2 * taskstats.c - Export per-task statistics to userland
4 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
5 * (C) Balbir Singh, IBM Corp. 2006
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
19 #include <linux/kernel.h>
20 #include <linux/taskstats_kern.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/delayacct.h>
23 #include <linux/cpumask.h>
24 #include <linux/percpu.h>
25 #include <linux/slab.h>
26 #include <linux/cgroupstats.h>
27 #include <linux/cgroup.h>
29 #include <linux/file.h>
30 #include <linux/pid_namespace.h>
31 #include <net/genetlink.h>
32 #include <linux/atomic.h>
33 #include <linux/sched/cputime.h>
36 * Maximum length of a cpumask that can be specified in
37 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
39 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
41 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
42 static int family_registered;
43 struct kmem_cache *taskstats_cache;
45 static struct genl_family family;
47 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
48 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
49 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
50 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
51 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
54 * We have to use TASKSTATS_CMD_ATTR_MAX here, it is the maxattr in the family.
55 * Make sure they are always aligned.
57 static const struct nla_policy cgroupstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
58 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
62 struct list_head list;
67 struct listener_list {
68 struct rw_semaphore sem;
69 struct list_head list;
71 static DEFINE_PER_CPU(struct listener_list, listener_array);
79 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
86 * If new attributes are added, please revisit this allocation
88 skb = genlmsg_new(size, GFP_KERNEL);
93 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
95 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
97 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
108 * Send taskstats data in @skb to listener with nl_pid @pid
110 static int send_reply(struct sk_buff *skb, struct genl_info *info)
112 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
113 void *reply = genlmsg_data(genlhdr);
115 genlmsg_end(skb, reply);
117 return genlmsg_reply(skb, info);
121 * Send taskstats data in @skb to listeners registered for @cpu's exit data
123 static void send_cpu_listeners(struct sk_buff *skb,
124 struct listener_list *listeners)
126 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
127 struct listener *s, *tmp;
128 struct sk_buff *skb_next, *skb_cur = skb;
129 void *reply = genlmsg_data(genlhdr);
130 int rc, delcount = 0;
132 genlmsg_end(skb, reply);
135 down_read(&listeners->sem);
136 list_for_each_entry(s, &listeners->list, list) {
138 if (!list_is_last(&s->list, &listeners->list)) {
139 skb_next = skb_clone(skb_cur, GFP_KERNEL);
143 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
144 if (rc == -ECONNREFUSED) {
150 up_read(&listeners->sem);
158 /* Delete invalidated entries */
159 down_write(&listeners->sem);
160 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
166 up_write(&listeners->sem);
169 static void fill_stats(struct user_namespace *user_ns,
170 struct pid_namespace *pid_ns,
171 struct task_struct *tsk, struct taskstats *stats)
173 memset(stats, 0, sizeof(*stats));
175 * Each accounting subsystem adds calls to its functions to
176 * fill in relevant parts of struct taskstsats as follows
178 * per-task-foo(stats, tsk);
181 delayacct_add_tsk(stats, tsk);
183 /* fill in basic acct fields */
184 stats->version = TASKSTATS_VERSION;
185 stats->nvcsw = tsk->nvcsw;
186 stats->nivcsw = tsk->nivcsw;
187 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
189 /* fill in extended acct fields */
190 xacct_add_tsk(stats, tsk);
193 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
195 struct task_struct *tsk;
197 tsk = find_get_task_by_vpid(pid);
200 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
201 put_task_struct(tsk);
205 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
207 struct task_struct *tsk, *first;
210 u64 delta, utime, stime;
214 * Add additional stats from live tasks except zombie thread group
215 * leaders who are already counted with the dead tasks
218 first = find_task_by_vpid(tgid);
220 if (!first || !lock_task_sighand(first, &flags))
223 if (first->signal->stats)
224 memcpy(stats, first->signal->stats, sizeof(*stats));
226 memset(stats, 0, sizeof(*stats));
229 start_time = ktime_get_ns();
234 * Accounting subsystem can call its functions here to
235 * fill in relevant parts of struct taskstsats as follows
237 * per-task-foo(stats, tsk);
239 delayacct_add_tsk(stats, tsk);
241 /* calculate task elapsed time in nsec */
242 delta = start_time - tsk->start_time;
243 /* Convert to micro seconds */
244 do_div(delta, NSEC_PER_USEC);
245 stats->ac_etime += delta;
247 task_cputime(tsk, &utime, &stime);
248 stats->ac_utime += div_u64(utime, NSEC_PER_USEC);
249 stats->ac_stime += div_u64(stime, NSEC_PER_USEC);
251 stats->nvcsw += tsk->nvcsw;
252 stats->nivcsw += tsk->nivcsw;
253 } while_each_thread(first, tsk);
255 unlock_task_sighand(first, &flags);
260 stats->version = TASKSTATS_VERSION;
262 * Accounting subsystems can also add calls here to modify
263 * fields of taskstats.
