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/delayacct.h>
22 #include <net/genetlink.h>
23 #include <asm/atomic.h>
25 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
26 static int family_registered;
27 kmem_cache_t *taskstats_cache;
29 static struct genl_family family = {
30 .id = GENL_ID_GENERATE,
31 .name = TASKSTATS_GENL_NAME,
32 .version = TASKSTATS_GENL_VERSION,
33 .maxattr = TASKSTATS_CMD_ATTR_MAX,
36 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
38 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
39 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
43 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
44 void **replyp, size_t size)
50 * If new attributes are added, please revisit this allocation
52 skb = nlmsg_new(size);
57 int seq = get_cpu_var(taskstats_seqnum)++;
58 put_cpu_var(taskstats_seqnum);
60 reply = genlmsg_put(skb, 0, seq,
64 reply = genlmsg_put(skb, info->snd_pid, info->snd_seq,
77 static int send_reply(struct sk_buff *skb, pid_t pid, int event)
79 struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
83 reply = genlmsg_data(genlhdr);
85 rc = genlmsg_end(skb, reply);
91 if (event == TASKSTATS_MSG_MULTICAST)
92 return genlmsg_multicast(skb, pid, TASKSTATS_LISTEN_GROUP);
93 return genlmsg_unicast(skb, pid);
96 static int fill_pid(pid_t pid, struct task_struct *pidtsk,
97 struct taskstats *stats)
100 struct task_struct *tsk = pidtsk;
103 read_lock(&tasklist_lock);
104 tsk = find_task_by_pid(pid);
106 read_unlock(&tasklist_lock);
109 get_task_struct(tsk);
110 read_unlock(&tasklist_lock);
112 get_task_struct(tsk);
115 * Each accounting subsystem adds calls to its functions to
116 * fill in relevant parts of struct taskstsats as follows
118 * rc = per-task-foo(stats, tsk);
123 rc = delayacct_add_tsk(stats, tsk);
124 stats->version = TASKSTATS_VERSION;
126 /* Define err: label here if needed */
127 put_task_struct(tsk);
132 static int fill_tgid(pid_t tgid, struct task_struct *tgidtsk,
133 struct taskstats *stats)
136 struct task_struct *tsk, *first;
139 read_lock(&tasklist_lock);
141 first = find_task_by_pid(tgid);
143 read_unlock(&tasklist_lock);
150 * Each accounting subsystem adds calls its functions to
151 * fill in relevant parts of struct taskstsats as follows
153 * rc = per-task-foo(stats, tsk);
158 rc = delayacct_add_tsk(stats, tsk);
162 } while_each_thread(first, tsk);
163 read_unlock(&tasklist_lock);
164 stats->version = TASKSTATS_VERSION;
168 * Accounting subsytems can also add calls here if they don't
169 * wish to aggregate statistics for per-tgid stats
175 static int taskstats_send_stats(struct sk_buff *skb, struct genl_info *info)
178 struct sk_buff *rep_skb;
179 struct taskstats stats;
185 * Size includes space for nested attributes
187 size = nla_total_size(sizeof(u32)) +
188 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
190 memset(&stats, 0, sizeof(stats));
191 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
195 if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
196 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
197 rc = fill_pid(pid, NULL, &stats);
201 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
202 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
203 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
205 } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
206 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
207 rc = fill_tgid(tgid, NULL, &stats);
211 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
212 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
213 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
220 nla_nest_end(rep_skb, na);
222 return send_reply(rep_skb, info->snd_pid, TASKSTATS_MSG_UNICAST);
225 return genlmsg_cancel(rep_skb, reply);
231 /* Send pid data out on exit */
232 void taskstats_exit_send(struct task_struct *tsk, struct taskstats *tidstats,
233 struct taskstats *tgidstats)
236 struct sk_buff *rep_skb;
242 if (!family_registered || !tidstats)
245 is_thread_group = !thread_group_empty(tsk);
249 * Size includes space for nested attributes
251 size = nla_total_size(sizeof(u32)) +
252 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
255 size = 2 * size; /* PID + STATS + TGID + STATS */
257 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
261 rc = fill_pid(tsk->pid, tsk, tidstats);
265 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
266 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
267 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
269 nla_nest_end(rep_skb, na);
271 if (!is_thread_group || !tgidstats) {
272 send_reply(rep_skb, 0, TASKSTATS_MSG_MULTICAST);
276 rc = fill_tgid(tsk->pid, tsk, tgidstats);
278 * If fill_tgid() failed then one probable reason could be that the
279 * thread group leader has exited. fill_tgid() will fail, send out
280 * the pid statistics collected earlier.
283 send_reply(rep_skb, 0, TASKSTATS_MSG_MULTICAST);
287 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
288 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
289 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
291 nla_nest_end(rep_skb, na);
293 send_reply(rep_skb, 0, TASKSTATS_MSG_MULTICAST);
297 genlmsg_cancel(rep_skb, reply);
305 static struct genl_ops taskstats_ops = {
306 .cmd = TASKSTATS_CMD_GET,
307 .doit = taskstats_send_stats,
308 .policy = taskstats_cmd_get_policy,
311 /* Needed early in initialization */
312 void __init taskstats_init_early(void)
314 taskstats_cache = kmem_cache_create("taskstats_cache",
315 sizeof(struct taskstats),
316 0, SLAB_PANIC, NULL, NULL);
319 static int __init taskstats_init(void)
323 rc = genl_register_family(&family);
327 rc = genl_register_ops(&family, &taskstats_ops);
331 family_registered = 1;
334 genl_unregister_family(&family);
339 * late initcall ensures initialization of statistics collection
340 * mechanisms precedes initialization of the taskstats interface
342 late_initcall(taskstats_init);