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;
150 idr_remove(&ns->idr, upid->nr);
152 spin_unlock_irqrestore(&pidmap_lock, flags);
154 call_rcu(&pid->rcu, delayed_put_pid);
157 struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
163 struct pid_namespace *tmp;
165 int retval = -ENOMEM;
168 * set_tid_size contains the size of the set_tid array. Starting at
169 * the most nested currently active PID namespace it tells alloc_pid()
170 * which PID to set for a process in that most nested PID namespace
171 * up to set_tid_size PID namespaces. It does not have to set the PID
172 * for a process in all nested PID namespaces but set_tid_size must
173 * never be greater than the current ns->level + 1.
175 if (set_tid_size > ns->level + 1)
176 return ERR_PTR(-EINVAL);
178 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
180 return ERR_PTR(retval);
183 pid->level = ns->level;
185 for (i = ns->level; i >= 0; i--) {
189 tid = set_tid[ns->level - i];
192 if (tid < 1 || tid >= pid_max)
195 * Also fail if a PID != 1 is requested and
198 if (tid != 1 && !tmp->child_reaper)
201 if (!ns_capable(tmp->user_ns, CAP_SYS_ADMIN))
206 idr_preload(GFP_KERNEL);
207 spin_lock_irq(&pidmap_lock);
210 nr = idr_alloc(&tmp->idr, NULL, tid,
211 tid + 1, GFP_ATOMIC);
213 * If ENOSPC is returned it means that the PID is
214 * alreay in use. Return EEXIST in that case.
221 * init really needs pid 1, but after reaching the
222 * maximum wrap back to RESERVED_PIDS
224 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
225 pid_min = RESERVED_PIDS;
228 * Store a null pointer so find_pid_ns does not find
229 * a partially initialized PID (see below).
231 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
232 pid_max, GFP_ATOMIC);
234 spin_unlock_irq(&pidmap_lock);
238 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
242 pid->numbers[i].nr = nr;
243 pid->numbers[i].ns = tmp;
248 refcount_set(&pid->count, 1);
249 for (type = 0; type < PIDTYPE_MAX; ++type)
250 INIT_HLIST_HEAD(&pid->tasks[type]);
252 init_waitqueue_head(&pid->wait_pidfd);
253 INIT_HLIST_HEAD(&pid->inodes);
255 upid = pid->numbers + ns->level;
256 spin_lock_irq(&pidmap_lock);
257 if (!(ns->pid_allocated & PIDNS_ADDING))
259 for ( ; upid >= pid->numbers; --upid) {
260 /* Make the PID visible to find_pid_ns. */
261 idr_replace(&upid->ns->idr, pid, upid->nr);
262 upid->ns->pid_allocated++;
264 spin_unlock_irq(&pidmap_lock);
269 spin_unlock_irq(&pidmap_lock);
273 spin_lock_irq(&pidmap_lock);
274 while (++i <= ns->level) {
275 upid = pid->numbers + i;
276 idr_remove(&upid->ns->idr, upid->nr);
279 /* On failure to allocate the first pid, reset the state */
280 if (ns->pid_allocated == PIDNS_ADDING)
281 idr_set_cursor(&ns->idr, 0);
283 spin_unlock_irq(&pidmap_lock);
285 kmem_cache_free(ns->pid_cachep, pid);
286 return ERR_PTR(retval);
289 void disable_pid_allocation(struct pid_namespace *ns)
291 spin_lock_irq(&pidmap_lock);
292 ns->pid_allocated &= ~PIDNS_ADDING;
293 spin_unlock_irq(&pidmap_lock);
296 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
298 return idr_find(&ns->idr, nr);
300 EXPORT_SYMBOL_GPL(find_pid_ns);
302 struct pid *find_vpid(int nr)
304 return find_pid_ns(nr, task_active_pid_ns(current));
306 EXPORT_SYMBOL_GPL(find_vpid);
308 static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
310 return (type == PIDTYPE_PID) ?
312 &task->signal->pids[type];
316 * attach_pid() must be called with the tasklist_lock write-held.
318 void attach_pid(struct task_struct *task, enum pid_type type)
320 struct pid *pid = *task_pid_ptr(task, type);
321 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
324 static void __change_pid(struct task_struct *task, enum pid_type type,
327 struct pid **pid_ptr = task_pid_ptr(task, type);
333 hlist_del_rcu(&task->pid_links[type]);
336 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
337 if (pid_has_task(pid, tmp))
343 void detach_pid(struct task_struct *task, enum pid_type type)
345 __change_pid(task, type, NULL);
348 void change_pid(struct task_struct *task, enum pid_type type,
351 __change_pid(task, type, pid);
352 attach_pid(task, type);
355 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
356 void transfer_pid(struct task_struct *old, struct task_struct *new,
359 if (type == PIDTYPE_PID)
360 new->thread_pid = old->thread_pid;
361 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
364 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
366 struct task_struct *result = NULL;
368 struct hlist_node *first;
369 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
370 lockdep_tasklist_lock_is_held());
372 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
376 EXPORT_SYMBOL(pid_task);
379 * Must be called under rcu_read_lock().
