Line data Source code
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 <net/sock.h>
46 : #include <uapi/linux/pidfd.h>
47 :
48 : struct pid init_struct_pid = {
49 : .count = REFCOUNT_INIT(1),
50 : .tasks = {
51 : { .first = NULL },
52 : { .first = NULL },
53 : { .first = NULL },
54 : },
55 : .level = 0,
56 : .numbers = { {
57 : .nr = 0,
58 : .ns = &init_pid_ns,
59 : }, }
60 : };
61 :
62 : int pid_max = PID_MAX_DEFAULT;
63 :
64 : #define RESERVED_PIDS 300
65 :
66 : int pid_max_min = RESERVED_PIDS + 1;
67 : int pid_max_max = PID_MAX_LIMIT;
68 :
69 : /*
70 : * PID-map pages start out as NULL, they get allocated upon
71 : * first use and are never deallocated. This way a low pid_max
72 : * value does not cause lots of bitmaps to be allocated, but
73 : * the scheme scales to up to 4 million PIDs, runtime.
74 : */
75 : struct pid_namespace init_pid_ns = {
76 : .ns.count = REFCOUNT_INIT(2),
77 : .idr = IDR_INIT(init_pid_ns.idr),
78 : .pid_allocated = PIDNS_ADDING,
79 : .level = 0,
80 : .child_reaper = &init_task,
81 : .user_ns = &init_user_ns,
82 : .ns.inum = PROC_PID_INIT_INO,
83 : #ifdef CONFIG_PID_NS
84 : .ns.ops = &pidns_operations,
85 : #endif
86 : };
87 : EXPORT_SYMBOL_GPL(init_pid_ns);
88 :
89 : /*
90 : * Note: disable interrupts while the pidmap_lock is held as an
91 : * interrupt might come in and do read_lock(&tasklist_lock).
92 : *
93 : * If we don't disable interrupts there is a nasty deadlock between
94 : * detach_pid()->free_pid() and another cpu that does
95 : * spin_lock(&pidmap_lock) followed by an interrupt routine that does
96 : * read_lock(&tasklist_lock);
97 : *
98 : * After we clean up the tasklist_lock and know there are no
99 : * irq handlers that take it we can leave the interrupts enabled.
100 : * For now it is easier to be safe than to prove it can't happen.
101 : */
102 :
103 : static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
104 :
105 1013 : void put_pid(struct pid *pid)
106 : {
107 : struct pid_namespace *ns;
108 :
109 1013 : if (!pid)
110 : return;
111 :
112 1013 : ns = pid->numbers[pid->level].ns;
113 2026 : if (refcount_dec_and_test(&pid->count)) {
114 332 : kmem_cache_free(ns->pid_cachep, pid);
115 332 : put_pid_ns(ns);
116 : }
117 : }
118 : EXPORT_SYMBOL_GPL(put_pid);
119 :
120 332 : static void delayed_put_pid(struct rcu_head *rhp)
121 : {
122 332 : struct pid *pid = container_of(rhp, struct pid, rcu);
123 332 : put_pid(pid);
124 332 : }
125 :
126 333 : void free_pid(struct pid *pid)
127 : {
128 : /* We can be called with write_lock_irq(&tasklist_lock) held */
129 : int i;
130 : unsigned long flags;
131 :
132 333 : spin_lock_irqsave(&pidmap_lock, flags);
133 666 : for (i = 0; i <= pid->level; i++) {
134 333 : struct upid *upid = pid->numbers + i;
135 333 : struct pid_namespace *ns = upid->ns;
136 333 : switch (--ns->pid_allocated) {
137 : case 2:
138 : case 1:
139 : /* When all that is left in the pid namespace
140 : * is the reaper wake up the reaper. The reaper
141 : * may be sleeping in zap_pid_ns_processes().
