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 <net/genetlink.h>
31 #include <asm/atomic.h>
34 * Maximum length of a cpumask that can be specified in
35 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
37 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
39 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
40 static int family_registered;
41 struct kmem_cache *taskstats_cache;
43 static struct genl_family family = {
44 .id = GENL_ID_GENERATE,
45 .name = TASKSTATS_GENL_NAME,
46 .version = TASKSTATS_GENL_VERSION,
47 .maxattr = TASKSTATS_CMD_ATTR_MAX,
50 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
51 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
52 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
53 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
54 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
56 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
57 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
61 struct list_head list;
66 struct listener_list {
67 struct rw_semaphore sem;
68 struct list_head list;
70 static DEFINE_PER_CPU(struct listener_list, listener_array);
78 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
85 * If new attributes are added, please revisit this allocation
87 skb = genlmsg_new(size, GFP_KERNEL);
92 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
94 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
96 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
107 * Send taskstats data in @skb to listener with nl_pid @pid
109 static int send_reply(struct sk_buff *skb, struct genl_info *info)
111 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
112 void *reply = genlmsg_data(genlhdr);
115 rc = genlmsg_end(skb, reply);
121 return genlmsg_reply(skb, info);
125 * Send taskstats data in @skb to listeners registered for @cpu's exit data
127 static void send_cpu_listeners(struct sk_buff *skb,
128 struct listener_list *listeners)
130 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
131 struct listener *s, *tmp;
132 struct sk_buff *skb_next, *skb_cur = skb;
133 void *reply = genlmsg_data(genlhdr);
134 int rc, delcount = 0;
136 rc = genlmsg_end(skb, reply);
143 down_read(&listeners->sem);
144 list_for_each_entry(s, &listeners->list, list) {
146 if (!list_is_last(&s->list, &listeners->list)) {
147 skb_next = skb_clone(skb_cur, GFP_KERNEL);
151 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
152 if (rc == -ECONNREFUSED) {
158 up_read(&listeners->sem);
166 /* Delete invalidated entries */
167 down_write(&listeners->sem);
168 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
174 up_write(&listeners->sem);
177 static void fill_stats(struct task_struct *tsk, struct taskstats *stats)
179 memset(stats, 0, sizeof(*stats));
181 * Each accounting subsystem adds calls to its functions to
182 * fill in relevant parts of struct taskstsats as follows
184 * per-task-foo(stats, tsk);
187 delayacct_add_tsk(stats, tsk);
189 /* fill in basic acct fields */
190 stats->version = TASKSTATS_VERSION;
191 stats->nvcsw = tsk->nvcsw;
192 stats->nivcsw = tsk->nivcsw;
193 bacct_add_tsk(stats, tsk);
195 /* fill in extended acct fields */
196 xacct_add_tsk(stats, tsk);
199 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
201 struct task_struct *tsk;
204 tsk = find_task_by_vpid(pid);
206 get_task_struct(tsk);
210 fill_stats(tsk, stats);
211 put_task_struct(tsk);
215 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
217 struct task_struct *tsk, *first;
222 * Add additional stats from live tasks except zombie thread group
223 * leaders who are already counted with the dead tasks
226 first = find_task_by_vpid(tgid);
228 if (!first || !lock_task_sighand(first, &flags))
231 if (first->signal->stats)
232 memcpy(stats, first->signal->stats, sizeof(*stats));
234 memset(stats, 0, sizeof(*stats));
241 * Accounting subsystem can call its functions here to
242 * fill in relevant parts of struct taskstsats as follows
244 * per-task-foo(stats, tsk);
246 delayacct_add_tsk(stats, tsk);
248 stats->nvcsw += tsk->nvcsw;
249 stats->nivcsw += tsk->nivcsw;
250 } while_each_thread(first, tsk);
252 unlock_task_sighand(first, &flags);
257 stats->version = TASKSTATS_VERSION;
259 * Accounting subsystems can also add calls here to modify
260 * fields of taskstats.
265 static void fill_tgid_exit(struct task_struct *tsk)
269 spin_lock_irqsave(&tsk->sighand->siglock, flags);
270 if (!tsk->signal->stats)
274 * Each accounting subsystem calls its functions here to
275 * accumalate its per-task stats for tsk, into the per-tgid structure
277 * per-task-foo(tsk->signal->stats, tsk);
279 delayacct_add_tsk(tsk->signal->stats, tsk);
281 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
285 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
287 struct listener_list *listeners;
288 struct listener *s, *tmp;
291 if (!cpumask_subset(mask, cpu_possible_mask))
294 if (isadd == REGISTER) {
295 for_each_cpu(cpu, mask) {
296 s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
301 INIT_LIST_HEAD(&s->list);
304 listeners = &per_cpu(listener_array, cpu);
305 down_write(&listeners->sem);
306 list_add(&s->list, &listeners->list);
307 up_write(&listeners->sem);
312 /* Deregister or cleanup */
314 for_each_cpu(cpu, mask) {
315 listeners = &per_cpu(listener_array, cpu);
316 down_write(&listeners->sem);
317 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
324 up_write(&listeners->sem);
329 static int parse(struct nlattr *na, struct cpumask *mask)
338 if (len > TASKSTATS_CPUMASK_MAXLEN)
342 data = kmalloc(len, GFP_KERNEL);
345 nla_strlcpy(data, na, len);
346 ret = cpulist_parse(data, mask);
351 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
353 struct nlattr *na, *ret;
356 /* If we don't pad, we end up with alignment on a 4 byte boundary.
