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 <net/genetlink.h>
  31 #include <asm/atomic.h>
  32
  33 /*
  34  * Maximum length of a cpumask that can be specified in
  35  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  36  */
  37 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  38
  39 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
  40 static int family_registered;
  41 struct kmem_cache *taskstats_cache;
  42
  43 static struct genl_family family = {
  44         .id             = GENL_ID_GENERATE,
  45         .name           = TASKSTATS_GENL_NAME,
  46         .version        = TASKSTATS_GENL_VERSION,
  47         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  48 };
  49
  50 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
  51         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  52         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  53         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  54         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  55
  56 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
  57         [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
  58 };
  59
  60 struct listener {
  61         struct list_head list;
  62         pid_t pid;
  63         char valid;
  64 };
  65
  66 struct listener_list {
  67         struct rw_semaphore sem;
  68         struct list_head list;
  69 };
  70 static DEFINE_PER_CPU(struct listener_list, listener_array);
  71
  72 enum actions {
  73         REGISTER,
  74         DEREGISTER,
  75         CPU_DONT_CARE
  76 };
  77
  78 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  79                                 size_t size)
  80 {
  81         struct sk_buff *skb;
  82         void *reply;
  83
  84         /*
  85          * If new attributes are added, please revisit this allocation
  86          */
  87         skb = genlmsg_new(size, GFP_KERNEL);
  88         if (!skb)
  89                 return -ENOMEM;
  90
  91         if (!info) {
  92                 int seq = get_cpu_var(taskstats_seqnum)++;
  93                 put_cpu_var(taskstats_seqnum);
  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         int rc;
 115
 116         rc = genlmsg_end(skb, reply);
 117         if (rc < 0) {
 118                 nlmsg_free(skb);
 119                 return rc;
 120         }
 121
 122         return genlmsg_reply(skb, info);
 123 }
 124
 125 /*
 126  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 127  */
 128 static void send_cpu_listeners(struct sk_buff *skb,
 129                                         struct listener_list *listeners)
 130 {
 131         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 132         struct listener *s, *tmp;
 133         struct sk_buff *skb_next, *skb_cur = skb;
 134         void *reply = genlmsg_data(genlhdr);
 135         int rc, delcount = 0;
 136
 137         rc = genlmsg_end(skb, reply);
 138         if (rc < 0) {
 139                 nlmsg_free(skb);
 140                 return;
 141         }
 142
 143         rc = 0;
 144         down_read(&listeners->sem);
 145         list_for_each_entry(s, &listeners->list, list) {
 146                 skb_next = NULL;
 147                 if (!list_is_last(&s->list, &listeners->list)) {
 148                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 149                         if (!skb_next)
 150                                 break;
 151                 }
 152                 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
 153                 if (rc == -ECONNREFUSED) {
 154                         s->valid = 0;
 155                         delcount++;
 156                 }
 157                 skb_cur = skb_next;
 158         }
 159         up_read(&listeners->sem);
 160
 161         if (skb_cur)
 162                 nlmsg_free(skb_cur);
 163
 164         if (!delcount)
 165                 return;
 166
 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);
 176 }
 177
 178 static void fill_stats(struct task_struct *tsk, struct taskstats *stats)
 179 {
 180         memset(stats, 0, sizeof(*stats));
 181         /*
 182          * Each accounting subsystem adds calls to its functions to
 183          * fill in relevant parts of struct taskstsats as follows
 184          *
 185          *      per-task-foo(stats, tsk);
 186          */
 187
 188         delayacct_add_tsk(stats, tsk);
 189
 190         /* fill in basic acct fields */
 191         stats->version = TASKSTATS_VERSION;
 192         stats->nvcsw = tsk->nvcsw;
 193         stats->nivcsw = tsk->nivcsw;
 194         bacct_add_tsk(stats, tsk);
 195
 196         /* fill in extended acct fields */
 197         xacct_add_tsk(stats, tsk);
 198 }
 199
 200 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
 201 {
 202         struct task_struct *tsk;
 203
 204         rcu_read_lock();
 205         tsk = find_task_by_vpid(pid);
