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.count = REFCOUNT_INIT(2),
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 .ns.inum = PROC_PID_INIT_INO,
81#ifdef CONFIG_PID_NS
82 .ns.ops = &pidns_operations,
83#endif
84 .pid_max = PID_MAX_DEFAULT,
85#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
86 .memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC,
87#endif
88};
89EXPORT_SYMBOL_GPL(init_pid_ns);
90
91static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
92seqcount_spinlock_t pidmap_lock_seq = SEQCNT_SPINLOCK_ZERO(pidmap_lock_seq, &pidmap_lock);
93
94void put_pid(struct pid *pid)
95{
96 struct pid_namespace *ns;
97
98 if (!pid)
99 return;
100
101 ns = pid->numbers[pid->level].ns;
102 if (refcount_dec_and_test(r: &pid->count)) {
103 WARN_ON_ONCE(pid->stashed);
104 kmem_cache_free(s: ns->pid_cachep, objp: pid);
105 put_pid_ns(ns);
106 }
107}
108EXPORT_SYMBOL_GPL(put_pid);
109
110static void delayed_put_pid(struct rcu_head *rhp)
111{
112 struct pid *pid = container_of(rhp, struct pid, rcu);
113 put_pid(pid);
114}
115
116void free_pid(struct pid *pid)
117{
118 int i;
119
120 lockdep_assert_not_held(&tasklist_lock);
121
122 spin_lock(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(tsk: 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, id: upid->nr);
143 }
144 pidfs_remove_pid(pid);
145 spin_unlock(lock: &pidmap_lock);
146
147 call_rcu(head: &pid->rcu, func: 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(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(error: -EINVAL);
182
183 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
184 if (!pid)
185 return ERR_PTR(error: 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(ns: tmp->user_ns))
208 goto out_free;
209 set_tid_size--;
210 }
211
212 idr_preload(GFP_KERNEL);
213 spin_lock(lock: &pidmap_lock);
214
215 if (tid) {
216 nr = idr_alloc(&tmp->idr, NULL, start: tid,
217 end: 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(idr: &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, start: pid_min,
238 end: pid_max, GFP_ATOMIC);
239 }
240 spin_unlock(lock: &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(r: &pid->count, n: 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(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, id: upid->nr);
281 upid->ns->pid_allocated++;
282 }
283 spin_unlock(lock: &pidmap_lock);
284 idr_preload_end();
285
286 return pid;
287
288out_unlock:
289 spin_unlock(lock: &pidmap_lock);
290 idr_preload_end();
291 put_pid_ns(ns);
292
293out_free:
294 spin_lock(lock: &pidmap_lock);
295 while (++i <= ns->level) {
296 upid = pid->numbers + i;
297 idr_remove(&upid->ns->idr, id: upid->nr);
298 }
299
300 /* On failure to allocate the first pid, reset the state */
301 if (ns->pid_allocated == PIDNS_ADDING)
302 idr_set_cursor(idr: &ns->idr, val: 0);
303
304 spin_unlock(lock: &pidmap_lock);
305
306 kmem_cache_free(s: ns->pid_cachep, objp: pid);
307 return ERR_PTR(error: retval);
308}
309
310void disable_pid_allocation(struct pid_namespace *ns)
311{
312 spin_lock(lock: &pidmap_lock);
313 ns->pid_allocated &= ~PIDNS_ADDING;
314 spin_unlock(lock: &pidmap_lock);
315}
316
317struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
318{
319 return idr_find(&ns->idr, id: nr);
320}
321EXPORT_SYMBOL_GPL(find_pid_ns);
322
323struct pid *find_vpid(int nr)
324{
325 return find_pid_ns(nr, task_active_pid_ns(current));
326}
327EXPORT_SYMBOL_GPL(find_vpid);
328
329static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
330{
331 return (type == PIDTYPE_PID) ?
332 &task->thread_pid :
333 &task->signal->pids[type];
334}
335
336/*
337 * attach_pid() must be called with the tasklist_lock write-held.
