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> | |
13 | #include <linux/syscalls.h> | |
14 | #include <linux/err.h> | |
0b6b030f | 15 | #include <linux/acct.h> |
5a0e3ad6 | 16 | #include <linux/slab.h> |
4308eebb | 17 | #include <linux/proc_fs.h> |
cf3f8921 | 18 | #include <linux/reboot.h> |
523a6a94 | 19 | #include <linux/export.h> |
74bd59bb PE |
20 | |
21 | #define BITS_PER_PAGE (PAGE_SIZE*8) | |
22 | ||
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 | ||
ed469a63 | 74 | static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns) |
74bd59bb PE |
75 | { |
76 | struct pid_namespace *ns; | |
ed469a63 | 77 | unsigned int level = parent_pid_ns->level + 1; |
4308eebb | 78 | int i, err = -ENOMEM; |
74bd59bb | 79 | |
84406c15 | 80 | ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL); |
74bd59bb PE |
81 | if (ns == NULL) |
82 | goto out; | |
83 | ||
84 | ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
85 | if (!ns->pidmap[0].page) | |
86 | goto out_free; | |
87 | ||
88 | ns->pid_cachep = create_pid_cachep(level + 1); | |
89 | if (ns->pid_cachep == NULL) | |
90 | goto out_free_map; | |
91 | ||
92 | kref_init(&ns->kref); | |
74bd59bb | 93 | ns->level = level; |
ed469a63 | 94 | ns->parent = get_pid_ns(parent_pid_ns); |
74bd59bb PE |
95 | |
96 | set_bit(0, ns->pidmap[0].page); | |
97 | atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1); | |
98 | ||
84406c15 | 99 | for (i = 1; i < PIDMAP_ENTRIES; i++) |
74bd59bb | 100 | atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE); |
74bd59bb | 101 | |
4308eebb EB |
102 | err = pid_ns_prepare_proc(ns); |
103 | if (err) | |
104 | goto out_put_parent_pid_ns; | |
105 | ||
74bd59bb PE |
106 | return ns; |
107 | ||
4308eebb EB |
108 | out_put_parent_pid_ns: |
109 | put_pid_ns(parent_pid_ns); | |
74bd59bb PE |
110 | out_free_map: |
111 | kfree(ns->pidmap[0].page); | |
112 | out_free: | |
113 | kmem_cache_free(pid_ns_cachep, ns); | |
114 | out: | |
4308eebb | 115 | return ERR_PTR(err); |
74bd59bb PE |
116 | } |
117 | ||
118 | static void destroy_pid_namespace(struct pid_namespace *ns) | |
119 | { | |
120 | int i; | |
121 | ||
122 | for (i = 0; i < PIDMAP_ENTRIES; i++) | |
123 | kfree(ns->pidmap[i].page); | |
124 | kmem_cache_free(pid_ns_cachep, ns); | |
125 | } | |
126 | ||
127 | struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns) | |
128 | { | |
74bd59bb | 129 | if (!(flags & CLONE_NEWPID)) |
dca4a979 | 130 | return get_pid_ns(old_ns); |
e5a47386 | 131 | if (flags & (CLONE_THREAD|CLONE_PARENT)) |
dca4a979 AD |
132 | return ERR_PTR(-EINVAL); |
133 | return create_pid_namespace(old_ns); | |
74bd59bb PE |
134 | } |
135 | ||
136 | void free_pid_ns(struct kref *kref) | |
137 | { | |
138 | struct pid_namespace *ns, *parent; | |
139 | ||
140 | ns = container_of(kref, struct pid_namespace, kref); | |
141 | ||
142 | parent = ns->parent; | |
143 | destroy_pid_namespace(ns); | |
144 | ||
145 | if (parent != NULL) | |
146 | put_pid_ns(parent); | |
147 | } | |
523a6a94 | 148 | EXPORT_SYMBOL_GPL(free_pid_ns); |
74bd59bb PE |
149 | |
150 | void zap_pid_ns_processes(struct pid_namespace *pid_ns) | |
151 | { | |
152 | int nr; | |
153 | int rc; | |
00c10bc1 EB |
154 | struct task_struct *task, *me = current; |
155 | ||
156 | /* Ignore SIGCHLD causing any terminated children to autoreap */ | |
157 | spin_lock_irq(&me->sighand->siglock); | |
158 | me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN; | |
159 | spin_unlock_irq(&me->sighand->siglock); | |
74bd59bb PE |
160 | |
161 | /* | |
162 | * The last thread in the cgroup-init thread group is terminating. | |
163 | * Find remaining pid_ts in the namespace, signal and wait for them | |
164 | * to exit. | |
165 | * | |
166 | * Note: This signals each threads in the namespace - even those that | |
167 | * belong to the same thread group, To avoid this, we would have | |
168 | * to walk the entire tasklist looking a processes in this | |
169 | * namespace, but that could be unnecessarily expensive if the | |
170 | * pid namespace has just a few processes. Or we need to | |
171 | * maintain a tasklist for each pid namespace. | |