 206         if (tsk)
 207                 get_task_struct(tsk);
 208         rcu_read_unlock();
 209         if (!tsk)
 210                 return -ESRCH;
 211         fill_stats(tsk, stats);
 212         put_task_struct(tsk);
 213         return 0;
 214 }
 215
 216 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
 217 {
 218         struct task_struct *tsk, *first;
 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         first = find_task_by_vpid(tgid);
 228
 229         if (!first || !lock_task_sighand(first, &flags))
 230                 goto out;
 231
 232         if (first->signal->stats)
 233                 memcpy(stats, first->signal->stats, sizeof(*stats));
 234         else
 235                 memset(stats, 0, sizeof(*stats));
 236
 237         tsk = first;
 238         do {
 239                 if (tsk->exit_state)
 240                         continue;
 241                 /*
 242                  * Accounting subsystem can call its functions here to
 243                  * fill in relevant parts of struct taskstsats as follows
 244                  *
 245                  *      per-task-foo(stats, tsk);
 246                  */
 247                 delayacct_add_tsk(stats, tsk);
 248
 249                 stats->nvcsw += tsk->nvcsw;
 250                 stats->nivcsw += tsk->nivcsw;
 251         } while_each_thread(first, tsk);
 252
 253         unlock_task_sighand(first, &flags);
 254         rc = 0;
 255 out:
 256         rcu_read_unlock();
 257
 258         stats->version = TASKSTATS_VERSION;
 259         /*
 260          * Accounting subsystems can also add calls here to modify
 261          * fields of taskstats.
 262          */
 263         return rc;
 264 }
 265
 266 static void fill_tgid_exit(struct task_struct *tsk)
 267 {
 268         unsigned long flags;
 269
 270         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 271         if (!tsk->signal->stats)
 272                 goto ret;
 273
 274         /*
 275          * Each accounting subsystem calls its functions here to
 276          * accumalate its per-task stats for tsk, into the per-tgid structure
 277          *
 278          *      per-task-foo(tsk->signal->stats, tsk);
 279          */
 280         delayacct_add_tsk(tsk->signal->stats, tsk);
 281 ret:
 282         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 283         return;
 284 }
 285
 286 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
 287 {
 288         struct listener_list *listeners;
 289         struct listener *s, *tmp;
 290         unsigned int cpu;
 291
 292         if (!cpumask_subset(mask, cpu_possible_mask))
 293                 return -EINVAL;
 294
 295         if (isadd == REGISTER) {
 296                 for_each_cpu(cpu, mask) {
 297                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 298                                          cpu_to_node(cpu));
 299                         if (!s)
 300                                 goto cleanup;
 301                         s->pid = pid;
 302                         INIT_LIST_HEAD(&s->list);
 303                         s->valid = 1;
 304
 305                         listeners = &per_cpu(listener_array, cpu);
 306                         down_write(&listeners->sem);
 307                         list_add(&s->list, &listeners->list);
 308                         up_write(&listeners->sem);
 309                 }
 310                 return 0;
 311         }
 312
 313         /* Deregister or cleanup */
 314 cleanup:
 315         for_each_cpu(cpu, mask) {
 316                 listeners = &per_cpu(listener_array, cpu);
 317                 down_write(&listeners->sem);
 318                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 319                         if (s->pid == pid) {
 320                                 list_del(&s->list);
 321                                 kfree(s);
 322                                 break;
 323                         }
 324                 }
 325                 up_write(&listeners->sem);
 326         }
 327         return 0;
 328 }
 329
 330 static int parse(struct nlattr *na, struct cpumask *mask)
 331 {
 332         char *data;
 333         int len;
 334         int ret;
 335
 336         if (na == NULL)
 337                 return 1;
 338         len = nla_len(na);
 339         if (len > TASKSTATS_CPUMASK_MAXLEN)
 340                 return -E2BIG;
 341         if (len < 1)
 342                 return -EINVAL;
 343         data = kmalloc(len, GFP_KERNEL);
 344         if (!data)
 345                 return -ENOMEM;
 346         nla_strlcpy(data, na, len);
 347         ret = cpulist_parse(data, mask);
 348         kfree(data);
 349         return ret;
 350 }
 351
 352 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
 353 {
 354         struct nlattr *na, *ret;
 355         int aggr;
 356
 357         /* If we don't pad, we end up with alignment on a 4 byte boundary.
