cfq: pass around cfq_io_cq instead of io_context
[linux-2.6.git] / kernel / taskstats.c
blobe66046456f4ffebab2ec0300e0537a1fb1911e56
1 /*
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>
28 #include <linux/fs.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 },
60 struct listener {
61 struct list_head list;
62 pid_t pid;
63 char valid;
66 struct listener_list {
67 struct rw_semaphore sem;
68 struct list_head list;
70 static DEFINE_PER_CPU(struct listener_list, listener_array);
72 enum actions {
73 REGISTER,
74 DEREGISTER,
75 CPU_DONT_CARE
78 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
79 size_t size)
81 struct sk_buff *skb;
82 void *reply;
85 * If new attributes are added, please revisit this allocation
87 skb = genlmsg_new(size, GFP_KERNEL);
88 if (!skb)
89 return -ENOMEM;
91 if (!info) {
92 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
94 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
95 } else
96 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
97 if (reply == NULL) {
98 nlmsg_free(skb);
99 return -EINVAL;
102 *skbp = skb;
103 return 0;
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);
113 int rc;
115 rc = genlmsg_end(skb, reply);
116 if (rc < 0) {
117 nlmsg_free(skb);
118 return rc;
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);
137 if (rc < 0) {
138 nlmsg_free(skb);
139 return;
142 rc = 0;
143 down_read(&listeners->sem);
144 list_for_each_entry(s, &listeners->list, list) {
145 skb_next = NULL;
146 if (!list_is_last(&s->list, &listeners->list)) {
147 skb_next = skb_clone(skb_cur, GFP_KERNEL);
148 if (!skb_next)
149 break;
151 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
152 if (rc == -ECONNREFUSED) {
153 s->valid = 0;
154 delcount++;
156 skb_cur = skb_next;
158 up_read(&listeners->sem);
160 if (skb_cur)
161 nlmsg_free(skb_cur);
163 if (!delcount)
164 return;
166 /* Delete invalidated entries */
167 down_write(&listeners->sem);
168 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
169 if (!s->valid) {
170 list_del(&s->list);
171 kfree(s);
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;
203 rcu_read_lock();
204 tsk = find_task_by_vpid(pid);
205 if (tsk)
206 get_task_struct(tsk);
207 rcu_read_unlock();
208 if (!tsk)
209 return -ESRCH;
210 fill_stats(tsk, stats);
211 put_task_struct(tsk);
212 return 0;
215 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
217 struct task_struct *tsk, *first;
218 unsigned long flags;
219 int rc = -ESRCH;
222 * Add additional stats from live tasks except zombie thread group
223 * leaders who are already counted with the dead tasks
225 rcu_read_lock();
226 first = find_task_by_vpid(tgid);
228 if (!first || !lock_task_sighand(first, &flags))
229 goto out;
231 if (first->signal->stats)
232 memcpy(stats, first->signal->stats, sizeof(*stats));
233 else
234 memset(stats, 0, sizeof(*stats));
236 tsk = first;
237 do {
238 if (tsk->exit_state)
239 continue;
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);
253 rc = 0;
254 out:
255 rcu_read_unlock();
257 stats->version = TASKSTATS_VERSION;
259 * Accounting subsystems can also add calls here to modify
260 * fields of taskstats.
