]>
Commit | Line | Data |
---|---|---|
74bd59bb PE |
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> | |
49f4d8b9 | 13 | #include <linux/user_namespace.h> |
74bd59bb | 14 | #include <linux/syscalls.h> |
5b825c3a | 15 | #include <linux/cred.h> |
74bd59bb | 16 | #include <linux/err.h> |
0b6b030f | 17 | #include <linux/acct.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
0bb80f24 | 19 | #include <linux/proc_ns.h> |
cf3f8921 | 20 | #include <linux/reboot.h> |
523a6a94 | 21 | #include <linux/export.h> |
29930025 | 22 | #include <linux/sched/task.h> |
74bd59bb | 23 | |
74bd59bb PE |
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 | ||
0a01f2cc EB |
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 | ||
f2302505 AV |
81 | /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ |
82 | #define MAX_PID_NS_LEVEL 32 | |
83 | ||
f333c700 EB |
84 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) |
85 | { | |
86 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); | |
87 | } | |
88 | ||
89 | static void dec_pid_namespaces(struct ucounts *ucounts) | |
90 | { | |
91 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); | |
92 | } | |
93 | ||
49f4d8b9 EB |
94 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
95 | struct pid_namespace *parent_pid_ns) | |
74bd59bb PE |
96 | { |
97 | struct pid_namespace *ns; | |
ed469a63 | 98 | unsigned int level = parent_pid_ns->level + 1; |
f333c700 | 99 | struct ucounts *ucounts; |
f2302505 AV |
100 | int i; |
101 | int err; | |
102 | ||
df75e774 | 103 | err = -ENOSPC; |
f333c700 EB |
104 | if (level > MAX_PID_NS_LEVEL) |
105 | goto out; | |
106 | ucounts = inc_pid_namespaces(user_ns); | |
107 | if (!ucounts) | |
f2302505 | 108 | goto out; |
74bd59bb | 109 | |
f2302505 | 110 | err = -ENOMEM; |
84406c15 | 111 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb | 112 | if (ns == NULL) |
f333c700 | 113 | goto out_dec; |
74bd59bb PE |
114 | |
115 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
116 | if (!ns->pidmap[0].page) | |
117 | goto out_free; | |
118 | ||
119 | ns->pid_cachep = create_pid_cachep(level + 1); | |
120 | if (ns->pid_cachep == NULL) | |
121 | goto out_free_map; | |
122 | ||
6344c433 | 123 | err = ns_alloc_inum(&ns->ns); |
98f842e6 EB |
124 | if (err) |
125 | goto out_free_map; | |
33c42940 | 126 | ns->ns.ops = &pidns_operations; |
98f842e6 | 127 | |
74bd59bb | 128 | kref_init(&ns->kref); |
74bd59bb | 129 | ns->level = level; |
ed469a63 | 130 | ns->parent = get_pid_ns(parent_pid_ns); |
49f4d8b9 | 131 | ns->user_ns = get_user_ns(user_ns); |
f333c700 | 132 | ns->ucounts = ucounts; |
c876ad76 | 133 | ns->nr_hashed = PIDNS_HASH_ADDING; |
0a01f2cc | 134 | INIT_WORK(&ns->proc_work, proc_cleanup_work); |
74bd59bb PE |
135 | |
136 | set_bit(0, ns->pidmap[0].page); | |
137 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); | |
138 | ||
84406c15 | 139 | for (i = 1; i < PIDMAP_ENTRIES; i++) |
74bd59bb | 140 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); |
74bd59bb PE |
141 | |
142 | return ns; | |
143 | ||
144 | out_free_map: | |
145 | kfree(ns->pidmap[0].page); | |
146 | out_free: | |
147 | kmem_cache_free(pid_ns_cachep, ns); | |
f333c700 EB |
148 | out_dec: |
149 | dec_pid_namespaces(ucounts); | |
74bd59bb | 150 | out: |
4308eebb | 151 | return ERR_PTR(err); |
74bd59bb PE |
152 | } |
153 | ||
1adfcb03 AV |
154 | static void delayed_free_pidns(struct rcu_head *p) |
155 | { | |
add7c65c AV |
156 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); |
157 | ||
158 | dec_pid_namespaces(ns->ucounts); | |
159 | put_user_ns(ns->user_ns); | |
160 | ||
161 | kmem_cache_free(pid_ns_cachep, ns); | |
