pid.c

  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 <linux/pidfs.h>
 46#include <linux/seqlock.h>
 47#include <net/sock.h>
 48#include <uapi/linux/pidfd.h>
 49
 50struct pid init_struct_pid = {
 51	.count		= REFCOUNT_INIT(1),
 52	.tasks		= {
 53		{ .first = NULL },
 54		{ .first = NULL },
 55		{ .first = NULL },
 56	},
 57	.level		= 0,
 58	.numbers	= { {
 59		.nr		= 0,
 60		.ns		= &init_pid_ns,
 61	}, }
 62};
 63
 64static int pid_max_min = RESERVED_PIDS + 1;
 65static int pid_max_max = PID_MAX_LIMIT;
 66
 67/*
 68 * PID-map pages start out as NULL, they get allocated upon
 69 * first use and are never deallocated. This way a low pid_max
 70 * value does not cause lots of bitmaps to be allocated, but
 71 * the scheme scales to up to 4 million PIDs, runtime.
 72 */
 73struct pid_namespace init_pid_ns = {
 74	.ns = NS_COMMON_INIT(init_pid_ns),
 75	.idr = IDR_INIT(init_pid_ns.idr),
 76	.pid_allocated = PIDNS_ADDING,
 77	.level = 0,
 78	.child_reaper = &init_task,
 79	.user_ns = &init_user_ns,
 80	.pid_max = PID_MAX_DEFAULT,
 81#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
 82	.memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC,
 83#endif
 84};
 85EXPORT_SYMBOL_GPL(init_pid_ns);
 86
 87static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
 88seqcount_spinlock_t pidmap_lock_seq = SEQCNT_SPINLOCK_ZERO(pidmap_lock_seq, &pidmap_lock);
 89
 90void put_pid(struct pid *pid)
 91{
 92	struct pid_namespace *ns;
 93
 94	if (!pid)
 95		return;
 96
 97	ns = pid->numbers[pid->level].ns;
 98	if (refcount_dec_and_test(&pid->count)) {
 99		pidfs_free_pid(pid);
100		kmem_cache_free(ns->pid_cachep, pid);
101		put_pid_ns(ns);
102	}
103}
104EXPORT_SYMBOL_GPL(put_pid);
105
106static void delayed_put_pid(struct rcu_head *rhp)
107{
108	struct pid *pid = container_of(rhp, struct pid, rcu);
109	put_pid(pid);
110}
111
112void free_pid(struct pid *pid)
113{
114	int i;
115	struct pid_namespace *active_ns;
116
117	lockdep_assert_not_held(&tasklist_lock);
118
119	active_ns = pid->numbers[pid->level].ns;
120	ns_ref_active_put(active_ns);
121
122	spin_lock(&pidmap_lock);
123	for (i = 0; i <= pid->level; i++) {
124		struct upid *upid = pid->numbers + i;
125		struct pid_namespace *ns = upid->ns;
126		switch (--ns->pid_allocated) {
127		case 2:
128		case 1:
129			/* When all that is left in the pid namespace
130			 * is the reaper wake up the reaper.  The reaper
131			 * may be sleeping in zap_pid_ns_processes().
132			 */
133			wake_up_process(ns->child_reaper);
134			break;
135		case PIDNS_ADDING:
136			/* Handle a fork failure of the first process */
137			WARN_ON(ns->child_reaper);
138			ns->pid_allocated = 0;
139			break;
140		}
141
142		idr_remove(&ns->idr, upid->nr);
143	}
144	pidfs_remove_pid(pid);
145	spin_unlock(&pidmap_lock);
146
147	call_rcu(&pid->rcu, delayed_put_pid);
148}
149
150void free_pids(struct pid **pids)
151{
152	int tmp;
153
154	/*
155	 * This can batch pidmap_lock.
