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
6 * (C) 2004 Nadia Yvette Chambers, Oracle
7 * (C) 2002-2004 Ingo Molnar, Red Hat
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.
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.
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).
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
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>
46 struct pid init_struct_pid = {
47 .count = REFCOUNT_INIT(1),
60 int pid_max = PID_MAX_DEFAULT;
62 #define RESERVED_PIDS 300
64 int pid_max_min = RESERVED_PIDS + 1;
65 int pid_max_max = PID_MAX_LIMIT;
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.
73 struct pid_namespace init_pid_ns = {
75 .idr = IDR_INIT(init_pid_ns.idr),
76 .pid_allocated = PIDNS_ADDING,
78 .child_reaper = &init_task,
79 .user_ns = &init_user_ns,
80 .ns.inum = PROC_PID_INIT_INO,
82 .ns.ops = &pidns_operations,
85 EXPORT_SYMBOL_GPL(init_pid_ns);
88 * Note: disable interrupts while the pidmap_lock is held as an
89 * interrupt might come in and do read_lock(&tasklist_lock).
91 * If we don't disable interrupts there is a nasty deadlock between
92 * detach_pid()->free_pid() and another cpu that does
93 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
94 * read_lock(&tasklist_lock);
96 * After we clean up the tasklist_lock and know there are no
97 * irq handlers that take it we can leave the interrupts enabled.
98 * For now it is easier to be safe than to prove it can't happen.
101 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
103 void put_pid(struct pid *pid)
105 struct pid_namespace *ns;
110 ns = pid->numbers[pid->level].ns;
111 if (refcount_dec_and_test(&pid->count)) {
112 kmem_cache_free(ns->pid_cachep, pid);
116 EXPORT_SYMBOL_GPL(put_pid);
118 static void delayed_put_pid(struct rcu_head *rhp)
120 struct pid *pid = container_of(rhp, struct pid, rcu);
124 void free_pid(struct pid *pid)
126 /* We can be called with write_lock_irq(&tasklist_lock) held */
130 spin_lock_irqsave(&pidmap_lock, flags);
131 for (i = 0; i <= pid->level; i++) {
132 struct upid *upid = pid->numbers + i;
133 struct pid_namespace *ns = upid->ns;
134 switch (--ns->pid_allocated) {
137 /* When all that is left in the pid namespace
138 * is the reaper wake up the reaper. The reaper
139 * may be sleeping in zap_pid_ns_processes().
141 wake_up_process(ns->child_reaper);
144 /* Handle a fork failure of the first process */
145 WARN_ON(ns->child_reaper);
146 ns->pid_allocated = 0;
149 schedule_work(&ns->proc_work);
153 idr_remove(&ns->idr, upid->nr);
155 spin_unlock_irqrestore(&pidmap_lock, flags);
157 call_rcu(&pid->rcu, delayed_put_pid);
160 struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
166 struct pid_namespace *tmp;
168 int retval = -ENOMEM;
171 * set_tid_size contains the size of the set_tid array. Starting at
172 * the most nested currently active PID namespace it tells alloc_pid()
173 * which PID to set for a process in that most nested PID namespace
174 * up to set_tid_size PID namespaces. It does not have to set the PID
175 * for a process in all nested PID namespaces but set_tid_size must
176 * never be greater than the current ns->level + 1.
178 if (set_tid_size > ns->level + 1)
179 return ERR_PTR(-EINVAL);
181 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
183 return ERR_PTR(retval);
186 pid->level = ns->level;
188 for (i = ns->level; i >= 0; i--) {
192 tid = set_tid[ns->level - i];
195 if (tid < 1 || tid >= pid_max)
198 * Also fail if a PID != 1 is requested and
201 if (tid != 1 && !tmp->child_reaper)
204 if (!ns_capable(tmp->user_ns, CAP_SYS_ADMIN))
209 idr_preload(GFP_KERNEL);
210 spin_lock_irq(&pidmap_lock);
213 nr = idr_alloc(&tmp->idr, NULL, tid,
214 tid + 1, GFP_ATOMIC);
216 * If ENOSPC is returned it means that the PID is
217 * alreay in use. Return EEXIST in that case.
