blob: 21f82c29c9143c9a39e658593b0ac4671ce9c1b9 [file] [log] [blame]
Shailabh Nagarc7572492006-07-14 00:24:40 -07001/*
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>
Jay Lanf3cef7a2006-09-30 23:28:55 -070021#include <linux/tsacct_kern.h>
Shailabh Nagar6f449932006-07-14 00:24:41 -070022#include <linux/delayacct.h>
Shailabh Nagarf9fd8912006-07-14 00:24:47 -070023#include <linux/cpumask.h>
24#include <linux/percpu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070026#include <linux/cgroupstats.h>
27#include <linux/cgroup.h>
28#include <linux/fs.h>
29#include <linux/file.h>
Eric W. Biederman4bd6e322012-02-07 17:56:49 -080030#include <linux/pid_namespace.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070031#include <net/genetlink.h>
Arun Sharma600634972011-07-26 16:09:06 -070032#include <linux/atomic.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070033
Shailabh Nagarf9fd8912006-07-14 00:24:47 -070034/*
35 * Maximum length of a cpumask that can be specified in
36 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
37 */
38#define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS)
39
Vegard Nossumb81f3ea2008-07-25 01:48:55 -070040static DEFINE_PER_CPU(__u32, taskstats_seqnum);
Shailabh Nagarc7572492006-07-14 00:24:40 -070041static int family_registered;
Christoph Lametere18b8902006-12-06 20:33:20 -080042struct kmem_cache *taskstats_cache;
Shailabh Nagarc7572492006-07-14 00:24:40 -070043
44static 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,
49};
50
Alexey Dobriyanb54452b2010-02-18 08:14:31 +000051static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
Shailabh Nagarc7572492006-07-14 00:24:40 -070052 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 },
53 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
Shailabh Nagarf9fd8912006-07-14 00:24:47 -070054 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
55 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
56
Alexey Dobriyanb54452b2010-02-18 08:14:31 +000057static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
Balbir Singh846c7bb2007-10-18 23:39:44 -070058 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
59};
60
Shailabh Nagarf9fd8912006-07-14 00:24:47 -070061struct listener {
62 struct list_head list;
63 pid_t pid;
Shailabh Nagarbb129992006-07-14 00:24:47 -070064 char valid;
Shailabh Nagarc7572492006-07-14 00:24:40 -070065};
66
Shailabh Nagarf9fd8912006-07-14 00:24:47 -070067struct listener_list {
68 struct rw_semaphore sem;
69 struct list_head list;
70};
71static DEFINE_PER_CPU(struct listener_list, listener_array);
72
73enum actions {
74 REGISTER,
75 DEREGISTER,
76 CPU_DONT_CARE
77};
Shailabh Nagarc7572492006-07-14 00:24:40 -070078
79static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
Oleg Nesterov371674852006-12-06 20:36:55 -080080 size_t size)
Shailabh Nagarc7572492006-07-14 00:24:40 -070081{
82 struct sk_buff *skb;
83 void *reply;
84
85 /*
86 * If new attributes are added, please revisit this allocation
87 */
Thomas Graf3dabc712006-11-14 19:44:52 -080088 skb = genlmsg_new(size, GFP_KERNEL);
Shailabh Nagarc7572492006-07-14 00:24:40 -070089 if (!skb)
90 return -ENOMEM;
91
92 if (!info) {
Christoph Lametercd85fc52010-12-08 17:42:22 +010093 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
Shailabh Nagarc7572492006-07-14 00:24:40 -070094
Thomas Graf17c157c2006-11-14 19:46:02 -080095 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
Shailabh Nagarc7572492006-07-14 00:24:40 -070096 } else
Thomas Graf17c157c2006-11-14 19:46:02 -080097 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
Shailabh Nagarc7572492006-07-14 00:24:40 -070098 if (reply == NULL) {
99 nlmsg_free(skb);
100 return -EINVAL;
101 }
102
103 *skbp = skb;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700104 return 0;
105}
106
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700107/*
108 * Send taskstats data in @skb to listener with nl_pid @pid
109 */
Johannes Berg134e6372009-07-10 09:51:34 +0000110static int send_reply(struct sk_buff *skb, struct genl_info *info)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700111{
Arnaldo Carvalho de Melob529ccf2007-04-25 19:08:35 -0700112 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700113 void *reply = genlmsg_data(genlhdr);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700114
Johannes Berg053c0952015-01-16 22:09:00 +0100115 genlmsg_end(skb, reply);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700116
