1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * Pid namespaces 3 * Pid namespaces
4 * 4 *
5 * Authors: 5 * Authors:
6 * (C) 2007 Pavel Emelyanov <xemul@openvz.o 6 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
7 * (C) 2007 Sukadev Bhattiprolu <sukadev@us 7 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
8 * Many thanks to Oleg Nesterov for commen 8 * Many thanks to Oleg Nesterov for comments and help
9 * 9 *
10 */ 10 */
11 11
12 #include <linux/pid.h> 12 #include <linux/pid.h>
13 #include <linux/pid_namespace.h> 13 #include <linux/pid_namespace.h>
14 #include <linux/user_namespace.h> 14 #include <linux/user_namespace.h>
15 #include <linux/syscalls.h> 15 #include <linux/syscalls.h>
16 #include <linux/cred.h> 16 #include <linux/cred.h>
17 #include <linux/err.h> 17 #include <linux/err.h>
18 #include <linux/acct.h> 18 #include <linux/acct.h>
19 #include <linux/slab.h> 19 #include <linux/slab.h>
20 #include <linux/proc_ns.h> 20 #include <linux/proc_ns.h>
21 #include <linux/reboot.h> 21 #include <linux/reboot.h>
22 #include <linux/export.h> 22 #include <linux/export.h>
23 #include <linux/sched/task.h> 23 #include <linux/sched/task.h>
24 #include <linux/sched/signal.h> 24 #include <linux/sched/signal.h>
25 #include <linux/idr.h> 25 #include <linux/idr.h>
26 #include <uapi/linux/wait.h> <<
27 #include "pid_sysctl.h" <<
28 26
29 static DEFINE_MUTEX(pid_caches_mutex); 27 static DEFINE_MUTEX(pid_caches_mutex);
30 static struct kmem_cache *pid_ns_cachep; 28 static struct kmem_cache *pid_ns_cachep;
31 /* Write once array, filled from the beginning 29 /* Write once array, filled from the beginning. */
32 static struct kmem_cache *pid_cache[MAX_PID_NS 30 static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
33 31
34 /* 32 /*
35 * creates the kmem cache to allocate pids fro 33 * creates the kmem cache to allocate pids from.
36 * @level: pid namespace level 34 * @level: pid namespace level
37 */ 35 */
38 36
39 static struct kmem_cache *create_pid_cachep(un 37 static struct kmem_cache *create_pid_cachep(unsigned int level)
40 { 38 {
41 /* Level 0 is init_pid_ns.pid_cachep * 39 /* Level 0 is init_pid_ns.pid_cachep */
42 struct kmem_cache **pkc = &pid_cache[l 40 struct kmem_cache **pkc = &pid_cache[level - 1];
43 struct kmem_cache *kc; 41 struct kmem_cache *kc;
44 char name[4 + 10 + 1]; 42 char name[4 + 10 + 1];
45 unsigned int len; 43 unsigned int len;
46 44
47 kc = READ_ONCE(*pkc); 45 kc = READ_ONCE(*pkc);
48 if (kc) 46 if (kc)
49 return kc; 47 return kc;
50 48
51 snprintf(name, sizeof(name), "pid_%u", 49 snprintf(name, sizeof(name), "pid_%u", level + 1);
52 len = struct_size_t(struct pid, number !! 50 len = sizeof(struct pid) + level * sizeof(struct upid);
53 mutex_lock(&pid_caches_mutex); 51 mutex_lock(&pid_caches_mutex);
54 /* Name collision forces to do allocat 52 /* Name collision forces to do allocation under mutex. */
55 if (!*pkc) 53 if (!*pkc)
56 *pkc = kmem_cache_create(name, !! 54 *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
57 SLAB_ <<
58 mutex_unlock(&pid_caches_mutex); 55 mutex_unlock(&pid_caches_mutex);
59 /* current can fail, but someone else 56 /* current can fail, but someone else can succeed. */
60 return READ_ONCE(*pkc); 57 return READ_ONCE(*pkc);
