Loading...
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
46struct pid init_struct_pid = {
47 .count = REFCOUNT_INIT(1),
48 .tasks = {
49 { .first = NULL },
50 { .first = NULL },
51 { .first = NULL },
52 },
53 .level = 0,
54 .numbers = { {
55 .nr = 0,
56 .ns = &init_pid_ns,
57 }, }
58};
59
60int pid_max = PID_MAX_DEFAULT;
61
62#define RESERVED_PIDS 300
63
64int pid_max_min = RESERVED_PIDS + 1;
65int 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 .kref = KREF_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};
85EXPORT_SYMBOL_GPL(init_pid_ns);
86
87/*
88 * Note: disable interrupts while the pidmap_lock is held as an
89 * interrupt might come in and do read_lock(&tasklist_lock).
90 *
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);
95 *
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.
99 */
100
101static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
102
103void put_pid(struct pid *pid)
104{
105 struct pid_namespace *ns;
106
107 if (!pid)
108 return;
109
110 ns = pid->numbers[pid->level].ns;
111 if (refcount_dec_and_test(&pid->count)) {
112 kmem_cache_free(ns->pid_cachep, pid);
113 put_pid_ns(ns);
114 }
115}
116EXPORT_SYMBOL_GPL(put_pid);
117
118static void delayed_put_pid(struct rcu_head *rhp)
119{
120 struct pid *pid = container_of(rhp, struct pid, rcu);
121 put_pid(pid);
122}
123
124void free_pid(struct pid *pid)
125{
126 /* We can be called with write_lock_irq(&tasklist_lock) held */
127 int i;
128 unsigned long flags;
129
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) {
135 case 2:
136 case 1:
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().
140 */
141 wake_up_process(ns->child_reaper);
142 break;
143 case PIDNS_ADDING:
144 /* Handle a fork failure of the first process */
145 WARN_ON(ns->child_reaper);
146 ns->pid_allocated = 0;
147 /* fall through */
148 case 0:
149 schedule_work(&ns->proc_work);
150 break;
151 }
152
153 idr_remove(&ns->idr, upid->nr);
154 }
155 spin_unlock_irqrestore(&pidmap_lock, flags);
156
157 call_rcu(&pid->rcu, delayed_put_pid);
158}
159
160struct pid *alloc_pid(struct pid_namespace *ns)
161{
162 struct pid *pid;
163 enum pid_type type;
164 int i, nr;
165 struct pid_namespace *tmp;
166 struct upid *upid;
167 int retval = -ENOMEM;
168
169 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
170 if (!pid)
171 return ERR_PTR(retval);
172
173 tmp = ns;
174 pid->level = ns->level;
175
176 for (i = ns->level; i >= 0; i--) {
177 int pid_min = 1;
178
179 idr_preload(GFP_KERNEL);
180 spin_lock_irq(&pidmap_lock);
181
182 /*
183 * init really needs pid 1, but after reaching the maximum
184 * wrap back to RESERVED_PIDS
185 */
186 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
187 pid_min = RESERVED_PIDS;
188
189 /*
190 * Store a null pointer so find_pid_ns does not find
191 * a partially initialized PID (see below).
