1 // SPDX-License-Identifier: GPL-2.0-only 1
2 /*
3 * Generic pidhash and scalable, time-bounded
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 task
10 * against. There is very little to them aside
11 * parking tasks using given ID's on a list.
12 *
13 * The hash is always changed with the tasklis
14 * and the hash is only accessed with the task
15 * read-acquired, so there's no additional SMP
16 *
17 * We have a list of bitmap pages, which bitma
18 * Allocating and freeing PIDs is completely l
19 * allocation scenario when all but one out of
20 * allocated already: the scanning of 32 list
21 * bytes. The typical fastpath is a single suc
22 *
23 * Pid namespaces:
24 * (C) 2007 Pavel Emelyanov <xemul@openvz.o
25 * (C) 2007 Sukadev Bhattiprolu <sukadev@us
26 * Many thanks to Oleg Nesterov for commen
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 <net/sock.h>
47 #include <uapi/linux/pidfd.h>
48
49 struct pid init_struct_pid = {
50 .count = REFCOUNT_INIT(1),
51 .tasks = {
52 { .first = NULL },
53 { .first = NULL },
54 { .first = NULL },
55 },
56 .level = 0,
57 .numbers = { {
58 .nr = 0,
59 .ns = &init_pid_ns
60 }, }
61 };
62
63 int pid_max = PID_MAX_DEFAULT;
64
65 int pid_max_min = RESERVED_PIDS + 1;
66 int pid_max_max = PID_MAX_LIMIT;
67 /*
68 * Pseudo filesystems start inode numbering af
69 * PIDs as a natural offset.
70 */
71 static u64 pidfs_ino = RESERVED_PIDS;
72
73 /*
74 * PID-map pages start out as NULL, they get a
75 * first use and are never deallocated. This w
76 * value does not cause lots of bitmaps to be
77 * the scheme scales to up to 4 million PIDs,
78 */
79 struct pid_namespace init_pid_ns = {
80 .ns.count = REFCOUNT_INIT(2),
81 .idr = IDR_INIT(init_pid_ns.idr),
82 .pid_allocated = PIDNS_ADDING,
83 .level = 0,
84 .child_reaper = &init_task,
85 .user_ns = &init_user_ns,
86 .ns.inum = PROC_PID_INIT_INO,
87 #ifdef CONFIG_PID_NS
88 .ns.ops = &pidns_operations,
89 #endif
90 #if defined(CONFIG_SYSCTL) && defined(CONFIG_M
91 .memfd_noexec_scope = MEMFD_NOEXEC_SCO
92 #endif
93 };
94 EXPORT_SYMBOL_GPL(init_pid_ns);
95
96 /*
97 * Note: disable interrupts while the pidmap_l
98 * interrupt might come in and do read_lock(&t
99 *
100 * If we don't disable interrupts there is a n
101 * detach_pid()->free_pid() and another cpu th
102 * spin_lock(&pidmap_lock) followed by an inte
103 * read_lock(&tasklist_lock);
104 *
105 * After we clean up the tasklist_lock and kno
106 * irq handlers that take it we can leave the
107 * For now it is easier to be safe than to pro
108 */
109
110 static __cacheline_aligned_in_smp DEFINE_SPIN
111
112 void put_pid(struct pid *pid)
113 {
114 struct pid_namespace *ns;
115
116 if (!pid)
117 return;
118
119 ns = pid->numbers[pid->level].ns;
120 if (refcount_dec_and_test(&pid->count)
121 kmem_cache_free(ns->pid_cachep
122 put_pid_ns(ns);
123 }
124 }
125 EXPORT_SYMBOL_GPL(put_pid);
126
127 static void delayed_put_pid(struct rcu_head *r
128 {
129 struct pid *pid = container_of(rhp, st
130 put_pid(pid);
131 }
132
133 void free_pid(struct pid *pid)
134 {
135 /* We can be called with write_lock_ir
136 int i;
137 unsigned long flags;
138
139 spin_lock_irqsave(&pidmap_lock, flags)
140 for (i = 0; i <= pid->level; i++) {
141 struct upid *upid = pid->numbe
142 struct pid_namespace *ns = upi
143 switch (--ns->pid_allocated) {
144 case 2:
145 case 1:
146 /* When all that is le
147 * is the reaper wake
148 * may be sleeping in
149 */
150 wake_up_process(ns->ch
151 break;
152 case PIDNS_ADDING:
153 /* Handle a fork failu
154 WARN_ON(ns->child_reap
155 ns->pid_allocated = 0;
156 break;
157 }
158
159 idr_remove(&ns->idr, upid->nr)
160 }
161 spin_unlock_irqrestore(&pidmap_lock, f
162
163 call_rcu(&pid->rcu, delayed_put_pid);
164 }
165
166 struct pid *alloc_pid(struct pid_namespace *ns
167 size_t set_tid_size)
168 {
169 struct pid *pid;
170 enum pid_type type;
171 int i, nr;
172 struct pid_namespace *tmp;
173 struct upid *upid;
174 int retval = -ENOMEM;
175
176 /*
177 * set_tid_size contains the size of t
178 * the most nested currently active PI
179 * which PID to set for a process in t
180 * up to set_tid_size PID namespaces.