268 static void fill_tgid_exit(struct task_struct *tsk)
272 spin_lock_irqsave(&tsk->sighand->siglock, flags);
273 if (!tsk->signal->stats)
277 * Each accounting subsystem calls its functions here to
278 * accumalate its per-task stats for tsk, into the per-tgid structure
280 * per-task-foo(tsk->signal->stats, tsk);
282 delayacct_add_tsk(tsk->signal->stats, tsk);
284 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
288 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
290 struct listener_list *listeners;
291 struct listener *s, *tmp, *s2;
295 if (!cpumask_subset(mask, cpu_possible_mask))
298 if (current_user_ns() != &init_user_ns)
301 if (task_active_pid_ns(current) != &init_pid_ns)
304 if (isadd == REGISTER) {
305 for_each_cpu(cpu, mask) {
306 s = kmalloc_node(sizeof(struct listener),
307 GFP_KERNEL, cpu_to_node(cpu));
315 listeners = &per_cpu(listener_array, cpu);
316 down_write(&listeners->sem);
317 list_for_each_entry(s2, &listeners->list, list) {
318 if (s2->pid == pid && s2->valid)
321 list_add(&s->list, &listeners->list);
324 up_write(&listeners->sem);
325 kfree(s); /* nop if NULL */
330 /* Deregister or cleanup */
332 for_each_cpu(cpu, mask) {
333 listeners = &per_cpu(listener_array, cpu);
334 down_write(&listeners->sem);
335 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
342 up_write(&listeners->sem);
347 static int parse(struct nlattr *na, struct cpumask *mask)
356 if (len > TASKSTATS_CPUMASK_MAXLEN)
360 data = kmalloc(len, GFP_KERNEL);
363 nla_strlcpy(data, na, len);
364 ret = cpulist_parse(data, mask);
369 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
371 struct nlattr *na, *ret;
374 aggr = (type == TASKSTATS_TYPE_PID)
375 ? TASKSTATS_TYPE_AGGR_PID
376 : TASKSTATS_TYPE_AGGR_TGID;
378 na = nla_nest_start(skb, aggr);
382 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
383 nla_nest_cancel(skb, na);
386 ret = nla_reserve_64bit(skb, TASKSTATS_TYPE_STATS,
387 sizeof(struct taskstats), TASKSTATS_TYPE_NULL);
389 nla_nest_cancel(skb, na);
392 nla_nest_end(skb, na);
394 return nla_data(ret);
399 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
402 struct sk_buff *rep_skb;
403 struct cgroupstats *stats;
409 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
413 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
418 size = nla_total_size(sizeof(struct cgroupstats));
420 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
425 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
426 sizeof(struct cgroupstats));
433 stats = nla_data(na);
434 memset(stats, 0, sizeof(*stats));
436 rc = cgroupstats_build(stats, f.file->f_path.dentry);
442 rc = send_reply(rep_skb, info);
449 static int cmd_attr_register_cpumask(struct genl_info *info)
454 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
456 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
459 rc = add_del_listener(info->snd_portid, mask, REGISTER);
461 free_cpumask_var(mask);
465 static int cmd_attr_deregister_cpumask(struct genl_info *info)
470 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
472 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
475 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
477 free_cpumask_var(mask);
481 static size_t taskstats_packet_size(void)
485 size = nla_total_size(sizeof(u32)) +
486 nla_total_size_64bit(sizeof(struct taskstats)) +
492 static int cmd_attr_pid(struct genl_info *info)
494 struct taskstats *stats;
495 struct sk_buff *rep_skb;
500 size = taskstats_packet_size();
502 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
507 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
508 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
512 rc = fill_stats_for_pid(pid, stats);
515 return send_reply(rep_skb, info);
