1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * taskstats.c - Export per-task statistics to userland
4 *
5 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
6 * (C) Balbir Singh, IBM Corp. 2006
7 */
8
9 #include <linux/kernel.h>
10 #include <linux/taskstats_kern.h>
11 #include <linux/tsacct_kern.h>
12 #include <linux/delayacct.h>
13 #include <linux/cpumask.h>
14 #include <linux/percpu.h>
15 #include <linux/slab.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/cgroup.h>
18 #include <linux/fs.h>
19 #include <linux/file.h>
20 #include <linux/pid_namespace.h>
21 #include <net/genetlink.h>
22 #include <linux/atomic.h>
23 #include <linux/sched/cputime.h>
24
25 /*
26 * Maximum length of a cpumask that can be specified in
27 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
28 */
29 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
30
31 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
32 static int family_registered;
33 struct kmem_cache *taskstats_cache;
34
35 static struct genl_family family;
36
37 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
38 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
39 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
40 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
41 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
42
43 /*
44 * We have to use TASKSTATS_CMD_ATTR_MAX here, it is the maxattr in the family.
45 * Make sure they are always aligned.
46 */
47 static const struct nla_policy cgroupstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
48 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
49 };
50
51 struct listener {
52 struct list_head list;
53 pid_t pid;
54 char valid;
55 };
56
57 struct listener_list {
58 struct rw_semaphore sem;
59 struct list_head list;
60 };
61 static DEFINE_PER_CPU(struct listener_list, listener_array);
62
63 enum actions {
64 REGISTER,
65 DEREGISTER,
66 CPU_DONT_CARE
67 };
68
prepare_reply(struct genl_info * info,u8 cmd,struct sk_buff ** skbp,size_t size)69 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
70 size_t size)
71 {
72 struct sk_buff *skb;
73 void *reply;
74
75 /*
76 * If new attributes are added, please revisit this allocation
77 */
78 skb = genlmsg_new(size, GFP_KERNEL);
79 if (!skb)
80 return -ENOMEM;
81
82 if (!info) {
83 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
84
85 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
86 } else
87 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
88 if (reply == NULL) {
89 nlmsg_free(skb);
90 return -EINVAL;
91 }
92
93 *skbp = skb;
94 return 0;
95 }
96
97 /*
98 * Send taskstats data in @skb to listener with nl_pid @pid
99 */
send_reply(struct sk_buff * skb,struct genl_info * info)100 static int send_reply(struct sk_buff *skb, struct genl_info *info)
101 {
102 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
103 void *reply = genlmsg_data(genlhdr);
104
105 genlmsg_end(skb, reply);
106
107 return genlmsg_reply(skb, info);
108 }
109
110 /*
111 * Send taskstats data in @skb to listeners registered for @cpu's exit data
112 */
send_cpu_listeners(struct sk_buff * skb,struct listener_list * listeners)113 static void send_cpu_listeners(struct sk_buff *skb,
114 struct listener_list *listeners)
115 {
116 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
117 struct listener *s, *tmp;
118 struct sk_buff *skb_next, *skb_cur = skb;
119 void *reply = genlmsg_data(genlhdr);
120 int rc, delcount = 0;
121
122 genlmsg_end(skb, reply);
123
124 rc = 0;
125 down_read(&listeners->sem);