338 */
339void attach_pid(struct task_struct *task, enum pid_type type)
340{
341 struct pid *pid;
342
343 lockdep_assert_held_write(&tasklist_lock);
344
345 pid = *task_pid_ptr(task, type);
346 hlist_add_head_rcu(n: &task->pid_links[type], h: &pid->tasks[type]);
347}
348
349static void __change_pid(struct pid **pids, struct task_struct *task,
350 enum pid_type type, struct pid *new)
351{
352 struct pid **pid_ptr, *pid;
353 int tmp;
354
355 lockdep_assert_held_write(&tasklist_lock);
356
357 pid_ptr = task_pid_ptr(task, type);
358 pid = *pid_ptr;
359
360 hlist_del_rcu(n: &task->pid_links[type]);
361 *pid_ptr = new;
362
363 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
364 if (pid_has_task(pid, type: tmp))
365 return;
366
367 WARN_ON(pids[type]);
368 pids[type] = pid;
369}
370
371void detach_pid(struct pid **pids, struct task_struct *task, enum pid_type type)
372{
373 __change_pid(pids, task, type, NULL);
374}
375
376void change_pid(struct pid **pids, struct task_struct *task, enum pid_type type,
377 struct pid *pid)
378{
379 __change_pid(pids, task, type, new: pid);
380 attach_pid(task, type);
381}
382
383void exchange_tids(struct task_struct *left, struct task_struct *right)
384{
385 struct pid *pid1 = left->thread_pid;
386 struct pid *pid2 = right->thread_pid;
387 struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
388 struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
389
390 lockdep_assert_held_write(&tasklist_lock);
391
392 /* Swap the single entry tid lists */
393 hlists_swap_heads_rcu(left: head1, right: head2);
394
395 /* Swap the per task_struct pid */
396 rcu_assign_pointer(left->thread_pid, pid2);
397 rcu_assign_pointer(right->thread_pid, pid1);
398
399 /* Swap the cached value */
400 WRITE_ONCE(left->pid, pid_nr(pid2));
401 WRITE_ONCE(right->pid, pid_nr(pid1));
402}
403
404/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
405void transfer_pid(struct task_struct *old, struct task_struct *new,
406 enum pid_type type)
407{
408 WARN_ON_ONCE(type == PIDTYPE_PID);
409 lockdep_assert_held_write(&tasklist_lock);
410 hlist_replace_rcu(old: &old->pid_links[type], new: &new->pid_links[type]);
411}
412
413struct task_struct *pid_task(struct pid *pid, enum pid_type type)
414{
415 struct task_struct *result = NULL;
416 if (pid) {
417 struct hlist_node *first;
418 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
419 lockdep_tasklist_lock_is_held());
420 if (first)
421 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
422 }
423 return result;
424}
425EXPORT_SYMBOL(pid_task);
426
427/*
428 * Must be called under rcu_read_lock().
429 */
430struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
431{
432 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
433 "find_task_by_pid_ns() needs rcu_read_lock() protection");
434 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
435}
436
437struct task_struct *find_task_by_vpid(pid_t vnr)
438{
439 return find_task_by_pid_ns(nr: vnr, ns: task_active_pid_ns(current));
440}
441
442struct task_struct *find_get_task_by_vpid(pid_t nr)
443{
444 struct task_struct *task;
445
446 rcu_read_lock();
447 task = find_task_by_vpid(vnr: nr);
448 if (task)
449 get_task_struct(t: task);
450 rcu_read_unlock();
451
452 return task;
453}
454
455struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
456{
457 struct pid *pid;
458 rcu_read_lock();
459 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
460 rcu_read_unlock();
461 return pid;
462}
463EXPORT_SYMBOL_GPL(get_task_pid);
464
465struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
466{
467 struct task_struct *result;
468 rcu_read_lock();
469 result = pid_task(pid, type);
470 if (result)
471 get_task_struct(t: result);
472 rcu_read_unlock();
473 return result;
474}
475EXPORT_SYMBOL_GPL(get_pid_task);
476
477struct pid *find_get_pid(pid_t nr)
478{
479 struct pid *pid;
480
481 rcu_read_lock();
482 pid = get_pid(pid: find_vpid(nr));
483 rcu_read_unlock();
484
485 return pid;
486}
487EXPORT_SYMBOL_GPL(find_get_pid);
488
489pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
490{
491 struct upid *upid;
492 pid_t nr = 0;
493
494 if (pid && ns->level <= pid->level) {
495 upid = &pid->numbers[ns->level];
496 if (upid->ns == ns)
497 nr = upid->nr;
498 }
499 return nr;
500}
501EXPORT_SYMBOL_GPL(pid_nr_ns);
502
503pid_t pid_vnr(struct pid *pid)
504{
505 return pid_nr_ns(pid, task_active_pid_ns(current));
506}
507EXPORT_SYMBOL_GPL(pid_vnr);
508
509pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
510 struct pid_namespace *ns)
511{
512 pid_t nr = 0;
513
514 rcu_read_lock();
515 if (!ns)
516 ns = task_active_pid_ns(current);
517 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
518 rcu_read_unlock();
519
520 return nr;
521}
522EXPORT_SYMBOL(__task_pid_nr_ns);
523
524struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
525{
526 return ns_of_pid(pid: task_pid(task: tsk));
527}
528EXPORT_SYMBOL_GPL(task_active_pid_ns);
529
530/*
531 * Used by proc to find the first pid that is greater than or equal to nr.
