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) = { 0 };
  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_pid(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_pid(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, cpumask_t *maskp, int isadd)
 294 {
 295         struct listener_list *listeners;
 296         struct listener *s, *tmp;
 297         unsigned int cpu;
 298         cpumask_t mask = *maskp;
 299
 300         if (!cpus_subset(mask, cpu_possible_map))
 301                 return -EINVAL;
 302
 303         if (isadd == REGISTER) {
 304                 for_each_cpu_mask(cpu, mask) {
 305                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 306                                          cpu_to_node(cpu));
 307                         if (!s)
 308                                 goto cleanup;
 309                         s->pid = pid;
 310                         INIT_LIST_HEAD(&s->list);
 311                         s->valid = 1;
 312
 313                         listeners = &per_cpu(listener_array, cpu);
 314                         down_write(&listeners->sem);
 315                         list_add(&s->list, &listeners->list);
 316                         up_write(&listeners->sem);
 317                 }
 318                 return 0;
 319         }
 320
 321         /* Deregister or cleanup */
 322 cleanup:
 323         for_each_cpu_mask(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 0;
 336 }
 337
 338 static int parse(struct nlattr *na, cpumask_t *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         if (nla_put(skb, type, sizeof(pid), &pid) < 0)
 373                 goto err;
 374         ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
 375         if (!ret)
 376                 goto err;
 377         nla_nest_end(skb, na);
 378
 379         return nla_data(ret);
 380 err:
 381         return NULL;
 382 }
 383
 384 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 385 {
 386         int rc = 0;
 387         struct sk_buff *rep_skb;
 388         struct cgroupstats *stats;
 389         struct nlattr *na;
 390         size_t size;
 391         u32 fd;
 392         struct file *file;
 393         int fput_needed;
 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         file = fget_light(fd, &fput_needed);
 401         if (!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         stats = nla_data(na);
 414         memset(stats, 0, sizeof(*stats));
 415
 416         rc = cgroupstats_build(stats, file->f_dentry);
 417         if (rc < 0) {
 418                 nlmsg_free(rep_skb);
 419                 goto err;
 420         }
 421
 422         rc = send_reply(rep_skb, info->snd_pid);
 423
 424 err:
 425         fput_light(file, fput_needed);
 426         return rc;
 427 }
 428
 429 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 430 {
 431         int rc = 0;
 432         struct sk_buff *rep_skb;
 433         struct taskstats *stats;
 434         size_t size;
 435         cpumask_t mask;
 436
 437         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 438         if (rc < 0)
 439                 return rc;
 440         if (rc == 0)
 441                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 442
 443         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 444         if (rc < 0)
 445                 return rc;
 446         if (rc == 0)
 447                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 448
 449         /*
 450          * Size includes space for nested attributes
 451          */
 452         size = nla_total_size(sizeof(u32)) +
 453                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 454
 455         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 456         if (rc < 0)
 457                 return rc;
 458
 459         rc = -EINVAL;
 460         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 461                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 462                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 463                 if (!stats)
 464                         goto err;
 465
 466                 rc = fill_pid(pid, NULL, stats);
 467                 if (rc < 0)
 468                         goto err;
 469         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 470                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 471                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 472                 if (!stats)
 473                         goto err;
 474
 475                 rc = fill_tgid(tgid, NULL, stats);
 476                 if (rc < 0)
 477                         goto err;
 478         } else
