~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/pid.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.72 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Generic pidhash and scalable, time-bounded PID allocator
  3  *
  4  * (C) 2002-2003 Nadia Yvette Chambers, IBM
  5  * (C) 2004 Nadia Yvette Chambers, 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 #include <linux/proc_fs.h>
 40 
 41 #define pid_hashfn(nr, ns)      \
 42         hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
 43 static struct hlist_head *pid_hash;
 44 static unsigned int pidhash_shift = 4;
 45 struct pid init_struct_pid = INIT_STRUCT_PID;
 46 
 47 int pid_max = PID_MAX_DEFAULT;
 48 
 49 #define RESERVED_PIDS           300
 50 
 51 int pid_max_min = RESERVED_PIDS + 1;
 52 int pid_max_max = PID_MAX_LIMIT;
 53 
 54 #define BITS_PER_PAGE           (PAGE_SIZE*8)
 55 #define BITS_PER_PAGE_MASK      (BITS_PER_PAGE-1)
 56 
 57 static inline int mk_pid(struct pid_namespace *pid_ns,
 58                 struct pidmap *map, int off)
 59 {
 60         return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
 61 }
 62 
 63 #define find_next_offset(map, off)                                      \
 64                 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
 65 
 66 /*
 67  * PID-map pages start out as NULL, they get allocated upon
 68  * first use and are never deallocated. This way a low pid_max
 69  * value does not cause lots of bitmaps to be allocated, but
 70  * the scheme scales to up to 4 million PIDs, runtime.
 71  */
 72 struct pid_namespace init_pid_ns = {
 73         .kref = {
 74                 .refcount       = ATOMIC_INIT(2),
 75         },
 76         .pidmap = {
 77                 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
 78         },
 79         .last_pid = 0,
 80         .level = 0,
 81         .child_reaper = &init_task,
 82         .user_ns = &init_user_ns,
 83         .proc_inum = PROC_PID_INIT_INO,
 84 };
 85 EXPORT_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 
101 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
102 
103 static void free_pidmap(struct upid *upid)
104 {
105         int nr = upid->nr;
106         struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
107         int offset = nr & BITS_PER_PAGE_MASK;
108 
109         clear_bit(offset, map->page);
110         atomic_inc(&map->nr_free);
111 }
112 
113 /*
114  * If we started walking pids at 'base', is 'a' seen before 'b'?
115  */
116 static int pid_before(int base, int a, int b)
117 {
118         /*
119          * This is the same as saying
120          *
121          * (a - base + MAXUINT) % MAXUINT < (b - base + MAXUINT) % MAXUINT
122          * and that mapping orders 'a' and 'b' with respect to 'base'.
123          */
124         return (unsigned)(a - base) < (unsigned)(b - base);
125 }
126 
127 /*
128  * We might be racing with someone else trying to set pid_ns->last_pid
129  * at the pid allocation time (there's also a sysctl for this, but racing
130  * with this one is OK, see comment in kernel/pid_namespace.c about it).
131  * We want the winner to have the "later" value, because if the
132  * "earlier" value prevails, then a pid may get reused immediately.
133  *
134  * Since pids rollover, it is not sufficient to just pick the bigger
135  * value.  We have to consider where we started counting from.
136  *
137  * 'base' is the value of pid_ns->last_pid that we observed when
138  * we started looking for a pid.
139  *
140  * 'pid' is the pid that we eventually found.
141  */
142 static void set_last_pid(struct pid_namespace *pid_ns, int base, int pid)
143 {
144         int prev;
145         int last_write = base;
146         do {
147                 prev = last_write;
148                 last_write = cmpxchg(&pid_ns->last_pid, prev, pid);
149         } while ((prev != last_write) && (pid_before(base, last_write, pid)));
150 }
151 
152 static int alloc_pidmap(struct pid_namespace *pid_ns)
153 {
154         int i, offset, max_scan, pid, last = pid_ns->last_pid;
155         struct pidmap *map;
156 
157         pid = last + 1;
158         if (pid >= pid_max)
159                 pid = RESERVED_PIDS;
160         offset = pid & BITS_PER_PAGE_MASK;
161         map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
162         /*
163          * If last_pid points into the middle of the map->page we
164          * want to scan this bitmap block twice, the second time
165          * we start with offset == 0 (or RESERVED_PIDS).