181 * for a process in all nested PID nam
182 * never be greater than the current n
183 */
184 if (set_tid_size > ns->level + 1)
185 return ERR_PTR(-EINVAL);
186
187 pid = kmem_cache_alloc(ns->pid_cachep,
188 if (!pid)
189 return ERR_PTR(retval);
190
191 tmp = ns;
192 pid->level = ns->level;
193
194 for (i = ns->level; i >= 0; i--) {
195 int tid = 0;
196
197 if (set_tid_size) {
198 tid = set_tid[ns->leve
199
200 retval = -EINVAL;
201 if (tid < 1 || tid >=
202 goto out_free;
203 /*
204 * Also fail if a PID
205 * no PID 1 exists.
206 */
207 if (tid != 1 && !tmp->
208 goto out_free;
209 retval = -EPERM;
210 if (!checkpoint_restor
211 goto out_free;
212 set_tid_size--;
213 }
214
215 idr_preload(GFP_KERNEL);
216 spin_lock_irq(&pidmap_lock);
217
218 if (tid) {
219 nr = idr_alloc(&tmp->i
220 tid + 1
221 /*
222 * If ENOSPC is return
223 * alreay in use. Retu
224 */
225 if (nr == -ENOSPC)
226 nr = -EEXIST;
227 } else {
228 int pid_min = 1;
229 /*
230 * init really needs p
231 * maximum wrap back t
232 */
233 if (idr_get_cursor(&tm
234 pid_min = RESE
235
236 /*
237 * Store a null pointe
238 * a partially initial
239 */
240 nr = idr_alloc_cyclic(
241
242 }
243 spin_unlock_irq(&pidmap_lock);
244 idr_preload_end();
245
246 if (nr < 0) {
247 retval = (nr == -ENOSP
248 goto out_free;
249 }
250
251 pid->numbers[i].nr = nr;
252 pid->numbers[i].ns = tmp;
253 tmp = tmp->parent;
254 }
255
256 /*
257 * ENOMEM is not the most obvious choi
258 * where the child subreaper has alrea
259 * namespace denies the creation of an
260 * is what we have exposed to userspac
261 * documented behavior for pid namespa
262 * change it even if there were an err
263 */
264 retval = -ENOMEM;
265
266 get_pid_ns(ns);
267 refcount_set(&pid->count, 1);
268 spin_lock_init(&pid->lock);
269 for (type = 0; type < PIDTYPE_MAX; ++t
270 INIT_HLIST_HEAD(&pid->tasks[ty
271
272 init_waitqueue_head(&pid->wait_pidfd);
273 INIT_HLIST_HEAD(&pid->inodes);
274
275 upid = pid->numbers + ns->level;
276 spin_lock_irq(&pidmap_lock);
277 if (!(ns->pid_allocated & PIDNS_ADDING
278 goto out_unlock;
279 pid->stashed = NULL;
280 pid->ino = ++pidfs_ino;
281 for ( ; upid >= pid->numbers; --upid)
282 /* Make the PID visible to fin
283 idr_replace(&upid->ns->idr, pi
284 upid->ns->pid_allocated++;
285 }
286 spin_unlock_irq(&pidmap_lock);
287
288 return pid;
289
290 out_unlock:
291 spin_unlock_irq(&pidmap_lock);
