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 PE |
14 | #include <linux/syscalls.h> |
15 | #include <linux/err.h> | |
0b6b030f | 16 | #include <linux/acct.h> |
5a0e3ad6 | 17 | #include <linux/slab.h> |
0bb80f24 | 18 | #include <linux/proc_ns.h> |
cf3f8921 | 19 | #include <linux/reboot.h> |
523a6a94 | 20 | #include <linux/export.h> |
74bd59bb | 21 | |
74bd59bb PE |
22 | struct pid_cache { |
23 | int nr_ids; | |
24 | char name[16]; | |
25 | struct kmem_cache *cachep; | |
26 | struct list_head list; | |
27 | }; | |
28 | ||
29 | static LIST_HEAD(pid_caches_lh); | |
30 | static DEFINE_MUTEX(pid_caches_mutex); | |
31 | static struct kmem_cache *pid_ns_cachep; | |
32 | ||
33 | /* | |
34 | * creates the kmem cache to allocate pids from. | |
35 | * @nr_ids: the number of numerical ids this pid will have to carry | |
36 | */ | |
37 | ||
38 | static struct kmem_cache *create_pid_cachep(int nr_ids) | |
39 | { | |
40 | struct pid_cache *pcache; | |
41 | struct kmem_cache *cachep; | |
42 | ||
43 | mutex_lock(&pid_caches_mutex); | |
44 | list_for_each_entry(pcache, &pid_caches_lh, list) | |
45 | if (pcache->nr_ids == nr_ids) | |
46 | goto out; | |
47 | ||
48 | pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL); | |
49 | if (pcache == NULL) | |
50 | goto err_alloc; | |
51 | ||
52 | snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids); | |
53 | cachep = kmem_cache_create(pcache->name, | |
54 | sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid), | |
55 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
56 | if (cachep == NULL) | |
57 | goto err_cachep; | |
58 | ||
59 | pcache->nr_ids = nr_ids; | |
60 | pcache->cachep = cachep; | |
61 | list_add(&pcache->list, &pid_caches_lh); | |
62 | out: | |
63 | mutex_unlock(&pid_caches_mutex); | |
64 | return pcache->cachep; | |
65 | ||
66 | err_cachep: | |
67 | kfree(pcache); | |
68 | err_alloc: | |
69 | mutex_unlock(&pid_caches_mutex); | |
70 | return NULL; | |
71 | } | |
72 | ||
0a01f2cc EB |
73 | static void proc_cleanup_work(struct work_struct *work) |
74 | { | |
75 | struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work); | |
76 | pid_ns_release_proc(ns); | |
77 | } | |
78 | ||
f2302505 AV |
79 | /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */ |
80 | #define MAX_PID_NS_LEVEL 32 | |
81 | ||
49f4d8b9 EB |
82 | static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns, |
83 | struct pid_namespace *parent_pid_ns) | |
74bd59bb PE |
84 | { |
85 | struct pid_namespace *ns; | |
ed469a63 | 86 | unsigned int level = parent_pid_ns->level + 1; |
f2302505 AV |
87 | int i; |
88 | int err; | |
89 | ||
90 | if (level > MAX_PID_NS_LEVEL) { | |
91 | err = -EINVAL; | |
92 | goto out; | |
93 | } | |
74bd59bb | 94 | |
f2302505 | 95 | err = -ENOMEM; |
84406c15 | 96 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb PE |
97 | if (ns == NULL) |
98 | goto out; | |
99 | ||
100 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
101 | if (!ns->pidmap[0].page) | |
102 | goto out_free; | |
103 | ||
104 | ns->pid_cachep = create_pid_cachep(level + 1); | |
105 | if (ns->pid_cachep == NULL) | |
106 | goto out_free_map; | |
107 | ||
6344c433 | 108 | err = ns_alloc_inum(&ns->ns); |
98f842e6 EB |
109 | if (err) |
110 | goto out_free_map; | |
33c42940 | 111 | ns->ns.ops = &pidns_operations; |
98f842e6 | 112 | |
74bd59bb | 113 | kref_init(&ns->kref); |