224 * init really needs pid 1, but after reaching the
225 * maximum wrap back to RESERVED_PIDS
227 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
228 pid_min = RESERVED_PIDS;
231 * Store a null pointer so find_pid_ns does not find
232 * a partially initialized PID (see below).
234 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
235 pid_max, GFP_ATOMIC);
237 spin_unlock_irq(&pidmap_lock);
241 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
245 pid->numbers[i].nr = nr;
246 pid->numbers[i].ns = tmp;
250 if (unlikely(is_child_reaper(pid))) {
251 if (pid_ns_prepare_proc(ns))
256 refcount_set(&pid->count, 1);
257 for (type = 0; type < PIDTYPE_MAX; ++type)
258 INIT_HLIST_HEAD(&pid->tasks[type]);
260 init_waitqueue_head(&pid->wait_pidfd);
262 upid = pid->numbers + ns->level;
263 spin_lock_irq(&pidmap_lock);
264 if (!(ns->pid_allocated & PIDNS_ADDING))
266 for ( ; upid >= pid->numbers; --upid) {
267 /* Make the PID visible to find_pid_ns. */
268 idr_replace(&upid->ns->idr, pid, upid->nr);
269 upid->ns->pid_allocated++;
271 spin_unlock_irq(&pidmap_lock);
276 spin_unlock_irq(&pidmap_lock);
280 spin_lock_irq(&pidmap_lock);
281 while (++i <= ns->level) {
282 upid = pid->numbers + i;
283 idr_remove(&upid->ns->idr, upid->nr);
286 /* On failure to allocate the first pid, reset the state */
287 if (ns->pid_allocated == PIDNS_ADDING)
288 idr_set_cursor(&ns->idr, 0);
290 spin_unlock_irq(&pidmap_lock);
292 kmem_cache_free(ns->pid_cachep, pid);
293 return ERR_PTR(retval);
296 void disable_pid_allocation(struct pid_namespace *ns)
298 spin_lock_irq(&pidmap_lock);
299 ns->pid_allocated &= ~PIDNS_ADDING;
300 spin_unlock_irq(&pidmap_lock);
303 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
305 return idr_find(&ns->idr, nr);
307 EXPORT_SYMBOL_GPL(find_pid_ns);
309 struct pid *find_vpid(int nr)
311 return find_pid_ns(nr, task_active_pid_ns(current));
313 EXPORT_SYMBOL_GPL(find_vpid);
315 static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
317 return (type == PIDTYPE_PID) ?
319 &task->signal->pids[type];
323 * attach_pid() must be called with the tasklist_lock write-held.
325 void attach_pid(struct task_struct *task, enum pid_type type)
327 struct pid *pid = *task_pid_ptr(task, type);
328 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
331 static void __change_pid(struct task_struct *task, enum pid_type type,
334 struct pid **pid_ptr = task_pid_ptr(task, type);
340 hlist_del_rcu(&task->pid_links[type]);
343 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
344 if (pid_has_task(pid, tmp))
350 void detach_pid(struct task_struct *task, enum pid_type type)
352 __change_pid(task, type, NULL);
355 void change_pid(struct task_struct *task, enum pid_type type,
358 __change_pid(task, type, pid);
359 attach_pid(task, type);
362 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
363 void transfer_pid(struct task_struct *old, struct task_struct *new,
366 if (type == PIDTYPE_PID)
367 new->thread_pid = old->thread_pid;
368 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
371 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
373 struct task_struct *result = NULL;
375 struct hlist_node *first;
376 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
377 lockdep_tasklist_lock_is_held());
379 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
383 EXPORT_SYMBOL(pid_task);
386 * Must be called under rcu_read_lock().