1adfcb03 AV |
162 | } |
163 | ||
74bd59bb PE |
164 | static void destroy_pid_namespace(struct pid_namespace *ns) |
165 | { | |
166 | int i; | |
167 | ||
6344c433 | 168 | ns_free_inum(&ns->ns); |
74bd59bb PE |
169 | for (i = 0; i < PIDMAP_ENTRIES; i++) |
170 | kfree(ns->pidmap[i].page); | |
1adfcb03 | 171 | call_rcu(&ns->rcu, delayed_free_pidns); |
74bd59bb PE |
172 | } |
173 | ||
49f4d8b9 EB |
174 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
175 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
74bd59bb | 176 | { |
74bd59bb | 177 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 178 | return get_pid_ns(old_ns); |
225778d6 EB |
179 | if (task_active_pid_ns(current) != old_ns) |
180 | return ERR_PTR(-EINVAL); | |
49f4d8b9 | 181 | return create_pid_namespace(user_ns, old_ns); |
74bd59bb PE |
182 | } |
183 | ||
bbc2e3ef | 184 | static void free_pid_ns(struct kref *kref) |
74bd59bb | 185 | { |
bbc2e3ef | 186 | struct pid_namespace *ns; |
74bd59bb PE |
187 | |
188 | ns = container_of(kref, struct pid_namespace, kref); | |
74bd59bb | 189 | destroy_pid_namespace(ns); |
bbc2e3ef | 190 | } |
74bd59bb | 191 | |
bbc2e3ef CG |
192 | void put_pid_ns(struct pid_namespace *ns) |
193 | { | |
194 | struct pid_namespace *parent; | |
195 | ||
196 | while (ns != &init_pid_ns) { | |
197 | parent = ns->parent; | |
198 | if (!kref_put(&ns->kref, free_pid_ns)) | |
199 | break; | |
200 | ns = parent; | |
201 | } | |
74bd59bb | 202 | } |
bbc2e3ef | 203 | EXPORT_SYMBOL_GPL(put_pid_ns); |
74bd59bb PE |
204 | |
205 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
206 | { | |
207 | int nr; | |
208 | int rc; | |
00c10bc1 | 209 | struct task_struct *task, *me = current; |
751c644b | 210 | int init_pids = thread_group_leader(me) ? 1 : 2; |
00c10bc1 | 211 | |
c876ad76 EB |
212 | /* Don't allow any more processes into the pid namespace */ |
213 | disable_pid_allocation(pid_ns); | |
214 | ||
a53b8315 ON |
215 | /* |
216 | * Ignore SIGCHLD causing any terminated children to autoreap. | |
217 | * This speeds up the namespace shutdown, plus see the comment | |
218 | * below. | |
219 | */ | |
00c10bc1 EB |
220 | spin_lock_irq(&me->sighand->siglock); |
221 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
222 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
223 | |
224 | /* | |
225 | * The last thread in the cgroup-init thread group is terminating. | |
226 | * Find remaining pid_ts in the namespace, signal and wait for them | |
227 | * to exit. | |
228 | * | |
229 | * Note: This signals each threads in the namespace - even those that | |
230 | * belong to the same thread group, To avoid this, we would have | |
231 | * to walk the entire tasklist looking a processes in this | |
232 | * namespace, but that could be unnecessarily expensive if the | |
233 | * pid namespace has just a few processes. Or we need to | |
234 | * maintain a tasklist for each pid namespace. | |
235 | * | |
236 | */ | |
237 | read_lock(&tasklist_lock); | |
238 | nr = next_pidmap(pid_ns, 1); | |
239 | while (nr > 0) { | |
e4da026f SB |
240 | rcu_read_lock(); |
241 | ||
e4da026f | 242 | task = pid_task(find_vpid(nr), PIDTYPE_PID); |
a02d6fd6 ON |
243 | if (task && !__fatal_signal_pending(task)) |
244 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); | |
e4da026f SB |
245 | |
246 | rcu_read_unlock(); | |
247 | ||
74bd59bb PE |
248 | nr = next_pidmap(pid_ns, nr); |
249 | } | |
250 | read_unlock(&tasklist_lock); | |
251 | ||
a53b8315 ON |
252 | /* |
253 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. | |
254 | * sys_wait4() will also block until our children traced from the | |