192 */
193 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
194 pid_max, GFP_ATOMIC);
195 spin_unlock_irq(&pidmap_lock);
196 idr_preload_end();
197
198 if (nr < 0) {
199 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
200 goto out_free;
201 }
202
203 pid->numbers[i].nr = nr;
204 pid->numbers[i].ns = tmp;
205 tmp = tmp->parent;
206 }
207
208 if (unlikely(is_child_reaper(pid))) {
209 if (pid_ns_prepare_proc(ns))
210 goto out_free;
211 }
212
213 get_pid_ns(ns);
214 refcount_set(&pid->count, 1);
215 for (type = 0; type < PIDTYPE_MAX; ++type)
216 INIT_HLIST_HEAD(&pid->tasks[type]);
217
218 init_waitqueue_head(&pid->wait_pidfd);
219
220 upid = pid->numbers + ns->level;
221 spin_lock_irq(&pidmap_lock);
222 if (!(ns->pid_allocated & PIDNS_ADDING))
223 goto out_unlock;
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++;
228 }
229 spin_unlock_irq(&pidmap_lock);
230
231 return pid;
232
233out_unlock:
234 spin_unlock_irq(&pidmap_lock);
235 put_pid_ns(ns);
236
237out_free:
238 spin_lock_irq(&pidmap_lock);
239 while (++i <= ns->level) {
240 upid = pid->numbers + i;
241 idr_remove(&upid->ns->idr, upid->nr);
242 }
243
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);
247
248 spin_unlock_irq(&pidmap_lock);
249
250 kmem_cache_free(ns->pid_cachep, pid);
251 return ERR_PTR(retval);
252}
253
254void disable_pid_allocation(struct pid_namespace *ns)
255{
256 spin_lock_irq(&pidmap_lock);
257 ns->pid_allocated &= ~PIDNS_ADDING;
258 spin_unlock_irq(&pidmap_lock);
259}
260
261struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
262{
263 return idr_find(&ns->idr, nr);
264}
265EXPORT_SYMBOL_GPL(find_pid_ns);
266
267struct pid *find_vpid(int nr)
268{
269 return find_pid_ns(nr, task_active_pid_ns(current));
270}
271EXPORT_SYMBOL_GPL(find_vpid);
272
273static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
274{
275 return (type == PIDTYPE_PID) ?
276 &task->thread_pid :
277 &task->signal->pids[type];
278}
279
280/*
281 * attach_pid() must be called with the tasklist_lock write-held.
282 */
283void attach_pid(struct task_struct *task, enum pid_type type)
284{
285 struct pid *pid = *task_pid_ptr(task, type);
286 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
287}
288
289static void __change_pid(struct task_struct *task, enum pid_type type,
290 struct pid *new)
291{
292 struct pid **pid_ptr = task_pid_ptr(task, type);
293 struct pid *pid;
294 int tmp;
295
296 pid = *pid_ptr;
297
298 hlist_del_rcu(&task->pid_links[type]);
299 *pid_ptr = new;
300
301 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
302 if (!hlist_empty(&pid->tasks[tmp]))
303 return;
304
305 free_pid(pid);
306}
307
308void detach_pid(struct task_struct *task, enum pid_type type)
309{
310 __change_pid(task, type, NULL);
311}
312
313void change_pid(struct task_struct *task, enum pid_type type,
314 struct pid *pid)
315{
316 __change_pid(task, type, pid);
317 attach_pid(task, type);
318}
319
320/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
321void transfer_pid(struct task_struct *old, struct task_struct *new,
322 enum pid_type type)
323{
324 if (type == PIDTYPE_PID)
325 new->thread_pid = old->thread_pid;
326 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
327}
328
329struct task_struct *pid_task(struct pid *pid, enum pid_type type)
330{
331 struct task_struct *result = NULL;
332 if (pid) {
333 struct hlist_node *first;
334 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
335 lockdep_tasklist_lock_is_held());
336 if (first)
337 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
338 }
339 return result;
340}
341EXPORT_SYMBOL(pid_task);
342
343/*
344 * Must be called under rcu_read_lock().