74bd59bb | 114 | ns->level = level; |
ed469a63 | 115 | ns->parent = get_pid_ns(parent_pid_ns); |
49f4d8b9 | 116 | ns->user_ns = get_user_ns(user_ns); |
c876ad76 | 117 | ns->nr_hashed = PIDNS_HASH_ADDING; |
0a01f2cc | 118 | INIT_WORK(&ns->proc_work, proc_cleanup_work); |
74bd59bb PE |
119 | |
120 | set_bit(0, ns->pidmap[0].page); | |
121 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); | |
122 | ||
84406c15 | 123 | for (i = 1; i < PIDMAP_ENTRIES; i++) |
74bd59bb | 124 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); |
74bd59bb PE |
125 | |
126 | return ns; | |
127 | ||
128 | out_free_map: | |
129 | kfree(ns->pidmap[0].page); | |
130 | out_free: | |
131 | kmem_cache_free(pid_ns_cachep, ns); | |
132 | out: | |
4308eebb | 133 | return ERR_PTR(err); |
74bd59bb PE |
134 | } |
135 | ||
1adfcb03 AV |
136 | static void delayed_free_pidns(struct rcu_head *p) |
137 | { | |
138 | kmem_cache_free(pid_ns_cachep, | |
139 | container_of(p, struct pid_namespace, rcu)); | |
140 | } | |
141 | ||
74bd59bb PE |
142 | static void destroy_pid_namespace(struct pid_namespace *ns) |
143 | { | |
144 | int i; | |
145 | ||
6344c433 | 146 | ns_free_inum(&ns->ns); |
74bd59bb PE |
147 | for (i = 0; i < PIDMAP_ENTRIES; i++) |
148 | kfree(ns->pidmap[i].page); | |
49f4d8b9 | 149 | put_user_ns(ns->user_ns); |
1adfcb03 | 150 | call_rcu(&ns->rcu, delayed_free_pidns); |
74bd59bb PE |
151 | } |
152 | ||
49f4d8b9 EB |
153 | struct pid_namespace *copy_pid_ns(unsigned long flags, |
154 | struct user_namespace *user_ns, struct pid_namespace *old_ns) | |
74bd59bb | 155 | { |
74bd59bb | 156 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 157 | return get_pid_ns(old_ns); |
225778d6 EB |
158 | if (task_active_pid_ns(current) != old_ns) |
159 | return ERR_PTR(-EINVAL); | |
49f4d8b9 | 160 | return create_pid_namespace(user_ns, old_ns); |
74bd59bb PE |
161 | } |
162 | ||
bbc2e3ef | 163 | static void free_pid_ns(struct kref *kref) |
74bd59bb | 164 | { |
bbc2e3ef | 165 | struct pid_namespace *ns; |
74bd59bb PE |
166 | |
167 | ns = container_of(kref, struct pid_namespace, kref); | |
74bd59bb | 168 | destroy_pid_namespace(ns); |
bbc2e3ef | 169 | } |
74bd59bb | 170 | |
bbc2e3ef CG |
171 | void put_pid_ns(struct pid_namespace *ns) |
172 | { | |
173 | struct pid_namespace *parent; | |
174 | ||
175 | while (ns != &init_pid_ns) { | |
176 | parent = ns->parent; | |
177 | if (!kref_put(&ns->kref, free_pid_ns)) | |
178 | break; | |
179 | ns = parent; | |
180 | } | |
74bd59bb | 181 | } |
bbc2e3ef | 182 | EXPORT_SYMBOL_GPL(put_pid_ns); |
74bd59bb PE |
183 | |
184 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
185 | { | |
186 | int nr; | |
187 | int rc; | |
00c10bc1 | 188 | struct task_struct *task, *me = current; |
751c644b | 189 | int init_pids = thread_group_leader(me) ? 1 : 2; |
00c10bc1 | 190 | |
c876ad76 EB |
191 | /* Don't allow any more processes into the pid namespace */ |
192 | disable_pid_allocation(pid_ns); | |
193 | ||
00c10bc1 EB |
194 | /* Ignore SIGCHLD causing any terminated children to autoreap */ |
195 | spin_lock_irq(&me->sighand->siglock); | |
196 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
197 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
198 | |
199 | /* | |
200 | * The last thread in the cgroup-init thread group is terminating. | |
201 | * Find remaining pid_ts in the namespace, signal and wait for them | |
