns->user_ns, user_ns))
88 goto out;
89
90 err = -ENOSPC;
91 if (level > MAX_PID_NS_LEVEL)
92 goto out;
93 ucounts = inc_pid_namespaces(user_ns);
94 if (!ucounts)
95 goto out;
96
97 err = -ENOMEM;
98 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
99 if (ns == NULL)
100 goto out_dec;
101
102 idr_init(&ns->idr);
103
104 ns->pid_cachep = create_pid_cachep(level);
105 if (ns->pid_cachep == NULL)
106 goto out_free_idr;
107
108 err = ns_alloc_inum(&ns->ns);
109 if (err)
110 goto out_free_idr;
111 ns->ns.ops = &pidns_operations;
112
113 kref_init(&ns->kref);
114 ns->level = level;
115 ns->parent = get_pid_ns(parent_pid_ns);
116 ns->user_ns = get_user_ns(user_ns);
117 ns->ucounts = ucounts;
118 ns->pid_allocated = PIDNS_ADDING;
119 INIT_WORK(&ns->proc_work, proc_cleanup_work);
120
121 return ns;
122
123out_free_idr:
124 idr_destroy(&ns->idr);
125 kmem_cache_free(pid_ns_cachep, ns);
126out_dec:
127 dec_pid_namespaces(ucounts);
128out:
129 return ERR_PTR(err);
130}
131
132static void delayed_free_pidns(struct rcu_head *p)
133{
134 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
135
136 dec_pid_namespaces(ns->ucounts);
137 put_user_ns(ns->user_ns);
138
139 kmem_cache_free(pid_ns_cachep, ns);
140}
141
142static void destroy_pid_namespace(struct pid_namespace *ns)
143{
144 ns_free_inum(&ns->ns);
145
146 idr_destroy(&ns->idr);
147 call_rcu(&ns->rcu, delayed_free_pidns);
148}
149
150struct pid_namespace *copy_pid_ns(unsigned long flags,
151 struct user_namespace *user_ns, struct pid_namespace *old_ns)
152{
153 if (!(flags & CLONE_NEWPID))
154 return get_pid_ns(old_ns);
155 if (task_active_pid_ns(current) != old_ns)
156 return ERR_PTR(-EINVAL);
157 return create_pid_namespace(user_ns, old_ns);
158}
159
160static void free_pid_ns(struct kref *kref)
161{
162 struct pid_namespace *ns;
163
164 ns = container_of(kref, struct pid_namespace, kref);
165 destroy_pid_namespace(ns);
166}
167
168void put_pid_ns(struct pid_namespace *ns)
169{
170 struct pid_namespace *parent;
171
172 while (ns != &init_pid_ns) {
173 parent = ns->parent;
174 if (!kref_put(&ns->kref, free_pid_ns))
175 break;
176 ns = parent;
177 }
178}
179EXPORT_SYMBOL_GPL(put_pid_ns);
180
181void zap_pid_ns_processes(struct pid_namespace *pid_ns)
182{
183 int nr;
184 int rc;
185 struct task_struct *task, *me = current;
186 int init_pids = thread_group_leader(me) ? 1 : 2;
187 struct pid *pid;
188
189 /* Don't allow any more processes into the pid namespace */
190 disable_pid_allocation(pid_ns);
191
192 /*
193 * Ignore SIGCHLD causing any terminated children to autoreap.
194 * This speeds up the namespace shutdown, plus see the comment
195 * below.
196 */
197 spin_lock_irq(&me->sighand->siglock);
198 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
199 spin_unlock_irq(&me->sighand->siglock);
200
201 /*
202 * The last thread in the cgroup-init thread group is terminating.
203 * Find remaining pid_ts in the namespace, signal and wait for them
204 * to exit.
205 *
206 * Note: This signals each threads in the namespace - even those that
207 * belong to the same thread group, To avoid this, we would have
208 * to walk the entire tasklist looking a processes in this
209 * namespace, but that could be unnecessarily expensive if the
210 * pid namespace has just a few processes. Or we need to
211 * maintain a tasklist for each pid namespace.
