projects/linux·blob v6.14 · refs

872 lines · 21.5 KiB

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 <[email protected]>, OpenVZ, SWsoft Inc.
25 * (C) 2007 Sukadev Bhattiprolu <[email protected]>, 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
91/*
92 * Note: disable interrupts while the pidmap_lock is held as an
93 * interrupt might come in and do read_lock(&tasklist_lock).
94 *
95 * If we don't disable interrupts there is a nasty deadlock between
96 * detach_pid()->free_pid() and another cpu that does
97 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
98 * read_lock(&tasklist_lock);
99 *
100 * After we clean up the tasklist_lock and know there are no
101 * irq handlers that take it we can leave the interrupts enabled.
102 * For now it is easier to be safe than to prove it can't happen.
103 */
104
105static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
106seqcount_spinlock_t pidmap_lock_seq = SEQCNT_SPINLOCK_ZERO(pidmap_lock_seq, &pidmap_lock);
107
108void put_pid(struct pid *pid)
109{
110 struct pid_namespace *ns;
111
112 if (!pid)
113 return;
114
115 ns = pid->numbers[pid->level].ns;
116 if (refcount_dec_and_test(&pid->count)) {
117 kmem_cache_free(ns->pid_cachep, pid);
118 put_pid_ns(ns);
119 }
120}
121EXPORT_SYMBOL_GPL(put_pid);
122
123static void delayed_put_pid(struct rcu_head *rhp)
124{
125 struct pid *pid = container_of(rhp, struct pid, rcu);
126 put_pid(pid);
127}
128
129void free_pid(struct pid *pid)
130{
131 /* We can be called with write_lock_irq(&tasklist_lock) held */
132 int i;
133 unsigned long flags;
134
135 spin_lock_irqsave(&pidmap_lock, flags);
136 for (i = 0; i <= pid->level; i++) {
137 struct upid *upid = pid->numbers + i;
138 struct pid_namespace *ns = upid->ns;
139 switch (--ns->pid_allocated) {
140 case 2:
141 case 1:
142 /* When all that is left in the pid namespace
143 * is the reaper wake up the reaper. The reaper
144 * may be sleeping in zap_pid_ns_processes().
145 */
146 wake_up_process(ns->child_reaper);
147 break;
148 case PIDNS_ADDING:
149 /* Handle a fork failure of the first process */
150 WARN_ON(ns->child_reaper);
151 ns->pid_allocated = 0;
152 break;
153 }
154
155 idr_remove(&ns->idr, upid->nr);
156 }
157 pidfs_remove_pid(pid);
158 spin_unlock_irqrestore(&pidmap_lock, flags);
159
160 call_rcu(&pid->rcu, delayed_put_pid);
161}
162
163struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
164 size_t set_tid_size)
165{
166 struct pid *pid;
167 enum pid_type type;
168 int i, nr;
169 struct pid_namespace *tmp;
170 struct upid *upid;
171 int retval = -ENOMEM;
172
173 /*
174 * set_tid_size contains the size of the set_tid array. Starting at
175 * the most nested currently active PID namespace it tells alloc_pid()
176 * which PID to set for a process in that most nested PID namespace
177 * up to set_tid_size PID namespaces. It does not have to set the PID
178 * for a process in all nested PID namespaces but set_tid_size must
179 * never be greater than the current ns->level + 1.
180 */
181 if (set_tid_size > ns->level + 1)
182 return ERR_PTR(-EINVAL);
183
184 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
185 if (!pid)
186 return ERR_PTR(retval);
187
188 tmp = ns;
189 pid->level = ns->level;
190
191 for (i = ns->level; i >= 0; i--) {
192 int tid = 0;
193 int pid_max = READ_ONCE(tmp->pid_max);
194
195 if (set_tid_size) {
196 tid = set_tid[ns->level - i];
197
198 retval = -EINVAL;
199 if (tid < 1 || tid >= pid_max)
200 goto out_free;
201 /*
202 * Also fail if a PID != 1 is requested and
203 * no PID 1 exists.