 358          * This causes lots of runtime warnings on systems requiring 8 byte
 359          * alignment */
 360         u32 pids[2] = { pid, 0 };
 361         int pid_size = ALIGN(sizeof(pid), sizeof(long));
 362
 363         aggr = (type == TASKSTATS_TYPE_PID)
 364                         ? TASKSTATS_TYPE_AGGR_PID
 365                         : TASKSTATS_TYPE_AGGR_TGID;
 366
 367         na = nla_nest_start(skb, aggr);
 368         if (!na)
 369                 goto err;
 370         if (nla_put(skb, type, pid_size, pids) < 0)
 371                 goto err;
 372         ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
 373         if (!ret)
 374                 goto err;
 375         nla_nest_end(skb, na);
 376
 377         return nla_data(ret);
 378 err:
 379         return NULL;
 380 }
 381
 382 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 383 {
 384         int rc = 0;
 385         struct sk_buff *rep_skb;
 386         struct cgroupstats *stats;
 387         struct nlattr *na;
 388         size_t size;
 389         u32 fd;
 390         struct file *file;
 391         int fput_needed;
 392
 393         na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
 394         if (!na)
 395                 return -EINVAL;
 396
 397         fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
 398         file = fget_light(fd, &fput_needed);
 399         if (!file)
 400                 return 0;
 401
 402         size = nla_total_size(sizeof(struct cgroupstats));
 403
 404         rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
 405                                 size);
 406         if (rc < 0)
 407                 goto err;
 408
 409         na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
 410                                 sizeof(struct cgroupstats));
 411         stats = nla_data(na);
 412         memset(stats, 0, sizeof(*stats));
 413
 414         rc = cgroupstats_build(stats, file->f_dentry);
 415         if (rc < 0) {
 416                 nlmsg_free(rep_skb);
 417                 goto err;
 418         }
 419
 420         rc = send_reply(rep_skb, info);
 421
 422 err:
 423         fput_light(file, fput_needed);
 424         return rc;
 425 }
 426
 427 static int cmd_attr_register_cpumask(struct genl_info *info)
 428 {
 429         cpumask_var_t mask;
 430         int rc;
 431
 432         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 433                 return -ENOMEM;
 434         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
 435         if (rc < 0)
 436                 goto out;
 437         rc = add_del_listener(info->snd_pid, mask, REGISTER);
 438 out:
 439         free_cpumask_var(mask);
 440         return rc;
 441 }
 442
 443 static int cmd_attr_deregister_cpumask(struct genl_info *info)
 444 {
 445         cpumask_var_t mask;
 446         int rc;
 447
 448         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 449                 return -ENOMEM;
 450         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
 451         if (rc < 0)
 452                 goto out;
 453         rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
 454 out:
 455         free_cpumask_var(mask);
 456         return rc;
 457 }
 458
 459 static int cmd_attr_pid(struct genl_info *info)
 460 {
 461         struct taskstats *stats;
 462         struct sk_buff *rep_skb;
 463         size_t size;
 464         u32 pid;
 465         int rc;
 466
 467         size = nla_total_size(sizeof(u32)) +
 468                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 469
 470         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 471         if (rc < 0)
 472                 return rc;
 473
 474         rc = -EINVAL;
 475         pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 476         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 477         if (!stats)
 478                 goto err;
 479
 480         rc = fill_stats_for_pid(pid, stats);
 481         if (rc < 0)
 482                 goto err;
 483         return send_reply(rep_skb, info);
 484 err:
 485         nlmsg_free(rep_skb);
 486         return rc;
 487 }
 488
 489 static int cmd_attr_tgid(struct genl_info *info)
 490 {
 491         struct taskstats *stats;
 492         struct sk_buff *rep_skb;
 493         size_t size;
 494         u32 tgid;
 495         int rc;
 496
 497         size = nla_total_size(sizeof(u32)) +
 498                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 499
 500         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 501         if (rc < 0)
 502                 return rc;
 503
 504         rc = -EINVAL;
 505         tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 506         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 507         if (!stats)
 508                 goto err;
 509
 510         rc = fill_stats_for_tgid(tgid, stats);
 511         if (rc < 0)
 512                 goto err;
 513         return send_reply(rep_skb, info);
 514 err:
 515         nlmsg_free(rep_skb);
 516         return rc;
 517 }
