blob: 13d2f7cd65dbfd851eaa61cf08645994c9fd0037 [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 Nesterov37167482006-12-06 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 int rc;
115
Shailabh Nagarc7572492006-07-14 00:24:40 -0700116 rc = genlmsg_end(skb, reply);
117 if (rc < 0) {
118 nlmsg_free(skb);
119 return rc;
120 }
121
Johannes Berg134e6372009-07-10 09:51:34 +0000122 return genlmsg_reply(skb, info);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700123}
124
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700125/*
126 * Send taskstats data in @skb to listeners registered for @cpu's exit data
127 */
Oleg Nesterov115085e2006-12-06 20:36:51 -0800128static void send_cpu_listeners(struct sk_buff *skb,
129 struct listener_list *listeners)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700130{
Arnaldo Carvalho de Melob529ccf2007-04-25 19:08:35 -0700131 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700132 struct listener *s, *tmp;
133 struct sk_buff *skb_next, *skb_cur = skb;
134 void *reply = genlmsg_data(genlhdr);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700135 int rc, delcount = 0;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700136
137 rc = genlmsg_end(skb, reply);
138 if (rc < 0) {
139 nlmsg_free(skb);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700140 return;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700141 }
142
143 rc = 0;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700144 down_read(&listeners->sem);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700145 list_for_each_entry(s, &listeners->list, list) {
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700146 skb_next = NULL;
147 if (!list_is_last(&s->list, &listeners->list)) {
148 skb_next = skb_clone(skb_cur, GFP_KERNEL);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700149 if (!skb_next)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700150 break;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700151 }
Johannes Berg134e6372009-07-10 09:51:34 +0000152 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
Shailabh Nagard94a0412006-07-30 03:03:11 -0700153 if (rc == -ECONNREFUSED) {
Shailabh Nagarbb129992006-07-14 00:24:47 -0700154 s->valid = 0;
155 delcount++;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700156 }
157 skb_cur = skb_next;
158 }
Shailabh Nagarbb129992006-07-14 00:24:47 -0700159 up_read(&listeners->sem);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700160
Shailabh Nagard94a0412006-07-30 03:03:11 -0700161 if (skb_cur)
162 nlmsg_free(skb_cur);
163
Shailabh Nagarbb129992006-07-14 00:24:47 -0700164 if (!delcount)
Shailabh Nagard94a0412006-07-30 03:03:11 -0700165 return;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700166
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);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700176}
177
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800178static void fill_stats(struct user_namespace *user_ns,
179 struct pid_namespace *pid_ns,
180 struct task_struct *tsk, struct taskstats *stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700181{
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800182 memset(stats, 0, sizeof(*stats));
Shailabh Nagarc7572492006-07-14 00:24:40 -0700183 /*
184 * Each accounting subsystem adds calls to its functions to
185 * fill in relevant parts of struct taskstsats as follows
186 *
Shailabh Nagar7d94ddd2006-07-30 03:03:10 -0700187 * per-task-foo(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700188 */
189
Shailabh Nagar7d94ddd2006-07-30 03:03:10 -0700190 delayacct_add_tsk(stats, tsk);
Jay Lanf3cef7a2006-09-30 23:28:55 -0700191
192 /* fill in basic acct fields */
Shailabh Nagar6f449932006-07-14 00:24:41 -0700193 stats->version = TASKSTATS_VERSION;
Maxim Uvarovb663a792007-07-15 23:40:48 -0700194 stats->nvcsw = tsk->nvcsw;
195 stats->nivcsw = tsk->nivcsw;
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800196 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700197
Jay Lan9acc1852006-09-30 23:28:58 -0700198 /* fill in extended acct fields */
199 xacct_add_tsk(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700200}
201
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700202static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700203{
Oleg Nesterova98b6092006-10-28 10:38:54 -0700204 struct task_struct *tsk;
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700205
206 rcu_read_lock();
207 tsk = find_task_by_vpid(pid);
208 if (tsk)
209 get_task_struct(tsk);
210 rcu_read_unlock();
