kernel/taskstats.c

   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/tsacct_kern.h>
  24 #include <linux/cpumask.h>
  25 #include <linux/percpu.h>
  26 #include <net/genetlink.h>
  27 #include <asm/atomic.h>
  28
  29 /*
  30  * Maximum length of a cpumask that can be specified in
  31  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  32  */
  33 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  34
  35 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
  36 static int family_registered;
  37 kmem_cache_t *taskstats_cache;
  38
  39 static struct genl_family family = {
  40         .id             = GENL_ID_GENERATE,
  41         .name           = TASKSTATS_GENL_NAME,
  42         .version        = TASKSTATS_GENL_VERSION,
  43         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  44 };
  45
  46 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
  47 __read_mostly = {
  48         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  49         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  50         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  51         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  52
  53 struct listener {
  54         struct list_head list;
  55         pid_t pid;
  56         char valid;
  57 };
  58
  59 struct listener_list {
  60         struct rw_semaphore sem;
  61         struct list_head list;
  62 };
  63 static DEFINE_PER_CPU(struct listener_list, listener_array);
  64
  65 enum actions {
  66         REGISTER,
  67         DEREGISTER,
  68         CPU_DONT_CARE
  69 };
  70
  71 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  72                         void **replyp, size_t size)
  73 {
  74         struct sk_buff *skb;
  75         void *reply;
  76
  77         /*
  78          * If new attributes are added, please revisit this allocation
  79          */
  80         skb = nlmsg_new(genlmsg_total_size(size), GFP_KERNEL);
  81         if (!skb)
  82                 return -ENOMEM;
  83
  84         if (!info) {
  85                 int seq = get_cpu_var(taskstats_seqnum)++;
  86                 put_cpu_var(taskstats_seqnum);
  87
  88                 reply = genlmsg_put(skb, 0, seq,
  89                                 family.id, 0, 0,
  90                                 cmd, family.version);
  91         } else
  92                 reply = genlmsg_put(skb, info->snd_pid, info->snd_seq,
  93                                 family.id, 0, 0,
  94                                 cmd, family.version);
  95         if (reply == NULL) {
  96                 nlmsg_free(skb);
  97                 return -EINVAL;
  98         }
  99
 100         *skbp = skb;
 101         *replyp = reply;
 102         return 0;
 103 }
 104
 105 /*
 106  * Send taskstats data in @skb to listener with nl_pid @pid
 107  */
 108 static int send_reply(struct sk_buff *skb, pid_t pid)
 109 {
 110         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 111         void *reply = genlmsg_data(genlhdr);
 112         int rc;
 113
 114         rc = genlmsg_end(skb, reply);
 115         if (rc < 0) {
 116                 nlmsg_free(skb);
 117                 return rc;
 118         }
 119
 120         return genlmsg_unicast(skb, pid);
 121 }
 122
 123 /*
 124  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 125  */
 126 static void send_cpu_listeners(struct sk_buff *skb, unsigned int cpu)
 127 {
 128         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 129         struct listener_list *listeners;
 130         struct listener *s, *tmp;
 131         struct sk_buff *skb_next, *skb_cur = skb;
 132         void *reply = genlmsg_data(genlhdr);
 133         int rc, delcount = 0;
 134
 135         rc = genlmsg_end(skb, reply);
 136         if (rc < 0) {
 137                 nlmsg_free(skb);
 138                 return;
 139         }
 140
 141         rc = 0;
 142         listeners = &per_cpu(listener_array, cpu);
 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;
 150                 }
 151                 rc = genlmsg_unicast(skb_cur, s->pid);
 152                 if (rc == -ECONNREFUSED) {
 153                         s->valid = 0;
 154                         delcount++;
 155                 }
 156                 skb_cur = skb_next;
 157         }
 158         up_read(&listeners->sem);
 159
 160         if (skb_cur)
 161                 nlmsg_free(skb_cur);
 162
 163         if (!delcount)
 164                 return;
 165
 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);
 172                 }
 173         }
 174         up_write(&listeners->sem);
 175 }
 176
 177 static int fill_pid(pid_t pid, struct task_struct *tsk,
 178                 struct taskstats *stats)
 179 {
 180         int rc = 0;
 181
 182         if (!tsk) {
 183                 rcu_read_lock();
 184                 tsk = find_task_by_pid(pid);
 185                 if (tsk)
 186                         get_task_struct(tsk);
 187                 rcu_read_unlock();
 188                 if (!tsk)
 189                         return -ESRCH;
 190         } else
 191                 get_task_struct(tsk);
 192
 193         /*
 194          * Each accounting subsystem adds calls to its functions to
 195          * fill in relevant parts of struct taskstsats as follows
 196          *
 197          *      per-task-foo(stats, tsk);
 198          */
 199
 200         delayacct_add_tsk(stats, tsk);
 201
 202         /* fill in basic acct fields */
 203         stats->version = TASKSTATS_VERSION;
 204         bacct_add_tsk(stats, tsk);
 205
 206         /* fill in extended acct fields */
 207         xacct_add_tsk(stats, tsk);
 208
 209         /* Define err: label here if needed */
 210         put_task_struct(tsk);
 211         return rc;
 212
 213 }
 214
 215 static int fill_tgid(pid_t tgid, struct task_struct *first,
 216                 struct taskstats *stats)
 217 {
 218         struct task_struct *tsk;
 219         unsigned long flags;
 220         int rc = -ESRCH;
 221
 222         /*
 223          * Add additional stats from live tasks except zombie thread group
 224          * leaders who are already counted with the dead tasks
 225          */
 226         rcu_read_lock();
 227         if (!first)
 228                 first = find_task_by_pid(tgid);
 229
 230         if (!first || !lock_task_sighand(first, &flags))
 231                 goto out;
 232
 233         if (first->signal->stats)
 234                 memcpy(stats, first->signal->stats, sizeof(*stats));
 235
 236         tsk = first;
 237         do {
 238                 if (tsk->exit_state)
 239                         continue;
 240                 /*
