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>
35 * Maximum length of a cpumask that can be specified in
36 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
38 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
40 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
41 static int family_registered;
42 struct kmem_cache *taskstats_cache;
44 static struct genl_family family = {
45 .id = GENL_ID_GENERATE,
46 .name = TASKSTATS_GENL_NAME,
47 .version = TASKSTATS_GENL_VERSION,
48 .maxattr = TASKSTATS_CMD_ATTR_MAX,
51 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
52 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
53 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
54 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
55 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
57 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_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);
116 rc = genlmsg_end(skb, reply);
122 return genlmsg_reply(skb, info);
126 * Send taskstats data in @skb to listeners registered for @cpu's exit data
128 static void send_cpu_listeners(struct sk_buff *skb,
129 struct listener_list *listeners)
131 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
132 struct listener *s, *tmp;
133 struct sk_buff *skb_next, *skb_cur = skb;
134 void *reply = genlmsg_data(genlhdr);
135 int rc, delcount = 0;
137 rc = genlmsg_end(skb, reply);
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(&init_net, 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 void fill_stats(struct user_namespace *user_ns,
179 struct pid_namespace *pid_ns,
180 struct task_struct *tsk, struct taskstats *stats)
182 memset(stats, 0, sizeof(*stats));
184 * Each accounting subsystem adds calls to its functions to
185 * fill in relevant parts of struct taskstsats as follows
187 * per-task-foo(stats, tsk);
190 delayacct_add_tsk(stats, tsk);
192 /* fill in basic acct fields */
193 stats->version = TASKSTATS_VERSION;
194 stats->nvcsw = tsk->nvcsw;
195 stats->nivcsw = tsk->nivcsw;
196 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
198 /* fill in extended acct fields */
199 xacct_add_tsk(stats, tsk);
202 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
204 struct task_struct *tsk;
207 tsk = find_task_by_vpid(pid);
209 get_task_struct(tsk);
213 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
214 put_task_struct(tsk);
218 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
220 struct task_struct *tsk, *first;
225 * Add additional stats from live tasks except zombie thread group
226 * leaders who are already counted with the dead tasks
229 first = find_task_by_vpid(tgid);
231 if (!first || !lock_task_sighand(first, &flags))
234 if (first->signal->stats)
235 memcpy(stats, first->signal->stats, sizeof(*stats));
237 memset(stats, 0, sizeof(*stats));
244 * Accounting subsystem can call its functions here to
245 * fill in relevant parts of struct taskstsats as follows
247 * per-task-foo(stats, tsk);
249 delayacct_add_tsk(stats, tsk);
251 stats->nvcsw += tsk->nvcsw;
252 stats->nivcsw += tsk->nivcsw;
253 } while_each_thread(first, tsk);
255 unlock_task_sighand(first, &flags);
260 stats->version = TASKSTATS_VERSION;
262 * Accounting subsystems can also add calls here to modify
263 * fields of taskstats.
268 static void fill_tgid_exit(struct task_struct *tsk)
272 spin_lock_irqsave(&tsk->sighand->siglock, flags);
273 if (!tsk->signal->stats)
277 * Each accounting subsystem calls its functions here to
278 * accumalate its per-task stats for tsk, into the per-tgid structure
280 * per-task-foo(tsk->signal->stats, tsk);
282 delayacct_add_tsk(tsk->signal->stats, tsk);
284 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
288 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
290 struct listener_list *listeners;
291 struct listener *s, *tmp, *s2;
294 if (!cpumask_subset(mask, cpu_possible_mask))
297 if (current_user_ns() != &init_user_ns)
300 if (task_active_pid_ns(current) != &init_pid_ns)
303 if (isadd == REGISTER) {
304 for_each_cpu(cpu, mask) {
305 s = kmalloc_node(sizeof(struct listener),
306 GFP_KERNEL, cpu_to_node(cpu));
313 listeners = &per_cpu(listener_array, cpu);
314 down_write(&listeners->sem);
315 list_for_each_entry(s2, &listeners->list, list) {
316 if (s2->pid == pid && s2->valid)
319 list_add(&s->list, &listeners->list);
322 up_write(&listeners->sem);
323 kfree(s); /* nop if NULL */
328 /* Deregister or cleanup */
330 for_each_cpu(cpu, mask) {
331 listeners = &per_cpu(listener_array, cpu);
332 down_write(&listeners->sem);
333 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
340 up_write(&listeners->sem);
345 static int parse(struct nlattr *na, struct cpumask *mask)
354 if (len > TASKSTATS_CPUMASK_MAXLEN)
358 data = kmalloc(len, GFP_KERNEL);
361 nla_strlcpy(data, na, len);
362 ret = cpulist_parse(data, mask);
367 #if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
368 #define TASKSTATS_NEEDS_PADDING 1
371 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
373 struct nlattr *na, *ret;
376 aggr = (type == TASKSTATS_TYPE_PID)
377 ? TASKSTATS_TYPE_AGGR_PID
378 : TASKSTATS_TYPE_AGGR_TGID;
381 * The taskstats structure is internally aligned on 8 byte
382 * boundaries but the layout of the aggregrate reply, with
383 * two NLA headers and the pid (each 4 bytes), actually
384 * force the entire structure to be unaligned. This causes
385 * the kernel to issue unaligned access warnings on some
386 * architectures like ia64. Unfortunately, some software out there
387 * doesn't properly unroll the NLA packet and assumes that the start
388 * of the taskstats structure will always be 20 bytes from the start
389 * of the netlink payload. Aligning the start of the taskstats
390 * structure breaks this software, which we don't want. So, for now
391 * the alignment only happens on architectures that require it
392 * and those users will have to update to fixed versions of those
393 * packages. Space is reserved in the packet only when needed.
