1/*
2 * taskstats.c - Export per-task statistics to userland
3 *
4 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
5 *           (C) Balbir Singh,   IBM Corp. 2006
6 *
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.
11 *
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.
16 *
17 */
18
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>
28#include <linux/fs.h>
29#include <linux/file.h>
30#include <net/genetlink.h>
31#include <asm/atomic.h>
32
33/*
34 * Maximum length of a cpumask that can be specified in
35 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
36 */
37#define TASKSTATS_CPUMASK_MAXLEN	(100+6*NR_CPUS)
38
39static DEFINE_PER_CPU(__u32, taskstats_seqnum);
40static int family_registered;
41struct kmem_cache *taskstats_cache;
42
43static 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,
48};
49
50static 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 },};
55
56static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
57	[CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
58};
59
60struct listener {
61	struct list_head list;
62	pid_t pid;
63	char valid;
64};
65
66struct listener_list {
67	struct rw_semaphore sem;
68	struct list_head list;
69};
70static DEFINE_PER_CPU(struct listener_list, listener_array);
71
72enum actions {
73	REGISTER,
74	DEREGISTER,
75	CPU_DONT_CARE
76};
77
78static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
79				size_t size)
80{
81	struct sk_buff *skb;
82	void *reply;
83
84	/*
85	 * If new attributes are added, please revisit this allocation
86	 */
87	skb = genlmsg_new(size, GFP_KERNEL);
88	if (!skb)
89		return -ENOMEM;
90
91	if (!info) {
92		int seq = get_cpu_var(taskstats_seqnum)++;
93		put_cpu_var(taskstats_seqnum);
94
95		reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
96	} else
97		reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
98	if (reply == NULL) {
99		nlmsg_free(skb);
100		return -EINVAL;
101	}
102
103	*skbp = skb;
104	return 0;
105}
106
107/*
108 * Send taskstats data in @skb to listener with nl_pid @pid
109 */
110static int send_reply(struct sk_buff *skb, struct genl_info *info)
111{
112	struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
113	void *reply = genlmsg_data(genlhdr);
114	int rc;
115
116	rc = genlmsg_end(skb, reply);
117	if (rc < 0) {
118		nlmsg_free(skb);
119		return rc;
120	}
121
122	return genlmsg_reply(skb, info);
123}
124
125/*
126 * Send taskstats data in @skb to listeners registered for @cpu's exit data
127 */
128static void send_cpu_listeners(struct sk_buff *skb,
129					struct listener_list *listeners)
130{
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;
136
137	rc = genlmsg_end(skb, reply);
138	if (rc < 0) {
139		nlmsg_free(skb);
140		return;
141	}
142
143	rc = 0;
144	down_read(&listeners->sem);
145	list_for_each_entry(s, &listeners->list, list) {
146		skb_next = NULL;
147		if (!list_is_last(&s->list, &listeners->list)) {
148			skb_next = skb_clone(skb_cur, GFP_KERNEL);
149			if (!skb_next)
150				break;
151		}
152		rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
153		if (rc == -ECONNREFUSED) {
154			s->valid = 0;
155			delcount++;
156		}
157		skb_cur = skb_next;
158	}
159	up_read(&listeners->sem);
160
161	if (skb_cur)
162		nlmsg_free(skb_cur);
163
164	if (!delcount)
165		return;
166
167	/* Delete invalidated entries */
168	down_write(&listeners->sem);
169	list_for_each_entry_safe(s, tmp, &listeners->list, list) {
170		if (!s->valid) {
171			list_del(&s->list);
172			kfree(s);
173		}
174	}
175	up_write(&listeners->sem);
176}
177
178static int fill_pid(pid_t pid, struct task_struct *tsk,
179		struct taskstats *stats)
180{
181	int rc = 0;
182
183	if (!tsk) {
184		rcu_read_lock();
185		tsk = find_task_by_vpid(pid);
186		if (tsk)
187			get_task_struct(tsk);
188		rcu_read_unlock();
189		if (!tsk)
190			return -ESRCH;
191	} else
192		get_task_struct(tsk);
193
194	memset(stats, 0, sizeof(*stats));
195	/*
196	 * Each accounting subsystem adds calls to its functions to
197	 * fill in relevant parts of struct taskstsats as follows
198	 *
199	 *	per-task-foo(stats, tsk);
200	 */
201
202	delayacct_add_tsk(stats, tsk);
203
204	/* fill in basic acct fields */
205	stats->version = TASKSTATS_VERSION;
206	stats->nvcsw = tsk->nvcsw;
207	stats->nivcsw = tsk->nivcsw;
208	bacct_add_tsk(stats, tsk);
209
210	/* fill in extended acct fields */
211	xacct_add_tsk(stats, tsk);
212
213	/* Define err: label here if needed */
214	put_task_struct(tsk);
215	return rc;
216
217}
218
219static int fill_tgid(pid_t tgid, struct task_struct *first,
220		struct taskstats *stats)
221{
222	struct task_struct *tsk;
223	unsigned long flags;
224	int rc = -ESRCH;
225
226	/*
227	 * Add additional stats from live tasks except zombie thread group
228	 * leaders who are already counted with the dead tasks
229	 */
230	rcu_read_lock();
231	if (!first)
232		first = find_task_by_vpid(tgid);
233
234	if (!first || !lock_task_sighand(first, &flags))
235		goto out;
236
237	if (first->signal->stats)
238		memcpy(stats, first->signal->stats, sizeof(*stats));
239	else
240		memset(stats, 0, sizeof(*stats));
241
242	tsk = first;
243	do {
244		if (tsk->exit_state)
245			continue;
246		/*
247		 * Accounting subsystem can call its functions here to
248		 * fill in relevant parts of struct taskstsats as follows
249		 *
250		 *	per-task-foo(stats, tsk);
251		 */
252		delayacct_add_tsk(stats, tsk);
253
254		stats->nvcsw += tsk->nvcsw;
255		stats->nivcsw += tsk->nivcsw;
256	} while_each_thread(first, tsk);
257
258	unlock_task_sighand(first, &flags);
259	rc = 0;
260out:
261	rcu_read_unlock();
262
263	stats->version = TASKSTATS_VERSION;
264	/*
265	 * Accounting subsystems can also add calls here to modify
266	 * fields of taskstats.