156	 */
157	for (tmp = PIDTYPE_MAX; --tmp >= 0; )
158		if (pids[tmp])
159			free_pid(pids[tmp]);
160}
161
162struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
163		      size_t set_tid_size)
164{
165	struct pid *pid;
166	enum pid_type type;
167	int i, nr;
168	struct pid_namespace *tmp;
169	struct upid *upid;
170	int retval = -ENOMEM;
171
172	/*
173	 * set_tid_size contains the size of the set_tid array. Starting at
174	 * the most nested currently active PID namespace it tells alloc_pid()
175	 * which PID to set for a process in that most nested PID namespace
176	 * up to set_tid_size PID namespaces. It does not have to set the PID
177	 * for a process in all nested PID namespaces but set_tid_size must
178	 * never be greater than the current ns->level + 1.
179	 */
180	if (set_tid_size > ns->level + 1)
181		return ERR_PTR(-EINVAL);
182
183	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
184	if (!pid)
185		return ERR_PTR(retval);
186
187	tmp = ns;
188	pid->level = ns->level;
189
190	for (i = ns->level; i >= 0; i--) {
191		int tid = 0;
192		int pid_max = READ_ONCE(tmp->pid_max);
193
194		if (set_tid_size) {
195			tid = set_tid[ns->level - i];
196
197			retval = -EINVAL;
198			if (tid < 1 || tid >= pid_max)
199				goto out_free;
200			/*
201			 * Also fail if a PID != 1 is requested and
202			 * no PID 1 exists.
203			 */
204			if (tid != 1 && !tmp->child_reaper)
205				goto out_free;
206			retval = -EPERM;
207			if (!checkpoint_restore_ns_capable(tmp->user_ns))
208				goto out_free;
209			set_tid_size--;
210		}
211
212		idr_preload(GFP_KERNEL);
213		spin_lock(&pidmap_lock);
214
215		if (tid) {
216			nr = idr_alloc(&tmp->idr, NULL, tid,
217				       tid + 1, GFP_ATOMIC);
218			/*
219			 * If ENOSPC is returned it means that the PID is
220			 * alreay in use. Return EEXIST in that case.
221			 */
222			if (nr == -ENOSPC)
223				nr = -EEXIST;
224		} else {
225			int pid_min = 1;
226			/*
227			 * init really needs pid 1, but after reaching the
228			 * maximum wrap back to RESERVED_PIDS
229			 */
230			if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
231				pid_min = RESERVED_PIDS;
232
233			/*
234			 * Store a null pointer so find_pid_ns does not find
235			 * a partially initialized PID (see below).
236			 */
237			nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
238					      pid_max, GFP_ATOMIC);
239		}
240		spin_unlock(&pidmap_lock);
241		idr_preload_end();
242
243		if (nr < 0) {
244			retval = (nr == -ENOSPC) ? -EAGAIN : nr;
245			goto out_free;
246		}
247
248		pid->numbers[i].nr = nr;
249		pid->numbers[i].ns = tmp;
250		tmp = tmp->parent;
251	}
252
253	/*
254	 * ENOMEM is not the most obvious choice especially for the case
255	 * where the child subreaper has already exited and the pid
256	 * namespace denies the creation of any new processes. But ENOMEM
257	 * is what we have exposed to userspace for a long time and it is
258	 * documented behavior for pid namespaces. So we can't easily
259	 * change it even if there were an error code better suited.