142 : */
143 0 : wake_up_process(ns->child_reaper);
144 0 : break;
145 : case PIDNS_ADDING:
146 : /* Handle a fork failure of the first process */
147 0 : WARN_ON(ns->child_reaper);
148 0 : ns->pid_allocated = 0;
149 0 : break;
150 : }
151 :
152 333 : idr_remove(&ns->idr, upid->nr);
153 : }
154 333 : spin_unlock_irqrestore(&pidmap_lock, flags);
155 :
156 333 : call_rcu(&pid->rcu, delayed_put_pid);
157 333 : }
158 :
159 348 : struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
160 : size_t set_tid_size)
161 : {
162 : struct pid *pid;
163 : enum pid_type type;
164 : int i, nr;
165 : struct pid_namespace *tmp;
166 : struct upid *upid;
167 348 : int retval = -ENOMEM;
168 :
169 : /*
170 : * set_tid_size contains the size of the set_tid array. Starting at
171 : * the most nested currently active PID namespace it tells alloc_pid()
172 : * which PID to set for a process in that most nested PID namespace
173 : * up to set_tid_size PID namespaces. It does not have to set the PID
174 : * for a process in all nested PID namespaces but set_tid_size must
175 : * never be greater than the current ns->level + 1.
176 : */
177 348 : if (set_tid_size > ns->level + 1)
178 : return ERR_PTR(-EINVAL);
179 :
180 348 : pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
181 348 : if (!pid)
182 : return ERR_PTR(retval);
183 :
184 348 : tmp = ns;
185 348 : pid->level = ns->level;
186 :
187 696 : for (i = ns->level; i >= 0; i--) {
188 348 : int tid = 0;
189 :
190 348 : if (set_tid_size) {
191 0 : tid = set_tid[ns->level - i];
192 :
193 0 : retval = -EINVAL;
194 0 : if (tid < 1 || tid >= pid_max)
195 : goto out_free;
196 : /*
197 : * Also fail if a PID != 1 is requested and
198 : * no PID 1 exists.
199 : */
200 0 : if (tid != 1 && !tmp->child_reaper)
201 : goto out_free;
202 0 : retval = -EPERM;
203 0 : if (!checkpoint_restore_ns_capable(tmp->user_ns))
204 : goto out_free;
205 0 : set_tid_size--;
206 : }
207 :
208 348 : idr_preload(GFP_KERNEL);
209 348 : spin_lock_irq(&pidmap_lock);
210 :
211 348 : if (tid) {
212 0 : nr = idr_alloc(&tmp->idr, NULL, tid,
213 : tid + 1, GFP_ATOMIC);
214 : /*
215 : * If ENOSPC is returned it means that the PID is
216 : * alreay in use. Return EEXIST in that case.
217 : */
218 0 : if (nr == -ENOSPC)
219 0 : nr = -EEXIST;
220 : } else {
221 348 : int pid_min = 1;
222 : /*
223 : * init really needs pid 1, but after reaching the
224 : * maximum wrap back to RESERVED_PIDS
225 : */
226 696 : if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
227 48 : pid_min = RESERVED_PIDS;
228 :
229 : /*
230 : * Store a null pointer so find_pid_ns does not find
231 : * a partially initialized PID (see below).
232 : */
233 348 : nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
234 : pid_max, GFP_ATOMIC);
235 : }
236 348 : spin_unlock_irq(&pidmap_lock);
237 : idr_preload_end();
238 :
239 348 : if (nr < 0) {
240 0 : retval = (nr == -ENOSPC) ? -EAGAIN : nr;
241 : goto out_free;
242 : }
243 :
244 348 : pid->numbers[i].nr = nr;
245 348 : pid->numbers[i].ns = tmp;
246 348 : tmp = tmp->parent;
247 : }
248 :
249 : /*
250 : * ENOMEM is not the most obvious choice especially for the case
251 : * where the child subreaper has already exited and the pid
252 : * namespace denies the creation of any new processes. But ENOMEM
253 : * is what we have exposed to userspace for a long time and it is
254 : * documented behavior for pid namespaces. So we can't easily
255 : * change it even if there were an error code better suited.