204 */
205 if (tid != 1 && !tmp->child_reaper)
206 goto out_free;
207 retval = -EPERM;
208 if (!checkpoint_restore_ns_capable(tmp->user_ns))
209 goto out_free;
210 set_tid_size--;
211 }
212
213 idr_preload(GFP_KERNEL);
214 spin_lock_irq(&pidmap_lock);
215
216 if (tid) {
217 nr = idr_alloc(&tmp->idr, NULL, tid,
218 tid + 1, GFP_ATOMIC);
219 /*
220 * If ENOSPC is returned it means that the PID is
221 * alreay in use. Return EEXIST in that case.
222 */
223 if (nr == -ENOSPC)
224 nr = -EEXIST;
225 } else {
226 int pid_min = 1;
227 /*
228 * init really needs pid 1, but after reaching the
229 * maximum wrap back to RESERVED_PIDS
230 */
231 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
232 pid_min = RESERVED_PIDS;
233
234 /*
235 * Store a null pointer so find_pid_ns does not find
236 * a partially initialized PID (see below).
237 */
238 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
239 pid_max, GFP_ATOMIC);
240 }
241 spin_unlock_irq(&pidmap_lock);
242 idr_preload_end();
243
244 if (nr < 0) {
245 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
246 goto out_free;
247 }
248
249 pid->numbers[i].nr = nr;
250 pid->numbers[i].ns = tmp;
251 tmp = tmp->parent;
252 }
253
254 /*
255 * ENOMEM is not the most obvious choice especially for the case
256 * where the child subreaper has already exited and the pid
257 * namespace denies the creation of any new processes. But ENOMEM
258 * is what we have exposed to userspace for a long time and it is
259 * documented behavior for pid namespaces. So we can't easily
260 * change it even if there were an error code better suited.
261 */
262 retval = -ENOMEM;
263
264 get_pid_ns(ns);
265 refcount_set(&pid->count, 1);
266 spin_lock_init(&pid->lock);
267 for (type = 0; type < PIDTYPE_MAX; ++type)
268 INIT_HLIST_HEAD(&pid->tasks[type]);
269
270 init_waitqueue_head(&pid->wait_pidfd);
271 INIT_HLIST_HEAD(&pid->inodes);
272
273 upid = pid->numbers + ns->level;
274 idr_preload(GFP_KERNEL);
275 spin_lock_irq(&pidmap_lock);
276 if (!(ns->pid_allocated & PIDNS_ADDING))
277 goto out_unlock;
278 pidfs_add_pid(pid);
279 for ( ; upid >= pid->numbers; --upid) {
280 /* Make the PID visible to find_pid_ns. */
281 idr_replace(&upid->ns->idr, pid, upid->nr);
282 upid->ns->pid_allocated++;
283 }
284 spin_unlock_irq(&pidmap_lock);
285 idr_preload_end();
286
287 return pid;
288
289out_unlock:
290 spin_unlock_irq(&pidmap_lock);
291 idr_preload_end();
292 put_pid_ns(ns);
293
294out_free:
295 spin_lock_irq(&pidmap_lock);
296 while (++i <= ns->level) {
297 upid = pid->numbers + i;
298 idr_remove(&upid->ns->idr, upid->nr);
299 }
300
301 /* On failure to allocate the first pid, reset the state */
302 if (ns->pid_allocated == PIDNS_ADDING)
303 idr_set_cursor(&ns->idr, 0);
304
305 spin_unlock_irq(&pidmap_lock);
306
307 kmem_cache_free(ns->pid_cachep, pid);
308 return ERR_PTR(retval);
309}
310
311void disable_pid_allocation(struct pid_namespace *ns)
312{
313 spin_lock_irq(&pidmap_lock);
314 ns->pid_allocated &= ~PIDNS_ADDING;
315 spin_unlock_irq(&pidmap_lock);
316}
317
318struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
319{
320 return idr_find(&ns->idr, nr);
321}
322EXPORT_SYMBOL_GPL(find_pid_ns);
323
324struct pid *find_vpid(int nr)
325{
326 return find_pid_ns(nr, task_active_pid_ns(current));
327}
328EXPORT_SYMBOL_GPL(find_vpid);
329
330static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
331{
332 return (type == PIDTYPE_PID) ?
