1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
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#include <linux/proc_ns.h>
40#include <linux/proc_fs.h>
41
42#define pid_hashfn(nr, ns) \
43 hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
44static struct hlist_head *pid_hash;
45static unsigned int pidhash_shift = 4;
46struct pid init_struct_pid = INIT_STRUCT_PID;
47
48int pid_max = PID_MAX_DEFAULT;
49
50#define RESERVED_PIDS 300
51
52int pid_max_min = RESERVED_PIDS + 1;
53int pid_max_max = PID_MAX_LIMIT;
54
55static inline int mk_pid(struct pid_namespace *pid_ns,
56 struct pidmap *map, int off)
57{
58 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
59}
60
61#define find_next_offset(map, off) \
62 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
63
64
65
66
67
68
69
70struct pid_namespace init_pid_ns = {
71 .kref = {
72 .refcount = ATOMIC_INIT(2),
73 },
74 .pidmap = {
75 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
76 },
77 .last_pid = 0,
78 .level = 0,
79 .child_reaper = &init_task,
80 .user_ns = &init_user_ns,
81 .proc_inum = PROC_PID_INIT_INO,
82};
83EXPORT_SYMBOL_GPL(init_pid_ns);
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
100
101static void free_pidmap(struct upid *upid)
102{
103 int nr = upid->nr;
104 struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
105 int offset = nr & BITS_PER_PAGE_MASK;
106
107 clear_bit(offset, map->page);
108 atomic_inc(&map->nr_free);
109}
110
111
112
113
114static int pid_before(int base, int a, int b)
115{
116
117
118
119
120
121
122 return (unsigned)(a - base) < (unsigned)(b - base);
123}
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
141{
142 int prev;
143 int last_write = base;
144 do {
145 prev = last_write;
146 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
147 } while ((prev != last_write) && (pid_before(base, last_write, pid)));
148}
149
150static int alloc_pidmap(struct pid_namespace *pid_ns)
151{
152 int i, offset, max_scan, pid, last = pid_ns->last_pid;
153 struct pidmap *map;
154
155 pid = last + 1;
156 if (pid >= pid_max)
157 pid = RESERVED_PIDS;
158 offset = pid & BITS_PER_PAGE_MASK;
159 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
160
161
162
163
164
165 max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
166 for (i = 0; i <= max_scan; ++i) {
167 if (unlikely(!map->page)) {
168 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
169
170
171
172
173 spin_lock_irq(&pidmap_lock);
174 if (!map->page) {
175 map->page = page;
176 page = NULL;
177 }
178 spin_unlock_irq(&pidmap_lock);
179 kfree(page);
180 if (unlikely(!map->page))
181 break;
182 }
183 if (likely(atomic_read(&map->nr_free))) {
184 for ( ; ; ) {
185 if (!test_and_set_bit(offset, map->page)) {
186 atomic_dec(&map->nr_free);
187 set_last_pid(pid_ns, last, pid);
188 return pid;
189 }
190 offset = find_next_offset(map, offset);
191 if (offset >= BITS_PER_PAGE)
192 break;
193 pid = mk_pid(pid_ns, map, offset);
194 if (pid >= pid_max)
195 break;
196 }
197 }
198 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
199 ++map;
200 offset = 0;
201 } else {
202 map = &pid_ns->pidmap[0];
203 offset = RESERVED_PIDS;
204 if (unlikely(last == offset))
205 break;
206 }
207 pid = mk_pid(pid_ns, map, offset);
208 }
209 return -1;
210}
211
212int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
213{
214 int offset;
215 struct pidmap *map, *end;
216
217 if (last >= PID_MAX_LIMIT)
218 return -1;
219
220 offset = (last + 1) & BITS_PER_PAGE_MASK;
221 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
222 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
223 for (; map < end; map++, offset = 0) {
224 if (unlikely(!map->page))
225 continue;
226 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
227 if (offset < BITS_PER_PAGE)
228 return mk_pid(pid_ns, map, offset);
229 }
230 return -1;
231}
232
233void put_pid(struct pid *pid)
234{
235 struct pid_namespace *ns;
236
237 if (!pid)
238 return;
239
240 ns = pid->numbers[pid->level].ns;
241 if ((atomic_read(&pid->count) == 1) ||
242 atomic_dec_and_test(&pid->count)) {
