1 // SPDX-License-Identifier: GPL-2.0-only
3  * Generic pidhash and scalable, time-bounded PID allocator
5  * (C) 2002-2003 Nadia Yvette Chambers, IBM
7  * (C) 2002-2004 Ingo Molnar, Red Hat
9  * pid-structures are backing objects for tasks sharing a given ID to chain
13  * The hash is always changed with the tasklist_lock write-acquired,
15  * read-acquired, so there's no additional SMP locking needed here.
18  * Allocating and freeing PIDs is completely lockless. The worst-case
20  * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
70  * PID-map pages start out as NULL, they get allocated upon
94  * detach_pid()->free_pid() and another cpu that does
112 	ns = pid->numbers[pid->level].ns;  in put_pid()
113 	if (refcount_dec_and_test(&pid->count)) {  in put_pid()
114 		kmem_cache_free(ns->pid_cachep, pid);  in put_pid()
133 	for (i = 0; i <= pid->level; i++) {  in free_pid()
134 		struct upid *upid = pid->numbers + i;  in free_pid()
135 		struct pid_namespace *ns = upid->ns;  in free_pid()
136 		switch (--ns->pid_allocated) {  in free_pid()
139 			/* When all that is left in the pid namespace  in free_pid()
143 			wake_up_process(ns->child_reaper);  in free_pid()
147 			WARN_ON(ns->child_reaper);  in free_pid()
148 			ns->pid_allocated = 0;  in free_pid()
152 		idr_remove(&ns->idr, upid->nr);  in free_pid()
156 	call_rcu(&pid->rcu, delayed_put_pid);  in free_pid()
167 	int retval = -ENOMEM;  in alloc_pid()
171 	 * the most nested currently active PID namespace it tells alloc_pid()  in alloc_pid()
172 	 * which PID to set for a process in that most nested PID namespace  in alloc_pid()
175 	 * never be greater than the current ns->level + 1.  in alloc_pid()
177 	if (set_tid_size > ns->level + 1)  in alloc_pid()
178 		return ERR_PTR(-EINVAL);  in alloc_pid()
180 	pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);  in alloc_pid()
185 	pid->level = ns->level;  in alloc_pid()
187 	for (i = ns->level; i >= 0; i--) {  in alloc_pid()
191 			tid = set_tid[ns->level - i];  in alloc_pid()
193 			retval = -EINVAL;  in alloc_pid()
200 			if (tid != 1 && !tmp->child_reaper)  in alloc_pid()
202 			retval = -EPERM;  in alloc_pid()
203 			if (!checkpoint_restore_ns_capable(tmp->user_ns))  in alloc_pid()
205 			set_tid_size--;  in alloc_pid()
212 			nr = idr_alloc(&tmp->idr, NULL, tid,  in alloc_pid()
218 			if (nr == -ENOSPC)  in alloc_pid()
219 				nr = -EEXIST;  in alloc_pid()
226 			if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)  in alloc_pid()
233 			nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,  in alloc_pid()
240 			retval = (nr == -ENOSPC) ? -EAGAIN : nr;  in alloc_pid()
244 		pid->numbers[i].nr = nr;  in alloc_pid()
245 		pid->numbers[i].ns = tmp;  in alloc_pid()
246 		tmp = tmp->parent;  in alloc_pid()
250 	 * ENOMEM is not the most obvious choice especially for the case  in alloc_pid()
257 	retval = -ENOMEM;  in alloc_pid()
260 	refcount_set(&pid->count, 1);  in alloc_pid()
261 	spin_lock_init(&pid->lock);  in alloc_pid()
263 		INIT_HLIST_HEAD(&pid->tasks[type]);  in alloc_pid()
265 	init_waitqueue_head(&pid->wait_pidfd);  in alloc_pid()
266 	INIT_HLIST_HEAD(&pid->inodes);  in alloc_pid()
268 	upid = pid->numbers + ns->level;  in alloc_pid()
270 	if (!(ns->pid_allocated & PIDNS_ADDING))  in alloc_pid()
272 	for ( ; upid >= pid->numbers; --upid) {  in alloc_pid()
274 		idr_replace(&upid->ns->idr, pid, upid->nr);  in alloc_pid()
275 		upid->ns->pid_allocated++;  in alloc_pid()
287 	while (++i <= ns->level) {  in alloc_pid()
288 		upid = pid->numbers + i;  in alloc_pid()
289 		idr_remove(&upid->ns->idr, upid->nr);  in alloc_pid()
293 	if (ns->pid_allocated == PIDNS_ADDING)  in alloc_pid()
294 		idr_set_cursor(&ns->idr, 0);  in alloc_pid()
298 	kmem_cache_free(ns->pid_cachep, pid);  in alloc_pid()
305 	ns->pid_allocated &= ~PIDNS_ADDING;  in disable_pid_allocation()
311 	return idr_find(&ns->idr, nr);  in find_pid_ns()
324 		&task->thread_pid :  in task_pid_ptr()
325 		&task->signal->pids[type];  in task_pid_ptr()
329  * attach_pid() must be called with the tasklist_lock write-held.
