Loading...
1/*
2 * Generic pidhash and scalable, time-bounded PID allocator
3 *
4 * (C) 2002-2003 Nadia Yvette Chambers, IBM
5 * (C) 2004 Nadia Yvette Chambers, Oracle
6 * (C) 2002-2004 Ingo Molnar, Red Hat
7 *
8 * pid-structures are backing objects for tasks sharing a given ID to chain
9 * against. There is very little to them aside from hashing them and
10 * parking tasks using given ID's on a list.
11 *
12 * The hash is always changed with the tasklist_lock write-acquired,
13 * and the hash is only accessed with the tasklist_lock at least
14 * read-acquired, so there's no additional SMP locking needed here.
15 *
16 * We have a list of bitmap pages, which bitmaps represent the PID space.
17 * Allocating and freeing PIDs is completely lockless. The worst-case
18 * allocation scenario when all but one out of 1 million PIDs possible are
19 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
20 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
21 *
22 * Pid namespaces:
23 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
24 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
25 * Many thanks to Oleg Nesterov for comments and help
26 *
27 */
28
29#include <linux/mm.h>
30#include <linux/export.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/rculist.h>
34#include <linux/bootmem.h>
35#include <linux/hash.h>
36#include <linux/pid_namespace.h>
37#include <linux/init_task.h>
38#include <linux/syscalls.h>
39#include <linux/proc_ns.h>
40#include <linux/proc_fs.h>
41#include <linux/sched/task.h>
42#include <linux/idr.h>
43
44struct pid init_struct_pid = {
45 .count = ATOMIC_INIT(1),
46 .tasks = {
47 { .first = NULL },
48 { .first = NULL },
49 { .first = NULL },
50 },
51 .level = 0,
52 .numbers = { {
53 .nr = 0,
54 .ns = &init_pid_ns,
55 }, }
56};
57
58int pid_max = PID_MAX_DEFAULT;
59
60#define RESERVED_PIDS 300
61
62int pid_max_min = RESERVED_PIDS + 1;
63int pid_max_max = PID_MAX_LIMIT;
64
65/*
66 * PID-map pages start out as NULL, they get allocated upon
67 * first use and are never deallocated. This way a low pid_max
68 * value does not cause lots of bitmaps to be allocated, but
69 * the scheme scales to up to 4 million PIDs, runtime.
70 */
71struct pid_namespace init_pid_ns = {
72 .kref = KREF_INIT(2),
73 .idr = IDR_INIT(init_pid_ns.idr),
74 .pid_allocated = PIDNS_ADDING,
75 .level = 0,
76 .child_reaper = &init_task,
77 .user_ns = &init_user_ns,
78 .ns.inum = PROC_PID_INIT_INO,
79#ifdef CONFIG_PID_NS
80 .ns.ops = &pidns_operations,
81#endif
82};
83EXPORT_SYMBOL_GPL(init_pid_ns);
84
85/*
86 * Note: disable interrupts while the pidmap_lock is held as an
87 * interrupt might come in and do read_lock(&tasklist_lock).
88 *
89 * If we don't disable interrupts there is a nasty deadlock between
90 * detach_pid()->free_pid() and another cpu that does
91 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
92 * read_lock(&tasklist_lock);
93 *
94 * After we clean up the tasklist_lock and know there are no
95 * irq handlers that take it we can leave the interrupts enabled.
96 * For now it is easier to be safe than to prove it can't happen.
97 */
98
99static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
100
101void put_pid(struct pid *pid)
102{
103 struct pid_namespace *ns;
104
105 if (!pid)
106 return;
107
108 ns = pid->numbers[pid->level].ns;
109 if ((atomic_read(&pid->count) == 1) ||
110 atomic_dec_and_test(&pid->count)) {
111 kmem_cache_free(ns->pid_cachep, pid);
112 put_pid_ns(ns);
113 }
114}
115EXPORT_SYMBOL_GPL(put_pid);
116
117static void delayed_put_pid(struct rcu_head *rhp)
118{
119 struct pid *pid = container_of(rhp, struct pid, rcu);
120 put_pid(pid);
121}
122
123void free_pid(struct pid *pid)
124{
125 /* We can be called with write_lock_irq(&tasklist_lock) held */
126 int i;
127 unsigned long flags;
128
129 spin_lock_irqsave(&pidmap_lock, flags);
130 for (i = 0; i <= pid->level; i++) {
131 struct upid *upid = pid->numbers + i;
132 struct pid_namespace *ns = upid->ns;
133 switch (--ns->pid_allocated) {
134 case 2:
135 case 1:
136 /* When all that is left in the pid namespace
137 * is the reaper wake up the reaper. The reaper
138 * may be sleeping in zap_pid_ns_processes().