 518
 519 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 520 {
 521         if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
 522                 return cmd_attr_register_cpumask(info);
 523         else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
 524                 return cmd_attr_deregister_cpumask(info);
 525         else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
 526                 return cmd_attr_pid(info);
 527         else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
 528                 return cmd_attr_tgid(info);
 529         else
 530                 return -EINVAL;
 531 }
 532
 533 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 534 {
 535         struct signal_struct *sig = tsk->signal;
 536         struct taskstats *stats;
 537
 538         if (sig->stats || thread_group_empty(tsk))
 539                 goto ret;
 540
 541         /* No problem if kmem_cache_zalloc() fails */
 542         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 543
 544         spin_lock_irq(&tsk->sighand->siglock);
 545         if (!sig->stats) {
 546                 sig->stats = stats;
 547                 stats = NULL;
 548         }
 549         spin_unlock_irq(&tsk->sighand->siglock);
 550
 551         if (stats)
 552                 kmem_cache_free(taskstats_cache, stats);
 553 ret:
 554         return sig->stats;
 555 }
 556
 557 /* Send pid data out on exit */
 558 void taskstats_exit(struct task_struct *tsk, int group_dead)
 559 {
 560         int rc;
 561         struct listener_list *listeners;
 562         struct taskstats *stats;
 563         struct sk_buff *rep_skb;
 564         size_t size;
 565         int is_thread_group;
 566
 567         if (!family_registered)
 568                 return;
 569
 570         /*
 571          * Size includes space for nested attributes
 572          */
 573         size = nla_total_size(sizeof(u32)) +
 574                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 575
 576         is_thread_group = !!taskstats_tgid_alloc(tsk);
 577         if (is_thread_group) {
 578                 /* PID + STATS + TGID + STATS */
 579                 size = 2 * size;
 580                 /* fill the tsk->signal->stats structure */
 581                 fill_tgid_exit(tsk);
 582         }
 583
 584         listeners = &__raw_get_cpu_var(listener_array);
 585         if (list_empty(&listeners->list))
 586                 return;
 587
 588         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 589         if (rc < 0)
 590                 return;
 591
 592         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
 593         if (!stats)
 594                 goto err;
 595
 596         fill_stats(tsk, stats);
 597
 598         /*
 599          * Doesn't matter if tsk is the leader or the last group member leaving
 600          */
 601         if (!is_thread_group || !group_dead)
 602                 goto send;
 603
 604         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
 605         if (!stats)
 606                 goto err;
 607
 608         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 609
 610 send:
 611         send_cpu_listeners(rep_skb, listeners);
 612         return;
 613 err:
 614         nlmsg_free(rep_skb);
 615 }
 616
 617 static struct genl_ops taskstats_ops = {
 618         .cmd            = TASKSTATS_CMD_GET,
 619         .doit           = taskstats_user_cmd,
 620         .policy         = taskstats_cmd_get_policy,
 621 };
 622
 623 static struct genl_ops cgroupstats_ops = {
 624         .cmd            = CGROUPSTATS_CMD_GET,
 625         .doit           = cgroupstats_user_cmd,
 626         .policy         = cgroupstats_cmd_get_policy,
 627 };
 628
 629 /* Needed early in initialization */
 630 void __init taskstats_init_early(void)
 631 {
 632         unsigned int i;
 633
 634         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 635         for_each_possible_cpu(i) {
 636                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 637                 init_rwsem(&(per_cpu(listener_array, i).sem));
 638         }
 639 }
 640
 641 static int __init taskstats_init(void)
 642 {
 643         int rc;
 644
 645         rc = genl_register_family(&family);
 646         if (rc)
 647                 return rc;
 648
 649         rc = genl_register_ops(&family, &taskstats_ops);
 650         if (rc < 0)
 651                 goto err;
 652
 653         rc = genl_register_ops(&family, &cgroupstats_ops);
 654         if (rc < 0)
 655                 goto err_cgroup_ops;
 656
 657         family_registered = 1;
 658         printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
 659         return 0;
 660 err_cgroup_ops:
 661         genl_unregister_ops(&family, &taskstats_ops);
 662 err:
 663         genl_unregister_family(&family);
 664         return rc;
 665 }
 666
 667 /*
 668  * late initcall ensures initialization of statistics collection
 669  * mechanisms precedes initialization of the taskstats interface
 670  */
 671 late_initcall(taskstats_init);