394 * This ifdef should be removed in several years e.g. 2012 once
395 * we can be confident that fixed versions are installed on most
396 * systems. We add the padding before the aggregate since the
397 * aggregate is already a defined type.
399 #ifdef TASKSTATS_NEEDS_PADDING
400 if (nla_put(skb, TASKSTATS_TYPE_NULL, 0, NULL) < 0)
403 na = nla_nest_start(skb, aggr);
407 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
408 nla_nest_cancel(skb, na);
411 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
413 nla_nest_cancel(skb, na);
416 nla_nest_end(skb, na);
418 return nla_data(ret);
423 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
426 struct sk_buff *rep_skb;
427 struct cgroupstats *stats;
433 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
437 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
442 size = nla_total_size(sizeof(struct cgroupstats));
444 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
449 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
450 sizeof(struct cgroupstats));
457 stats = nla_data(na);
458 memset(stats, 0, sizeof(*stats));
460 rc = cgroupstats_build(stats, f.file->f_dentry);
466 rc = send_reply(rep_skb, info);
473 static int cmd_attr_register_cpumask(struct genl_info *info)
478 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
480 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
483 rc = add_del_listener(info->snd_portid, mask, REGISTER);
485 free_cpumask_var(mask);
489 static int cmd_attr_deregister_cpumask(struct genl_info *info)
494 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
496 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
499 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
501 free_cpumask_var(mask);
505 static size_t taskstats_packet_size(void)
509 size = nla_total_size(sizeof(u32)) +
510 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
511 #ifdef TASKSTATS_NEEDS_PADDING
512 size += nla_total_size(0); /* Padding for alignment */
517 static int cmd_attr_pid(struct genl_info *info)
519 struct taskstats *stats;
520 struct sk_buff *rep_skb;
525 size = taskstats_packet_size();
527 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
532 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
533 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
537 rc = fill_stats_for_pid(pid, stats);
540 return send_reply(rep_skb, info);
546 static int cmd_attr_tgid(struct genl_info *info)
548 struct taskstats *stats;
549 struct sk_buff *rep_skb;
554 size = taskstats_packet_size();
556 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
561 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
562 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
566 rc = fill_stats_for_tgid(tgid, stats);
569 return send_reply(rep_skb, info);
575 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
577 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
578 return cmd_attr_register_cpumask(info);
579 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
580 return cmd_attr_deregister_cpumask(info);
581 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
582 return cmd_attr_pid(info);
583 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
584 return cmd_attr_tgid(info);
589 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
591 struct signal_struct *sig = tsk->signal;
592 struct taskstats *stats;
594 if (sig->stats || thread_group_empty(tsk))
597 /* No problem if kmem_cache_zalloc() fails */
598 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
600 spin_lock_irq(&tsk->sighand->siglock);
605 spin_unlock_irq(&tsk->sighand->siglock);
608 kmem_cache_free(taskstats_cache, stats);
613 /* Send pid data out on exit */
614 void taskstats_exit(struct task_struct *tsk, int group_dead)
617 struct listener_list *listeners;
618 struct taskstats *stats;
619 struct sk_buff *rep_skb;
623 if (!family_registered)
627 * Size includes space for nested attributes
629 size = taskstats_packet_size();
631 is_thread_group = !!taskstats_tgid_alloc(tsk);
632 if (is_thread_group) {
633 /* PID + STATS + TGID + STATS */
635 /* fill the tsk->signal->stats structure */
639 listeners = __this_cpu_ptr(&listener_array);
640 if (list_empty(&listeners->list))
643 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
647 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
648 task_pid_nr_ns(tsk, &init_pid_ns));
652 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
655 * Doesn't matter if tsk is the leader or the last group member leaving
657 if (!is_thread_group || !group_dead)
660 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
661 task_tgid_nr_ns(tsk, &init_pid_ns));
665 memcpy(stats, tsk->signal->stats, sizeof(*stats));
668 send_cpu_listeners(rep_skb, listeners);
674 static struct genl_ops taskstats_ops = {
675 .cmd = TASKSTATS_CMD_GET,
676 .doit = taskstats_user_cmd,
677 .policy = taskstats_cmd_get_policy,
678 .flags = GENL_ADMIN_PERM,
681 static struct genl_ops cgroupstats_ops = {
682 .cmd = CGROUPSTATS_CMD_GET,
683 .doit = cgroupstats_user_cmd,
684 .policy = cgroupstats_cmd_get_policy,
687 /* Needed early in initialization */
688 void __init taskstats_init_early(void)
692 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
693 for_each_possible_cpu(i) {
694 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
695 init_rwsem(&(per_cpu(listener_array, i).sem));
699 static int __init taskstats_init(void)
703 rc = genl_register_family(&family);
707 rc = genl_register_ops(&family, &taskstats_ops);
711 rc = genl_register_ops(&family, &cgroupstats_ops);
715 family_registered = 1;
716 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
719 genl_unregister_ops(&family, &taskstats_ops);
721 genl_unregister_family(&family);
726 * late initcall ensures initialization of statistics collection
727 * mechanisms precedes initialization of the taskstats interface
729 late_initcall(taskstats_init);