 479                 goto err;
 480
 481         return send_reply(rep_skb, info->snd_pid);
 482 err:
 483         nlmsg_free(rep_skb);
 484         return rc;
 485 }
 486
 487 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 488 {
 489         struct signal_struct *sig = tsk->signal;
 490         struct taskstats *stats;
 491
 492         if (sig->stats || thread_group_empty(tsk))
 493                 goto ret;
 494
 495         /* No problem if kmem_cache_zalloc() fails */
 496         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 497
 498         spin_lock_irq(&tsk->sighand->siglock);
 499         if (!sig->stats) {
 500                 sig->stats = stats;
 501                 stats = NULL;
 502         }
 503         spin_unlock_irq(&tsk->sighand->siglock);
 504
 505         if (stats)
 506                 kmem_cache_free(taskstats_cache, stats);
 507 ret:
 508         return sig->stats;
 509 }
 510
 511 /* Send pid data out on exit */
 512 void taskstats_exit(struct task_struct *tsk, int group_dead)
 513 {
 514         int rc;
 515         struct listener_list *listeners;
 516         struct taskstats *stats;
 517         struct sk_buff *rep_skb;
 518         size_t size;
 519         int is_thread_group;
 520
 521         if (!family_registered)
 522                 return;
 523
 524         /*
 525          * Size includes space for nested attributes
 526          */
 527         size = nla_total_size(sizeof(u32)) +
 528                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 529
 530         is_thread_group = !!taskstats_tgid_alloc(tsk);
 531         if (is_thread_group) {
 532                 /* PID + STATS + TGID + STATS */
 533                 size = 2 * size;
 534                 /* fill the tsk->signal->stats structure */
 535                 fill_tgid_exit(tsk);
 536         }
 537
 538         listeners = &__raw_get_cpu_var(listener_array);
 539         if (list_empty(&listeners->list))
 540                 return;
 541
 542         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 543         if (rc < 0)
 544                 return;
 545
 546         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
 547         if (!stats)
 548                 goto err;
 549
 550         rc = fill_pid(tsk->pid, tsk, stats);
 551         if (rc < 0)
 552                 goto err;
 553
 554         /*
 555          * Doesn't matter if tsk is the leader or the last group member leaving
 556          */
 557         if (!is_thread_group || !group_dead)
 558                 goto send;
 559
 560         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
 561         if (!stats)
 562                 goto err;
 563
 564         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 565
 566 send:
 567         send_cpu_listeners(rep_skb, listeners);
 568         return;
 569 err:
 570         nlmsg_free(rep_skb);
 571 }
 572
 573 static struct genl_ops taskstats_ops = {
 574         .cmd            = TASKSTATS_CMD_GET,
 575         .doit           = taskstats_user_cmd,
 576         .policy         = taskstats_cmd_get_policy,
 577 };
 578
 579 static struct genl_ops cgroupstats_ops = {
 580         .cmd            = CGROUPSTATS_CMD_GET,
 581         .doit           = cgroupstats_user_cmd,
 582         .policy         = cgroupstats_cmd_get_policy,
 583 };
 584
 585 /* Needed early in initialization */
 586 void __init taskstats_init_early(void)
 587 {
 588         unsigned int i;
 589
 590         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 591         for_each_possible_cpu(i) {
 592                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 593                 init_rwsem(&(per_cpu(listener_array, i).sem));
 594         }
 595 }
 596
 597 static int __init taskstats_init(void)
 598 {
 599         int rc;
 600
 601         rc = genl_register_family(&family);
 602         if (rc)
 603                 return rc;
 604
 605         rc = genl_register_ops(&family, &taskstats_ops);
 606         if (rc < 0)
 607                 goto err;
 608
 609         rc = genl_register_ops(&family, &cgroupstats_ops);
 610         if (rc < 0)
 611                 goto err_cgroup_ops;
 612
 613         family_registered = 1;
 614         printk("registered taskstats version %d\n", TASKSTATS_GENL_VERSION);
 615         return 0;
 616 err_cgroup_ops:
 617         genl_unregister_ops(&family, &taskstats_ops);
 618 err:
 619         genl_unregister_family(&family);
 620         return rc;
 621 }
 622
 623 /*
 624  * late initcall ensures initialization of statistics collection
 625  * mechanisms precedes initialization of the taskstats interface
 626  */
 627 late_initcall(taskstats_init);