126 list_for_each_entry(s, &listeners->list, list) {
127 skb_next = NULL;
128 if (!list_is_last(&s->list, &listeners->list)) {
129 skb_next = skb_clone(skb_cur, GFP_KERNEL);
130 if (!skb_next)
131 break;
132 }
133 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
134 if (rc == -ECONNREFUSED) {
135 s->valid = 0;
136 delcount++;
137 }
138 skb_cur = skb_next;
139 }
140 up_read(&listeners->sem);
141
142 if (skb_cur)
143 nlmsg_free(skb_cur);
144
145 if (!delcount)
146 return;
147
148 /* Delete invalidated entries */
149 down_write(&listeners->sem);
150 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
151 if (!s->valid) {
152 list_del(&s->list);
153 kfree(s);
154 }
155 }
156 up_write(&listeners->sem);
157 }
158
fill_stats(struct user_namespace * user_ns,struct pid_namespace * pid_ns,struct task_struct * tsk,struct taskstats * stats)159 static void fill_stats(struct user_namespace *user_ns,
160 struct pid_namespace *pid_ns,
161 struct task_struct *tsk, struct taskstats *stats)
162 {
163 memset(stats, 0, sizeof(*stats));
164 /*
165 * Each accounting subsystem adds calls to its functions to
166 * fill in relevant parts of struct taskstsats as follows
167 *
168 * per-task-foo(stats, tsk);
169 */
170
171 delayacct_add_tsk(stats, tsk);
172
173 /* fill in basic acct fields */
174 stats->version = TASKSTATS_VERSION;
175 stats->nvcsw = tsk->nvcsw;
176 stats->nivcsw = tsk->nivcsw;
177 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
178
179 /* fill in extended acct fields */
180 xacct_add_tsk(stats, tsk);
181 }
182
fill_stats_for_pid(pid_t pid,struct taskstats * stats)183 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
184 {
185 struct task_struct *tsk;
186
187 tsk = find_get_task_by_vpid(pid);
188 if (!tsk)
189 return -ESRCH;
190 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
191 put_task_struct(tsk);
192 return 0;
193 }
194
fill_stats_for_tgid(pid_t tgid,struct taskstats * stats)195 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
196 {
197 struct task_struct *tsk, *first;
198 unsigned long flags;
199 int rc = -ESRCH;
200 u64 delta, utime, stime;
201 u64 start_time;
202
203 /*
204 * Add additional stats from live tasks except zombie thread group
205 * leaders who are already counted with the dead tasks
206 */
207 rcu_read_lock();
208 first = find_task_by_vpid(tgid);
209
210 if (!first || !lock_task_sighand(first, &flags))
211 goto out;
212
213 if (first->signal->stats)
214 memcpy(stats, first->signal->stats, sizeof(*stats));
215 else
216 memset(stats, 0, sizeof(*stats));
217
218 tsk = first;
219 start_time = ktime_get_ns();
220 do {
221 if (tsk->exit_state)
222 continue;
223 /*
224 * Accounting subsystem can call its functions here to
225 * fill in relevant parts of struct taskstsats as follows
226 *
227 * per-task-foo(stats, tsk);
228 */
229 delayacct_add_tsk(stats, tsk);
230
231 /* calculate task elapsed time in nsec */
232 delta = start_time - tsk->start_time;
233 /* Convert to micro seconds */
234 do_div(delta, NSEC_PER_USEC);
235 stats->ac_etime += delta;
236
237 task_cputime(tsk, &utime, &stime);
238 stats->ac_utime += div_u64(utime, NSEC_PER_USEC);
239 stats->ac_stime += div_u64(stime, NSEC_PER_USEC);
240
241 stats->nvcsw += tsk->nvcsw;
242 stats->nivcsw += tsk->nivcsw;
243 } while_each_thread(first, tsk);
244
245 unlock_task_sighand(first, &flags);
246 rc = 0;
247 out:
248 rcu_read_unlock();
249
250 stats->version = TASKSTATS_VERSION;
251 /*
252 * Accounting subsystems can also add calls here to modify
253 * fields of taskstats.