256 : */
257 348 : retval = -ENOMEM;
258 :
259 348 : get_pid_ns(ns);
260 696 : refcount_set(&pid->count, 1);
261 348 : spin_lock_init(&pid->lock);
262 1740 : for (type = 0; type < PIDTYPE_MAX; ++type)
263 1392 : INIT_HLIST_HEAD(&pid->tasks[type]);
264 :
265 348 : init_waitqueue_head(&pid->wait_pidfd);
266 348 : INIT_HLIST_HEAD(&pid->inodes);
267 :
268 348 : upid = pid->numbers + ns->level;
269 348 : spin_lock_irq(&pidmap_lock);
270 348 : if (!(ns->pid_allocated & PIDNS_ADDING))
271 : goto out_unlock;
272 348 : for ( ; upid >= pid->numbers; --upid) {
273 : /* Make the PID visible to find_pid_ns. */
274 348 : idr_replace(&upid->ns->idr, pid, upid->nr);
275 348 : upid->ns->pid_allocated++;
276 : }
277 348 : spin_unlock_irq(&pidmap_lock);
278 :
279 348 : return pid;
280 :
281 : out_unlock:
282 0 : spin_unlock_irq(&pidmap_lock);
283 0 : put_pid_ns(ns);
284 :
285 : out_free:
286 : spin_lock_irq(&pidmap_lock);
287 0 : while (++i <= ns->level) {
288 0 : upid = pid->numbers + i;
289 0 : idr_remove(&upid->ns->idr, upid->nr);
290 : }
291 :
292 : /* On failure to allocate the first pid, reset the state */
293 0 : if (ns->pid_allocated == PIDNS_ADDING)
294 0 : idr_set_cursor(&ns->idr, 0);
295 :
296 0 : spin_unlock_irq(&pidmap_lock);
297 :
298 0 : kmem_cache_free(ns->pid_cachep, pid);
299 0 : return ERR_PTR(retval);
300 : }
301 :
302 0 : void disable_pid_allocation(struct pid_namespace *ns)
303 : {
304 0 : spin_lock_irq(&pidmap_lock);
305 0 : ns->pid_allocated &= ~PIDNS_ADDING;
306 0 : spin_unlock_irq(&pidmap_lock);
307 0 : }
308 :
309 0 : struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
310 : {
311 2 : return idr_find(&ns->idr, nr);
312 : }
313 : EXPORT_SYMBOL_GPL(find_pid_ns);
314 :
315 0 : struct pid *find_vpid(int nr)
316 : {
317 0 : return find_pid_ns(nr, task_active_pid_ns(current));
318 : }
319 : EXPORT_SYMBOL_GPL(find_vpid);
320 :
321 : static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
322 : {
323 : return (type == PIDTYPE_PID) ?
324 5448 : &task->thread_pid :
325 2043 : &task->signal->pids[type];
326 : }
327 :
328 : /*
329 : * attach_pid() must be called with the tasklist_lock write-held.
330 : */
331 1392 : void attach_pid(struct task_struct *task, enum pid_type type)
332 : {
333 1392 : struct pid *pid = *task_pid_ptr(task, type);
334 2784 : hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
335 1392 : }
336 :
337 1332 : static void __change_pid(struct task_struct *task, enum pid_type type,
338 : struct pid *new)
339 : {
340 1332 : struct pid **pid_ptr = task_pid_ptr(task, type);
341 : struct pid *pid;
342 : int tmp;
343 :
344 1332 : pid = *pid_ptr;
345 :
346 2664 : hlist_del_rcu(&task->pid_links[type]);
347 1332 : *pid_ptr = new;
348 :
349 4662 : for (tmp = PIDTYPE_MAX; --tmp >= 0; )
350 5994 : if (pid_has_task(pid, tmp))
351 : return;
352 :
353 333 : free_pid(pid);
354 : }
355 :
356 1332 : void detach_pid(struct task_struct *task, enum pid_type type)
357 : {
358 1332 : __change_pid(task, type, NULL);
359 1332 : }
360 :
361 0 : void change_pid(struct task_struct *task, enum pid_type type,
362 : struct pid *pid)
363 : {
364 0 : __change_pid(task, type, pid);
365 0 : attach_pid(task, type);
366 0 : }
367 :
368 0 : void exchange_tids(struct task_struct *left, struct task_struct *right)
369 : {
370 0 : struct pid *pid1 = left->thread_pid;
371 0 : struct pid *pid2 = right->thread_pid;
372 0 : struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
373 0 : struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
374 :
375 : /* Swap the single entry tid lists */
376 0 : hlists_swap_heads_rcu(head1, head2);
377 :
378 : /* Swap the per task_struct pid */
379 0 : rcu_assign_pointer(left->thread_pid, pid2);
380 0 : rcu_assign_pointer(right->thread_pid, pid1);