333 &task->thread_pid :
334 &task->signal->pids[type];
335}
336
337/*
338 * attach_pid() must be called with the tasklist_lock write-held.
339 */
340void attach_pid(struct task_struct *task, enum pid_type type)
341{
342 struct pid *pid = *task_pid_ptr(task, type);
343 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
344}
345
346static void __change_pid(struct task_struct *task, enum pid_type type,
347 struct pid *new)
348{
349 struct pid **pid_ptr = task_pid_ptr(task, type);
350 struct pid *pid;
351 int tmp;
352
353 pid = *pid_ptr;
354
355 hlist_del_rcu(&task->pid_links[type]);
356 *pid_ptr = new;
357
358 if (type == PIDTYPE_PID) {
359 WARN_ON_ONCE(pid_has_task(pid, PIDTYPE_PID));
360 wake_up_all(&pid->wait_pidfd);
361 }
362
363 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
364 if (pid_has_task(pid, tmp))
365 return;
366
367 free_pid(pid);
368}
369
370void detach_pid(struct task_struct *task, enum pid_type type)
371{
372 __change_pid(task, type, NULL);
373}
374
375void change_pid(struct task_struct *task, enum pid_type type,
376 struct pid *pid)
377{
378 __change_pid(task, type, pid);
379 attach_pid(task, type);
380}
381
382void exchange_tids(struct task_struct *left, struct task_struct *right)
383{
384 struct pid *pid1 = left->thread_pid;
385 struct pid *pid2 = right->thread_pid;
386 struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
387 struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
388
389 /* Swap the single entry tid lists */
390 hlists_swap_heads_rcu(head1, head2);
391
392 /* Swap the per task_struct pid */
393 rcu_assign_pointer(left->thread_pid, pid2);
394 rcu_assign_pointer(right->thread_pid, pid1);
395
396 /* Swap the cached value */
397 WRITE_ONCE(left->pid, pid_nr(pid2));
398 WRITE_ONCE(right->pid, pid_nr(pid1));
399}
400
401/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
402void transfer_pid(struct task_struct *old, struct task_struct *new,
403 enum pid_type type)
404{
405 WARN_ON_ONCE(type == PIDTYPE_PID);
406 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
407}
408
409struct task_struct *pid_task(struct pid *pid, enum pid_type type)
410{
411 struct task_struct *result = NULL;
412 if (pid) {
413 struct hlist_node *first;
414 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
415 lockdep_tasklist_lock_is_held());
416 if (first)
417 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
418 }
419 return result;
420}
421EXPORT_SYMBOL(pid_task);
422
423/*
424 * Must be called under rcu_read_lock().
425 */
426struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
427{
428 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
429 "find_task_by_pid_ns() needs rcu_read_lock() protection");
430 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
431}
432
433struct task_struct *find_task_by_vpid(pid_t vnr)
434{
435 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
436}
437
438struct task_struct *find_get_task_by_vpid(pid_t nr)
439{
440 struct task_struct *task;
441
442 rcu_read_lock();
443 task = find_task_by_vpid(nr);
444 if (task)
445 get_task_struct(task);
446 rcu_read_unlock();
447
448 return task;
449}
450
451struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
452{
453 struct pid *pid;
454 rcu_read_lock();
455 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
456 rcu_read_unlock();
457 return pid;
458}
459EXPORT_SYMBOL_GPL(get_task_pid);
460
461struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
462{
463 struct task_struct *result;
464 rcu_read_lock();
465 result = pid_task(pid, type);
466 if (result)
467 get_task_struct(result);
468 rcu_read_unlock();
469 return result;
470}
471EXPORT_SYMBOL_GPL(get_pid_task);
472
473struct pid *find_get_pid(pid_t nr)
474{
475 struct pid *pid;
476
477 rcu_read_lock();