292 put_pid_ns(ns);
293
294 out_free:
295 spin_lock_irq(&pidmap_lock);
296 while (++i <= ns->level) {
297 upid = pid->numbers + i;
298 idr_remove(&upid->ns->idr, upi
299 }
300
301 /* On failure to allocate the first pi
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
311 void disable_pid_allocation(struct pid_namespa
312 {
313 spin_lock_irq(&pidmap_lock);
314 ns->pid_allocated &= ~PIDNS_ADDING;
315 spin_unlock_irq(&pidmap_lock);
316 }
317
318 struct pid *find_pid_ns(int nr, struct pid_nam
319 {
320 return idr_find(&ns->idr, nr);
321 }
322 EXPORT_SYMBOL_GPL(find_pid_ns);
323
324 struct pid *find_vpid(int nr)
325 {
326 return find_pid_ns(nr, task_active_pid
327 }
328 EXPORT_SYMBOL_GPL(find_vpid);
329
330 static struct pid **task_pid_ptr(struct task_s
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 taskli
339 */
340 void attach_pid(struct task_struct *task, enum
341 {
342 struct pid *pid = *task_pid_ptr(task,
343 hlist_add_head_rcu(&task->pid_links[ty
344 }
345
346 static void __change_pid(struct task_struct *t
347 struct pid *new)
348 {
349 struct pid **pid_ptr = task_pid_ptr(ta
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,
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
370 void detach_pid(struct task_struct *task, enum
371 {
372 __change_pid(task, type, NULL);
373 }
374
375 void change_pid(struct task_struct *task, enum
376 struct pid *pid)
377 {
378 __change_pid(task, type, pid);
379 attach_pid(task, type);
380 }
381
382 void exchange_tids(struct task_struct *left, s
383 {
384 struct pid *pid1 = left->thread_pid;
385 struct pid *pid2 = right->thread_pid;
386 struct hlist_head *head1 = &pid1->task
387 struct hlist_head *head2 = &pid2->task
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, p
394 rcu_assign_pointer(right->thread_pid,
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_p
402 void transfer_pid(struct task_struct *old, str
403 enum pid_type type)
404 {
405 WARN_ON_ONCE(type == PIDTYPE_PID);
406 hlist_replace_rcu(&old->pid_links[type
407 }
408
409 struct task_struct *pid_task(struct pid *pid,
410 {
411 struct task_struct *result = NULL;
412 if (pid) {
413 struct hlist_node *first;
414 first = rcu_dereference_check(
415
416 if (first)
417 result = hlist_entry(f
418 }
419 return result;
420 }
421 EXPORT_SYMBOL(pid_task);
422
423 /*
424 * Must be called under rcu_read_lock().
