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/tsacct_kern.h>
  24 #include <linux/cpumask.h>
  25 #include <linux/percpu.h>
  26 #include <net/genetlink.h>
  27 #include <asm/atomic.h>
  28
  29 /*
  30  * Maximum length of a cpumask that can be specified in
  31  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  32  */
  33 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  34
  35 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
  36 static int family_registered;
  37 struct kmem_cache *taskstats_cache;
  38
  39 static struct genl_family family = {
  40         .id             = GENL_ID_GENERATE,
  41         .name           = TASKSTATS_GENL_NAME,
  42         .version        = TASKSTATS_GENL_VERSION,
  43         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  44 };
  45
  46 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
  47 __read_mostly = {
  48         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  49         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  50         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  51         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  52
  53 struct listener {
  54         struct list_head list;
  55         pid_t pid;
  56         char valid;
  57 };
  58
  59 struct listener_list {
  60         struct rw_semaphore sem;
  61         struct list_head list;
  62 };
  63 static DEFINE_PER_CPU(struct listener_list, listener_array);
  64
  65 enum actions {
  66         REGISTER,
  67         DEREGISTER,
  68         CPU_DONT_CARE
  69 };
  70
  71 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  72                         void **replyp, size_t size)
  73 {
  74         struct sk_buff *skb;
  75         void *reply;
  76
  77         /*
  78          * If new attributes are added, please revisit this allocation
  79          */
  80         skb = genlmsg_new(size, GFP_KERNEL);
  81         if (!skb)
  82                 return -ENOMEM;
  83
  84         if (!info) {
  85                 int seq = get_cpu_var(taskstats_seqnum)++;
  86                 put_cpu_var(taskstats_seqnum);
  87
  88                 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd);
  89         } else
  90                 reply = genlmsg_put_reply(skb, info, &family, 0, cmd);
  91         if (reply == NULL) {
  92                 nlmsg_free(skb);
  93                 return -EINVAL;
  94         }
  95
  96         *skbp = skb;
  97         *replyp = reply;
  98         return 0;
  99 }
 100
 101 /*
 102  * Send taskstats data in @skb to listener with nl_pid @pid
 103  */
 104 static int send_reply(struct sk_buff *skb, pid_t pid)
 105 {
 106         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 107         void *reply = genlmsg_data(genlhdr);
 108         int rc;
 109
 110         rc = genlmsg_end(skb, reply);
 111         if (rc < 0) {
 112                 nlmsg_free(skb);
 113                 return rc;
 114         }
 115
 116         return genlmsg_unicast(skb, pid);
 117 }
 118
 119 /*
 120  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 121  */
 122 static void send_cpu_listeners(struct sk_buff *skb,
 123                                         struct listener_list *listeners)
 124 {
 125         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 126         struct listener *s, *tmp;
 127         struct sk_buff *skb_next, *skb_cur = skb;
 128         void *reply = genlmsg_data(genlhdr);
 129         int rc, delcount = 0;
 130
 131         rc = genlmsg_end(skb, reply);
 132         if (rc < 0) {
 133                 nlmsg_free(skb);
 134                 return;
 135         }
 136
 137         rc = 0;
 138         down_read(&listeners->sem);
 139         list_for_each_entry(s, &listeners->list, list) {
 140                 skb_next = NULL;