Johannes Berg134e6372009-07-10 09:51:34 +0000117 return genlmsg_reply(skb, info);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700118}
119
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700120/*
121 * Send taskstats data in @skb to listeners registered for @cpu's exit data
122 */
Oleg Nesterov115085e2006-12-06 20:36:51 -0800123static void send_cpu_listeners(struct sk_buff *skb,
124 struct listener_list *listeners)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700125{
Arnaldo Carvalho de Melob529ccf2007-04-25 19:08:35 -0700126 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700127 struct listener *s, *tmp;
128 struct sk_buff *skb_next, *skb_cur = skb;
129 void *reply = genlmsg_data(genlhdr);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700130 int rc, delcount = 0;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700131
Johannes Berg053c0952015-01-16 22:09:00 +0100132 genlmsg_end(skb, reply);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700133
134 rc = 0;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700135 down_read(&listeners->sem);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700136 list_for_each_entry(s, &listeners->list, list) {
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700137 skb_next = NULL;
138 if (!list_is_last(&s->list, &listeners->list)) {
139 skb_next = skb_clone(skb_cur, GFP_KERNEL);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700140 if (!skb_next)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700141 break;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700142 }
Johannes Berg134e6372009-07-10 09:51:34 +0000143 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700144 if (rc == -ECONNREFUSED) {
Shailabh Nagarbb129992006-07-14 00:24:47 -0700145 s->valid = 0;
146 delcount++;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700147 }
148 skb_cur = skb_next;
149 }
Shailabh Nagarbb129992006-07-14 00:24:47 -0700150 up_read(&listeners->sem);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700151
Shailabh Nagard94a0412006-07-30 03:03:11 -0700152 if (skb_cur)
153 nlmsg_free(skb_cur);
154
Shailabh Nagarbb129992006-07-14 00:24:47 -0700155 if (!delcount)
Shailabh Nagard94a0412006-07-30 03:03:11 -0700156 return;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700157
158 /* Delete invalidated entries */
159 down_write(&listeners->sem);
160 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
161 if (!s->valid) {
162 list_del(&s->list);
163 kfree(s);
164 }
165 }
166 up_write(&listeners->sem);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700167}
168
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800169static void fill_stats(struct user_namespace *user_ns,
170 struct pid_namespace *pid_ns,
171 struct task_struct *tsk, struct taskstats *stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700172{
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800173 memset(stats, 0, sizeof(*stats));
Shailabh Nagarc7572492006-07-14 00:24:40 -0700174 /*
175 * Each accounting subsystem adds calls to its functions to
176 * fill in relevant parts of struct taskstsats as follows
177 *
Shailabh Nagar7d94ddd2006-07-30 03:03:10 -0700178 * per-task-foo(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700179 */
180
Shailabh Nagar7d94ddd2006-07-30 03:03:10 -0700181 delayacct_add_tsk(stats, tsk);
Jay Lanf3cef7a2006-09-30 23:28:55 -0700182
183 /* fill in basic acct fields */
Shailabh Nagar6f449932006-07-14 00:24:41 -0700184 stats->version = TASKSTATS_VERSION;
Maxim Uvarovb663a792007-07-15 23:40:48 -0700185 stats->nvcsw = tsk->nvcsw;
186 stats->nivcsw = tsk->nivcsw;
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800187 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700188
Jay Lan9acc1852006-09-30 23:28:58 -0700189 /* fill in extended acct fields */
190 xacct_add_tsk(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700191}
192
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700193static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700194{
Oleg Nesterova98b6092006-10-28 10:38:54 -0700195 struct task_struct *tsk;
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700196
197 rcu_read_lock();
198 tsk = find_task_by_vpid(pid);
199 if (tsk)
200 get_task_struct(tsk);
201 rcu_read_unlock();
202 if (!tsk)
203 return -ESRCH;
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800204 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700205 put_task_struct(tsk);
206 return 0;
207}
208