211 if (!tsk)
212 return -ESRCH;
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800213 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700214 put_task_struct(tsk);
215 return 0;
216}
217
218static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
219{
220 struct task_struct *tsk, *first;
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700221 unsigned long flags;
Oleg Nesterova98b6092006-10-28 10:38:54 -0700222 int rc = -ESRCH;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700223
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700224 /*
225 * Add additional stats from live tasks except zombie thread group
226 * leaders who are already counted with the dead tasks
227 */
Oleg Nesterova98b6092006-10-28 10:38:54 -0700228 rcu_read_lock();
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700229 first = find_task_by_vpid(tgid);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700230
Oleg Nesterova98b6092006-10-28 10:38:54 -0700231 if (!first || !lock_task_sighand(first, &flags))
232 goto out;
233
234 if (first->signal->stats)
235 memcpy(stats, first->signal->stats, sizeof(*stats));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800236 else
237 memset(stats, 0, sizeof(*stats));
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700238
Shailabh Nagarc7572492006-07-14 00:24:40 -0700239 tsk = first;
240 do {
Oleg Nesterovd7c3f5f2006-10-28 10:38:54 -0700241 if (tsk->exit_state)
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700242 continue;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700243 /*
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700244 * Accounting subsystem can call its functions here to
Shailabh Nagarc7572492006-07-14 00:24:40 -0700245 * fill in relevant parts of struct taskstsats as follows
246 *
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700247 * per-task-foo(stats, tsk);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700248 */
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700249 delayacct_add_tsk(stats, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700250
Maxim Uvarovb663a792007-07-15 23:40:48 -0700251 stats->nvcsw += tsk->nvcsw;
252 stats->nivcsw += tsk->nivcsw;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700253 } while_each_thread(first, tsk);
Shailabh Nagar6f449932006-07-14 00:24:41 -0700254
Oleg Nesterova98b6092006-10-28 10:38:54 -0700255 unlock_task_sighand(first, &flags);
256 rc = 0;
257out:
258 rcu_read_unlock();
259
260 stats->version = TASKSTATS_VERSION;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700261 /*
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +0200262 * Accounting subsystems can also add calls here to modify
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700263 * fields of taskstats.
Shailabh Nagarc7572492006-07-14 00:24:40 -0700264 */
Oleg Nesterova98b6092006-10-28 10:38:54 -0700265 return rc;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700266}
267
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700268static void fill_tgid_exit(struct task_struct *tsk)
269{
270 unsigned long flags;
271
Oleg Nesterovb8534d72006-10-28 10:38:53 -0700272 spin_lock_irqsave(&tsk->sighand->siglock, flags);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700273 if (!tsk->signal->stats)
274 goto ret;
275
276 /*
277 * Each accounting subsystem calls its functions here to
278 * accumalate its per-task stats for tsk, into the per-tgid structure
279 *
280 * per-task-foo(tsk->signal->stats, tsk);
281 */
282 delayacct_add_tsk(tsk->signal->stats, tsk);
283ret:
Oleg Nesterovb8534d72006-10-28 10:38:53 -0700284 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700285 return;
286}
287
Rusty Russell41c7bb92009-01-01 10:12:28 +1030288static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700289{
290 struct listener_list *listeners;
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700291 struct listener *s, *tmp, *s2;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700292 unsigned int cpu;
Chen Gang0d206332013-11-12 15:11:23 -0800293 int ret = 0;
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700294
Rusty Russell41c7bb92009-01-01 10:12:28 +1030295 if (!cpumask_subset(mask, cpu_possible_mask))
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700296 return -EINVAL;
297
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800298 if (current_user_ns() != &init_user_ns)
299 return -EINVAL;
300
301 if (task_active_pid_ns(current) != &init_pid_ns)
302 return -EINVAL;
303
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700304 if (isadd == REGISTER) {
