]>
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 | /* Write once array, filled from the beginning. */ |
30 | static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL]; | |
74bd59bb PE |
31 | |
32 | /* | |
33 | * creates the kmem cache to allocate pids from. | |
dd206bec | 34 | * @level: pid namespace level |
74bd59bb PE |
35 | */ |
36 | ||
dd206bec | 37 | static struct kmem_cache *create_pid_cachep(unsigned int level) |
74bd59bb | 38 | { |
dd206bec AD |
39 | /* Level 0 is init_pid_ns.pid_cachep */ |
40 | struct kmem_cache **pkc = &pid_cache[level - 1]; | |
41 | struct kmem_cache *kc; | |
42 | char name[4 + 10 + 1]; | |
43 | unsigned int len; | |
44 | ||
45 | kc = READ_ONCE(*pkc); | |
46 | if (kc) | |
47 | return kc; | |
48 | ||
49 | snprintf(name, sizeof(name), "pid_%u", level + 1); | |
50 | len = sizeof(struct pid) + level * sizeof(struct upid); | |
74bd59bb | 51 | mutex_lock(&pid_caches_mutex); |
dd206bec AD |
52 | /* Name collision forces to do allocation under mutex. */ |
53 | if (!*pkc) | |
54 | *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0); | |
74bd59bb | 55 | mutex_unlock(&pid_caches_mutex); |
dd206bec AD |
56 | /* current can fail, but someone else can succeed. */ |
57 | return READ_ONCE(*pkc); | |
74bd59bb PE |
58 | } |
59 | ||
f333c700 EB |
60 | static struct ucounts *inc_pid_namespaces(struct user_namespace *ns) |
61 | { | |
62 | return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES); | |
63 | } | |
64 | ||
65 | static void dec_pid_namespaces(struct ucounts *ucounts) | |
66 | { | |
67 | dec_ucount(ucounts, UCOUNT_PID_NAMESPACES); | |
68 | } | |
69 | ||
49f4d8b9 EB |
70 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
71 | struct pid_namespace *parent_pid_ns) | |
74bd59bb PE |
72 | { |
73 | struct pid_namespace *ns; | |
ed469a63 | 74 | unsigned int level = parent_pid_ns->level + 1; |
f333c700 | 75 | struct ucounts *ucounts; |
f2302505 AV |
76 | int err; |
77 | ||
a2b42626 EB |
78 | err = -EINVAL; |
79 | if (!in_userns(parent_pid_ns->user_ns, user_ns)) | |
80 | goto out; | |
81 | ||
df75e774 | 82 | err = -ENOSPC; |
f333c700 EB |
83 | if (level > MAX_PID_NS_LEVEL) |
84 | goto out; | |
85 | ucounts = inc_pid_namespaces(user_ns); | |
86 | if (!ucounts) | |
f2302505 | 87 | goto out; |
74bd59bb | 88 | |
f2302505 | 89 | err = -ENOMEM; |
84406c15 | 90 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb | 91 | if (ns == NULL) |
f333c700 | 92 | goto out_dec; |
74bd59bb | 93 | |
95846ecf | 94 | idr_init(&ns->idr); |
74bd59bb | 95 | |
dd206bec | 96 | ns->pid_cachep = create_pid_cachep(level); |
74bd59bb | 97 | if (ns->pid_cachep == NULL) |
95846ecf | 98 | goto out_free_idr; |
74bd59bb | 99 | |
6344c433 | 100 | err = ns_alloc_inum(&ns->ns); |
98f842e6 | 101 | if (err) |
95846ecf | 102 | goto out_free_idr; |
33c42940 | 103 | ns->ns.ops = &pidns_operations; |
98f842e6 | 104 | |
74bd59bb | 105 | kref_init(&ns->kref); |
74bd59bb | 106 | ns->level = level; |
ed469a63 | 107 | ns->parent = get_pid_ns(parent_pid_ns); |
49f4d8b9 | 108 | ns->user_ns = get_user_ns(user_ns); |
f333c700 | 109 | ns->ucounts = ucounts; |
e8cfbc24 | 110 | ns->pid_allocated = PIDNS_ADDING; |
74bd59bb | 111 | |
74bd59bb PE |
112 | return ns; |
113 | ||
95846ecf GS |
