]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
74bd59bb PE |
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> | |
49f4d8b9 | 14 | #include <linux/user_namespace.h> |
74bd59bb | 15 | #include <linux/syscalls.h> |
5b825c3a | 16 | #include <linux/cred.h> |
74bd59bb | 17 | #include <linux/err.h> |
0b6b030f | 18 | #include <linux/acct.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
0bb80f24 | 20 | #include <linux/proc_ns.h> |
cf3f8921 | 21 | #include <linux/reboot.h> |
523a6a94 | 22 | #include <linux/export.h> |
29930025 | 23 | #include <linux/sched/task.h> |
f361bf4a | 24 | #include <linux/sched/signal.h> |
95846ecf | 25 | #include <linux/idr.h> |
74bd59bb | 26 | |
74bd59bb PE |
27 | static DEFINE_MUTEX(pid_caches_mutex); |
28 | static struct kmem_cache *pid_ns_cachep; | |
dd206bec AD |
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. */ | |
32 | static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL]; | |
74bd59bb PE |
33 | |
34 | /* | |
35 | * creates the kmem cache to allocate pids from. | |
dd206bec | 36 | * @level: pid namespace level |
74bd59bb PE |
37 | */ |
38 | ||
dd206bec | 39 | static struct kmem_cache *create_pid_cachep(unsigned int level) |
74bd59bb | 40 | { |
dd206bec AD |
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); | |
74bd59bb | 53 | mutex_lock(&pid_caches_mutex); |
dd206bec AD |
54 | /* Name collision forces to do allocation under mutex. */ |
55 | if (!*pkc) | |
d5f2b7a5 VA |
56 | *pkc = kmem_cache_create(name, len, 0, |
57 | SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, 0); | |
74bd59bb | 58 | mutex_unlock(&pid_caches_mutex); |
dd206bec AD |
59 | /* current can fail, but someone else can succeed. */ |
60 | return READ_ONCE(*pkc); | |
74bd59bb PE |
61 | } |
62 | ||
0a01f2cc EB |
63 | static void proc_cleanup_work(struct work_struct *work) |
64 | { | |
65 | struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); | |
66 | pid_ns_release_proc(ns); | |
67 | } | |
68 | ||
f333c700 EB |
69 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) |
70 | { | |
71 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); | |
72 | } | |
73 | ||
74 | static void dec_pid_namespaces(struct ucounts *ucounts) | |
75 | { | |
76 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); | |
77 | } | |
78 | ||
49f4d8b9 EB |
79 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
80 | struct pid_namespace *parent_pid_ns) | |
74bd59bb PE |
81 | { |
82 | struct pid_namespace *ns; | |
ed469a63 | 83 | unsigned int level = parent_pid_ns->level + 1; |
f333c700 | 84 | struct ucounts *ucounts; |
f2302505 AV |
85 | int err; |
86 | ||
a2b42626 EB |
87 | err = -EINVAL; |
88 | if (!in_userns(parent_pid_ns->user_ns, user_ns)) | |
89 | goto out; | |
90 | ||
df75e774 | 91 | err = -ENOSPC; |
f333c700 EB |
92 | if (level > MAX_PID_NS_LEVEL) |
93 | goto out; | |
94 | ucounts = inc_pid_namespaces(user_ns); | |
95 | if (!ucounts) | |
f2302505 | 96 | goto out; |
74bd59bb | 97 | |
f2302505 | 98 | err = -ENOMEM; |
84406c15 | 99 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb | 100 | if (ns == NULL) |
f333c700 | 101 | goto out_dec; |
74bd59bb | 102 | |
95846ecf | 103 | idr_init(&ns->idr); |
74bd59bb | 104 | |
dd206bec | 105 | ns->pid_cachep = create_pid_cachep(level); |
74bd59bb | 106 | if (ns->pid_cachep == NULL) |
95846ecf | 107 | goto out_free_idr; |