139 */
140 wake_up_process(ns->child_reaper);
141 break;
142 case PIDNS_ADDING:
143 /* Handle a fork failure of the first process */
144 WARN_ON(ns->child_reaper);
145 ns->pid_allocated = 0;
146 /* fall through */
147 case 0:
148 schedule_work(&ns->proc_work);
149 break;
150 }
151
152 idr_remove(&ns->idr, upid->nr);
153 }
154 spin_unlock_irqrestore(&pidmap_lock, flags);
155
156 call_rcu(&pid->rcu, delayed_put_pid);
157}
158
159struct pid *alloc_pid(struct pid_namespace *ns)
160{
161 struct pid *pid;
162 enum pid_type type;
163 int i, nr;
164 struct pid_namespace *tmp;
165 struct upid *upid;
166 int retval = -ENOMEM;
167
168 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
169 if (!pid)
170 return ERR_PTR(retval);
171
172 tmp = ns;
173 pid->level = ns->level;
174
175 for (i = ns->level; i >= 0; i--) {
176 int pid_min = 1;
177
178 idr_preload(GFP_KERNEL);
179 spin_lock_irq(&pidmap_lock);
180
181 /*
182 * init really needs pid 1, but after reaching the maximum
183 * wrap back to RESERVED_PIDS
184 */
185 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
186 pid_min = RESERVED_PIDS;
187
188 /*
189 * Store a null pointer so find_pid_ns does not find
190 * a partially initialized PID (see below).
191 */
192 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
193 pid_max, GFP_ATOMIC);
194 spin_unlock_irq(&pidmap_lock);
195 idr_preload_end();
196
197 if (nr < 0) {
198 retval = nr;
199 goto out_free;
200 }
201
202 pid->numbers[i].nr = nr;
203 pid->numbers[i].ns = tmp;
204 tmp = tmp->parent;
205 }
206
207 if (unlikely(is_child_reaper(pid))) {
208 if (pid_ns_prepare_proc(ns))
209 goto out_free;
210 }
211
212 get_pid_ns(ns);
213 atomic_set(&pid->count, 1);
214 for (type = 0; type < PIDTYPE_MAX; ++type)
215 INIT_HLIST_HEAD(&pid->tasks[type]);
216
217 upid = pid->numbers + ns->level;
218 spin_lock_irq(&pidmap_lock);
219 if (!(ns->pid_allocated & PIDNS_ADDING))
220 goto out_unlock;
221 for ( ; upid >= pid->numbers; --upid) {
222 /* Make the PID visible to find_pid_ns. */
223 idr_replace(&upid->ns->idr, pid, upid->nr);
224 upid->ns->pid_allocated++;
225 }
226 spin_unlock_irq(&pidmap_lock);
227
228 return pid;
229
230out_unlock:
231 spin_unlock_irq(&pidmap_lock);
232 put_pid_ns(ns);
233
234out_free:
235 spin_lock_irq(&pidmap_lock);
236 while (++i <= ns->level)
237 idr_remove(&ns->idr, (pid->numbers + i)->nr);
238
239 /* On failure to allocate the first pid, reset the state */
240 if (ns->pid_allocated == PIDNS_ADDING)
241 idr_set_cursor(&ns->idr, 0);
242
243 spin_unlock_irq(&pidmap_lock);
244
245 kmem_cache_free(ns->pid_cachep, pid);
246 return ERR_PTR(retval);
247}
248
249void disable_pid_allocation(struct pid_namespace *ns)
250{
251 spin_lock_irq(&pidmap_lock);
252 ns->pid_allocated &= ~PIDNS_ADDING;
253 spin_unlock_irq(&pidmap_lock);
254}
255
256struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
257{
258 return idr_find(&ns->idr, nr);
259}
260EXPORT_SYMBOL_GPL(find_pid_ns);
261
262struct pid *find_vpid(int nr)
263{
264 return find_pid_ns(nr, task_active_pid_ns(current));
265}
266EXPORT_SYMBOL_GPL(find_vpid);
267
268/*
269 * attach_pid() must be called with the tasklist_lock write-held.