334 	hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);  in attach_pid()
346 	hlist_del_rcu(&task->pid_links[type]);  in __change_pid()
349 	for (tmp = PIDTYPE_MAX; --tmp >= 0; )  in __change_pid()
368 void exchange_tids(struct task_struct *left, struct task_struct *right)  in exchange_tids()  argument
370 	struct pid *pid1 = left->thread_pid;  in exchange_tids()
371 	struct pid *pid2 = right->thread_pid;  in exchange_tids()
372 	struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];  in exchange_tids()
373 	struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];  in exchange_tids()
379 	rcu_assign_pointer(left->thread_pid, pid2);  in exchange_tids()
380 	rcu_assign_pointer(right->thread_pid, pid1);  in exchange_tids()
383 	WRITE_ONCE(left->pid, pid_nr(pid2));  in exchange_tids()
384 	WRITE_ONCE(right->pid, pid_nr(pid1));  in exchange_tids()
392 		new->thread_pid = old->thread_pid;  in transfer_pid()
393 	hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);  in transfer_pid()
401 		first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),  in pid_task()
477 	if (pid && ns->level <= pid->level) {  in pid_nr_ns()
478 		upid = &pid->numbers[ns->level];  in pid_nr_ns()
479 		if (upid->ns == ns)  in pid_nr_ns()
480 			nr = upid->nr;  in pid_nr_ns()
520 	return idr_get_next(&ns->idr, &nr);  in find_ge_pid()
530 		return ERR_PTR(-EBADF);  in pidfd_get_pid()
535 		*flags = f.file->f_flags;  in pidfd_get_pid()
543  * pidfd_create() - Create a new pid file descriptor.
569  * pidfd_open() - Open new pid file descriptor.
576  * @pid must be a thread-group leader. This restriction currently exists
590 		return -EINVAL;  in SYSCALL_DEFINE2()
593 		return -EINVAL;  in SYSCALL_DEFINE2()
597 		return -ESRCH;  in SYSCALL_DEFINE2()
602 		fd = -EINVAL;  in SYSCALL_DEFINE2()
631 	ret = mutex_lock_killable(&task->signal->exec_update_mutex);  in __pidfd_fget()
638 		file = ERR_PTR(-EPERM);  in __pidfd_fget()
640 	mutex_unlock(&task->signal->exec_update_mutex);  in __pidfd_fget()
642 	return file ?: ERR_PTR(-EBADF);  in __pidfd_fget()
653 		return -ESRCH;  in pidfd_getfd()
667  * sys_pidfd_getfd() - Get a file descriptor from another process
689 	/* flags is currently unused - make sure it's unset */  in SYSCALL_DEFINE3()
691 		return -EINVAL;  in SYSCALL_DEFINE3()
695 		return -EBADF;  in SYSCALL_DEFINE3()