 241                  * Accounting subsystem can call its functions here to
 242                  * fill in relevant parts of struct taskstsats as follows
 243                  *
 244                  *      per-task-foo(stats, tsk);
 245                  */
 246                 delayacct_add_tsk(stats, tsk);
 247
 248         } while_each_thread(first, tsk);
 249
 250         unlock_task_sighand(first, &flags);
 251         rc = 0;
 252 out:
 253         rcu_read_unlock();
 254
 255         stats->version = TASKSTATS_VERSION;
 256         /*
 257          * Accounting subsytems can also add calls here to modify
 258          * fields of taskstats.
 259          */
 260         return rc;
 261 }
 262
 263
 264 static void fill_tgid_exit(struct task_struct *tsk)
 265 {
 266         unsigned long flags;
 267
 268         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 269         if (!tsk->signal->stats)
 270                 goto ret;
 271
 272         /*
 273          * Each accounting subsystem calls its functions here to
 274          * accumalate its per-task stats for tsk, into the per-tgid structure
 275          *
 276          *      per-task-foo(tsk->signal->stats, tsk);
 277          */
 278         delayacct_add_tsk(tsk->signal->stats, tsk);
 279 ret:
 280         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 281         return;
 282 }
 283
 284 static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
 285 {
 286         struct listener_list *listeners;
 287         struct listener *s, *tmp;
 288         unsigned int cpu;
 289         cpumask_t mask = *maskp;
 290
 291         if (!cpus_subset(mask, cpu_possible_map))
 292                 return -EINVAL;
 293
 294         if (isadd == REGISTER) {
 295                 for_each_cpu_mask(cpu, mask) {
 296                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 297                                          cpu_to_node(cpu));
 298                         if (!s)
 299                                 goto cleanup;
 300                         s->pid = pid;
 301                         INIT_LIST_HEAD(&s->list);
 302                         s->valid = 1;
 303
 304                         listeners = &per_cpu(listener_array, cpu);
 305                         down_write(&listeners->sem);
 306                         list_add(&s->list, &listeners->list);
 307                         up_write(&listeners->sem);
 308                 }
 309                 return 0;
 310         }
 311
 312         /* Deregister or cleanup */
 313 cleanup:
 314         for_each_cpu_mask(cpu, mask) {
 315                 listeners = &per_cpu(listener_array, cpu);
 316                 down_write(&listeners->sem);
 317                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 318                         if (s->pid == pid) {
 319                                 list_del(&s->list);
 320                                 kfree(s);
 321                                 break;
 322                         }
 323                 }
 324                 up_write(&listeners->sem);
 325         }
 326         return 0;
 327 }
 328
 329 static int parse(struct nlattr *na, cpumask_t *mask)
 330 {
 331         char *data;
 332         int len;
 333         int ret;
 334
 335         if (na == NULL)
 336                 return 1;
 337         len = nla_len(na);
 338         if (len > TASKSTATS_CPUMASK_MAXLEN)
 339                 return -E2BIG;
 340         if (len < 1)
 341                 return -EINVAL;
 342         data = kmalloc(len, GFP_KERNEL);
 343         if (!data)
 344                 return -ENOMEM;
 345         nla_strlcpy(data, na, len);
 346         ret = cpulist_parse(data, *mask);
 347         kfree(data);
 348         return ret;
 349 }
 350
 351 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 352 {
 353         int rc = 0;
 354         struct sk_buff *rep_skb;
 355         struct taskstats stats;
 356         void *reply;
 357         size_t size;
 358         struct nlattr *na;
 359         cpumask_t mask;
 360
 361         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 362         if (rc < 0)
 363                 return rc;
 364         if (rc == 0)
 365                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 366
 367         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 368         if (rc < 0)
 369                 return rc;
 370         if (rc == 0)
 371                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 372
 373         /*
 374          * Size includes space for nested attributes
 375          */
 376         size = nla_total_size(sizeof(u32)) +
 377                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 378
 379         memset(&stats, 0, sizeof(stats));
 380         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 381         if (rc < 0)
 382                 return rc;
 383
 384         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 385                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 386                 rc = fill_pid(pid, NULL, &stats);
 387                 if (rc < 0)
 388                         goto err;
 389
 390                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 391                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
 392                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 393                                 stats);
 394         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 395                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 396                 rc = fill_tgid(tgid, NULL, &stats);
 397                 if (rc < 0)
 398                         goto err;
 399
 400                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 401                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 402                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 403                                 stats);
 404         } else {
 405                 rc = -EINVAL;
 406                 goto err;
 407         }
 408
 409         nla_nest_end(rep_skb, na);
 410
 411         return send_reply(rep_skb, info->snd_pid);
 412
 413 nla_put_failure:
 414         rc = genlmsg_cancel(rep_skb, reply);
 415 err:
 416         nlmsg_free(rep_skb);
 417         return rc;
 418 }
 419
 420 void taskstats_exit_alloc(struct taskstats **ptidstats, unsigned int *mycpu)
 421 {
 422         struct listener_list *listeners;
 423         struct taskstats *tmp;
 424         /*
 425          * This is the cpu on which the task is exiting currently and will
 426          * be the one for which the exit event is sent, even if the cpu
 427          * on which this function is running changes later.
