Loading...
1/*
2 * Generic pidhash and scalable, time-bounded PID allocator
3 *
4 * (C) 2002-2003 William Irwin, IBM
5 * (C) 2004 William Irwin, Oracle
6 * (C) 2002-2004 Ingo Molnar, Red Hat
7 *
8 * pid-structures are backing objects for tasks sharing a given ID to chain
9 * against. There is very little to them aside from hashing them and
10 * parking tasks using given ID's on a list.
11 *
12 * The hash is always changed with the tasklist_lock write-acquired,
13 * and the hash is only accessed with the tasklist_lock at least
14 * read-acquired, so there's no additional SMP locking needed here.
15 *
16 * We have a list of bitmap pages, which bitmaps represent the PID space.
17 * Allocating and freeing PIDs is completely lockless. The worst-case
18 * allocation scenario when all but one out of 1 million PIDs possible are
19 * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
20 * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
21 *
22 * Pid namespaces:
23 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
24 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
25 * Many thanks to Oleg Nesterov for comments and help
26 *
27 */
28
29#include <linux/mm.h>
30#include <linux/export.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/rculist.h>
34#include <linux/bootmem.h>
35#include <linux/hash.h>
36#include <linux/pid_namespace.h>
37#include <linux/init_task.h>
38#include <linux/syscalls.h>
39
40#define pid_hashfn(nr, ns) \
41 hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
42static struct hlist_head *pid_hash;
43static unsigned int pidhash_shift = 4;
44struct pid init_struct_pid = INIT_STRUCT_PID;
45
46int pid_max = PID_MAX_DEFAULT;
47
48#define RESERVED_PIDS 300
49
50int pid_max_min = RESERVED_PIDS + 1;
51int pid_max_max = PID_MAX_LIMIT;
52
53#define BITS_PER_PAGE (PAGE_SIZE*8)
54#define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
55
56static inline int mk_pid(struct pid_namespace *pid_ns,
57 struct pidmap *map, int off)
58{
59 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
60}
61
62#define find_next_offset(map, off) \
63 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
64
65/*
66 * PID-map pages start out as NULL, they get allocated upon
67 * first use and are never deallocated. This way a low pid_max
68 * value does not cause lots of bitmaps to be allocated, but
69 * the scheme scales to up to 4 million PIDs, runtime.
70 */
71struct pid_namespace init_pid_ns = {
72 .kref = {
73 .refcount = ATOMIC_INIT(2),
74 },
75 .pidmap = {
76 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
77 },
78 .last_pid = 0,
79 .level = 0,
80 .child_reaper = &init_task,
81};
82EXPORT_SYMBOL_GPL(init_pid_ns);
83
84int is_container_init(struct task_struct *tsk)
85{
86 int ret = 0;
87 struct pid *pid;
88
89 rcu_read_lock();
90 pid = task_pid(tsk);
91 if (pid != NULL && pid->numbers[pid->level].nr == 1)
92 ret = 1;
93 rcu_read_unlock();
94
95 return ret;
96}
97EXPORT_SYMBOL(is_container_init);
98
99/*
100 * Note: disable interrupts while the pidmap_lock is held as an
101 * interrupt might come in and do read_lock(&tasklist_lock).
102 *
103 * If we don't disable interrupts there is a nasty deadlock between
104 * detach_pid()->free_pid() and another cpu that does
105 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
106 * read_lock(&tasklist_lock);
107 *
108 * After we clean up the tasklist_lock and know there are no
109 * irq handlers that take it we can leave the interrupts enabled.
110 * For now it is easier to be safe than to prove it can't happen.
111 */
112
113static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
114
115static void free_pidmap(struct upid *upid)
116{
117 int nr = upid->nr;
118 struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
119 int offset = nr & BITS_PER_PAGE_MASK;
120
121 clear_bit(offset, map->page);
122 atomic_inc(&map->nr_free);
123}
124
125/*
126 * If we started walking pids at 'base', is 'a' seen before 'b'?
127 */
128static int pid_before(int base, int a, int b)
129{
130 /*
131 * This is the same as saying
132 *
133 * (a - base + MAXUINT) % MAXUINT < (b - base + MAXUINT) % MAXUINT
134 * and that mapping orders 'a' and 'b' with respect to 'base'.
135 */
136 return (unsigned)(a - base) < (unsigned)(b - base);
137}
138
139/*
140 * We might be racing with someone else trying to set pid_ns->last_pid
141 * at the pid allocation time (there's also a sysctl for this, but racing
142 * with this one is OK, see comment in kernel/pid_namespace.c about it).