166          */
167         max_scan = DIV_ROUND_UP(pid_max, BITS_PER_PAGE) - !offset;
168         for (i = 0; i <= max_scan; ++i) {
169                 if (unlikely(!map->page)) {
170                         void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
171                         /*
172                          * Free the page if someone raced with us
173                          * installing it:
174                          */
175                         spin_lock_irq(&pidmap_lock);
176                         if (!map->page) {
177                                 map->page = page;
178                                 page = NULL;
179                         }
180                         spin_unlock_irq(&pidmap_lock);
181                         kfree(page);
182                         if (unlikely(!map->page))
183                                 break;
184                 }
185                 if (likely(atomic_read(&map->nr_free))) {
186                         do {
187                                 if (!test_and_set_bit(offset, map->page)) {
188                                         atomic_dec(&map->nr_free);
189                                         set_last_pid(pid_ns, last, pid);
190                                         return pid;
191                                 }
192                                 offset = find_next_offset(map, offset);
193                                 pid = mk_pid(pid_ns, map, offset);
194                         } while (offset < BITS_PER_PAGE && pid < pid_max);
195                 }
196                 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
197                         ++map;
198                         offset = 0;
199                 } else {
200                         map = &pid_ns->pidmap[0];
201                         offset = RESERVED_PIDS;
202                         if (unlikely(last == offset))
203                                 break;
204                 }
205                 pid = mk_pid(pid_ns, map, offset);
206         }
207         return -1;
208 }
209 
210 int next_pidmap(struct pid_namespace *pid_ns, unsigned int last)
211 {
212         int offset;
213         struct pidmap *map, *end;
214 
215         if (last >= PID_MAX_LIMIT)
216                 return -1;
217 
218         offset = (last + 1) & BITS_PER_PAGE_MASK;
219         map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
220         end = &pid_ns->pidmap[PIDMAP_ENTRIES];
221         for (; map < end; map++, offset = 0) {
222                 if (unlikely(!map->page))
223                         continue;
224                 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
225                 if (offset < BITS_PER_PAGE)
226                         return mk_pid(pid_ns, map, offset);
227         }
228         return -1;
229 }
230 
231 void put_pid(struct pid *pid)
232 {
233         struct pid_namespace *ns;
234 
235         if (!pid)
236                 return;
237 
238         ns = pid->numbers[pid->level].ns;
239         if ((atomic_read(&pid->count) == 1) ||
240              atomic_dec_and_test(&pid->count)) {
241                 kmem_cache_free(ns->pid_cachep, pid);
242                 put_pid_ns(ns);
243         }
244 }
245 EXPORT_SYMBOL_GPL(put_pid);
246 
247 static void delayed_put_pid(struct rcu_head *rhp)
248 {
249         struct pid *pid = container_of(rhp, struct pid, rcu);
250         put_pid(pid);
251 }
252 
253 void free_pid(struct pid *pid)
254 {
255         /* We can be called with write_lock_irq(&tasklist_lock) held */
256         int i;
257         unsigned long flags;
258 
259         spin_lock_irqsave(&pidmap_lock, flags);
260         for (i = 0; i <= pid->level; i++) {
261                 struct upid *upid = pid->numbers + i;
262                 struct pid_namespace *ns = upid->ns;
263                 hlist_del_rcu(&upid->pid_chain);
264                 switch(--ns->nr_hashed) {
265                 case 1:
266                         /* When all that is left in the pid namespace
267                          * is the reaper wake up the reaper.  The reaper
268                          * may be sleeping in zap_pid_ns_processes().