209static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
210{
211 struct task_struct *tsk, *first;
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700212 unsigned long flags;
Oleg Nesterova98b6092006-10-28 10:38:54 -0700213 int rc = -ESRCH;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700214
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700215 /*
216 * Add additional stats from live tasks except zombie thread group
217 * leaders who are already counted with the dead tasks
218 */
Oleg Nesterova98b6092006-10-28 10:38:54 -0700219 rcu_read_lock();
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700220 first = find_task_by_vpid(tgid);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700221
Oleg Nesterova98b6092006-10-28 10:38:54 -0700222 if (!first || !lock_task_sighand(first, &flags))
223 goto out;
224
225 if (first->signal->stats)
226 memcpy(stats, first->signal->stats, sizeof(*stats));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800227 else
228 memset(stats, 0, sizeof(*stats));
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700229
Shailabh Nagarc7572492006-07-14 00:24:40 -0700230 tsk = first;
231 do {
Oleg Nesterovd7c3f5f2006-10-28 10:38:54 -0700232 if (tsk->exit_state)
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700233 continue;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700234 /*
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700235 * Accounting subsystem can call its functions here to
Shailabh Nagarc7572492006-07-14 00:24:40 -0700236 * fill in relevant parts of struct taskstsats as follows
237 *
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700238 * per-task-foo(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700239 */
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700240 delayacct_add_tsk(stats, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700241
Maxim Uvarovb663a792007-07-15 23:40:48 -0700242 stats->nvcsw += tsk->nvcsw;
243 stats->nivcsw += tsk->nivcsw;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700244 } while_each_thread(first, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700245
Oleg Nesterova98b6092006-10-28 10:38:54 -0700246 unlock_task_sighand(first, &flags);
247 rc = 0;
248out:
249 rcu_read_unlock();
250
251 stats->version = TASKSTATS_VERSION;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700252 /*
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +0200253 * Accounting subsystems can also add calls here to modify
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700254 * fields of taskstats.
Shailabh Nagarc7572492006-07-14 00:24:40 -0700255 */
Oleg Nesterova98b6092006-10-28 10:38:54 -0700256 return rc;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700257}
258
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700259static void fill_tgid_exit(struct task_struct *tsk)
260{
261 unsigned long flags;
262
Oleg Nesterovb8534d72006-10-28 10:38:53 -0700263 spin_lock_irqsave(&tsk->sighand->siglock, flags);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700264 if (!tsk->signal->stats)
265 goto ret;
266
267 /*
268 * Each accounting subsystem calls its functions here to
269 * accumalate its per-task stats for tsk, into the per-tgid structure
270 *
271 * per-task-foo(tsk->signal->stats, tsk);
272 */
273 delayacct_add_tsk(tsk->signal->stats, tsk);
274ret:
Oleg Nesterovb8534d72006-10-28 10:38:53 -0700275 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700276 return;
277}
278
Rusty Russell41c7bb92009-01-01 10:12:28 +1030279static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700280{
281 struct listener_list *listeners;
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700282 struct listener *s, *tmp, *s2;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700283 unsigned int cpu;
Chen Gang0d206332013-11-12 15:11:23 -0800284 int ret = 0;
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700285
Rusty Russell41c7bb92009-01-01 10:12:28 +1030286 if (!cpumask_subset(mask, cpu_possible_mask))
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700287 return -EINVAL;
288
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800289 if (current_user_ns() != &init_user_ns)
290 return -EINVAL;
291
292 if (task_active_pid_ns(current) != &init_pid_ns)
293 return -EINVAL;
294
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700295 if (isadd == REGISTER) {
Rusty Russell41c7bb92009-01-01 10:12:28 +1030296 for_each_cpu(cpu, mask) {
