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 <linux/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, *s2;
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),
297 GFP_KERNEL, cpu_to_node(cpu));
304 listeners = &per_cpu(listener_array, cpu);
305 down_write(&listeners->sem);
306 list_for_each_entry(s2, &listeners->list, list) {
307 if (s2->pid == pid && s2->valid)
310 list_add(&s->list, &listeners->list);
313 up_write(&listeners->sem);
314 kfree(s); /* nop if NULL */
319 /* Deregister or cleanup */
321 for_each_cpu(cpu, mask) {
322 listeners = &per_cpu(listener_array, cpu);
323 down_write(&listeners->sem);
324 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
331 up_write(&listeners->sem);
336 static int parse(struct nlattr *na, struct cpumask *mask)
345 if (len > TASKSTATS_CPUMASK_MAXLEN)
349 data = kmalloc(len, GFP_KERNEL);
352 nla_strlcpy(data, na, len);
353 ret = cpulist_parse(data, mask);
358 #if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
359 #define TASKSTATS_NEEDS_PADDING 1
362 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
364 struct nlattr *na, *ret;
367 aggr = (type == TASKSTATS_TYPE_PID)
368 ? TASKSTATS_TYPE_AGGR_PID
369 : TASKSTATS_TYPE_AGGR_TGID;
372 * The taskstats structure is internally aligned on 8 byte
373 * boundaries but the layout of the aggregrate reply, with
374 * two NLA headers and the pid (each 4 bytes), actually
375 * force the entire structure to be unaligned. This causes
376 * the kernel to issue unaligned access warnings on some
377 * architectures like ia64. Unfortunately, some software out there
378 * doesn't properly unroll the NLA packet and assumes that the start
379 * of the taskstats structure will always be 20 bytes from the start
380 * of the netlink payload. Aligning the start of the taskstats
381 * structure breaks this software, which we don't want. So, for now
382 * the alignment only happens on architectures that require it
383 * and those users will have to update to fixed versions of those
384 * packages. Space is reserved in the packet only when needed.
385 * This ifdef should be removed in several years e.g. 2012 once
386 * we can be confident that fixed versions are installed on most
387 * systems. We add the padding before the aggregate since the
388 * aggregate is already a defined type.
390 #ifdef TASKSTATS_NEEDS_PADDING
391 if (nla_put(skb, TASKSTATS_TYPE_NULL, 0, NULL) < 0)
394 na = nla_nest_start(skb, aggr);
398 if (nla_put(skb, type, sizeof(pid), &pid) < 0)
400 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
403 nla_nest_end(skb, na);
405 return nla_data(ret);
410 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
413 struct sk_buff *rep_skb;
414 struct cgroupstats *stats;
421 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
425 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
426 file = fget_light(fd, &fput_needed);
430 size = nla_total_size(sizeof(struct cgroupstats));
432 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
437 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
438 sizeof(struct cgroupstats));
444 stats = nla_data(na);
445 memset(stats, 0, sizeof(*stats));
447 rc = cgroupstats_build(stats, file->f_dentry);
453 rc = send_reply(rep_skb, info);
456 fput_light(file, fput_needed);
460 static int cmd_attr_register_cpumask(struct genl_info *info)
465 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
467 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
470 rc = add_del_listener(info->snd_pid, mask, REGISTER);
472 free_cpumask_var(mask);
476 static int cmd_attr_deregister_cpumask(struct genl_info *info)
481 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
483 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
486 rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
488 free_cpumask_var(mask);
492 static size_t taskstats_packet_size(void)
496 size = nla_total_size(sizeof(u32)) +
497 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
498 #ifdef TASKSTATS_NEEDS_PADDING
499 size += nla_total_size(0); /* Padding for alignment */
504 static int cmd_attr_pid(struct genl_info *info)
506 struct taskstats *stats;
507 struct sk_buff *rep_skb;
512 size = taskstats_packet_size();
514 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
519 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
520 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
524 rc = fill_stats_for_pid(pid, stats);
527 return send_reply(rep_skb, info);
533 static int cmd_attr_tgid(struct genl_info *info)
535 struct taskstats *stats;
536 struct sk_buff *rep_skb;
541 size = taskstats_packet_size();
543 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
548 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
549 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
553 rc = fill_stats_for_tgid(tgid, stats);
556 return send_reply(rep_skb, info);
562 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
564 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
565 return cmd_attr_register_cpumask(info);
566 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
567 return cmd_attr_deregister_cpumask(info);
568 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
569 return cmd_attr_pid(info);
570 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
571 return cmd_attr_tgid(info);
576 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
578 struct signal_struct *sig = tsk->signal;
579 struct taskstats *stats;
581 if (sig->stats || thread_group_empty(tsk))
584 /* No problem if kmem_cache_zalloc() fails */
585 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
587 spin_lock_irq(&tsk->sighand->siglock);
592 spin_unlock_irq(&tsk->sighand->siglock);
595 kmem_cache_free(taskstats_cache, stats);
600 /* Send pid data out on exit */
601 void taskstats_exit(struct task_struct *tsk, int group_dead)
604 struct listener_list *listeners;
605 struct taskstats *stats;
606 struct sk_buff *rep_skb;
610 if (!family_registered)
614 * Size includes space for nested attributes
616 size = taskstats_packet_size();
618 is_thread_group = !!taskstats_tgid_alloc(tsk);
619 if (is_thread_group) {
620 /* PID + STATS + TGID + STATS */
622 /* fill the tsk->signal->stats structure */
626 listeners = __this_cpu_ptr(&listener_array);
627 if (list_empty(&listeners->list))
630 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
634 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
638 fill_stats(tsk, stats);
641 * Doesn't matter if tsk is the leader or the last group member leaving
643 if (!is_thread_group || !group_dead)
646 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
650 memcpy(stats, tsk->signal->stats, sizeof(*stats));
653 send_cpu_listeners(rep_skb, listeners);
659 static struct genl_ops taskstats_ops = {
660 .cmd = TASKSTATS_CMD_GET,
661 .doit = taskstats_user_cmd,
662 .policy = taskstats_cmd_get_policy,
663 .flags = GENL_ADMIN_PERM,
666 static struct genl_ops cgroupstats_ops = {
667 .cmd = CGROUPSTATS_CMD_GET,
668 .doit = cgroupstats_user_cmd,
669 .policy = cgroupstats_cmd_get_policy,
672 /* Needed early in initialization */
673 void __init taskstats_init_early(void)
677 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
678 for_each_possible_cpu(i) {
679 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
680 init_rwsem(&(per_cpu(listener_array, i).sem));
684 static int __init taskstats_init(void)
688 rc = genl_register_family(&family);
692 rc = genl_register_ops(&family, &taskstats_ops);
696 rc = genl_register_ops(&family, &cgroupstats_ops);
700 family_registered = 1;
701 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
704 genl_unregister_ops(&family, &taskstats_ops);
706 genl_unregister_family(&family);
711 * late initcall ensures initialization of statistics collection
712 * mechanisms precedes initialization of the taskstats interface
714 late_initcall(taskstats_init);