270 */
271void attach_pid(struct task_struct *task, enum pid_type type)
272{
273 struct pid_link *link = &task->pids[type];
274 hlist_add_head_rcu(&link->node, &link->pid->tasks[type]);
275}
276
277static void __change_pid(struct task_struct *task, enum pid_type type,
278 struct pid *new)
279{
280 struct pid_link *link;
281 struct pid *pid;
282 int tmp;
283
284 link = &task->pids[type];
285 pid = link->pid;
286
287 hlist_del_rcu(&link->node);
288 link->pid = new;
289
290 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
291 if (!hlist_empty(&pid->tasks[tmp]))
292 return;
293
294 free_pid(pid);
295}
296
297void detach_pid(struct task_struct *task, enum pid_type type)
298{
299 __change_pid(task, type, NULL);
300}
301
302void change_pid(struct task_struct *task, enum pid_type type,
303 struct pid *pid)
304{
305 __change_pid(task, type, pid);
306 attach_pid(task, type);
307}
308
309/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
310void transfer_pid(struct task_struct *old, struct task_struct *new,
311 enum pid_type type)
312{
313 new->pids[type].pid = old->pids[type].pid;
314 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
315}
316
317struct task_struct *pid_task(struct pid *pid, enum pid_type type)
318{
319 struct task_struct *result = NULL;
320 if (pid) {
321 struct hlist_node *first;
322 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
323 lockdep_tasklist_lock_is_held());
324 if (first)
325 result = hlist_entry(first, struct task_struct, pids[(type)].node);
326 }
327 return result;
328}
329EXPORT_SYMBOL(pid_task);
330
331/*
332 * Must be called under rcu_read_lock().
333 */
334struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
335{
336 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
337 "find_task_by_pid_ns() needs rcu_read_lock() protection");
338 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
339}
340
341struct task_struct *find_task_by_vpid(pid_t vnr)
342{
343 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
344}
345
346struct task_struct *find_get_task_by_vpid(pid_t nr)
347{
348 struct task_struct *task;
349
350 rcu_read_lock();
351 task = find_task_by_vpid(nr);
352 if (task)
353 get_task_struct(task);
354 rcu_read_unlock();
355
356 return task;
357}
358
359struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
360{
361 struct pid *pid;
362 rcu_read_lock();
363 if (type != PIDTYPE_PID)
364 task = task->group_leader;
365 pid = get_pid(rcu_dereference(task->pids[type].pid));
366 rcu_read_unlock();
367 return pid;
368}
369EXPORT_SYMBOL_GPL(get_task_pid);
370
371struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
372{
373 struct task_struct *result;
374 rcu_read_lock();
375 result = pid_task(pid, type);
376 if (result)
377 get_task_struct(result);
378 rcu_read_unlock();
379 return result;
380}
381EXPORT_SYMBOL_GPL(get_pid_task);
382
383struct pid *find_get_pid(pid_t nr)
384{
385 struct pid *pid;
386
387 rcu_read_lock();
388 pid = get_pid(find_vpid(nr));
389 rcu_read_unlock();
390
391 return pid;
392}
393EXPORT_SYMBOL_GPL(find_get_pid);
394
395pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
396{
397 struct upid *upid;
398 pid_t nr = 0;
399
400 if (pid && ns->level <= pid->level) {
401 upid = &pid->numbers[ns->level];
402 if (upid->ns == ns)
403 nr = upid->nr;
404 }
405 return nr;
406}
407EXPORT_SYMBOL_GPL(pid_nr_ns);
408
409pid_t pid_vnr(struct pid *pid)
410{
411 return pid_nr_ns(pid, task_active_pid_ns(current));
412}
413EXPORT_SYMBOL_GPL(pid_vnr);
414
415pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
416 struct pid_namespace *ns)
417{
418 pid_t nr = 0;
419
420 rcu_read_lock();
421 if (!ns)
422 ns = task_active_pid_ns(current);
423 if (likely(pid_alive(task))) {
424 if (type != PIDTYPE_PID) {
425 if (type == __PIDTYPE_TGID)
426 type = PIDTYPE_PID;
427
428 task = task->group_leader;
429 }
430 nr = pid_nr_ns(rcu_dereference(task->pids[type].pid), ns);
431 }
432 rcu_read_unlock();
433
434 return nr;
435}
436EXPORT_SYMBOL(__task_pid_nr_ns);
437
438struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
439{
440 return ns_of_pid(task_pid(tsk));
441}
442EXPORT_SYMBOL_GPL(task_active_pid_ns);
443
444/*
445 * Used by proc to find the first pid that is greater than or equal to nr.
