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 <net/genetlink.h>
  26 #include <asm/atomic.h>
  27
  28 /*
  29  * Maximum length of a cpumask that can be specified in
  30  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  31  */
  32 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  33
  34 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
  35 static int family_registered;
  36 struct kmem_cache *taskstats_cache;
  37
  38 static struct genl_family family = {
  39         .id             = GENL_ID_GENERATE,
  40         .name           = TASKSTATS_GENL_NAME,
  41         .version        = TASKSTATS_GENL_VERSION,
  42         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  43 };
  44
  45 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
  46 __read_mostly = {
  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 struct listener {
  53         struct list_head list;
  54         pid_t pid;
  55         char valid;
  56 };
  57
  58 struct listener_list {
  59         struct rw_semaphore sem;
  60         struct list_head list;
  61 };
  62 static DEFINE_PER_CPU(struct listener_list, listener_array);
  63
  64 enum actions {
  65         REGISTER,
  66         DEREGISTER,
  67         CPU_DONT_CARE
  68 };
  69
  70 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  71                                 size_t size)
  72 {
  73         struct sk_buff *skb;
  74         void *reply;
  75
  76         /*
  77          * If new attributes are added, please revisit this allocation
  78          */
  79         skb = genlmsg_new(size, GFP_KERNEL);
  80         if (!skb)
  81                 return -ENOMEM;
  82
  83         if (!info) {
  84                 int seq = get_cpu_var(taskstats_seqnum)++;
  85                 put_cpu_var(taskstats_seqnum);
  86
  87                 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
  88         } else
  89                 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
  90         if (reply == NULL) {
  91                 nlmsg_free(skb);
  92                 return -EINVAL;
  93         }
  94
  95         *skbp = skb;
  96         return 0;
  97 }
  98
  99 /*
 100  * Send taskstats data in @skb to listener with nl_pid @pid
 101  */
 102 static int send_reply(struct sk_buff *skb, pid_t pid)
 103 {
 104         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 105         void *reply = genlmsg_data(genlhdr);
 106         int rc;
 107
 108         rc = genlmsg_end(skb, reply);
 109         if (rc < 0) {
 110                 nlmsg_free(skb);
 111                 return rc;
 112         }
 113
 114         return genlmsg_unicast(skb, pid);
 115 }
 116
 117 /*
 118  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 119  */
 120 static void send_cpu_listeners(struct sk_buff *skb,
 121                                         struct listener_list *listeners)
 122 {
 123         struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb));
 124         struct listener *s, *tmp;
 125         struct sk_buff *skb_next, *skb_cur = skb;
 126         void *reply = genlmsg_data(genlhdr);
 127         int rc, delcount = 0;
 128
 129         rc = genlmsg_end(skb, reply);
 130         if (rc < 0) {
 131                 nlmsg_free(skb);
 132                 return;
 133         }
 134
 135         rc = 0;
 136         down_read(&listeners->sem);
 137         list_for_each_entry(s, &listeners->list, list) {
 138                 skb_next = NULL;
 139                 if (!list_is_last(&s->list, &listeners->list)) {
 140                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 141                         if (!skb_next)
 142                                 break;
 143                 }
 144                 rc = genlmsg_unicast(skb_cur, s->pid);
 145                 if (rc == -ECONNREFUSED) {
 146                         s->valid = 0;
 147                         delcount++;
 148                 }
 149                 skb_cur = skb_next;
 150         }
 151         up_read(&listeners->sem);
 152
 153         if (skb_cur)
 154                 nlmsg_free(skb_cur);
 155
 156         if (!delcount)
 157                 return;
 158
 159         /* Delete invalidated entries */
 160         down_write(&listeners->sem);
 161         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 162                 if (!s->valid) {
 163                         list_del(&s->list);
 164                         kfree(s);
 165                 }
 166         }
 167         up_write(&listeners->sem);
 168 }
 169
 170 static int fill_pid(pid_t pid, struct task_struct *tsk,
 171                 struct taskstats *stats)
 172 {
 173         int rc = 0;
 174
 175         if (!tsk) {
 176                 rcu_read_lock();
 177                 tsk = find_task_by_pid(pid);
 178                 if (tsk)
 179                         get_task_struct(tsk);
 180                 rcu_read_unlock();
 181                 if (!tsk)
 182                         return -ESRCH;
 183         } else
 184                 get_task_struct(tsk);
 185
 186         memset(stats, 0, sizeof(*stats));
 187         /*
 188          * Each accounting subsystem adds calls to its functions to
 189          * fill in relevant parts of struct taskstsats as follows
 190          *
 191          *      per-task-foo(stats, tsk);
 192          */
 193
 194         delayacct_add_tsk(stats, tsk);
 195
 196         /* fill in basic acct fields */
 197         stats->version = TASKSTATS_VERSION;
 198         stats->nvcsw = tsk->nvcsw;
 199         stats->nivcsw = tsk->nivcsw;
 200         bacct_add_tsk(stats, tsk);
 201
 202         /* fill in extended acct fields */
 203         xacct_add_tsk(stats, tsk);
 204
 205         /* Define err: label here if needed */
 206         put_task_struct(tsk);
 207         return rc;
 208