143 * We want the winner to have the "later" value, because if the
144 * "earlier" value prevails, then a pid may get reused immediately.
145 *
146 * Since pids rollover, it is not sufficient to just pick the bigger
147 * value. We have to consider where we started counting from.
148 *
149 * 'base' is the value of pid_ns->last_pid that we observed when
150 * we started looking for a pid.
151 *
152 * 'pid' is the pid that we eventually found.
153 */
154static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
155{
156 int prev;
157 int last_write = base;
158 do {
159 prev = last_write;
160 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
161 } while ((prev != last_write) && (pid_before(base, last_write, pid)));
162}
163
164static int alloc_pidmap(struct pid_namespace *pid_ns)
165{
166 int i, offset, max_scan, pid, last = pid_ns->last_pid;
167 struct pidmap *map;
168
169 pid = last + 1;
170 if (pid >= pid_max)
171 pid = RESERVED_PIDS;
172 offset = pid & BITS_PER_PAGE_MASK;
173 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
174 /*
175 * If last_pid points into the middle of the map->page we
176 * want to scan this bitmap block twice, the second time
177 * we start with offset == 0 (or RESERVED_PIDS).
178 */
179 max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
180 for (i = 0; i <= max_scan; ++i) {
181 if (unlikely(!map->page)) {
182 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
183 /*
184 * Free the page if someone raced with us
185 * installing it:
186 */
187 spin_lock_irq(&pidmap_lock);
188 if (!map->page) {
189 map->page = page;
190 page = NULL;
191 }
192 spin_unlock_irq(&pidmap_lock);
193 kfree(page);
194 if (unlikely(!map->page))
195 break;
196 }
197 if (likely(atomic_read(&map->nr_free))) {
198 do {
199 if (!test_and_set_bit(offset, map->page)) {
200 atomic_dec(&map->nr_free);
201 set_last_pid(pid_ns, last, pid);
202 return pid;
203 }
204 offset = find_next_offset(map, offset);
205 pid = mk_pid(pid_ns, map, offset);
206 } while (offset < BITS_PER_PAGE && pid < pid_max);
207 }
208 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
209 ++map;
210 offset = 0;
211 } else {
212 map = &pid_ns->pidmap[0];
213 offset = RESERVED_PIDS;
214 if (unlikely(last == offset))
215 break;
216 }
217 pid = mk_pid(pid_ns, map, offset);
218 }
219 return -1;
220}
221
222int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
223{
224 int offset;
225 struct pidmap *map, *end;
226
227 if (last >= PID_MAX_LIMIT)
228 return -1;
229
230 offset = (last + 1) & BITS_PER_PAGE_MASK;
231 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
232 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
233 for (; map < end; map++, offset = 0) {
234 if (unlikely(!map->page))
235 continue;
236 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
237 if (offset < BITS_PER_PAGE)
238 return mk_pid(pid_ns, map, offset);
239 }
240 return -1;
241}
242
243void put_pid(struct pid *pid)
244{
245 struct pid_namespace *ns;
246
247 if (!pid)
248 return;
249
250 ns = pid->numbers[pid->level].ns;
251 if ((atomic_read(&pid->count) == 1) ||
252 atomic_dec_and_test(&pid->count)) {
253 kmem_cache_free(ns->pid_cachep, pid);
254 put_pid_ns(ns);
255 }
256}
257EXPORT_SYMBOL_GPL(put_pid);
258
259static void delayed_put_pid(struct rcu_head *rhp)
260{
261 struct pid *pid = container_of(rhp, struct pid, rcu);
262 put_pid(pid);
263}
264
265void free_pid(struct pid *pid)
266{
267 /* We can be called with write_lock_irq(&tasklist_lock) held */
268 int i;
269 unsigned long flags;
270
271 spin_lock_irqsave(&pidmap_lock, flags);
272 for (i = 0; i <= pid->level; i++)