446 *
447 * If there is a pid at nr this function is exactly the same as find_pid_ns.
448 */
449struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
450{
451 return idr_get_next(&ns->idr, &nr);
452}
453
454void __init pid_idr_init(void)
455{
456 /* Verify no one has done anything silly: */
457 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
458
459 /* bump default and minimum pid_max based on number of cpus */
460 pid_max = min(pid_max_max, max_t(int, pid_max,
461 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
462 pid_max_min = max_t(int, pid_max_min,
463 PIDS_PER_CPU_MIN * num_possible_cpus());
464 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
465
466 idr_init(&init_pid_ns.idr);
467
468 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
469 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
470}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Generic pidhash and scalable, time-bounded PID allocator
4 *
5 * (C) 2002-2003 Nadia Yvette Chambers, IBM
6 * (C) 2004 Nadia Yvette Chambers, Oracle
7 * (C) 2002-2004 Ingo Molnar, Red Hat
8 *
9 * pid-structures are backing objects for tasks sharing a given ID to chain
10 * against. There is very little to them aside from hashing them and
11 * parking tasks using given ID's on a list.
12 *
13 * The hash is always changed with the tasklist_lock write-acquired,
14 * and the hash is only accessed with the tasklist_lock at least
15 * read-acquired, so there's no additional SMP locking needed here.
16 *
17 * We have a list of bitmap pages, which bitmaps represent the PID space.
18 * Allocating and freeing PIDs is completely lockless. The worst-case
19 * allocation scenario when all but one out of 1 million PIDs possible are
20 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
21 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
22 *
23 * Pid namespaces:
24 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
25 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
26 * Many thanks to Oleg Nesterov for comments and help
27 *
28 */
29
30#include <linux/mm.h>
31#include <linux/export.h>
32#include <linux/slab.h>
33#include <linux/init.h>
34#include <linux/rculist.h>
35#include <linux/memblock.h>
36#include <linux/pid_namespace.h>
37#include <linux/init_task.h>
38#include <linux/syscalls.h>
39#include <linux/proc_ns.h>
40#include <linux/refcount.h>
41#include <linux/anon_inodes.h>
42#include <linux/sched/signal.h>
43#include <linux/sched/task.h>
44#include <linux/idr.h>
45#include <net/sock.h>
46
47struct pid init_struct_pid = {
48 .count = REFCOUNT_INIT(1),
49 .tasks = {
50 { .first = NULL },
51 { .first = NULL },
52 { .first = NULL },
53 },
54 .level = 0,
55 .numbers = { {
56 .nr = 0,
57 .ns = &init_pid_ns,
58 }, }
59};
60
61int pid_max = PID_MAX_DEFAULT;
62
63#define RESERVED_PIDS 300
64
65int pid_max_min = RESERVED_PIDS + 1;
66int pid_max_max = PID_MAX_LIMIT;
67
68/*
69 * PID-map pages start out as NULL, they get allocated upon
70 * first use and are never deallocated. This way a low pid_max
71 * value does not cause lots of bitmaps to be allocated, but
72 * the scheme scales to up to 4 million PIDs, runtime.
73 */
74struct pid_namespace init_pid_ns = {
75 .kref = KREF_INIT(2),
76 .idr = IDR_INIT(init_pid_ns.idr),
77 .pid_allocated = PIDNS_ADDING,
78 .level = 0,
79 .child_reaper = &init_task,
80 .user_ns = &init_user_ns,
81 .ns.inum = PROC_PID_INIT_INO,
82#ifdef CONFIG_PID_NS
83 .ns.ops = &pidns_operations,
84#endif
85};
86EXPORT_SYMBOL_GPL(init_pid_ns);
87
88/*
89 * Note: disable interrupts while the pidmap_lock is held as an
90 * interrupt might come in and do read_lock(&tasklist_lock).