 209 }
 210
 211 static int fill_tgid(pid_t tgid, struct task_struct *first,
 212                 struct taskstats *stats)
 213 {
 214         struct task_struct *tsk;
 215         unsigned long flags;
 216         int rc = -ESRCH;
 217
 218         /*
 219          * Add additional stats from live tasks except zombie thread group
 220          * leaders who are already counted with the dead tasks
 221          */
 222         rcu_read_lock();
 223         if (!first)
 224                 first = find_task_by_pid(tgid);
 225
 226         if (!first || !lock_task_sighand(first, &flags))
 227                 goto out;
 228
 229         if (first->signal->stats)
 230                 memcpy(stats, first->signal->stats, sizeof(*stats));
 231         else
 232                 memset(stats, 0, sizeof(*stats));
 233
 234         tsk = first;
 235         do {
 236                 if (tsk->exit_state)
 237                         continue;
 238                 /*
 239                  * Accounting subsystem can call its functions here to
 240                  * fill in relevant parts of struct taskstsats as follows
 241                  *
 242                  *      per-task-foo(stats, tsk);
 243                  */
 244                 delayacct_add_tsk(stats, tsk);
 245
 246                 stats->nvcsw += tsk->nvcsw;
 247                 stats->nivcsw += tsk->nivcsw;
 248         } while_each_thread(first, tsk);
 249
 250         unlock_task_sighand(first, &flags);
 251         rc = 0;
 252 out:
 253         rcu_read_unlock();
 254
 255         stats->version = TASKSTATS_VERSION;
 256         /*
 257          * Accounting subsytems can also add calls here to modify
 258          * fields of taskstats.
 259          */
 260         return rc;
 261 }
 262
 263
 264 static void fill_tgid_exit(struct task_struct *tsk)
 265 {
 266         unsigned long flags;
 267
 268         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 269         if (!tsk->signal->stats)
 270                 goto ret;
 271
 272         /*
 273          * Each accounting subsystem calls its functions here to
 274          * accumalate its per-task stats for tsk, into the per-tgid structure
 275          *
 276          *      per-task-foo(tsk->signal->stats, tsk);
 277          */
 278         delayacct_add_tsk(tsk->signal->stats, tsk);
 279 ret:
 280         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 281         return;
 282 }
 283
 284 static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
 285 {
 286         struct listener_list *listeners;
 287         struct listener *s, *tmp;
 288         unsigned int cpu;
 289         cpumask_t mask = *maskp;
 290
 291         if (!cpus_subset(mask, cpu_possible_map))
 292                 return -EINVAL;
 293
 294         if (isadd == REGISTER) {
 295                 for_each_cpu_mask(cpu, mask) {
 296                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 297                                          cpu_to_node(cpu));
 298                         if (!s)
 299                                 goto cleanup;
 300                         s->pid = pid;
 301                         INIT_LIST_HEAD(&s->list);
 302                         s->valid = 1;
 303
 304                         listeners = &per_cpu(listener_array, cpu);
 305                         down_write(&listeners->sem);
 306                         list_add(&s->list, &listeners->list);
 307                         up_write(&listeners->sem);
 308                 }
 309                 return 0;
 310         }
 311
 312         /* Deregister or cleanup */
 313 cleanup:
 314         for_each_cpu_mask(cpu, mask) {
 315                 listeners = &per_cpu(listener_array, cpu);
 316                 down_write(&listeners->sem);
 317                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 318                         if (s->pid == pid) {
 319                                 list_del(&s->list);
 320                                 kfree(s);
 321                                 break;
 322                         }
 323                 }
 324                 up_write(&listeners->sem);
 325         }
 326         return 0;
 327 }
 328
 329 static int parse(struct nlattr *na, cpumask_t *mask)
 330 {
 331         char *data;
 332         int len;
 333         int ret;
 334
 335         if (na == NULL)
 336                 return 1;
 337         len = nla_len(na);
 338         if (len > TASKSTATS_CPUMASK_MAXLEN)
 339                 return -E2BIG;
 340         if (len < 1)
 341                 return -EINVAL;
 342         data = kmalloc(len, GFP_KERNEL);
 343         if (!data)
 344                 return -ENOMEM;
 345         nla_strlcpy(data, na, len);
 346         ret = cpulist_parse(data, *mask);
 347         kfree(data);
 348         return ret;
 349 }
 350
 351 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid)
 352 {
 353         struct nlattr *na, *ret;
 354         int aggr;
 355
 356         aggr = (type == TASKSTATS_TYPE_PID)
 357                         ? TASKSTATS_TYPE_AGGR_PID
 358                         : TASKSTATS_TYPE_AGGR_TGID;
 359
 360         na = nla_nest_start(skb, aggr);
 361         if (!na)
 362                 goto err;
 363         if (nla_put(skb, type, sizeof(pid), &pid) < 0)
 364                 goto err;
 365         ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats));
 366         if (!ret)
 367                 goto err;
 368         nla_nest_end(skb, na);
 369
 370         return nla_data(ret);
 371 err:
 372         return NULL;
 373 }
 374
 375 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 376 {
 377         int rc = 0;
 378         struct sk_buff *rep_skb;
 379         struct taskstats *stats;
 380         size_t size;
 381         cpumask_t mask;
 382