61 } 58 }
62 59
63 static struct ucounts *inc_pid_namespaces(stru 60 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
64 { 61 {
65 return inc_ucount(ns, current_euid(), 62 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
66 } 63 }
67 64
68 static void dec_pid_namespaces(struct ucounts 65 static void dec_pid_namespaces(struct ucounts *ucounts)
69 { 66 {
70 dec_ucount(ucounts, UCOUNT_PID_NAMESPA 67 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
71 } 68 }
72 69
73 static struct pid_namespace *create_pid_namesp 70 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
74 struct pid_namespace *parent_pid_ns) 71 struct pid_namespace *parent_pid_ns)
75 { 72 {
76 struct pid_namespace *ns; 73 struct pid_namespace *ns;
77 unsigned int level = parent_pid_ns->le 74 unsigned int level = parent_pid_ns->level + 1;
78 struct ucounts *ucounts; 75 struct ucounts *ucounts;
79 int err; 76 int err;
80 77
81 err = -EINVAL; 78 err = -EINVAL;
82 if (!in_userns(parent_pid_ns->user_ns, 79 if (!in_userns(parent_pid_ns->user_ns, user_ns))
83 goto out; 80 goto out;
84 81
85 err = -ENOSPC; 82 err = -ENOSPC;
86 if (level > MAX_PID_NS_LEVEL) 83 if (level > MAX_PID_NS_LEVEL)
87 goto out; 84 goto out;
88 ucounts = inc_pid_namespaces(user_ns); 85 ucounts = inc_pid_namespaces(user_ns);
89 if (!ucounts) 86 if (!ucounts)
90 goto out; 87 goto out;
91 88
92 err = -ENOMEM; 89 err = -ENOMEM;
93 ns = kmem_cache_zalloc(pid_ns_cachep, 90 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
94 if (ns == NULL) 91 if (ns == NULL)
95 goto out_dec; 92 goto out_dec;
96 93
97 idr_init(&ns->idr); 94 idr_init(&ns->idr);
98 95
99 ns->pid_cachep = create_pid_cachep(lev 96 ns->pid_cachep = create_pid_cachep(level);
100 if (ns->pid_cachep == NULL) 97 if (ns->pid_cachep == NULL)
101 goto out_free_idr; 98 goto out_free_idr;
102 99
103 err = ns_alloc_inum(&ns->ns); 100 err = ns_alloc_inum(&ns->ns);
104 if (err) 101 if (err)
105 goto out_free_idr; 102 goto out_free_idr;
106 ns->ns.ops = &pidns_operations; 103 ns->ns.ops = &pidns_operations;
107 104
108 refcount_set(&ns->ns.count, 1); !! 105 kref_init(&ns->kref);
109 ns->level = level; 106 ns->level = level;
110 ns->parent = get_pid_ns(parent_pid_ns) 107 ns->parent = get_pid_ns(parent_pid_ns);
111 ns->user_ns = get_user_ns(user_ns); 108 ns->user_ns = get_user_ns(user_ns);
112 ns->ucounts = ucounts; 109 ns->ucounts = ucounts;
113 ns->pid_allocated = PIDNS_ADDING; 110 ns->pid_allocated = PIDNS_ADDING;
114 #if defined(CONFIG_SYSCTL) && defined(CONFIG_M !! 111
115 ns->memfd_noexec_scope = pidns_memfd_n <<
116 #endif <<
117 return ns; 112 return ns;
118 113
119 out_free_idr: 114 out_free_idr:
120 idr_destroy(&ns->idr); 115 idr_destroy(&ns->idr);
121 kmem_cache_free(pid_ns_cachep, ns); 116 kmem_cache_free(pid_ns_cachep, ns);
122 out_dec: 117 out_dec:
123 dec_pid_namespaces(ucounts); 118 dec_pid_namespaces(ucounts);
124 out: 119 out:
125 return ERR_PTR(err); 120 return ERR_PTR(err);
126 } 121 }
127 122
128 static void delayed_free_pidns(struct rcu_head 123 static void delayed_free_pidns(struct rcu_head *p)
129 { 124 {
130 struct pid_namespace *ns = container_o 125 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
131 126