260	 */
261	retval = -ENOMEM;
262
263	get_pid_ns(ns);
264	refcount_set(&pid->count, 1);
265	spin_lock_init(&pid->lock);
266	for (type = 0; type < PIDTYPE_MAX; ++type)
267		INIT_HLIST_HEAD(&pid->tasks[type]);
268
269	init_waitqueue_head(&pid->wait_pidfd);
270	INIT_HLIST_HEAD(&pid->inodes);
271
272	upid = pid->numbers + ns->level;
273	idr_preload(GFP_KERNEL);
274	spin_lock(&pidmap_lock);
275	if (!(ns->pid_allocated & PIDNS_ADDING))
276		goto out_unlock;
277	pidfs_add_pid(pid);
278	for ( ; upid >= pid->numbers; --upid) {
279		/* Make the PID visible to find_pid_ns. */
280		idr_replace(&upid->ns->idr, pid, upid->nr);
281		upid->ns->pid_allocated++;
282	}
283	spin_unlock(&pidmap_lock);
284	idr_preload_end();
285	ns_ref_active_get(ns);
286
287	return pid;
288
289out_unlock:
290	spin_unlock(&pidmap_lock);
291	idr_preload_end();
292	put_pid_ns(ns);
293
294out_free:
295	spin_lock(&pidmap_lock);
296	while (++i <= ns->level) {
297		upid = pid->numbers + i;
298		idr_remove(&upid->ns->idr, upid->nr);
299	}
300
301	/* On failure to allocate the first pid, reset the state */
302	if (ns->pid_allocated == PIDNS_ADDING)
303		idr_set_cursor(&ns->idr, 0);
304
305	spin_unlock(&pidmap_lock);
306
307	kmem_cache_free(ns->pid_cachep, pid);
308	return ERR_PTR(retval);
309}
310
311void disable_pid_allocation(struct pid_namespace *ns)
312{
313	spin_lock(&pidmap_lock);
314	ns->pid_allocated &= ~PIDNS_ADDING;
315	spin_unlock(&pidmap_lock);
316}
317
318struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
319{
320	return idr_find(&ns->idr, nr);
321}
322EXPORT_SYMBOL_GPL(find_pid_ns);
323
324struct pid *find_vpid(int nr)
325{
326	return find_pid_ns(nr, task_active_pid_ns(current));
327}
328EXPORT_SYMBOL_GPL(find_vpid);
329
330static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
331{
332	return (type == PIDTYPE_PID) ?
333		&task->thread_pid :
334		&task->signal->pids[type];
335}
336
337/*
338 * attach_pid() must be called with the tasklist_lock write-held.
339 */
340void attach_pid(struct task_struct *task, enum pid_type type)
341{
342	struct pid *pid;
343
344	lockdep_assert_held_write(&tasklist_lock);
345
346	pid = *task_pid_ptr(task, type);
347	hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
348}
349
350static void __change_pid(struct pid **pids, struct task_struct *task,
351			 enum pid_type type, struct pid *new)
352{
353	struct pid **pid_ptr, *pid;
354	int tmp;
355
356	lockdep_assert_held_write(&tasklist_lock);
357
358	pid_ptr = task_pid_ptr(task, type);
359	pid = *pid_ptr;
360
361	hlist_del_rcu(&task->pid_links[type]);
362	*pid_ptr = new;
363
364	for (tmp = PIDTYPE_MAX; --tmp >= 0; )
365		if (pid_has_task(pid, tmp))
366			return;
367
368	WARN_ON(pids[type]);
369	pids[type] = pid;
370}
371
372void detach_pid(struct pid **pids, struct task_struct *task, enum pid_type type)
373{
374	__change_pid(pids, task, type, NULL);
375}
376
377void change_pid(struct pid **pids, struct task_struct *task, enum pid_type type,
378		struct pid *pid)
379{
380	__change_pid(pids, task, type, pid);
381	attach_pid(task, type);
382}
383
384void exchange_tids(struct task_struct *left, struct task_struct *right)
385{
386	struct pid *pid1 = left->thread_pid;
387	struct pid *pid2 = right->thread_pid;
388	struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
389	struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
390
391	lockdep_assert_held_write(&tasklist_lock);
392
393	/* Swap the single entry tid lists */
394	hlists_swap_heads_rcu(head1, head2);
395
396	/* Swap the per task_struct pid */
397	rcu_assign_pointer(left->thread_pid, pid2);
398	rcu_assign_pointer(right->thread_pid, pid1);
399
400	/* Swap the cached value */
401	WRITE_ONCE(left->pid, pid_nr(pid2));
402	WRITE_ONCE(right->pid, pid_nr(pid1));
403}
404
405/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
406void transfer_pid(struct task_struct *old, struct task_struct *new,
407			   enum pid_type type)
408{
409	WARN_ON_ONCE(type == PIDTYPE_PID);
410	lockdep_assert_held_write(&tasklist_lock);
411	hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
412}
413
414struct task_struct *pid_task(struct pid *pid, enum pid_type type)
415{
416	struct task_struct *result = NULL;
417	if (pid) {
418		struct hlist_node *first;
419		first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
420					      lockdep_tasklist_lock_is_held());
421		if (first)
422			result = hlist_entry(first, struct task_struct, pid_links[(type)]);
423	}
424	return result;
425}
426EXPORT_SYMBOL(pid_task);
427
428/*
429 * Must be called under rcu_read_lock().
