blob: 3976dd57db78665a783f54d5361a59c621e37bcf
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_ns.h> |
19 | #include <linux/reboot.h> |
20 | #include <linux/export.h> |
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 void proc_cleanup_work(struct work_struct *work) |
74 | { |
75 | struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); |
76 | pid_ns_release_proc(ns); |
77 | } |
78 | |
79 | /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ |
80 | #define MAX_PID_NS_LEVEL 32 |
81 | |
82 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) |
83 | { |
84 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); |
85 | } |
86 | |
87 | static void dec_pid_namespaces(struct ucounts *ucounts) |
88 | { |
89 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); |
90 | } |
91 | |
92 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
93 | struct pid_namespace *parent_pid_ns) |
94 | { |
95 | struct pid_namespace *ns; |
96 | unsigned int level = parent_pid_ns->level + 1; |
97 | struct ucounts *ucounts; |
98 | int i; |
99 | int err; |
100 | |
101 | err = -ENOSPC; |
102 | if (level > MAX_PID_NS_LEVEL) |
103 | goto out; |
104 | ucounts = inc_pid_namespaces(user_ns); |
105 | if (!ucounts) |
106 | goto out; |
107 | |
108 | err = -ENOMEM; |
109 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
110 | if (ns == NULL) |
111 | goto out_dec; |
112 | |
113 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
114 | if (!ns->pidmap[0].page) |
115 | goto out_free; |
116 | |
117 | ns->pid_cachep = create_pid_cachep(level + 1); |
118 | if (ns->pid_cachep == NULL) |
119 | goto out_free_map; |
120 | |
121 | err = ns_alloc_inum(&ns->ns); |
122 | if (err) |
123 | goto out_free_map; |
124 | ns->ns.ops = &pidns_operations; |
125 | |
126 | kref_init(&ns->kref); |
127 | ns->level = level; |
128 | ns->parent = get_pid_ns(parent_pid_ns); |
129 | ns->user_ns = get_user_ns(user_ns); |
130 | ns->ucounts = ucounts; |
131 | ns->nr_hashed = PIDNS_HASH_ADDING; |
132 | INIT_WORK(&ns->proc_work, proc_cleanup_work); |
133 | |
134 | set_bit(0, ns->pidmap[0].page); |
135 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); |
136 | |
137 | for (i = 1; i < PIDMAP_ENTRIES; i++) |
138 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); |
139 | |
140 | return ns; |
141 | |
142 | out_free_map: |
143 | kfree(ns->pidmap[0].page); |
144 | out_free: |
145 | kmem_cache_free(pid_ns_cachep, ns); |
146 | out_dec: |
147 | dec_pid_namespaces(ucounts); |
148 | out: |
149 | return ERR_PTR(err); |
150 | } |
151 | |
152 | static void delayed_free_pidns(struct rcu_head *p) |
153 | { |
154 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); |
155 | |
156 | dec_pid_namespaces(ns->ucounts); |
157 | put_user_ns(ns->user_ns); |
158 | |
159 | kmem_cache_free(pid_ns_cachep, ns); |
160 | } |
161 | |
162 | static void destroy_pid_namespace(struct pid_namespace *ns) |
163 | { |
164 | int i; |
165 | |
166 | ns_free_inum(&ns->ns); |
167 | for (i = 0; i < PIDMAP_ENTRIES; i++) |
168 | kfree(ns->pidmap[i].page); |
169 | call_rcu(&ns->rcu, delayed_free_pidns); |
170 | } |
171 | |
172 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
173 | struct user_namespace *user_ns, struct pid_namespace *old_ns) |
174 | { |
175 | if (!(flags & CLONE_NEWPID)) |
176 | return get_pid_ns(old_ns); |
177 | if (task_active_pid_ns(current) != old_ns) |
178 | return ERR_PTR(-EINVAL); |
179 | return create_pid_namespace(user_ns, old_ns); |
180 | } |
181 | |
182 | static void free_pid_ns(struct kref *kref) |
183 | { |
184 | struct pid_namespace *ns; |
185 | |
186 | ns = container_of(kref, struct pid_namespace, kref); |
187 | destroy_pid_namespace(ns); |
188 | } |
189 | |
190 | void put_pid_ns(struct pid_namespace *ns) |
191 | { |
192 | struct pid_namespace *parent; |
193 | |
194 | while (ns != &init_pid_ns) { |
195 | parent = ns->parent; |
196 | if (!kref_put(&ns->kref, free_pid_ns)) |
197 | break; |
198 | ns = parent; |
199 | } |
200 | } |
201 | EXPORT_SYMBOL_GPL(put_pid_ns); |
202 | |
203 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) |