357 * This causes lots of runtime warnings on systems requiring 8 byte
359 u32 pids[2] = { pid, 0 };
360 int pid_size = ALIGN(sizeof(pid), sizeof(long));
362 aggr = (type == TASKSTATS_TYPE_PID)
363 ? TASKSTATS_TYPE_AGGR_PID
364 : TASKSTATS_TYPE_AGGR_TGID;
366 na = nla_nest_start(skb, aggr);
369 if (nla_put(skb, type, pid_size, pids) < 0)
371 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
374 nla_nest_end(skb, na);
376 return nla_data(ret);
381 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
384 struct sk_buff *rep_skb;
385 struct cgroupstats *stats;
392 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
396 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
397 file = fget_light(fd, &fput_needed);
401 size = nla_total_size(sizeof(struct cgroupstats));
403 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
408 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
409 sizeof(struct cgroupstats));
410 stats = nla_data(na);
411 memset(stats, 0, sizeof(*stats));
413 rc = cgroupstats_build(stats, file->f_dentry);
419 rc = send_reply(rep_skb, info);
422 fput_light(file, fput_needed);
426 static int cmd_attr_register_cpumask(struct genl_info *info)
431 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
433 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
436 rc = add_del_listener(info->snd_pid, mask, REGISTER);
438 free_cpumask_var(mask);
442 static int cmd_attr_deregister_cpumask(struct genl_info *info)
447 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
449 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
452 rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
454 free_cpumask_var(mask);
458 static int cmd_attr_pid(struct genl_info *info)
460 struct taskstats *stats;
461 struct sk_buff *rep_skb;
466 size = nla_total_size(sizeof(u32)) +
467 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
469 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
474 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
475 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
479 rc = fill_stats_for_pid(pid, stats);
482 return send_reply(rep_skb, info);
488 static int cmd_attr_tgid(struct genl_info *info)
490 struct taskstats *stats;
491 struct sk_buff *rep_skb;
496 size = nla_total_size(sizeof(u32)) +
497 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
499 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
504 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
505 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
509 rc = fill_stats_for_tgid(tgid, stats);
512 return send_reply(rep_skb, info);
518 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
520 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
521 return cmd_attr_register_cpumask(info);
522 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
523 return cmd_attr_deregister_cpumask(info);
524 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
525 return cmd_attr_pid(info);
526 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
527 return cmd_attr_tgid(info);
532 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
534 struct signal_struct *sig = tsk->signal;
535 struct taskstats *stats;
537 if (sig->stats || thread_group_empty(tsk))
540 /* No problem if kmem_cache_zalloc() fails */
541 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
543 spin_lock_irq(&tsk->sighand->siglock);
548 spin_unlock_irq(&tsk->sighand->siglock);
551 kmem_cache_free(taskstats_cache, stats);
556 /* Send pid data out on exit */
557 void taskstats_exit(struct task_struct *tsk, int group_dead)
560 struct listener_list *listeners;
561 struct taskstats *stats;
562 struct sk_buff *rep_skb;
566 if (!family_registered)
570 * Size includes space for nested attributes
572 size = nla_total_size(sizeof(u32)) +
573 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
575 is_thread_group = !!taskstats_tgid_alloc(tsk);
576 if (is_thread_group) {
577 /* PID + STATS + TGID + STATS */
579 /* fill the tsk->signal->stats structure */
583 listeners = __this_cpu_ptr(&listener_array);
584 if (list_empty(&listeners->list))
587 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
591 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
595 fill_stats(tsk, stats);
598 * Doesn't matter if tsk is the leader or the last group member leaving
600 if (!is_thread_group || !group_dead)
603 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
607 memcpy(stats, tsk->signal->stats, sizeof(*stats));
610 send_cpu_listeners(rep_skb, listeners);
616 static struct genl_ops taskstats_ops = {
617 .cmd = TASKSTATS_CMD_GET,
618 .doit = taskstats_user_cmd,
619 .policy = taskstats_cmd_get_policy,
622 static struct genl_ops cgroupstats_ops = {
623 .cmd = CGROUPSTATS_CMD_GET,
624 .doit = cgroupstats_user_cmd,
625 .policy = cgroupstats_cmd_get_policy,
628 /* Needed early in initialization */
629 void __init taskstats_init_early(void)
633 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
634 for_each_possible_cpu(i) {
635 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
636 init_rwsem(&(per_cpu(listener_array, i).sem));
640 static int __init taskstats_init(void)
644 rc = genl_register_family(&family);
648 rc = genl_register_ops(&family, &taskstats_ops);
652 rc = genl_register_ops(&family, &cgroupstats_ops);
656 family_registered = 1;
657 printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
660 genl_unregister_ops(&family, &taskstats_ops);
662 genl_unregister_family(&family);
667 * late initcall ensures initialization of statistics collection
668 * mechanisms precedes initialization of the taskstats interface
670 late_initcall(taskstats_init);