255 | * parent namespace are detached and become EXIT_DEAD. | |
256 | */ | |
74bd59bb PE |
257 | do { |
258 | clear_thread_flag(TIF_SIGPENDING); | |
259 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
260 | } while (rc != -ECHILD); | |
261 | ||
6347e900 | 262 | /* |
a53b8315 ON |
263 | * sys_wait4() above can't reap the EXIT_DEAD children but we do not |
264 | * really care, we could reparent them to the global init. We could | |
265 | * exit and reap ->child_reaper even if it is not the last thread in | |
266 | * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(), | |
267 | * pid_ns can not go away until proc_kill_sb() drops the reference. | |
268 | * | |
269 | * But this ns can also have other tasks injected by setns()+fork(). | |
270 | * Again, ignoring the user visible semantics we do not really need | |
271 | * to wait until they are all reaped, but they can be reparented to | |
272 | * us and thus we need to ensure that pid->child_reaper stays valid | |
273 | * until they all go away. See free_pid()->wake_up_process(). | |
274 | * | |
275 | * We rely on ignored SIGCHLD, an injected zombie must be autoreaped | |
276 | * if reparented. | |
6347e900 EB |
277 | */ |
278 | for (;;) { | |
af4b8a83 | 279 | set_current_state(TASK_UNINTERRUPTIBLE); |
751c644b | 280 | if (pid_ns->nr_hashed == init_pids) |
6347e900 EB |
281 | break; |
282 | schedule(); | |
283 | } | |
af4b8a83 | 284 | __set_current_state(TASK_RUNNING); |
6347e900 | 285 | |
cf3f8921 DL |
286 | if (pid_ns->reboot) |
287 | current->signal->group_exit_code = pid_ns->reboot; | |
288 | ||
0b6b030f | 289 | acct_exit_ns(pid_ns); |
74bd59bb PE |
290 | return; |
291 | } | |
292 | ||
98ed57ee | 293 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 PE |
294 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
295 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
296 | { | |
49f4d8b9 | 297 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
b8f566b0 PE |
298 | struct ctl_table tmp = *table; |
299 | ||
49f4d8b9 | 300 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) |
b8f566b0 PE |
301 | return -EPERM; |
302 | ||
303 | /* | |
304 | * Writing directly to ns' last_pid field is OK, since this field | |
305 | * is volatile in a living namespace anyway and a code writing to | |
306 | * it should synchronize its usage with external means. | |
307 | */ | |
308 | ||
49f4d8b9 | 309 | tmp.data = &pid_ns->last_pid; |
579035dc | 310 | return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); |
b8f566b0 PE |
311 | } |
312 | ||
579035dc AV |
313 | extern int pid_max; |
314 | static int zero = 0; | |
b8f566b0 PE |
315 | static struct ctl_table pid_ns_ctl_table[] = { |
316 | { | |
317 | .procname = "ns_last_pid", | |
318 | .maxlen = sizeof(int), | |
319 | .mode = 0666, /* permissions are checked in the handler */ | |
320 | .proc_handler = pid_ns_ctl_handler, | |
579035dc AV |
321 | .extra1 = &zero, |
322 | .extra2 = &pid_max, | |
b8f566b0 PE |
323 | }, |
324 | { } | |
325 | }; | |
b8f566b0 | 326 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 327 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 328 | |
cf3f8921 DL |
329 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
330 | { | |
331 | if (pid_ns == &init_pid_ns) | |
332 | return 0; | |
333 | ||
334 | switch (cmd) { | |
335 | case LINUX_REBOOT_CMD_RESTART2: | |
336 | case LINUX_REBOOT_CMD_RESTART: | |
337 | pid_ns->reboot = SIGHUP; | |
338 | break; | |
339 | ||
340 | case LINUX_REBOOT_CMD_POWER_OFF: | |
341 | case LINUX_REBOOT_CMD_HALT: | |
342 | pid_ns->reboot = SIGINT; | |
343 | break; | |
344 | default: | |
345 | return -EINVAL; | |
346 | } | |
347 | ||
