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/cgroupstats.h>
  26 #include <linux/cgroup.h>
  27 #include <linux/fs.h>
  28 #include <linux/file.h>
  29 #include <net/genetlink.h>
  30 #include <asm/atomic.h>
  31
  32 /*
  33  * Maximum length of a cpumask that can be specified in
  34  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  35  */
  36 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  37
  38 static DEFINE_PER_CPU(__u32, taskstats_seqnum);
  39 static int family_registered;
  40 struct kmem_cache *taskstats_cache;
  41
  42 static struct genl_family family = {
  43         .id             = GENL_ID_GENERATE,
  44         .name           = TASKSTATS_GENL_NAME,
  45         .version        = TASKSTATS_GENL_VERSION,
  46         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  47 };
  48
  49 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
  50 __read_mostly = {
  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 struct nla_policy
  57 cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] __read_mostly = {
  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 = get_cpu_var(taskstats_seqnum)++;
  94                 put_cpu_var(taskstats_seqnum);
  95
  96                 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
  97         } else
  98                 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
  99         if (reply == NULL) {
 100                 nlmsg_free(skb);
 101                 return -EINVAL;
 102         }
 103
 104         *skbp = skb;
 105         return 0;
 106 }
 107
 108 /*
 109  * Send taskstats data in @skb to listener with nl_pid @pid
 110  */
 111 static int send_reply(struct sk_buff *skb, pid_t pid)
 112 {
 113         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 114         void *reply = genlmsg_data(genlhdr);
 115         int rc;
 116
 117         rc = genlmsg_end(skb, reply);
 118         if (rc < 0) {
 119                 nlmsg_free(skb);
 120                 return rc;
 121         }
 122
 123         return genlmsg_unicast(skb, pid);
 124 }
 125
 126 /*
 127  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 128  */
 129 static void send_cpu_listeners(struct sk_buff *skb,
 130                                         struct listener_list *listeners)
 131 {
 132         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 133         struct listener *s, *tmp;
 134         struct sk_buff *skb_next, *skb_cur = skb;
 135         void *reply = genlmsg_data(genlhdr);
 136         int rc, delcount = 0;
 137
 138         rc = genlmsg_end(skb, reply);
 139         if (rc < 0) {
 140                 nlmsg_free(skb);
 141                 return;
 142         }
 143
 144         rc = 0;
 145         down_read(&listeners->sem);
 146         list_for_each_entry(s, &listeners->list, list) {
 147                 skb_next = NULL;
 148                 if (!list_is_last(&s->list, &listeners->list)) {
 149                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 150                         if (!skb_next)
 151                                 break;
 152                 }
 153                 rc = genlmsg_unicast(skb_cur, s->pid);
 154                 if (rc == -ECONNREFUSED) {
 155                         s->valid = 0;
 156                         delcount++;
 157                 }
 158                 skb_cur = skb_next;
 159         }
 160         up_read(&listeners->sem);
 161
 162         if (skb_cur)
 163                 nlmsg_free(skb_cur);
 164
 165         if (!delcount)
 166                 return;
 167
 168         /* Delete invalidated entries */
 169         down_write(&listeners->sem);
 170         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 171                 if (!s->valid) {
 172                         list_del(&s->list);
 173                         kfree(s);
 174                 }
 175         }
 176         up_write(&listeners->sem);
 177 }
 178
 179 static int fill_pid(pid_t pid, struct task_struct *tsk,
 180                 struct taskstats *stats)
 181 {
 182         int rc = 0;
 183
 184         if (!tsk) {
 185                 rcu_read_lock();
 186                 tsk = find_task_by_vpid(pid);
 187                 if (tsk)
 188                         get_task_struct(tsk);
 189                 rcu_read_unlock();
 190                 if (!tsk)
 191                         return -ESRCH;
 192         } else
 193                 get_task_struct(tsk);
 194
 195         memset(stats, 0, sizeof(*stats));
 196         /*
 197          * Each accounting subsystem adds calls to its functions to
 198          * fill in relevant parts of struct taskstsats as follows
 199          *
 200          *      per-task-foo(stats, tsk);
 201          */
 202