Rusty Russell41c7bb92009-01-01 10:12:28 +1030305 for_each_cpu(cpu, mask) {
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700306 s = kmalloc_node(sizeof(struct listener),
307 GFP_KERNEL, cpu_to_node(cpu));
Chen Gang0d206332013-11-12 15:11:23 -0800308 if (!s) {
309 ret = -ENOMEM;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700310 goto cleanup;
Chen Gang0d206332013-11-12 15:11:23 -0800311 }
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700312 s->pid = pid;
Shailabh Nagarbb129992006-07-14 00:24:47 -0700313 s->valid = 1;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700314
315 listeners = &per_cpu(listener_array, cpu);
316 down_write(&listeners->sem);
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700317 list_for_each_entry(s2, &listeners->list, list) {
Oleg Nesterova7295892011-08-03 16:21:05 -0700318 if (s2->pid == pid && s2->valid)
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700319 goto exists;
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700320 }
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700321 list_add(&s->list, &listeners->list);
Vasiliy Kulikov26c4cae2011-06-27 16:18:11 -0700322 s = NULL;
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700323exists:
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700324 up_write(&listeners->sem);
Oleg Nesterovdfc428b2011-08-03 16:21:04 -0700325 kfree(s); /* nop if NULL */
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700326 }
327 return 0;
328 }
329
330 /* Deregister or cleanup */
331cleanup:
Rusty Russell41c7bb92009-01-01 10:12:28 +1030332 for_each_cpu(cpu, mask) {
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700333 listeners = &per_cpu(listener_array, cpu);
334 down_write(&listeners->sem);
335 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
336 if (s->pid == pid) {
337 list_del(&s->list);
338 kfree(s);
339 break;
340 }
341 }
342 up_write(&listeners->sem);
343 }
Chen Gang0d206332013-11-12 15:11:23 -0800344 return ret;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700345}
346
Rusty Russell41c7bb92009-01-01 10:12:28 +1030347static int parse(struct nlattr *na, struct cpumask *mask)
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700348{
349 char *data;
350 int len;
351 int ret;
352
353 if (na == NULL)
354 return 1;
355 len = nla_len(na);
356 if (len > TASKSTATS_CPUMASK_MAXLEN)
357 return -E2BIG;
358 if (len < 1)
359 return -EINVAL;
360 data = kmalloc(len, GFP_KERNEL);
361 if (!data)
362 return -ENOMEM;
363 nla_strlcpy(data, na, len);
Rusty Russell29c01772008-12-13 21:20:25 +1030364 ret = cpulist_parse(data, mask);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700365 kfree(data);
366 return ret;
367}
368
Jeff Mahoney9ab020c2011-01-12 17:00:48 -0800369#if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800370#define TASKSTATS_NEEDS_PADDING 1
371#endif
372
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800373static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
Oleg Nesterov68062b82006-12-06 20:36:53 -0800374{
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800375 struct nlattr *na, *ret;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800376 int aggr;
377
Oleg Nesterov37167482006-12-06 20:36:55 -0800378 aggr = (type == TASKSTATS_TYPE_PID)
379 ? TASKSTATS_TYPE_AGGR_PID
380 : TASKSTATS_TYPE_AGGR_TGID;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800381
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800382 /*
383 * The taskstats structure is internally aligned on 8 byte
384 * boundaries but the layout of the aggregrate reply, with
385 * two NLA headers and the pid (each 4 bytes), actually
386 * force the entire structure to be unaligned. This causes
387 * the kernel to issue unaligned access warnings on some
388 * architectures like ia64. Unfortunately, some software out there
389 * doesn't properly unroll the NLA packet and assumes that the start
390 * of the taskstats structure will always be 20 bytes from the start
391 * of the netlink payload. Aligning the start of the taskstats
392 * structure breaks this software, which we don't want. So, for now
393 * the alignment only happens on architectures that require it
394 * and those users will have to update to fixed versions of those
395 * packages. Space is reserved in the packet only when needed.
396 * This ifdef should be removed in several years e.g. 2012 once
397 * we can be confident that fixed versions are installed on most
398 * systems. We add the padding before the aggregate since the
399 * aggregate is already a defined type.