114 | out_free_idr: |
115 | idr_destroy(&ns->idr); | |
74bd59bb | 116 | kmem_cache_free(pid_ns_cachep, ns); |
f333c700 EB |
117 | out_dec: |
118 | dec_pid_namespaces(ucounts); | |
74bd59bb | 119 | out: |
4308eebb | 120 | return ERR_PTR(err); |
74bd59bb PE |
121 | } |
122 | ||
1adfcb03 AV |
123 | static void delayed_free_pidns(struct rcu_head *p) |
124 | { | |
add7c65c AV |
125 | struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu); |
126 | ||
127 | dec_pid_namespaces(ns->ucounts); | |
128 | put_user_ns(ns->user_ns); | |
129 | ||
130 | kmem_cache_free(pid_ns_cachep, ns); | |
1adfcb03 AV |
131 | } |
132 | ||
74bd59bb PE |
133 | static void destroy_pid_namespace(struct pid_namespace *ns) |
134 | { | |
6344c433 | 135 | ns_free_inum(&ns->ns); |
95846ecf GS |
136 | |
137 | idr_destroy(&ns->idr); | |
1adfcb03 | 138 | call_rcu(&ns->rcu, delayed_free_pidns); |
74bd59bb PE |
139 | } |
140 | ||
49f4d8b9 EB |
141 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
142 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
74bd59bb | 143 | { |
74bd59bb | 144 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 145 | return get_pid_ns(old_ns); |
225778d6 EB |
146 | if (task_active_pid_ns(current) != old_ns) |
147 | return ERR_PTR(-EINVAL); | |
49f4d8b9 | 148 | return create_pid_namespace(user_ns, old_ns); |
74bd59bb PE |
149 | } |
150 | ||
bbc2e3ef | 151 | static void free_pid_ns(struct kref *kref) |
74bd59bb | 152 | { |
bbc2e3ef | 153 | struct pid_namespace *ns; |
74bd59bb PE |
154 | |
155 | ns = container_of(kref, struct pid_namespace, kref); | |
74bd59bb | 156 | destroy_pid_namespace(ns); |
bbc2e3ef | 157 | } |
74bd59bb | 158 | |
bbc2e3ef CG |
159 | void put_pid_ns(struct pid_namespace *ns) |
160 | { | |
161 | struct pid_namespace *parent; | |
162 | ||
163 | while (ns != &init_pid_ns) { | |
164 | parent = ns->parent; | |
165 | if (!kref_put(&ns->kref, free_pid_ns)) | |
166 | break; | |
167 | ns = parent; | |
168 | } | |
74bd59bb | 169 | } |
bbc2e3ef | 170 | EXPORT_SYMBOL_GPL(put_pid_ns); |
74bd59bb PE |
171 | |
172 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
173 | { | |
174 | int nr; | |
175 | int rc; | |
00c10bc1 | 176 | struct task_struct *task, *me = current; |
751c644b | 177 | int init_pids = thread_group_leader(me) ? 1 : 2; |
95846ecf | 178 | struct pid *pid; |
00c10bc1 | 179 | |
c876ad76 EB |
180 | /* Don't allow any more processes into the pid namespace */ |
181 | disable_pid_allocation(pid_ns); | |
182 | ||
a53b8315 ON |
183 | /* |
184 | * Ignore SIGCHLD causing any terminated children to autoreap. | |
185 | * This speeds up the namespace shutdown, plus see the comment | |
186 | * below. | |
187 | */ | |
00c10bc1 EB |
188 | spin_lock_irq(&me->sighand->siglock); |
189 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
190 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
191 | |
192 | /* | |
193 | * The last thread in the cgroup-init thread group is terminating. | |
194 | * Find remaining pid_ts in the namespace, signal and wait for them | |
195 | * to exit. | |
196 | * | |
197 | * Note: This signals each threads in the namespace - even those that | |
198 | * belong to the same thread group, To avoid this, we would have | |