 203         delayacct_add_tsk(stats, tsk);
 204
 205         /* fill in basic acct fields */
 206         stats->version = TASKSTATS_VERSION;
 207         stats->nvcsw = tsk->nvcsw;
 208         stats->nivcsw = tsk->nivcsw;
 209         bacct_add_tsk(stats, tsk);
 210
 211         /* fill in extended acct fields */
 212         xacct_add_tsk(stats, tsk);
 213
 214         /* Define err: label here if needed */
 215         put_task_struct(tsk);
 216         return rc;
 217
 218 }
 219
 220 static int fill_tgid(pid_t tgid, struct task_struct *first,
 221                 struct taskstats *stats)
 222 {
 223         struct task_struct *tsk;
 224         unsigned long flags;
 225         int rc = -ESRCH;
 226
 227         /*
 228          * Add additional stats from live tasks except zombie thread group
 229          * leaders who are already counted with the dead tasks
 230          */
 231         rcu_read_lock();
 232         if (!first)
 233                 first = find_task_by_vpid(tgid);
 234
 235         if (!first || !lock_task_sighand(first, &flags))
 236                 goto out;
 237
 238         if (first->signal->stats)
 239                 memcpy(stats, first->signal->stats, sizeof(*stats));
 240         else
 241                 memset(stats, 0, sizeof(*stats));
 242
 243         tsk = first;
 244         do {
 245                 if (tsk->exit_state)
 246                         continue;
 247                 /*
 248                  * Accounting subsystem can call its functions here to
 249                  * fill in relevant parts of struct taskstsats as follows
 250                  *
 251                  *      per-task-foo(stats, tsk);
 252                  */
 253                 delayacct_add_tsk(stats, tsk);
 254
 255                 stats->nvcsw += tsk->nvcsw;
 256                 stats->nivcsw += tsk->nivcsw;
 257         } while_each_thread(first, tsk);
 258
 259         unlock_task_sighand(first, &flags);
 260         rc = 0;
 261 out:
 262         rcu_read_unlock();
 263
 264         stats->version = TASKSTATS_VERSION;
 265         /*
 266          * Accounting subsystems can also add calls here to modify
 267          * fields of taskstats.
 268          */
 269         return rc;
 270 }
 271
 272
 273 static void fill_tgid_exit(struct task_struct *tsk)
 274 {
 275         unsigned long flags;
 276
 277         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 278         if (!tsk->signal->stats)
 279                 goto ret;
 280
 281         /*
 282          * Each accounting subsystem calls its functions here to
 283          * accumalate its per-task stats for tsk, into the per-tgid structure
 284          *
 285          *      per-task-foo(tsk->signal->stats, tsk);
 286          */
 287         delayacct_add_tsk(tsk->signal->stats, tsk);
 288 ret:
 289         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 290         return;
 291 }
 292
 293 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
 294 {
 295         struct listener_list *listeners;
 296         struct listener *s, *tmp;
 297         unsigned int cpu;
 298
 299         if (!cpumask_subset(mask, cpu_possible_mask))
 300                 return -EINVAL;
 301
 302         if (isadd == REGISTER) {
 303                 for_each_cpu(cpu, mask) {
 304                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 305                                          cpu_to_node(cpu));
 306                         if (!s)
 307                                 goto cleanup;
 308                         s->pid = pid;
 309                         INIT_LIST_HEAD(&s->list);
 310                         s->valid = 1;
 311
 312                         listeners = &per_cpu(listener_array, cpu);
 313                         down_write(&listeners->sem);
 314                         list_add(&s->list, &listeners->list);
 315                         up_write(&listeners->sem);
 316                 }
 317                 return 0;
 318         }
 319
 320         /* Deregister or cleanup */
 321 cleanup:
 322         for_each_cpu(cpu, mask) {
 323                 listeners = &per_cpu(listener_array, cpu);
 324                 down_write(&listeners->sem);
 325                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 326                         if (s->pid == pid) {
 327                                 list_del(&s->list);
 328                                 kfree(s);
 329                                 break;
 330                         }
 331                 }
 332                 up_write(&listeners->sem);
 333         }
 334         return 0;
 335 }
 336
 337 static int parse(struct nlattr *na, struct cpumask *mask)
 338 {
 339         char *data;
 340         int len;
 341         int ret;
 342
 343         if (na == NULL)