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700297 s = kmalloc_node(sizeof(struct listener),
298 GFP_KERNEL, cpu_to_node(cpu));
Chen Gang0d206332013-11-12 15:11:23 -0800299 if (!s) {
300 ret = -ENOMEM;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700301 goto cleanup;
Chen Gang0d206332013-11-12 15:11:23 -0800302 }
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700303 s->pid = pid;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700304 s->valid = 1;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700305
306 listeners = &per_cpu(listener_array, cpu);
307 down_write(&listeners->sem);
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700308 list_for_each_entry(s2, &listeners->list, list) {
Oleg Nesterova7295892011-08-03 16:21:05 -0700309 if (s2->pid == pid && s2->valid)
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700310 goto exists;
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700311 }
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700312 list_add(&s->list, &listeners->list);
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700313 s = NULL;
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700314exists:
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700315 up_write(&listeners->sem);
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700316 kfree(s); /* nop if NULL */
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700317 }
318 return 0;
319 }
320
321 /* Deregister or cleanup */
322cleanup:
Rusty Russell41c7bb92009-01-01 10:12:28 +1030323 for_each_cpu(cpu, mask) {
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700324 listeners = &per_cpu(listener_array, cpu);
325 down_write(&listeners->sem);
326 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
327 if (s->pid == pid) {
328 list_del(&s->list);
329 kfree(s);
330 break;
331 }
332 }
333 up_write(&listeners->sem);
334 }
Chen Gang0d206332013-11-12 15:11:23 -0800335 return ret;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700336}
337
Rusty Russell41c7bb92009-01-01 10:12:28 +1030338static int parse(struct nlattr *na, struct cpumask *mask)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700339{
340 char *data;
341 int len;
342 int ret;
343
344 if (na == NULL)
345 return 1;
346 len = nla_len(na);
347 if (len > TASKSTATS_CPUMASK_MAXLEN)
348 return -E2BIG;
349 if (len < 1)
350 return -EINVAL;
351 data = kmalloc(len, GFP_KERNEL);
352 if (!data)
353 return -ENOMEM;
354 nla_strlcpy(data, na, len);
Rusty Russell29c01772008-12-13 21:20:25 +1030355 ret = cpulist_parse(data, mask);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700356 kfree(data);
357 return ret;
358}
359
Jeff Mahoney9ab020c2011-01-12 17:00:48 -0800360#if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800361#define TASKSTATS_NEEDS_PADDING 1
362#endif
363
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800364static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
Oleg Nesterov68062b82006-12-06 20:36:53 -0800365{
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800366 struct nlattr *na, *ret;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800367 int aggr;
368
Oleg Nesterov371674852006-12-06 20:36:55 -0800369 aggr = (type == TASKSTATS_TYPE_PID)
370 ? TASKSTATS_TYPE_AGGR_PID
371 : TASKSTATS_TYPE_AGGR_TGID;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800372
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800373 /*
374 * The taskstats structure is internally aligned on 8 byte
375 * boundaries but the layout of the aggregrate reply, with
376 * two NLA headers and the pid (each 4 bytes), actually
377 * force the entire structure to be unaligned. This causes
378 * the kernel to issue unaligned access warnings on some
379 * architectures like ia64. Unfortunately, some software out there
380 * doesn't properly unroll the NLA packet and assumes that the start
381 * of the taskstats structure will always be 20 bytes from the start
382 * of the netlink payload. Aligning the start of the taskstats
383 * structure breaks this software, which we don't want. So, for now
384 * the alignment only happens on architectures that require it
385 * and those users will have to update to fixed versions of those
386 * packages. Space is reserved in the packet only when needed.
387 * This ifdef should be removed in several years e.g. 2012 once
388 * we can be confident that fixed versions are installed on most
389 * systems. We add the padding before the aggregate since the
390 * aggregate is already a defined type.