381 :
382 : /* Swap the cached value */
383 0 : WRITE_ONCE(left->pid, pid_nr(pid2));
384 0 : WRITE_ONCE(right->pid, pid_nr(pid1));
385 0 : }
386 :
387 : /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
388 0 : void transfer_pid(struct task_struct *old, struct task_struct *new,
389 : enum pid_type type)
390 : {
391 0 : if (type == PIDTYPE_PID)
392 0 : new->thread_pid = old->thread_pid;
393 0 : hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
394 0 : }
395 :
396 0 : struct task_struct *pid_task(struct pid *pid, enum pid_type type)
397 : {
398 2 : struct task_struct *result = NULL;
399 2 : if (pid) {
400 : struct hlist_node *first;
401 2 : first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
402 : lockdep_tasklist_lock_is_held());
403 2 : if (first)
404 2 : result = hlist_entry(first, struct task_struct, pid_links[(type)]);
405 : }
406 0 : return result;
407 : }
408 : EXPORT_SYMBOL(pid_task);
409 :
410 : /*
411 : * Must be called under rcu_read_lock().
412 : */
413 2 : struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
414 : {
415 : RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
416 : "find_task_by_pid_ns() needs rcu_read_lock() protection");
417 4 : return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
418 : }
419 :
420 0 : struct task_struct *find_task_by_vpid(pid_t vnr)
421 : {
422 0 : return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
423 : }
424 :
425 0 : struct task_struct *find_get_task_by_vpid(pid_t nr)
426 : {
427 : struct task_struct *task;
428 :
429 : rcu_read_lock();
430 0 : task = find_task_by_vpid(nr);
431 0 : if (task)
432 : get_task_struct(task);
433 : rcu_read_unlock();
434 :
435 0 : return task;
436 : }
437 :
438 348 : struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
439 : {
440 : struct pid *pid;
441 : rcu_read_lock();
442 696 : pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
443 : rcu_read_unlock();
444 348 : return pid;
445 : }
446 : EXPORT_SYMBOL_GPL(get_task_pid);
447 :
448 0 : struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
449 : {
450 : struct task_struct *result;
451 : rcu_read_lock();
452 0 : result = pid_task(pid, type);
453 0 : if (result)
454 : get_task_struct(result);
455 : rcu_read_unlock();
456 0 : return result;
457 : }
458 : EXPORT_SYMBOL_GPL(get_pid_task);
459 :
460 0 : struct pid *find_get_pid(pid_t nr)
461 : {
462 : struct pid *pid;
463 :
464 : rcu_read_lock();
465 0 : pid = get_pid(find_vpid(nr));
466 : rcu_read_unlock();
467 :
468 0 : return pid;
469 : }
470 : EXPORT_SYMBOL_GPL(find_get_pid);
471 :
472 0 : pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
473 : {
474 : struct upid *upid;
475 681 : pid_t nr = 0;
476 :
477 681 : if (pid && ns->level <= pid->level) {
478 681 : upid = &pid->numbers[ns->level];
479 681 : if (upid->ns == ns)
480 681 : nr = upid->nr;
481 : }
482 0 : return nr;
483 : }
484 : EXPORT_SYMBOL_GPL(pid_nr_ns);
485 :
486 348 : pid_t pid_vnr(struct pid *pid)
487 : {
488 1044 : return pid_nr_ns(pid, task_active_pid_ns(current));
489 : }
490 : EXPORT_SYMBOL_GPL(pid_vnr);
491 :
492 333 : pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
493 : struct pid_namespace *ns)
494 : {
495 333 : pid_t nr = 0;
496 :
497 : rcu_read_lock();
498 333 : if (!ns)
499 0 : ns = task_active_pid_ns(current);
500 666 : nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
501 : rcu_read_unlock();
502 :
503 333 : return nr;
504 : }
505 : EXPORT_SYMBOL(__task_pid_nr_ns);
506 :
507 667 : struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
508 : {
509 2030 : return ns_of_pid(task_pid(tsk));
510 : }
511 : EXPORT_SYMBOL_GPL(task_active_pid_ns);
512 :
513 : /*
514 : * Used by proc to find the first pid that is greater than or equal to nr.