132 dec_pid_namespaces(ns->ucounts); 127 dec_pid_namespaces(ns->ucounts);
133 put_user_ns(ns->user_ns); 128 put_user_ns(ns->user_ns);
134 129
135 kmem_cache_free(pid_ns_cachep, ns); 130 kmem_cache_free(pid_ns_cachep, ns);
136 } 131 }
137 132
138 static void destroy_pid_namespace(struct pid_n 133 static void destroy_pid_namespace(struct pid_namespace *ns)
139 { 134 {
140 ns_free_inum(&ns->ns); 135 ns_free_inum(&ns->ns);
141 136
142 idr_destroy(&ns->idr); 137 idr_destroy(&ns->idr);
143 call_rcu(&ns->rcu, delayed_free_pidns) 138 call_rcu(&ns->rcu, delayed_free_pidns);
144 } 139 }
145 140
146 struct pid_namespace *copy_pid_ns(unsigned lon 141 struct pid_namespace *copy_pid_ns(unsigned long flags,
147 struct user_namespace *user_ns, struct 142 struct user_namespace *user_ns, struct pid_namespace *old_ns)
148 { 143 {
149 if (!(flags & CLONE_NEWPID)) 144 if (!(flags & CLONE_NEWPID))
150 return get_pid_ns(old_ns); 145 return get_pid_ns(old_ns);
151 if (task_active_pid_ns(current) != old 146 if (task_active_pid_ns(current) != old_ns)
152 return ERR_PTR(-EINVAL); 147 return ERR_PTR(-EINVAL);
153 return create_pid_namespace(user_ns, o 148 return create_pid_namespace(user_ns, old_ns);
154 } 149 }
155 150
>> 151 static void free_pid_ns(struct kref *kref)
>> 152 {
>> 153 struct pid_namespace *ns;
>> 154
>> 155 ns = container_of(kref, struct pid_namespace, kref);
>> 156 destroy_pid_namespace(ns);
>> 157 }
>> 158
156 void put_pid_ns(struct pid_namespace *ns) 159 void put_pid_ns(struct pid_namespace *ns)
157 { 160 {
158 struct pid_namespace *parent; 161 struct pid_namespace *parent;
159 162
160 while (ns != &init_pid_ns) { 163 while (ns != &init_pid_ns) {
161 parent = ns->parent; 164 parent = ns->parent;
162 if (!refcount_dec_and_test(&ns !! 165 if (!kref_put(&ns->kref, free_pid_ns))
163 break; 166 break;
164 destroy_pid_namespace(ns); <<
165 ns = parent; 167 ns = parent;
166 } 168 }
167 } 169 }
168 EXPORT_SYMBOL_GPL(put_pid_ns); 170 EXPORT_SYMBOL_GPL(put_pid_ns);
169 171
170 void zap_pid_ns_processes(struct pid_namespace 172 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
171 { 173 {
172 int nr; 174 int nr;
173 int rc; 175 int rc;
174 struct task_struct *task, *me = curren 176 struct task_struct *task, *me = current;
175 int init_pids = thread_group_leader(me 177 int init_pids = thread_group_leader(me) ? 1 : 2;
176 struct pid *pid; 178 struct pid *pid;
177 179
178 /* Don't allow any more processes into 180 /* Don't allow any more processes into the pid namespace */
179 disable_pid_allocation(pid_ns); 181 disable_pid_allocation(pid_ns);
180 182
181 /* 183 /*
182 * Ignore SIGCHLD causing any terminat 184 * Ignore SIGCHLD causing any terminated children to autoreap.
183 * This speeds up the namespace shutdo 185 * This speeds up the namespace shutdown, plus see the comment
184 * below. 186 * below.
185 */ 187 */
186 spin_lock_irq(&me->sighand->siglock); 188 spin_lock_irq(&me->sighand->siglock);
187 me->sighand->action[SIGCHLD - 1].sa.sa 189 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
188 spin_unlock_irq(&me->sighand->siglock) 190 spin_unlock_irq(&me->sighand->siglock);
189 191
190 /* 192 /*
191 * The last thread in the cgroup-init 193 * The last thread in the cgroup-init thread group is terminating.