212 *
213 */
214 rcu_read_lock();
215 read_lock(&tasklist_lock);
216 nr = 2;
217 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
218 task = pid_task(pid, PIDTYPE_PID);
219 if (task && !__fatal_signal_pending(task))
220 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
221 }
222 read_unlock(&tasklist_lock);
223 rcu_read_unlock();
224
225 /*
226 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
227 * kernel_wait4() will also block until our children traced from the
228 * parent namespace are detached and become EXIT_DEAD.
229 */
230 do {
231 clear_thread_flag(TIF_SIGPENDING);
232 rc = kernel_wait4(-1, NULL, __WALL, NULL);
233 } while (rc != -ECHILD);
234
235 /*
236 * kernel_wait4() above can't reap the EXIT_DEAD children but we do not
237 * really care, we could reparent them to the global init. We could
238 * exit and reap ->child_reaper even if it is not the last thread in
239 * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(),
240 * pid_ns can not go away until proc_kill_sb() drops the reference.
241 *
242 * But this ns can also have other tasks injected by setns()+fork().
243 * Again, ignoring the user visible semantics we do not really need
244 * to wait until they are all reaped, but they can be reparented to
245 * us and thus we need to ensure that pid->child_reaper stays valid
246 * until they all go away. See free_pid()->wake_up_process().
247 *
248 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
249 * if reparented.
250 */
251 for (;;) {
252 set_current_state(TASK_INTERRUPTIBLE);
253 if (pid_ns->pid_allocated == init_pids)
254 break;
255 schedule();
256 }
257 __set_current_state(TASK_RUNNING);
258
259 if (pid_ns->reboot)
260 current->signal->group_exit_code = pid_ns->reboot;
261
262 acct_exit_ns(pid_ns);
263 return;
264}
265
266#ifdef CONFIG_CHECKPOINT_RESTORE
267static int pid_ns_ctl_handler(struct ctl_table *table, int write,
268 void __user *buffer, size_t *lenp, loff_t *ppos)
269{
270 struct pid_namespace *pid_ns = task_active_pid_ns(current);
271 struct ctl_table tmp = *table;
272 int ret, next;
273
274 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
275 return -EPERM;
276
277 /*
278 * Writing directly to ns' last_pid field is OK, since this field
279 * is volatile in a living namespace anyway and a code writing to
280 * it should synchronize its usage with external means.
281 */
282
283 next = idr_get_cursor(&pid_ns->idr) - 1;
284
285 tmp.data = &next;
286 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
287 if (!ret && write)
288 idr_set_cursor(&pid_ns->idr, next + 1);
289
290 return ret;
291}
292
293extern int pid_max;
294static struct ctl_table pid_ns_ctl_table[] = {
295 {
296 .procname = "ns_last_pid",
297 .maxlen = sizeof(int),
298 .mode = 0666, /* permissions are checked in the handler */
299 .proc_handler = pid_ns_ctl_handler,
300 .extra1 = SYSCTL_ZERO,
301 .extra2 = &pid_max,
302 },
303 { }
304};
305static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
306#endif /* CONFIG_CHECKPOINT_RESTORE */
307
308int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
309{
310 if (pid_ns == &init_pid_ns)
311 return 0;
312
313 switch (cmd) {
314 case LINUX_REBOOT_CMD_RESTART2:
315 case LINUX_REBOOT_CMD_RESTART:
316 pid_ns->reboot = SIGHUP;
317 break;
318
319 case LINUX_REBOOT_CMD_POWER_OFF:
320 case LINUX_REBOOT_CMD_HALT:
321 pid_ns->reboot = SIGINT;
322 break;
323 default:
324 return -EINVAL;
325 }
326
327 read_lock(&tasklist_lock);
328 send_sig(SIGKILL, pid_ns->child_reaper, 1);
329 read_unlock(&tasklist_lock);
330
331 do_exit(0);
332
333 /* Not reached */
334 return 0;
335}
336
337static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
338{
339 return container_of(ns, struct pid_namespace, ns);
340}
341
342static struct ns_common *pidns_get(struct task_struct *task)
343{
344 struct pid_namespace *ns;
345
346 rcu_read_lock();
347 ns = task_active_pid_ns(task);
348 if (ns)
349 get_pid_ns(ns);
350 rcu_read_unlock();
351
352 return ns ? &ns->ns : NULL;
353}
354
355static struct ns_common *pidns_for_children_get(struct task_struct *task)
356{
357 struct pid_namespace *ns = NULL;
358
359 task_lock(task);
360 if (task->nsproxy) {
361 ns = task->nsproxy->pid_ns_for_children;
362 get_pid_ns(ns);
363 }
364 task_unlock(task);
365
366 if (ns) {
367 read_lock(&tasklist_lock);
368 if (!ns->child_reaper) {
369 put_pid_ns(ns);
370 ns = NULL;
371 }
372 read_unlock(&tasklist_lock);
373 }
374
375 return ns ? &ns->ns : NULL;
376}
377
378static void pidns_put(struct ns_common *ns)
379{
380 put_pid_ns(to_pid_ns(ns));
381}
382
383static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
384{
385 struct pid_namespace *active = task_active_pid_ns(current);
386 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
387
388 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
389 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
390 return -EPERM;
391
392 /*
393 * Only allow entering the current active pid namespace
394 * or a child of the current active pid namespace.