391 */
392#ifdef TASKSTATS_NEEDS_PADDING
393 if (nla_put(skb, TASKSTATS_TYPE_NULL, 0, NULL) < 0)
394 goto err;
395#endif
Oleg Nesterov68062b82006-12-06 20:36:53 -0800396 na = nla_nest_start(skb, aggr);
Oleg Nesterov371674852006-12-06 20:36:55 -0800397 if (!na)
398 goto err;
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800399
Chen Gang3fa58262013-11-12 15:11:22 -0800400 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
401 nla_nest_cancel(skb, na);
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800402 goto err;
Chen Gang3fa58262013-11-12 15:11:22 -0800403 }
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800404 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
Chen Gang3fa58262013-11-12 15:11:22 -0800405 if (!ret) {
406 nla_nest_cancel(skb, na);
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800407 goto err;
Chen Gang3fa58262013-11-12 15:11:22 -0800408 }
Oleg Nesterov68062b82006-12-06 20:36:53 -0800409 nla_nest_end(skb, na);
410
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800411 return nla_data(ret);
412err:
413 return NULL;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800414}
415
Balbir Singh846c7bb2007-10-18 23:39:44 -0700416static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
417{
418 int rc = 0;
419 struct sk_buff *rep_skb;
420 struct cgroupstats *stats;
421 struct nlattr *na;
422 size_t size;
423 u32 fd;
Al Viro2903ff02012-08-28 12:52:22 -0400424 struct fd f;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700425
426 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
427 if (!na)
428 return -EINVAL;
429
430 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
Al Viro2903ff02012-08-28 12:52:22 -0400431 f = fdget(fd);
432 if (!f.file)
Adrian Bunkf9615982007-11-14 17:00:37 -0800433 return 0;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700434
Adrian Bunkf9615982007-11-14 17:00:37 -0800435 size = nla_total_size(sizeof(struct cgroupstats));
Balbir Singh846c7bb2007-10-18 23:39:44 -0700436
Adrian Bunkf9615982007-11-14 17:00:37 -0800437 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
438 size);
439 if (rc < 0)
440 goto err;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700441
Adrian Bunkf9615982007-11-14 17:00:37 -0800442 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
443 sizeof(struct cgroupstats));
Alan Cox25353b32012-07-30 14:42:49 -0700444 if (na == NULL) {
Jesper Juhl0324b5a2012-10-04 17:16:52 -0700445 nlmsg_free(rep_skb);
Alan Cox25353b32012-07-30 14:42:49 -0700446 rc = -EMSGSIZE;
447 goto err;
448 }
449
Adrian Bunkf9615982007-11-14 17:00:37 -0800450 stats = nla_data(na);
451 memset(stats, 0, sizeof(*stats));
Balbir Singh846c7bb2007-10-18 23:39:44 -0700452
Al Virob5830432014-10-31 01:22:04 -0400453 rc = cgroupstats_build(stats, f.file->f_path.dentry);
Adrian Bunkf9615982007-11-14 17:00:37 -0800454 if (rc < 0) {
455 nlmsg_free(rep_skb);
456 goto err;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700457 }
458
Johannes Berg134e6372009-07-10 09:51:34 +0000459 rc = send_reply(rep_skb, info);
Adrian Bunkf9615982007-11-14 17:00:37 -0800460
Balbir Singh846c7bb2007-10-18 23:39:44 -0700461err:
Al Viro2903ff02012-08-28 12:52:22 -0400462 fdput(f);
Balbir Singh846c7bb2007-10-18 23:39:44 -0700463 return rc;
464}
465
Michael Holzheu93233122010-10-27 15:34:44 -0700466static int cmd_attr_register_cpumask(struct genl_info *info)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700467{
Rusty Russell41c7bb92009-01-01 10:12:28 +1030468 cpumask_var_t mask;
Michael Holzheu93233122010-10-27 15:34:44 -0700469 int rc;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700470
Rusty Russell41c7bb92009-01-01 10:12:28 +1030471 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
472 return -ENOMEM;
Rusty Russell41c7bb92009-01-01 10:12:28 +1030473 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700474 if (rc < 0)
Michael Holzheu93233122010-10-27 15:34:44 -0700475 goto out;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000476 rc = add_del_listener(info->snd_portid, mask, REGISTER);
Michael Holzheu93233122010-10-27 15:34:44 -0700477out:
478 free_cpumask_var(mask);
479 return rc;
480}
Rusty Russell41c7bb92009-01-01 10:12:28 +1030481
Michael Holzheu93233122010-10-27 15:34:44 -0700482static int cmd_attr_deregister_cpumask(struct genl_info *info)
483{
484 cpumask_var_t mask;
485 int rc;
486
487 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
488 return -ENOMEM;
Rusty Russell41c7bb92009-01-01 10:12:28 +1030489 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
490 if (rc < 0)