345 */
346struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
347{
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);
351}
352
353struct task_struct *find_task_by_vpid(pid_t vnr)
354{
355 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
356}
357
358struct task_struct *find_get_task_by_vpid(pid_t nr)
359{
360 struct task_struct *task;
361
362 rcu_read_lock();
363 task = find_task_by_vpid(nr);
364 if (task)
365 get_task_struct(task);
366 rcu_read_unlock();
367
368 return task;
369}
370
371struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
372{
373 struct pid *pid;
374 rcu_read_lock();
375 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
376 rcu_read_unlock();
377 return pid;
378}
379EXPORT_SYMBOL_GPL(get_task_pid);
380
381struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
382{
383 struct task_struct *result;
384 rcu_read_lock();
385 result = pid_task(pid, type);
386 if (result)
387 get_task_struct(result);
388 rcu_read_unlock();
389 return result;
390}
391EXPORT_SYMBOL_GPL(get_pid_task);
392
393struct pid *find_get_pid(pid_t nr)
394{
395 struct pid *pid;
396
397 rcu_read_lock();
398 pid = get_pid(find_vpid(nr));
399 rcu_read_unlock();
400
401 return pid;
402}
403EXPORT_SYMBOL_GPL(find_get_pid);
404
405pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
406{
407 struct upid *upid;
408 pid_t nr = 0;
409
410 if (pid && ns->level <= pid->level) {
411 upid = &pid->numbers[ns->level];
412 if (upid->ns == ns)
413 nr = upid->nr;
414 }
415 return nr;
416}
417EXPORT_SYMBOL_GPL(pid_nr_ns);
418
419pid_t pid_vnr(struct pid *pid)
420{
421 return pid_nr_ns(pid, task_active_pid_ns(current));
422}
423EXPORT_SYMBOL_GPL(pid_vnr);
424
425pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
426 struct pid_namespace *ns)
427{
428 pid_t nr = 0;
429
430 rcu_read_lock();
431 if (!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);
435 rcu_read_unlock();
436
437 return nr;
438}
439EXPORT_SYMBOL(__task_pid_nr_ns);
440
441struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
442{
443 return ns_of_pid(task_pid(tsk));
444}
445EXPORT_SYMBOL_GPL(task_active_pid_ns);
446
447/*
448 * Used by proc to find the first pid that is greater than or equal to nr.
449 *
450 * If there is a pid at nr this function is exactly the same as find_pid_ns.
451 */
452struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
453{
454 return idr_get_next(&ns->idr, &nr);
455}
456
457/**
458 * pidfd_create() - Create a new pid file descriptor.
459 *
460 * @pid: struct pid that the pidfd will reference
461 *
462 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
463 *
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.
466 *
467 * Return: On success, a cloexec pidfd is returned.
468 * On error, a negative errno number will be returned.
469 */
470static int pidfd_create(struct pid *pid)
471{
472 int fd;
473
474 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
475 O_RDWR | O_CLOEXEC);
476 if (fd < 0)
477 put_pid(pid);
478
479 return fd;
480}
481
482/**
483 * pidfd_open() - Open new pid file descriptor.
484 *
485 * @pid: pid for which to retrieve a pidfd
486 * @flags: flags to pass
487 *
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
493 * leaders).
494 *
495 * Return: On success, a cloexec pidfd is returned.
496 * On error, a negative errno number will be returned.
497 */
498SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
499{
500 int fd, ret;
501 struct pid *p;
502
503 if (flags)
504 return -EINVAL;
505
506 if (pid <= 0)
507 return -EINVAL;
508
509 p = find_get_pid(pid);
510 if (!p)
511 return -ESRCH;
512
513 ret = 0;
514 rcu_read_lock();
515 if (!pid_task(p, PIDTYPE_TGID))
516 ret = -EINVAL;
517 rcu_read_unlock();
518
519 fd = ret ?: pidfd_create(p);
520 put_pid(p);
521 return fd;
522}
523
524void __init pid_idr_init(void)
525{
526 /* Verify no one has done anything silly: */
527 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
528
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);
535
536 idr_init(&init_pid_ns.idr);
537
538 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
539 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT);
540}
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 <net/sock.h>
47#include <uapi/linux/pidfd.h>
48
49struct 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
63int pid_max = PID_MAX_DEFAULT;
64
65int pid_max_min = RESERVED_PIDS + 1;
66int pid_max_max = PID_MAX_LIMIT;
67/*
68 * Pseudo filesystems start inode numbering after one. We use Reserved
69 * PIDs as a natural offset.
70 */
71static u64 pidfs_ino = RESERVED_PIDS;
72
73/*
74 * PID-map pages start out as NULL, they get allocated upon
75 * first use and are never deallocated. This way a low pid_max
76 * value does not cause lots of bitmaps to be allocated, but
77 * the scheme scales to up to 4 million PIDs, runtime.
78 */
79struct 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_MEMFD_CREATE)
91 .memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC,
92#endif
93};
94EXPORT_SYMBOL_GPL(init_pid_ns);
95
96/*
97 * Note: disable interrupts while the pidmap_lock is held as an
98 * interrupt might come in and do read_lock(&tasklist_lock).