192 * Find remaining pid_ts in the namesp 194 * Find remaining pid_ts in the namespace, signal and wait for them
193 * to exit. 195 * to exit.
194 * 196 *
195 * Note: This signals each threads in 197 * Note: This signals each threads in the namespace - even those that
196 * belong to the same thread gr 198 * belong to the same thread group, To avoid this, we would have
197 * to walk the entire tasklist 199 * to walk the entire tasklist looking a processes in this
198 * namespace, but that could be 200 * namespace, but that could be unnecessarily expensive if the
199 * pid namespace has just a few 201 * pid namespace has just a few processes. Or we need to
200 * maintain a tasklist for each 202 * maintain a tasklist for each pid namespace.
201 * 203 *
202 */ 204 */
203 rcu_read_lock(); 205 rcu_read_lock();
204 read_lock(&tasklist_lock); 206 read_lock(&tasklist_lock);
205 nr = 2; 207 nr = 2;
206 idr_for_each_entry_continue(&pid_ns->i 208 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
207 task = pid_task(pid, PIDTYPE_P 209 task = pid_task(pid, PIDTYPE_PID);
208 if (task && !__fatal_signal_pe 210 if (task && !__fatal_signal_pending(task))
209 group_send_sig_info(SI 211 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
210 } 212 }
211 read_unlock(&tasklist_lock); 213 read_unlock(&tasklist_lock);
212 rcu_read_unlock(); 214 rcu_read_unlock();
213 215
214 /* 216 /*
215 * Reap the EXIT_ZOMBIE children we ha 217 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
216 * kernel_wait4() will also block unti 218 * kernel_wait4() will also block until our children traced from the
217 * parent namespace are detached and b 219 * parent namespace are detached and become EXIT_DEAD.
218 */ 220 */
219 do { 221 do {
220 clear_thread_flag(TIF_SIGPENDI 222 clear_thread_flag(TIF_SIGPENDING);
221 clear_thread_flag(TIF_NOTIFY_S <<
222 rc = kernel_wait4(-1, NULL, __ 223 rc = kernel_wait4(-1, NULL, __WALL, NULL);
223 } while (rc != -ECHILD); 224 } while (rc != -ECHILD);
224 225
225 /* 226 /*
226 * kernel_wait4() misses EXIT_DEAD chi 227 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
227 * process whose parents processes are 228 * process whose parents processes are outside of the pid
228 * namespace. Such processes are crea 229 * namespace. Such processes are created with setns()+fork().
229 * 230 *
230 * If those EXIT_ZOMBIE processes are 231 * If those EXIT_ZOMBIE processes are not reaped by their
231 * parents before their parents exit, 232 * parents before their parents exit, they will be reparented
232 * to pid_ns->child_reaper. Thus pidn 233 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
233 * stay valid until they all go away. 234 * stay valid until they all go away.
234 * 235 *
235 * The code relies on the pid_ns->chil !! 236 * The code relies on the the pid_ns->child_reaper ignoring
236 * SIGCHILD to cause those EXIT_ZOMBIE 237 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
237 * autoreaped if reparented. 238 * autoreaped if reparented.
238 * 239 *
239 * Semantically it is also desirable t 240 * Semantically it is also desirable to wait for EXIT_ZOMBIE
240 * processes before allowing the child 241 * processes before allowing the child_reaper to be reaped, as
241 * that gives the invariant that when 242 * that gives the invariant that when the init process of a
242 * pid namespace is reaped all of the 243 * pid namespace is reaped all of the processes in the pid
243 * namespace are gone. 244 * namespace are gone.
244 * 245 *
245 * Once all of the other tasks are gon 246 * Once all of the other tasks are gone from the pid_namespace
246 * free_pid() will awaken this task. 247 * free_pid() will awaken this task.