269                          */
270                         wake_up_process(ns->child_reaper);
271                         break;
272                 case 0:
273                         schedule_work(&ns->proc_work);
274                         break;
275                 }
276         }
277         spin_unlock_irqrestore(&pidmap_lock, flags);
278 
279         for (i = 0; i <= pid->level; i++)
280                 free_pidmap(pid->numbers + i);
281 
282         call_rcu(&pid->rcu, delayed_put_pid);
283 }
284 
285 struct pid *alloc_pid(struct pid_namespace *ns)
286 {
287         struct pid *pid;
288         enum pid_type type;
289         int i, nr;
290         struct pid_namespace *tmp;
291         struct upid *upid;
292 
293         pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
294         if (!pid)
295                 goto out;
296 
297         tmp = ns;
298         pid->level = ns->level;
299         for (i = ns->level; i >= 0; i--) {
300                 nr = alloc_pidmap(tmp);
301                 if (nr < 0)
302                         goto out_free;
303 
304                 pid->numbers[i].nr = nr;
305                 pid->numbers[i].ns = tmp;
306                 tmp = tmp->parent;
307         }
308 
309         if (unlikely(is_child_reaper(pid))) {
310                 if (pid_ns_prepare_proc(ns))
311                         goto out_free;
312         }
313 
314         get_pid_ns(ns);
315         atomic_set(&pid->count, 1);
316         for (type = 0; type < PIDTYPE_MAX; ++type)
317                 INIT_HLIST_HEAD(&pid->tasks[type]);
318 
319         upid = pid->numbers + ns->level;
320         spin_lock_irq(&pidmap_lock);
321         if (!(ns->nr_hashed & PIDNS_HASH_ADDING))
322                 goto out_unlock;
323         for ( ; upid >= pid->numbers; --upid) {
324                 hlist_add_head_rcu(&upid->pid_chain,
325                                 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
326                 upid->ns->nr_hashed++;
327         }
328         spin_unlock_irq(&pidmap_lock);
329 
330 out:
331         return pid;
332 
333 out_unlock:
334         spin_unlock_irq(&pidmap_lock);
335 out_free:
336         while (++i <= ns->level)
337                 free_pidmap(pid->numbers + i);
338 
339         kmem_cache_free(ns->pid_cachep, pid);
340         pid = NULL;
341         goto out;
342 }
343 
344 void disable_pid_allocation(struct pid_namespace *ns)
345 {
346         spin_lock_irq(&pidmap_lock);
347         ns->nr_hashed &= ~PIDNS_HASH_ADDING;
348         spin_unlock_irq(&pidmap_lock);
349 }
350 
351 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
352 {
353         struct hlist_node *elem;
354         struct upid *pnr;
355 
356         hlist_for_each_entry_rcu(pnr, elem,
357                         &pid_hash[pid_hashfn(nr, ns)], pid_chain)
358                 if (pnr->nr == nr && pnr->ns == ns)
359                         return container_of(pnr, struct pid,
360                                         numbers[ns->level]);
361 
362         return NULL;
363 }
364 EXPORT_SYMBOL_GPL(find_pid_ns);
365 
366 struct pid *find_vpid(int nr)
367 {
368         return find_pid_ns(nr, task_active_pid_ns(current));
369 }
370 EXPORT_SYMBOL_GPL(find_vpid);
371 
372 /*
373  * attach_pid() must be called with the tasklist_lock write-held.
374  */
375 void attach_pid(struct task_struct *task, enum pid_type type,
376                 struct pid *pid)
377 {
378         struct pid_link *link;
379 
380         link = &task->pids[type];
381         link->pid = pid;
382         hlist_add_head_rcu(&link->node, &pid->tasks[type]);
383 }
384 
385 static void __change_pid(struct task_struct *task, enum pid_type type,
386                         struct pid *new)
387 {
388         struct pid_link *link;
389         struct pid *pid;
390         int tmp;
391 
392         link = &task->pids[type];
393         pid = link->pid;
394 
395         hlist_del_rcu(&link->node);
396         link->pid = new;
397 
398         for (tmp = PIDTYPE_MAX; --tmp >= 0; )
399                 if (!hlist_empty(&pid->tasks[tmp]))
400                         return;
401 
402         free_pid(pid);
403 }
404 
405 void detach_pid(struct task_struct *task, enum pid_type type)
406 {
407         __change_pid(task, type, NULL);
408 }
409 
410 void change_pid(struct task_struct *task, enum pid_type type,
411                 struct pid *pid)
412 {
413         __change_pid(task, type, pid);
414         attach_pid(task, type, pid);
415 }
416 
417 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
418 void transfer_pid(struct task_struct *old, struct task_struct *new,
419                            enum pid_type type)
420 {
421         new->pids[type].pid = old->pids[type].pid;
422         hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
423 }
424 
425 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
426 {
427         struct task_struct *result = NULL;
428         if (pid) {
429                 struct hlist_node *first;
430                 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
431                                               lockdep_tasklist_lock_is_held());
432                 if (first)
433                         result = hlist_entry(first, struct task_struct, pids[(type)].node);
434         }
435         return result;
436 }
437 EXPORT_SYMBOL(pid_task);
438 
439 /*
440  * Must be called under rcu_read_lock().