395 *
396 * This is required for fork to return a usable pid value and
397 * this maintains the property that processes and their
398 * children can not escape their current pid namespace.
399 */
400 if (new->level < active->level)
401 return -EINVAL;
402
403 ancestor = new;
404 while (ancestor->level > active->level)
405 ancestor = ancestor->parent;
406 if (ancestor != active)
407 return -EINVAL;
408
409 put_pid_ns(nsproxy->pid_ns_for_children);
410 nsproxy->pid_ns_for_children = get_pid_ns(new);
411 return 0;
412}
413
414static struct ns_common *pidns_get_parent(struct ns_common *ns)
415{
416 struct pid_namespace *active = task_active_pid_ns(current);
417 struct pid_namespace *pid_ns, *p;
418
419 /* See if the parent is in the current namespace */
420 pid_ns = p = to_pid_ns(ns)->parent;
421 for (;;) {
422 if (!p)
423 return ERR_PTR(-EPERM);
424 if (p == active)
425 break;
426 p = p->parent;
427 }
428
429 return &get_pid_ns(pid_ns)->ns;
430}
431
432static struct user_namespace *pidns_owner(struct ns_common *ns)
433{
434 return to_pid_ns(ns)->user_ns;
435}
436
437const struct proc_ns_operations pidns_operations = {
438 .name = "pid",
439 .type = CLONE_NEWPID,
440 .get = pidns_get,
441 .put = pidns_put,
442 .install = pidns_install,
443 .owner = pidns_owner,
444 .get_parent = pidns_get_parent,
445};
446
447const struct proc_ns_operations pidns_for_children_operations = {
448 .name = "pid_for_children",
449 .real_ns_name = "pid",
450 .type = CLONE_NEWPID,
451 .get = pidns_for_children_get,
452 .put = pidns_put,
453 .install = pidns_install,
454 .owner = pidns_owner,
455 .get_parent = pidns_get_parent,
456};
457
458static __init int pid_namespaces_init(void)
459{
460 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
461
462#ifdef CONFIG_CHECKPOINT_RESTORE
463 register_sysctl_paths(kern_path, pid_ns_ctl_table);
464#endif
465 return 0;
466}
467
468__initcall(pid_namespaces_init);
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
19#define BITS_PER_PAGE (PAGE_SIZE*8)
20
21struct pid_cache {
22 int nr_ids;
23 char name[16];
24 struct kmem_cache *cachep;
25 struct list_head list;
26};
27
28static LIST_HEAD(pid_caches_lh);
29static DEFINE_MUTEX(pid_caches_mutex);
30static struct kmem_cache *pid_ns_cachep;
31
32/*
33 * creates the kmem cache to allocate pids from.