202 | * to exit. | |
203 | * | |
204 | * Note: This signals each threads in the namespace - even those that | |
205 | * belong to the same thread group, To avoid this, we would have | |
206 | * to walk the entire tasklist looking a processes in this | |
207 | * namespace, but that could be unnecessarily expensive if the | |
208 | * pid namespace has just a few processes. Or we need to | |
209 | * maintain a tasklist for each pid namespace. | |
210 | * | |
211 | */ | |
212 | read_lock(&tasklist_lock); | |
213 | nr = next_pidmap(pid_ns, 1); | |
214 | while (nr > 0) { | |
e4da026f SB |
215 | rcu_read_lock(); |
216 | ||
e4da026f | 217 | task = pid_task(find_vpid(nr), PIDTYPE_PID); |
a02d6fd6 ON |
218 | if (task && !__fatal_signal_pending(task)) |
219 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); | |
e4da026f SB |
220 | |
221 | rcu_read_unlock(); | |
222 | ||
74bd59bb PE |
223 | nr = next_pidmap(pid_ns, nr); |
224 | } | |
225 | read_unlock(&tasklist_lock); | |
226 | ||
6347e900 | 227 | /* Firstly reap the EXIT_ZOMBIE children we may have. */ |
74bd59bb PE |
228 | do { |
229 | clear_thread_flag(TIF_SIGPENDING); | |
230 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
231 | } while (rc != -ECHILD); | |
232 | ||
6347e900 EB |
233 | /* |
234 | * sys_wait4() above can't reap the TASK_DEAD children. | |
af4b8a83 | 235 | * Make sure they all go away, see free_pid(). |
6347e900 EB |
236 | */ |
237 | for (;;) { | |
af4b8a83 | 238 | set_current_state(TASK_UNINTERRUPTIBLE); |
751c644b | 239 | if (pid_ns->nr_hashed == init_pids) |
6347e900 EB |
240 | break; |
241 | schedule(); | |
242 | } | |
af4b8a83 | 243 | __set_current_state(TASK_RUNNING); |
6347e900 | 244 | |
cf3f8921 DL |
245 | if (pid_ns->reboot) |
246 | current->signal->group_exit_code = pid_ns->reboot; | |
247 | ||
0b6b030f | 248 | acct_exit_ns(pid_ns); |
74bd59bb PE |
249 | return; |
250 | } | |
251 | ||
98ed57ee | 252 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 PE |
253 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
254 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
255 | { | |
49f4d8b9 | 256 | struct pid_namespace *pid_ns = task_active_pid_ns(current); |
b8f566b0 PE |
257 | struct ctl_table tmp = *table; |
258 | ||
49f4d8b9 | 259 | if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN)) |
b8f566b0 PE |
260 | return -EPERM; |
261 | ||
262 | /* | |
263 | * Writing directly to ns' last_pid field is OK, since this field | |
264 | * is volatile in a living namespace anyway and a code writing to | |
265 | * it should synchronize its usage with external means. | |
266 | */ | |
267 | ||
49f4d8b9 | 268 | tmp.data = &pid_ns->last_pid; |
579035dc | 269 | return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); |
b8f566b0 PE |
270 | } |
271 | ||
579035dc AV |
272 | extern int pid_max; |
273 | static int zero = 0; | |
b8f566b0 PE |
274 | static struct ctl_table pid_ns_ctl_table[] = { |
275 | { | |
276 | .procname = "ns_last_pid", | |
277 | .maxlen = sizeof(int), | |
278 | .mode = 0666, /* permissions are checked in the handler */ | |
279 | .proc_handler = pid_ns_ctl_handler, | |
579035dc AV |
280 | .extra1 = &zero, |
281 | .extra2 = &pid_max, | |
b8f566b0 PE |
282 | }, |
283 | { } | |
284 | }; | |
b8f566b0 | 285 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 286 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 287 | |