400 */
401#ifdef TASKSTATS_NEEDS_PADDING
402 if (nla_put(skb, TASKSTATS_TYPE_NULL, 0, NULL) < 0)
403 goto err;
404#endif
Oleg Nesterov68062b82006-12-06 20:36:53 -0800405 na = nla_nest_start(skb, aggr);
Oleg Nesterov37167482006-12-06 20:36:55 -0800406 if (!na)
407 goto err;
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800408
Chen Gang3fa58262013-11-12 15:11:22 -0800409 if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
410 nla_nest_cancel(skb, na);
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800411 goto err;
Chen Gang3fa58262013-11-12 15:11:22 -0800412 }
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800413 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
Chen Gang3fa58262013-11-12 15:11:22 -0800414 if (!ret) {
415 nla_nest_cancel(skb, na);
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800416 goto err;
Chen Gang3fa58262013-11-12 15:11:22 -0800417 }
Oleg Nesterov68062b82006-12-06 20:36:53 -0800418 nla_nest_end(skb, na);
419
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800420 return nla_data(ret);
421err:
422 return NULL;
Oleg Nesterov68062b82006-12-06 20:36:53 -0800423}
424
Balbir Singh846c7bb2007-10-18 23:39:44 -0700425static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
426{
427 int rc = 0;
428 struct sk_buff *rep_skb;
429 struct cgroupstats *stats;
430 struct nlattr *na;
431 size_t size;
432 u32 fd;
Al Viro2903ff02012-08-28 12:52:22 -0400433 struct fd f;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700434
435 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
436 if (!na)
437 return -EINVAL;
438
439 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
Al Viro2903ff02012-08-28 12:52:22 -0400440 f = fdget(fd);
441 if (!f.file)
Adrian Bunkf9615982007-11-14 17:00:37 -0800442 return 0;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700443
Adrian Bunkf9615982007-11-14 17:00:37 -0800444 size = nla_total_size(sizeof(struct cgroupstats));
Balbir Singh846c7bb2007-10-18 23:39:44 -0700445
Adrian Bunkf9615982007-11-14 17:00:37 -0800446 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
447 size);
448 if (rc < 0)
449 goto err;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700450
Adrian Bunkf9615982007-11-14 17:00:37 -0800451 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
452 sizeof(struct cgroupstats));
Alan Cox25353b32012-07-30 14:42:49 -0700453 if (na == NULL) {
Jesper Juhl0324b5a2012-10-04 17:16:52 -0700454 nlmsg_free(rep_skb);
Alan Cox25353b32012-07-30 14:42:49 -0700455 rc = -EMSGSIZE;
456 goto err;
457 }
458
Adrian Bunkf9615982007-11-14 17:00:37 -0800459 stats = nla_data(na);
460 memset(stats, 0, sizeof(*stats));
Balbir Singh846c7bb2007-10-18 23:39:44 -0700461
Al Viro2903ff02012-08-28 12:52:22 -0400462 rc = cgroupstats_build(stats, f.file->f_dentry);
Adrian Bunkf9615982007-11-14 17:00:37 -0800463 if (rc < 0) {
464 nlmsg_free(rep_skb);
465 goto err;
Balbir Singh846c7bb2007-10-18 23:39:44 -0700466 }
467
Johannes Berg134e6372009-07-10 09:51:34 +0000468 rc = send_reply(rep_skb, info);
Adrian Bunkf9615982007-11-14 17:00:37 -0800469
Balbir Singh846c7bb2007-10-18 23:39:44 -0700470err:
Al Viro2903ff02012-08-28 12:52:22 -0400471 fdput(f);
Balbir Singh846c7bb2007-10-18 23:39:44 -0700472 return rc;
473}
474
Michael Holzheu93233122010-10-27 15:34:44 -0700475static int cmd_attr_register_cpumask(struct genl_info *info)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700476{
Rusty Russell41c7bb92009-01-01 10:12:28 +1030477 cpumask_var_t mask;
Michael Holzheu93233122010-10-27 15:34:44 -0700478 int rc;
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700479
Rusty Russell41c7bb92009-01-01 10:12:28 +1030480 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
481 return -ENOMEM;
Rusty Russell41c7bb92009-01-01 10:12:28 +1030482 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700483 if (rc < 0)
Michael Holzheu93233122010-10-27 15:34:44 -0700484 goto out;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000485 rc = add_del_listener(info->snd_portid, mask, REGISTER);
Michael Holzheu93233122010-10-27 15:34:44 -0700486out:
487 free_cpumask_var(mask);
488 return rc;
489}
Rusty Russell41c7bb92009-01-01 10:12:28 +1030490
Michael Holzheu93233122010-10-27 15:34:44 -0700491static int cmd_attr_deregister_cpumask(struct genl_info *info)
492{
493 cpumask_var_t mask;
494 int rc;
495
496 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
497 return -ENOMEM;
Rusty Russell41c7bb92009-01-01 10:12:28 +1030498 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
499 if (rc < 0)
Michael Holzheu93233122010-10-27 15:34:44 -0700500 goto out;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000501 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