 344                 return 1;
 345         len = nla_len(na);
 346         if (len > TASKSTATS_CPUMASK_MAXLEN)
 347                 return -E2BIG;
 348         if (len < 1)
 349                 return -EINVAL;
 350         data = kmalloc(len, GFP_KERNEL);
 351         if (!data)
 352                 return -ENOMEM;
 353         nla_strlcpy(data, na, len);
 354         ret = cpulist_parse(data, mask);
 355         kfree(data);
 356         return ret;
 357 }
 358
 359 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
 360 {
 361         struct nlattr *na, *ret;
 362         int aggr;
 363
 364         aggr = (type == TASKSTATS_TYPE_PID)
 365                         ? TASKSTATS_TYPE_AGGR_PID
 366                         : TASKSTATS_TYPE_AGGR_TGID;
 367
 368         na = nla_nest_start(skb, aggr);
 369         if (!na)
 370                 goto err;
 371         if (nla_put(skb, type, sizeof(pid), &pid) < 0)
 372                 goto err;
 373         ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
 374         if (!ret)
 375                 goto err;
 376         nla_nest_end(skb, na);
 377
 378         return nla_data(ret);
 379 err:
 380         return NULL;
 381 }
 382
 383 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 384 {
 385         int rc = 0;
 386         struct sk_buff *rep_skb;
 387         struct cgroupstats *stats;
 388         struct nlattr *na;
 389         size_t size;
 390         u32 fd;
 391         struct file *file;
 392         int fput_needed;
 393
 394         na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
 395         if (!na)
 396                 return -EINVAL;
 397
 398         fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
 399         file = fget_light(fd, &fput_needed);
 400         if (!file)
 401                 return 0;
 402
 403         size = nla_total_size(sizeof(struct cgroupstats));
 404
 405         rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb,
 406                                 size);
 407         if (rc < 0)
 408                 goto err;
 409
 410         na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
 411                                 sizeof(struct cgroupstats));
 412         stats = nla_data(na);
 413         memset(stats, 0, sizeof(*stats));
 414
 415         rc = cgroupstats_build(stats, file->f_dentry);
 416         if (rc < 0) {
 417                 nlmsg_free(rep_skb);
 418                 goto err;
 419         }
 420
 421         rc = send_reply(rep_skb, info->snd_pid);
 422
 423 err:
 424         fput_light(file, fput_needed);
 425         return rc;
 426 }
 427
 428 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 429 {
 430         int rc;
 431         struct sk_buff *rep_skb;
 432         struct taskstats *stats;
 433         size_t size;
 434         cpumask_var_t mask;
 435
 436         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 437                 return -ENOMEM;
 438
 439         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
 440         if (rc < 0)
 441                 goto free_return_rc;
 442         if (rc == 0) {
 443                 rc = add_del_listener(info->snd_pid, mask, REGISTER);
 444                 goto free_return_rc;
 445         }
 446
 447         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
 448         if (rc < 0)
 449                 goto free_return_rc;
 450         if (rc == 0) {
 451                 rc = add_del_listener(info->snd_pid, mask, DEREGISTER);
 452 free_return_rc:
 453                 free_cpumask_var(mask);
 454                 return rc;
 455         }
 456         free_cpumask_var(mask);
 457
 458         /*
 459          * Size includes space for nested attributes
 460          */
 461         size = nla_total_size(sizeof(u32)) +
 462                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 463
 464         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 465         if (rc < 0)
 466                 return rc;
 467
 468         rc = -EINVAL;
 469         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 470                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 471                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 472                 if (!stats)
 473                         goto err;
 474
 475                 rc = fill_pid(pid, NULL, stats);
 476                 if (rc < 0)
 477                         goto err;
 478         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 479                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 480                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 481                 if (!stats)
 482                         goto err;
 483
 484                 rc = fill_tgid(tgid, NULL, stats);
