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
  46 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = {
  47         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  48         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  49         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  50         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  51
  52 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = {
  53         [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 },
  54 };
  55
  56 struct listener {
  57         struct list_head list;
  58         pid_t pid;
  59         char valid;
  60 };
  61
  62 struct listener_list {
  63         struct rw_semaphore sem;
  64         struct list_head list;
  65 };
  66 static DEFINE_PER_CPU(struct listener_list, listener_array);
  67
  68 enum actions {
  69         REGISTER,
  70         DEREGISTER,
  71         CPU_DONT_CARE
  72 };
  73
  74 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  75                                 size_t size)
  76 {
  77         struct sk_buff *skb;
  78         void *reply;
  79
  80         /*
  81          * If new attributes are added, please revisit this allocation
  82          */
  83         skb = genlmsg_new(size, GFP_KERNEL);
  84         if (!skb)
  85                 return -ENOMEM;
  86
  87         if (!info) {
  88                 int seq = this_cpu_inc_return(taskstats_seqnum) - 1;
  89
  90                 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
  91         } else
  92                 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
  93         if (reply == NULL) {
  94                 nlmsg_free(skb);
  95                 return -EINVAL;
  96         }
  97
  98         *skbp = skb;
  99         return 0;
 100 }
 101
 102 /*
 103  * Send taskstats data in @skb to listener with nl_pid @pid
 104  */
 105 static int send_reply(struct sk_buff *skb, struct genl_info *info)
 106 {
 107         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 108         void *reply = genlmsg_data(genlhdr);
 109
 110         genlmsg_end(skb, reply);
 111
 112         return genlmsg_reply(skb, info);
 113 }
 114
 115 /*
 116  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 117  */
 118 static void send_cpu_listeners(struct sk_buff *skb,
 119                                         struct listener_list *listeners)
 120 {
 121         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 122         struct listener *s, *tmp;
 123         struct sk_buff *skb_next, *skb_cur = skb;
 124         void *reply = genlmsg_data(genlhdr);
 125         int rc, delcount = 0;
 126
 127         genlmsg_end(skb, reply);
 128
 129         rc = 0;
 130         down_read(&listeners->sem);
 131         list_for_each_entry(s, &listeners->list, list) {
 132                 skb_next = NULL;
 133                 if (!list_is_last(&s->list, &listeners->list)) {
 134                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 135                         if (!skb_next)
 136                                 break;
 137                 }
 138                 rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
 139                 if (rc == -ECONNREFUSED) {
 140                         s->valid = 0;
 141                         delcount++;
 142                 }
 143                 skb_cur = skb_next;
 144         }
 145         up_read(&listeners->sem);
 146
 147         if (skb_cur)
 148                 nlmsg_free(skb_cur);
 149
 150         if (!delcount)
 151                 return;
 152
 153         /* Delete invalidated entries */
 154         down_write(&listeners->sem);
 155         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 156                 if (!s->valid) {
 157                         list_del(&s->list);
 158                         kfree(s);
 159                 }
 160         }
 161         up_write(&listeners->sem);
 162 }
 163
 164 static void fill_stats(struct user_namespace *user_ns,
 165                        struct pid_namespace *pid_ns,
 166                        struct task_struct *tsk, struct taskstats *stats)
 167 {
 168         memset(stats, 0, sizeof(*stats));
 169         /*
 170          * Each accounting subsystem adds calls to its functions to
 171          * fill in relevant parts of struct taskstsats as follows
 172          *
 173          *      per-task-foo(stats, tsk);
 174          */
 175
 176         delayacct_add_tsk(stats, tsk);
 177
 178         /* fill in basic acct fields */
 179         stats->version = TASKSTATS_VERSION;
 180         stats->nvcsw = tsk->nvcsw;
 181         stats->nivcsw = tsk->nivcsw;
 182         bacct_add_tsk(user_ns, pid_ns, stats, tsk);
 183
 184         /* fill in extended acct fields */
 185         xacct_add_tsk(stats, tsk);
 186 }
 187
 188 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
 189 {
 190         struct task_struct *tsk;
 191
 192         rcu_read_lock();
 193         tsk = find_task_by_vpid(pid);
 194         if (tsk)
 195                 get_task_struct(tsk);
 196         rcu_read_unlock();
 197         if (!tsk)
 198                 return -ESRCH;
 199         fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
 200         put_task_struct(tsk);
 201         return 0;
 202 }
 203
 204 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats)
 205 {
 206         struct task_struct *tsk, *first;
 207         unsigned long flags;
 208         int rc = -ESRCH;
 209
 210         /*
 211          * Add additional stats from live tasks except zombie thread group
 212          * leaders who are already counted with the dead tasks
 213          */
 214         rcu_read_lock();
 215         first = find_task_by_vpid(tgid);
 216
 217         if (!first || !lock_task_sighand(first, &flags))
 218                 goto out;
 219
 220         if (first->signal->stats)
 221                 memcpy(stats, first->signal->stats, sizeof(*stats));
 222         else
 223                 memset(stats, 0, sizeof(*stats));
 224
 225         tsk = first;
 226         do {
 227                 if (tsk->exit_state)
 228                         continue;
 229                 /*
 230                  * Accounting subsystem can call its functions here to
 231                  * fill in relevant parts of struct taskstsats as follows
 232                  *
 233                  *      per-task-foo(stats, tsk);
 234                  */
 235                 delayacct_add_tsk(stats, tsk);
 236
 237                 stats->nvcsw += tsk->nvcsw;
 238                 stats->nivcsw += tsk->nivcsw;
 239         } while_each_thread(first, tsk);
 240
 241         unlock_task_sighand(first, &flags);
 242         rc = 0;
 243 out:
 244         rcu_read_unlock();
 245
 246         stats->version = TASKSTATS_VERSION;
 247         /*
 248          * Accounting subsystems can also add calls here to modify
 249          * fields of taskstats.