199 | * to walk the entire tasklist looking a processes in this | |
200 | * namespace, but that could be unnecessarily expensive if the | |
201 | * pid namespace has just a few processes. Or we need to | |
202 | * maintain a tasklist for each pid namespace. | |
203 | * | |
204 | */ | |
95846ecf | 205 | rcu_read_lock(); |
74bd59bb | 206 | read_lock(&tasklist_lock); |
95846ecf GS |
207 | nr = 2; |
208 | idr_for_each_entry_continue(&pid_ns->idr, pid, nr) { | |
209 | task = pid_task(pid, PIDTYPE_PID); | |
a02d6fd6 | 210 | if (task && !__fatal_signal_pending(task)) |
82058d66 | 211 | group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX); |
74bd59bb PE |
212 | } |
213 | read_unlock(&tasklist_lock); | |
95846ecf | 214 | rcu_read_unlock(); |
74bd59bb | 215 | |
a53b8315 ON |
216 | /* |
217 | * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD. | |
d300b610 | 218 | * kernel_wait4() will also block until our children traced from the |
a53b8315 ON |
219 | * parent namespace are detached and become EXIT_DEAD. |
220 | */ | |
74bd59bb PE |
221 | do { |
222 | clear_thread_flag(TIF_SIGPENDING); | |
d300b610 | 223 | rc = kernel_wait4(-1, NULL, __WALL, NULL); |
74bd59bb PE |
224 | } while (rc != -ECHILD); |
225 | ||
6347e900 | 226 | /* |
af9fe6d6 EB |
227 | * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE |
228 | * process whose parents processes are outside of the pid | |
229 | * namespace. Such processes are created with setns()+fork(). | |
a53b8315 | 230 | * |
af9fe6d6 EB |
231 | * If those EXIT_ZOMBIE processes are not reaped by their |
232 | * parents before their parents exit, they will be reparented | |
233 | * to pid_ns->child_reaper. Thus pidns->child_reaper needs to | |
234 | * stay valid until they all go away. | |
a53b8315 | 235 | * |
af9fe6d6 EB |
236 | * The code relies on the the pid_ns->child_reaper ignoring |
237 | * SIGCHILD to cause those EXIT_ZOMBIE processes to be | |
238 | * autoreaped if reparented. | |
239 | * | |
240 | * Semantically it is also desirable to wait for EXIT_ZOMBIE | |
241 | * processes before allowing the child_reaper to be reaped, as | |
242 | * that gives the invariant that when the init process of a | |
243 | * pid namespace is reaped all of the processes in the pid | |
244 | * namespace are gone. | |
245 | * | |
246 | * Once all of the other tasks are gone from the pid_namespace | |
247 | * free_pid() will awaken this task. | |
6347e900 EB |
248 | */ |
249 | for (;;) { | |
b9a985db | 250 | set_current_state(TASK_INTERRUPTIBLE); |
e8cfbc24 | 251 | if (pid_ns->pid_allocated == init_pids) |
6347e900 EB |
252 | break; |
253 | schedule(); | |
254 | } | |
af4b8a83 | 255 | __set_current_state(TASK_RUNNING); |
6347e900 | 256 | |
cf3f8921 DL |
257 | if (pid_ns->reboot) |
258 | current->signal->group_exit_code = pid_ns->reboot; | |
259 | ||
0b6b030f | 260 | acct_exit_ns(pid_ns); |
74bd59bb PE |
261 | return; |
262 | } | |
263 | ||
98ed57ee | 264 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 | 265 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
32927393 | 266 | void *buffer, size_t *lenp, loff_t *ppos) |
b8f566b0 | 267 | { |
49f4d8b9 | 268 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
b8f566b0 | 269 | struct ctl_table tmp = *table; |