478 pid = get_pid(find_vpid(nr));
479 rcu_read_unlock();
480
481 return pid;
482}
483EXPORT_SYMBOL_GPL(find_get_pid);
484
485pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
486{
487 struct upid *upid;
488 pid_t nr = 0;
489
490 if (pid && ns->level <= pid->level) {
491 upid = &pid->numbers[ns->level];
492 if (upid->ns == ns)
493 nr = upid->nr;
494 }
495 return nr;
496}
497EXPORT_SYMBOL_GPL(pid_nr_ns);
498
499pid_t pid_vnr(struct pid *pid)
500{
501 return pid_nr_ns(pid, task_active_pid_ns(current));
502}
503EXPORT_SYMBOL_GPL(pid_vnr);
504
505pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
506 struct pid_namespace *ns)
507{
508 pid_t nr = 0;
509
510 rcu_read_lock();
511 if (!ns)
512 ns = task_active_pid_ns(current);
513 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
514 rcu_read_unlock();
515
516 return nr;
517}
518EXPORT_SYMBOL(__task_pid_nr_ns);
519
520struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
521{
522 return ns_of_pid(task_pid(tsk));
523}
524EXPORT_SYMBOL_GPL(task_active_pid_ns);
525
526/*
527 * Used by proc to find the first pid that is greater than or equal to nr.
528 *
529 * If there is a pid at nr this function is exactly the same as find_pid_ns.
530 */
531struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
532{
533 return idr_get_next(&ns->idr, &nr);
534}
535EXPORT_SYMBOL_GPL(find_ge_pid);
536
537struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
538{
539 CLASS(fd, f)(fd);
540 struct pid *pid;
541
542 if (fd_empty(f))
543 return ERR_PTR(-EBADF);
544
545 pid = pidfd_pid(fd_file(f));
546 if (!IS_ERR(pid)) {
547 get_pid(pid);
548 *flags = fd_file(f)->f_flags;
549 }
550 return pid;
551}
552
553/**
554 * pidfd_get_task() - Get the task associated with a pidfd
555 *
556 * @pidfd: pidfd for which to get the task
557 * @flags: flags associated with this pidfd
558 *
559 * Return the task associated with @pidfd. The function takes a reference on
560 * the returned task. The caller is responsible for releasing that reference.
561 *
562 * Return: On success, the task_struct associated with the pidfd.
563 * On error, a negative errno number will be returned.
564 */
565struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
566{
567 unsigned int f_flags;
568 struct pid *pid;
569 struct task_struct *task;
570
571 pid = pidfd_get_pid(pidfd, &f_flags);
572 if (IS_ERR(pid))
573 return ERR_CAST(pid);
574
575 task = get_pid_task(pid, PIDTYPE_TGID);
576 put_pid(pid);
577 if (!task)
578 return ERR_PTR(-ESRCH);
579
580 *flags = f_flags;
581 return task;
582}
583
584/**
585 * pidfd_create() - Create a new pid file descriptor.
586 *
587 * @pid: struct pid that the pidfd will reference
588 * @flags: flags to pass
589 *
590 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
591 *
592 * Note, that this function can only be called after the fd table has
593 * been unshared to avoid leaking the pidfd to the new process.
594 *
595 * This symbol should not be explicitly exported to loadable modules.
596 *
597 * Return: On success, a cloexec pidfd is returned.
598 * On error, a negative errno number will be returned.
599 */
600static int pidfd_create(struct pid *pid, unsigned int flags)
601{
602 int pidfd;
603 struct file *pidfd_file;
604
605 pidfd = pidfd_prepare(pid, flags, &pidfd_file);
606 if (pidfd < 0)
607 return pidfd;
608
609 fd_install(pidfd, pidfd_file);
610 return pidfd;
611}
612
613/**
614 * sys_pidfd_open() - Open new pid file descriptor.
615 *
616 * @pid: pid for which to retrieve a pidfd
617 * @flags: flags to pass
618 *
619 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
620 * the task identified by @pid. Without PIDFD_THREAD flag the target task
621 * must be a thread-group leader.
622 *
623 * Return: On success, a cloexec pidfd is returned.