247 */ 248 */
248 for (;;) { 249 for (;;) {
249 set_current_state(TASK_INTERRU 250 set_current_state(TASK_INTERRUPTIBLE);
250 if (pid_ns->pid_allocated == i 251 if (pid_ns->pid_allocated == init_pids)
251 break; 252 break;
252 schedule(); 253 schedule();
253 } 254 }
254 __set_current_state(TASK_RUNNING); 255 __set_current_state(TASK_RUNNING);
255 256
256 if (pid_ns->reboot) 257 if (pid_ns->reboot)
257 current->signal->group_exit_co 258 current->signal->group_exit_code = pid_ns->reboot;
258 259
259 acct_exit_ns(pid_ns); 260 acct_exit_ns(pid_ns);
260 return; 261 return;
261 } 262 }
262 263
263 #ifdef CONFIG_CHECKPOINT_RESTORE 264 #ifdef CONFIG_CHECKPOINT_RESTORE
264 static int pid_ns_ctl_handler(const struct ctl !! 265 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
265 void *buffer, size_t *lenp, lo !! 266 void __user *buffer, size_t *lenp, loff_t *ppos)
266 { 267 {
267 struct pid_namespace *pid_ns = task_ac 268 struct pid_namespace *pid_ns = task_active_pid_ns(current);
268 struct ctl_table tmp = *table; 269 struct ctl_table tmp = *table;
269 int ret, next; 270 int ret, next;
270 271
271 if (write && !checkpoint_restore_ns_ca !! 272 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
272 return -EPERM; 273 return -EPERM;
273 274
>> 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
274 next = idr_get_cursor(&pid_ns->idr) - 281 next = idr_get_cursor(&pid_ns->idr) - 1;
275 282
276 tmp.data = &next; 283 tmp.data = &next;
277 ret = proc_dointvec_minmax(&tmp, write 284 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
278 if (!ret && write) 285 if (!ret && write)
279 idr_set_cursor(&pid_ns->idr, n 286 idr_set_cursor(&pid_ns->idr, next + 1);
280 287
281 return ret; 288 return ret;
282 } 289 }
283 290
284 extern int pid_max; 291 extern int pid_max;
285 static struct ctl_table pid_ns_ctl_table[] = { 292 static struct ctl_table pid_ns_ctl_table[] = {
286 { 293 {
287 .procname = "ns_last_pid", 294 .procname = "ns_last_pid",
288 .maxlen = sizeof(int), 295 .maxlen = sizeof(int),
289 .mode = 0666, /* permissions a 296 .mode = 0666, /* permissions are checked in the handler */
290 .proc_handler = pid_ns_ctl_han 297 .proc_handler = pid_ns_ctl_handler,
291 .extra1 = SYSCTL_ZERO, 298 .extra1 = SYSCTL_ZERO,
292 .extra2 = &pid_max, 299 .extra2 = &pid_max,
293 }, 300 },
>> 301 { }
294 }; 302 };
>> 303 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
295 #endif /* CONFIG_CHECKPOINT_RESTORE */ 304 #endif /* CONFIG_CHECKPOINT_RESTORE */
296 305
297 int reboot_pid_ns(struct pid_namespace *pid_ns 306 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
298 { 307 {
299 if (pid_ns == &init_pid_ns) 308 if (pid_ns == &init_pid_ns)
300 return 0; 309 return 0;
301 310
302 switch (cmd) { 311 switch (cmd) {
303 case LINUX_REBOOT_CMD_RESTART2: 312 case LINUX_REBOOT_CMD_RESTART2:
304 case LINUX_REBOOT_CMD_RESTART: 313 case LINUX_REBOOT_CMD_RESTART:
305 pid_ns->reboot = SIGHUP; 314 pid_ns->reboot = SIGHUP;
306 break; 315 break;
307 316
308 case LINUX_REBOOT_CMD_POWER_OFF: 317 case LINUX_REBOOT_CMD_POWER_OFF:
309 case LINUX_REBOOT_CMD_HALT: 318 case LINUX_REBOOT_CMD_HALT:
310 pid_ns->reboot = SIGINT; 319 pid_ns->reboot = SIGINT;
311 break; 320 break;
312 default: 321 default:
313 return -EINVAL; 322 return -EINVAL;
314 } 323 }
315 324
316 read_lock(&tasklist_lock); 325 read_lock(&tasklist_lock);