Michael Holzheu93233122010-10-27 15:34:44 -0700491 goto out;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000492 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
Michael Holzheu93233122010-10-27 15:34:44 -0700493out:
Rusty Russell41c7bb92009-01-01 10:12:28 +1030494 free_cpumask_var(mask);
Michael Holzheu93233122010-10-27 15:34:44 -0700495 return rc;
496}
Shailabh Nagarc7572492006-07-14 00:24:40 -0700497
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800498static size_t taskstats_packet_size(void)
499{
500 size_t size;
501
502 size = nla_total_size(sizeof(u32)) +
503 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
504#ifdef TASKSTATS_NEEDS_PADDING
505 size += nla_total_size(0); /* Padding for alignment */
506#endif
507 return size;
508}
509
Michael Holzheu93233122010-10-27 15:34:44 -0700510static int cmd_attr_pid(struct genl_info *info)
511{
512 struct taskstats *stats;
513 struct sk_buff *rep_skb;
514 size_t size;
515 u32 pid;
516 int rc;
517
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800518 size = taskstats_packet_size();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700519
Oleg Nesterov371674852006-12-06 20:36:55 -0800520 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700521 if (rc < 0)
522 return rc;
523
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800524 rc = -EINVAL;
Michael Holzheu93233122010-10-27 15:34:44 -0700525 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
526 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
527 if (!stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700528 goto err;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700529
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700530 rc = fill_stats_for_pid(pid, stats);
Michael Holzheu93233122010-10-27 15:34:44 -0700531 if (rc < 0)
532 goto err;
Johannes Berg134e6372009-07-10 09:51:34 +0000533 return send_reply(rep_skb, info);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700534err:
535 nlmsg_free(rep_skb);
536 return rc;
537}
538
Michael Holzheu93233122010-10-27 15:34:44 -0700539static int cmd_attr_tgid(struct genl_info *info)
540{
541 struct taskstats *stats;
542 struct sk_buff *rep_skb;
543 size_t size;
544 u32 tgid;
545 int rc;
546
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800547 size = taskstats_packet_size();
Michael Holzheu93233122010-10-27 15:34:44 -0700548
549 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
550 if (rc < 0)
551 return rc;
552
553 rc = -EINVAL;
554 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
555 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
556 if (!stats)
557 goto err;
558
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700559 rc = fill_stats_for_tgid(tgid, stats);
Michael Holzheu93233122010-10-27 15:34:44 -0700560 if (rc < 0)
561 goto err;
562 return send_reply(rep_skb, info);
563err:
564 nlmsg_free(rep_skb);
565 return rc;
566}
567
568static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
569{
570 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
571 return cmd_attr_register_cpumask(info);
572 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
573 return cmd_attr_deregister_cpumask(info);
574 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
575 return cmd_attr_pid(info);
576 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
577 return cmd_attr_tgid(info);
578 else
579 return -EINVAL;
580}
581
Oleg Nesterov34ec1232006-12-06 20:36:52 -0800582static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
583{
584 struct signal_struct *sig = tsk->signal;
585 struct taskstats *stats;
586
587 if (sig->stats || thread_group_empty(tsk))
588 goto ret;
589
590 /* No problem if kmem_cache_zalloc() fails */
591 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
592
593 spin_lock_irq(&tsk->sighand->siglock);
594 if (!sig->stats) {
595 sig->stats = stats;
596 stats = NULL;
597 }
598 spin_unlock_irq(&tsk->sighand->siglock);
599
600 if (stats)
601 kmem_cache_free(taskstats_cache, stats);
602ret:
603 return sig->stats;
604}
605
Shailabh Nagarc7572492006-07-14 00:24:40 -0700606/* Send pid data out on exit */
Oleg Nesterov115085e2006-12-06 20:36:51 -0800607void taskstats_exit(struct task_struct *tsk, int group_dead)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700608{
609 int rc;
Oleg Nesterov115085e2006-12-06 20:36:51 -0800610 struct listener_list *listeners;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800611 struct taskstats *stats;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700612 struct sk_buff *rep_skb;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700613 size_t size;
614 int is_thread_group;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700615