515 : *
516 : * If there is a pid at nr this function is exactly the same as find_pid_ns.
517 : */
518 0 : struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
519 : {
520 0 : return idr_get_next(&ns->idr, &nr);
521 : }
522 : EXPORT_SYMBOL_GPL(find_ge_pid);
523 :
524 0 : struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
525 : {
526 : struct fd f;
527 : struct pid *pid;
528 :
529 0 : f = fdget(fd);
530 0 : if (!f.file)
531 : return ERR_PTR(-EBADF);
532 :
533 0 : pid = pidfd_pid(f.file);
534 0 : if (!IS_ERR(pid)) {
535 0 : get_pid(pid);
536 0 : *flags = f.file->f_flags;
537 : }
538 :
539 0 : fdput(f);
540 : return pid;
541 : }
542 :
543 : /**
544 : * pidfd_get_task() - Get the task associated with a pidfd
545 : *
546 : * @pidfd: pidfd for which to get the task
547 : * @flags: flags associated with this pidfd
548 : *
549 : * Return the task associated with @pidfd. The function takes a reference on
550 : * the returned task. The caller is responsible for releasing that reference.
551 : *
552 : * Currently, the process identified by @pidfd is always a thread-group leader.
553 : * This restriction currently exists for all aspects of pidfds including pidfd
554 : * creation (CLONE_PIDFD cannot be used with CLONE_THREAD) and pidfd polling
555 : * (only supports thread group leaders).
556 : *
557 : * Return: On success, the task_struct associated with the pidfd.
558 : * On error, a negative errno number will be returned.
559 : */
560 0 : struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
561 : {
562 : unsigned int f_flags;
563 : struct pid *pid;
564 : struct task_struct *task;
565 :
566 0 : pid = pidfd_get_pid(pidfd, &f_flags);
567 0 : if (IS_ERR(pid))
568 : return ERR_CAST(pid);
569 :
570 0 : task = get_pid_task(pid, PIDTYPE_TGID);
571 0 : put_pid(pid);
572 0 : if (!task)
573 : return ERR_PTR(-ESRCH);
574 :
575 0 : *flags = f_flags;
576 0 : return task;
577 : }
578 :
579 : /**
580 : * pidfd_create() - Create a new pid file descriptor.
581 : *
582 : * @pid: struct pid that the pidfd will reference
583 : * @flags: flags to pass
584 : *
585 : * This creates a new pid file descriptor with the O_CLOEXEC flag set.
586 : *
587 : * Note, that this function can only be called after the fd table has
588 : * been unshared to avoid leaking the pidfd to the new process.
589 : *
590 : * This symbol should not be explicitly exported to loadable modules.
591 : *
592 : * Return: On success, a cloexec pidfd is returned.
593 : * On error, a negative errno number will be returned.
594 : */
595 0 : int pidfd_create(struct pid *pid, unsigned int flags)
596 : {
597 : int fd;
598 :
599 0 : if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
600 : return -EINVAL;
601 :
602 0 : if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
603 : return -EINVAL;
604 :
605 0 : fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
606 0 : flags | O_RDWR | O_CLOEXEC);
607 0 : if (fd < 0)
608 0 : put_pid(pid);
609 :
610 : return fd;
611 : }
612 :
613 : /**
614 : * 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 process identified by @pid. Currently, the process identified by
621 : * @pid must be a thread-group leader. This restriction currently exists
622 : * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
623 : * be used with CLONE_THREAD) and pidfd polling (only supports thread group
624 : * leaders).