267	 */
268	return rc;
269}
270
271
272static void fill_tgid_exit(struct task_struct *tsk)
273{
274	unsigned long flags;
275
276	spin_lock_irqsave(&tsk->sighand->siglock, flags);
277	if (!tsk->signal->stats)
278		goto ret;
279
280	/*
281	 * Each accounting subsystem calls its functions here to
282	 * accumalate its per-task stats for tsk, into the per-tgid structure
283	 *
284	 *	per-task-foo(tsk->signal->stats, tsk);
285	 */
286	delayacct_add_tsk(tsk->signal->stats, tsk);
287ret:
288	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
289	return;
290}
291
292static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
293{
294	struct listener_list *listeners;
295	struct listener *s, *tmp;
296	unsigned int cpu;
297
298	if (!cpumask_subset(mask, cpu_possible_mask))
299		return -EINVAL;
300
301	if (isadd == REGISTER) {
302		for_each_cpu(cpu, mask) {
303			s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
304					 cpu_to_node(cpu));
305			if (!s)
306				goto cleanup;
307			s->pid = pid;
308			INIT_LIST_HEAD(&s->list);
309			s->valid = 1;
310
311			listeners = &per_cpu(listener_array, cpu);
312			down_write(&listeners->sem);
313			list_add(&s->list, &listeners->list);
314			up_write(&listeners->sem);
315		}
316		return 0;
317	}
318
319	/* Deregister or cleanup */
320cleanup:
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) {
325			if (s->pid == pid) {
326				list_del(&s->list);
327				kfree(s);
328				break;
329			}
330		}
331		up_write(&listeners->sem);
332	}
333	return 0;
334}
335
336static int parse(struct nlattr *na, struct cpumask *mask)
337{
338	char *data;
339	int len;
340	int ret;
341
342	if (na == NULL)
343		return 1;
344	len = nla_len(na);
345	if (len > TASKSTATS_CPUMASK_MAXLEN)
346		return -E2BIG;
347	if (len < 1)
348		return -EINVAL;
349	data = kmalloc(len, GFP_KERNEL);
350	if (!data)
351		return -ENOMEM;
352	nla_strlcpy(data, na, len);
353	ret = cpulist_parse(data, mask);
354	kfree(data);
355	return ret;
356}
357
358static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
359{
360	struct nlattr *na, *ret;
361	int aggr;
362
363	aggr = (type == TASKSTATS_TYPE_PID)
364			? TASKSTATS_TYPE_AGGR_PID
365			: TASKSTATS_TYPE_AGGR_TGID;
366
367	na = nla_nest_start(skb, aggr);
368	if (!na)
369		goto err;
370	if (nla_put(skb, type, sizeof(pid), &pid) < 0)
371		goto err;
372	ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
373	if (!ret)
374		goto err;
375	nla_nest_end(skb, na);
376
377	return nla_data(ret);
378err:
379	return NULL;
380}
381
382static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
383{
384	int rc = 0;
385	struct sk_buff *rep_skb;
386	struct cgroupstats *stats;
387	struct nlattr *na;
388	size_t size;
389	u32 fd;
390	struct file *file;
391	int fput_needed;
392
393	na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
394	if (!na)
395		return -EINVAL;
396
397	fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
398	file = fget_light(fd, &fput_needed);
399	if (!file)
400		return 0;
401
402	size = nla_total_size(sizeof(struct cgroupstats));
403
404	rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
405				size);
406	if (rc < 0)
407		goto err;
408
409	na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
410				sizeof(struct cgroupstats));
411	stats = nla_data(na);
412	memset(stats, 0, sizeof(*stats));
413
414	rc = cgroupstats_build(stats, file->f_dentry);
415	if (rc < 0) {
416		nlmsg_free(rep_skb);
417		goto err;
418	}
419
420	rc = send_reply(rep_skb, info);
421
422err:
423	fput_light(file, fput_needed);
424	return rc;
425}
426
427static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
428{
429	int rc;
430	struct sk_buff *rep_skb;
431	struct taskstats *stats;
432	size_t size;
433	cpumask_var_t mask;
434
435	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
436		return -ENOMEM;
437
438	rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
439	if (rc < 0)
440		goto free_return_rc;
441	if (rc == 0) {
442		rc = add_del_listener(info->snd_pid, mask, REGISTER);
443		goto free_return_rc;
444	}
445
446	rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
447	if (rc < 0)
448		goto free_return_rc;
449	if (rc == 0) {
450		rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
451free_return_rc:
452		free_cpumask_var(mask);
453		return rc;
454	}
455	free_cpumask_var(mask);
456
457	/*
458	 * Size includes space for nested attributes
459	 */