 485                 if (rc < 0)
 486                         goto err;
 487         } else
 488                 goto err;
 489
 490         return send_reply(rep_skb, info->snd_pid);
 491 err:
 492         nlmsg_free(rep_skb);
 493         return rc;
 494 }
 495
 496 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 497 {
 498         struct signal_struct *sig = tsk->signal;
 499         struct taskstats *stats;
 500
 501         if (sig->stats || thread_group_empty(tsk))
 502                 goto ret;
 503
 504         /* No problem if kmem_cache_zalloc() fails */
 505         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 506
 507         spin_lock_irq(&tsk->sighand->siglock);
 508         if (!sig->stats) {
 509                 sig->stats = stats;
 510                 stats = NULL;
 511         }
 512         spin_unlock_irq(&tsk->sighand->siglock);
 513
 514         if (stats)
 515                 kmem_cache_free(taskstats_cache, stats);
 516 ret:
 517         return sig->stats;
 518 }
 519
 520 /* Send pid data out on exit */
 521 void taskstats_exit(struct task_struct *tsk, int group_dead)
 522 {
 523         int rc;
 524         struct listener_list *listeners;
 525         struct taskstats *stats;
 526         struct sk_buff *rep_skb;
 527         size_t size;
 528         int is_thread_group;
 529
 530         if (!family_registered)
 531                 return;
 532
 533         /*
 534          * Size includes space for nested attributes
 535          */
 536         size = nla_total_size(sizeof(u32)) +
 537                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 538
 539         is_thread_group = !!taskstats_tgid_alloc(tsk);
 540         if (is_thread_group) {
 541                 /* PID + STATS + TGID + STATS */
 542                 size = 2 * size;
 543                 /* fill the tsk->signal->stats structure */
 544                 fill_tgid_exit(tsk);
 545         }
 546
 547         listeners = &__raw_get_cpu_var(listener_array);
 548         if (list_empty(&listeners->list))
 549                 return;
 550
 551         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 552         if (rc < 0)
 553                 return;
 554
 555         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
 556         if (!stats)
 557                 goto err;
 558
 559         rc = fill_pid(-1, tsk, stats);
 560         if (rc < 0)
 561                 goto err;
 562
 563         /*
 564          * Doesn't matter if tsk is the leader or the last group member leaving
 565          */
 566         if (!is_thread_group || !group_dead)
 567                 goto send;
 568
 569         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
 570         if (!stats)
 571                 goto err;
 572
 573         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 574
 575 send:
 576         send_cpu_listeners(rep_skb, listeners);
 577         return;
 578 err:
 579         nlmsg_free(rep_skb);
 580 }
 581
 582 static struct genl_ops taskstats_ops = {
 583         .cmd            = TASKSTATS_CMD_GET,
 584         .doit           = taskstats_user_cmd,
 585         .policy         = taskstats_cmd_get_policy,
 586 };
 587
 588 static struct genl_ops cgroupstats_ops = {
 589         .cmd            = CGROUPSTATS_CMD_GET,
 590         .doit           = cgroupstats_user_cmd,
 591         .policy         = cgroupstats_cmd_get_policy,
 592 };
 593
 594 /* Needed early in initialization */
 595 void __init taskstats_init_early(void)
 596 {
 597         unsigned int i;
 598
 599         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 600         for_each_possible_cpu(i) {
 601                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 602                 init_rwsem(&(per_cpu(listener_array, i).sem));
 603         }
 604 }
 605
 606 static int __init taskstats_init(void)
 607 {
 608         int rc;
 609
 610         rc = genl_register_family(&family);
 611         if (rc)
 612                 return rc;
 613
 614         rc = genl_register_ops(&family, &taskstats_ops);
 615         if (rc < 0)
 616                 goto err;
 617
 618         rc = genl_register_ops(&family, &cgroupstats_ops);
 619         if (rc < 0)
 620                 goto err_cgroup_ops;
 621
 622         family_registered = 1;
 623         printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
 624         return 0;
 625 err_cgroup_ops:
 626         genl_unregister_ops(&family, &taskstats_ops);
 627 err:
 628         genl_unregister_family(&family);
 629         return rc;
 630 }
 631
 632 /*
 633  * late initcall ensures initialization of statistics collection
 634  * mechanisms precedes initialization of the taskstats interface
 635  */
 636 late_initcall(taskstats_init);