cf3f8921 DL |
288 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
289 | { | |
290 | if (pid_ns == &init_pid_ns) | |
291 | return 0; | |
292 | ||
293 | switch (cmd) { | |
294 | case LINUX_REBOOT_CMD_RESTART2: | |
295 | case LINUX_REBOOT_CMD_RESTART: | |
296 | pid_ns->reboot = SIGHUP; | |
297 | break; | |
298 | ||
299 | case LINUX_REBOOT_CMD_POWER_OFF: | |
300 | case LINUX_REBOOT_CMD_HALT: | |
301 | pid_ns->reboot = SIGINT; | |
302 | break; | |
303 | default: | |
304 | return -EINVAL; | |
305 | } | |
306 | ||
307 | read_lock(&tasklist_lock); | |
308 | force_sig(SIGKILL, pid_ns->child_reaper); | |
309 | read_unlock(&tasklist_lock); | |
310 | ||
311 | do_exit(0); | |
312 | ||
313 | /* Not reached */ | |
314 | return 0; | |
315 | } | |
316 | ||
3c041184 AV |
317 | static inline struct pid_namespace *to_pid_ns(struct ns_common *ns) |
318 | { | |
319 | return container_of(ns, struct pid_namespace, ns); | |
320 | } | |
321 | ||
64964528 | 322 | static struct ns_common *pidns_get(struct task_struct *task) |
57e8391d EB |
323 | { |
324 | struct pid_namespace *ns; | |
325 | ||
326 | rcu_read_lock(); | |
d2308225 ON |
327 | ns = task_active_pid_ns(task); |
328 | if (ns) | |
329 | get_pid_ns(ns); | |
57e8391d EB |
330 | rcu_read_unlock(); |
331 | ||
3c041184 | 332 | return ns ? &ns->ns : NULL; |
57e8391d EB |
333 | } |
334 | ||
64964528 | 335 | static void pidns_put(struct ns_common *ns) |
57e8391d | 336 | { |
3c041184 | 337 | put_pid_ns(to_pid_ns(ns)); |
57e8391d EB |
338 | } |
339 | ||
64964528 | 340 | static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
57e8391d EB |
341 | { |
342 | struct pid_namespace *active = task_active_pid_ns(current); | |
3c041184 | 343 | struct pid_namespace *ancestor, *new = to_pid_ns(ns); |
57e8391d | 344 | |
5e4a0847 | 345 | if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) || |
c7b96acf | 346 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
57e8391d EB |
347 | return -EPERM; |
348 | ||
349 | /* | |
350 | * Only allow entering the current active pid namespace | |
351 | * or a child of the current active pid namespace. | |
352 | * | |
353 | * This is required for fork to return a usable pid value and | |
354 | * this maintains the property that processes and their | |
355 | * children can not escape their current pid namespace. | |
356 | */ | |
357 | if (new->level < active->level) | |
358 | return -EINVAL; | |
359 | ||
360 | ancestor = new; | |
361 | while (ancestor->level > active->level) | |
362 | ancestor = ancestor->parent; | |
363 | if (ancestor != active) | |
364 | return -EINVAL; | |
365 | ||
c2b1df2e AL |
366 | put_pid_ns(nsproxy->pid_ns_for_children); |
367 | nsproxy->pid_ns_for_children = get_pid_ns(new); | |
57e8391d EB |
368 | return 0; |
369 | } | |
370 | ||
371 | const struct proc_ns_operations pidns_operations = { | |
372 | .name = "pid", | |
373 | .type = CLONE_NEWPID, | |
374 | .get = pidns_get, | |
375 | .put = pidns_put, | |
376 | .install = pidns_install, | |
377 | }; | |
378 | ||
74bd59bb PE |
379 | static __init int pid_namespaces_init(void) |
380 | { | |
381 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
382 | |
383 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 384 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 385 | #endif |
74bd59bb PE |
386 | return 0; |
387 | } | |
388 | ||
389 | __initcall(pid_namespaces_init); |