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 kmem_cache_t *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,
72 void **replyp, size_t size)
78 * If new attributes are added, please revisit this allocation
80 size = nlmsg_total_size(genlmsg_total_size(size));
81 skb = nlmsg_new(size, GFP_KERNEL);
86 int seq = get_cpu_var(taskstats_seqnum)++;
87 put_cpu_var(taskstats_seqnum);
89 reply = genlmsg_put(skb, 0, seq,
93 reply = genlmsg_put(skb, info->snd_pid, info->snd_seq,
107 * Send taskstats data in @skb to listener with nl_pid @pid
109 static int send_reply(struct sk_buff *skb, pid_t pid)
111 struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
112 void *reply = genlmsg_data(genlhdr);
115 rc = genlmsg_end(skb, reply);
121 return genlmsg_unicast(skb, pid);
125 * Send taskstats data in @skb to listeners registered for @cpu's exit data
127 static void send_cpu_listeners(struct sk_buff *skb, unsigned int cpu)
129 struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
130 struct listener_list *listeners;
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 listeners = &per_cpu(listener_array, cpu);
144 down_read(&listeners->sem);
145 list_for_each_entry(s, &listeners->list, list) {
147 if (!list_is_last(&s->list, &listeners->list)) {
148 skb_next = skb_clone(skb_cur, GFP_KERNEL);
152 rc = genlmsg_unicast(skb_cur, s->pid);
153 if (rc == -ECONNREFUSED) {
159 up_read(&listeners->sem);
167 /* Delete invalidated entries */
168 down_write(&listeners->sem);
169 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
175 up_write(&listeners->sem);
178 static int fill_pid(pid_t pid, struct task_struct *tsk,
179 struct taskstats *stats)
185 tsk = find_task_by_pid(pid);
187 get_task_struct(tsk);
192 get_task_struct(tsk);
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 bacct_add_tsk(stats, tsk);
207 /* fill in extended acct fields */
208 xacct_add_tsk(stats, tsk);
210 /* Define err: label here if needed */
211 put_task_struct(tsk);
216 static int fill_tgid(pid_t tgid, struct task_struct *first,
217 struct taskstats *stats)
219 struct task_struct *tsk;
224 * Add additional stats from live tasks except zombie thread group
225 * leaders who are already counted with the dead tasks
229 first = find_task_by_pid(tgid);
231 if (!first || !lock_task_sighand(first, &flags))
234 if (first->signal->stats)
235 memcpy(stats, first->signal->stats, sizeof(*stats));
242 * Accounting subsystem can call its functions here to
243 * fill in relevant parts of struct taskstsats as follows
245 * per-task-foo(stats, tsk);
247 delayacct_add_tsk(stats, tsk);
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 int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
355 struct sk_buff *rep_skb;
356 struct taskstats stats;
362 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
366 return add_del_listener(info->snd_pid, &mask, REGISTER);
368 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
372 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
375 * Size includes space for nested attributes
377 size = nla_total_size(sizeof(u32)) +
378 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
380 memset(&stats, 0, sizeof(stats));
381 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
385 if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
386 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
387 rc = fill_pid(pid, NULL, &stats);
391 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
392 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
393 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
395 } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
396 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
397 rc = fill_tgid(tgid, NULL, &stats);
401 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
402 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
403 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
410 nla_nest_end(rep_skb, na);
412 return send_reply(rep_skb, info->snd_pid);
415 rc = genlmsg_cancel(rep_skb, reply);
421 void taskstats_exit_alloc(struct taskstats **ptidstats, unsigned int *mycpu)
423 struct listener_list *listeners;
424 struct taskstats *tmp;
426 * This is the cpu on which the task is exiting currently and will
427 * be the one for which the exit event is sent, even if the cpu
428 * on which this function is running changes later.
430 *mycpu = raw_smp_processor_id();
433 tmp = kmem_cache_zalloc(taskstats_cache, SLAB_KERNEL);
437 listeners = &per_cpu(listener_array, *mycpu);
438 down_read(&listeners->sem);
439 if (!list_empty(&listeners->list)) {
443 up_read(&listeners->sem);
447 /* Send pid data out on exit */
448 void taskstats_exit_send(struct task_struct *tsk, struct taskstats *tidstats,
449 int group_dead, unsigned int mycpu)
452 struct sk_buff *rep_skb;
458 if (!family_registered || !tidstats)
463 * Size includes space for nested attributes
465 size = nla_total_size(sizeof(u32)) +
466 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
468 is_thread_group = (tsk->signal->stats != NULL);
470 size = 2 * size; /* PID + STATS + TGID + STATS */
472 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
476 rc = fill_pid(tsk->pid, tsk, tidstats);
480 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
481 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
482 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
484 nla_nest_end(rep_skb, na);
486 if (!is_thread_group)
490 * tsk has/had a thread group so fill the tsk->signal->stats structure
491 * Doesn't matter if tsk is the leader or the last group member leaving
498 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
499 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
500 /* No locking needed for tsk->signal->stats since group is dead */
501 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
502 *tsk->signal->stats);
503 nla_nest_end(rep_skb, na);
506 send_cpu_listeners(rep_skb, mycpu);
510 genlmsg_cancel(rep_skb, reply);
517 static struct genl_ops taskstats_ops = {
518 .cmd = TASKSTATS_CMD_GET,
519 .doit = taskstats_user_cmd,
520 .policy = taskstats_cmd_get_policy,
523 /* Needed early in initialization */
524 void __init taskstats_init_early(void)
528 taskstats_cache = kmem_cache_create("taskstats_cache",
529 sizeof(struct taskstats),
530 0, SLAB_PANIC, NULL, NULL);
531 for_each_possible_cpu(i) {
532 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
533 init_rwsem(&(per_cpu(listener_array, i).sem));
537 static int __init taskstats_init(void)
541 rc = genl_register_family(&family);
545 rc = genl_register_ops(&family, &taskstats_ops);
549 family_registered = 1;
552 genl_unregister_family(&family);
557 * late initcall ensures initialization of statistics collection
558 * mechanisms precedes initialization of the taskstats interface
560 late_initcall(taskstats_init);