172 | * | |
173 | */ | |
174 | read_lock(&tasklist_lock); | |
175 | nr = next_pidmap(pid_ns, 1); | |
176 | while (nr > 0) { | |
e4da026f SB |
177 | rcu_read_lock(); |
178 | ||
e4da026f | 179 | task = pid_task(find_vpid(nr), PIDTYPE_PID); |
a02d6fd6 ON |
180 | if (task && !__fatal_signal_pending(task)) |
181 | send_sig_info(SIGKILL, SEND_SIG_FORCED, task); | |
e4da026f SB |
182 | |
183 | rcu_read_unlock(); | |
184 | ||
74bd59bb PE |
185 | nr = next_pidmap(pid_ns, nr); |
186 | } | |
187 | read_unlock(&tasklist_lock); | |
188 | ||
6347e900 | 189 | /* Firstly reap the EXIT_ZOMBIE children we may have. */ |
74bd59bb PE |
190 | do { |
191 | clear_thread_flag(TIF_SIGPENDING); | |
192 | rc = sys_wait4(-1, NULL, __WALL, NULL); | |
193 | } while (rc != -ECHILD); | |
194 | ||
6347e900 EB |
195 | /* |
196 | * sys_wait4() above can't reap the TASK_DEAD children. | |
197 | * Make sure they all go away, see __unhash_process(). | |
198 | */ | |
199 | for (;;) { | |
200 | bool need_wait = false; | |
201 | ||
202 | read_lock(&tasklist_lock); | |
203 | if (!list_empty(¤t->children)) { | |
204 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
205 | need_wait = true; | |
206 | } | |
207 | read_unlock(&tasklist_lock); | |
208 | ||
209 | if (!need_wait) | |
210 | break; | |
211 | schedule(); | |
212 | } | |
213 | ||
cf3f8921 DL |
214 | if (pid_ns->reboot) |
215 | current->signal->group_exit_code = pid_ns->reboot; | |
216 | ||
0b6b030f | 217 | acct_exit_ns(pid_ns); |
74bd59bb PE |
218 | return; |
219 | } | |
220 | ||
98ed57ee | 221 | #ifdef CONFIG_CHECKPOINT_RESTORE |
b8f566b0 PE |
222 | static int pid_ns_ctl_handler(struct ctl_table *table, int write, |
223 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
224 | { | |
225 | struct ctl_table tmp = *table; | |
226 | ||
227 | if (write && !capable(CAP_SYS_ADMIN)) | |
228 | return -EPERM; | |
229 | ||
230 | /* | |
231 | * Writing directly to ns' last_pid field is OK, since this field | |
232 | * is volatile in a living namespace anyway and a code writing to | |
233 | * it should synchronize its usage with external means. | |
234 | */ | |
235 | ||
236 | tmp.data = ¤t->nsproxy->pid_ns->last_pid; | |
579035dc | 237 | return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); |
b8f566b0 PE |
238 | } |
239 | ||
579035dc AV |
240 | extern int pid_max; |
241 | static int zero = 0; | |
b8f566b0 PE |
242 | static struct ctl_table pid_ns_ctl_table[] = { |
243 | { | |
244 | .procname = "ns_last_pid", | |
245 | .maxlen = sizeof(int), | |
246 | .mode = 0666, /* permissions are checked in the handler */ | |
247 | .proc_handler = pid_ns_ctl_handler, | |
579035dc AV |
248 | .extra1 = &zero, |
249 | .extra2 = &pid_max, | |
b8f566b0 PE |
250 | }, |
251 | { } | |
252 | }; | |
b8f566b0 | 253 | static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } }; |
98ed57ee | 254 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
b8f566b0 | 255 | |
cf3f8921 DL |
256 | int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd) |
257 | { | |
258 | if (pid_ns == &init_pid_ns) | |
259 | return 0; | |
260 | ||
261 | switch (cmd) { | |
262 | case LINUX_REBOOT_CMD_RESTART2: | |
263 | case LINUX_REBOOT_CMD_RESTART: | |
264 | pid_ns->reboot = SIGHUP; | |
265 | break; | |
266 | ||
267 | case LINUX_REBOOT_CMD_POWER_OFF: | |
268 | case LINUX_REBOOT_CMD_HALT: | |
269 | pid_ns->reboot = SIGINT; | |
270 | break; | |
271 | default: | |
272 | return -EINVAL; | |
273 | } | |
274 | ||
275 | read_lock(&tasklist_lock); | |
276 | force_sig(SIGKILL, pid_ns->child_reaper); | |
277 | read_unlock(&tasklist_lock); | |
278 | ||
279 | do_exit(0); | |
280 | ||
281 | /* Not reached */ | |
282 | return 0; | |
283 | } | |
284 | ||
74bd59bb PE |
285 | static __init int pid_namespaces_init(void) |
286 | { | |
287 | pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC); | |
98ed57ee CG |
288 | |
289 | #ifdef CONFIG_CHECKPOINT_RESTORE | |
b8f566b0 | 290 | register_sysctl_paths(kern_path, pid_ns_ctl_table); |
98ed57ee | 291 | #endif |
74bd59bb PE |
292 | return 0; |
293 | } | |
294 | ||
295 | __initcall(pid_namespaces_init); |