441  */
442 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
443 {
444         rcu_lockdep_assert(rcu_read_lock_held(),
445                            "find_task_by_pid_ns() needs rcu_read_lock()"
446                            " protection");
447         return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
448 }
449 
450 struct task_struct *find_task_by_vpid(pid_t vnr)
451 {
452         return find_task_by_pid_ns(vnr, task_active_pid_ns(current));
453 }
454 
455 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
456 {
457         struct pid *pid;
458         rcu_read_lock();
459         if (type != PIDTYPE_PID)
460                 task = task->group_leader;
461         pid = get_pid(task->pids[type].pid);
462         rcu_read_unlock();
463         return pid;
464 }
465 EXPORT_SYMBOL_GPL(get_task_pid);
466 
467 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
468 {
469         struct task_struct *result;
470         rcu_read_lock();
471         result = pid_task(pid, type);
472         if (result)
473                 get_task_struct(result);
474         rcu_read_unlock();
475         return result;
476 }
477 EXPORT_SYMBOL_GPL(get_pid_task);
478 
479 struct pid *find_get_pid(pid_t nr)
480 {
481         struct pid *pid;
482 
483         rcu_read_lock();
484         pid = get_pid(find_vpid(nr));
485         rcu_read_unlock();
486 
487         return pid;
488 }
489 EXPORT_SYMBOL_GPL(find_get_pid);
490 
491 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
492 {
493         struct upid *upid;
494         pid_t nr = 0;
495 
496         if (pid && ns->level <= pid->level) {
497                 upid = &pid->numbers[ns->level];
498                 if (upid->ns == ns)
499                         nr = upid->nr;
500         }
501         return nr;
502 }
503 EXPORT_SYMBOL_GPL(pid_nr_ns);
504 
505 pid_t pid_vnr(struct pid *pid)
506 {
507         return pid_nr_ns(pid, task_active_pid_ns(current));
508 }
509 EXPORT_SYMBOL_GPL(pid_vnr);
510 
511 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
512                         struct pid_namespace *ns)
513 {
514         pid_t nr = 0;
515 
516         rcu_read_lock();
517         if (!ns)
518                 ns = task_active_pid_ns(current);
519         if (likely(pid_alive(task))) {
520                 if (type != PIDTYPE_PID)
521                         task = task->group_leader;
522                 nr = pid_nr_ns(task->pids[type].pid, ns);
523         }
524         rcu_read_unlock();
525 
526         return nr;
527 }
528 EXPORT_SYMBOL(__task_pid_nr_ns);
529 
530 pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
531 {
532         return pid_nr_ns(task_tgid(tsk), ns);
533 }
534 EXPORT_SYMBOL(task_tgid_nr_ns);
535 
536 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
537 {
538         return ns_of_pid(task_pid(tsk));
539 }
540 EXPORT_SYMBOL_GPL(task_active_pid_ns);
541 
542 /*
543  * Used by proc to find the first pid that is greater than or equal to nr.
544  *
545  * If there is a pid at nr this function is exactly the same as find_pid_ns.
546  */
547 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
548 {
549         struct pid *pid;
550 
551         do {
552                 pid = find_pid_ns(nr, ns);
553                 if (pid)
554                         break;
555                 nr = next_pidmap(ns, nr);
556         } while (nr > 0);
557 
558         return pid;
559 }
560 
561 /*
562  * The pid hash table is scaled according to the amount of memory in the
563  * machine.  From a minimum of 16 slots up to 4096 slots at one gigabyte or
564  * more.
565  */
566 void __init pidhash_init(void)
567 {
568         unsigned int i, pidhash_size;
569 
570         pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
571                                            HASH_EARLY | HASH_SMALL,
572                                            &pidhash_shift, NULL,
573                                            0, 4096);
574         pidhash_size = 1U << pidhash_shift;
575 
576         for (i = 0; i < pidhash_size; i++)
577                 INIT_HLIST_HEAD(&pid_hash[i]);
578 }
579 
580 void __init pidmap_init(void)
581 {
582         /* Veryify no one has done anything silly */
583         BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_HASH_ADDING);
584 
585         /* bump default and minimum pid_max based on number of cpus */
586         pid_max = min(pid_max_max, max_t(int, pid_max,
587                                 PIDS_PER_CPU_DEFAULT * num_possible_cpus()));
588         pid_max_min = max_t(int, pid_max_min,
589                                 PIDS_PER_CPU_MIN * num_possible_cpus());
590         pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min);
591 
592         init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
593         /* Reserve PID 0. We never call free_pidmap(0) */
594         set_bit(0, init_pid_ns.pidmap[0].page);
595         atomic_dec(&init_pid_ns.pidmap[0].nr_free);
596         init_pid_ns.nr_hashed = PIDNS_HASH_ADDING;
597 
598         init_pid_ns.pid_cachep = KMEM_CACHE(pid,
599                         SLAB_HWCACHE_ALIGN | SLAB_PANIC);
600 }
601 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp