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/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 <linux/pid_namespace.h>
  31 #include <net/genetlink.h>
  32 #include <linux/atomic.h>
  33
  34 /*
  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
  40 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
  41 static int family_registered;
  42 struct kmem_cache *taskstats_cache;
  43
  44 static 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
  51 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
  52         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  53         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  54         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  55         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  56
  57 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
  58         [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
  59 };
  60
  61 struct listener {
  62         struct list_head list;
  63         pid_t pid;
  64         char valid;
  65 };
  66
  67 struct listener_list {
  68         struct rw_semaphore sem;
  69         struct list_head list;
  70 };
  71 static DEFINE_PER_CPU(struct listener_list, listener_array);
  72
  73 enum actions {
  74         REGISTER,
  75         DEREGISTER,
  76         CPU_DONT_CARE
  77 };
  78
  79 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  80                                 size_t size)
  81 {
  82         struct sk_buff *skb;
  83         void *reply;
  84
  85         /*
  86          * If new attributes are added, please revisit this allocation
  87          */
  88         skb = genlmsg_new(size, GFP_KERNEL);
  89         if (!skb)
  90                 return -ENOMEM;
  91
  92         if (!info) {
  93                 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
  94
  95                 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
  96         } else
  97                 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
  98         if (reply == NULL) {
  99                 nlmsg_free(skb);
 100                 return -EINVAL;
 101         }
 102
 103         *skbp = skb;
 104         return 0;
 105 }
 106
 107 /*
 108  * Send taskstats data in @skb to listener with nl_pid @pid
 109  */
 110 static int send_reply(struct sk_buff *skb, struct genl_info *info)
 111 {
 112         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 113         void *reply = genlmsg_data(genlhdr);
 114
 115         genlmsg_end(skb, reply);
 116
 117         return genlmsg_reply(skb, info);
 118 }
 119
 120 /*
 121  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 122  */
 123 static void send_cpu_listeners(struct sk_buff *skb,
 124                                         struct listener_list *listeners)
 125 {
 126         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 127         struct listener *s, *tmp;
 128         struct sk_buff *skb_next, *skb_cur = skb;
 129         void *reply = genlmsg_data(genlhdr);
 130         int rc, delcount = 0;
 131
 132         genlmsg_end(skb, reply);
 133
 134         rc = 0;
 135         down_read(&listeners->sem);
 136         list_for_each_entry(s, &listeners->list, list) {
 137                 skb_next = NULL;
 138                 if (!list_is_last(&s->list, &listeners->list)) {
 139                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 140                         if (!skb_next)
 141                                 break;
 142                 }
 143                 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
 144                 if (rc == -ECONNREFUSED) {
 145                         s->valid = 0;
 146                         delcount++;
 147                 }
 148                 skb_cur = skb_next;
 149         }
 150         up_read(&listeners->sem);
 151
 152         if (skb_cur)
 153                 nlmsg_free(skb_cur);
 154
 155         if (!delcount)
 156                 return;
 157
 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);
 167 }
 168
 169 static void fill_stats(struct user_namespace *user_ns,
 170                        struct pid_namespace *pid_ns,
 171                        struct task_struct *tsk, struct taskstats *stats)
 172 {
 173         memset(stats, 0, sizeof(*stats));
 174         /*
 175          * Each accounting subsystem adds calls to its functions to
 176          * fill in relevant parts of struct taskstsats as follows
 177          *
 178          *      per-task-foo(stats, tsk);
 179          */
 180
 181         delayacct_add_tsk(stats, tsk);
 182
 183         /* fill in basic acct fields */
 184         stats->version = TASKSTATS_VERSION;
 185         stats->nvcsw = tsk->nvcsw;
 186         stats->nivcsw = tsk->nivcsw;
 187         bacct_add_tsk(user_ns, pid_ns, stats, tsk);
 188
 189         /* fill in extended acct fields */
 190         xacct_add_tsk(stats, tsk);
 191 }
 192
 193 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
 194 {
 195         struct task_struct *tsk;
 196
 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;
 204         fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
 205         put_task_struct(tsk);
 206         return 0;
 207 }
 208
 209 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
 210 {
 211         struct task_struct *tsk, *first;
 212         unsigned long flags;
 213         int rc = -ESRCH;
 214
 215         /*
 216          * Add additional stats from live tasks except zombie thread group
 217          * leaders who are already counted with the dead tasks
 218          */
 219         rcu_read_lock();
 220         first = find_task_by_vpid(tgid);
 221
 222         if (!first || !lock_task_sighand(first, &flags))
 223                 goto out;
 224
 225         if (first->signal->stats)
 226                 memcpy(stats, first->signal->stats, sizeof(*stats));
 227         else
 228                 memset(stats, 0, sizeof(*stats));
 229
 230         tsk = first;
 231         do {
 232                 if (tsk->exit_state)
 233                         continue;
 234                 /*
 235                  * Accounting subsystem can call its functions here to
 236                  * fill in relevant parts of struct taskstsats as follows
 237                  *
 238                  *      per-task-foo(stats, tsk);
 239                  */
 240                 delayacct_add_tsk(stats, tsk);
 241
 242                 stats->nvcsw += tsk->nvcsw;
 243                 stats->nivcsw += tsk->nivcsw;
 244         } while_each_thread(first, tsk);
 245
 246         unlock_task_sighand(first, &flags);
 247         rc = 0;
 248 out:
 249         rcu_read_unlock();
 250
 251         stats->version = TASKSTATS_VERSION;
 252         /*
 253          * Accounting subsystems can also add calls here to modify
 254          * fields of taskstats.
 255          */
 256         return rc;
 257 }
 258
 259 static void fill_tgid_exit(struct task_struct *tsk)
 260 {
 261         unsigned long flags;
 262
 263         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 264         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);
 274 ret:
 275         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 276         return;
 277 }
 278
 279 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
 280 {
 281         struct listener_list *listeners;
 282         struct listener *s, *tmp, *s2;
 283         unsigned int cpu;
 284         int ret = 0;
 285
 286         if (!cpumask_subset(mask, cpu_possible_mask))
 287                 return -EINVAL;
 288
 289         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
 295         if (isadd == REGISTER) {
 296                 for_each_cpu(cpu, mask) {
 297                         s = kmalloc_node(sizeof(struct listener),
 298                                         GFP_KERNEL, cpu_to_node(cpu));
 299                         if (!s) {
 300                                 ret = -ENOMEM;
 301                                 goto cleanup;
 302                         }
 303                         s->pid = pid;
 304                         s->valid = 1;
 305
 306                         listeners = &per_cpu(listener_array, cpu);
 307                         down_write(&listeners->sem);
 308                         list_for_each_entry(s2, &listeners->list, list) {
 309                                 if (s2->pid == pid && s2->valid)
 310                                         goto exists;
 311                         }
 312                         list_add(&s->list, &listeners->list);
 313                         s = NULL;
 314 exists:
 315                         up_write(&listeners->sem);
 316                         kfree(s); /* nop if NULL */
 317                 }
 318                 return 0;
 319         }
 320
 321         /* Deregister or cleanup */
 322 cleanup:
 323         for_each_cpu(cpu, mask) {
 324                 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         }
 335         return ret;
 336 }
 337
 338 static int parse(struct nlattr *na, struct cpumask *mask)
 339 {
 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);
 355         ret = cpulist_parse(data, mask);
 356         kfree(data);
 357         return ret;
 358 }
 359
 360 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
 361 {
 362         struct nlattr *na, *ret;
 363         int aggr;
 364
 365         aggr = (type == TASKSTATS_TYPE_PID)
 366                         ? TASKSTATS_TYPE_AGGR_PID
 367                         : TASKSTATS_TYPE_AGGR_TGID;
 368
 369         na = nla_nest_start(skb, aggr);
 370         if (!na)
 371                 goto err;
 372
 373         if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
 374                 nla_nest_cancel(skb, na);
 375                 goto err;
 376         }
 377         ret = nla_reserve_64bit(skb, TASKSTATS_TYPE_STATS,
 378                                 sizeof(struct taskstats), TASKSTATS_TYPE_NULL);
 379         if (!ret) {
 380                 nla_nest_cancel(skb, na);
 381                 goto err;
 382         }
 383         nla_nest_end(skb, na);
 384
 385         return nla_data(ret);
 386 err:
 387         return NULL;
 388 }
 389
 390 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 391 {
 392         int rc = 0;
 393         struct sk_buff *rep_skb;
 394         struct cgroupstats *stats;
 395         struct nlattr *na;
 396         size_t size;
 397         u32 fd;
 398         struct fd f;
 399
 400         na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
 401         if (!na)
 402                 return -EINVAL;
 403
 404         fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
 405         f = fdget(fd);
 406         if (!f.file)
 407                 return 0;
 408
 409         size = nla_total_size(sizeof(struct cgroupstats));
 410
 411         rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
 412                                 size);
 413         if (rc < 0)
 414                 goto err;
 415
 416         na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
 417                                 sizeof(struct cgroupstats));
 418         if (na == NULL) {
 419                 nlmsg_free(rep_skb);
 420                 rc = -EMSGSIZE;
 421                 goto err;
 422         }
 423
 424         stats = nla_data(na);
 425         memset(stats, 0, sizeof(*stats));
 426
 427         rc = cgroupstats_build(stats, f.file->f_path.dentry);
 428         if (rc < 0) {
 429                 nlmsg_free(rep_skb);
 430                 goto err;
 431         }
 432
 433         rc = send_reply(rep_skb, info);
 434
 435 err:
 436         fdput(f);
 437         return rc;
 438 }
 439
 440 static int cmd_attr_register_cpumask(struct genl_info *info)
 441 {
 442         cpumask_var_t mask;
 443         int rc;
 444
 445         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 446                 return -ENOMEM;
 447         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
 448         if (rc < 0)
 449                 goto out;
 450         rc = add_del_listener(info->snd_portid, mask, REGISTER);
 451 out:
 452         free_cpumask_var(mask);
 453         return rc;
 454 }
 455
 456 static int cmd_attr_deregister_cpumask(struct genl_info *info)
 457 {
 458         cpumask_var_t mask;
 459         int rc;
 460
 461         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 462                 return -ENOMEM;
 463         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
 464         if (rc < 0)
 465                 goto out;
 466         rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
 467 out:
 468         free_cpumask_var(mask);
 469         return rc;
 470 }
 471
 472 static size_t taskstats_packet_size(void)
 473 {
 474         size_t size;
 475
 476         size = nla_total_size(sizeof(u32)) +
 477                 nla_total_size_64bit(sizeof(struct taskstats)) +
 478                 nla_total_size(0);
 479
 480         return size;
 481 }
 482
 483 static int cmd_attr_pid(struct genl_info *info)
 484 {
 485         struct taskstats *stats;
 486         struct sk_buff *rep_skb;
 487         size_t size;
 488         u32 pid;
 489         int rc;
 490
 491         size = taskstats_packet_size();
 492
 493         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 494         if (rc < 0)
 495                 return rc;
 496
 497         rc = -EINVAL;
 498         pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 499         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 500         if (!stats)
 501                 goto err;
 502
 503         rc = fill_stats_for_pid(pid, stats);
 504         if (rc < 0)
 505                 goto err;
 506         return send_reply(rep_skb, info);
 507 err:
 508         nlmsg_free(rep_skb);
 509         return rc;
 510 }
 511
 512 static int cmd_attr_tgid(struct genl_info *info)
 513 {
 514         struct taskstats *stats;
 515         struct sk_buff *rep_skb;
 516         size_t size;
 517         u32 tgid;
 518         int rc;
 519
 520         size = taskstats_packet_size();
 521
 522         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 523         if (rc < 0)
 524                 return rc;
 525
 526         rc = -EINVAL;
 527         tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 528         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 529         if (!stats)
 530                 goto err;
 531
 532         rc = fill_stats_for_tgid(tgid, stats);
 533         if (rc < 0)
 534                 goto err;
 535         return send_reply(rep_skb, info);
 536 err:
 537         nlmsg_free(rep_skb);