254 */
255 return rc;
256 }
257
fill_tgid_exit(struct task_struct * tsk)258 static void fill_tgid_exit(struct task_struct *tsk)
259 {
260 unsigned long flags;
261
262 spin_lock_irqsave(&tsk->sighand->siglock, flags);
263 if (!tsk->signal->stats)
264 goto ret;
265
266 /*
267 * Each accounting subsystem calls its functions here to
268 * accumalate its per-task stats for tsk, into the per-tgid structure
269 *
270 * per-task-foo(tsk->signal->stats, tsk);
271 */
272 delayacct_add_tsk(tsk->signal->stats, tsk);
273 ret:
274 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
275 return;
276 }
277
add_del_listener(pid_t pid,const struct cpumask * mask,int isadd)278 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
279 {
280 struct listener_list *listeners;
281 struct listener *s, *tmp, *s2;
282 unsigned int cpu;
283 int ret = 0;
284
285 if (!cpumask_subset(mask, cpu_possible_mask))
286 return -EINVAL;
287
288 if (current_user_ns() != &init_user_ns)
289 return -EINVAL;
290
291 if (task_active_pid_ns(current) != &init_pid_ns)
292 return -EINVAL;
293
294 if (isadd == REGISTER) {
295 for_each_cpu(cpu, mask) {
296 s = kmalloc_node(sizeof(struct listener),
297 GFP_KERNEL, cpu_to_node(cpu));
298 if (!s) {
299 ret = -ENOMEM;
300 goto cleanup;
301 }
302 s->pid = pid;
303 s->valid = 1;
304
305 listeners = &per_cpu(listener_array, cpu);
306 down_write(&listeners->sem);
307 list_for_each_entry(s2, &listeners->list, list) {
308 if (s2->pid == pid && s2->valid)
309 goto exists;
310 }
311 list_add(&s->list, &listeners->list);
312 s = NULL;
313 exists:
314 up_write(&listeners->sem);
315 kfree(s); /* nop if NULL */
316 }
317 return 0;
318 }
319
320 /* Deregister or cleanup */
321 cleanup:
322 for_each_cpu(cpu, mask) {
323 listeners = &per_cpu(listener_array, cpu);
324 down_write(&listeners->sem);
325 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
326 if (s->pid == pid) {
327 list_del(&s->list);
328 kfree(s);
329 break;
330 }
331 }
332 up_write(&listeners->sem);
333 }
334 return ret;
335 }
336
parse(struct nlattr * na,struct cpumask * mask)337 static int parse(struct nlattr *na, struct cpumask *mask)
338 {
339 char *data;
340 int len;
341 int ret;
342
343 if (na == NULL)
344 return 1;
345 len = nla_len(na);
346 if (len > TASKSTATS_CPUMASK_MAXLEN)
347 return -E2BIG;
348 if (len < 1)
349 return -EINVAL;
350 data = kmalloc(len, GFP_KERNEL);
351 if (!data)
352 return -ENOMEM;
353 nla_strlcpy(data, na, len);
354 ret = cpulist_parse(data, mask);
355 kfree(data);
356 return ret;
357 }
358
mk_reply(struct sk_buff * skb,int type,u32 pid)359 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
360 {
361 struct nlattr *na, *ret;
362 int aggr;
363
364 aggr = (type == TASKSTATS_TYPE_PID)
365 ? TASKSTATS_TYPE_AGGR_PID
366 : TASKSTATS_TYPE_AGGR_TGID;
367
368 na = nla_nest_start_noflag(skb, aggr);
369 if (!na)
370 goto err;
371
372 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
373 nla_nest_cancel(skb, na);
374 goto err;
375 }
376 ret = nla_reserve_64bit(skb, TASKSTATS_TYPE_STATS,
377 sizeof(struct taskstats), TASKSTATS_TYPE_NULL);
378 if (!ret) {
379 nla_nest_cancel(skb, na);
380 goto err;
381 }
382 nla_nest_end(skb, na);
383
384 return nla_data(ret);
385 err:
386 return NULL;
387 }
388
cgroupstats_user_cmd(struct sk_buff * skb,struct genl_info * info)389 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
390 {
391 int rc = 0;
392 struct sk_buff *rep_skb;
393 struct cgroupstats *stats;
394 struct nlattr *na;
395 size_t size;
396 u32 fd;
397 struct fd f;
398
399 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
400 if (!na)
401 return -EINVAL;
402
403 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
404 f = fdget(fd);
405 if (!f.file)
406 return 0;
407
408 size = nla_total_size(sizeof(struct cgroupstats));
409
410 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
411 size);
412 if (rc < 0)
413 goto err;
414
415 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