95846ecf | 270 | int ret, next; |
b8f566b0 | 271 | |
b9a3db92 | 272 | if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns)) |
b8f566b0 PE |
273 | return -EPERM; |
274 | ||
275 | /* | |
276 | * Writing directly to ns' last_pid field is OK, since this field | |
277 | * is volatile in a living namespace anyway and a code writing to | |
278 | * it should synchronize its usage with external means. | |
279 | */ | |
280 | ||
95846ecf GS |
281 | next = idr_get_cursor(&pid_ns->idr) - 1; |
282 | ||
283 | tmp.data = &next; | |
284 | ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); | |
285 | if (!ret && write) | |
286 | idr_set_cursor(&pid_ns->idr, next + 1); | |
287 | ||
288 | return ret; | |
b8f566b0 PE |
289 | } |
290 | ||
579035dc | 291 | extern int pid_max; |
b8f566b0 PE |
292 | static struct ctl_table pid_ns_ctl_table[] = { |
293 | { | |
294 | .procname = "ns_last_pid", | |
295 | .maxlen = sizeof(int), | |
296 | .mode = 0666, /* permissions are checked in the handler */ | |
297 | .proc_handler = pid_ns_ctl_handler, | |
eec4844f | 298 | .extra1 = SYSCTL_ZERO, |
579035dc | 299 | .extra2 = &pid_max, |
b8f566b0 PE |
300 | }, |
301 | { } | |
302 | }; | |
b8f566b0 | 303 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 304 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 305 | |
cf3f8921 DL |
306 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
307 | { | |
308 | if (pid_ns == &init_pid_ns) | |
309 | return 0; | |
310 | ||
311 | switch (cmd) { | |
312 | case LINUX_REBOOT_CMD_RESTART2: | |
313 | case LINUX_REBOOT_CMD_RESTART: | |
314 | pid_ns->reboot = SIGHUP; | |
315 | break; | |
316 | ||
317 | case LINUX_REBOOT_CMD_POWER_OFF: | |
318 | case LINUX_REBOOT_CMD_HALT: | |
319 | pid_ns->reboot = SIGINT; | |
320 | break; | |
321 | default: | |
322 | return -EINVAL; | |
323 | } | |
324 | ||
325 | read_lock(&tasklist_lock); | |
f9070dc9 | 326 | send_sig(SIGKILL, pid_ns->child_reaper, 1); |
cf3f8921 DL |
327 | read_unlock(&tasklist_lock); |
328 | ||
329 | do_exit(0); | |
330 | ||
331 | /* Not reached */ | |
332 | return 0; | |
333 | } | |
334 | ||
3c041184 AV |
335 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
336 | { | |
337 | return container_of(ns, struct pid_namespace, ns); | |
338 | } | |
339 | ||
64964528 | 340 | static struct ns_common *pidns_get(struct task_struct *task) |
57e8391d EB |
341 | { |
342 | struct pid_namespace *ns; | |
343 | ||
344 | rcu_read_lock(); | |
d2308225 ON |
345 | ns = task_active_pid_ns(task); |
346 | if (ns) | |
347 | get_pid_ns(ns); | |
57e8391d EB |
348 | rcu_read_unlock(); |
349 | ||
3c041184 | 350 | return ns ? &ns->ns : NULL; |
57e8391d EB |
351 | } |
352 | ||
eaa0d190 KT |
353 | static struct ns_common *pidns_for_children_get(struct task_struct *task) |
354 | { | |
355 | struct pid_namespace *ns = NULL; | |
356 | ||
357 | task_lock(task); | |
358 | if (task->nsproxy) { | |
359 | ns = task->nsproxy->pid_ns_for_children; | |
360 | get_pid_ns(ns); | |
361 | } | |
362 | task_unlock(task); | |
363 | ||
364 | if (ns) { | |
365 | read_lock(&tasklist_lock); | |
366 | if (!ns->child_reaper) { | |
367 | put_pid_ns(ns); | |
368 | ns = NULL; | |
369 | } | |
370 | read_unlock(&tasklist_lock); | |
371 | } | |
372 | ||