430 */
431struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
432{
433	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
434			 "find_task_by_pid_ns() needs rcu_read_lock() protection");
435	return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
436}
437
438struct task_struct *find_task_by_vpid(pid_t vnr)
439{
440	return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
441}
442
443struct task_struct *find_get_task_by_vpid(pid_t nr)
444{
445	struct task_struct *task;
446
447	rcu_read_lock();
448	task = find_task_by_vpid(nr);
449	if (task)
450		get_task_struct(task);
451	rcu_read_unlock();
452
453	return task;
454}
455
456struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
457{
458	struct pid *pid;
459	rcu_read_lock();
460	pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
461	rcu_read_unlock();
462	return pid;
463}
464EXPORT_SYMBOL_GPL(get_task_pid);
465
466struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
467{
468	struct task_struct *result;
469	rcu_read_lock();
470	result = pid_task(pid, type);
471	if (result)
472		get_task_struct(result);
473	rcu_read_unlock();
474	return result;
475}
476EXPORT_SYMBOL_GPL(get_pid_task);
477
478struct pid *find_get_pid(pid_t nr)
479{
480	struct pid *pid;
481
482	rcu_read_lock();
483	pid = get_pid(find_vpid(nr));
484	rcu_read_unlock();
485
486	return pid;
487}
488EXPORT_SYMBOL_GPL(find_get_pid);
489
490pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
491{
492	struct upid *upid;
493	pid_t nr = 0;
494
495	if (pid && ns && ns->level <= pid->level) {
496		upid = &pid->numbers[ns->level];
497		if (upid->ns == ns)
498			nr = upid->nr;
499	}
500	return nr;
501}
502EXPORT_SYMBOL_GPL(pid_nr_ns);
503
504pid_t pid_vnr(struct pid *pid)
505{
506	return pid_nr_ns(pid, task_active_pid_ns(current));
507}
508EXPORT_SYMBOL_GPL(pid_vnr);
509
510pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
511			struct pid_namespace *ns)
512{
513	pid_t nr = 0;
514
515	rcu_read_lock();
516	if (!ns)
517		ns = task_active_pid_ns(current);
518	if (ns)
519		nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
520	rcu_read_unlock();
521
522	return nr;
523}
524EXPORT_SYMBOL(__task_pid_nr_ns);
525
526struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
527{
528	return ns_of_pid(task_pid(tsk));
529}
530EXPORT_SYMBOL_GPL(task_active_pid_ns);
531
532/*
533 * Used by proc to find the first pid that is greater than or equal to nr.
534 *
535 * If there is a pid at nr this function is exactly the same as find_pid_ns.
536 */
537struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
538{
539	return idr_get_next(&ns->idr, &nr);
540}
541EXPORT_SYMBOL_GPL(find_ge_pid);
542
543struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
544{
545	CLASS(fd, f)(fd);
546	struct pid *pid;
547
548	if (fd_empty(f))
549		return ERR_PTR(-EBADF);
550
551	pid = pidfd_pid(fd_file(f));
552	if (!IS_ERR(pid)) {
553		get_pid(pid);
554		*flags = fd_file(f)->f_flags;
555	}
556	return pid;
557}
558
559/**
560 * pidfd_get_task() - Get the task associated with a pidfd
561 *
562 * @pidfd: pidfd for which to get the task
563 * @flags: flags associated with this pidfd
564 *
565 * Return the task associated with @pidfd. The function takes a reference on
566 * the returned task. The caller is responsible for releasing that reference.
