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)
165 struct pid_namespace *tmp;
167 int retval = -ENOMEM;
169 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
171 return ERR_PTR(retval);
174 pid->level = ns->level;
176 for (i = ns->level; i >= 0; i--) {
179 idr_preload(GFP_KERNEL);
180 spin_lock_irq(&pidmap_lock);
183 * init really needs pid 1, but after reaching the maximum
184 * wrap back to RESERVED_PIDS
186 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
187 pid_min = RESERVED_PIDS;
190 * Store a null pointer so find_pid_ns does not find
191 * a partially initialized PID (see below).
193 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
194 pid_max, GFP_ATOMIC);
195 spin_unlock_irq(&pidmap_lock);
199 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
203 pid->numbers[i].nr = nr;
204 pid->numbers[i].ns = tmp;
208 if (unlikely(is_child_reaper(pid))) {
209 if (pid_ns_prepare_proc(ns))
214 refcount_set(&pid->count, 1);
215 for (type = 0; type < PIDTYPE_MAX; ++type)
216 INIT_HLIST_HEAD(&pid->tasks[type]);
218 init_waitqueue_head(&pid->wait_pidfd);
220 upid = pid->numbers + ns->level;
221 spin_lock_irq(&pidmap_lock);
222 if (!(ns->pid_allocated & PIDNS_ADDING))
224 for ( ; upid >= pid->numbers; --upid) {
225 /* Make the PID visible to find_pid_ns. */
226 idr_replace(&upid->ns->idr, pid, upid->nr);
227 upid->ns->pid_allocated++;
229 spin_unlock_irq(&pidmap_lock);
234 spin_unlock_irq(&pidmap_lock);
238 spin_lock_irq(&pidmap_lock);
239 while (++i <= ns->level) {
240 upid = pid->numbers + i;
241 idr_remove(&upid->ns->idr, upid->nr);
244 /* On failure to allocate the first pid, reset the state */
245 if (ns->pid_allocated == PIDNS_ADDING)
246 idr_set_cursor(&ns->idr, 0);
248 spin_unlock_irq(&pidmap_lock);
250 kmem_cache_free(ns->pid_cachep, pid);
251 return ERR_PTR(retval);
254 void disable_pid_allocation(struct pid_namespace *ns)
256 spin_lock_irq(&pidmap_lock);
257 ns->pid_allocated &= ~PIDNS_ADDING;
258 spin_unlock_irq(&pidmap_lock);
261 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
263 return idr_find(&ns->idr, nr);
265 EXPORT_SYMBOL_GPL(find_pid_ns);
267 struct pid *find_vpid(int nr)
269 return find_pid_ns(nr, task_active_pid_ns(current));
271 EXPORT_SYMBOL_GPL(find_vpid);
273 static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
275 return (type == PIDTYPE_PID) ?
277 &task->signal->pids[type];
281 * attach_pid() must be called with the tasklist_lock write-held.
283 void attach_pid(struct task_struct *task, enum pid_type type)
285 struct pid *pid = *task_pid_ptr(task, type);
286 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
289 static void __change_pid(struct task_struct *task, enum pid_type type,
292 struct pid **pid_ptr = task_pid_ptr(task, type);
298 hlist_del_rcu(&task->pid_links[type]);
301 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
302 if (!hlist_empty(&pid->tasks[tmp]))
308 void detach_pid(struct task_struct *task, enum pid_type type)
310 __change_pid(task, type, NULL);
313 void change_pid(struct task_struct *task, enum pid_type type,
316 __change_pid(task, type, pid);
317 attach_pid(task, type);
320 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
321 void transfer_pid(struct task_struct *old, struct task_struct *new,
324 if (type == PIDTYPE_PID)
325 new->thread_pid = old->thread_pid;
326 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
329 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
331 struct task_struct *result = NULL;
333 struct hlist_node *first;
334 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
335 lockdep_tasklist_lock_is_held());
337 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
341 EXPORT_SYMBOL(pid_task);
344 * Must be called under rcu_read_lock().
346 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
348 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
349 "find_task_by_pid_ns() needs rcu_read_lock() protection");
350 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
353 struct task_struct *find_task_by_vpid(pid_t vnr)
355 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
358 struct task_struct *find_get_task_by_vpid(pid_t nr)
360 struct task_struct *task;
363 task = find_task_by_vpid(nr);
365 get_task_struct(task);
371 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
375 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
379 EXPORT_SYMBOL_GPL(get_task_pid);
381 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
383 struct task_struct *result;
385 result = pid_task(pid, type);
387 get_task_struct(result);
391 EXPORT_SYMBOL_GPL(get_pid_task);
393 struct pid *find_get_pid(pid_t nr)
398 pid = get_pid(find_vpid(nr));
403 EXPORT_SYMBOL_GPL(find_get_pid);
405 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
410 if (pid && ns->level <= pid->level) {
411 upid = &pid->numbers[ns->level];
417 EXPORT_SYMBOL_GPL(pid_nr_ns);
419 pid_t pid_vnr(struct pid *pid)
421 return pid_nr_ns(pid, task_active_pid_ns(current));
423 EXPORT_SYMBOL_GPL(pid_vnr);
425 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
426 struct pid_namespace *ns)
432 ns = task_active_pid_ns(current);
433 if (likely(pid_alive(task)))
434 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
439 EXPORT_SYMBOL(__task_pid_nr_ns);
441 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
443 return ns_of_pid(task_pid(tsk));
445 EXPORT_SYMBOL_GPL(task_active_pid_ns);
448 * Used by proc to find the first pid that is greater than or equal to nr.
450 * If there is a pid at nr this function is exactly the same as find_pid_ns.
452 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
454 return idr_get_next(&ns->idr, &nr);
458 * pidfd_create() - Create a new pid file descriptor.
460 * @pid: struct pid that the pidfd will reference
462 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
464 * Note, that this function can only be called after the fd table has
465 * been unshared to avoid leaking the pidfd to the new process.
467 * Return: On success, a cloexec pidfd is returned.
468 * On error, a negative errno number will be returned.
470 static int pidfd_create(struct pid *pid)
474 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
483 * pidfd_open() - Open new pid file descriptor.
485 * @pid: pid for which to retrieve a pidfd
486 * @flags: flags to pass
488 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
489 * the process identified by @pid. Currently, the process identified by
490 * @pid must be a thread-group leader. This restriction currently exists
491 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
492 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
495 * Return: On success, a cloexec pidfd is returned.
496 * On error, a negative errno number will be returned.
498 SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
509 p = find_get_pid(pid);
515 if (!pid_task(p, PIDTYPE_TGID))
519 fd = ret ?: pidfd_create(p);
524 void __init pid_idr_init(void)
526 /* Verify no one has done anything silly: */
527 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
529 /* bump default and minimum pid_max based on number of cpus */
530 pid_max = min(pid_max_max, max_t(int, pid_max,
531 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
532 pid_max_min = max_t(int, pid_max_min,
533 PIDS_PER_CPU_MIN * num_possible_cpus());
534 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
536 idr_init(&init_pid_ns.idr);
538 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
539 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);