Michael Holzheu93233122010-10-27 15:34:44 -0700502out:
Rusty Russell41c7bb92009-01-01 10:12:28 +1030503 free_cpumask_var(mask);
Michael Holzheu93233122010-10-27 15:34:44 -0700504 return rc;
505}
Shailabh Nagarc7572492006-07-14 00:24:40 -0700506
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800507static size_t taskstats_packet_size(void)
508{
509 size_t size;
510
511 size = nla_total_size(sizeof(u32)) +
512 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
513#ifdef TASKSTATS_NEEDS_PADDING
514 size += nla_total_size(0); /* Padding for alignment */
515#endif
516 return size;
517}
518
Michael Holzheu93233122010-10-27 15:34:44 -0700519static int cmd_attr_pid(struct genl_info *info)
520{
521 struct taskstats *stats;
522 struct sk_buff *rep_skb;
523 size_t size;
524 u32 pid;
525 int rc;
526
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800527 size = taskstats_packet_size();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700528
Oleg Nesterov37167482006-12-06 20:36:55 -0800529 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700530 if (rc < 0)
531 return rc;
532
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800533 rc = -EINVAL;
Michael Holzheu93233122010-10-27 15:34:44 -0700534 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
535 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
536 if (!stats)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700537 goto err;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700538
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700539 rc = fill_stats_for_pid(pid, stats);
Michael Holzheu93233122010-10-27 15:34:44 -0700540 if (rc < 0)
541 goto err;
Johannes Berg134e6372009-07-10 09:51:34 +0000542 return send_reply(rep_skb, info);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700543err:
544 nlmsg_free(rep_skb);
545 return rc;
546}
547
Michael Holzheu93233122010-10-27 15:34:44 -0700548static int cmd_attr_tgid(struct genl_info *info)
549{
550 struct taskstats *stats;
551 struct sk_buff *rep_skb;
552 size_t size;
553 u32 tgid;
554 int rc;
555
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800556 size = taskstats_packet_size();
Michael Holzheu93233122010-10-27 15:34:44 -0700557
558 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
559 if (rc < 0)
560 return rc;
561
562 rc = -EINVAL;
563 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
564 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
565 if (!stats)
566 goto err;
567
Michael Holzheu3d9e0cf2010-10-27 15:34:44 -0700568 rc = fill_stats_for_tgid(tgid, stats);
Michael Holzheu93233122010-10-27 15:34:44 -0700569 if (rc < 0)
570 goto err;
571 return send_reply(rep_skb, info);
572err:
573 nlmsg_free(rep_skb);
574 return rc;
575}
576
577static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
578{
579 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
580 return cmd_attr_register_cpumask(info);
581 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
582 return cmd_attr_deregister_cpumask(info);
583 else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
584 return cmd_attr_pid(info);
585 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
586 return cmd_attr_tgid(info);
587 else
588 return -EINVAL;
589}
590
Oleg Nesterov34ec1232006-12-06 20:36:52 -0800591static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
592{
593 struct signal_struct *sig = tsk->signal;
594 struct taskstats *stats;
595
596 if (sig->stats || thread_group_empty(tsk))
597 goto ret;
598
599 /* No problem if kmem_cache_zalloc() fails */
600 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
601
602 spin_lock_irq(&tsk->sighand->siglock);
603 if (!sig->stats) {
604 sig->stats = stats;
605 stats = NULL;
606 }
607 spin_unlock_irq(&tsk->sighand->siglock);
608
609 if (stats)
610 kmem_cache_free(taskstats_cache, stats);
611ret:
612 return sig->stats;
613}
614
Shailabh Nagarc7572492006-07-14 00:24:40 -0700615/* Send pid data out on exit */
Oleg Nesterov115085e2006-12-06 20:36:51 -0800616void taskstats_exit(struct task_struct *tsk, int group_dead)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700617{
618 int rc;
Oleg Nesterov115085e2006-12-06 20:36:51 -0800619 struct listener_list *listeners;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800620 struct taskstats *stats;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700621 struct sk_buff *rep_skb;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700622 size_t size;
623 int is_thread_group;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700624