 141                 if (!list_is_last(&s->list, &listeners->list)) {
 142                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 143                         if (!skb_next)
 144                                 break;
 145                 }
 146                 rc = genlmsg_unicast(skb_cur, s->pid);
 147                 if (rc == -ECONNREFUSED) {
 148                         s->valid = 0;
 149                         delcount++;
 150                 }
 151                 skb_cur = skb_next;
 152         }
 153         up_read(&listeners->sem);
 154
 155         if (skb_cur)
 156                 nlmsg_free(skb_cur);
 157
 158         if (!delcount)
 159                 return;
 160
 161         /* Delete invalidated entries */
 162         down_write(&listeners->sem);
 163         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 164                 if (!s->valid) {
 165                         list_del(&s->list);
 166                         kfree(s);
 167                 }
 168         }
 169         up_write(&listeners->sem);
 170 }
 171
 172 static int fill_pid(pid_t pid, struct task_struct *tsk,
 173                 struct taskstats *stats)
 174 {
 175         int rc = 0;
 176
 177         if (!tsk) {
 178                 rcu_read_lock();
 179                 tsk = find_task_by_pid(pid);
 180                 if (tsk)
 181                         get_task_struct(tsk);
 182                 rcu_read_unlock();
 183                 if (!tsk)
 184                         return -ESRCH;
 185         } else
 186                 get_task_struct(tsk);
 187
 188         /*
 189          * Each accounting subsystem adds calls to its functions to
 190          * fill in relevant parts of struct taskstsats as follows
 191          *
 192          *      per-task-foo(stats, tsk);
 193          */
 194
 195         delayacct_add_tsk(stats, tsk);
 196
 197         /* fill in basic acct fields */
 198         stats->version = TASKSTATS_VERSION;
 199         bacct_add_tsk(stats, tsk);
 200
 201         /* fill in extended acct fields */
 202         xacct_add_tsk(stats, tsk);
 203
 204         /* Define err: label here if needed */
 205         put_task_struct(tsk);
 206         return rc;
 207
 208 }
 209
 210 static int fill_tgid(pid_t tgid, struct task_struct *first,
 211                 struct taskstats *stats)
 212 {
 213         struct task_struct *tsk;
 214         unsigned long flags;
 215         int rc = -ESRCH;
 216
 217         /*
 218          * Add additional stats from live tasks except zombie thread group
 219          * leaders who are already counted with the dead tasks
 220          */
 221         rcu_read_lock();
 222         if (!first)
 223                 first = find_task_by_pid(tgid);
 224
 225         if (!first || !lock_task_sighand(first, &flags))
 226                 goto out;
 227
 228         if (first->signal->stats)
 229                 memcpy(stats, first->signal->stats, sizeof(*stats));
 230
 231         tsk = first;
 232         do {
 233                 if (tsk->exit_state)
 234                         continue;
 235                 /*
 236                  * Accounting subsystem can call its functions here to
 237                  * fill in relevant parts of struct taskstsats as follows
 238                  *
 239                  *      per-task-foo(stats, tsk);
 240                  */
 241                 delayacct_add_tsk(stats, tsk);
 242
 243         } while_each_thread(first, tsk);
 244
 245         unlock_task_sighand(first, &flags);
 246         rc = 0;
 247 out:
 248         rcu_read_unlock();
 249
 250         stats->version = TASKSTATS_VERSION;
 251         /*
 252          * Accounting subsytems can also add calls here to modify
 253          * fields of taskstats.