91 *
92 * If we don't disable interrupts there is a nasty deadlock between
93 * detach_pid()->free_pid() and another cpu that does
94 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
95 * read_lock(&tasklist_lock);
96 *
97 * After we clean up the tasklist_lock and know there are no
98 * irq handlers that take it we can leave the interrupts enabled.
99 * For now it is easier to be safe than to prove it can't happen.
100 */
101
102static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
103
104void put_pid(struct pid *pid)
105{
106 struct pid_namespace *ns;
107
108 if (!pid)
109 return;
110
111 ns = pid->numbers[pid->level].ns;
112 if (refcount_dec_and_test(&pid->count)) {
113 kmem_cache_free(ns->pid_cachep, pid);
114 put_pid_ns(ns);
115 }
116}
117EXPORT_SYMBOL_GPL(put_pid);
118
119static void delayed_put_pid(struct rcu_head *rhp)
120{
121 struct pid *pid = container_of(rhp, struct pid, rcu);
122 put_pid(pid);
123}
124
125void free_pid(struct pid *pid)
126{
127 /* We can be called with write_lock_irq(&tasklist_lock) held */
128 int i;
129 unsigned long flags;
130
131 spin_lock_irqsave(&pidmap_lock, flags);
132 for (i = 0; i <= pid->level; i++) {
133 struct upid *upid = pid->numbers + i;
134 struct pid_namespace *ns = upid->ns;
135 switch (--ns->pid_allocated) {
136 case 2:
137 case 1:
138 /* When all that is left in the pid namespace
139 * is the reaper wake up the reaper. The reaper
140 * may be sleeping in zap_pid_ns_processes().
141 */
142 wake_up_process(ns->child_reaper);
143 break;
144 case PIDNS_ADDING:
145 /* Handle a fork failure of the first process */
146 WARN_ON(ns->child_reaper);
147 ns->pid_allocated = 0;
148 break;
149 }
150
151 idr_remove(&ns->idr, upid->nr);
152 }
153 spin_unlock_irqrestore(&pidmap_lock, flags);
154
155 call_rcu(&pid->rcu, delayed_put_pid);
156}
157
158struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
159 size_t set_tid_size)
160{
161 struct pid *pid;
162 enum pid_type type;
163 int i, nr;
164 struct pid_namespace *tmp;
165 struct upid *upid;
166 int retval = -ENOMEM;
167
168 /*
169 * set_tid_size contains the size of the set_tid array. Starting at
170 * the most nested currently active PID namespace it tells alloc_pid()
171 * which PID to set for a process in that most nested PID namespace
172 * up to set_tid_size PID namespaces. It does not have to set the PID
173 * for a process in all nested PID namespaces but set_tid_size must
174 * never be greater than the current ns->level + 1.
175 */
176 if (set_tid_size > ns->level + 1)
177 return ERR_PTR(-EINVAL);
178
179 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
180 if (!pid)
181 return ERR_PTR(retval);
182
183 tmp = ns;
184 pid->level = ns->level;
185
186 for (i = ns->level; i >= 0; i--) {
187 int tid = 0;
188
189 if (set_tid_size) {
190 tid = set_tid[ns->level - i];
191
192 retval = -EINVAL;
193 if (tid < 1 || tid >= pid_max)
194 goto out_free;
195 /*
196 * Also fail if a PID != 1 is requested and
197 * no PID 1 exists.
198 */
199 if (tid != 1 && !tmp->child_reaper)
200 goto out_free;
201 retval = -EPERM;
202 if (!checkpoint_restore_ns_capable(tmp->user_ns))
203 goto out_free;
204 set_tid_size--;
205 }
206
207 idr_preload(GFP_KERNEL);
208 spin_lock_irq(&pidmap_lock);
209
210 if (tid) {
211 nr = idr_alloc(&tmp->idr, NULL, tid,
212 tid + 1, GFP_ATOMIC);
213 /*
214 * If ENOSPC is returned it means that the PID is
215 * alreay in use. Return EEXIST in that case.