348 | read_lock(&tasklist_lock); | |
349 | force_sig(SIGKILL, pid_ns->child_reaper); | |
350 | read_unlock(&tasklist_lock); | |
351 | ||
352 | do_exit(0); | |
353 | ||
354 | /* Not reached */ | |
355 | return 0; | |
356 | } | |
357 | ||
3c041184 AV |
358 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
359 | { | |
360 | return container_of(ns, struct pid_namespace, ns); | |
361 | } | |
362 | ||
64964528 | 363 | static struct ns_common *pidns_get(struct task_struct *task) |
57e8391d EB |
364 | { |
365 | struct pid_namespace *ns; | |
366 | ||
367 | rcu_read_lock(); | |
d2308225 ON |
368 | ns = task_active_pid_ns(task); |
369 | if (ns) | |
370 | get_pid_ns(ns); | |
57e8391d EB |
371 | rcu_read_unlock(); |
372 | ||
3c041184 | 373 | return ns ? &ns->ns : NULL; |
57e8391d EB |
374 | } |
375 | ||
64964528 | 376 | static void pidns_put(struct ns_common *ns) |
57e8391d | 377 | { |
3c041184 | 378 | put_pid_ns(to_pid_ns(ns)); |
57e8391d EB |
379 | } |
380 | ||
64964528 | 381 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
57e8391d EB |
382 | { |
383 | struct pid_namespace *active = task_active_pid_ns(current); | |
3c041184 | 384 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
57e8391d | 385 | |
5e4a0847 | 386 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
c7b96acf | 387 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
57e8391d EB |
388 | return -EPERM; |
389 | ||
390 | /* | |
391 | * Only allow entering the current active pid namespace | |
392 | * or a child of the current active pid namespace. | |
393 | * | |
394 | * This is required for fork to return a usable pid value and | |
395 | * this maintains the property that processes and their | |
396 | * children can not escape their current pid namespace. | |
397 | */ | |
398 | if (new->level < active->level) | |
399 | return -EINVAL; | |
400 | ||
401 | ancestor = new; | |
402 | while (ancestor->level > active->level) | |
403 | ancestor = ancestor->parent; | |
404 | if (ancestor != active) | |
405 | return -EINVAL; | |
406 | ||
c2b1df2e AL |
407 | put_pid_ns(nsproxy->pid_ns_for_children); |
408 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
57e8391d EB |
409 | return 0; |
410 | } | |
411 | ||
a7306ed8 AV |
412 | static struct ns_common *pidns_get_parent(struct ns_common *ns) |
413 | { | |
414 | struct pid_namespace *active = task_active_pid_ns(current); | |
415 | struct pid_namespace *pid_ns, *p; | |
416 | ||
417 | /* See if the parent is in the current namespace */ | |
418 | pid_ns = p = to_pid_ns(ns)->parent; | |
419 | for (;;) { | |
420 | if (!p) | |
421 | return ERR_PTR(-EPERM); | |
422 | if (p == active) | |
423 | break; | |
424 | p = p->parent; | |
425 | } | |
426 | ||
427 | return &get_pid_ns(pid_ns)->ns; | |
428 | } | |
429 | ||
bcac25a5 AV |
430 | static struct user_namespace *pidns_owner(struct ns_common *ns) |
431 | { | |
432 | return to_pid_ns(ns)->user_ns; | |
433 | } | |
434 | ||
57e8391d EB |
435 | const struct proc_ns_operations pidns_operations = { |
436 | .name = "pid", | |
437 | .type = CLONE_NEWPID, | |
438 | .get = pidns_get, | |
439 | .put = pidns_put, | |
440 | .install = pidns_install, | |
bcac25a5 | 441 | .owner = pidns_owner, |
a7306ed8 | 442 | .get_parent = pidns_get_parent, |
57e8391d EB |
443 | }; |
444 | ||
74bd59bb PE |
445 | static __init int pid_namespaces_init(void) |
446 | { | |
447 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
448 | |
449 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 450 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 451 | #endif |
74bd59bb PE |
452 | return 0; |
453 | } | |
454 | ||
455 | __initcall(pid_namespaces_init); |