204 | { |
205 | int nr; |
206 | int rc; |
207 | struct task_struct *task, *me = current; |
208 | int init_pids = thread_group_leader(me) ? 1 : 2; |
209 | |
210 | /* Don't allow any more processes into the pid namespace */ |
211 | disable_pid_allocation(pid_ns); |
212 | |
213 | /* |
214 | * Ignore SIGCHLD causing any terminated children to autoreap. |
215 | * This speeds up the namespace shutdown, plus see the comment |
216 | * below. |
217 | */ |
218 | spin_lock_irq(&me->sighand->siglock); |
219 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; |
220 | spin_unlock_irq(&me->sighand->siglock); |
221 | |
222 | /* |
223 | * The last thread in the cgroup-init thread group is terminating. |
224 | * Find remaining pid_ts in the namespace, signal and wait for them |
225 | * to exit. |
226 | * |
227 | * Note: This signals each threads in the namespace - even those that |
228 | * belong to the same thread group, To avoid this, we would have |
229 | * to walk the entire tasklist looking a processes in this |
230 | * namespace, but that could be unnecessarily expensive if the |
231 | * pid namespace has just a few processes. Or we need to |
232 | * maintain a tasklist for each pid namespace. |
233 | * |
234 | */ |
235 | read_lock(&tasklist_lock); |
236 | nr = next_pidmap(pid_ns, 1); |
237 | while (nr > 0) { |
238 | rcu_read_lock(); |
239 | |
240 | task = pid_task(find_vpid(nr), PIDTYPE_PID); |
241 | if (task && !__fatal_signal_pending(task)) |
242 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); |
243 | |
244 | rcu_read_unlock(); |
245 | |
246 | nr = next_pidmap(pid_ns, nr); |
247 | } |
248 | read_unlock(&tasklist_lock); |
249 | |
250 | /* |
251 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. |
252 | * sys_wait4() will also block until our children traced from the |
253 | * parent namespace are detached and become EXIT_DEAD. |
254 | */ |
255 | do { |
256 | clear_thread_flag(TIF_SIGPENDING); |
257 | rc = sys_wait4(-1, NULL, __WALL, NULL); |
258 | } while (rc != -ECHILD); |
259 | |
260 | /* |
261 | * sys_wait4() above can't reap the EXIT_DEAD children but we do not |
262 | * really care, we could reparent them to the global init. We could |
263 | * exit and reap ->child_reaper even if it is not the last thread in |
264 | * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(), |
265 | * pid_ns can not go away until proc_kill_sb() drops the reference. |
266 | * |
267 | * But this ns can also have other tasks injected by setns()+fork(). |
268 | * Again, ignoring the user visible semantics we do not really need |
269 | * to wait until they are all reaped, but they can be reparented to |
270 | * us and thus we need to ensure that pid->child_reaper stays valid |
271 | * until they all go away. See free_pid()->wake_up_process(). |
272 | * |
273 | * We rely on ignored SIGCHLD, an injected zombie must be autoreaped |
274 | * if reparented. |
275 | */ |
276 | for (;;) { |
277 | set_current_state(TASK_INTERRUPTIBLE); |
278 | if (pid_ns->nr_hashed == init_pids) |
279 | break; |
280 | schedule(); |
281 | } |
282 | __set_current_state(TASK_RUNNING); |
283 | |
284 | if (pid_ns->reboot) |
285 | current->signal->group_exit_code = pid_ns->reboot; |
286 | |
287 | acct_exit_ns(pid_ns); |
288 | return; |
289 | } |
290 | |
291 | #ifdef CONFIG_CHECKPOINT_RESTORE |
292 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
293 | void __user *buffer, size_t *lenp, loff_t *ppos) |
294 | { |
295 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
296 | struct ctl_table tmp = *table; |
297 | |
298 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) |
299 | return -EPERM; |
300 | |
301 | /* |
302 | * Writing directly to ns' last_pid field is OK, since this field |
303 | * is volatile in a living namespace anyway and a code writing to |
304 | * it should synchronize its usage with external means. |
305 | */ |
306 | |
307 | tmp.data = &pid_ns->last_pid; |
308 | return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); |
309 | } |
310 | |
311 | extern int pid_max; |
312 | static int zero = 0; |