624 * On error, a negative errno number will be returned.
625 */
626SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
627{
628 int fd;
629 struct pid *p;
630
631 if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD))
632 return -EINVAL;
633
634 if (pid <= 0)
635 return -EINVAL;
636
637 p = find_get_pid(pid);
638 if (!p)
639 return -ESRCH;
640
641 fd = pidfd_create(p, flags);
642
643 put_pid(p);
644 return fd;
645}
646
647#ifdef CONFIG_SYSCTL
648static struct ctl_table_set *pid_table_root_lookup(struct ctl_table_root *root)
649{
650 return &task_active_pid_ns(current)->set;
651}
652
653static int set_is_seen(struct ctl_table_set *set)
654{
655 return &task_active_pid_ns(current)->set == set;
656}
657
658static int pid_table_root_permissions(struct ctl_table_header *head,
659 const struct ctl_table *table)
660{
661 struct pid_namespace *pidns =
662 container_of(head->set, struct pid_namespace, set);
663 int mode = table->mode;
664
665 if (ns_capable(pidns->user_ns, CAP_SYS_ADMIN) ||
666 uid_eq(current_euid(), make_kuid(pidns->user_ns, 0)))
667 mode = (mode & S_IRWXU) >> 6;
668 else if (in_egroup_p(make_kgid(pidns->user_ns, 0)))
669 mode = (mode & S_IRWXG) >> 3;
670 else
671 mode = mode & S_IROTH;
672 return (mode << 6) | (mode << 3) | mode;
673}
674
675static void pid_table_root_set_ownership(struct ctl_table_header *head,
676 kuid_t *uid, kgid_t *gid)
677{
678 struct pid_namespace *pidns =
679 container_of(head->set, struct pid_namespace, set);
680 kuid_t ns_root_uid;
681 kgid_t ns_root_gid;
682
683 ns_root_uid = make_kuid(pidns->user_ns, 0);
684 if (uid_valid(ns_root_uid))
685 *uid = ns_root_uid;
686
687 ns_root_gid = make_kgid(pidns->user_ns, 0);
688 if (gid_valid(ns_root_gid))
689 *gid = ns_root_gid;
690}
691
692static struct ctl_table_root pid_table_root = {
693 .lookup = pid_table_root_lookup,
694 .permissions = pid_table_root_permissions,
695 .set_ownership = pid_table_root_set_ownership,
696};
697
698static const struct ctl_table pid_table[] = {
699 {
700 .procname = "pid_max",
701 .data = &init_pid_ns.pid_max,
702 .maxlen = sizeof(int),
703 .mode = 0644,
704 .proc_handler = proc_dointvec_minmax,
705 .extra1 = &pid_max_min,
706 .extra2 = &pid_max_max,
707 },
708};
709#endif
710
711int register_pidns_sysctls(struct pid_namespace *pidns)
712{
713#ifdef CONFIG_SYSCTL
714 struct ctl_table *tbl;
715
716 setup_sysctl_set(&pidns->set, &pid_table_root, set_is_seen);
717
718 tbl = kmemdup(pid_table, sizeof(pid_table), GFP_KERNEL);
719 if (!tbl)
720 return -ENOMEM;
721 tbl->data = &pidns->pid_max;
722 pidns->pid_max = min(pid_max_max, max_t(int, pidns->pid_max,
723 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
724
725 pidns->sysctls = __register_sysctl_table(&pidns->set, "kernel", tbl,
726 ARRAY_SIZE(pid_table));
727 if (!pidns->sysctls) {
728 kfree(tbl);
729 retire_sysctl_set(&pidns->set);
730 return -ENOMEM;
731 }
732#endif
733 return 0;
734}
735
736void unregister_pidns_sysctls(struct pid_namespace *pidns)
737{
738#ifdef CONFIG_SYSCTL
739 const struct ctl_table *tbl;
740
741 tbl = pidns->sysctls->ctl_table_arg;
742 unregister_sysctl_table(pidns->sysctls);
743 retire_sysctl_set(&pidns->set);
744 kfree(tbl);
745#endif
746}
747
748void __init pid_idr_init(void)
749{
750 /* Verify no one has done anything silly: */
751 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
752
753 /* bump default and minimum pid_max based on number of cpus */
