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/cgroupstats.h>
26 #include <linux/cgroup.h>
28 #include <linux/file.h>
29 #include <net/genetlink.h>
30 #include <asm/atomic.h>
33 * Maximum length of a cpumask that can be specified in
34 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
36 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
38 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
39 static int family_registered;
40 struct kmem_cache *taskstats_cache;
42 static struct genl_family family = {
43 .id = GENL_ID_GENERATE,
44 .name = TASKSTATS_GENL_NAME,
45 .version = TASKSTATS_GENL_VERSION,
46 .maxattr = TASKSTATS_CMD_ATTR_MAX,
49 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
50 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
51 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
52 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
53 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
55 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
56 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
60 struct list_head list;
65 struct listener_list {
66 struct rw_semaphore sem;
67 struct list_head list;
69 static DEFINE_PER_CPU(struct listener_list, listener_array);
77 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
84 * If new attributes are added, please revisit this allocation
86 skb = genlmsg_new(size, GFP_KERNEL);
91 int seq = get_cpu_var(taskstats_seqnum)++;
92 put_cpu_var(taskstats_seqnum);
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 int fill_pid(pid_t pid, struct task_struct *tsk,
178 struct taskstats *stats)
184 tsk = find_task_by_vpid(pid);
186 get_task_struct(tsk);
191 get_task_struct(tsk);
193 memset(stats, 0, sizeof(*stats));
195 * Each accounting subsystem adds calls to its functions to
196 * fill in relevant parts of struct taskstsats as follows
198 * per-task-foo(stats, tsk);
201 delayacct_add_tsk(stats, tsk);
203 /* fill in basic acct fields */
204 stats->version = TASKSTATS_VERSION;
205 stats->nvcsw = tsk->nvcsw;
206 stats->nivcsw = tsk->nivcsw;
207 bacct_add_tsk(stats, tsk);
209 /* fill in extended acct fields */
210 xacct_add_tsk(stats, tsk);
212 /* Define err: label here if needed */
213 put_task_struct(tsk);
218 static int fill_tgid(pid_t tgid, struct task_struct *first,
219 struct taskstats *stats)
221 struct task_struct *tsk;
226 * Add additional stats from live tasks except zombie thread group
227 * leaders who are already counted with the dead tasks
231 first = find_task_by_vpid(tgid);
233 if (!first || !lock_task_sighand(first, &flags))
236 if (first->signal->stats)
237 memcpy(stats, first->signal->stats, sizeof(*stats));
239 memset(stats, 0, sizeof(*stats));
246 * Accounting subsystem can call its functions here to
247 * fill in relevant parts of struct taskstsats as follows
249 * per-task-foo(stats, tsk);
251 delayacct_add_tsk(stats, tsk);
253 stats->nvcsw += tsk->nvcsw;
254 stats->nivcsw += tsk->nivcsw;
255 } while_each_thread(first, tsk);
257 unlock_task_sighand(first, &flags);
262 stats->version = TASKSTATS_VERSION;
264 * Accounting subsystems can also add calls here to modify
265 * fields of taskstats.