273 hlist_del_rcu(&pid->numbers[i].pid_chain);
274 spin_unlock_irqrestore(&pidmap_lock, flags);
275
276 for (i = 0; i <= pid->level; i++)
277 free_pidmap(pid->numbers + i);
278
279 call_rcu(&pid->rcu, delayed_put_pid);
280}
281
282struct pid *alloc_pid(struct pid_namespace *ns)
283{
284 struct pid *pid;
285 enum pid_type type;
286 int i, nr;
287 struct pid_namespace *tmp;
288 struct upid *upid;
289
290 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
291 if (!pid)
292 goto out;
293
294 tmp = ns;
295 for (i = ns->level; i >= 0; i--) {
296 nr = alloc_pidmap(tmp);
297 if (nr < 0)
298 goto out_free;
299
300 pid->numbers[i].nr = nr;
301 pid->numbers[i].ns = tmp;
302 tmp = tmp->parent;
303 }
304
305 get_pid_ns(ns);
306 pid->level = ns->level;
307 atomic_set(&pid->count, 1);
308 for (type = 0; type < PIDTYPE_MAX; ++type)
309 INIT_HLIST_HEAD(&pid->tasks[type]);
310
311 upid = pid->numbers + ns->level;
312 spin_lock_irq(&pidmap_lock);
313 for ( ; upid >= pid->numbers; --upid)
314 hlist_add_head_rcu(&upid->pid_chain,
315 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
316 spin_unlock_irq(&pidmap_lock);
317
318out:
319 return pid;
320
321out_free:
322 while (++i <= ns->level)
323 free_pidmap(pid->numbers + i);
324
325 kmem_cache_free(ns->pid_cachep, pid);
326 pid = NULL;
327 goto out;
328}
329
330struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
331{
332 struct hlist_node *elem;
333 struct upid *pnr;
334
335 hlist_for_each_entry_rcu(pnr, elem,
336 &pid_hash[pid_hashfn(nr, ns)], pid_chain)
337 if (pnr->nr == nr && pnr->ns == ns)
338 return container_of(pnr, struct pid,
339 numbers[ns->level]);
340
341 return NULL;
342}
343EXPORT_SYMBOL_GPL(find_pid_ns);
344
345struct pid *find_vpid(int nr)
346{
347 return find_pid_ns(nr, current->nsproxy->pid_ns);
348}
349EXPORT_SYMBOL_GPL(find_vpid);
350
351/*
352 * attach_pid() must be called with the tasklist_lock write-held.
353 */
354void attach_pid(struct task_struct *task, enum pid_type type,
355 struct pid *pid)
356{
357 struct pid_link *link;
358
359 link = &task->pids[type];
360 link->pid = pid;
361 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
362}
363
364static void __change_pid(struct task_struct *task, enum pid_type type,
365 struct pid *new)
366{
367 struct pid_link *link;
368 struct pid *pid;
369 int tmp;
370
371 link = &task->pids[type];
372 pid = link->pid;
373
374 hlist_del_rcu(&link->node);
375 link->pid = new;
376
377 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
378 if (!hlist_empty(&pid->tasks[tmp]))
379 return;
380
381 free_pid(pid);
382}
383
384void detach_pid(struct task_struct *task, enum pid_type type)
385{
386 __change_pid(task, type, NULL);
387}
388
389void change_pid(struct task_struct *task, enum pid_type type,
390 struct pid *pid)
391{
392 __change_pid(task, type, pid);
393 attach_pid(task, type, pid);
394}
395
396/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
397void transfer_pid(struct task_struct *old, struct task_struct *new,
398 enum pid_type type)
399{
400 new->pids[type].pid = old->pids[type].pid;
401 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
402}
403
404struct task_struct *pid_task(struct pid *pid, enum pid_type type)
405{
406 struct task_struct *result = NULL;
407 if (pid) {
408 struct hlist_node *first;
409 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
410 lockdep_tasklist_lock_is_held());
411 if (first)
412 result = hlist_entry(first, struct task_struct, pids[(type)].node);
413 }
414 return result;
415}
416EXPORT_SYMBOL(pid_task);
417
418/*
419 * Must be called under rcu_read_lock().