262 return rc;
265 static void fill_tgid_exit(struct task_struct *tsk)
267 unsigned long flags;
269 spin_lock_irqsave(&tsk->sighand->siglock, flags);
270 if (!tsk->signal->stats)
271 goto ret;
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);
280 ret:
281 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
282 return;
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;
289 unsigned int cpu;
291 if (!cpumask_subset(mask, cpu_possible_mask))
292 return -EINVAL;
294 if (isadd == REGISTER) {
295 for_each_cpu(cpu, mask) {
296 s = kmalloc_node(sizeof(struct listener),
297 GFP_KERNEL, cpu_to_node(cpu));
298 if (!s)
299 goto cleanup;
301 s->pid = pid;
302 s->valid = 1;
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)
308 goto exists;
310 list_add(&s->list, &listeners->list);
311 s = NULL;
312 exists:
313 up_write(&listeners->sem);
314 kfree(s); /* nop if NULL */
316 return 0;
319 /* Deregister or cleanup */
320 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) {
325 if (s->pid == pid) {
326 list_del(&s->list);
327 kfree(s);
328 break;
331 up_write(&listeners->sem);
333 return 0;
336 static int parse(struct nlattr *na, struct cpumask *mask)
338 char *data;
339 int len;
340 int ret;
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;
358 #if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
359 #define TASKSTATS_NEEDS_PADDING 1
360 #endif
362 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
364 struct nlattr *na, *ret;
365 int aggr;
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)
392 goto err;
393 #endif
394 na = nla_nest_start(skb, aggr);
395 if (!na)
396 goto err;
398 if (nla_put(skb, type, sizeof(pid), &pid) < 0)
399 goto err;
400 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
401 if (!ret)
402 goto err;
403 nla_nest_end(skb, na);
405 return nla_data(ret);
406 err:
407 return NULL;
410 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
412 int rc = 0;
413 struct sk_buff *rep_skb;
414 struct cgroupstats *stats;
415 struct nlattr *na;
416 size_t size;
417 u32 fd;
418 struct file *file;
419 int fput_needed;
421 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
422 if (!na)
423 return -EINVAL;
425 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
426 file = fget_light(fd, &fput_needed);
427 if (!file)
428 return 0;
430 size = nla_total_size(sizeof(struct cgroupstats));
432 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
433 size);
434 if (rc < 0)
435 goto err;
437 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
438 sizeof(struct cgroupstats));
439 stats = nla_data(na);
440 memset(stats, 0, sizeof(*stats));
442 rc = cgroupstats_build(stats, file->f_dentry);
443 if (rc < 0) {
444 nlmsg_free(rep_skb);
445 goto err;
448 rc = send_reply(rep_skb, info);
450 err:
451 fput_light(file, fput_needed);
452 return rc;
455 static int cmd_attr_register_cpumask(struct genl_info *info)
457 cpumask_var_t mask;
458 int rc;
460 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
461 return -ENOMEM;
462 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
463 if (rc < 0)
464 goto out;
465 rc = add_del_listener(info->snd_pid, mask, REGISTER);
466 out:
467 free_cpumask_var(mask);
468 return rc;
471 static int cmd_attr_deregister_cpumask(struct genl_info *info)
473 cpumask_var_t mask;
474 int rc;
476 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
477 return -ENOMEM;
478 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
479 if (rc < 0)
480 goto out;
481 rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
482 out:
483 free_cpumask_var(mask);
484 return rc;
487 static size_t taskstats_packet_size(void)
489 size_t size;
491 size = nla_total_size(sizeof(u32)) +
492 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
493 #ifdef TASKSTATS_NEEDS_PADDING
494 size += nla_total_size(0); /* Padding for alignment */
495 #endif
496 return size;
499 static int cmd_attr_pid(struct genl_info *info)
501 struct taskstats *stats;
502 struct sk_buff *rep_skb;
503 size_t size;
504 u32 pid;
505 int rc;
507 size = taskstats_packet_size();
509 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
510 if (rc < 0)
511 return rc;
513 rc = -EINVAL;
514 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
515 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
516 if (!stats)
517 goto err;
519 rc = fill_stats_for_pid(pid, stats);