99 *
100 * If we don't disable interrupts there is a nasty deadlock between
101 * detach_pid()->free_pid() and another cpu that does
102 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
103 * read_lock(&tasklist_lock);
104 *
105 * After we clean up the tasklist_lock and know there are no
106 * irq handlers that take it we can leave the interrupts enabled.
107 * For now it is easier to be safe than to prove it can't happen.
108 */
109
110static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
111
112void 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, pid);
122 put_pid_ns(ns);
123 }
124}
125EXPORT_SYMBOL_GPL(put_pid);
126
127static void delayed_put_pid(struct rcu_head *rhp)
128{
129 struct pid *pid = container_of(rhp, struct pid, rcu);
130 put_pid(pid);
131}
132
133void free_pid(struct pid *pid)
134{
135 /* We can be called with write_lock_irq(&tasklist_lock) held */
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->numbers + i;
142 struct pid_namespace *ns = upid->ns;
143 switch (--ns->pid_allocated) {
144 case 2:
145 case 1:
146 /* When all that is left in the pid namespace
147 * is the reaper wake up the reaper. The reaper
148 * may be sleeping in zap_pid_ns_processes().
149 */
150 wake_up_process(ns->child_reaper);
151 break;
152 case PIDNS_ADDING:
153 /* Handle a fork failure of the first process */
154 WARN_ON(ns->child_reaper);
155 ns->pid_allocated = 0;
156 break;
157 }
158
159 idr_remove(&ns->idr, upid->nr);
160 }
161 spin_unlock_irqrestore(&pidmap_lock, flags);
162
163 call_rcu(&pid->rcu, delayed_put_pid);
164}
165
166struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
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 the set_tid array. Starting at
178 * the most nested currently active PID namespace it tells alloc_pid()
179 * which PID to set for a process in that most nested PID namespace
180 * up to set_tid_size PID namespaces. It does not have to set the PID
181 * for a process in all nested PID namespaces but set_tid_size must
182 * never be greater than the current ns->level + 1.
183 */
184 if (set_tid_size > ns->level + 1)
185 return ERR_PTR(-EINVAL);
186
187 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
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->level - i];
199
200 retval = -EINVAL;
201 if (tid < 1 || tid >= pid_max)
202 goto out_free;
203 /*
204 * Also fail if a PID != 1 is requested and
205 * no PID 1 exists.
206 */
207 if (tid != 1 && !tmp->child_reaper)
208 goto out_free;
209 retval = -EPERM;
210 if (!checkpoint_restore_ns_capable(tmp->user_ns))
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->idr, NULL, tid,
220 tid + 1, GFP_ATOMIC);
221 /*
222 * If ENOSPC is returned it means that the PID is
223 * alreay in use. Return EEXIST in that case.
224 */
225 if (nr == -ENOSPC)
226 nr = -EEXIST;
227 } else {
228 int pid_min = 1;
229 /*
230 * init really needs pid 1, but after reaching the
231 * maximum wrap back to RESERVED_PIDS
232 */
233 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
234 pid_min = RESERVED_PIDS;
235
236 /*
237 * Store a null pointer so find_pid_ns does not find
238 * a partially initialized PID (see below).
239 */
240 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
241 pid_max, GFP_ATOMIC);
242 }
243 spin_unlock_irq(&pidmap_lock);
244 idr_preload_end();
245
246 if (nr < 0) {
247 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
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 choice especially for the case
258 * where the child subreaper has already exited and the pid
259 * namespace denies the creation of any new processes. But ENOMEM
260 * is what we have exposed to userspace for a long time and it is
261 * documented behavior for pid namespaces. So we can't easily
262 * change it even if there were an error code better suited.
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; ++type)
270 INIT_HLIST_HEAD(&pid->tasks[type]);
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 find_pid_ns. */
283 idr_replace(&upid->ns->idr, pid, upid->nr);
284 upid->ns->pid_allocated++;
285 }
286 spin_unlock_irq(&pidmap_lock);
287
288 return pid;
289
290out_unlock:
291 spin_unlock_irq(&pidmap_lock);
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 struct fd f;
540 struct pid *pid;
541
542 f = fdget(fd);
543 if (!f.file)
544 return ERR_PTR(-EBADF);
545
546 pid = pidfd_pid(f.file);
547 if (!IS_ERR(pid)) {
548 get_pid(pid);
549 *flags = f.file->f_flags;
550 }
551
552 fdput(f);
553 return pid;
554}
555
556/**
557 * pidfd_get_task() - Get the task associated with a pidfd
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 function takes a reference on
563 * the returned task. The caller is responsible for releasing that reference.