420 */
421struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
422{
423 rcu_lockdep_assert(rcu_read_lock_held(),
424 "find_task_by_pid_ns() needs rcu_read_lock()"
425 " protection");
426 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
427}
428
429struct task_struct *find_task_by_vpid(pid_t vnr)
430{
431 return find_task_by_pid_ns(vnr, current->nsproxy->pid_ns);
432}
433
434struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
435{
436 struct pid *pid;
437 rcu_read_lock();
438 if (type != PIDTYPE_PID)
439 task = task->group_leader;
440 pid = get_pid(task->pids[type].pid);
441 rcu_read_unlock();
442 return pid;
443}
444EXPORT_SYMBOL_GPL(get_task_pid);
445
446struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
447{
448 struct task_struct *result;
449 rcu_read_lock();
450 result = pid_task(pid, type);
451 if (result)
452 get_task_struct(result);
453 rcu_read_unlock();
454 return result;
455}
456EXPORT_SYMBOL_GPL(get_pid_task);
457
458struct pid *find_get_pid(pid_t nr)
459{
460 struct pid *pid;
461
462 rcu_read_lock();
463 pid = get_pid(find_vpid(nr));
464 rcu_read_unlock();
465
466 return pid;
467}
468EXPORT_SYMBOL_GPL(find_get_pid);
469
470pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
471{
472 struct upid *upid;
473 pid_t nr = 0;
474
475 if (pid && ns->level <= pid->level) {
476 upid = &pid->numbers[ns->level];
477 if (upid->ns == ns)
478 nr = upid->nr;
479 }
480 return nr;
481}
482
483pid_t pid_vnr(struct pid *pid)
484{
485 return pid_nr_ns(pid, current->nsproxy->pid_ns);
486}
487EXPORT_SYMBOL_GPL(pid_vnr);
488
489pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
490 struct pid_namespace *ns)
491{
492 pid_t nr = 0;
493
494 rcu_read_lock();
495 if (!ns)
496 ns = current->nsproxy->pid_ns;
497 if (likely(pid_alive(task))) {
498 if (type != PIDTYPE_PID)
499 task = task->group_leader;
500 nr = pid_nr_ns(task->pids[type].pid, ns);
501 }
502 rcu_read_unlock();
503
504 return nr;
505}
506EXPORT_SYMBOL(__task_pid_nr_ns);
507
508pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
509{
510 return pid_nr_ns(task_tgid(tsk), ns);
511}
512EXPORT_SYMBOL(task_tgid_nr_ns);
513
514struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
515{
516 return ns_of_pid(task_pid(tsk));
517}
518EXPORT_SYMBOL_GPL(task_active_pid_ns);
519
520/*
521 * Used by proc to find the first pid that is greater than or equal to nr.
522 *
523 * If there is a pid at nr this function is exactly the same as find_pid_ns.
524 */
525struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
526{
527 struct pid *pid;
528
529 do {
530 pid = find_pid_ns(nr, ns);
531 if (pid)
532 break;
533 nr = next_pidmap(ns, nr);
534 } while (nr > 0);
535
536 return pid;
537}
538
539/*
540 * The pid hash table is scaled according to the amount of memory in the
541 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
542 * more.
543 */
544void __init pidhash_init(void)
545{
546 unsigned int i, pidhash_size;
547
548 pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
549 HASH_EARLY | HASH_SMALL,
550 &pidhash_shift, NULL,
551 0, 4096);
552 pidhash_size = 1U << pidhash_shift;
553
554 for (i = 0; i < pidhash_size; i++)
555 INIT_HLIST_HEAD(&pid_hash[i]);
556}
557
558void __init pidmap_init(void)
559{
560 /* bump default and minimum pid_max based on number of cpus */
561 pid_max = min(pid_max_max, max_t(int, pid_max,
562 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
563 pid_max_min = max_t(int, pid_max_min,
564 PIDS_PER_CPU_MIN * num_possible_cpus());
565 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
566
567 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
568 /* Reserve PID 0. We never call free_pidmap(0) */
569 set_bit(0, init_pid_ns.pidmap[0].page);
570 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
571
572 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
573 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
574}
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
48struct 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
62int pid_max = PID_MAX_DEFAULT;
63
64#define RESERVED_PIDS 300
65
66int pid_max_min = RESERVED_PIDS + 1;
67int 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 */
75struct 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#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
87 .memfd_noexec_scope = MEMFD_NOEXEC_SCOPE_EXEC,
88#endif
89};
90EXPORT_SYMBOL_GPL(init_pid_ns);
91
92/*
93 * Note: disable interrupts while the pidmap_lock is held as an
94 * interrupt might come in and do read_lock(&tasklist_lock).
95 *
96 * If we don't disable interrupts there is a nasty deadlock between
97 * detach_pid()->free_pid() and another cpu that does
98 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
99 * read_lock(&tasklist_lock);
100 *
101 * After we clean up the tasklist_lock and know there are no
102 * irq handlers that take it we can leave the interrupts enabled.
103 * For now it is easier to be safe than to prove it can't happen.
104 */
105
106static __cacheline_aligned_in_smp DEFINE_SPINLOCK(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 spin_unlock_irqrestore(&pidmap_lock, flags);
158
159 call_rcu(&pid->rcu, delayed_put_pid);
160}
161
162struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
163 size_t set_tid_size)
164{
165 struct pid *pid;
166 enum pid_type type;
167 int i, nr;
168 struct pid_namespace *tmp;
169 struct upid *upid;
170 int retval = -ENOMEM;
171
172 /*
173 * set_tid_size contains the size of the set_tid array. Starting at
174 * the most nested currently active PID namespace it tells alloc_pid()
175 * which PID to set for a process in that most nested PID namespace
176 * up to set_tid_size PID namespaces. It does not have to set the PID
177 * for a process in all nested PID namespaces but set_tid_size must
178 * never be greater than the current ns->level + 1.
179 */
180 if (set_tid_size > ns->level + 1)
181 return ERR_PTR(-EINVAL);
182
183 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
184 if (!pid)
185 return ERR_PTR(retval);
186
187 tmp = ns;
188 pid->level = ns->level;
189
190 for (i = ns->level; i >= 0; i--) {
191 int tid = 0;
192
193 if (set_tid_size) {
194 tid = set_tid[ns->level - i];
195
196 retval = -EINVAL;
197 if (tid < 1 || tid >= pid_max)
198 goto out_free;
199 /*
200 * Also fail if a PID != 1 is requested and
201 * no PID 1 exists.
202 */
203 if (tid != 1 && !tmp->child_reaper)
204 goto out_free;
205 retval = -EPERM;
206 if (!checkpoint_restore_ns_capable(tmp->user_ns))
207 goto out_free;
208 set_tid_size--;
209 }
210
211 idr_preload(GFP_KERNEL);
212 spin_lock_irq(&pidmap_lock);
213
214 if (tid) {
215 nr = idr_alloc(&tmp->idr, NULL, tid,
216 tid + 1, GFP_ATOMIC);
217 /*
218 * If ENOSPC is returned it means that the PID is
219 * alreay in use. Return EEXIST in that case.
220 */
221 if (nr == -ENOSPC)
222 nr = -EEXIST;
223 } else {
224 int pid_min = 1;
225 /*
226 * init really needs pid 1, but after reaching the
227 * maximum wrap back to RESERVED_PIDS
228 */
229 if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS)
230 pid_min = RESERVED_PIDS;
231
232 /*
233 * Store a null pointer so find_pid_ns does not find
234 * a partially initialized PID (see below).
235 */
236 nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min,
237 pid_max, GFP_ATOMIC);
238 }
239 spin_unlock_irq(&pidmap_lock);
240 idr_preload_end();
241
242 if (nr < 0) {
243 retval = (nr == -ENOSPC) ? -EAGAIN : nr;
244 goto out_free;
245 }
246
247 pid->numbers[i].nr = nr;
248 pid->numbers[i].ns = tmp;
249 tmp = tmp->parent;
250 }
251
252 /*
253 * ENOMEM is not the most obvious choice especially for the case
254 * where the child subreaper has already exited and the pid
255 * namespace denies the creation of any new processes. But ENOMEM
256 * is what we have exposed to userspace for a long time and it is
257 * documented behavior for pid namespaces. So we can't easily
258 * change it even if there were an error code better suited.
259 */
260 retval = -ENOMEM;
261
262 get_pid_ns(ns);
263 refcount_set(&pid->count, 1);
264 spin_lock_init(&pid->lock);
265 for (type = 0; type < PIDTYPE_MAX; ++type)
266 INIT_HLIST_HEAD(&pid->tasks[type]);
267
268 init_waitqueue_head(&pid->wait_pidfd);
269 INIT_HLIST_HEAD(&pid->inodes);
270
271 upid = pid->numbers + ns->level;
272 spin_lock_irq(&pidmap_lock);
273 if (!(ns->pid_allocated & PIDNS_ADDING))
274 goto out_unlock;
275 for ( ; upid >= pid->numbers; --upid) {
276 /* Make the PID visible to find_pid_ns. */
277 idr_replace(&upid->ns->idr, pid, upid->nr);
278 upid->ns->pid_allocated++;
279 }
280 spin_unlock_irq(&pidmap_lock);
281
282 return pid;
283
284out_unlock:
285 spin_unlock_irq(&pidmap_lock);
286 put_pid_ns(ns);
287
288out_free:
289 spin_lock_irq(&pidmap_lock);
290 while (++i <= ns->level) {
291 upid = pid->numbers + i;
292 idr_remove(&upid->ns->idr, upid->nr);
293 }
294
295 /* On failure to allocate the first pid, reset the state */
296 if (ns->pid_allocated == PIDNS_ADDING)
297 idr_set_cursor(&ns->idr, 0);
298
299 spin_unlock_irq(&pidmap_lock);
300
301 kmem_cache_free(ns->pid_cachep, pid);
302 return ERR_PTR(retval);
303}
304
305void disable_pid_allocation(struct pid_namespace *ns)
306{
307 spin_lock_irq(&pidmap_lock);
308 ns->pid_allocated &= ~PIDNS_ADDING;
309 spin_unlock_irq(&pidmap_lock);
310}
311
312struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
313{
314 return idr_find(&ns->idr, nr);
315}
316EXPORT_SYMBOL_GPL(find_pid_ns);
317
318struct pid *find_vpid(int nr)
319{
320 return find_pid_ns(nr, task_active_pid_ns(current));
321}
322EXPORT_SYMBOL_GPL(find_vpid);
323
324static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type)
325{
326 return (type == PIDTYPE_PID) ?
327 &task->thread_pid :
328 &task->signal->pids[type];
329}
330
331/*
332 * attach_pid() must be called with the tasklist_lock write-held.
333 */
334void attach_pid(struct task_struct *task, enum pid_type type)
335{
336 struct pid *pid = *task_pid_ptr(task, type);
337 hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]);
338}
339
340static void __change_pid(struct task_struct *task, enum pid_type type,
341 struct pid *new)
342{
343 struct pid **pid_ptr = task_pid_ptr(task, type);
344 struct pid *pid;
345 int tmp;
346
347 pid = *pid_ptr;
348
349 hlist_del_rcu(&task->pid_links[type]);
350 *pid_ptr = new;
351
352 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
353 if (pid_has_task(pid, tmp))
354 return;
355
356 free_pid(pid);
357}
358
359void detach_pid(struct task_struct *task, enum pid_type type)
360{
361 __change_pid(task, type, NULL);
362}
363
364void change_pid(struct task_struct *task, enum pid_type type,
365 struct pid *pid)
366{
367 __change_pid(task, type, pid);
368 attach_pid(task, type);
369}
370
371void exchange_tids(struct task_struct *left, struct task_struct *right)
372{
373 struct pid *pid1 = left->thread_pid;
374 struct pid *pid2 = right->thread_pid;
375 struct hlist_head *head1 = &pid1->tasks[PIDTYPE_PID];
376 struct hlist_head *head2 = &pid2->tasks[PIDTYPE_PID];
377
378 /* Swap the single entry tid lists */
379 hlists_swap_heads_rcu(head1, head2);
380
381 /* Swap the per task_struct pid */
382 rcu_assign_pointer(left->thread_pid, pid2);
383 rcu_assign_pointer(right->thread_pid, pid1);
384
385 /* Swap the cached value */
386 WRITE_ONCE(left->pid, pid_nr(pid2));
387 WRITE_ONCE(right->pid, pid_nr(pid1));
388}
389
390/* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
391void transfer_pid(struct task_struct *old, struct task_struct *new,
392 enum pid_type type)
393{
394 if (type == PIDTYPE_PID)
395 new->thread_pid = old->thread_pid;
396 hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]);
397}
398
399struct task_struct *pid_task(struct pid *pid, enum pid_type type)
400{
401 struct task_struct *result = NULL;
402 if (pid) {
403 struct hlist_node *first;
404 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
405 lockdep_tasklist_lock_is_held());
406 if (first)
407 result = hlist_entry(first, struct task_struct, pid_links[(type)]);
408 }
409 return result;
410}
411EXPORT_SYMBOL(pid_task);
412
413/*
414 * Must be called under rcu_read_lock().
415 */
416struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
417{
418 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
419 "find_task_by_pid_ns() needs rcu_read_lock() protection");
420 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
421}
422
423struct task_struct *find_task_by_vpid(pid_t vnr)
424{
425 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
426}
427
428struct task_struct *find_get_task_by_vpid(pid_t nr)
429{
430 struct task_struct *task;
431
432 rcu_read_lock();
433 task = find_task_by_vpid(nr);
434 if (task)
435 get_task_struct(task);
436 rcu_read_unlock();
437
438 return task;
439}
440
441struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
442{
443 struct pid *pid;
444 rcu_read_lock();
445 pid = get_pid(rcu_dereference(*task_pid_ptr(task, type)));
446 rcu_read_unlock();
447 return pid;
448}
449EXPORT_SYMBOL_GPL(get_task_pid);
450
451struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
452{
453 struct task_struct *result;
454 rcu_read_lock();
455 result = pid_task(pid, type);
456 if (result)
457 get_task_struct(result);
458 rcu_read_unlock();
459 return result;
460}
461EXPORT_SYMBOL_GPL(get_pid_task);
462
463struct pid *find_get_pid(pid_t nr)
464{
465 struct pid *pid;
466
467 rcu_read_lock();
468 pid = get_pid(find_vpid(nr));
469 rcu_read_unlock();
470
471 return pid;
472}
473EXPORT_SYMBOL_GPL(find_get_pid);
474
475pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
476{
477 struct upid *upid;
478 pid_t nr = 0;
479
480 if (pid && ns->level <= pid->level) {
481 upid = &pid->numbers[ns->level];
482 if (upid->ns == ns)
483 nr = upid->nr;
484 }
485 return nr;
486}
487EXPORT_SYMBOL_GPL(pid_nr_ns);
488
489pid_t pid_vnr(struct pid *pid)
490{
491 return pid_nr_ns(pid, task_active_pid_ns(current));
492}
493EXPORT_SYMBOL_GPL(pid_vnr);
494
495pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
496 struct pid_namespace *ns)
497{
498 pid_t nr = 0;
499
500 rcu_read_lock();
501 if (!ns)
502 ns = task_active_pid_ns(current);
503 nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
504 rcu_read_unlock();
505
506 return nr;
507}
508EXPORT_SYMBOL(__task_pid_nr_ns);
509
510struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
511{
512 return ns_of_pid(task_pid(tsk));
513}
514EXPORT_SYMBOL_GPL(task_active_pid_ns);
515
516/*
517 * Used by proc to find the first pid that is greater than or equal to nr.
518 *
519 * If there is a pid at nr this function is exactly the same as find_pid_ns.
520 */
521struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
522{
523 return idr_get_next(&ns->idr, &nr);
524}
525EXPORT_SYMBOL_GPL(find_ge_pid);
526
527struct pid *pidfd_get_pid(unsigned int fd, unsigned int *flags)
528{
529 struct fd f;
530 struct pid *pid;
531
532 f = fdget(fd);
533 if (!f.file)
534 return ERR_PTR(-EBADF);
535
536 pid = pidfd_pid(f.file);
537 if (!IS_ERR(pid)) {
538 get_pid(pid);
539 *flags = f.file->f_flags;
540 }
541
542 fdput(f);
543 return pid;
544}
545
546/**
547 * pidfd_get_task() - Get the task associated with a pidfd
548 *
549 * @pidfd: pidfd for which to get the task
550 * @flags: flags associated with this pidfd
551 *
552 * Return the task associated with @pidfd. The function takes a reference on
553 * the returned task. The caller is responsible for releasing that reference.
554 *
555 * Currently, the process identified by @pidfd is always a thread-group leader.
556 * This restriction currently exists for all aspects of pidfds including pidfd
557 * creation (CLONE_PIDFD cannot be used with CLONE_THREAD) and pidfd polling
558 * (only supports thread group leaders).
559 *
560 * Return: On success, the task_struct associated with the pidfd.
561 * On error, a negative errno number will be returned.
562 */
563struct task_struct *pidfd_get_task(int pidfd, unsigned int *flags)
564{
565 unsigned int f_flags;
566 struct pid *pid;
567 struct task_struct *task;
568
569 pid = pidfd_get_pid(pidfd, &f_flags);
570 if (IS_ERR(pid))
571 return ERR_CAST(pid);
572
573 task = get_pid_task(pid, PIDTYPE_TGID);
574 put_pid(pid);
575 if (!task)
576 return ERR_PTR(-ESRCH);
577
578 *flags = f_flags;
579 return task;
580}
581
582/**
583 * pidfd_create() - Create a new pid file descriptor.
584 *
585 * @pid: struct pid that the pidfd will reference
586 * @flags: flags to pass
587 *
588 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
589 *
590 * Note, that this function can only be called after the fd table has
591 * been unshared to avoid leaking the pidfd to the new process.
592 *
593 * This symbol should not be explicitly exported to loadable modules.
594 *
595 * Return: On success, a cloexec pidfd is returned.
596 * On error, a negative errno number will be returned.
597 */
598int pidfd_create(struct pid *pid, unsigned int flags)
599{
600 int pidfd;
601 struct file *pidfd_file;
602
603 pidfd = pidfd_prepare(pid, flags, &pidfd_file);
604 if (pidfd < 0)
605 return pidfd;
606
607 fd_install(pidfd, pidfd_file);
608 return pidfd;
609}
610
611/**
612 * sys_pidfd_open() - Open new pid file descriptor.
613 *
614 * @pid: pid for which to retrieve a pidfd
615 * @flags: flags to pass
616 *
617 * This creates a new pid file descriptor with the O_CLOEXEC flag set for
618 * the process identified by @pid. Currently, the process identified by
619 * @pid must be a thread-group leader. This restriction currently exists
620 * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
621 * be used with CLONE_THREAD) and pidfd polling (only supports thread group
622 * leaders).
623 *
624 * Return: On success, a cloexec pidfd is returned.
625 * On error, a negative errno number will be returned.
626 */
627SYSCALL_DEFINE2(pidfd_open, pid_t, pid, unsigned int, flags)
628{
629 int fd;
630 struct pid *p;
631
632 if (flags & ~PIDFD_NONBLOCK)
633 return -EINVAL;
634
635 if (pid <= 0)
636 return -EINVAL;
637
638 p = find_get_pid(pid);
639 if (!p)
640 return -ESRCH;
641
642 fd = pidfd_create(p, flags);
643
644 put_pid(p);
645 return fd;
646}
647
648void __init pid_idr_init(void)
649{
650 /* Verify no one has done anything silly: */
651 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING);
652
653 /* bump default and minimum pid_max based on number of cpus */
654 pid_max = min(pid_max_max, max_t(int, pid_max,
655 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
656 pid_max_min = max_t(int, pid_max_min,
657 PIDS_PER_CPU_MIN * num_possible_cpus());
658 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
659
660 idr_init(&init_pid_ns.idr);
661
662 init_pid_ns.pid_cachep = kmem_cache_create("pid",
663 struct_size_t(struct pid, numbers, 1),
664 __alignof__(struct pid),
665 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT,
666 NULL);
667}
668
669static struct file *__pidfd_fget(struct task_struct *task, int fd)
670{
671 struct file *file;
672 int ret;
673
674 ret = down_read_killable(&task->signal->exec_update_lock);
675 if (ret)
676 return ERR_PTR(ret);
677
678 if (ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS))
679 file = fget_task(task, fd);
680 else
681 file = ERR_PTR(-EPERM);
682
683 up_read(&task->signal->exec_update_lock);
684
685 return file ?: ERR_PTR(-EBADF);
686}
687
688static int pidfd_getfd(struct pid *pid, int fd)
689{
690 struct task_struct *task;
691 struct file *file;
692 int ret;
693
694 task = get_pid_task(pid, PIDTYPE_PID);
695 if (!task)
696 return -ESRCH;
697
698 file = __pidfd_fget(task, fd);
699 put_task_struct(task);
700 if (IS_ERR(file))
701 return PTR_ERR(file);
702
703 ret = receive_fd(file, NULL, O_CLOEXEC);
704 fput(file);
705
706 return ret;
707}
708
709/**
710 * sys_pidfd_getfd() - Get a file descriptor from another process
711 *
712 * @pidfd: the pidfd file descriptor of the process
713 * @fd: the file descriptor number to get
714 * @flags: flags on how to get the fd (reserved)
715 *
716 * This syscall gets a copy of a file descriptor from another process
717 * based on the pidfd, and file descriptor number. It requires that
718 * the calling process has the ability to ptrace the process represented
719 * by the pidfd. The process which is having its file descriptor copied
720 * is otherwise unaffected.
721 *
722 * Return: On success, a cloexec file descriptor is returned.
723 * On error, a negative errno number will be returned.
724 */
725SYSCALL_DEFINE3(pidfd_getfd, int, pidfd, int, fd,
726 unsigned int, flags)
727{
728 struct pid *pid;
729 struct fd f;
730 int ret;
731
732 /* flags is currently unused - make sure it's unset */
733 if (flags)
734 return -EINVAL;
735
736 f = fdget(pidfd);
737 if (!f.file)
738 return -EBADF;
739
740 pid = pidfd_pid(f.file);
741 if (IS_ERR(pid))
742 ret = PTR_ERR(pid);
743 else
744 ret = pidfd_getfd(pid, fd);
745
746 fdput(f);
747 return ret;
748}