74bd59bb | 108 | |
6344c433 | 109 | err = ns_alloc_inum(&ns->ns); |
98f842e6 | 110 | if (err) |
95846ecf | 111 | goto out_free_idr; |
33c42940 | 112 | ns->ns.ops = &pidns_operations; |
98f842e6 | 113 | |
74bd59bb | 114 | kref_init(&ns->kref); |
74bd59bb | 115 | ns->level = level; |
ed469a63 | 116 | ns->parent = get_pid_ns(parent_pid_ns); |
49f4d8b9 | 117 | ns->user_ns = get_user_ns(user_ns); |
f333c700 | 118 | ns->ucounts = ucounts; |
e8cfbc24 | 119 | ns->pid_allocated = PIDNS_ADDING; |
0a01f2cc | 120 | INIT_WORK(&ns->proc_work, proc_cleanup_work); |
74bd59bb | 121 | |
74bd59bb PE |
122 | return ns; |
123 | ||
95846ecf GS |
124 | out_free_idr: |
125 | idr_destroy(&ns->idr); | |
74bd59bb | 126 | kmem_cache_free(pid_ns_cachep, ns); |
f333c700 EB |
127 | out_dec: |
128 | dec_pid_namespaces(ucounts); | |
74bd59bb | 129 | out: |
4308eebb | 130 | return ERR_PTR(err); |
74bd59bb PE |
131 | } |
132 | ||
1adfcb03 AV |
133 | static void delayed_free_pidns(struct rcu_head *p) |
134 | { | |
add7c65c AV |
135 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); |
136 | ||
137 | dec_pid_namespaces(ns->ucounts); | |
138 | put_user_ns(ns->user_ns); | |
139 | ||
140 | kmem_cache_free(pid_ns_cachep, ns); | |
1adfcb03 AV |
141 | } |
142 | ||
74bd59bb PE |
143 | static void destroy_pid_namespace(struct pid_namespace *ns) |
144 | { | |
6344c433 | 145 | ns_free_inum(&ns->ns); |
95846ecf GS |
146 | |
147 | idr_destroy(&ns->idr); | |
1adfcb03 | 148 | call_rcu(&ns->rcu, delayed_free_pidns); |
74bd59bb PE |
149 | } |
150 | ||
49f4d8b9 EB |
151 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
152 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
74bd59bb | 153 | { |
74bd59bb | 154 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 155 | return get_pid_ns(old_ns); |
225778d6 EB |
156 | if (task_active_pid_ns(current) != old_ns) |
157 | return ERR_PTR(-EINVAL); | |
49f4d8b9 | 158 | return create_pid_namespace(user_ns, old_ns); |
74bd59bb PE |
159 | } |
160 | ||
bbc2e3ef | 161 | static void free_pid_ns(struct kref *kref) |
74bd59bb | 162 | { |
bbc2e3ef | 163 | struct pid_namespace *ns; |
74bd59bb PE |
164 | |
165 | ns = container_of(kref, struct pid_namespace, kref); | |
74bd59bb | 166 | destroy_pid_namespace(ns); |
bbc2e3ef | 167 | } |
74bd59bb | 168 | |
bbc2e3ef CG |
169 | void put_pid_ns(struct pid_namespace *ns) |
170 | { | |
171 | struct pid_namespace *parent; | |
172 | ||
173 | while (ns != &init_pid_ns) { | |
174 | parent = ns->parent; | |
175 | if (!kref_put(&ns->kref, free_pid_ns)) | |
176 | break; | |
177 | ns = parent; | |
178 | } | |
74bd59bb | 179 | } |
bbc2e3ef | 180 | EXPORT_SYMBOL_GPL(put_pid_ns); |
74bd59bb PE |
181 | |
182 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
183 | { | |
184 | int nr; | |
185 | int rc; | |
00c10bc1 | 186 | struct task_struct *task, *me = current; |
751c644b | 187 | int init_pids = thread_group_leader(me) ? 1 : 2; |
95846ecf | 188 | struct pid *pid; |
00c10bc1 | 189 | |
c876ad76 EB |
190 | /* Don't allow any more processes into the pid namespace */ |
191 | disable_pid_allocation(pid_ns); | |
192 | ||
a53b8315 ON |
193 | /* |
194 | * Ignore SIGCHLD causing any terminated children to autoreap. | |
195 | * This speeds up the namespace shutdown, plus see the comment | |
196 | * below. | |
197 | */ | |
00c10bc1 EB |