271 static void fill_tgid_exit(struct task_struct *tsk)
275 spin_lock_irqsave(&tsk->sighand->siglock, flags);
276 if (!tsk->signal->stats)
280 * Each accounting subsystem calls its functions here to
281 * accumalate its per-task stats for tsk, into the per-tgid structure
283 * per-task-foo(tsk->signal->stats, tsk);
285 delayacct_add_tsk(tsk->signal->stats, tsk);
287 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
291 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
293 struct listener_list *listeners;
294 struct listener *s, *tmp;
297 if (!cpumask_subset(mask, cpu_possible_mask))
300 if (isadd == REGISTER) {
301 for_each_cpu(cpu, mask) {
302 s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
307 INIT_LIST_HEAD(&s->list);
310 listeners = &per_cpu(listener_array, cpu);
311 down_write(&listeners->sem);
312 list_add(&s->list, &listeners->list);
313 up_write(&listeners->sem);
318 /* Deregister or cleanup */
320 for_each_cpu(cpu, mask) {
321 listeners = &per_cpu(listener_array, cpu);
322 down_write(&listeners->sem);
323 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
330 up_write(&listeners->sem);
335 static int parse(struct nlattr *na, struct cpumask *mask)
344 if (len > TASKSTATS_CPUMASK_MAXLEN)
348 data = kmalloc(len, GFP_KERNEL);
351 nla_strlcpy(data, na, len);
352 ret = cpulist_parse(data, mask);
357 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
359 struct nlattr *na, *ret;
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, sizeof(pid), &pid) < 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 taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
429 struct sk_buff *rep_skb;
430 struct taskstats *stats;
434 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
437 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
441 rc = add_del_listener(info->snd_pid, mask, REGISTER);
445 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
449 rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
451 free_cpumask_var(mask);
454 free_cpumask_var(mask);
457 * Size includes space for nested attributes
459 size = nla_total_size(sizeof(u32)) +
460 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
462 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
467 if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
468 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
469 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
473 rc = fill_pid(pid, NULL, stats);
476 } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
477 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
478 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
482 rc = fill_tgid(tgid, NULL, stats);
488 return send_reply(rep_skb, info);
494 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
496 struct signal_struct *sig = tsk->signal;
497 struct taskstats *stats;
499 if (sig->stats || thread_group_empty(tsk))
502 /* No problem if kmem_cache_zalloc() fails */
503 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
505 spin_lock_irq(&tsk->sighand->siglock);
510 spin_unlock_irq(&tsk->sighand->siglock);
513 kmem_cache_free(taskstats_cache, stats);
518 /* Send pid data out on exit */
519 void taskstats_exit(struct task_struct *tsk, int group_dead)
522 struct listener_list *listeners;
523 struct taskstats *stats;
524 struct sk_buff *rep_skb;
528 if (!family_registered)
532 * Size includes space for nested attributes
534 size = nla_total_size(sizeof(u32)) +
535 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
537 is_thread_group = !!taskstats_tgid_alloc(tsk);
538 if (is_thread_group) {
539 /* PID + STATS + TGID + STATS */
541 /* fill the tsk->signal->stats structure */
545 listeners = &__raw_get_cpu_var(listener_array);
546 if (list_empty(&listeners->list))
549 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
553 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
557 rc = fill_pid(-1, tsk, stats);
562 * Doesn't matter if tsk is the leader or the last group member leaving
564 if (!is_thread_group || !group_dead)
567 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
571 memcpy(stats, tsk->signal->stats, sizeof(*stats));
574 send_cpu_listeners(rep_skb, listeners);
580 static struct genl_ops taskstats_ops = {
581 .cmd = TASKSTATS_CMD_GET,
582 .doit = taskstats_user_cmd,
583 .policy = taskstats_cmd_get_policy,
586 static struct genl_ops cgroupstats_ops = {
587 .cmd = CGROUPSTATS_CMD_GET,
588 .doit = cgroupstats_user_cmd,
589 .policy = cgroupstats_cmd_get_policy,
592 /* Needed early in initialization */
593 void __init taskstats_init_early(void)
597 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
598 for_each_possible_cpu(i) {
599 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
600 init_rwsem(&(per_cpu(listener_array, i).sem));
604 static int __init taskstats_init(void)
608 rc = genl_register_family(&family);
612 rc = genl_register_ops(&family, &taskstats_ops);
616 rc = genl_register_ops(&family, &cgroupstats_ops);
620 family_registered = 1;
621 printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
624 genl_unregister_ops(&family, &taskstats_ops);
626 genl_unregister_family(&family);
631 * late initcall ensures initialization of statistics collection
632 * mechanisms precedes initialization of the taskstats interface
634 late_initcall(taskstats_init);