381 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
383 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
384 "find_task_by_pid_ns() needs rcu_read_lock() protection");
385 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
388 struct task_struct *find_task_by_vpid(pid_t vnr)
390 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
393 struct task_struct *find_get_task_by_vpid(pid_t nr)
395 struct task_struct *task;
398 task = find_task_by_vpid(nr);
400 get_task_struct(task);
406 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
410 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
414 EXPORT_SYMBOL_GPL(get_task_pid);
416 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
418 struct task_struct *result;
420 result = pid_task(pid, type);
422 get_task_struct(result);
426 EXPORT_SYMBOL_GPL(get_pid_task);
428 struct pid *find_get_pid(pid_t nr)
433 pid = get_pid(find_vpid(nr));
438 EXPORT_SYMBOL_GPL(find_get_pid);
440 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
445 if (pid && ns->level <= pid->level) {
446 upid = &pid->numbers[ns->level];
452 EXPORT_SYMBOL_GPL(pid_nr_ns);
454 pid_t pid_vnr(struct pid *pid)
456 return pid_nr_ns(pid, task_active_pid_ns(current));
458 EXPORT_SYMBOL_GPL(pid_vnr);
460 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
461 struct pid_namespace *ns)
467 ns = task_active_pid_ns(current);
468 if (likely(pid_alive(task)))
469 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
474 EXPORT_SYMBOL(__task_pid_nr_ns);
476 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
478 return ns_of_pid(task_pid(tsk));
480 EXPORT_SYMBOL_GPL(task_active_pid_ns);
483 * Used by proc to find the first pid that is greater than or equal to nr.
485 * If there is a pid at nr this function is exactly the same as find_pid_ns.
487 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
489 return idr_get_next(&ns->idr, &nr);
493 * pidfd_create() - Create a new pid file descriptor.
495 * @pid: struct pid that the pidfd will reference
497 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
499 * Note, that this function can only be called after the fd table has
500 * been unshared to avoid leaking the pidfd to the new process.
502 * Return: On success, a cloexec pidfd is returned.
503 * On error, a negative errno number will be returned.
505 static int pidfd_create(struct pid *pid)
509 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
518 * pidfd_open() - Open new pid file descriptor.
520 * @pid: pid for which to retrieve a pidfd
521 * @flags: flags to pass
523 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
524 * the process identified by @pid. Currently, the process identified by
525 * @pid must be a thread-group leader. This restriction currently exists
526 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
527 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
530 * Return: On success, a cloexec pidfd is returned.
531 * On error, a negative errno number will be returned.
533 SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
544 p = find_get_pid(pid);
548 if (pid_has_task(p, PIDTYPE_TGID))
549 fd = pidfd_create(p);
557 void __init pid_idr_init(void)
559 /* Verify no one has done anything silly: */
560 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
562 /* bump default and minimum pid_max based on number of cpus */
563 pid_max = min(pid_max_max, max_t(int, pid_max,
564 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
565 pid_max_min = max_t(int, pid_max_min,
566 PIDS_PER_CPU_MIN * num_possible_cpus());
567 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
569 idr_init(&init_pid_ns.idr);
571 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
572 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
575 static struct file *__pidfd_fget(struct task_struct *task, int fd)
580 ret = mutex_lock_killable(&task->signal->exec_update_mutex);
584 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
585 file = fget_task(task, fd);
587 file = ERR_PTR(-EPERM);
589 mutex_unlock(&task->signal->exec_update_mutex);
591 return file ?: ERR_PTR(-EBADF);
594 static int pidfd_getfd(struct pid *pid, int fd)
596 struct task_struct *task;
600 task = get_pid_task(pid, PIDTYPE_PID);
604 file = __pidfd_fget(task, fd);
605 put_task_struct(task);
607 return PTR_ERR(file);
609 ret = security_file_receive(file);
615 ret = get_unused_fd_flags(O_CLOEXEC);
619 fd_install(ret, file);
625 * sys_pidfd_getfd() - Get a file descriptor from another process
627 * @pidfd: the pidfd file descriptor of the process
628 * @fd: the file descriptor number to get
629 * @flags: flags on how to get the fd (reserved)
631 * This syscall gets a copy of a file descriptor from another process
632 * based on the pidfd, and file descriptor number. It requires that
633 * the calling process has the ability to ptrace the process represented
634 * by the pidfd. The process which is having its file descriptor copied
635 * is otherwise unaffected.
637 * Return: On success, a cloexec file descriptor is returned.
638 * On error, a negative errno number will be returned.
640 SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
647 /* flags is currently unused - make sure it's unset */
655 pid = pidfd_pid(f.file);
659 ret = pidfd_getfd(pid, fd);