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/tsacct_kern.h>
24 #include <linux/cpumask.h>
25 #include <linux/percpu.h>
26 #include <net/genetlink.h>
27 #include <asm/atomic.h>
30 * Maximum length of a cpumask that can be specified in
31 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
33 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
35 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
36 static int family_registered;
37 struct kmem_cache *taskstats_cache;
39 static struct genl_family family = {
40 .id = GENL_ID_GENERATE,
41 .name = TASKSTATS_GENL_NAME,
42 .version = TASKSTATS_GENL_VERSION,
43 .maxattr = TASKSTATS_CMD_ATTR_MAX,
46 static 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 struct list_head list;
59 struct listener_list {
60 struct rw_semaphore sem;
61 struct list_head list;
63 static DEFINE_PER_CPU(struct listener_list, listener_array);
71 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
78 * If new attributes are added, please revisit this allocation
80 skb = genlmsg_new(size, GFP_KERNEL);
85 int seq = get_cpu_var(taskstats_seqnum)++;
86 put_cpu_var(taskstats_seqnum);
88 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
90 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
101 * Send taskstats data in @skb to listener with nl_pid @pid
103 static int send_reply(struct sk_buff *skb, pid_t pid)
105 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
106 void *reply = genlmsg_data(genlhdr);
109 rc = genlmsg_end(skb, reply);
115 return genlmsg_unicast(skb, pid);
119 * Send taskstats data in @skb to listeners registered for @cpu's exit data
121 static void send_cpu_listeners(struct sk_buff *skb,
122 struct listener_list *listeners)
124 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
125 struct listener *s, *tmp;
126 struct sk_buff *skb_next, *skb_cur = skb;
127 void *reply = genlmsg_data(genlhdr);
128 int rc, delcount = 0;
130 rc = genlmsg_end(skb, reply);
137 down_read(&listeners->sem);
138 list_for_each_entry(s, &listeners->list, list) {
140 if (!list_is_last(&s->list, &listeners->list)) {
141 skb_next = skb_clone(skb_cur, GFP_KERNEL);
145 rc = genlmsg_unicast(skb_cur, s->pid);
146 if (rc == -ECONNREFUSED) {
152 up_read(&listeners->sem);
160 /* Delete invalidated entries */
161 down_write(&listeners->sem);
162 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
168 up_write(&listeners->sem);
171 static int fill_pid(pid_t pid, struct task_struct *tsk,
172 struct taskstats *stats)
178 tsk = find_task_by_pid(pid);
180 get_task_struct(tsk);
185 get_task_struct(tsk);
187 memset(stats, 0, sizeof(*stats));
189 * Each accounting subsystem adds calls to its functions to
190 * fill in relevant parts of struct taskstsats as follows
192 * per-task-foo(stats, tsk);
195 delayacct_add_tsk(stats, tsk);
197 /* fill in basic acct fields */
198 stats->version = TASKSTATS_VERSION;
199 stats->nvcsw = tsk->nvcsw;
200 stats->nivcsw = tsk->nivcsw;
201 bacct_add_tsk(stats, tsk);
203 /* fill in extended acct fields */
204 xacct_add_tsk(stats, tsk);
206 /* Define err: label here if needed */
207 put_task_struct(tsk);
212 static int fill_tgid(pid_t tgid, struct task_struct *first,
213 struct taskstats *stats)
215 struct task_struct *tsk;
220 * Add additional stats from live tasks except zombie thread group
221 * leaders who are already counted with the dead tasks
225 first = find_task_by_pid(tgid);
227 if (!first || !lock_task_sighand(first, &flags))
230 if (first->signal->stats)
231 memcpy(stats, first->signal->stats, sizeof(*stats));
233 memset(stats, 0, sizeof(*stats));
240 * Accounting subsystem can call its functions here to
241 * fill in relevant parts of struct taskstsats as follows
243 * per-task-foo(stats, tsk);
245 delayacct_add_tsk(stats, tsk);
247 stats->nvcsw += tsk->nvcsw;
248 stats->nivcsw += tsk->nivcsw;
249 } while_each_thread(first, tsk);
251 unlock_task_sighand(first, &flags);
256 stats->version = TASKSTATS_VERSION;
258 * Accounting subsytems can also add calls here to modify
259 * 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, cpumask_t *maskp, int isadd)
287 struct listener_list *listeners;
288 struct listener *s, *tmp;
290 cpumask_t mask = *maskp;
292 if (!cpus_subset(mask, cpu_possible_map))
295 if (isadd == REGISTER) {
296 for_each_cpu_mask(cpu, mask) {
297 s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