754 init_pid_ns.pid_max = min(pid_max_max, max_t(int, init_pid_ns.pid_max,
755 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
756 pid_max_min = max_t(int, pid_max_min,
757 PIDS_PER_CPU_MIN * num_possible_cpus());
758 pr_info("pid_max: default: %u minimum: %u\n", init_pid_ns.pid_max, pid_max_min);
759
760 idr_init(&init_pid_ns.idr);
761
762 init_pid_ns.pid_cachep = kmem_cache_create("pid",
763 struct_size_t(struct pid, numbers, 1),
764 __alignof__(struct pid),
765 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
766 NULL);
767}
768
769static __init int pid_namespace_sysctl_init(void)
770{
771#ifdef CONFIG_SYSCTL
772 /* "kernel" directory will have already been initialized. */
773 BUG_ON(register_pidns_sysctls(&init_pid_ns));
774#endif
775 return 0;
776}
777subsys_initcall(pid_namespace_sysctl_init);
778
779static struct file *__pidfd_fget(struct task_struct *task, int fd)
780{
781 struct file *file;
782 int ret;
783
784 ret = down_read_killable(&task->signal->exec_update_lock);
785 if (ret)
786 return ERR_PTR(ret);
787
788 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
789 file = fget_task(task, fd);
790 else
791 file = ERR_PTR(-EPERM);
792
793 up_read(&task->signal->exec_update_lock);
794
795 if (!file) {
796 /*
797 * It is possible that the target thread is exiting; it can be
798 * either:
799 * 1. before exit_signals(), which gives a real fd
800 * 2. before exit_files() takes the task_lock() gives a real fd
801 * 3. after exit_files() releases task_lock(), ->files is NULL;
802 * this has PF_EXITING, since it was set in exit_signals(),
803 * __pidfd_fget() returns EBADF.
804 * In case 3 we get EBADF, but that really means ESRCH, since
805 * the task is currently exiting and has freed its files
806 * struct, so we fix it up.
807 */
808 if (task->flags & PF_EXITING)
809 file = ERR_PTR(-ESRCH);
810 else
811 file = ERR_PTR(-EBADF);
812 }
813
814 return file;
815}
816
817static int pidfd_getfd(struct pid *pid, int fd)
818{
819 struct task_struct *task;
820 struct file *file;
821 int ret;
822
823 task = get_pid_task(pid, PIDTYPE_PID);
824 if (!task)
825 return -ESRCH;
826
827 file = __pidfd_fget(task, fd);
828 put_task_struct(task);
829 if (IS_ERR(file))
830 return PTR_ERR(file);
831
832 ret = receive_fd(file, NULL, O_CLOEXEC);
833 fput(file);
834
835 return ret;
836}
837
838/**
839 * sys_pidfd_getfd() - Get a file descriptor from another process
840 *
841 * @pidfd: the pidfd file descriptor of the process
842 * @fd: the file descriptor number to get
843 * @flags: flags on how to get the fd (reserved)
844 *
845 * This syscall gets a copy of a file descriptor from another process
846 * based on the pidfd, and file descriptor number. It requires that
847 * the calling process has the ability to ptrace the process represented
848 * by the pidfd. The process which is having its file descriptor copied
849 * is otherwise unaffected.
850 *
851 * Return: On success, a cloexec file descriptor is returned.
852 * On error, a negative errno number will be returned.
853 */
854SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
855 unsigned int, flags)
856{
857 struct pid *pid;
858
859 /* flags is currently unused - make sure it's unset */
860 if (flags)
861 return -EINVAL;
862
863 CLASS(fd, f)(pidfd);
864 if (fd_empty(f))
865 return -EBADF;
866
867 pid = pidfd_pid(fd_file(f));
868 if (IS_ERR(pid))
869 return PTR_ERR(pid);
870
871 return pidfd_getfd(pid, fd);
872}