388 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
390 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
391 "find_task_by_pid_ns() needs rcu_read_lock() protection");
392 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
395 struct task_struct *find_task_by_vpid(pid_t vnr)
397 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
400 struct task_struct *find_get_task_by_vpid(pid_t nr)
402 struct task_struct *task;
405 task = find_task_by_vpid(nr);
407 get_task_struct(task);
413 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
417 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
421 EXPORT_SYMBOL_GPL(get_task_pid);
423 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
425 struct task_struct *result;
427 result = pid_task(pid, type);
429 get_task_struct(result);
433 EXPORT_SYMBOL_GPL(get_pid_task);
435 struct pid *find_get_pid(pid_t nr)
440 pid = get_pid(find_vpid(nr));
445 EXPORT_SYMBOL_GPL(find_get_pid);
447 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
452 if (pid && ns->level <= pid->level) {
453 upid = &pid->numbers[ns->level];
459 EXPORT_SYMBOL_GPL(pid_nr_ns);
461 pid_t pid_vnr(struct pid *pid)
463 return pid_nr_ns(pid, task_active_pid_ns(current));
465 EXPORT_SYMBOL_GPL(pid_vnr);
467 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
468 struct pid_namespace *ns)
474 ns = task_active_pid_ns(current);
475 if (likely(pid_alive(task)))
476 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
481 EXPORT_SYMBOL(__task_pid_nr_ns);
483 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
485 return ns_of_pid(task_pid(tsk));
487 EXPORT_SYMBOL_GPL(task_active_pid_ns);
490 * Used by proc to find the first pid that is greater than or equal to nr.
492 * If there is a pid at nr this function is exactly the same as find_pid_ns.
494 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
496 return idr_get_next(&ns->idr, &nr);
500 * pidfd_create() - Create a new pid file descriptor.
502 * @pid: struct pid that the pidfd will reference
504 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
506 * Note, that this function can only be called after the fd table has
507 * been unshared to avoid leaking the pidfd to the new process.
509 * Return: On success, a cloexec pidfd is returned.
510 * On error, a negative errno number will be returned.
512 static int pidfd_create(struct pid *pid)
516 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
525 * pidfd_open() - Open new pid file descriptor.
527 * @pid: pid for which to retrieve a pidfd
528 * @flags: flags to pass
530 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
531 * the process identified by @pid. Currently, the process identified by
532 * @pid must be a thread-group leader. This restriction currently exists
533 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
534 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
537 * Return: On success, a cloexec pidfd is returned.
538 * On error, a negative errno number will be returned.
540 SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
551 p = find_get_pid(pid);
555 if (pid_has_task(p, PIDTYPE_TGID))
556 fd = pidfd_create(p);
564 void __init pid_idr_init(void)
566 /* Verify no one has done anything silly: */
567 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
569 /* bump default and minimum pid_max based on number of cpus */
570 pid_max = min(pid_max_max, max_t(int, pid_max,
571 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
572 pid_max_min = max_t(int, pid_max_min,
573 PIDS_PER_CPU_MIN * num_possible_cpus());
574 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
576 idr_init(&init_pid_ns.idr);
578 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
579 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
582 static struct file *__pidfd_fget(struct task_struct *task, int fd)
587 ret = mutex_lock_killable(&task->signal->cred_guard_mutex);
591 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
592 file = fget_task(task, fd);
594 file = ERR_PTR(-EPERM);
596 mutex_unlock(&task->signal->cred_guard_mutex);
598 return file ?: ERR_PTR(-EBADF);
601 static int pidfd_getfd(struct pid *pid, int fd)
603 struct task_struct *task;
607 task = get_pid_task(pid, PIDTYPE_PID);
611 file = __pidfd_fget(task, fd);
612 put_task_struct(task);
614 return PTR_ERR(file);
616 ret = security_file_receive(file);
622 ret = get_unused_fd_flags(O_CLOEXEC);
626 fd_install(ret, file);
632 * sys_pidfd_getfd() - Get a file descriptor from another process
634 * @pidfd: the pidfd file descriptor of the process
635 * @fd: the file descriptor number to get
636 * @flags: flags on how to get the fd (reserved)
638 * This syscall gets a copy of a file descriptor from another process
639 * based on the pidfd, and file descriptor number. It requires that
640 * the calling process has the ability to ptrace the process represented
641 * by the pidfd. The process which is having its file descriptor copied
642 * is otherwise unaffected.
644 * Return: On success, a cloexec file descriptor is returned.
645 * On error, a negative errno number will be returned.
647 SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
654 /* flags is currently unused - make sure it's unset */
662 pid = pidfd_pid(f.file);
666 ret = pidfd_getfd(pid, fd);