532 *
533 * If there is a pid at nr this function is exactly the same as find_pid_ns.
534 */
535struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
536{
537 return idr_get_next(&ns->idr, nextid: &nr);
538}
539EXPORT_SYMBOL_GPL(find_ge_pid);
540
541struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
542{
543 CLASS(fd, f)(fd);
544 struct pid *pid;
545
546 if (fd_empty(f))
547 return ERR_PTR(error: -EBADF);
548
549 pid = pidfd_pid(fd_file(f));
550 if (!IS_ERR(ptr: pid)) {
551 get_pid(pid);
552 *flags = fd_file(f)->f_flags;
553 }
554 return pid;
555}
556
557/**
558 * pidfd_get_task() - Get the task associated with a pidfd
559 *
560 * @pidfd: pidfd for which to get the task
561 * @flags: flags associated with this pidfd
562 *
563 * Return the task associated with @pidfd. The function takes a reference on
564 * the returned task. The caller is responsible for releasing that reference.
565 *
566 * Return: On success, the task_struct associated with the pidfd.
567 * On error, a negative errno number will be returned.
568 */
569struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
570{
571 unsigned int f_flags = 0;
572 struct pid *pid;
573 struct task_struct *task;
574 enum pid_type type;
575
576 switch (pidfd) {
577 case PIDFD_SELF_THREAD:
578 type = PIDTYPE_PID;
579 pid = get_task_pid(current, type);
580 break;
581 case PIDFD_SELF_THREAD_GROUP:
582 type = PIDTYPE_TGID;
583 pid = get_task_pid(current, type);
584 break;
585 default:
586 pid = pidfd_get_pid(fd: pidfd, flags: &f_flags);
587 if (IS_ERR(ptr: pid))
588 return ERR_CAST(ptr: pid);
589 type = PIDTYPE_TGID;
590 break;
591 }
592
593 task = get_pid_task(pid, type);
594 put_pid(pid);
595 if (!task)
596 return ERR_PTR(error: -ESRCH);
597
598 *flags = f_flags;
599 return task;
600}
601
602/**
603 * pidfd_create() - Create a new pid file descriptor.
604 *
605 * @pid: struct pid that the pidfd will reference
606 * @flags: flags to pass
607 *
608 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
609 *
610 * Note, that this function can only be called after the fd table has
611 * been unshared to avoid leaking the pidfd to the new process.
612 *
613 * This symbol should not be explicitly exported to loadable modules.
614 *
615 * Return: On success, a cloexec pidfd is returned.
616 * On error, a negative errno number will be returned.
617 */
618static int pidfd_create(struct pid *pid, unsigned int flags)
619{
620 int pidfd;
621 struct file *pidfd_file;
622
623 pidfd = pidfd_prepare(pid, flags, ret_file: &pidfd_file);
624 if (pidfd < 0)
625 return pidfd;
626
627 fd_install(fd: pidfd, file: pidfd_file);
628 return pidfd;
629}
630
631/**
632 * sys_pidfd_open() - Open new pid file descriptor.
633 *
634 * @pid: pid for which to retrieve a pidfd
635 * @flags: flags to pass
636 *
637 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
638 * the task identified by @pid. Without PIDFD_THREAD flag the target task
639 * must be a thread-group leader.
