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;
198 tsk = find_task_by_vpid(pid);
200 get_task_struct(tsk);
204 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
205 put_task_struct(tsk);
209 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
211 struct task_struct *tsk, *first;
214 u64 delta, utime, stime;
218 * Add additional stats from live tasks except zombie thread group
219 * leaders who are already counted with the dead tasks
222 first = find_task_by_vpid(tgid);
224 if (!first || !lock_task_sighand(first, &flags))
227 if (first->signal->stats)
228 memcpy(stats, first->signal->stats, sizeof(*stats));
230 memset(stats, 0, sizeof(*stats));
233 start_time = ktime_get_ns();
238 * Accounting subsystem can call its functions here to
239 * fill in relevant parts of struct taskstsats as follows
241 * per-task-foo(stats, tsk);
243 delayacct_add_tsk(stats, tsk);
245 /* calculate task elapsed time in nsec */
246 delta = start_time - tsk->start_time;
247 /* Convert to micro seconds */
248 do_div(delta, NSEC_PER_USEC);
249 stats->ac_etime += delta;
251 task_cputime(tsk, &utime, &stime);
252 stats->ac_utime += div_u64(utime, NSEC_PER_USEC);
253 stats->ac_stime += div_u64(stime, NSEC_PER_USEC);
255 stats->nvcsw += tsk->nvcsw;
256 stats->nivcsw += tsk->nivcsw;
257 } while_each_thread(first, tsk);
259 unlock_task_sighand(first, &flags);
264 stats->version = TASKSTATS_VERSION;
266 * Accounting subsystems can also add calls here to modify
267 * fields of taskstats.
272 static void fill_tgid_exit(struct task_struct *tsk)
276 spin_lock_irqsave(&tsk->sighand->siglock, flags);
277 if (!tsk->signal->stats)
281 * Each accounting subsystem calls its functions here to
282 * accumalate its per-task stats for tsk, into the per-tgid structure
284 * per-task-foo(tsk->signal->stats, tsk);
286 delayacct_add_tsk(tsk->signal->stats, tsk);
288 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
292 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
294 struct listener_list *listeners;
295 struct listener *s, *tmp, *s2;
299 if (!cpumask_subset(mask, cpu_possible_mask))
302 if (current_user_ns() != &init_user_ns)
305 if (task_active_pid_ns(current) != &init_pid_ns)
308 if (isadd == REGISTER) {
309 for_each_cpu(cpu, mask) {
310 s = kmalloc_node(sizeof(struct listener),
311 GFP_KERNEL, cpu_to_node(cpu));
319 listeners = &per_cpu(listener_array, cpu);
320 down_write(&listeners->sem);
321 list_for_each_entry(s2, &listeners->list, list) {
322 if (s2->pid == pid && s2->valid)
325 list_add(&s->list, &listeners->list);
328 up_write(&listeners->sem);
329 kfree(s); /* nop if NULL */
334 /* Deregister or cleanup */
336 for_each_cpu(cpu, mask) {
337 listeners = &per_cpu(listener_array, cpu);
338 down_write(&listeners->sem);
339 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
346 up_write(&listeners->sem);
351 static int parse(struct nlattr *na, struct cpumask *mask)
360 if (len > TASKSTATS_CPUMASK_MAXLEN)
364 data = kmalloc(len, GFP_KERNEL);
367 nla_strlcpy(data, na, len);
368 ret = cpulist_parse(data, mask);
373 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
375 struct nlattr *na, *ret;
378 aggr = (type == TASKSTATS_TYPE_PID)
379 ? TASKSTATS_TYPE_AGGR_PID
380 : TASKSTATS_TYPE_AGGR_TGID;
382 na = nla_nest_start(skb, aggr);
386 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
387 nla_nest_cancel(skb, na);
390 ret = nla_reserve_64bit(skb, TASKSTATS_TYPE_STATS,
391 sizeof(struct taskstats), TASKSTATS_TYPE_NULL);
393 nla_nest_cancel(skb, na);
396 nla_nest_end(skb, na);
398 return nla_data(ret);
403 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
406 struct sk_buff *rep_skb;
407 struct cgroupstats *stats;
413 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
417 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
422 size = nla_total_size(sizeof(struct cgroupstats));
424 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
429 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
430 sizeof(struct cgroupstats));
437 stats = nla_data(na);
438 memset(stats, 0, sizeof(*stats));
440 rc = cgroupstats_build(stats, f.file->f_path.dentry);
446 rc = send_reply(rep_skb, info);
453 static int cmd_attr_register_cpumask(struct genl_info *info)
458 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
460 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
463 rc = add_del_listener(info->snd_portid, mask, REGISTER);