Oleg Nesterov4a279ff2006-10-30 22:07:15 -0800616 if (!family_registered)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700617 return;
618
Shailabh Nagarc7572492006-07-14 00:24:40 -0700619 /*
620 * Size includes space for nested attributes
621 */
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800622 size = taskstats_packet_size();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700623
Oleg Nesterov34ec1232006-12-06 20:36:52 -0800624 is_thread_group = !!taskstats_tgid_alloc(tsk);
Oleg Nesterov4a279ff2006-10-30 22:07:15 -0800625 if (is_thread_group) {
626 /* PID + STATS + TGID + STATS */
627 size = 2 * size;
628 /* fill the tsk->signal->stats structure */
629 fill_tgid_exit(tsk);
630 }
631
Christoph Lameter4a32fea2014-08-17 12:30:27 -0500632 listeners = raw_cpu_ptr(&listener_array);
Oleg Nesterov115085e2006-12-06 20:36:51 -0800633 if (list_empty(&listeners->list))
634 return;
635
Oleg Nesterov371674852006-12-06 20:36:55 -0800636 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700637 if (rc < 0)
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800638 return;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700639
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800640 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
641 task_pid_nr_ns(tsk, &init_pid_ns));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800642 if (!stats)
Oleg Nesterov371674852006-12-06 20:36:55 -0800643 goto err;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800644
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800645 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700646
Shailabh Nagarc7572492006-07-14 00:24:40 -0700647 /*
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700648 * Doesn't matter if tsk is the leader or the last group member leaving
Shailabh Nagarc7572492006-07-14 00:24:40 -0700649 */
Oleg Nesterov68062b82006-12-06 20:36:53 -0800650 if (!is_thread_group || !group_dead)
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700651 goto send;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700652
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800653 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
654 task_tgid_nr_ns(tsk, &init_pid_ns));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800655 if (!stats)
Oleg Nesterov371674852006-12-06 20:36:55 -0800656 goto err;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800657
658 memcpy(stats, tsk->signal->stats, sizeof(*stats));
Shailabh Nagarc7572492006-07-14 00:24:40 -0700659
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700660send:
Oleg Nesterov115085e2006-12-06 20:36:51 -0800661 send_cpu_listeners(rep_skb, listeners);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700662 return;
Oleg Nesterov371674852006-12-06 20:36:55 -0800663err:
Shailabh Nagarc7572492006-07-14 00:24:40 -0700664 nlmsg_free(rep_skb);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700665}
666
Johannes Berg4534de82013-11-14 17:14:46 +0100667static const struct genl_ops taskstats_ops[] = {
Johannes Berg88d36a92013-11-14 17:14:39 +0100668 {
669 .cmd = TASKSTATS_CMD_GET,
670 .doit = taskstats_user_cmd,
671 .policy = taskstats_cmd_get_policy,
672 .flags = GENL_ADMIN_PERM,
673 },
674 {
675 .cmd = CGROUPSTATS_CMD_GET,
676 .doit = cgroupstats_user_cmd,
677 .policy = cgroupstats_cmd_get_policy,
678 },
Balbir Singh846c7bb2007-10-18 23:39:44 -0700679};
680
Shailabh Nagarc7572492006-07-14 00:24:40 -0700681/* Needed early in initialization */
682void __init taskstats_init_early(void)
683{
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700684 unsigned int i;
685
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700686 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700687 for_each_possible_cpu(i) {
688 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
689 init_rwsem(&(per_cpu(listener_array, i).sem));
690 }
Shailabh Nagarc7572492006-07-14 00:24:40 -0700691}
692
693static int __init taskstats_init(void)
694{
695 int rc;
696
Johannes Bergc53ed742013-11-19 15:19:31 +0100697 rc = genl_register_family_with_ops(&family, taskstats_ops);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700698 if (rc)
699 return rc;
700
Shailabh Nagarc7572492006-07-14 00:24:40 -0700701 family_registered = 1;
Mandeep Singh Bainesf9b182e2011-03-23 16:43:27 -0700702 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700703 return 0;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700704}
705
706/*
707 * late initcall ensures initialization of statistics collection
708 * mechanisms precedes initialization of the taskstats interface
709 */
710late_initcall(taskstats_init);