198 | spin_lock_irq(&me->sighand->siglock); |
199 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
200 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
201 | |
202 | /* | |
203 | * The last thread in the cgroup-init thread group is terminating. | |
204 | * Find remaining pid_ts in the namespace, signal and wait for them | |
205 | * to exit. | |
206 | * | |
207 | * Note: This signals each threads in the namespace - even those that | |
208 | * belong to the same thread group, To avoid this, we would have | |
209 | * to walk the entire tasklist looking a processes in this | |
210 | * namespace, but that could be unnecessarily expensive if the | |
211 | * pid namespace has just a few processes. Or we need to | |
212 | * maintain a tasklist for each pid namespace. | |
213 | * | |
214 | */ | |
95846ecf | 215 | rcu_read_lock(); |
74bd59bb | 216 | read_lock(&tasklist_lock); |
95846ecf GS |
217 | nr = 2; |
218 | idr_for_each_entry_continue(&pid_ns->idr, pid, nr) { | |
219 | task = pid_task(pid, PIDTYPE_PID); | |
a02d6fd6 | 220 | if (task && !__fatal_signal_pending(task)) |
82058d66 | 221 | group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX); |
74bd59bb PE |
222 | } |
223 | read_unlock(&tasklist_lock); | |
95846ecf | 224 | rcu_read_unlock(); |
74bd59bb | 225 | |
a53b8315 ON |
226 | /* |
227 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. | |
d300b610 | 228 | * kernel_wait4() will also block until our children traced from the |
a53b8315 ON |
229 | * parent namespace are detached and become EXIT_DEAD. |
230 | */ | |
74bd59bb PE |
231 | do { |
232 | clear_thread_flag(TIF_SIGPENDING); | |
d300b610 | 233 | rc = kernel_wait4(-1, NULL, __WALL, NULL); |
74bd59bb PE |
234 | } while (rc != -ECHILD); |
235 | ||
6347e900 | 236 | /* |
d300b610 | 237 | * kernel_wait4() above can't reap the EXIT_DEAD children but we do not |
a53b8315 ON |
238 | * really care, we could reparent them to the global init. We could |
239 | * exit and reap ->child_reaper even if it is not the last thread in | |
e8cfbc24 | 240 | * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(), |
a53b8315 ON |
241 | * pid_ns can not go away until proc_kill_sb() drops the reference. |
242 | * | |
243 | * But this ns can also have other tasks injected by setns()+fork(). | |
244 | * Again, ignoring the user visible semantics we do not really need | |
245 | * to wait until they are all reaped, but they can be reparented to | |
246 | * us and thus we need to ensure that pid->child_reaper stays valid | |
247 | * until they all go away. See free_pid()->wake_up_process(). | |
248 | * | |
249 | * We rely on ignored SIGCHLD, an injected zombie must be autoreaped | |
250 | * if reparented. | |
6347e900 EB |
251 | */ |
252 | for (;;) { | |
b9a985db | 253 | set_current_state(TASK_INTERRUPTIBLE); |
e8cfbc24 | 254 | if (pid_ns->pid_allocated == init_pids) |
6347e900 EB |
255 | break; |
256 | schedule(); | |
257 | } | |
af4b8a83 | 258 | __set_current_state(TASK_RUNNING); |
6347e900 | 259 | |
cf3f8921 DL |
260 | if (pid_ns->reboot) |
261 | current->signal->group_exit_code = pid_ns->reboot; | |
262 | ||
0b6b030f | 263 | acct_exit_ns(pid_ns); |
74bd59bb PE |
264 | return; |
265 | } | |
266 | ||
98ed57ee | 267 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 PE |
268 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