 250          */
 251         return rc;
 252 }
 253
 254 static void fill_tgid_exit(struct task_struct *tsk)
 255 {
 256         unsigned long flags;
 257
 258         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 259         if (!tsk->signal->stats)
 260                 goto ret;
 261
 262         /*
 263          * Each accounting subsystem calls its functions here to
 264          * accumalate its per-task stats for tsk, into the per-tgid structure
 265          *
 266          *      per-task-foo(tsk->signal->stats, tsk);
 267          */
 268         delayacct_add_tsk(tsk->signal->stats, tsk);
 269 ret:
 270         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 271         return;
 272 }
 273
 274 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
 275 {
 276         struct listener_list *listeners;
 277         struct listener *s, *tmp, *s2;
 278         unsigned int cpu;
 279         int ret = 0;
 280
 281         if (!cpumask_subset(mask, cpu_possible_mask))
 282                 return -EINVAL;
 283
 284         if (current_user_ns() != &init_user_ns)
 285                 return -EINVAL;
 286
 287         if (task_active_pid_ns(current) != &init_pid_ns)
 288                 return -EINVAL;
 289
 290         if (isadd == REGISTER) {
 291                 for_each_cpu(cpu, mask) {
 292                         s = kmalloc_node(sizeof(struct listener),
 293                                         GFP_KERNEL, cpu_to_node(cpu));
 294                         if (!s) {
 295                                 ret = -ENOMEM;
 296                                 goto cleanup;
 297                         }
 298                         s->pid = pid;
 299                         s->valid = 1;
 300
 301                         listeners = &per_cpu(listener_array, cpu);
 302                         down_write(&listeners->sem);
 303                         list_for_each_entry(s2, &listeners->list, list) {
 304                                 if (s2->pid == pid && s2->valid)
 305                                         goto exists;
 306                         }
 307                         list_add(&s->list, &listeners->list);
 308                         s = NULL;
 309 exists:
 310                         up_write(&listeners->sem);
 311                         kfree(s); /* nop if NULL */
 312                 }
 313                 return 0;
 314         }
 315
 316         /* Deregister or cleanup */
 317 cleanup:
 318         for_each_cpu(cpu, mask) {
 319                 listeners = &per_cpu(listener_array, cpu);
 320                 down_write(&listeners->sem);
 321                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 322                         if (s->pid == pid) {
 323                                 list_del(&s->list);
 324                                 kfree(s);
 325                                 break;
 326                         }
 327                 }
 328                 up_write(&listeners->sem);
 329         }
 330         return ret;
 331 }
 332
 333 static int parse(struct nlattr *na, struct cpumask *mask)
 334 {
 335         char *data;
 336         int len;
 337         int ret;
 338
 339         if (na == NULL)
 340                 return 1;
 341         len = nla_len(na);
 342         if (len > TASKSTATS_CPUMASK_MAXLEN)
 343                 return -E2BIG;
 344         if (len < 1)
 345                 return -EINVAL;
 346         data = kmalloc(len, GFP_KERNEL);
 347         if (!data)
 348                 return -ENOMEM;
 349         nla_strlcpy(data, na, len);
 350         ret = cpulist_parse(data, mask);
 351         kfree(data);
 352         return ret;
 353 }
 354
 355 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
 356 {
 357         struct nlattr *na, *ret;
 358         int aggr;
 359
 360         aggr = (type == TASKSTATS_TYPE_PID)
 361                         ? TASKSTATS_TYPE_AGGR_PID
 362                         : TASKSTATS_TYPE_AGGR_TGID;
 363
 364         na = nla_nest_start(skb, aggr);
 365         if (!na)
 366                 goto err;
 367
 368         if (nla_put(skb, type, sizeof(pid), &pid) < 0) {
 369                 nla_nest_cancel(skb, na);
 370                 goto err;
 371         }
 372         ret = nla_reserve_64bit(skb, TASKSTATS_TYPE_STATS,
 373                                 sizeof(struct taskstats), TASKSTATS_TYPE_NULL);
 374         if (!ret) {
 375                 nla_nest_cancel(skb, na);
 376                 goto err;
 377         }
 378         nla_nest_end(skb, na);
 379
 380         return nla_data(ret);
 381 err:
 382         return NULL;
 383 }
 384
 385 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 386 {
 387         int rc = 0;
 388         struct sk_buff *rep_skb;
 389         struct cgroupstats *stats;