34 * @nr_ids: the number of numerical ids this pid will have to carry
35 */
36
37static struct kmem_cache *create_pid_cachep(int nr_ids)
38{
39 struct pid_cache *pcache;
40 struct kmem_cache *cachep;
41
42 mutex_lock(&pid_caches_mutex);
43 list_for_each_entry(pcache, &pid_caches_lh, list)
44 if (pcache->nr_ids == nr_ids)
45 goto out;
46
47 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
48 if (pcache == NULL)
49 goto err_alloc;
50
51 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
52 cachep = kmem_cache_create(pcache->name,
53 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
54 0, SLAB_HWCACHE_ALIGN, NULL);
55 if (cachep == NULL)
56 goto err_cachep;
57
58 pcache->nr_ids = nr_ids;
59 pcache->cachep = cachep;
60 list_add(&pcache->list, &pid_caches_lh);
61out:
62 mutex_unlock(&pid_caches_mutex);
63 return pcache->cachep;
64
65err_cachep:
66 kfree(pcache);
67err_alloc:
68 mutex_unlock(&pid_caches_mutex);
69 return NULL;
70}
71
72static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
73{
74 struct pid_namespace *ns;
75 unsigned int level = parent_pid_ns->level + 1;
76 int i, err = -ENOMEM;
77
78 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
79 if (ns == NULL)
80 goto out;
81
82 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
83 if (!ns->pidmap[0].page)
84 goto out_free;
85
86 ns->pid_cachep = create_pid_cachep(level + 1);
87 if (ns->pid_cachep == NULL)
88 goto out_free_map;
89
90 kref_init(&ns->kref);
91 ns->level = level;
92 ns->parent = get_pid_ns(parent_pid_ns);
93
94 set_bit(0, ns->pidmap[0].page);
95 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
96
97 for (i = 1; i < PIDMAP_ENTRIES; i++)
98 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
99
100 err = pid_ns_prepare_proc(ns);
101 if (err)
102 goto out_put_parent_pid_ns;
103
104 return ns;
105
106out_put_parent_pid_ns:
107 put_pid_ns(parent_pid_ns);
108out_free_map:
109 kfree(ns->pidmap[0].page);
110out_free:
111 kmem_cache_free(pid_ns_cachep, ns);
112out:
113 return ERR_PTR(err);
114}
115
116static void destroy_pid_namespace(struct pid_namespace *ns)
117{
118 int i;
119
120 for (i = 0; i < PIDMAP_ENTRIES; i++)
121 kfree(ns->pidmap[i].page);
122 kmem_cache_free(pid_ns_cachep, ns);
123}
124
125struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
126{
127 if (!(flags & CLONE_NEWPID))
128 return get_pid_ns(old_ns);
129 if (flags & (CLONE_THREAD|CLONE_PARENT))
130 return ERR_PTR(-EINVAL);
131 return create_pid_namespace(old_ns);
132}
133
134void free_pid_ns(struct kref *kref)
135{
136 struct pid_namespace *ns, *parent;
137
138 ns = container_of(kref, struct pid_namespace, kref);
139
140 parent = ns->parent;
141 destroy_pid_namespace(ns);
142
143 if (parent != NULL)
144 put_pid_ns(parent);
145}
146
147void zap_pid_ns_processes(struct pid_namespace *pid_ns)
148{
149 int nr;
150 int rc;
151 struct task_struct *task;
152
153 /*
154 * The last thread in the cgroup-init thread group is terminating.
155 * Find remaining pid_ts in the namespace, signal and wait for them
156 * to exit.
157 *
158 * Note: This signals each threads in the namespace - even those that
159 * belong to the same thread group, To avoid this, we would have
160 * to walk the entire tasklist looking a processes in this
161 * namespace, but that could be unnecessarily expensive if the
162 * pid namespace has just a few processes. Or we need to
163 * maintain a tasklist for each pid namespace.
164 *
165 */
166 read_lock(&tasklist_lock);
167 nr = next_pidmap(pid_ns, 1);
168 while (nr > 0) {
169 rcu_read_lock();
170
171 /*
172 * Any nested-container's init processes won't ignore the
173 * SEND_SIG_NOINFO signal, see send_signal()->si_fromuser().
174 */
175 task = pid_task(find_vpid(nr), PIDTYPE_PID);
176 if (task)
177 send_sig_info(SIGKILL, SEND_SIG_NOINFO, task);
178
179 rcu_read_unlock();
180
181 nr = next_pidmap(pid_ns, nr);
182 }
183 read_unlock(&tasklist_lock);
184
185 do {
186 clear_thread_flag(TIF_SIGPENDING);
187 rc = sys_wait4(-1, NULL, __WALL, NULL);
188 } while (rc != -ECHILD);
189
190 acct_exit_ns(pid_ns);
191 return;
192}
193
194static __init int pid_namespaces_init(void)
195{
196 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
197 return 0;
198}
199
200__initcall(pid_namespaces_init);