269 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
270 | { | |
49f4d8b9 | 271 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
b8f566b0 | 272 | struct ctl_table tmp = *table; |
95846ecf | 273 | int ret, next; |
b8f566b0 | 274 | |
49f4d8b9 | 275 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) |
b8f566b0 PE |
276 | return -EPERM; |
277 | ||
278 | /* | |
279 | * Writing directly to ns' last_pid field is OK, since this field | |
280 | * is volatile in a living namespace anyway and a code writing to | |
281 | * it should synchronize its usage with external means. | |
282 | */ | |
283 | ||
95846ecf GS |
284 | next = idr_get_cursor(&pid_ns->idr) - 1; |
285 | ||
286 | tmp.data = &next; | |
287 | ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); | |
288 | if (!ret && write) | |
289 | idr_set_cursor(&pid_ns->idr, next + 1); | |
290 | ||
291 | return ret; | |
b8f566b0 PE |
292 | } |
293 | ||
579035dc | 294 | extern int pid_max; |
b8f566b0 PE |
295 | static struct ctl_table pid_ns_ctl_table[] = { |
296 | { | |
297 | .procname = "ns_last_pid", | |
298 | .maxlen = sizeof(int), | |
299 | .mode = 0666, /* permissions are checked in the handler */ | |
300 | .proc_handler = pid_ns_ctl_handler, | |
eec4844f | 301 | .extra1 = SYSCTL_ZERO, |
579035dc | 302 | .extra2 = &pid_max, |
b8f566b0 PE |
303 | }, |
304 | { } | |
305 | }; | |
b8f566b0 | 306 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 307 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 308 | |
cf3f8921 DL |
309 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
310 | { | |
311 | if (pid_ns == &init_pid_ns) | |
312 | return 0; | |
313 | ||
314 | switch (cmd) { | |
315 | case LINUX_REBOOT_CMD_RESTART2: | |
316 | case LINUX_REBOOT_CMD_RESTART: | |
317 | pid_ns->reboot = SIGHUP; | |
318 | break; | |
319 | ||
320 | case LINUX_REBOOT_CMD_POWER_OFF: | |
321 | case LINUX_REBOOT_CMD_HALT: | |
322 | pid_ns->reboot = SIGINT; | |
323 | break; | |
324 | default: | |
325 | return -EINVAL; | |
326 | } | |
327 | ||
328 | read_lock(&tasklist_lock); | |
f9070dc9 | 329 | send_sig(SIGKILL, pid_ns->child_reaper, 1); |
cf3f8921 DL |
330 | read_unlock(&tasklist_lock); |
331 | ||
332 | do_exit(0); | |
333 | ||
334 | /* Not reached */ | |
335 | return 0; | |
336 | } | |
337 | ||
3c041184 AV |
338 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
339 | { | |
340 | return container_of(ns, struct pid_namespace, ns); | |
341 | } | |
342 | ||
64964528 | 343 | static struct ns_common *pidns_get(struct task_struct *task) |
57e8391d EB |
344 | { |
345 | struct pid_namespace *ns; | |
346 | ||
347 | rcu_read_lock(); | |
d2308225 ON |
348 | ns = task_active_pid_ns(task); |
349 | if (ns) | |
350 | get_pid_ns(ns); | |
57e8391d EB |
351 | rcu_read_unlock(); |
352 | ||
3c041184 | 353 | return ns ? &ns->ns : NULL; |
57e8391d EB |
354 | } |
355 | ||
eaa0d190 KT |
356 | static struct ns_common *pidns_for_children_get(struct task_struct *task) |
357 | { | |
358 | struct pid_namespace *ns = NULL; | |
359 | ||
360 | task_lock(task); | |
361 | if (task->nsproxy) { | |
362 | ns = task->nsproxy->pid_ns_for_children; | |
363 | get_pid_ns(ns); | |
364 | } | |
365 | task_unlock(task); | |
366 | ||
367 | if (ns) { | |
368 | read_lock(&tasklist_lock); | |
369 | if (!ns->child_reaper) { | |
370 | put_pid_ns(ns); | |
371 | ns = NULL; | |
372 | } | |
373 | read_unlock(&tasklist_lock); | |
374 | } | |
375 | ||
376 | return ns ? &ns->ns : NULL; | |
377 | } | |
378 | ||
64964528 | 379 | static void pidns_put(struct ns_common *ns) |
57e8391d | 380 | { |
3c041184 | 381 | put_pid_ns(to_pid_ns(ns)); |
57e8391d EB |
382 | } |
383 | ||
64964528 | 384 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
57e8391d EB |
385 | { |
386 | struct pid_namespace *active = task_active_pid_ns(current); | |
3c041184 | 387 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
57e8391d | 388 | |
5e4a0847 | 389 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
c7b96acf | 390 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
57e8391d EB |
391 | return -EPERM; |
392 | ||
393 | /* | |
394 | * Only allow entering the current active pid namespace | |
395 | * or a child of the current active pid namespace. | |
396 | * | |
397 | * This is required for fork to return a usable pid value and | |
398 | * this maintains the property that processes and their | |
399 | * children can not escape their current pid namespace. | |
400 | */ | |
401 | if (new->level < active->level) | |
402 | return -EINVAL; | |
403 | ||
404 | ancestor = new; | |
405 | while (ancestor->level > active->level) | |
406 | ancestor = ancestor->parent; | |
407 | if (ancestor != active) | |
408 | return -EINVAL; | |
409 | ||
c2b1df2e AL |
410 | put_pid_ns(nsproxy->pid_ns_for_children); |
411 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
57e8391d EB |
412 | return 0; |
413 | } | |
414 | ||
a7306ed8 AV |
415 | static struct ns_common *pidns_get_parent(struct ns_common *ns) |
416 | { | |
417 | struct pid_namespace *active = task_active_pid_ns(current); | |
418 | struct pid_namespace *pid_ns, *p; | |
419 | ||
420 | /* See if the parent is in the current namespace */ | |
421 | pid_ns = p = to_pid_ns(ns)->parent; | |
422 | for (;;) { | |
423 | if (!p) | |
424 | return ERR_PTR(-EPERM); | |
425 | if (p == active) | |
426 | break; | |
427 | p = p->parent; | |
428 | } | |
429 | ||
430 | return &get_pid_ns(pid_ns)->ns; | |
431 | } | |
432 | ||
bcac25a5 AV |
433 | static struct user_namespace *pidns_owner(struct ns_common *ns) |
434 | { | |
435 | return to_pid_ns(ns)->user_ns; | |
436 | } | |
437 | ||
57e8391d EB |
438 | const struct proc_ns_operations pidns_operations = { |
439 | .name = "pid", | |
440 | .type = CLONE_NEWPID, | |
441 | .get = pidns_get, | |
442 | .put = pidns_put, | |
443 | .install = pidns_install, | |
bcac25a5 | 444 | .owner = pidns_owner, |
a7306ed8 | 445 | .get_parent = pidns_get_parent, |
57e8391d EB |
446 | }; |
447 | ||
eaa0d190 KT |
448 | const struct proc_ns_operations pidns_for_children_operations = { |
449 | .name = "pid_for_children", | |
450 | .real_ns_name = "pid", | |
451 | .type = CLONE_NEWPID, | |
452 | .get = pidns_for_children_get, | |
453 | .put = pidns_put, | |
454 | .install = pidns_install, | |
455 | .owner = pidns_owner, | |
456 | .get_parent = pidns_get_parent, | |
457 | }; | |
458 | ||
74bd59bb PE |
459 | static __init int pid_namespaces_init(void) |
460 | { | |
461 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
462 | |
463 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 464 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 465 | #endif |
74bd59bb PE |
466 | return 0; |
467 | } | |
468 | ||
469 | __initcall(pid_namespaces_init); |