 390         struct nlattr *na;
 391         size_t size;
 392         u32 fd;
 393         struct fd f;
 394
 395         na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
 396         if (!na)
 397                 return -EINVAL;
 398
 399         fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
 400         f = fdget(fd);
 401         if (!f.file)
 402                 return 0;
 403
 404         size = nla_total_size(sizeof(struct cgroupstats));
 405
 406         rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
 407                                 size);
 408         if (rc < 0)
 409                 goto err;
 410
 411         na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
 412                                 sizeof(struct cgroupstats));
 413         if (na == NULL) {
 414                 nlmsg_free(rep_skb);
 415                 rc = -EMSGSIZE;
 416                 goto err;
 417         }
 418
 419         stats = nla_data(na);
 420         memset(stats, 0, sizeof(*stats));
 421
 422         rc = cgroupstats_build(stats, f.file->f_path.dentry);
 423         if (rc < 0) {
 424                 nlmsg_free(rep_skb);
 425                 goto err;
 426         }
 427
 428         rc = send_reply(rep_skb, info);
 429
 430 err:
 431         fdput(f);
 432         return rc;
 433 }
 434
 435 static int cmd_attr_register_cpumask(struct genl_info *info)
 436 {
 437         cpumask_var_t mask;
 438         int rc;
 439
 440         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 441                 return -ENOMEM;
 442         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
 443         if (rc < 0)
 444                 goto out;
 445         rc = add_del_listener(info->snd_portid, mask, REGISTER);
 446 out:
 447         free_cpumask_var(mask);
 448         return rc;
 449 }
 450
 451 static int cmd_attr_deregister_cpumask(struct genl_info *info)
 452 {
 453         cpumask_var_t mask;
 454         int rc;
 455
 456         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 457                 return -ENOMEM;
 458         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
 459         if (rc < 0)
 460                 goto out;
 461         rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
 462 out:
 463         free_cpumask_var(mask);
 464         return rc;
 465 }
 466
 467 static size_t taskstats_packet_size(void)
 468 {
 469         size_t size;
 470
 471         size = nla_total_size(sizeof(u32)) +
 472                 nla_total_size_64bit(sizeof(struct taskstats)) +
 473                 nla_total_size(0);
 474
 475         return size;
 476 }
 477
 478 static int cmd_attr_pid(struct genl_info *info)
 479 {
 480         struct taskstats *stats;
 481         struct sk_buff *rep_skb;
 482         size_t size;
 483         u32 pid;
 484         int rc;
 485
 486         size = taskstats_packet_size();
 487
 488         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 489         if (rc < 0)
 490                 return rc;
 491
 492         rc = -EINVAL;
 493         pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 494         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 495         if (!stats)
 496                 goto err;
 497
 498         rc = fill_stats_for_pid(pid, stats);
 499         if (rc < 0)
 500                 goto err;
 501         return send_reply(rep_skb, info);
 502 err:
 503         nlmsg_free(rep_skb);
 504         return rc;
 505 }
 506
 507 static int cmd_attr_tgid(struct genl_info *info)
 508 {
 509         struct taskstats *stats;
 510         struct sk_buff *rep_skb;
 511         size_t size;
 512         u32 tgid;
 513         int rc;
 514
 515         size = taskstats_packet_size();
 516
 517         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 518         if (rc < 0)
 519                 return rc;
 520
 521         rc = -EINVAL;
 522         tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 523         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 524         if (!stats)
 525                 goto err;
 526
 527         rc = fill_stats_for_tgid(tgid, stats);
 528         if (rc < 0)
 529                 goto err;
 530         return send_reply(rep_skb, info);
 531 err:
 532         nlmsg_free(rep_skb);
 533         return rc;
 534 }
 535
 536 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 537 {
 538         if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK])
 539                 return cmd_attr_register_cpumask(info);
 540         else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK])
 541                 return cmd_attr_deregister_cpumask(info);
 542         else if (info->attrs[TASKSTATS_CMD_ATTR_PID])
 543                 return cmd_attr_pid(info);
