]>
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> |
74bd59bb | 22 | |
74bd59bb PE |
23 | struct pid_cache { |
24 | int nr_ids; | |
25 | char name[16]; | |
26 | struct kmem_cache *cachep; | |
27 | struct list_head list; | |
28 | }; | |
29 | ||
30 | static LIST_HEAD(pid_caches_lh); | |
31 | static DEFINE_MUTEX(pid_caches_mutex); | |
32 | static struct kmem_cache *pid_ns_cachep; | |
33 | ||
34 | /* | |
35 | * creates the kmem cache to allocate pids from. | |
36 | * @nr_ids: the number of numerical ids this pid will have to carry | |
37 | */ | |
38 | ||
39 | static struct kmem_cache *create_pid_cachep(int nr_ids) | |
40 | { | |
41 | struct pid_cache *pcache; | |
42 | struct kmem_cache *cachep; | |
43 | ||
44 | mutex_lock(&pid_caches_mutex); | |
45 | list_for_each_entry(pcache, &pid_caches_lh, list) | |
46 | if (pcache->nr_ids == nr_ids) | |
47 | goto out; | |
48 | ||
49 | pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); | |
50 | if (pcache == NULL) | |
51 | goto err_alloc; | |
52 | ||
53 | snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); | |
54 | cachep = kmem_cache_create(pcache->name, | |
55 | sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid), | |
56 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
57 | if (cachep == NULL) | |
58 | goto err_cachep; | |
59 | ||
60 | pcache->nr_ids = nr_ids; | |
61 | pcache->cachep = cachep; | |
62 | list_add(&pcache->list, &pid_caches_lh); | |
63 | out: | |
64 | mutex_unlock(&pid_caches_mutex); | |
65 | return pcache->cachep; | |
66 | ||
67 | err_cachep: | |
68 | kfree(pcache); | |
69 | err_alloc: | |
70 | mutex_unlock(&pid_caches_mutex); | |
71 | return NULL; | |
72 | } | |
73 | ||
0a01f2cc EB |
74 | static void proc_cleanup_work(struct work_struct *work) |
75 | { | |
76 | struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); | |
77 | pid_ns_release_proc(ns); | |
78 | } | |
79 | ||
f2302505 AV |
80 | /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ |
81 | #define MAX_PID_NS_LEVEL 32 | |
82 | ||
f333c700 EB |
83 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) |
84 | { | |
85 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); | |
86 | } | |
87 | ||
88 | static void dec_pid_namespaces(struct ucounts *ucounts) | |
89 | { | |
90 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); | |
91 | } | |
92 | ||
49f4d8b9 EB |
93 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
94 | struct pid_namespace *parent_pid_ns) | |
74bd59bb PE |
95 | { |
96 | struct pid_namespace *ns; | |
ed469a63 | 97 | unsigned int level = parent_pid_ns->level + 1; |
f333c700 | 98 | struct ucounts *ucounts; |
f2302505 AV |
99 | int i; |
100 | int err; | |
101 | ||
df75e774 | 102 | err = -ENOSPC; |
f333c700 EB |
103 | if (level > MAX_PID_NS_LEVEL) |
104 | goto out; | |
105 | ucounts = inc_pid_namespaces(user_ns); | |
106 | if (!ucounts) | |
f2302505 | 107 | goto out; |
74bd59bb | 108 | |
f2302505 | 109 | err = -ENOMEM; |
84406c15 | 110 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb | 111 | if (ns == NULL) |
f333c700 | 112 | goto out_dec; |
74bd59bb PE |
113 | |
114 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
115 | if (!ns->pidmap[0].page) | |
116 | goto out_free; | |
117 | ||
118 | ns->pid_cachep = create_pid_cachep(level + 1); | |
119 | if (ns->pid_cachep == NULL) | |
120 | goto out_free_map; | |
121 | ||
6344c433 | 122 | err = ns_alloc_inum(&ns->ns); |
98f842e6 EB |
123 | if (err) |
124 | goto out_free_map; | |
33c42940 | 125 | ns->ns.ops = &pidns_operations; |
98f842e6 | 126 | |
74bd59bb | 127 | kref_init(&ns->kref); |
74bd59bb | 128 | ns->level = level; |
ed469a63 | 129 | ns->parent = get_pid_ns(parent_pid_ns); |
49f4d8b9 | 130 | ns->user_ns = get_user_ns(user_ns); |
f333c700 | 131 | ns->ucounts = ucounts; |
c876ad76 | 132 | ns->nr_hashed = PIDNS_HASH_ADDING; |
0a01f2cc | 133 | INIT_WORK(&ns->proc_work, proc_cleanup_work); |
74bd59bb PE |
134 | |
135 | set_bit(0, ns->pidmap[0].page); | |