 428          */
 429         *mycpu = raw_smp_processor_id();
 430
 431         *ptidstats = NULL;
 432         tmp = kmem_cache_zalloc(taskstats_cache, SLAB_KERNEL);
 433         if (!tmp)
 434                 return;
 435
 436         listeners = &per_cpu(listener_array, *mycpu);
 437         down_read(&listeners->sem);
 438         if (!list_empty(&listeners->list)) {
 439                 *ptidstats = tmp;
 440                 tmp = NULL;
 441         }
 442         up_read(&listeners->sem);
 443         kfree(tmp);
 444 }
 445
 446 /* Send pid data out on exit */
 447 void taskstats_exit_send(struct task_struct *tsk, struct taskstats *tidstats,
 448                         int group_dead, unsigned int mycpu)
 449 {
 450         int rc;
 451         struct sk_buff *rep_skb;
 452         void *reply;
 453         size_t size;
 454         int is_thread_group;
 455         struct nlattr *na;
 456
 457         if (!family_registered)
 458                 return;
 459
 460         /*
 461          * Size includes space for nested attributes
 462          */
 463         size = nla_total_size(sizeof(u32)) +
 464                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 465
 466         is_thread_group = (tsk->signal->stats != NULL);
 467         if (is_thread_group) {
 468                 /* PID + STATS + TGID + STATS */
 469                 size = 2 * size;
 470                 /* fill the tsk->signal->stats structure */
 471                 fill_tgid_exit(tsk);
 472         }
 473
 474         if (!tidstats)
 475                 return;
 476
 477         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 478         if (rc < 0)
 479                 goto ret;
 480
 481         rc = fill_pid(tsk->pid, tsk, tidstats);
 482         if (rc < 0)
 483                 goto err_skb;
 484
 485         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 486         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
 487         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 488                         *tidstats);
 489         nla_nest_end(rep_skb, na);
 490
 491         if (!is_thread_group)
 492                 goto send;
 493
 494         /*
 495          * Doesn't matter if tsk is the leader or the last group member leaving
 496          */
 497         if (!group_dead)
 498                 goto send;
 499
 500         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 501         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
 502         /* No locking needed for tsk->signal->stats since group is dead */
 503         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 504                         *tsk->signal->stats);
 505         nla_nest_end(rep_skb, na);
 506
 507 send:
 508         send_cpu_listeners(rep_skb, mycpu);
 509         return;
 510
 511 nla_put_failure:
 512         genlmsg_cancel(rep_skb, reply);
 513 err_skb:
 514         nlmsg_free(rep_skb);
 515 ret:
 516         return;
 517 }
 518
 519 static struct genl_ops taskstats_ops = {
 520         .cmd            = TASKSTATS_CMD_GET,
 521         .doit           = taskstats_user_cmd,
 522         .policy         = taskstats_cmd_get_policy,
 523 };
 524
 525 /* Needed early in initialization */
 526 void __init taskstats_init_early(void)
 527 {
 528         unsigned int i;
 529
 530         taskstats_cache = kmem_cache_create("taskstats_cache",
 531                                                 sizeof(struct taskstats),
 532                                                 0, SLAB_PANIC, NULL, NULL);
 533         for_each_possible_cpu(i) {
 534                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 535                 init_rwsem(&(per_cpu(listener_array, i).sem));
 536         }
 537 }
 538
 539 static int __init taskstats_init(void)
 540 {
 541         int rc;
 542
 543         rc = genl_register_family(&family);
 544         if (rc)
 545                 return rc;
 546
 547         rc = genl_register_ops(&family, &taskstats_ops);
 548         if (rc < 0)
 549                 goto err;
 550
 551         family_registered = 1;
 552         return 0;
 553 err:
 554         genl_unregister_family(&family);
 555         return rc;
 556 }
 557
 558 /*
 559  * late initcall ensures initialization of statistics collection
 560  * mechanisms precedes initialization of the taskstats interface
 561  */
 562 late_initcall(taskstats_init);