625 : *
626 : * Return: On success, a cloexec pidfd is returned.
627 : * On error, a negative errno number will be returned.
628 : */
629 0 : SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
630 : {
631 : int fd;
632 : struct pid *p;
633 :
634 0 : if (flags & ~PIDFD_NONBLOCK)
635 : return -EINVAL;
636 :
637 0 : if (pid <= 0)
638 : return -EINVAL;
639 :
640 0 : p = find_get_pid(pid);
641 0 : if (!p)
642 : return -ESRCH;
643 :
644 0 : fd = pidfd_create(p, flags);
645 :
646 0 : put_pid(p);
647 0 : return fd;
648 : }
649 :
650 1 : void __init pid_idr_init(void)
651 : {
652 : /* Verify no one has done anything silly: */
653 : BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
654 :
655 : /* bump default and minimum pid_max based on number of cpus */
656 1 : pid_max = min(pid_max_max, max_t(int, pid_max,
657 : PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
658 1 : pid_max_min = max_t(int, pid_max_min,
659 : PIDS_PER_CPU_MIN * num_possible_cpus());
660 1 : pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
661 :
662 1 : idr_init(&init_pid_ns.idr);
663 :
664 1 : init_pid_ns.pid_cachep = KMEM_CACHE(pid,
665 : SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
666 1 : }
667 :
668 0 : static struct file *__pidfd_fget(struct task_struct *task, int fd)
669 : {
670 : struct file *file;
671 : int ret;
672 :
673 0 : ret = down_read_killable(&task->signal->exec_update_lock);
674 0 : if (ret)
675 0 : return ERR_PTR(ret);
676 :
677 0 : if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
678 0 : file = fget_task(task, fd);
679 : else
680 : file = ERR_PTR(-EPERM);
681 :
682 0 : up_read(&task->signal->exec_update_lock);
683 :
684 0 : return file ?: ERR_PTR(-EBADF);
685 : }
686 :
687 0 : static int pidfd_getfd(struct pid *pid, int fd)
688 : {
689 : struct task_struct *task;
690 : struct file *file;
691 : int ret;
692 :
693 0 : task = get_pid_task(pid, PIDTYPE_PID);
694 0 : if (!task)
695 : return -ESRCH;
696 :
697 0 : file = __pidfd_fget(task, fd);
698 0 : put_task_struct(task);
699 0 : if (IS_ERR(file))
700 0 : return PTR_ERR(file);
701 :
702 0 : ret = receive_fd(file, O_CLOEXEC);
703 0 : fput(file);
704 :
705 0 : return ret;
706 : }
707 :
708 : /**
709 : * sys_pidfd_getfd() - Get a file descriptor from another process
710 : *
711 : * @pidfd: the pidfd file descriptor of the process
712 : * @fd: the file descriptor number to get
713 : * @flags: flags on how to get the fd (reserved)
714 : *
715 : * This syscall gets a copy of a file descriptor from another process
716 : * based on the pidfd, and file descriptor number. It requires that
717 : * the calling process has the ability to ptrace the process represented
718 : * by the pidfd. The process which is having its file descriptor copied
719 : * is otherwise unaffected.
720 : *
721 : * Return: On success, a cloexec file descriptor is returned.
722 : * On error, a negative errno number will be returned.
723 : */
724 0 : SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
725 : unsigned int, flags)
726 : {
727 : struct pid *pid;
728 : struct fd f;
729 : int ret;
730 :
731 : /* flags is currently unused - make sure it's unset */
732 0 : if (flags)
733 : return -EINVAL;
734 :
735 0 : f = fdget(pidfd);
736 0 : if (!f.file)
737 : return -EBADF;
738 :
739 0 : pid = pidfd_pid(f.file);
740 0 : if (IS_ERR(pid))
741 0 : ret = PTR_ERR(pid);
742 : else
743 0 : ret = pidfd_getfd(pid, fd);
744 :
745 0 : fdput(f);
746 0 : return ret;
747 : }
|