243 kmem_cache_free(ns->pid_cachep, pid);
244 put_pid_ns(ns);
245 }
246}
247EXPORT_SYMBOL_GPL(put_pid);
248
249static void delayed_put_pid(struct rcu_head *rhp)
250{
251 struct pid *pid = container_of(rhp, struct pid, rcu);
252 put_pid(pid);
253}
254
255void free_pid(struct pid *pid)
256{
257
258 int i;
259 unsigned long flags;
260
261 spin_lock_irqsave(&pidmap_lock, flags);
262 for (i = 0; i <= pid->level; i++) {
263 struct upid *upid = pid->numbers + i;
264 struct pid_namespace *ns = upid->ns;
265 hlist_del_rcu(&upid->pid_chain);
266 switch(--ns->nr_hashed) {
267 case 1:
268
269
270
271
272 wake_up_process(ns->child_reaper);
273 break;
274 case 0:
275 schedule_work(&ns->proc_work);
276 break;
277 }
278 }
279 spin_unlock_irqrestore(&pidmap_lock, flags);
280
281 for (i = 0; i <= pid->level; i++)
282 free_pidmap(pid->numbers + i);
283
284 call_rcu(&pid->rcu, delayed_put_pid);
285}
286
287struct pid *alloc_pid(struct pid_namespace *ns)
288{
289 struct pid *pid;
290 enum pid_type type;
291 int i, nr;
292 struct pid_namespace *tmp;
293 struct upid *upid;
294
295 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
296 if (!pid)
297 goto out;
298
299 tmp = ns;
300 pid->level = ns->level;
301 for (i = ns->level; i >= 0; i--) {
302 nr = alloc_pidmap(tmp);
303 if (nr < 0)
304 goto out_free;
305
306 pid->numbers[i].nr = nr;
307 pid->numbers[i].ns = tmp;
308 tmp = tmp->parent;
309 }
310
311 if (unlikely(is_child_reaper(pid))) {
312 if (pid_ns_prepare_proc(ns))
313 goto out_free;
314 }
315
316 get_pid_ns(ns);
317 atomic_set(&pid->count, 1);
318 for (type = 0; type < PIDTYPE_MAX; ++type)
319 INIT_HLIST_HEAD(&pid->tasks[type]);
320
321 upid = pid->numbers + ns->level;
322 spin_lock_irq(&pidmap_lock);
323 if (!(ns->nr_hashed & PIDNS_HASH_ADDING))
324 goto out_unlock;
325 for ( ; upid >= pid->numbers; --upid) {
326 hlist_add_head_rcu(&upid->pid_chain,
327 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
328 upid->ns->nr_hashed++;
329 }
330 spin_unlock_irq(&pidmap_lock);
331
332out:
333 return pid;
334
335out_unlock:
336 spin_unlock_irq(&pidmap_lock);
337out_free:
338 while (++i <= ns->level)
339 free_pidmap(pid->numbers + i);
340
341 kmem_cache_free(ns->pid_cachep, pid);
342 pid = NULL;
343 goto out;
344}
345
346void disable_pid_allocation(struct pid_namespace *ns)
347{
348 spin_lock_irq(&pidmap_lock);
349 ns->nr_hashed &= ~PIDNS_HASH_ADDING;
350 spin_unlock_irq(&pidmap_lock);
351}
352
353struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
354{
355 struct upid *pnr;
356
357 hlist_for_each_entry_rcu(pnr,
358 &pid_hash[pid_hashfn(nr, ns)], pid_chain)
359 if (pnr->nr == nr && pnr->ns == ns)
360 return container_of(pnr, struct pid,
361 numbers[ns->level]);
362
363 return NULL;
364}
365EXPORT_SYMBOL_GPL(find_pid_ns);
366
367struct pid *find_vpid(int nr)
368{
369 return find_pid_ns(nr, task_active_pid_ns(current));
370}
371EXPORT_SYMBOL_GPL(find_vpid);
372
373
374
375
376void attach_pid(struct task_struct *task, enum pid_type type,
377 struct pid *pid)
378{
379 struct pid_link *link;
380
381 link = &task->pids[type];
382 link->pid = pid;
383 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
384}
385
386static void __change_pid(struct task_struct *task, enum pid_type type,
387 struct pid *new)
388{
389 struct pid_link *link;
390 struct pid *pid;
391 int tmp;
392
393 link = &task->pids[type];
394 pid = link->pid;
395
396 hlist_del_rcu(&link->node);
397 link->pid = new;
398
399 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
400 if (!hlist_empty(&pid->tasks[tmp]))
401 return;
402
403 free_pid(pid);
404}
405
406void detach_pid(struct task_struct *task, enum pid_type type)
407{
408 __change_pid(task, type, NULL);
409}
410
411void change_pid(struct task_struct *task, enum pid_type type,
412 struct pid *pid)
413{
414 __change_pid(task, type, pid);
415 attach_pid(task, type, pid);
416}
417
418
419void transfer_pid(struct task_struct *old, struct task_struct *new,
420 enum pid_type type)
421{
422 new->pids[type].pid = old->pids[type].pid;
423 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
