]>
Commit | Line | Data |
---|---|---|
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/cred.h> | |
16 | #include <linux/err.h> | |
17 | #include <linux/acct.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/proc_ns.h> | |
20 | #include <linux/reboot.h> | |
21 | #include <linux/export.h> | |
22 | #include <linux/sched/task.h> | |
23 | #include <linux/sched/signal.h> | |
24 | #include <linux/idr.h> | |
25 | ||
26 | struct pid_cache { | |
27 | int nr_ids; | |
28 | char name[16]; | |
29 | struct kmem_cache *cachep; | |
30 | struct list_head list; | |
31 | }; | |
32 | ||
33 | static LIST_HEAD(pid_caches_lh); | |
34 | static DEFINE_MUTEX(pid_caches_mutex); | |
35 | static struct kmem_cache *pid_ns_cachep; | |
36 | ||
37 | /* | |
38 | * creates the kmem cache to allocate pids from. | |
39 | * @nr_ids: the number of numerical ids this pid will have to carry | |
40 | */ | |
41 | ||
42 | static struct kmem_cache *create_pid_cachep(int nr_ids) | |
43 | { | |
44 | struct pid_cache *pcache; | |
45 | struct kmem_cache *cachep; | |
46 | ||
47 | mutex_lock(&pid_caches_mutex); | |
48 | list_for_each_entry(pcache, &pid_caches_lh, list) | |
49 | if (pcache->nr_ids == nr_ids) | |
50 | goto out; | |
51 | ||
52 | pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); | |
53 | if (pcache == NULL) | |
54 | goto err_alloc; | |
55 | ||
56 | snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); | |
57 | cachep = kmem_cache_create(pcache->name, | |
58 | sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid), | |
59 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
60 | if (cachep == NULL) | |
61 | goto err_cachep; | |
62 | ||
63 | pcache->nr_ids = nr_ids; | |
64 | pcache->cachep = cachep; | |
65 | list_add(&pcache->list, &pid_caches_lh); | |
66 | out: | |
67 | mutex_unlock(&pid_caches_mutex); | |
68 | return pcache->cachep; | |
69 | ||
70 | err_cachep: | |
71 | kfree(pcache); | |
72 | err_alloc: | |
73 | mutex_unlock(&pid_caches_mutex); | |
74 | return NULL; | |
75 | } | |
76 | ||
77 | static void proc_cleanup_work(struct work_struct *work) | |
78 | { | |
79 | struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); | |
80 | pid_ns_release_proc(ns); | |
81 | } | |
82 | ||
83 | /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ | |
84 | #define MAX_PID_NS_LEVEL 32 | |
85 | ||
86 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) | |
87 | { | |
88 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); | |
89 | } | |
90 | ||
91 | static void dec_pid_namespaces(struct ucounts *ucounts) | |
92 | { | |
93 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); | |
94 | } | |
95 | ||
96 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, | |
97 | struct pid_namespace *parent_pid_ns) | |
98 | { | |
99 | struct pid_namespace *ns; | |
100 | unsigned int level = parent_pid_ns->level + 1; | |
101 | struct ucounts *ucounts; | |
102 | int err; | |
103 | ||
104 | err = -EINVAL; | |
105 | if (!in_userns(parent_pid_ns->user_ns, user_ns)) | |
106 | goto out; | |
107 | ||
108 | err = -ENOSPC; | |
109 | if (level > MAX_PID_NS_LEVEL) | |
110 | goto out; | |
111 | ucounts = inc_pid_namespaces(user_ns); | |
112 | if (!ucounts) | |
113 | goto out; | |
114 | ||
115 | err = -ENOMEM; | |
116 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); | |
117 | if (ns == NULL) | |
118 | goto out_dec; | |
119 | ||
120 | idr_init(&ns->idr); | |
121 | ||
122 | ns->pid_cachep = create_pid_cachep(level + 1); | |
123 | if (ns->pid_cachep == NULL) | |
124 | goto out_free_idr; | |
125 | ||
126 | err = ns_alloc_inum(&ns->ns); | |
127 | if (err) | |
128 | goto out_free_idr; | |
129 | ns->ns.ops = &pidns_operations; | |
130 | ||
131 | kref_init(&ns->kref); | |
132 | ns->level = level; | |
133 | ns->parent = get_pid_ns(parent_pid_ns); | |
134 | ns->user_ns = get_user_ns(user_ns); | |
135 | ns->ucounts = ucounts; | |
136 | ns->pid_allocated = PIDNS_ADDING; | |
137 | INIT_WORK(&ns->proc_work, proc_cleanup_work); | |
138 | ||
139 | return ns; | |
140 | ||
141 | out_free_idr: | |
142 | idr_destroy(&ns->idr); | |
143 | kmem_cache_free(pid_ns_cachep, ns); | |
144 | out_dec: | |
145 | dec_pid_namespaces(ucounts); | |
146 | out: | |
147 | return ERR_PTR(err); | |
148 | } | |
149 | ||
150 | static void delayed_free_pidns(struct rcu_head *p) | |
151 | { | |
152 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); | |
153 | ||
154 | dec_pid_namespaces(ns->ucounts); | |
155 | put_user_ns(ns->user_ns); | |
156 | ||
157 | kmem_cache_free(pid_ns_cachep, ns); | |
158 | } | |
159 | ||
160 | static void destroy_pid_namespace(struct pid_namespace *ns) | |