216 */
217 if (nr == -ENOSPC)
218 nr = -EEXIST;
219 } else {
220 int pid_min = 1;
221 /*
222 * init really needs pid 1, but after reaching the
223 * maximum wrap back to RESERVED_PIDS
224 */
225 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
226 pid_min = RESERVED_PIDS;
227
228 /*
229 * Store a null pointer so find_pid_ns does not find
230 * a partially initialized PID (see below).
231 */
232 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
233 pid_max, GFP_ATOMIC);
234 }
235 spin_unlock_irq(&pidmap_lock);
236 idr_preload_end();
237
238 if (nr < 0) {
239 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
240 goto out_free;
241 }
242
243 pid->numbers[i].nr = nr;
244 pid->numbers[i].ns = tmp;
245 tmp = tmp->parent;
246 }
247
248 /*
249 * ENOMEM is not the most obvious choice especially for the case
250 * where the child subreaper has already exited and the pid
251 * namespace denies the creation of any new processes. But ENOMEM
252 * is what we have exposed to userspace for a long time and it is
253 * documented behavior for pid namespaces. So we can't easily
254 * change it even if there were an error code better suited.
255 */
256 retval = -ENOMEM;
257
258 get_pid_ns(ns);
259 refcount_set(&pid->count, 1);
260 spin_lock_init(&pid->lock);
261 for (type = 0; type < PIDTYPE_MAX; ++type)
262 INIT_HLIST_HEAD(&pid->tasks[type]);
263
264 init_waitqueue_head(&pid->wait_pidfd);
265 INIT_HLIST_HEAD(&pid->inodes);
266
267 upid = pid->numbers + ns->level;
268 spin_lock_irq(&pidmap_lock);
269 if (!(ns->pid_allocated & PIDNS_ADDING))
270 goto out_unlock;
271 for ( ; upid >= pid->numbers; --upid) {
272 /* Make the PID visible to find_pid_ns. */
273 idr_replace(&upid->ns->idr, pid, upid->nr);
274 upid->ns->pid_allocated++;
275 }
276 spin_unlock_irq(&pidmap_lock);
277
278 return pid;
279
280out_unlock:
281 spin_unlock_irq(&pidmap_lock);
282 put_pid_ns(ns);
283
284out_free:
285 spin_lock_irq(&pidmap_lock);
286 while (++i <= ns->level) {
287 upid = pid->numbers + i;
288 idr_remove(&upid->ns->idr, upid->nr);
289 }
290
291 /* On failure to allocate the first pid, reset the state */
292 if (ns->pid_allocated == PIDNS_ADDING)
293 idr_set_cursor(&ns->idr, 0);
294
295 spin_unlock_irq(&pidmap_lock);
296
297 kmem_cache_free(ns->pid_cachep, pid);
298 return ERR_PTR(retval);
299}
300
301void disable_pid_allocation(struct pid_namespace *ns)
302{
303 spin_lock_irq(&pidmap_lock);
304 ns->pid_allocated &= ~PIDNS_ADDING;
305 spin_unlock_irq(&pidmap_lock);
306}
307
308struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
309{
310 return idr_find(&ns->idr, nr);
311}
312EXPORT_SYMBOL_GPL(find_pid_ns);
313
314struct pid *find_vpid(int nr)
315{
316 return find_pid_ns(nr, task_active_pid_ns(current));
317}
318EXPORT_SYMBOL_GPL(find_vpid);
319
320static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
321{
322 return (type == PIDTYPE_PID) ?
323 &task->thread_pid :
324 &task->signal->pids[type];
325}
326
327/*
328 * attach_pid() must be called with the tasklist_lock write-held.
329 */
330void attach_pid(struct task_struct *task, enum pid_type type)
331{
332 struct pid *pid = *task_pid_ptr(task, type);
333 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
334}
335
336static void __change_pid(struct task_struct *task, enum pid_type type,
337 struct pid *new)
338{
339 struct pid **pid_ptr = task_pid_ptr(task, type);
340 struct pid *pid;
341 int tmp;
342
343 pid = *pid_ptr;
344
345 hlist_del_rcu(&task->pid_links[type]);
346 *pid_ptr = new;
347
348 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
349 if (pid_has_task(pid, tmp))
350 return;
351
352 free_pid(pid);
353}
354
355void detach_pid(struct task_struct *task, enum pid_type type)
356{
357 __change_pid(task, type, NULL);
358}
359
360void change_pid(struct task_struct *task, enum pid_type type,
361 struct pid *pid)
362{
363 __change_pid(task, type, pid);
364 attach_pid(task, type);
365}
366
367void exchange_tids(struct task_struct *left, struct task_struct *right)
368{
369 struct pid *pid1 = left->thread_pid;
370 struct pid *pid2 = right->thread_pid;
371 struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
372 struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
373
374 /* Swap the single entry tid lists */
375 hlists_swap_heads_rcu(head1, head2);
376
377 /* Swap the per task_struct pid */
378 rcu_assign_pointer(left->thread_pid, pid2);
379 rcu_assign_pointer(right->thread_pid, pid1);
380
381 /* Swap the cached value */
382 WRITE_ONCE(left->pid, pid_nr(pid2));
383 WRITE_ONCE(right->pid, pid_nr(pid1));
384}
385
386/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
387void transfer_pid(struct task_struct *old, struct task_struct *new,
388 enum pid_type type)
389{
390 if (type == PIDTYPE_PID)
391 new->thread_pid = old->thread_pid;
392 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
393}
394
395struct task_struct *pid_task(struct pid *pid, enum pid_type type)
396{
397 struct task_struct *result = NULL;
398 if (pid) {
399 struct hlist_node *first;
400 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
401 lockdep_tasklist_lock_is_held());
402 if (first)
403 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
404 }
405 return result;
406}
407EXPORT_SYMBOL(pid_task);
408
409/*
410 * Must be called under rcu_read_lock().
411 */
412struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
413{
414 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
415 "find_task_by_pid_ns() needs rcu_read_lock() protection");
416 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
417}
418
419struct task_struct *find_task_by_vpid(pid_t vnr)
420{
421 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
422}
423
424struct task_struct *find_get_task_by_vpid(pid_t nr)
425{
426 struct task_struct *task;
427
428 rcu_read_lock();
429 task = find_task_by_vpid(nr);
430 if (task)
431 get_task_struct(task);
432 rcu_read_unlock();
433
434 return task;
435}
436
437struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
438{
439 struct pid *pid;
440 rcu_read_lock();
441 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
442 rcu_read_unlock();
443 return pid;
444}
445EXPORT_SYMBOL_GPL(get_task_pid);
446
447struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
448{
449 struct task_struct *result;
450 rcu_read_lock();
451 result = pid_task(pid, type);
452 if (result)
453 get_task_struct(result);
454 rcu_read_unlock();
455 return result;
456}
457EXPORT_SYMBOL_GPL(get_pid_task);
458
459struct pid *find_get_pid(pid_t nr)
460{
461 struct pid *pid;
462
463 rcu_read_lock();
464 pid = get_pid(find_vpid(nr));
465 rcu_read_unlock();
466
467 return pid;
468}
469EXPORT_SYMBOL_GPL(find_get_pid);
470
471pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
472{
473 struct upid *upid;
474 pid_t nr = 0;
475
476 if (pid && ns->level <= pid->level) {
477 upid = &pid->numbers[ns->level];
478 if (upid->ns == ns)
479 nr = upid->nr;
480 }
481 return nr;
482}
483EXPORT_SYMBOL_GPL(pid_nr_ns);
484
485pid_t pid_vnr(struct pid *pid)
486{
487 return pid_nr_ns(pid, task_active_pid_ns(current));
488}
489EXPORT_SYMBOL_GPL(pid_vnr);
490
491pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
492 struct pid_namespace *ns)
493{
494 pid_t nr = 0;
495
496 rcu_read_lock();
497 if (!ns)
498 ns = task_active_pid_ns(current);
499 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
500 rcu_read_unlock();
501
502 return nr;
503}
504EXPORT_SYMBOL(__task_pid_nr_ns);
505
506struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
507{
508 return ns_of_pid(task_pid(tsk));
509}
510EXPORT_SYMBOL_GPL(task_active_pid_ns);
511
512/*
513 * Used by proc to find the first pid that is greater than or equal to nr.
514 *
515 * If there is a pid at nr this function is exactly the same as find_pid_ns.
516 */
517struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
518{
519 return idr_get_next(&ns->idr, &nr);
520}
521
522/**
523 * pidfd_create() - Create a new pid file descriptor.
524 *
525 * @pid: struct pid that the pidfd will reference
526 *
527 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
528 *
529 * Note, that this function can only be called after the fd table has
530 * been unshared to avoid leaking the pidfd to the new process.
531 *
532 * Return: On success, a cloexec pidfd is returned.
533 * On error, a negative errno number will be returned.
534 */
535static int pidfd_create(struct pid *pid)
536{
537 int fd;
538
539 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
540 O_RDWR | O_CLOEXEC);
541 if (fd < 0)
542 put_pid(pid);
543
544 return fd;
545}
546
547/**
548 * pidfd_open() - Open new pid file descriptor.
549 *
550 * @pid: pid for which to retrieve a pidfd
551 * @flags: flags to pass
552 *
553 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
554 * the process identified by @pid. Currently, the process identified by
555 * @pid must be a thread-group leader. This restriction currently exists
556 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
557 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
558 * leaders).
559 *
560 * Return: On success, a cloexec pidfd is returned.
561 * On error, a negative errno number will be returned.
562 */
563SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
564{
565 int fd;
566 struct pid *p;
567
568 if (flags)
569 return -EINVAL;
570
571 if (pid <= 0)
572 return -EINVAL;
573
574 p = find_get_pid(pid);
575 if (!p)
576 return -ESRCH;
577
578 if (pid_has_task(p, PIDTYPE_TGID))
579 fd = pidfd_create(p);
580 else
581 fd = -EINVAL;
582
583 put_pid(p);
584 return fd;
585}
586
587void __init pid_idr_init(void)
588{
589 /* Verify no one has done anything silly: */
590 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
591
592 /* bump default and minimum pid_max based on number of cpus */
593 pid_max = min(pid_max_max, max_t(int, pid_max,
594 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
595 pid_max_min = max_t(int, pid_max_min,
596 PIDS_PER_CPU_MIN * num_possible_cpus());
597 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
598
599 idr_init(&init_pid_ns.idr);
600
601 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
602 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
603}
604
605static struct file *__pidfd_fget(struct task_struct *task, int fd)
606{
607 struct file *file;
608 int ret;
609
610 ret = mutex_lock_killable(&task->signal->exec_update_mutex);
611 if (ret)
612 return ERR_PTR(ret);
613
614 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
615 file = fget_task(task, fd);
616 else
617 file = ERR_PTR(-EPERM);
618
619 mutex_unlock(&task->signal->exec_update_mutex);
620
621 return file ?: ERR_PTR(-EBADF);
622}
623
624static int pidfd_getfd(struct pid *pid, int fd)
625{
626 struct task_struct *task;
627 struct file *file;
628 int ret;
629
630 task = get_pid_task(pid, PIDTYPE_PID);
631 if (!task)
632 return -ESRCH;
633
634 file = __pidfd_fget(task, fd);
635 put_task_struct(task);
636 if (IS_ERR(file))
637 return PTR_ERR(file);
638
639 ret = receive_fd(file, O_CLOEXEC);
640 fput(file);
641
642 return ret;
643}
644
645/**
646 * sys_pidfd_getfd() - Get a file descriptor from another process
647 *
648 * @pidfd: the pidfd file descriptor of the process
649 * @fd: the file descriptor number to get
650 * @flags: flags on how to get the fd (reserved)
651 *
652 * This syscall gets a copy of a file descriptor from another process
653 * based on the pidfd, and file descriptor number. It requires that
654 * the calling process has the ability to ptrace the process represented
655 * by the pidfd. The process which is having its file descriptor copied
656 * is otherwise unaffected.
657 *
658 * Return: On success, a cloexec file descriptor is returned.
659 * On error, a negative errno number will be returned.
660 */
661SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
662 unsigned int, flags)
663{
664 struct pid *pid;
665 struct fd f;
666 int ret;
667
668 /* flags is currently unused - make sure it's unset */
669 if (flags)
670 return -EINVAL;
671
672 f = fdget(pidfd);
673 if (!f.file)
674 return -EBADF;
675
676 pid = pidfd_pid(f.file);
677 if (IS_ERR(pid))
678 ret = PTR_ERR(pid);
679 else
680 ret = pidfd_getfd(pid, fd);
681
682 fdput(f);
683 return ret;
684}