 383         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 384         if (rc < 0)
 385                 return rc;
 386         if (rc == 0)
 387                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 388
 389         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 390         if (rc < 0)
 391                 return rc;
 392         if (rc == 0)
 393                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 394
 395         /*
 396          * Size includes space for nested attributes
 397          */
 398         size = nla_total_size(sizeof(u32)) +
 399                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 400
 401         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size);
 402         if (rc < 0)
 403                 return rc;
 404
 405         rc = -EINVAL;
 406         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 407                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 408                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid);
 409                 if (!stats)
 410                         goto err;
 411
 412                 rc = fill_pid(pid, NULL, stats);
 413                 if (rc < 0)
 414                         goto err;
 415         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 416                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 417                 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 418                 if (!stats)
 419                         goto err;
 420
 421                 rc = fill_tgid(tgid, NULL, stats);
 422                 if (rc < 0)
 423                         goto err;
 424         } else
 425                 goto err;
 426
 427         return send_reply(rep_skb, info->snd_pid);
 428 err:
 429         nlmsg_free(rep_skb);
 430         return rc;
 431 }
 432
 433 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 434 {
 435         struct signal_struct *sig = tsk->signal;
 436         struct taskstats *stats;
 437
 438         if (sig->stats || thread_group_empty(tsk))
 439                 goto ret;
 440
 441         /* No problem if kmem_cache_zalloc() fails */
 442         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 443
 444         spin_lock_irq(&tsk->sighand->siglock);
 445         if (!sig->stats) {
 446                 sig->stats = stats;
 447                 stats = NULL;
 448         }
 449         spin_unlock_irq(&tsk->sighand->siglock);
 450
 451         if (stats)
 452                 kmem_cache_free(taskstats_cache, stats);
 453 ret:
 454         return sig->stats;
 455 }
 456
 457 /* Send pid data out on exit */
 458 void taskstats_exit(struct task_struct *tsk, int group_dead)
 459 {
 460         int rc;
 461         struct listener_list *listeners;
 462         struct taskstats *stats;
 463         struct sk_buff *rep_skb;
 464         size_t size;
 465         int is_thread_group;
 466
 467         if (!family_registered)
 468                 return;
 469
 470         /*
 471          * Size includes space for nested attributes
 472          */
 473         size = nla_total_size(sizeof(u32)) +
 474                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 475
 476         is_thread_group = !!taskstats_tgid_alloc(tsk);
 477         if (is_thread_group) {
 478                 /* PID + STATS + TGID + STATS */
 479                 size = 2 * size;
 480                 /* fill the tsk->signal->stats structure */
 481                 fill_tgid_exit(tsk);
 482         }
 483
 484         listeners = &__raw_get_cpu_var(listener_array);
 485         if (list_empty(&listeners->list))
 486                 return;
 487
 488         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size);
 489         if (rc < 0)
 490                 return;
 491
 492         stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid);
 493         if (!stats)
 494                 goto err;
 495
 496         rc = fill_pid(tsk->pid, tsk, stats);
 497         if (rc < 0)
 498                 goto err;
 499
 500         /*
 501          * Doesn't matter if tsk is the leader or the last group member leaving
 502          */
 503         if (!is_thread_group || !group_dead)
 504                 goto send;
 505
 506         stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid);
 507         if (!stats)
 508                 goto err;
 509
 510         memcpy(stats, tsk->signal->stats, sizeof(*stats));
 511
 512 send:
 513         send_cpu_listeners(rep_skb, listeners);
 514         return;
 515 err:
 516         nlmsg_free(rep_skb);
 517 }
 518
 519 static struct genl_ops taskstats_ops = {
 520         .cmd            = TASKSTATS_CMD_GET,
 521         .doit           = taskstats_user_cmd,
 522         .policy         = taskstats_cmd_get_policy,
 523 };
 524
 525 /* Needed early in initialization */
 526 void __init taskstats_init_early(void)
 527 {
 528         unsigned int i;
 529
 530         taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC);
 531         for_each_possible_cpu(i) {
 532                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 533                 init_rwsem(&(per_cpu(listener_array, i).sem));
 534         }
 535 }
 536
 537 static int __init taskstats_init(void)
 538 {
 539         int rc;
 540
 541         rc = genl_register_family(&family);
 542         if (rc)
 543                 return rc;
 544
 545         rc = genl_register_ops(&family, &taskstats_ops);
 546         if (rc < 0)
 547                 goto err;
 548
 549         family_registered = 1;
 550         return 0;
 551 err:
 552         genl_unregister_family(&family);
 553         return rc;
 554 }
 555
 556 /*
 557  * late initcall ensures initialization of statistics collection
 558  * mechanisms precedes initialization of the taskstats interface
 559  */
 560 late_initcall(taskstats_init);