424}
425
426struct task_struct *pid_task(struct pid *pid, enum pid_type type)
427{
428 struct task_struct *result = NULL;
429 if (pid) {
430 struct hlist_node *first;
431 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
432 lockdep_tasklist_lock_is_held());
433 if (first)
434 result = hlist_entry(first, struct task_struct, pids[(type)].node);
435 }
436 return result;
437}
438EXPORT_SYMBOL(pid_task);
439
440
441
442
443struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
444{
445 rcu_lockdep_assert(rcu_read_lock_held(),
446 "find_task_by_pid_ns() needs rcu_read_lock()"
447 " protection");
448 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
449}
450
451struct task_struct *find_task_by_vpid(pid_t vnr)
452{
453 return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
454}
455
456struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
457{
458 struct pid *pid;
459 rcu_read_lock();
460 if (type != PIDTYPE_PID)
461 task = task->group_leader;
462 pid = get_pid(task->pids[type].pid);
463 rcu_read_unlock();
464 return pid;
465}
466EXPORT_SYMBOL_GPL(get_task_pid);
467
468struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
469{
470 struct task_struct *result;
471 rcu_read_lock();
472 result = pid_task(pid, type);
473 if (result)
474 get_task_struct(result);
475 rcu_read_unlock();
476 return result;
477}
478EXPORT_SYMBOL_GPL(get_pid_task);
479
480struct pid *find_get_pid(pid_t nr)
481{
482 struct pid *pid;
483
484 rcu_read_lock();
485 pid = get_pid(find_vpid(nr));
486 rcu_read_unlock();
487
488 return pid;
489}
490EXPORT_SYMBOL_GPL(find_get_pid);
491
492pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
493{
494 struct upid *upid;
495 pid_t nr = 0;
496
497 if (pid && ns->level <= pid->level) {
498 upid = &pid->numbers[ns->level];
499 if (upid->ns == ns)
500 nr = upid->nr;
501 }
502 return nr;
503}
504EXPORT_SYMBOL_GPL(pid_nr_ns);
505
506pid_t pid_vnr(struct pid *pid)
507{
508 return pid_nr_ns(pid, task_active_pid_ns(current));
509}
510EXPORT_SYMBOL_GPL(pid_vnr);
511
512pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
513 struct pid_namespace *ns)
514{
515 pid_t nr = 0;
516
517 rcu_read_lock();
518 if (!ns)
519 ns = task_active_pid_ns(current);
520 if (likely(pid_alive(task))) {
521 if (type != PIDTYPE_PID)
522 task = task->group_leader;
523 nr = pid_nr_ns(task->pids[type].pid, ns);
524 }
525 rcu_read_unlock();
526
527 return nr;
528}
529EXPORT_SYMBOL(__task_pid_nr_ns);
530
531pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
532{
533 return pid_nr_ns(task_tgid(tsk), ns);
534}
535EXPORT_SYMBOL(task_tgid_nr_ns);
536
537struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
538{
539 return ns_of_pid(task_pid(tsk));
540}
541EXPORT_SYMBOL_GPL(task_active_pid_ns);
542
543
544
545
546
547
548struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
549{
550 struct pid *pid;
551
552 do {
553 pid = find_pid_ns(nr, ns);
554 if (pid)
555 break;
556 nr = next_pidmap(ns, nr);
557 } while (nr > 0);
558
559 return pid;
560}
561
562
563
564
565
566
567void __init pidhash_init(void)
568{
569 unsigned int i, pidhash_size;
570
571 pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
572 HASH_EARLY | HASH_SMALL,
573 &pidhash_shift, NULL,
574 0, 4096);
575 pidhash_size = 1U << pidhash_shift;
576
577 for (i = 0; i < pidhash_size; i++)
578 INIT_HLIST_HEAD(&pid_hash[i]);
579}
580
581void __init pidmap_init(void)
582{
583
584 BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_HASH_ADDING);
585
586
587 pid_max = min(pid_max_max, max_t(int, pid_max,
588 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
589 pid_max_min = max_t(int, pid_max_min,
590 PIDS_PER_CPU_MIN * num_possible_cpus());
591 pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
592
593 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
594
595 set_bit(0, init_pid_ns.pidmap[0].page);
596 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
597 init_pid_ns.nr_hashed = PIDNS_HASH_ADDING;
598
599 init_pid_ns.pid_cachep = KMEM_CACHE(pid,
600 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
601}
602