313 | static struct ctl_table pid_ns_ctl_table[] = { |
314 | { |
315 | .procname = "ns_last_pid", |
316 | .maxlen = sizeof(int), |
317 | .mode = 0666, /* permissions are checked in the handler */ |
318 | .proc_handler = pid_ns_ctl_handler, |
319 | .extra1 = &zero, |
320 | .extra2 = &pid_max, |
321 | }, |
322 | { } |
323 | }; |
324 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
325 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
326 | |
327 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
328 | { |
329 | if (pid_ns == &init_pid_ns) |
330 | return 0; |
331 | |
332 | switch (cmd) { |
333 | case LINUX_REBOOT_CMD_RESTART2: |
334 | case LINUX_REBOOT_CMD_RESTART: |
335 | pid_ns->reboot = SIGHUP; |
336 | break; |
337 | |
338 | case LINUX_REBOOT_CMD_POWER_OFF: |
339 | case LINUX_REBOOT_CMD_HALT: |
340 | pid_ns->reboot = SIGINT; |
341 | break; |
342 | default: |
343 | return -EINVAL; |
344 | } |
345 | |
346 | read_lock(&tasklist_lock); |
347 | force_sig(SIGKILL, pid_ns->child_reaper); |
348 | read_unlock(&tasklist_lock); |
349 | |
350 | do_exit(0); |
351 | |
352 | /* Not reached */ |
353 | return 0; |
354 | } |
355 | |
356 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
357 | { |
358 | return container_of(ns, struct pid_namespace, ns); |
359 | } |
360 | |
361 | static struct ns_common *pidns_get(struct task_struct *task) |
362 | { |
363 | struct pid_namespace *ns; |
364 | |
365 | rcu_read_lock(); |
366 | ns = task_active_pid_ns(task); |
367 | if (ns) |
368 | get_pid_ns(ns); |
369 | rcu_read_unlock(); |
370 | |
371 | return ns ? &ns->ns : NULL; |
372 | } |
373 | |
374 | static void pidns_put(struct ns_common *ns) |
375 | { |
376 | put_pid_ns(to_pid_ns(ns)); |
377 | } |
378 | |
379 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
380 | { |
381 | struct pid_namespace *active = task_active_pid_ns(current); |
382 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
383 | |
384 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
385 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
386 | return -EPERM; |
387 | |
388 | /* |
389 | * Only allow entering the current active pid namespace |
390 | * or a child of the current active pid namespace. |
391 | * |
392 | * This is required for fork to return a usable pid value and |
393 | * this maintains the property that processes and their |
394 | * children can not escape their current pid namespace. |
395 | */ |
396 | if (new->level < active->level) |
397 | return -EINVAL; |
398 | |
399 | ancestor = new; |
400 | while (ancestor->level > active->level) |
401 | ancestor = ancestor->parent; |
402 | if (ancestor != active) |
403 | return -EINVAL; |
404 | |
405 | put_pid_ns(nsproxy->pid_ns_for_children); |
406 | nsproxy->pid_ns_for_children = get_pid_ns(new); |
407 | return 0; |
408 | } |
409 | |
410 | static struct ns_common *pidns_get_parent(struct ns_common *ns) |
411 | { |
412 | struct pid_namespace *active = task_active_pid_ns(current); |
413 | struct pid_namespace *pid_ns, *p; |
414 | |
415 | /* See if the parent is in the current namespace */ |
416 | pid_ns = p = to_pid_ns(ns)->parent; |
417 | for (;;) { |
418 | if (!p) |
419 | return ERR_PTR(-EPERM); |
420 | if (p == active) |
421 | break; |
422 | p = p->parent; |
423 | } |
424 | |
425 | return &get_pid_ns(pid_ns)->ns; |
426 | } |
427 | |
428 | static struct user_namespace *pidns_owner(struct ns_common *ns) |
429 | { |
430 | return to_pid_ns(ns)->user_ns; |
431 | } |
432 | |
433 | const struct proc_ns_operations pidns_operations = { |
434 | .name = "pid", |
435 | .type = CLONE_NEWPID, |
436 | .get = pidns_get, |
437 | .put = pidns_put, |
438 | .install = pidns_install, |
439 | .owner = pidns_owner, |
440 | .get_parent = pidns_get_parent, |
441 | }; |
442 | |
443 | static __init int pid_namespaces_init(void) |
444 | { |
445 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); |
446 | |
447 | #ifdef CONFIG_CHECKPOINT_RESTORE |
448 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
449 | #endif |
450 | return 0; |
451 | } |
452 | |
453 | __initcall(pid_namespaces_init); |
454 |