Oleg Nesterov4a279ff2006-10-30 22:07:15 -0800625 if (!family_registered)
Shailabh Nagarc7572492006-07-14 00:24:40 -0700626 return;
627
Shailabh Nagarc7572492006-07-14 00:24:40 -0700628 /*
629 * Size includes space for nested attributes
630 */
Jeff Mahoney4be2c952010-12-21 17:24:30 -0800631 size = taskstats_packet_size();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700632
Oleg Nesterov34ec1232006-12-06 20:36:52 -0800633 is_thread_group = !!taskstats_tgid_alloc(tsk);
Oleg Nesterov4a279ff2006-10-30 22:07:15 -0800634 if (is_thread_group) {
635 /* PID + STATS + TGID + STATS */
636 size = 2 * size;
637 /* fill the tsk->signal->stats structure */
638 fill_tgid_exit(tsk);
639 }
640
Christoph Lametercd85fc52010-12-08 17:42:22 +0100641 listeners = __this_cpu_ptr(&listener_array);
Oleg Nesterov115085e2006-12-06 20:36:51 -0800642 if (list_empty(&listeners->list))
643 return;
644
Oleg Nesterov37167482006-12-06 20:36:55 -0800645 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700646 if (rc < 0)
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800647 return;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700648
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800649 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
650 task_pid_nr_ns(tsk, &init_pid_ns));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800651 if (!stats)
Oleg Nesterov37167482006-12-06 20:36:55 -0800652 goto err;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800653
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800654 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700655
Shailabh Nagarc7572492006-07-14 00:24:40 -0700656 /*
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700657 * Doesn't matter if tsk is the leader or the last group member leaving
Shailabh Nagarc7572492006-07-14 00:24:40 -0700658 */
Oleg Nesterov68062b82006-12-06 20:36:53 -0800659 if (!is_thread_group || !group_dead)
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700660 goto send;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700661
Eric W. Biederman4bd6e322012-02-07 17:56:49 -0800662 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
663 task_tgid_nr_ns(tsk, &init_pid_ns));
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800664 if (!stats)
Oleg Nesterov37167482006-12-06 20:36:55 -0800665 goto err;
Oleg Nesterov51de4d92006-12-06 20:36:54 -0800666
667 memcpy(stats, tsk->signal->stats, sizeof(*stats));
Shailabh Nagarc7572492006-07-14 00:24:40 -0700668
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700669send:
Oleg Nesterov115085e2006-12-06 20:36:51 -0800670 send_cpu_listeners(rep_skb, listeners);
Shailabh Nagarad4ecbc2006-07-14 00:24:44 -0700671 return;
Oleg Nesterov37167482006-12-06 20:36:55 -0800672err:
Shailabh Nagarc7572492006-07-14 00:24:40 -0700673 nlmsg_free(rep_skb);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700674}
675
Johannes Berg4534de82013-11-14 17:14:46 +0100676static const struct genl_ops taskstats_ops[] = {
Johannes Berg88d36a92013-11-14 17:14:39 +0100677 {
678 .cmd = TASKSTATS_CMD_GET,
679 .doit = taskstats_user_cmd,
680 .policy = taskstats_cmd_get_policy,
681 .flags = GENL_ADMIN_PERM,
682 },
683 {
684 .cmd = CGROUPSTATS_CMD_GET,
685 .doit = cgroupstats_user_cmd,
686 .policy = cgroupstats_cmd_get_policy,
687 },
Balbir Singh846c7bb2007-10-18 23:39:44 -0700688};
689
Shailabh Nagarc7572492006-07-14 00:24:40 -0700690/* Needed early in initialization */
691void __init taskstats_init_early(void)
692{
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700693 unsigned int i;
694
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700695 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
Shailabh Nagarf9fd8912006-07-14 00:24:47 -0700696 for_each_possible_cpu(i) {
697 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
698 init_rwsem(&(per_cpu(listener_array, i).sem));
699 }
Shailabh Nagarc7572492006-07-14 00:24:40 -0700700}
701
702static int __init taskstats_init(void)
703{
704 int rc;
705
Johannes Bergc53ed742013-11-19 15:19:31 +0100706 rc = genl_register_family_with_ops(&family, taskstats_ops);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700707 if (rc)
708 return rc;
709
Shailabh Nagarc7572492006-07-14 00:24:40 -0700710 family_registered = 1;
Mandeep Singh Bainesf9b182e2011-03-23 16:43:27 -0700711 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
Shailabh Nagarc7572492006-07-14 00:24:40 -0700712 return 0;
Shailabh Nagarc7572492006-07-14 00:24:40 -0700713}
714
715/*
716 * late initcall ensures initialization of statistics collection
717 * mechanisms precedes initialization of the taskstats interface
718 */
719late_initcall(taskstats_init);