373 | return ns ? &ns->ns : NULL; | |
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 | ||
f2a8d52e | 381 | static int pidns_install(struct nsset *nsset, struct ns_common *ns) |
57e8391d | 382 | { |
f2a8d52e | 383 | struct nsproxy *nsproxy = nsset->nsproxy; |
57e8391d | 384 | struct pid_namespace *active = task_active_pid_ns(current); |
3c041184 | 385 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
57e8391d | 386 | |
5e4a0847 | 387 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e | 388 | !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN)) |
57e8391d EB |
389 | return -EPERM; |
390 | ||
391 | /* | |
392 | * Only allow entering the current active pid namespace | |
393 | * or a child of the current active pid namespace. | |
394 | * | |
395 | * This is required for fork to return a usable pid value and | |
396 | * this maintains the property that processes and their | |
397 | * children can not escape their current pid namespace. | |
398 | */ | |
399 | if (new->level < active->level) | |
400 | return -EINVAL; | |
401 | ||
402 | ancestor = new; | |
403 | while (ancestor->level > active->level) | |
404 | ancestor = ancestor->parent; | |
405 | if (ancestor != active) | |
406 | return -EINVAL; | |
407 | ||
c2b1df2e AL |
408 | put_pid_ns(nsproxy->pid_ns_for_children); |
409 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
57e8391d EB |
410 | return 0; |
411 | } | |
412 | ||
a7306ed8 AV |
413 | static struct ns_common *pidns_get_parent(struct ns_common *ns) |
414 | { | |
415 | struct pid_namespace *active = task_active_pid_ns(current); | |
416 | struct pid_namespace *pid_ns, *p; | |
417 | ||
418 | /* See if the parent is in the current namespace */ | |
419 | pid_ns = p = to_pid_ns(ns)->parent; | |
420 | for (;;) { | |
421 | if (!p) | |
422 | return ERR_PTR(-EPERM); | |
423 | if (p == active) | |
424 | break; | |
425 | p = p->parent; | |
426 | } | |
427 | ||
428 | return &get_pid_ns(pid_ns)->ns; | |
429 | } | |
430 | ||
bcac25a5 AV |
431 | static struct user_namespace *pidns_owner(struct ns_common *ns) |
432 | { | |
433 | return to_pid_ns(ns)->user_ns; | |
434 | } | |
435 | ||
57e8391d EB |
436 | const struct proc_ns_operations pidns_operations = { |
437 | .name = "pid", | |
438 | .type = CLONE_NEWPID, | |
439 | .get = pidns_get, | |
440 | .put = pidns_put, | |
441 | .install = pidns_install, | |
bcac25a5 | 442 | .owner = pidns_owner, |
a7306ed8 | 443 | .get_parent = pidns_get_parent, |
57e8391d EB |
444 | }; |
445 | ||
eaa0d190 KT |
446 | const struct proc_ns_operations pidns_for_children_operations = { |
447 | .name = "pid_for_children", | |
448 | .real_ns_name = "pid", | |
449 | .type = CLONE_NEWPID, | |
450 | .get = pidns_for_children_get, | |
451 | .put = pidns_put, | |
452 | .install = pidns_install, | |
453 | .owner = pidns_owner, | |
454 | .get_parent = pidns_get_parent, | |
455 | }; | |
456 | ||
74bd59bb PE |
457 | static __init int pid_namespaces_init(void) |
458 | { | |
459 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
460 | |
461 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 462 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 463 | #endif |
74bd59bb PE |
464 | return 0; |
465 | } | |
466 | ||
467 | __initcall(pid_namespaces_init); |