416 sizeof(struct cgroupstats));
417 if (na == NULL) {
418 nlmsg_free(rep_skb);
419 rc = -EMSGSIZE;
420 goto err;
421 }
422
423 stats = nla_data(na);
424 memset(stats, 0, sizeof(*stats));
425
426 rc = cgroupstats_build(stats, f.file->f_path.dentry);
427 if (rc < 0) {
428 nlmsg_free(rep_skb);
429 goto err;
430 }
431
432 rc = send_reply(rep_skb, info);
433
434 err:
435 fdput(f);
436 return rc;
437 }
438
cmd_attr_register_cpumask(struct genl_info * info)439 static int cmd_attr_register_cpumask(struct genl_info *info)
440 {
441 cpumask_var_t mask;
442 int rc;
443
444 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
445 return -ENOMEM;
446 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
447 if (rc < 0)
448 goto out;
449 rc = add_del_listener(info->snd_portid, mask, REGISTER);
450 out:
451 free_cpumask_var(mask);
452 return rc;
453 }
454
cmd_attr_deregister_cpumask(struct genl_info * info)455 static int cmd_attr_deregister_cpumask(struct genl_info *info)
456 {
457 cpumask_var_t mask;
458 int rc;
459
460 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
461 return -ENOMEM;
462 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
463 if (rc < 0)
464 goto out;
465 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
466 out:
467 free_cpumask_var(mask);
468 return rc;
469 }
470
taskstats_packet_size(void)471 static size_t taskstats_packet_size(void)
472 {
473 size_t size;
474
475 size = nla_total_size(sizeof(u32)) +
476 nla_total_size_64bit(sizeof(struct taskstats)) +
477 nla_total_size(0);
478
479 return size;
480 }
481
cmd_attr_pid(struct genl_info * info)482 static int cmd_attr_pid(struct genl_info *info)
483 {
484 struct taskstats *stats;
485 struct sk_buff *rep_skb;
486 size_t size;
487 u32 pid;
488 int rc;
489
490 size = taskstats_packet_size();
491
492 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
493 if (rc < 0)
494 return rc;
495
496 rc = -EINVAL;
497 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
498 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
499 if (!stats)
500 goto err;
501
502 rc = fill_stats_for_pid(pid, stats);
503 if (rc < 0)
504 goto err;
505 return send_reply(rep_skb, info);
506 err:
507 nlmsg_free(rep_skb);
508 return rc;
509 }
510
cmd_attr_tgid(struct genl_info * info)511 static int cmd_attr_tgid(struct genl_info *info)
512 {
513 struct taskstats *stats;
514 struct sk_buff *rep_skb;
515 size_t size;
516 u32 tgid;
517 int rc;
518
519 size = taskstats_packet_size();
520
521 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
522 if (rc < 0)
523 return rc;
524
525 rc = -EINVAL;
526 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
527 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
528 if (!stats)
529 goto err;
530
531 rc = fill_stats_for_tgid(tgid, stats);
532 if (rc < 0)
533 goto err;
534 return send_reply(rep_skb, info);
535 err:
536 nlmsg_free(rep_skb);
537 return rc;
538 }
539
taskstats_user_cmd(struct sk_buff * skb,struct genl_info * info)540 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
541 {
542 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
543 return cmd_attr_register_cpumask(info);
544 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
545 return cmd_attr_deregister_cpumask(info);
546 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
547 return cmd_attr_pid(info);
548 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
549 return cmd_attr_tgid(info);
550 else
551 return -EINVAL;
552 }
553
taskstats_tgid_alloc(struct task_struct * tsk)554 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
555 {
556 struct signal_struct *sig = tsk->signal;
557 struct taskstats *stats;
558
559 if (sig->stats || thread_group_empty(tsk))
560 goto ret;
561
562 /* No problem if kmem_cache_zalloc() fails */
563 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
564
565 spin_lock_irq(&tsk->sighand->siglock);
566 if (!sig->stats) {
567 sig->stats = stats;