136 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); | |
137 | ||
84406c15 | 138 | for (i = 1; i < PIDMAP_ENTRIES; i++) |
74bd59bb | 139 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); |
74bd59bb PE |
140 | |
141 | return ns; | |
142 | ||
143 | out_free_map: | |
144 | kfree(ns->pidmap[0].page); | |
145 | out_free: | |
146 | kmem_cache_free(pid_ns_cachep, ns); | |
f333c700 EB |
147 | out_dec: |
148 | dec_pid_namespaces(ucounts); | |
74bd59bb | 149 | out: |
4308eebb | 150 | return ERR_PTR(err); |
74bd59bb PE |
151 | } |
152 | ||
1adfcb03 AV |
153 | static void delayed_free_pidns(struct rcu_head *p) |
154 | { | |
add7c65c AV |
155 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); |
156 | ||
157 | dec_pid_namespaces(ns->ucounts); | |
158 | put_user_ns(ns->user_ns); | |
159 | ||
160 | kmem_cache_free(pid_ns_cachep, ns); | |
1adfcb03 AV |
161 | } |
162 | ||
74bd59bb PE |
163 | static void destroy_pid_namespace(struct pid_namespace *ns) |
164 | { | |
165 | int i; | |
166 | ||
6344c433 | 167 | ns_free_inum(&ns->ns); |
74bd59bb PE |
168 | for (i = 0; i < PIDMAP_ENTRIES; i++) |
169 | kfree(ns->pidmap[i].page); | |
1adfcb03 | 170 | call_rcu(&ns->rcu, delayed_free_pidns); |
74bd59bb PE |
171 | } |
172 | ||
49f4d8b9 EB |
173 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
174 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
74bd59bb | 175 | { |
74bd59bb | 176 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 177 | return get_pid_ns(old_ns); |
225778d6 EB |
178 | if (task_active_pid_ns(current) != old_ns) |
179 | return ERR_PTR(-EINVAL); | |
49f4d8b9 | 180 | return create_pid_namespace(user_ns, old_ns); |
74bd59bb PE |
181 | } |
182 | ||
bbc2e3ef | 183 | static void free_pid_ns(struct kref *kref) |
74bd59bb | 184 | { |
bbc2e3ef | 185 | struct pid_namespace *ns; |
74bd59bb PE |
186 | |
187 | ns = container_of(kref, struct pid_namespace, kref); | |
74bd59bb | 188 | destroy_pid_namespace(ns); |
bbc2e3ef | 189 | } |
74bd59bb | 190 | |
bbc2e3ef CG |
191 | void put_pid_ns(struct pid_namespace *ns) |
192 | { | |
193 | struct pid_namespace *parent; | |
194 | ||
195 | while (ns != &init_pid_ns) { | |
196 | parent = ns->parent; | |
197 | if (!kref_put(&ns->kref, free_pid_ns)) | |
198 | break; | |
199 | ns = parent; | |
200 | } | |
74bd59bb | 201 | } |
bbc2e3ef | 202 | EXPORT_SYMBOL_GPL(put_pid_ns); |
74bd59bb PE |
203 | |
204 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
205 | { | |
206 | int nr; | |
207 | int rc; | |
00c10bc1 | 208 | struct task_struct *task, *me = current; |
751c644b | 209 | int init_pids = thread_group_leader(me) ? 1 : 2; |
00c10bc1 | 210 | |
c876ad76 EB |
211 | /* Don't allow any more processes into the pid namespace */ |
212 | disable_pid_allocation(pid_ns); | |
213 | ||
a53b8315 ON |
214 | /* |
215 | * Ignore SIGCHLD causing any terminated children to autoreap. | |
216 | * This speeds up the namespace shutdown, plus see the comment | |
217 | * below. | |
218 | */ | |
00c10bc1 EB |
219 | spin_lock_irq(&me->sighand->siglock); |
220 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
221 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
222 | |
223 | /* | |
224 | * The last thread in the cgroup-init thread group is terminating. | |
225 | * Find remaining pid_ts in the namespace, signal and wait for them | |
226 | * to exit. | |
227 | * | |
228 | * Note: This signals each threads in the namespace - even those that | |
229 | * belong to the same thread group, To avoid this, we would have | |
230 | * to walk the entire tasklist looking a processes in this | |
231 | * namespace, but that could be unnecessarily expensive if the | |
232 | * pid namespace has just a few processes. Or we need to | |
233 | * maintain a tasklist for each pid namespace. | |
234 | * | |
235 | */ | |
236 | read_lock(&tasklist_lock); | |
237 | nr = next_pidmap(pid_ns, 1); | |
238 | while (nr > 0) { | |
e4da026f SB |
239 | rcu_read_lock(); |
240 | ||
e4da026f | 241 | task = pid_task(find_vpid(nr), PIDTYPE_PID); |
a02d6fd6 ON |
242 | if (task && !__fatal_signal_pending(task)) |
243 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); | |
e4da026f SB |
244 | |
245 | rcu_read_unlock(); | |
246 | ||
74bd59bb PE |
247 | nr = next_pidmap(pid_ns, nr); |
248 | } | |
249 | read_unlock(&tasklist_lock); | |
250 | ||
a53b8315 ON |
251 | /* |
252 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. | |
253 | * sys_wait4() will also block until our children traced from the | |
254 | * parent namespace are detached and become EXIT_DEAD. | |
255 | */ | |
74bd59bb PE |
256 | do { |
257 | clear_thread_flag(TIF_SIGPENDING); | |
258 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
259 | } while (rc != -ECHILD); | |
260 | ||
6347e900 | 261 | /* |
a53b8315 ON |
262 | * sys_wait4() above can't reap the EXIT_DEAD children but we do not |
263 | * really care, we could reparent them to the global init. We could | |
264 | * exit and reap ->child_reaper even if it is not the last thread in | |
265 | * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(), | |
266 | * pid_ns can not go away until proc_kill_sb() drops the reference. | |
267 | * | |
268 | * But this ns can also have other tasks injected by setns()+fork(). | |
269 | * Again, ignoring the user visible semantics we do not really need | |
270 | * to wait until they are all reaped, but they can be reparented to | |
271 | * us and thus we need to ensure that pid->child_reaper stays valid | |
272 | * until they all go away. See free_pid()->wake_up_process(). | |
273 | * | |
274 | * We rely on ignored SIGCHLD, an injected zombie must be autoreaped | |
275 | * if reparented. | |
6347e900 EB |
276 | */ |
277 | for (;;) { | |
af4b8a83 | 278 | set_current_state(TASK_UNINTERRUPTIBLE); |
751c644b | 279 | if (pid_ns->nr_hashed == init_pids) |
6347e900 EB |
280 | break; |
281 | schedule(); | |
282 | } | |
af4b8a83 | 283 | __set_current_state(TASK_RUNNING); |
6347e900 | 284 | |
cf3f8921 DL |
285 | if (pid_ns->reboot) |
286 | current->signal->group_exit_code = pid_ns->reboot; | |
287 | ||
0b6b030f | 288 | acct_exit_ns(pid_ns); |
74bd59bb PE |
289 | return; |
290 | } | |
291 | ||
98ed57ee | 292 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 PE |
293 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
294 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
295 | { | |
49f4d8b9 | 296 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
b8f566b0 PE |
297 | struct ctl_table tmp = *table; |
298 | ||
49f4d8b9 | 299 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) |
b8f566b0 PE |
300 | return -EPERM; |
301 | ||
302 | /* | |
303 | * Writing directly to ns' last_pid field is OK, since this field | |
304 | * is volatile in a living namespace anyway and a code writing to | |
305 | * it should synchronize its usage with external means. | |
306 | */ | |
307 | ||
49f4d8b9 | 308 | tmp.data = &pid_ns->last_pid; |
579035dc | 309 | return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); |
b8f566b0 PE |
310 | } |
311 | ||
579035dc AV |
312 | extern int pid_max; |
313 | static int zero = 0; | |
b8f566b0 PE |
314 | static struct ctl_table pid_ns_ctl_table[] = { |
315 | { | |
316 | .procname = "ns_last_pid", | |
317 | .maxlen = sizeof(int), | |
318 | .mode = 0666, /* permissions are checked in the handler */ | |
319 | .proc_handler = pid_ns_ctl_handler, | |
579035dc AV |
320 | .extra1 = &zero, |
321 | .extra2 = &pid_max, | |
b8f566b0 PE |
322 | }, |
323 | { } | |
324 | }; | |
b8f566b0 | 325 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 326 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 327 | |
cf3f8921 DL |
328 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
329 | { | |
330 | if (pid_ns == &init_pid_ns) | |
331 | return 0; | |
332 | ||
333 | switch (cmd) { | |
334 | case LINUX_REBOOT_CMD_RESTART2: | |
335 | case LINUX_REBOOT_CMD_RESTART: | |
336 | pid_ns->reboot = SIGHUP; | |
337 | break; | |
338 | ||
339 | case LINUX_REBOOT_CMD_POWER_OFF: | |
340 | case LINUX_REBOOT_CMD_HALT: | |
341 | pid_ns->reboot = SIGINT; | |
342 | break; | |
343 | default: | |
344 | return -EINVAL; | |
345 | } | |
346 | ||
347 | read_lock(&tasklist_lock); | |
348 | force_sig(SIGKILL, pid_ns->child_reaper); | |
349 | read_unlock(&tasklist_lock); | |
350 | ||
351 | do_exit(0); | |
352 | ||
353 | /* Not reached */ | |
354 | return 0; | |
355 | } | |
356 | ||
3c041184 AV |
357 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
358 | { | |
359 | return container_of(ns, struct pid_namespace, ns); | |
360 | } | |
361 | ||
64964528 | 362 | static struct ns_common *pidns_get(struct task_struct *task) |
57e8391d EB |
363 | { |
364 | struct pid_namespace *ns; | |
365 | ||
366 | rcu_read_lock(); | |
d2308225 ON |
367 | ns = task_active_pid_ns(task); |
368 | if (ns) | |
369 | get_pid_ns(ns); | |
57e8391d EB |
370 | rcu_read_unlock(); |
371 | ||
3c041184 | 372 | return ns ? &ns->ns : NULL; |
57e8391d EB |
373 | } |
374 | ||
64964528 | 375 | static void pidns_put(struct ns_common *ns) |
57e8391d | 376 | { |
3c041184 | 377 | put_pid_ns(to_pid_ns(ns)); |
57e8391d EB |
378 | } |
379 | ||
64964528 | 380 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
57e8391d EB |
381 | { |
382 | struct pid_namespace *active = task_active_pid_ns(current); | |
3c041184 | 383 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
57e8391d | 384 | |
5e4a0847 | 385 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
c7b96acf | 386 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
57e8391d EB |
387 | return -EPERM; |
388 | ||
389 | /* | |
390 | * Only allow entering the current active pid namespace | |
391 | * or a child of the current active pid namespace. | |
392 | * | |
393 | * This is required for fork to return a usable pid value and | |
394 | * this maintains the property that processes and their | |
395 | * children can not escape their current pid namespace. | |
396 | */ | |
397 | if (new->level < active->level) | |
398 | return -EINVAL; | |
399 | ||
400 | ancestor = new; | |
401 | while (ancestor->level > active->level) | |
402 | ancestor = ancestor->parent; | |
403 | if (ancestor != active) | |
404 | return -EINVAL; | |
405 | ||
c2b1df2e AL |
406 | put_pid_ns(nsproxy->pid_ns_for_children); |
407 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
57e8391d EB |
408 | return 0; |
409 | } | |
410 | ||
a7306ed8 AV |
411 | static struct ns_common *pidns_get_parent(struct ns_common *ns) |
412 | { | |
413 | struct pid_namespace *active = task_active_pid_ns(current); | |
414 | struct pid_namespace *pid_ns, *p; | |
415 | ||
416 | /* See if the parent is in the current namespace */ | |
417 | pid_ns = p = to_pid_ns(ns)->parent; | |
418 | for (;;) { | |
419 | if (!p) | |
420 | return ERR_PTR(-EPERM); | |
421 | if (p == active) | |
422 | break; | |
423 | p = p->parent; | |
424 | } | |
425 | ||
426 | return &get_pid_ns(pid_ns)->ns; | |
427 | } | |
428 | ||
bcac25a5 AV |
429 | static struct user_namespace *pidns_owner(struct ns_common *ns) |
430 | { | |
431 | return to_pid_ns(ns)->user_ns; | |
432 | } | |
433 | ||
57e8391d EB |
434 | const struct proc_ns_operations pidns_operations = { |
435 | .name = "pid", | |
436 | .type = CLONE_NEWPID, | |
437 | .get = pidns_get, | |
438 | .put = pidns_put, | |
439 | .install = pidns_install, | |
bcac25a5 | 440 | .owner = pidns_owner, |
a7306ed8 | 441 | .get_parent = pidns_get_parent, |
57e8391d EB |
442 | }; |
443 | ||
74bd59bb PE |
444 | static __init int pid_namespaces_init(void) |
445 | { | |
446 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
447 | |
448 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 449 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 450 | #endif |
74bd59bb PE |
451 | return 0; |
452 | } | |
453 | ||
454 | __initcall(pid_namespaces_init); |