161 | { | |
162 | ns_free_inum(&ns->ns); | |
163 | ||
164 | idr_destroy(&ns->idr); | |
165 | call_rcu(&ns->rcu, delayed_free_pidns); | |
166 | } | |
167 | ||
168 | struct pid_namespace *copy_pid_ns(unsigned long flags, | |
169 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
170 | { | |
171 | if (!(flags & CLONE_NEWPID)) | |
172 | return get_pid_ns(old_ns); | |
173 | if (task_active_pid_ns(current) != old_ns) | |
174 | return ERR_PTR(-EINVAL); | |
175 | return create_pid_namespace(user_ns, old_ns); | |
176 | } | |
177 | ||
178 | static void free_pid_ns(struct kref *kref) | |
179 | { | |
180 | struct pid_namespace *ns; | |
181 | ||
182 | ns = container_of(kref, struct pid_namespace, kref); | |
183 | destroy_pid_namespace(ns); | |
184 | } | |
185 | ||
186 | void put_pid_ns(struct pid_namespace *ns) | |
187 | { | |
188 | struct pid_namespace *parent; | |
189 | ||
190 | while (ns != &init_pid_ns) { | |
191 | parent = ns->parent; | |
192 | if (!kref_put(&ns->kref, free_pid_ns)) | |
193 | break; | |
194 | ns = parent; | |
195 | } | |
196 | } | |
197 | EXPORT_SYMBOL_GPL(put_pid_ns); | |
198 | ||
199 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
200 | { | |
201 | int nr; | |
202 | int rc; | |
203 | struct task_struct *task, *me = current; | |
204 | int init_pids = thread_group_leader(me) ? 1 : 2; | |
205 | struct pid *pid; | |
206 | ||
207 | /* Don't allow any more processes into the pid namespace */ | |
208 | disable_pid_allocation(pid_ns); | |
209 | ||
210 | /* | |
211 | * Ignore SIGCHLD causing any terminated children to autoreap. | |
212 | * This speeds up the namespace shutdown, plus see the comment | |
213 | * below. | |
214 | */ | |
215 | spin_lock_irq(&me->sighand->siglock); | |
216 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
217 | spin_unlock_irq(&me->sighand->siglock); | |
218 | ||
219 | /* | |
220 | * The last thread in the cgroup-init thread group is terminating. | |
221 | * Find remaining pid_ts in the namespace, signal and wait for them | |
222 | * to exit. | |
223 | * | |
224 | * Note: This signals each threads in the namespace - even those that | |
225 | * belong to the same thread group, To avoid this, we would have | |
226 | * to walk the entire tasklist looking a processes in this | |
227 | * namespace, but that could be unnecessarily expensive if the | |
228 | * pid namespace has just a few processes. Or we need to | |
229 | * maintain a tasklist for each pid namespace. | |
230 | * | |
231 | */ | |
232 | rcu_read_lock(); | |
233 | read_lock(&tasklist_lock); | |
234 | nr = 2; | |
235 | idr_for_each_entry_continue(&pid_ns->idr, pid, nr) { | |
236 | task = pid_task(pid, PIDTYPE_PID); | |
237 | if (task && !__fatal_signal_pending(task)) | |
238 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); | |
239 | } | |
240 | read_unlock(&tasklist_lock); | |
241 | rcu_read_unlock(); | |
242 | ||
243 | /* | |
244 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. | |
245 | * sys_wait4() will also block until our children traced from the | |
246 | * parent namespace are detached and become EXIT_DEAD. | |
247 | */ | |
248 | do { | |
249 | clear_thread_flag(TIF_SIGPENDING); | |
250 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
251 | } while (rc != -ECHILD); | |
252 | ||
253 | /* | |
254 | * sys_wait4() above can't reap the EXIT_DEAD children but we do not | |
255 | * really care, we could reparent them to the global init. We could | |
256 | * exit and reap ->child_reaper even if it is not the last thread in | |
257 | * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(), | |
258 | * pid_ns can not go away until proc_kill_sb() drops the reference. | |
259 | * | |
260 | * But this ns can also have other tasks injected by setns()+fork(). | |
261 | * Again, ignoring the user visible semantics we do not really need | |
262 | * to wait until they are all reaped, but they can be reparented to | |
263 | * us and thus we need to ensure that pid->child_reaper stays valid | |
264 | * until they all go away. See free_pid()->wake_up_process(). | |
265 | * | |
266 | * We rely on ignored SIGCHLD, an injected zombie must be autoreaped | |
267 | * if reparented. | |
268 | */ | |
269 | for (;;) { | |
270 | set_current_state(TASK_INTERRUPTIBLE); | |
271 | if (pid_ns->pid_allocated == init_pids) | |
272 | break; | |
273 | schedule(); | |
274 | } | |
275 | __set_current_state(TASK_RUNNING); | |
276 | ||
277 | if (pid_ns->reboot) | |
278 | current->signal->group_exit_code = pid_ns->reboot; | |
279 | ||
280 | acct_exit_ns(pid_ns); | |
281 | return; | |
282 | } | |
283 | ||