465 free_cpumask_var(mask);
469 static int cmd_attr_deregister_cpumask(struct genl_info *info)
474 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
476 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
479 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
481 free_cpumask_var(mask);
485 static size_t taskstats_packet_size(void)
489 size = nla_total_size(sizeof(u32)) +
490 nla_total_size_64bit(sizeof(struct taskstats)) +
496 static int cmd_attr_pid(struct genl_info *info)
498 struct taskstats *stats;
499 struct sk_buff *rep_skb;
504 size = taskstats_packet_size();
506 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
511 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
512 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
516 rc = fill_stats_for_pid(pid, stats);
519 return send_reply(rep_skb, info);
525 static int cmd_attr_tgid(struct genl_info *info)
527 struct taskstats *stats;
528 struct sk_buff *rep_skb;
533 size = taskstats_packet_size();
535 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
540 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
541 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
545 rc = fill_stats_for_tgid(tgid, stats);
548 return send_reply(rep_skb, info);
554 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
556 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
557 return cmd_attr_register_cpumask(info);
558 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
559 return cmd_attr_deregister_cpumask(info);
560 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
561 return cmd_attr_pid(info);
562 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
563 return cmd_attr_tgid(info);
568 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
570 struct signal_struct *sig = tsk->signal;
571 struct taskstats *stats;
573 if (sig->stats || thread_group_empty(tsk))
576 /* No problem if kmem_cache_zalloc() fails */
577 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
579 spin_lock_irq(&tsk->sighand->siglock);
584 spin_unlock_irq(&tsk->sighand->siglock);
587 kmem_cache_free(taskstats_cache, stats);
592 /* Send pid data out on exit */
593 void taskstats_exit(struct task_struct *tsk, int group_dead)
596 struct listener_list *listeners;
597 struct taskstats *stats;
598 struct sk_buff *rep_skb;
602 if (!family_registered)
606 * Size includes space for nested attributes
608 size = taskstats_packet_size();
610 is_thread_group = !!taskstats_tgid_alloc(tsk);
611 if (is_thread_group) {
612 /* PID + STATS + TGID + STATS */
614 /* fill the tsk->signal->stats structure */
618 listeners = raw_cpu_ptr(&listener_array);
619 if (list_empty(&listeners->list))
622 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
626 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
627 task_pid_nr_ns(tsk, &init_pid_ns));
631 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
634 * Doesn't matter if tsk is the leader or the last group member leaving
636 if (!is_thread_group || !group_dead)
639 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
640 task_tgid_nr_ns(tsk, &init_pid_ns));
644 memcpy(stats, tsk->signal->stats, sizeof(*stats));
647 send_cpu_listeners(rep_skb, listeners);
653 static const struct genl_ops taskstats_ops[] = {
655 .cmd = TASKSTATS_CMD_GET,
656 .doit = taskstats_user_cmd,
657 .policy = taskstats_cmd_get_policy,
658 .flags = GENL_ADMIN_PERM,
661 .cmd = CGROUPSTATS_CMD_GET,
662 .doit = cgroupstats_user_cmd,
663 .policy = cgroupstats_cmd_get_policy,
667 static struct genl_family family __ro_after_init = {
668 .name = TASKSTATS_GENL_NAME,
669 .version = TASKSTATS_GENL_VERSION,
670 .maxattr = TASKSTATS_CMD_ATTR_MAX,
671 .module = THIS_MODULE,
672 .ops = taskstats_ops,
673 .n_ops = ARRAY_SIZE(taskstats_ops),
676 /* Needed early in initialization */
677 void __init taskstats_init_early(void)
681 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
682 for_each_possible_cpu(i) {
683 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
684 init_rwsem(&(per_cpu(listener_array, i).sem));
688 static int __init taskstats_init(void)
692 rc = genl_register_family(&family);
696 family_registered = 1;
697 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
702 * late initcall ensures initialization of statistics collection
703 * mechanisms precedes initialization of the taskstats interface
705 late_initcall(taskstats_init);