317 send_sig(SIGKILL, pid_ns->child_reaper 326 send_sig(SIGKILL, pid_ns->child_reaper, 1);
318 read_unlock(&tasklist_lock); 327 read_unlock(&tasklist_lock);
319 328
320 do_exit(0); 329 do_exit(0);
321 330
322 /* Not reached */ 331 /* Not reached */
323 return 0; 332 return 0;
324 } 333 }
325 334
326 static inline struct pid_namespace *to_pid_ns( 335 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
327 { 336 {
328 return container_of(ns, struct pid_nam 337 return container_of(ns, struct pid_namespace, ns);
329 } 338 }
330 339
331 static struct ns_common *pidns_get(struct task 340 static struct ns_common *pidns_get(struct task_struct *task)
332 { 341 {
333 struct pid_namespace *ns; 342 struct pid_namespace *ns;
334 343
335 rcu_read_lock(); 344 rcu_read_lock();
336 ns = task_active_pid_ns(task); 345 ns = task_active_pid_ns(task);
337 if (ns) 346 if (ns)
338 get_pid_ns(ns); 347 get_pid_ns(ns);
339 rcu_read_unlock(); 348 rcu_read_unlock();
340 349
341 return ns ? &ns->ns : NULL; 350 return ns ? &ns->ns : NULL;
342 } 351 }
343 352
344 static struct ns_common *pidns_for_children_ge 353 static struct ns_common *pidns_for_children_get(struct task_struct *task)
345 { 354 {
346 struct pid_namespace *ns = NULL; 355 struct pid_namespace *ns = NULL;
347 356
348 task_lock(task); 357 task_lock(task);
349 if (task->nsproxy) { 358 if (task->nsproxy) {
350 ns = task->nsproxy->pid_ns_for 359 ns = task->nsproxy->pid_ns_for_children;
351 get_pid_ns(ns); 360 get_pid_ns(ns);
352 } 361 }
353 task_unlock(task); 362 task_unlock(task);
354 363
355 if (ns) { 364 if (ns) {
356 read_lock(&tasklist_lock); 365 read_lock(&tasklist_lock);
357 if (!ns->child_reaper) { 366 if (!ns->child_reaper) {
358 put_pid_ns(ns); 367 put_pid_ns(ns);
359 ns = NULL; 368 ns = NULL;
360 } 369 }
361 read_unlock(&tasklist_lock); 370 read_unlock(&tasklist_lock);
362 } 371 }
363 372
364 return ns ? &ns->ns : NULL; 373 return ns ? &ns->ns : NULL;
365 } 374 }
366 375
367 static void pidns_put(struct ns_common *ns) 376 static void pidns_put(struct ns_common *ns)
368 { 377 {
369 put_pid_ns(to_pid_ns(ns)); 378 put_pid_ns(to_pid_ns(ns));
370 } 379 }
371 380
372 static int pidns_install(struct nsset *nsset, !! 381 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
373 { 382 {
374 struct nsproxy *nsproxy = nsset->nspro <<
375 struct pid_namespace *active = task_ac 383 struct pid_namespace *active = task_active_pid_ns(current);
376 struct pid_namespace *ancestor, *new = 384 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
377 385
378 if (!ns_capable(new->user_ns, CAP_SYS_ 386 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
379 !ns_capable(nsset->cred->user_ns, !! 387 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
380 return -EPERM; 388 return -EPERM;
381 389
382 /* 390 /*
383 * Only allow entering the current act 391 * Only allow entering the current active pid namespace
384 * or a child of the current active pi 392 * or a child of the current active pid namespace.
385 * 393 *
386 * This is required for fork to return 394 * This is required for fork to return a usable pid value and
387 * this maintains the property that pr 395 * this maintains the property that processes and their
388 * children can not escape their curre 396 * children can not escape their current pid namespace.