425 */
426 struct task_struct *find_task_by_pid_ns(pid_t
427 {
428 RCU_LOCKDEP_WARN(!rcu_read_lock_held()
429 "find_task_by_pid_ns(
430 return pid_task(find_pid_ns(nr, ns), P
431 }
432
433 struct task_struct *find_task_by_vpid(pid_t vn
434 {
435 return find_task_by_pid_ns(vnr, task_a
436 }
437
438 struct task_struct *find_get_task_by_vpid(pid_
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
451 struct pid *get_task_pid(struct task_struct *t
452 {
453 struct pid *pid;
454 rcu_read_lock();
455 pid = get_pid(rcu_dereference(*task_pi
456 rcu_read_unlock();
457 return pid;
458 }
459 EXPORT_SYMBOL_GPL(get_task_pid);
460
461 struct task_struct *get_pid_task(struct pid *p
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 }
471 EXPORT_SYMBOL_GPL(get_pid_task);
472
473 struct 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 }
483 EXPORT_SYMBOL_GPL(find_get_pid);
484
485 pid_t pid_nr_ns(struct pid *pid, struct pid_na
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 }
497 EXPORT_SYMBOL_GPL(pid_nr_ns);
498
499 pid_t pid_vnr(struct pid *pid)
500 {
501 return pid_nr_ns(pid, task_active_pid_
502 }
503 EXPORT_SYMBOL_GPL(pid_vnr);
504
505 pid_t __task_pid_nr_ns(struct task_struct *tas
506 struct pid_namespace *
507 {
508 pid_t nr = 0;
509
510 rcu_read_lock();
511 if (!ns)
512 ns = task_active_pid_ns(curren
513 nr = pid_nr_ns(rcu_dereference(*task_p
514 rcu_read_unlock();
515
516 return nr;
517 }
518 EXPORT_SYMBOL(__task_pid_nr_ns);
519
520 struct pid_namespace *task_active_pid_ns(struc
521 {
522 return ns_of_pid(task_pid(tsk));
523 }
524 EXPORT_SYMBOL_GPL(task_active_pid_ns);
525
526 /*
527 * Used by proc to find the first pid that is
528 *
529 * If there is a pid at nr this function is ex
530 */
531 struct pid *find_ge_pid(int nr, struct pid_nam
532 {
533 return idr_get_next(&ns->idr, &nr);
534 }
535 EXPORT_SYMBOL_GPL(find_ge_pid);
536
537 struct pid *pidfd_get_pid(unsigned int fd, uns
538 {
539 struct fd f;
540 struct pid *pid;
541
542 f = fdget(fd);
543 if (!fd_file(f))
544 return ERR_PTR(-EBADF);
545
546 pid = pidfd_pid(fd_file(f));
547 if (!IS_ERR(pid)) {
548 get_pid(pid);
549 *flags = fd_file(f)->f_flags;
550 }
551
552 fdput(f);
553 return pid;
554 }
555
556 /**
557 * pidfd_get_task() - Get the task associated
558 *
559 * @pidfd: pidfd for which to get the task
560 * @flags: flags associated with this pidfd
561 *
562 * Return the task associated with @pidfd. The
563 * the returned task. The caller is responsibl
564 *
565 * Return: On success, the task_struct associa
566 * On error, a negative errno number w
567 */
568 struct task_struct *pidfd_get_task(int pidfd,
569 {
570 unsigned int f_flags;
571 struct pid *pid;
572 struct task_struct *task;
573
574 pid = pidfd_get_pid(pidfd, &f_flags);
575 if (IS_ERR(pid))
576 return ERR_CAST(pid);
577
578 task = get_pid_task(pid, PIDTYPE_TGID)
579 put_pid(pid);
580 if (!task)
581 return ERR_PTR(-ESRCH);
582
583 *flags = f_flags;
584 return task;
585 }
586
587 /**
588 * pidfd_create() - Create a new pid file desc
589 *
590 * @pid: struct pid that the pidfd will refe
591 * @flags: flags to pass
592 *
593 * This creates a new pid file descriptor with
594 *
595 * Note, that this function can only be called
596 * been unshared to avoid leaking the pidfd to
597 *
598 * This symbol should not be explicitly export
599 *
600 * Return: On success, a cloexec pidfd is retu
601 * On error, a negative errno number w
602 */
603 static int pidfd_create(struct pid *pid, unsig
604 {
605 int pidfd;
606 struct file *pidfd_file;
607
608 pidfd = pidfd_prepare(pid, flags, &pid
609 if (pidfd < 0)
610 return pidfd;
611
612 fd_install(pidfd, pidfd_file);
613 return pidfd;
614 }
615
616 /**
617 * sys_pidfd_open() - Open new pid file descri
618 *
619 * @pid: pid for which to retrieve a pidfd
620 * @flags: flags to pass
621 *
622 * This creates a new pid file descriptor with
623 * the task identified by @pid. Without PIDFD_
624 * must be a thread-group leader.