520 if (rc < 0)
521 goto err;
522 return send_reply(rep_skb, info);
523 err:
524 nlmsg_free(rep_skb);
525 return rc;
528 static int cmd_attr_tgid(struct genl_info *info)
530 struct taskstats *stats;
531 struct sk_buff *rep_skb;
532 size_t size;
533 u32 tgid;
534 int rc;
536 size = taskstats_packet_size();
538 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
539 if (rc < 0)
540 return rc;
542 rc = -EINVAL;
543 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
544 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
545 if (!stats)
546 goto err;
548 rc = fill_stats_for_tgid(tgid, stats);
549 if (rc < 0)
550 goto err;
551 return send_reply(rep_skb, info);
552 err:
553 nlmsg_free(rep_skb);
554 return rc;
557 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
559 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
560 return cmd_attr_register_cpumask(info);
561 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
562 return cmd_attr_deregister_cpumask(info);
563 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
564 return cmd_attr_pid(info);
565 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
566 return cmd_attr_tgid(info);
567 else
568 return -EINVAL;
571 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
573 struct signal_struct *sig = tsk->signal;
574 struct taskstats *stats;
576 if (sig->stats || thread_group_empty(tsk))
577 goto ret;
579 /* No problem if kmem_cache_zalloc() fails */
580 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
582 spin_lock_irq(&tsk->sighand->siglock);
583 if (!sig->stats) {
584 sig->stats = stats;
585 stats = NULL;
587 spin_unlock_irq(&tsk->sighand->siglock);
589 if (stats)
590 kmem_cache_free(taskstats_cache, stats);
591 ret:
592 return sig->stats;
595 /* Send pid data out on exit */
596 void taskstats_exit(struct task_struct *tsk, int group_dead)
598 int rc;
599 struct listener_list *listeners;
600 struct taskstats *stats;
601 struct sk_buff *rep_skb;
602 size_t size;
603 int is_thread_group;
605 if (!family_registered)
606 return;
609 * Size includes space for nested attributes
611 size = taskstats_packet_size();
613 is_thread_group = !!taskstats_tgid_alloc(tsk);
614 if (is_thread_group) {
615 /* PID + STATS + TGID + STATS */
616 size = 2 * size;
617 /* fill the tsk->signal->stats structure */
618 fill_tgid_exit(tsk);
621 listeners = __this_cpu_ptr(&listener_array);
622 if (list_empty(&listeners->list))
623 return;
625 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
626 if (rc < 0)
627 return;
629 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
630 if (!stats)
631 goto err;
633 fill_stats(tsk, stats);
636 * Doesn't matter if tsk is the leader or the last group member leaving
638 if (!is_thread_group || !group_dead)
639 goto send;
641 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
642 if (!stats)
643 goto err;
645 memcpy(stats, tsk->signal->stats, sizeof(*stats));
647 send:
648 send_cpu_listeners(rep_skb, listeners);
649 return;
650 err:
651 nlmsg_free(rep_skb);
654 static struct genl_ops taskstats_ops = {
655 .cmd = TASKSTATS_CMD_GET,
656 .doit = taskstats_user_cmd,
657 .policy = taskstats_cmd_get_policy,
658 .flags = GENL_ADMIN_PERM,
661 static struct genl_ops cgroupstats_ops = {
662 .cmd = CGROUPSTATS_CMD_GET,
663 .doit = cgroupstats_user_cmd,
664 .policy = cgroupstats_cmd_get_policy,
667 /* Needed early in initialization */
668 void __init taskstats_init_early(void)
670 unsigned int i;
672 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
673 for_each_possible_cpu(i) {
674 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
675 init_rwsem(&(per_cpu(listener_array, i).sem));
679 static int __init taskstats_init(void)
681 int rc;
683 rc = genl_register_family(&family);
684 if (rc)
685 return rc;
687 rc = genl_register_ops(&family, &taskstats_ops);
688 if (rc < 0)
689 goto err;
691 rc = genl_register_ops(&family, &cgroupstats_ops);
692 if (rc < 0)
693 goto err_cgroup_ops;
695 family_registered = 1;
696 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
697 return 0;
698 err_cgroup_ops:
699 genl_unregister_ops(&family, &taskstats_ops);
700 err:
701 genl_unregister_family(&family);
702 return rc;
706 * late initcall ensures initialization of statistics collection
707 * mechanisms precedes initialization of the taskstats interface
709 late_initcall(taskstats_init);