460	size = nla_total_size(sizeof(u32)) +
461		nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
462
463	rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
464	if (rc < 0)
465		return rc;
466
467	rc = -EINVAL;
468	if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
469		u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
470		stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
471		if (!stats)
472			goto err;
473
474		rc = fill_pid(pid, NULL, stats);
475		if (rc < 0)
476			goto err;
477	} else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
478		u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
479		stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
480		if (!stats)
481			goto err;
482
483		rc = fill_tgid(tgid, NULL, stats);
484		if (rc < 0)
485			goto err;
486	} else
487		goto err;
488
489	return send_reply(rep_skb, info);
490err:
491	nlmsg_free(rep_skb);
492	return rc;
493}
494
495static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
496{
497	struct signal_struct *sig = tsk->signal;
498	struct taskstats *stats;
499
500	if (sig->stats || thread_group_empty(tsk))
501		goto ret;
502
503	/* No problem if kmem_cache_zalloc() fails */
504	stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
505
506	spin_lock_irq(&tsk->sighand->siglock);
507	if (!sig->stats) {
508		sig->stats = stats;
509		stats = NULL;
510	}
511	spin_unlock_irq(&tsk->sighand->siglock);
512
513	if (stats)
514		kmem_cache_free(taskstats_cache, stats);
515ret:
516	return sig->stats;
517}
518
519/* Send pid data out on exit */
520void taskstats_exit(struct task_struct *tsk, int group_dead)
521{
522	int rc;
523	struct listener_list *listeners;
524	struct taskstats *stats;
525	struct sk_buff *rep_skb;
526	size_t size;
527	int is_thread_group;
528
529	if (!family_registered)
530		return;
531
532	/*
533	 * Size includes space for nested attributes
534	 */
535	size = nla_total_size(sizeof(u32)) +
536		nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
537
538	is_thread_group = !!taskstats_tgid_alloc(tsk);
539	if (is_thread_group) {
540		/* PID + STATS + TGID + STATS */
541		size = 2 * size;
542		/* fill the tsk->signal->stats structure */
543		fill_tgid_exit(tsk);
544	}
545
546	listeners = &__raw_get_cpu_var(listener_array);
547	if (list_empty(&listeners->list))
548		return;
549
550	rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
551	if (rc < 0)
552		return;
553
554	stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
555	if (!stats)
556		goto err;
557
558	rc = fill_pid(-1, tsk, stats);
559	if (rc < 0)
560		goto err;
561
562	/*
563	 * Doesn't matter if tsk is the leader or the last group member leaving
564	 */
565	if (!is_thread_group || !group_dead)
566		goto send;
567
568	stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
569	if (!stats)
570		goto err;
571
572	memcpy(stats, tsk->signal->stats, sizeof(*stats));
573
574send:
575	send_cpu_listeners(rep_skb, listeners);
576	return;
577err:
578	nlmsg_free(rep_skb);
579}
580
581static struct genl_ops taskstats_ops = {
582	.cmd		= TASKSTATS_CMD_GET,
583	.doit		= taskstats_user_cmd,
584	.policy		= taskstats_cmd_get_policy,
585};
586
587static struct genl_ops cgroupstats_ops = {
588	.cmd		= CGROUPSTATS_CMD_GET,
589	.doit		= cgroupstats_user_cmd,
590	.policy		= cgroupstats_cmd_get_policy,
591};
592
593/* Needed early in initialization */
594void __init taskstats_init_early(void)
595{
596	unsigned int i;
597
598	taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
599	for_each_possible_cpu(i) {
600		INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
601		init_rwsem(&(per_cpu(listener_array, i).sem));
602	}
603}
604
605static int __init taskstats_init(void)
606{
607	int rc;
608
609	rc = genl_register_family(&family);
610	if (rc)
611		return rc;
612
613	rc = genl_register_ops(&family, &taskstats_ops);
614	if (rc < 0)
615		goto err;
616
617	rc = genl_register_ops(&family, &cgroupstats_ops);
618	if (rc < 0)
619		goto err_cgroup_ops;
620
621	family_registered = 1;
622	printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
623	return 0;
624err_cgroup_ops:
625	genl_unregister_ops(&family, &taskstats_ops);
626err:
627	genl_unregister_family(&family);
628	return rc;
629}
630
631/*
632 * late initcall ensures initialization of statistics collection
633 * mechanisms precedes initialization of the taskstats interface
634 */
635late_initcall(taskstats_init);
636