625 *
626 * Return: On success, a cloexec pidfd is retu
627 * On error, a negative errno number w
628 */
629 SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsign
630 {
631 int fd;
632 struct pid *p;
633
634 if (flags & ~(PIDFD_NONBLOCK | PIDFD_T
635 return -EINVAL;
636
637 if (pid <= 0)
638 return -EINVAL;
639
640 p = find_get_pid(pid);
641 if (!p)
642 return -ESRCH;
643
644 fd = pidfd_create(p, flags);
645
646 put_pid(p);
647 return fd;
648 }
649
650 void __init pid_idr_init(void)
651 {
652 /* Verify no one has done anything sil
653 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_AD
654
655 /* bump default and minimum pid_max ba
656 pid_max = min(pid_max_max, max_t(int,
657 PIDS_PER_CPU_D
658 pid_max_min = max_t(int, pid_max_min,
659 PIDS_PER_CPU_M
660 pr_info("pid_max: default: %u minimum:
661
662 idr_init(&init_pid_ns.idr);
663
664 init_pid_ns.pid_cachep = kmem_cache_cr
665 struct_size_t(struct p
666 __alignof__(struct pid
667 SLAB_HWCACHE_ALIGN | S
668 NULL);
669 }
670
671 static struct file *__pidfd_fget(struct task_s
672 {
673 struct file *file;
674 int ret;
675
676 ret = down_read_killable(&task->signal
677 if (ret)
678 return ERR_PTR(ret);
679
680 if (ptrace_may_access(task, PTRACE_MOD
681 file = fget_task(task, fd);
682 else
683 file = ERR_PTR(-EPERM);
684
685 up_read(&task->signal->exec_update_loc
686
687 if (!file) {
688 /*
689 * It is possible that the tar
690 * either:
691 * 1. before exit_signals(), w
692 * 2. before exit_files() take
693 * 3. after exit_files() relea
694 * this has PF_EXITING, sin
695 * __pidfd_fget() returns E
696 * In case 3 we get EBADF, but
697 * the task is currently exiti
698 * struct, so we fix it up.
699 */
700 if (task->flags & PF_EXITING)
701 file = ERR_PTR(-ESRCH)
702 else
703 file = ERR_PTR(-EBADF)
704 }
705
706 return file;
707 }
708
709 static int pidfd_getfd(struct pid *pid, int fd
710 {
711 struct task_struct *task;
712 struct file *file;
713 int ret;
714
715 task = get_pid_task(pid, PIDTYPE_PID);
716 if (!task)
717 return -ESRCH;
718
719 file = __pidfd_fget(task, fd);
720 put_task_struct(task);
721 if (IS_ERR(file))
722 return PTR_ERR(file);
723
724 ret = receive_fd(file, NULL, O_CLOEXEC
725 fput(file);
726
727 return ret;
728 }
729
730 /**
731 * sys_pidfd_getfd() - Get a file descriptor f
732 *
733 * @pidfd: the pidfd file descriptor of t
734 * @fd: the file descriptor number to
735 * @flags: flags on how to get the fd (re
736 *
737 * This syscall gets a copy of a file descript
738 * based on the pidfd, and file descriptor num
739 * the calling process has the ability to ptra
740 * by the pidfd. The process which is having i
741 * is otherwise unaffected.
742 *
743 * Return: On success, a cloexec file descript
744 * On error, a negative errno number w
745 */
746 SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int,
747 unsigned int, flags)
748 {
749 struct pid *pid;
750 struct fd f;
751 int ret;
752
753 /* flags is currently unused - make su
754 if (flags)
755 return -EINVAL;
756
757 f = fdget(pidfd);
758 if (!fd_file(f))
759 return -EBADF;
760
761 pid = pidfd_pid(fd_file(f));
762 if (IS_ERR(pid))
763 ret = PTR_ERR(pid);
764 else
765 ret = pidfd_getfd(pid, fd);
766
767 fdput(f);
768 return ret;
769 }
770
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.