640 *
641 * Return: On success, a cloexec pidfd is returned.
642 * On error, a negative errno number will be returned.
643 */
644SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
645{
646 int fd;
647 struct pid *p;
648
649 if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD))
650 return -EINVAL;
651
652 if (pid <= 0)
653 return -EINVAL;
654
655 p = find_get_pid(pid);
656 if (!p)
657 return -ESRCH;
658
659 fd = pidfd_create(pid: p, flags);
660
661 put_pid(p);
662 return fd;
663}
664
665#ifdef CONFIG_SYSCTL
666static struct ctl_table_set *pid_table_root_lookup(struct ctl_table_root *root)
667{
668 return &task_active_pid_ns(current)->set;
669}
670
671static int set_is_seen(struct ctl_table_set *set)
672{
673 return &task_active_pid_ns(current)->set == set;
674}
675
676static int pid_table_root_permissions(struct ctl_table_header *head,
677 const struct ctl_table *table)
678{
679 struct pid_namespace *pidns =
680 container_of(head->set, struct pid_namespace, set);
681 int mode = table->mode;
682
683 if (ns_capable(ns: pidns->user_ns, CAP_SYS_ADMIN) ||
684 uid_eq(current_euid(), right: make_kuid(from: pidns->user_ns, uid: 0)))
685 mode = (mode & S_IRWXU) >> 6;
686 else if (in_egroup_p(make_kgid(from: pidns->user_ns, gid: 0)))
687 mode = (mode & S_IRWXG) >> 3;
688 else
689 mode = mode & S_IROTH;
690 return (mode << 6) | (mode << 3) | mode;
691}
692
693static void pid_table_root_set_ownership(struct ctl_table_header *head,
694 kuid_t *uid, kgid_t *gid)
695{
696 struct pid_namespace *pidns =
697 container_of(head->set, struct pid_namespace, set);
698 kuid_t ns_root_uid;
699 kgid_t ns_root_gid;
700
701 ns_root_uid = make_kuid(from: pidns->user_ns, uid: 0);
702 if (uid_valid(uid: ns_root_uid))
703 *uid = ns_root_uid;
704
705 ns_root_gid = make_kgid(from: pidns->user_ns, gid: 0);
706 if (gid_valid(gid: ns_root_gid))
707 *gid = ns_root_gid;
708}
709
710static struct ctl_table_root pid_table_root = {
711 .lookup = pid_table_root_lookup,
712 .permissions = pid_table_root_permissions,
713 .set_ownership = pid_table_root_set_ownership,
714};
715
716static const struct ctl_table pid_table[] = {
717 {
718 .procname = "pid_max",
719 .data = &init_pid_ns.pid_max,
720 .maxlen = sizeof(int),
721 .mode = 0644,
722 .proc_handler = proc_dointvec_minmax,
723 .extra1 = &pid_max_min,
724 .extra2 = &pid_max_max,
725 },
726};
727#endif
728
729int register_pidns_sysctls(struct pid_namespace *pidns)
730{
731#ifdef CONFIG_SYSCTL
732 struct ctl_table *tbl;
733
734 setup_sysctl_set(p: &pidns->set, root: &pid_table_root, is_seen: set_is_seen);
735
736 tbl = kmemdup(pid_table, sizeof(pid_table), GFP_KERNEL);
737 if (!tbl)
738 return -ENOMEM;
739 tbl->data = &pidns->pid_max;
740 pidns->pid_max = min(pid_max_max, max_t(int, pidns->pid_max,
741 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
742
743 pidns->sysctls = __register_sysctl_table(set: &pidns->set, path: "kernel", table: tbl,
744 ARRAY_SIZE(pid_table));
745 if (!pidns->sysctls) {
746 kfree(objp: tbl);
747 retire_sysctl_set(set: &pidns->set);
748 return -ENOMEM;
749 }
750#endif
751 return 0;
752}
753
754void unregister_pidns_sysctls(struct pid_namespace *pidns)
755{
756#ifdef CONFIG_SYSCTL
757 const struct ctl_table *tbl;
758
759 tbl = pidns->sysctls->ctl_table_arg;
760 unregister_sysctl_table(table: pidns->sysctls);
761 retire_sysctl_set(set: &pidns->set);
762 kfree(objp: tbl);
763#endif
764}
765
766void __init pid_idr_init(void)
767{
768 /* Verify no one has done anything silly: */
769 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
770