564 *
565 * Return: On success, the task_struct associated with the pidfd.
566 * On error, a negative errno number will be returned.
567 */
568struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
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 descriptor.
589 *
590 * @pid: struct pid that the pidfd will reference
591 * @flags: flags to pass
592 *
593 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
594 *
595 * Note, that this function can only be called after the fd table has
596 * been unshared to avoid leaking the pidfd to the new process.
597 *
598 * This symbol should not be explicitly exported to loadable modules.
599 *
600 * Return: On success, a cloexec pidfd is returned.
601 * On error, a negative errno number will be returned.
602 */
603static int pidfd_create(struct pid *pid, unsigned int flags)
604{
605 int pidfd;
606 struct file *pidfd_file;
607
608 pidfd = pidfd_prepare(pid, flags, &pidfd_file);
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 descriptor.
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 the O_CLOEXEC flag set for
623 * the task identified by @pid. Without PIDFD_THREAD flag the target task
624 * must be a thread-group leader.
625 *
626 * Return: On success, a cloexec pidfd is returned.
627 * On error, a negative errno number will be returned.
628 */
629SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
630{
631 int fd;
632 struct pid *p;
633
634 if (flags & ~(PIDFD_NONBLOCK | PIDFD_THREAD))
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
650void __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 pid_max = min(pid_max_max, max_t(int, pid_max,
657 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
658 pid_max_min = max_t(int, pid_max_min,
659 PIDS_PER_CPU_MIN * num_possible_cpus());
660 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
661
662 idr_init(&init_pid_ns.idr);
663
664 init_pid_ns.pid_cachep = kmem_cache_create("pid",
665 struct_size_t(struct pid, numbers, 1),
666 __alignof__(struct pid),
667 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
668 NULL);
669}
670
671static struct file *__pidfd_fget(struct task_struct *task, int fd)
672{
673 struct file *file;
674 int ret;
675
676 ret = down_read_killable(&task->signal->exec_update_lock);
677 if (ret)
678 return ERR_PTR(ret);
679
680 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
681 file = fget_task(task, fd);
682 else
683 file = ERR_PTR(-EPERM);
684
685 up_read(&task->signal->exec_update_lock);
686
687 if (!file) {
688 /*
689 * It is possible that the target thread is exiting; it can be
690 * either:
691 * 1. before exit_signals(), which gives a real fd
692 * 2. before exit_files() takes the task_lock() gives a real fd
693 * 3. after exit_files() releases task_lock(), ->files is NULL;
694 * this has PF_EXITING, since it was set in exit_signals(),
695 * __pidfd_fget() returns EBADF.
696 * In case 3 we get EBADF, but that really means ESRCH, since
697 * the task is currently exiting and has freed its files
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
709static 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 from another process
732 *
733 * @pidfd: the pidfd file descriptor of the process
734 * @fd: the file descriptor number to get
735 * @flags: flags on how to get the fd (reserved)
736 *
737 * This syscall gets a copy of a file descriptor from another process
738 * based on the pidfd, and file descriptor number. It requires that
739 * the calling process has the ability to ptrace the process represented
740 * by the pidfd. The process which is having its file descriptor copied
741 * is otherwise unaffected.
742 *
743 * Return: On success, a cloexec file descriptor is returned.
744 * On error, a negative errno number will be returned.
745 */
746SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
747 unsigned int, flags)
748{
749 struct pid *pid;
750 struct fd f;
751 int ret;
752
753 /* flags is currently unused - make sure it's unset */
754 if (flags)
755 return -EINVAL;
756
757 f = fdget(pidfd);
758 if (!f.file)
759 return -EBADF;
760
761 pid = pidfd_pid(f.file);
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}