302 INIT_LIST_HEAD(&s->list);
305 listeners = &per_cpu(listener_array, cpu);
306 down_write(&listeners->sem);
307 list_add(&s->list, &listeners->list);
308 up_write(&listeners->sem);
313 /* Deregister or cleanup */
315 for_each_cpu_mask(cpu, mask) {
316 listeners = &per_cpu(listener_array, cpu);
317 down_write(&listeners->sem);
318 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
325 up_write(&listeners->sem);
330 static int parse(struct nlattr *na, cpumask_t *mask)
339 if (len > TASKSTATS_CPUMASK_MAXLEN)
343 data = kmalloc(len, GFP_KERNEL);
346 nla_strlcpy(data, na, len);
347 ret = cpulist_parse(data, *mask);
352 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
354 struct nlattr *na, *ret;
357 aggr = (type == TASKSTATS_TYPE_PID)
358 ? TASKSTATS_TYPE_AGGR_PID
359 : TASKSTATS_TYPE_AGGR_TGID;
361 na = nla_nest_start(skb, aggr);
364 if (nla_put(skb, type, sizeof(pid), &pid) < 0)
366 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
369 nla_nest_end(skb, na);
371 return nla_data(ret);
376 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
379 struct sk_buff *rep_skb;
380 struct taskstats *stats;
384 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
388 return add_del_listener(info->snd_pid, &mask, REGISTER);
390 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
394 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
397 * Size includes space for nested attributes
399 size = nla_total_size(sizeof(u32)) +
400 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
402 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
407 if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
408 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
409 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
413 rc = fill_pid(pid, NULL, stats);
416 } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
417 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
418 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
422 rc = fill_tgid(tgid, NULL, stats);
428 return send_reply(rep_skb, info->snd_pid);
434 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
436 struct signal_struct *sig = tsk->signal;
437 struct taskstats *stats;
439 if (sig->stats || thread_group_empty(tsk))
442 /* No problem if kmem_cache_zalloc() fails */
443 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
445 spin_lock_irq(&tsk->sighand->siglock);
450 spin_unlock_irq(&tsk->sighand->siglock);
453 kmem_cache_free(taskstats_cache, stats);
458 /* Send pid data out on exit */
459 void taskstats_exit(struct task_struct *tsk, int group_dead)
462 struct listener_list *listeners;
463 struct taskstats *stats;
464 struct sk_buff *rep_skb;
468 if (!family_registered)
472 * Size includes space for nested attributes
474 size = nla_total_size(sizeof(u32)) +
475 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
477 is_thread_group = !!taskstats_tgid_alloc(tsk);
478 if (is_thread_group) {
479 /* PID + STATS + TGID + STATS */
481 /* fill the tsk->signal->stats structure */
485 listeners = &__raw_get_cpu_var(listener_array);
486 if (list_empty(&listeners->list))
489 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
493 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
497 rc = fill_pid(tsk->pid, tsk, stats);
502 * Doesn't matter if tsk is the leader or the last group member leaving
504 if (!is_thread_group || !group_dead)
507 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
511 memcpy(stats, tsk->signal->stats, sizeof(*stats));
514 send_cpu_listeners(rep_skb, listeners);
520 static struct genl_ops taskstats_ops = {
521 .cmd = TASKSTATS_CMD_GET,
522 .doit = taskstats_user_cmd,
523 .policy = taskstats_cmd_get_policy,
526 /* Needed early in initialization */
527 void __init taskstats_init_early(void)
531 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
532 for_each_possible_cpu(i) {
533 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
534 init_rwsem(&(per_cpu(listener_array, i).sem));
538 static int __init taskstats_init(void)
542 rc = genl_register_family(&family);
546 rc = genl_register_ops(&family, &taskstats_ops);
550 family_registered = 1;
553 genl_unregister_family(&family);
558 * late initcall ensures initialization of statistics collection
559 * mechanisms precedes initialization of the taskstats interface
561 late_initcall(taskstats_init);