771 /* bump default and minimum pid_max based on number of cpus */
772 init_pid_ns.pid_max = min(pid_max_max, max_t(int, init_pid_ns.pid_max,
773 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
774 pid_max_min = max_t(int, pid_max_min,
775 PIDS_PER_CPU_MIN * num_possible_cpus());
776 pr_info("pid_max: default: %u minimum: %u\n", init_pid_ns.pid_max, pid_max_min);
777
778 idr_init(idr: &init_pid_ns.idr);
779
780 init_pid_ns.pid_cachep = kmem_cache_create("pid",
781 struct_size_t(struct pid, numbers, 1),
782 __alignof__(struct pid),
783 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
784 NULL);
785}
786
787static __init int pid_namespace_sysctl_init(void)
788{
789#ifdef CONFIG_SYSCTL
790 /* "kernel" directory will have already been initialized. */
791 BUG_ON(register_pidns_sysctls(&init_pid_ns));
792#endif
793 return 0;
794}
795subsys_initcall(pid_namespace_sysctl_init);
796
797static struct file *__pidfd_fget(struct task_struct *task, int fd)
798{
799 struct file *file;
800 int ret;
801
802 ret = down_read_killable(sem: &task->signal->exec_update_lock);
803 if (ret)
804 return ERR_PTR(error: ret);
805
806 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
807 file = fget_task(task, fd);
808 else
809 file = ERR_PTR(error: -EPERM);
810
811 up_read(sem: &task->signal->exec_update_lock);
812
813 if (!file) {
814 /*
815 * It is possible that the target thread is exiting; it can be
816 * either:
817 * 1. before exit_signals(), which gives a real fd
818 * 2. before exit_files() takes the task_lock() gives a real fd
819 * 3. after exit_files() releases task_lock(), ->files is NULL;
820 * this has PF_EXITING, since it was set in exit_signals(),
821 * __pidfd_fget() returns EBADF.
822 * In case 3 we get EBADF, but that really means ESRCH, since
823 * the task is currently exiting and has freed its files
824 * struct, so we fix it up.
825 */
826 if (task->flags & PF_EXITING)
827 file = ERR_PTR(error: -ESRCH);
828 else
829 file = ERR_PTR(error: -EBADF);
830 }
831
832 return file;
833}
834
835static int pidfd_getfd(struct pid *pid, int fd)
836{
837 struct task_struct *task;
838 struct file *file;
839 int ret;
840
841 task = get_pid_task(pid, PIDTYPE_PID);
842 if (!task)
843 return -ESRCH;
844
845 file = __pidfd_fget(task, fd);
846 put_task_struct(t: task);
847 if (IS_ERR(ptr: file))
848 return PTR_ERR(ptr: file);
849
850 ret = receive_fd(file, NULL, O_CLOEXEC);
851 fput(file);
852
853 return ret;
854}
855
856/**
857 * sys_pidfd_getfd() - Get a file descriptor from another process
858 *
859 * @pidfd: the pidfd file descriptor of the process
860 * @fd: the file descriptor number to get
861 * @flags: flags on how to get the fd (reserved)
862 *
863 * This syscall gets a copy of a file descriptor from another process
864 * based on the pidfd, and file descriptor number. It requires that
865 * the calling process has the ability to ptrace the process represented
866 * by the pidfd. The process which is having its file descriptor copied
867 * is otherwise unaffected.
868 *
869 * Return: On success, a cloexec file descriptor is returned.
870 * On error, a negative errno number will be returned.
871 */
872SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
873 unsigned int, flags)
874{
875 struct pid *pid;
876
877 /* flags is currently unused - make sure it's unset */
878 if (flags)
879 return -EINVAL;
880
881 CLASS(fd, f)(fd: pidfd);
882 if (fd_empty(f))
883 return -EBADF;
884
885 pid = pidfd_pid(fd_file(f));
886 if (IS_ERR(ptr: pid))
887 return PTR_ERR(ptr: pid);
888
889 return pidfd_getfd(pid, fd);
890}
891

source code of linux/kernel/pid.c