 538         return rc;
 539 }
 540
 541 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 542 {
 543         if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
 544                 return cmd_attr_register_cpumask(info);
 545         else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
 546                 return cmd_attr_deregister_cpumask(info);
 547         else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
 548                 return cmd_attr_pid(info);
 549         else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
 550                 return cmd_attr_tgid(info);
 551         else
 552                 return -EINVAL;
 553 }
 554
 555 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 556 {
 557         struct signal_struct *sig = tsk->signal;
 558         struct taskstats *stats;
 559
 560         if (sig->stats || thread_group_empty(tsk))
 561                 goto ret;
 562
 563         /* No problem if kmem_cache_zalloc() fails */
 564         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 565
 566         spin_lock_irq(&tsk->sighand->siglock);
 567         if (!sig->stats) {
 568                 sig->stats = stats;
 569                 stats = NULL;
 570         }
 571         spin_unlock_irq(&tsk->sighand->siglock);
 572
 573         if (stats)
 574                 kmem_cache_free(taskstats_cache, stats);
 575 ret:
 576         return sig->stats;
 577 }
 578
 579 /* Send pid data out on exit */
 580 void taskstats_exit(struct task_struct *tsk, int group_dead)
 581 {
 582         int rc;
 583         struct listener_list *listeners;
 584         struct taskstats *stats;
 585         struct sk_buff *rep_skb;
 586         size_t size;
 587         int is_thread_group;
 588
 589         if (!family_registered)
 590                 return;
 591
 592         /*
 593          * Size includes space for nested attributes
 594          */
 595         size = taskstats_packet_size();
 596
 597         is_thread_group = !!taskstats_tgid_alloc(tsk);
 598         if (is_thread_group) {
 599                 /* PID + STATS + TGID + STATS */
 600                 size = 2 * size;
 601                 /* fill the tsk->signal->stats structure */
 602                 fill_tgid_exit(tsk);
 603         }
 604
 605         listeners = raw_cpu_ptr(&listener_array);
 606         if (list_empty(&listeners->list))
 607                 return;
 608
 609         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 610         if (rc < 0)
 611                 return;
 612
 613         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
 614                          task_pid_nr_ns(tsk, &init_pid_ns));
 615         if (!stats)
 616                 goto err;
 617
 618         fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
 619
 620         /*
 621          * Doesn't matter if tsk is the leader or the last group member leaving
 622          */
 623         if (!is_thread_group || !group_dead)
 624                 goto send;
 625
 626         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
 627                          task_tgid_nr_ns(tsk, &init_pid_ns));
 628         if (!stats)
 629                 goto err;
 630
 631         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 632
 633 send:
 634         send_cpu_listeners(rep_skb, listeners);
 635         return;
 636 err:
 637         nlmsg_free(rep_skb);
 638 }
 639
 640 static const struct genl_ops taskstats_ops[] = {
 641         {
 642                 .cmd            = TASKSTATS_CMD_GET,
 643                 .doit           = taskstats_user_cmd,
 644                 .policy         = taskstats_cmd_get_policy,
 645                 .flags          = GENL_ADMIN_PERM,
 646         },
 647         {
 648                 .cmd            = CGROUPSTATS_CMD_GET,
 649                 .doit           = cgroupstats_user_cmd,
 650                 .policy         = cgroupstats_cmd_get_policy,
 651         },
 652 };
 653
 654 /* Needed early in initialization */
 655 void __init taskstats_init_early(void)
 656 {
 657         unsigned int i;
 658
 659         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 660         for_each_possible_cpu(i) {
 661                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 662                 init_rwsem(&(per_cpu(listener_array, i).sem));
 663         }
 664 }
 665
 666 static int __init taskstats_init(void)
 667 {
 668         int rc;
 669
 670         rc = genl_register_family_with_ops(&family, taskstats_ops);
 671         if (rc)
 672                 return rc;
 673
 674         family_registered = 1;
 675         pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
 676         return 0;
 677 }
 678
 679 /*
 680  * late initcall ensures initialization of statistics collection
 681  * mechanisms precedes initialization of the taskstats interface
 682  */
 683 late_initcall(taskstats_init);