521 static int cmd_attr_tgid(struct genl_info *info)
523 struct taskstats *stats;
524 struct sk_buff *rep_skb;
529 size = taskstats_packet_size();
531 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
536 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
537 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
541 rc = fill_stats_for_tgid(tgid, stats);
544 return send_reply(rep_skb, info);
550 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
552 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
553 return cmd_attr_register_cpumask(info);
554 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
555 return cmd_attr_deregister_cpumask(info);
556 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
557 return cmd_attr_pid(info);
558 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
559 return cmd_attr_tgid(info);
564 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
566 struct signal_struct *sig = tsk->signal;
567 struct taskstats *stats_new, *stats;
569 /* Pairs with smp_store_release() below. */
570 stats = smp_load_acquire(&sig->stats);
571 if (stats || thread_group_empty(tsk))
574 /* No problem if kmem_cache_zalloc() fails */
575 stats_new = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
577 spin_lock_irq(&tsk->sighand->siglock);
581 * Pairs with smp_store_release() above and order the
582 * kmem_cache_zalloc().
584 smp_store_release(&sig->stats, stats_new);
588 spin_unlock_irq(&tsk->sighand->siglock);
591 kmem_cache_free(taskstats_cache, stats_new);
596 /* Send pid data out on exit */
597 void taskstats_exit(struct task_struct *tsk, int group_dead)
600 struct listener_list *listeners;
601 struct taskstats *stats;
602 struct sk_buff *rep_skb;
606 if (!family_registered)
610 * Size includes space for nested attributes
612 size = taskstats_packet_size();
614 is_thread_group = !!taskstats_tgid_alloc(tsk);
615 if (is_thread_group) {
616 /* PID + STATS + TGID + STATS */
618 /* fill the tsk->signal->stats structure */
622 listeners = raw_cpu_ptr(&listener_array);
623 if (list_empty(&listeners->list))
626 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
630 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
631 task_pid_nr_ns(tsk, &init_pid_ns));
635 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
638 * Doesn't matter if tsk is the leader or the last group member leaving
640 if (!is_thread_group || !group_dead)
643 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
644 task_tgid_nr_ns(tsk, &init_pid_ns));
648 memcpy(stats, tsk->signal->stats, sizeof(*stats));
651 send_cpu_listeners(rep_skb, listeners);
657 static const struct genl_ops taskstats_ops[] = {
659 .cmd = TASKSTATS_CMD_GET,
660 .doit = taskstats_user_cmd,
661 .policy = taskstats_cmd_get_policy,
662 .flags = GENL_ADMIN_PERM,
665 .cmd = CGROUPSTATS_CMD_GET,
666 .doit = cgroupstats_user_cmd,
667 .policy = cgroupstats_cmd_get_policy,
671 static struct genl_family family __ro_after_init = {
672 .name = TASKSTATS_GENL_NAME,
673 .version = TASKSTATS_GENL_VERSION,
674 .maxattr = TASKSTATS_CMD_ATTR_MAX,
675 .module = THIS_MODULE,
676 .ops = taskstats_ops,
677 .n_ops = ARRAY_SIZE(taskstats_ops),
680 /* Needed early in initialization */
681 void __init taskstats_init_early(void)
685 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
686 for_each_possible_cpu(i) {
687 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
688 init_rwsem(&(per_cpu(listener_array, i).sem));
692 static int __init taskstats_init(void)
696 rc = genl_register_family(&family);
700 family_registered = 1;
701 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
706 * late initcall ensures initialization of statistics collection
707 * mechanisms precedes initialization of the taskstats interface
709 late_initcall(taskstats_init);