568 stats = NULL;
569 }
570 spin_unlock_irq(&tsk->sighand->siglock);
571
572 if (stats)
573 kmem_cache_free(taskstats_cache, stats);
574 ret:
575 return sig->stats;
576 }
577
578 /* Send pid data out on exit */
taskstats_exit(struct task_struct * tsk,int group_dead)579 void taskstats_exit(struct task_struct *tsk, int group_dead)
580 {
581 int rc;
582 struct listener_list *listeners;
583 struct taskstats *stats;
584 struct sk_buff *rep_skb;
585 size_t size;
586 int is_thread_group;
587
588 if (!family_registered)
589 return;
590
591 /*
592 * Size includes space for nested attributes
593 */
594 size = taskstats_packet_size();
595
596 is_thread_group = !!taskstats_tgid_alloc(tsk);
597 if (is_thread_group) {
598 /* PID + STATS + TGID + STATS */
599 size = 2 * size;
600 /* fill the tsk->signal->stats structure */
601 fill_tgid_exit(tsk);
602 }
603
604 listeners = raw_cpu_ptr(&listener_array);
605 if (list_empty(&listeners->list))
606 return;
607
608 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
609 if (rc < 0)
610 return;
611
612 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
613 task_pid_nr_ns(tsk, &init_pid_ns));
614 if (!stats)
615 goto err;
616
617 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
618
619 /*
620 * Doesn't matter if tsk is the leader or the last group member leaving
621 */
622 if (!is_thread_group || !group_dead)
623 goto send;
624
625 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
626 task_tgid_nr_ns(tsk, &init_pid_ns));
627 if (!stats)
628 goto err;
629
630 memcpy(stats, tsk->signal->stats, sizeof(*stats));
631
632 send:
633 send_cpu_listeners(rep_skb, listeners);
634 return;
635 err:
636 nlmsg_free(rep_skb);
637 }
638
639 static const struct genl_ops taskstats_ops[] = {
640 {
641 .cmd = TASKSTATS_CMD_GET,
642 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
643 .doit = taskstats_user_cmd,
644 /* policy enforced later */
645 .flags = GENL_ADMIN_PERM | GENL_CMD_CAP_HASPOL,
646 },
647 {
648 .cmd = CGROUPSTATS_CMD_GET,
649 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
650 .doit = cgroupstats_user_cmd,
651 /* policy enforced later */
652 .flags = GENL_CMD_CAP_HASPOL,
653 },
654 };
655
taskstats_pre_doit(const struct genl_ops * ops,struct sk_buff * skb,struct genl_info * info)656 static int taskstats_pre_doit(const struct genl_ops *ops, struct sk_buff *skb,
657 struct genl_info *info)
658 {
659 const struct nla_policy *policy = NULL;
660
661 switch (ops->cmd) {
662 case TASKSTATS_CMD_GET:
663 policy = taskstats_cmd_get_policy;
664 break;
665 case CGROUPSTATS_CMD_GET:
666 policy = cgroupstats_cmd_get_policy;
667 break;
668 default:
669 return -EINVAL;
670 }
671
672 return nlmsg_validate_deprecated(info->nlhdr, GENL_HDRLEN,
673 TASKSTATS_CMD_ATTR_MAX, policy,
674 info->extack);
675 }
676
677 static struct genl_family family __ro_after_init = {
678 .name = TASKSTATS_GENL_NAME,
679 .version = TASKSTATS_GENL_VERSION,
680 .maxattr = TASKSTATS_CMD_ATTR_MAX,
681 .module = THIS_MODULE,
682 .ops = taskstats_ops,
683 .n_ops = ARRAY_SIZE(taskstats_ops),
684 .pre_doit = taskstats_pre_doit,
685 };
686
687 /* Needed early in initialization */
taskstats_init_early(void)688 void __init taskstats_init_early(void)
689 {
690 unsigned int i;
691
692 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
693 for_each_possible_cpu(i) {
694 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
695 init_rwsem(&(per_cpu(listener_array, i).sem));
696 }
697 }
698
taskstats_init(void)699 static int __init taskstats_init(void)
700 {
701 int rc;
702
703 rc = genl_register_family(&family);
704 if (rc)
705 return rc;
706
707 family_registered = 1;
708 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
709 return 0;
710 }
711
712 /*
713 * late initcall ensures initialization of statistics collection
714 * mechanisms precedes initialization of the taskstats interface
715 */
716 late_initcall(taskstats_init);
717