567 *
568 * Return: On success, the task_struct associated with the pidfd.
569 *	   On error, a negative errno number will be returned.
570 */
571struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
572{
573	unsigned int f_flags = 0;
574	struct pid *pid;
575	struct task_struct *task;
576	enum pid_type type;
577
578	switch (pidfd) {
579	case  PIDFD_SELF_THREAD:
580		type = PIDTYPE_PID;
581		pid = get_task_pid(current, type);
582		break;
583	case  PIDFD_SELF_THREAD_GROUP:
584		type = PIDTYPE_TGID;
585		pid = get_task_pid(current, type);
586		break;
587	default:
588		pid = pidfd_get_pid(pidfd, &f_flags);
589		if (IS_ERR(pid))
590			return ERR_CAST(pid);
591		type = PIDTYPE_TGID;
592		break;
593	}
594
595	task = get_pid_task(pid, type);
596	put_pid(pid);
597	if (!task)
598		return ERR_PTR(-ESRCH);
599
600	*flags = f_flags;
601	return task;
602}
603
604/**
605 * pidfd_create() - Create a new pid file descriptor.
606 *
607 * @pid:   struct pid that the pidfd will reference
608 * @flags: flags to pass
609 *
610 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
611 *
612 * Note, that this function can only be called after the fd table has
613 * been unshared to avoid leaking the pidfd to the new process.
614 *
615 * This symbol should not be explicitly exported to loadable modules.
616 *
617 * Return: On success, a cloexec pidfd is returned.
618 *         On error, a negative errno number will be returned.
619 */
620static int pidfd_create(struct pid *pid, unsigned int flags)
621{
622	int pidfd;
623	struct file *pidfd_file;
624
625	pidfd = pidfd_prepare(pid, flags, &pidfd_file);
626	if (pidfd < 0)
627		return pidfd;
628
629	fd_install(pidfd, pidfd_file);
630	return pidfd;
631}
632
633/**
634 * sys_pidfd_open() - Open new pid file descriptor.
635 *
636 * @pid:   pid for which to retrieve a pidfd
637 * @flags: flags to pass
638 *
639 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
640 * the task identified by @pid. Without PIDFD_THREAD flag the target task
641 * must be a thread-group leader.
642 *
643 * Return: On success, a cloexec pidfd is returned.
644 *         On error, a negative errno number will be returned.
645 */
646SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
647{
648	int fd;
649	struct pid *p;
650
651	if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD))
652		return -EINVAL;
653
654	if (pid <= 0)
655		return -EINVAL;
656
657	p = find_get_pid(pid);
658	if (!p)
659		return -ESRCH;
660
661	fd = pidfd_create(p, flags);
662
663	put_pid(p);
664	return fd;
665}
666
667#ifdef CONFIG_SYSCTL
668static struct ctl_table_set *pid_table_root_lookup(struct ctl_table_root *root)
669{
670	return &task_active_pid_ns(current)->set;
671}
672
673static int set_is_seen(struct ctl_table_set *set)
674{
675	return &task_active_pid_ns(current)->set == set;
676}
677
678static int pid_table_root_permissions(struct ctl_table_header *head,
679				      const struct ctl_table *table)
680{
681	struct pid_namespace *pidns =
682		container_of(head->set, struct pid_namespace, set);
683	int mode = table->mode;
684
685	if (ns_capable_noaudit(pidns->user_ns, CAP_SYS_ADMIN) ||
686	    uid_eq(current_euid(), make_kuid(pidns->user_ns, 0)))
687		mode = (mode & S_IRWXU) >> 6;
688	else if (in_egroup_p(make_kgid(pidns->user_ns, 0)))
689		mode = (mode & S_IRWXG) >> 3;
690	else
691		mode = mode & S_IROTH;
692	return (mode << 6) | (mode << 3) | mode;
693}
694
695static void pid_table_root_set_ownership(struct ctl_table_header *head,
696					 kuid_t *uid, kgid_t *gid)
697{
698	struct pid_namespace *pidns =
699		container_of(head->set, struct pid_namespace, set);
700	kuid_t ns_root_uid;
701	kgid_t ns_root_gid;
702
703	ns_root_uid = make_kuid(pidns->user_ns, 0);
704	if (uid_valid(ns_root_uid))
705		*uid = ns_root_uid;
706
707	ns_root_gid = make_kgid(pidns->user_ns, 0);
708	if (gid_valid(ns_root_gid))
709		*gid = ns_root_gid;
710}
711
712static struct ctl_table_root pid_table_root = {
713	.lookup		= pid_table_root_lookup,
714	.permissions	= pid_table_root_permissions,
715	.set_ownership	= pid_table_root_set_ownership,
716};
717
718static int proc_do_cad_pid(const struct ctl_table *table, int write, void *buffer,
719		size_t *lenp, loff_t *ppos)
720{
721	struct pid *new_pid;
722	pid_t tmp_pid;
723	int r;
724	struct ctl_table tmp_table = *table;
725
726	tmp_pid = pid_vnr(cad_pid);
727	tmp_table.data = &tmp_pid;
728
729	r = proc_dointvec(&tmp_table, write, buffer, lenp, ppos);
730	if (r || !write)
731		return r;
732
733	new_pid = find_get_pid(tmp_pid);
734	if (!new_pid)
735		return -ESRCH;
736
737	put_pid(xchg(&cad_pid, new_pid));
738	return 0;
739}
740
741static const struct ctl_table pid_table[] = {
742	{
743		.procname	= "pid_max",
744		.data		= &init_pid_ns.pid_max,
745		.maxlen		= sizeof(int),
746		.mode		= 0644,
747		.proc_handler	= proc_dointvec_minmax,
748		.extra1		= &pid_max_min,
749		.extra2		= &pid_max_max,
750	},
751#ifdef CONFIG_PROC_SYSCTL
752	{
753		.procname	= "cad_pid",
754		.maxlen		= sizeof(int),
755		.mode		= 0600,
756		.proc_handler	= proc_do_cad_pid,
757	},
758#endif
759};
760#endif
761
762int register_pidns_sysctls(struct pid_namespace *pidns)
763{
764#ifdef CONFIG_SYSCTL
765	struct ctl_table *tbl;
766
767	setup_sysctl_set(&pidns->set, &pid_table_root, set_is_seen);
768
769	tbl = kmemdup(pid_table, sizeof(pid_table), GFP_KERNEL);
770	if (!tbl)
771		return -ENOMEM;
772	tbl->data = &pidns->pid_max;
773	pidns->pid_max = min(pid_max_max, max_t(int, pidns->pid_max,
774			     PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
775
776	pidns->sysctls = __register_sysctl_table(&pidns->set, "kernel", tbl,
777						 ARRAY_SIZE(pid_table));
778	if (!pidns->sysctls) {
779		kfree(tbl);
780		retire_sysctl_set(&pidns->set);
781		return -ENOMEM;
782	}
783#endif
784	return 0;
785}
786
787void unregister_pidns_sysctls(struct pid_namespace *pidns)
788{
789#ifdef CONFIG_SYSCTL
790	const struct ctl_table *tbl;
791
792	tbl = pidns->sysctls->ctl_table_arg;
793	unregister_sysctl_table(pidns->sysctls);
794	retire_sysctl_set(&pidns->set);
795	kfree(tbl);
796#endif
797}
798
799void __init pid_idr_init(void)
800{
801	/* Verify no one has done anything silly: */
802	BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
803
804	/* bump default and minimum pid_max based on number of cpus */
805	init_pid_ns.pid_max = min(pid_max_max, max_t(int, init_pid_ns.pid_max,
806				  PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
807	pid_max_min = max_t(int, pid_max_min,
808				PIDS_PER_CPU_MIN * num_possible_cpus());
809	pr_info("pid_max: default: %u minimum: %u\n", init_pid_ns.pid_max, pid_max_min);
810
811	idr_init(&init_pid_ns.idr);
812
813	init_pid_ns.pid_cachep = kmem_cache_create("pid",
814			struct_size_t(struct pid, numbers, 1),
815			__alignof__(struct pid),
816			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
817			NULL);
818}
819
820static __init int pid_namespace_sysctl_init(void)
821{
822#ifdef CONFIG_SYSCTL
823	/* "kernel" directory will have already been initialized. */
824	BUG_ON(register_pidns_sysctls(&init_pid_ns));
825#endif
826	return 0;
827}
828subsys_initcall(pid_namespace_sysctl_init);
829
830static struct file *__pidfd_fget(struct task_struct *task, int fd)
831{
832	struct file *file;
833	int ret;
834
835	ret = down_read_killable(&task->signal->exec_update_lock);
836	if (ret)
837		return ERR_PTR(ret);
838
839	if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
840		file = fget_task(task, fd);
841	else
842		file = ERR_PTR(-EPERM);
843
844	up_read(&task->signal->exec_update_lock);
845
846	if (!file) {
847		/*
848		 * It is possible that the target thread is exiting; it can be
849		 * either:
850		 * 1. before exit_signals(), which gives a real fd
851		 * 2. before exit_files() takes the task_lock() gives a real fd
852		 * 3. after exit_files() releases task_lock(), ->files is NULL;
853		 *    this has PF_EXITING, since it was set in exit_signals(),
854		 *    __pidfd_fget() returns EBADF.
855		 * In case 3 we get EBADF, but that really means ESRCH, since
856		 * the task is currently exiting and has freed its files
857		 * struct, so we fix it up.
858		 */
859		if (task->flags & PF_EXITING)
860			file = ERR_PTR(-ESRCH);
861		else
862			file = ERR_PTR(-EBADF);
863	}
864
865	return file;
866}
867
868static int pidfd_getfd(struct pid *pid, int fd)
869{
870	struct task_struct *task;
871	struct file *file;
872	int ret;
873
874	task = get_pid_task(pid, PIDTYPE_PID);
875	if (!task)
876		return -ESRCH;
877
878	file = __pidfd_fget(task, fd);
879	put_task_struct(task);
880	if (IS_ERR(file))
881		return PTR_ERR(file);
882
883	ret = receive_fd(file, NULL, O_CLOEXEC);
884	fput(file);
885
886	return ret;
887}
888
889/**
890 * sys_pidfd_getfd() - Get a file descriptor from another process
891 *
892 * @pidfd:	the pidfd file descriptor of the process
893 * @fd:		the file descriptor number to get
894 * @flags:	flags on how to get the fd (reserved)
895 *
896 * This syscall gets a copy of a file descriptor from another process
897 * based on the pidfd, and file descriptor number. It requires that
898 * the calling process has the ability to ptrace the process represented
899 * by the pidfd. The process which is having its file descriptor copied
900 * is otherwise unaffected.
901 *
902 * Return: On success, a cloexec file descriptor is returned.
903 *         On error, a negative errno number will be returned.
904 */
905SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
906		unsigned int, flags)
907{
908	struct pid *pid;
909
910	/* flags is currently unused - make sure it's unset */
911	if (flags)
912		return -EINVAL;
913
914	CLASS(fd, f)(pidfd);
915	if (fd_empty(f))
916		return -EBADF;
917
918	pid = pidfd_pid(fd_file(f));
919	if (IS_ERR(pid))
920		return PTR_ERR(pid);
921
922	return pidfd_getfd(pid, fd);
923}