284 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
285 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, | |
286 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
287 | { | |
288 | struct pid_namespace *pid_ns = task_active_pid_ns(current); | |
289 | struct ctl_table tmp = *table; | |
290 | int ret, next; | |
291 | ||
292 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) | |
293 | return -EPERM; | |
294 | ||
295 | /* | |
296 | * Writing directly to ns' last_pid field is OK, since this field | |
297 | * is volatile in a living namespace anyway and a code writing to | |
298 | * it should synchronize its usage with external means. | |
299 | */ | |
300 | ||
301 | next = idr_get_cursor(&pid_ns->idr) - 1; | |
302 | ||
303 | tmp.data = &next; | |
304 | ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); | |
305 | if (!ret && write) | |
306 | idr_set_cursor(&pid_ns->idr, next + 1); | |
307 | ||
308 | return ret; | |
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 struct ns_common *pidns_for_children_get(struct task_struct *task) | |
375 | { | |
376 | struct pid_namespace *ns = NULL; | |
377 | ||
378 | task_lock(task); | |
379 | if (task->nsproxy) { | |
380 | ns = task->nsproxy->pid_ns_for_children; | |
381 | get_pid_ns(ns); | |
382 | } | |
383 | task_unlock(task); | |
384 | ||
385 | if (ns) { | |
386 | read_lock(&tasklist_lock); | |
387 | if (!ns->child_reaper) { | |
388 | put_pid_ns(ns); | |
389 | ns = NULL; | |
390 | } | |
391 | read_unlock(&tasklist_lock); | |
392 | } | |
393 | ||
394 | return ns ? &ns->ns : NULL; | |
395 | } | |
396 | ||
397 | static void pidns_put(struct ns_common *ns) | |
398 | { | |
399 | put_pid_ns(to_pid_ns(ns)); | |
400 | } | |
401 | ||
402 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) | |
403 | { | |
404 | struct pid_namespace *active = task_active_pid_ns(current); | |
405 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); | |
406 | ||
407 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || | |
408 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
409 | return -EPERM; | |
410 | ||
411 | /* | |
412 | * Only allow entering the current active pid namespace | |
413 | * or a child of the current active pid namespace. | |
414 | * | |
415 | * This is required for fork to return a usable pid value and | |
416 | * this maintains the property that processes and their | |
417 | * children can not escape their current pid namespace. | |
418 | */ | |
419 | if (new->level < active->level) | |
420 | return -EINVAL; | |
421 | ||
422 | ancestor = new; | |
423 | while (ancestor->level > active->level) | |
424 | ancestor = ancestor->parent; | |
425 | if (ancestor != active) | |
426 | return -EINVAL; | |
427 | ||
428 | put_pid_ns(nsproxy->pid_ns_for_children); | |
429 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
430 | return 0; | |
431 | } | |
432 | ||
433 | static struct ns_common *pidns_get_parent(struct ns_common *ns) | |
434 | { | |
435 | struct pid_namespace *active = task_active_pid_ns(current); | |
436 | struct pid_namespace *pid_ns, *p; | |
437 | ||
438 | /* See if the parent is in the current namespace */ | |
439 | pid_ns = p = to_pid_ns(ns)->parent; | |
440 | for (;;) { | |
441 | if (!p) | |
442 | return ERR_PTR(-EPERM); | |
443 | if (p == active) | |
444 | break; | |
445 | p = p->parent; | |
446 | } | |
447 | ||
448 | return &get_pid_ns(pid_ns)->ns; | |
449 | } | |
450 | ||
451 | static struct user_namespace *pidns_owner(struct ns_common *ns) | |
452 | { | |
453 | return to_pid_ns(ns)->user_ns; | |
454 | } | |
455 | ||
456 | const struct proc_ns_operations pidns_operations = { | |
457 | .name = "pid", | |
458 | .type = CLONE_NEWPID, | |
459 | .get = pidns_get, | |
460 | .put = pidns_put, | |
461 | .install = pidns_install, | |
462 | .owner = pidns_owner, | |
463 | .get_parent = pidns_get_parent, | |
464 | }; | |
465 | ||
466 | const struct proc_ns_operations pidns_for_children_operations = { | |
467 | .name = "pid_for_children", | |
468 | .real_ns_name = "pid", | |
469 | .type = CLONE_NEWPID, | |
470 | .get = pidns_for_children_get, | |
471 | .put = pidns_put, | |
472 | .install = pidns_install, | |
473 | .owner = pidns_owner, | |
474 | .get_parent = pidns_get_parent, | |
475 | }; | |
476 | ||
477 | static __init int pid_namespaces_init(void) | |
478 | { | |
479 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
480 | ||
481 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
482 | register_sysctl_paths(kern_path, pid_ns_ctl_table); | |
483 | #endif | |
484 | return 0; | |
485 | } | |
486 | ||
487 | __initcall(pid_namespaces_init); |