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