389 */ 397 */
390 if (new->level < active->level) 398 if (new->level < active->level)
391 return -EINVAL; 399 return -EINVAL;
392 400
393 ancestor = new; 401 ancestor = new;
394 while (ancestor->level > active->level 402 while (ancestor->level > active->level)
395 ancestor = ancestor->parent; 403 ancestor = ancestor->parent;
396 if (ancestor != active) 404 if (ancestor != active)
397 return -EINVAL; 405 return -EINVAL;
398 406
399 put_pid_ns(nsproxy->pid_ns_for_childre 407 put_pid_ns(nsproxy->pid_ns_for_children);
400 nsproxy->pid_ns_for_children = get_pid 408 nsproxy->pid_ns_for_children = get_pid_ns(new);
401 return 0; 409 return 0;
402 } 410 }
403 411
404 static struct ns_common *pidns_get_parent(stru 412 static struct ns_common *pidns_get_parent(struct ns_common *ns)
405 { 413 {
406 struct pid_namespace *active = task_ac 414 struct pid_namespace *active = task_active_pid_ns(current);
407 struct pid_namespace *pid_ns, *p; 415 struct pid_namespace *pid_ns, *p;
408 416
409 /* See if the parent is in the current 417 /* See if the parent is in the current namespace */
410 pid_ns = p = to_pid_ns(ns)->parent; 418 pid_ns = p = to_pid_ns(ns)->parent;
411 for (;;) { 419 for (;;) {
412 if (!p) 420 if (!p)
413 return ERR_PTR(-EPERM) 421 return ERR_PTR(-EPERM);
414 if (p == active) 422 if (p == active)
415 break; 423 break;
416 p = p->parent; 424 p = p->parent;
417 } 425 }
418 426
419 return &get_pid_ns(pid_ns)->ns; 427 return &get_pid_ns(pid_ns)->ns;
420 } 428 }
421 429
422 static struct user_namespace *pidns_owner(stru 430 static struct user_namespace *pidns_owner(struct ns_common *ns)
423 { 431 {
424 return to_pid_ns(ns)->user_ns; 432 return to_pid_ns(ns)->user_ns;
425 } 433 }
426 434
427 const struct proc_ns_operations pidns_operatio 435 const struct proc_ns_operations pidns_operations = {
428 .name = "pid", 436 .name = "pid",
429 .type = CLONE_NEWPID, 437 .type = CLONE_NEWPID,
430 .get = pidns_get, 438 .get = pidns_get,
431 .put = pidns_put, 439 .put = pidns_put,
432 .install = pidns_install, 440 .install = pidns_install,
433 .owner = pidns_owner, 441 .owner = pidns_owner,
434 .get_parent = pidns_get_parent, 442 .get_parent = pidns_get_parent,
435 }; 443 };
436 444
437 const struct proc_ns_operations pidns_for_chil 445 const struct proc_ns_operations pidns_for_children_operations = {
438 .name = "pid_for_children", 446 .name = "pid_for_children",
439 .real_ns_name = "pid", 447 .real_ns_name = "pid",
440 .type = CLONE_NEWPID, 448 .type = CLONE_NEWPID,
441 .get = pidns_for_children_g 449 .get = pidns_for_children_get,
442 .put = pidns_put, 450 .put = pidns_put,
443 .install = pidns_install, 451 .install = pidns_install,
444 .owner = pidns_owner, 452 .owner = pidns_owner,
445 .get_parent = pidns_get_parent, 453 .get_parent = pidns_get_parent,
446 }; 454 };
447 455
448 static __init int pid_namespaces_init(void) 456 static __init int pid_namespaces_init(void)
449 { 457 {
450 pid_ns_cachep = KMEM_CACHE(pid_namespa !! 458 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
451 459
452 #ifdef CONFIG_CHECKPOINT_RESTORE 460 #ifdef CONFIG_CHECKPOINT_RESTORE
453 register_sysctl_init("kernel", pid_ns_ !! 461 register_sysctl_paths(kern_path, pid_ns_ctl_table);
454 #endif 462 #endif
455 <<
456 register_pid_ns_sysctl_table_vm(); <<
457 return 0; 463 return 0;
458 } 464 }
459 465
460 __initcall(pid_namespaces_init); 466 __initcall(pid_namespaces_init);
461 467
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.