 544         else if (info->attrs[TASKSTATS_CMD_ATTR_TGID])
 545                 return cmd_attr_tgid(info);
 546         else
 547                 return -EINVAL;
 548 }
 549
 550 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 551 {
 552         struct signal_struct *sig = tsk->signal;
 553         struct taskstats *stats;
 554
 555         if (sig->stats || thread_group_empty(tsk))
 556                 goto ret;
 557
 558         /* No problem if kmem_cache_zalloc() fails */
 559         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 560
 561         spin_lock_irq(&tsk->sighand->siglock);
 562         if (!sig->stats) {
 563                 sig->stats = stats;
 564                 stats = NULL;
 565         }
 566         spin_unlock_irq(&tsk->sighand->siglock);
 567
 568         if (stats)
 569                 kmem_cache_free(taskstats_cache, stats);
 570 ret:
 571         return sig->stats;
 572 }
 573
 574 /* Send pid data out on exit */
 575 void taskstats_exit(struct task_struct *tsk, int group_dead)
 576 {
 577         int rc;
 578         struct listener_list *listeners;
 579         struct taskstats *stats;
 580         struct sk_buff *rep_skb;
 581         size_t size;
 582         int is_thread_group;
 583
 584         if (!family_registered)
 585                 return;
 586
 587         /*
 588          * Size includes space for nested attributes
 589          */
 590         size = taskstats_packet_size();
 591
 592         is_thread_group = !!taskstats_tgid_alloc(tsk);
 593         if (is_thread_group) {
 594                 /* PID + STATS + TGID + STATS */
 595                 size = 2 * size;
 596                 /* fill the tsk->signal->stats structure */
 597                 fill_tgid_exit(tsk);
 598         }
 599
 600         listeners = raw_cpu_ptr(&listener_array);
 601         if (list_empty(&listeners->list))
 602                 return;
 603
 604         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 605         if (rc < 0)
 606                 return;
 607
 608         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
 609                          task_pid_nr_ns(tsk, &init_pid_ns));
 610         if (!stats)
 611                 goto err;
 612
 613         fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
 614
 615         /*
 616          * Doesn't matter if tsk is the leader or the last group member leaving
 617          */
 618         if (!is_thread_group || !group_dead)
 619                 goto send;
 620
 621         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
 622                          task_tgid_nr_ns(tsk, &init_pid_ns));
 623         if (!stats)
 624                 goto err;
 625
 626         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 627
 628 send:
 629         send_cpu_listeners(rep_skb, listeners);
 630         return;
 631 err:
 632         nlmsg_free(rep_skb);
 633 }
 634
 635 static const struct genl_ops taskstats_ops[] = {
 636         {
 637                 .cmd            = TASKSTATS_CMD_GET,
 638                 .doit           = taskstats_user_cmd,
 639                 .policy         = taskstats_cmd_get_policy,
 640                 .flags          = GENL_ADMIN_PERM,
 641         },
 642         {
 643                 .cmd            = CGROUPSTATS_CMD_GET,
 644                 .doit           = cgroupstats_user_cmd,
 645                 .policy         = cgroupstats_cmd_get_policy,
 646         },
 647 };
 648
 649 static struct genl_family family __ro_after_init = {
 650         .name           = TASKSTATS_GENL_NAME,
 651         .version        = TASKSTATS_GENL_VERSION,
 652         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
 653         .module         = THIS_MODULE,
 654         .ops            = taskstats_ops,
 655         .n_ops          = ARRAY_SIZE(taskstats_ops),
 656 };
 657
 658 /* Needed early in initialization */
 659 void __init taskstats_init_early(void)
 660 {
 661         unsigned int i;
 662
 663         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 664         for_each_possible_cpu(i) {
 665                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 666                 init_rwsem(&(per_cpu(listener_array, i).sem));
 667         }
 668 }
 669
 670 static int __init taskstats_init(void)
 671 {
 672         int rc;
 673
 674         rc = genl_register_family(&family);
 675         if (rc)
 676                 return rc;
 677
 678         family_registered = 1;
 679         pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
 680         return 0;
 681 }
 682
 683 /*
 684  * late initcall ensures initialization of statistics collection
 685  * mechanisms precedes initialization of the taskstats interface
 686  */
 687 late_initcall(taskstats_init);