 254          */
 255         return rc;
 256 }
 257
 258
 259 static void fill_tgid_exit(struct task_struct *tsk)
 260 {
 261         unsigned long flags;
 262
 263         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 264         if (!tsk->signal->stats)
 265                 goto ret;
 266
 267         /*
 268          * Each accounting subsystem calls its functions here to
 269          * accumalate its per-task stats for tsk, into the per-tgid structure
 270          *
 271          *      per-task-foo(tsk->signal->stats, tsk);
 272          */
 273         delayacct_add_tsk(tsk->signal->stats, tsk);
 274 ret:
 275         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 276         return;
 277 }
 278
 279 static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
 280 {
 281         struct listener_list *listeners;
 282         struct listener *s, *tmp;
 283         unsigned int cpu;
 284         cpumask_t mask = *maskp;
 285
 286         if (!cpus_subset(mask, cpu_possible_map))
 287                 return -EINVAL;
 288
 289         if (isadd == REGISTER) {
 290                 for_each_cpu_mask(cpu, mask) {
 291                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 292                                          cpu_to_node(cpu));
 293                         if (!s)
 294                                 goto cleanup;
 295                         s->pid = pid;
 296                         INIT_LIST_HEAD(&s->list);
 297                         s->valid = 1;
 298
 299                         listeners = &per_cpu(listener_array, cpu);
 300                         down_write(&listeners->sem);
 301                         list_add(&s->list, &listeners->list);
 302                         up_write(&listeners->sem);
 303                 }
 304                 return 0;
 305         }
 306
 307         /* Deregister or cleanup */
 308 cleanup:
 309         for_each_cpu_mask(cpu, mask) {
 310                 listeners = &per_cpu(listener_array, cpu);
 311                 down_write(&listeners->sem);
 312                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 313                         if (s->pid == pid) {
 314                                 list_del(&s->list);
 315                                 kfree(s);
 316                                 break;
 317                         }
 318                 }
 319                 up_write(&listeners->sem);
 320         }
 321         return 0;
 322 }
 323
 324 static int parse(struct nlattr *na, cpumask_t *mask)
 325 {
 326         char *data;
 327         int len;
 328         int ret;
 329
 330         if (na == NULL)
 331                 return 1;
 332         len = nla_len(na);
 333         if (len > TASKSTATS_CPUMASK_MAXLEN)
 334                 return -E2BIG;
 335         if (len < 1)
 336                 return -EINVAL;
 337         data = kmalloc(len, GFP_KERNEL);
 338         if (!data)
 339                 return -ENOMEM;
 340         nla_strlcpy(data, na, len);
 341         ret = cpulist_parse(data, *mask);
 342         kfree(data);
 343         return ret;
 344 }
 345
 346 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 347 {
 348         int rc = 0;
 349         struct sk_buff *rep_skb;
 350         struct taskstats stats;
 351         void *reply;
 352         size_t size;
 353         struct nlattr *na;
 354         cpumask_t mask;
 355
 356         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 357         if (rc < 0)
 358                 return rc;
 359         if (rc == 0)
 360                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 361
 362         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 363         if (rc < 0)
 364                 return rc;
 365         if (rc == 0)
 366                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 367
 368         /*
 369          * Size includes space for nested attributes
 370          */
 371         size = nla_total_size(sizeof(u32)) +
 372                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 373
 374         memset(&stats, 0, sizeof(stats));
 375         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 376         if (rc < 0)
 377                 return rc;
 378
 379         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 380                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 381                 rc = fill_pid(pid, NULL, &stats);
 382                 if (rc < 0)
 383                         goto err;
 384
 385                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 386                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
 387                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 388                                 stats);
 389         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 390                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 391                 rc = fill_tgid(tgid, NULL, &stats);
 392                 if (rc < 0)
 393                         goto err;
 394
 395                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 396                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 397                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 398                                 stats);
 399         } else {
 400                 rc = -EINVAL;
 401                 goto err;
 402         }
 403
 404         nla_nest_end(rep_skb, na);
 405
 406         return send_reply(rep_skb, info->snd_pid);
 407
 408 nla_put_failure:
 409         rc = genlmsg_cancel(rep_skb, reply);
 410 err:
 411         nlmsg_free(rep_skb);
 412         return rc;
 413 }
 414
 415 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk)
 416 {
 417         struct signal_struct *sig = tsk->signal;
 418         struct taskstats *stats;
 419
 420         if (sig->stats || thread_group_empty(tsk))
 421                 goto ret;
 422
 423         /* No problem if kmem_cache_zalloc() fails */
 424         stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 425
 426         spin_lock_irq(&tsk->sighand->siglock);
 427         if (!sig->stats) {
 428                 sig->stats = stats;
 429                 stats = NULL;
 430         }
 431         spin_unlock_irq(&tsk->sighand->siglock);
 432
 433         if (stats)
 434                 kmem_cache_free(taskstats_cache, stats);
 435 ret:
 436         return sig->stats;
 437 }
 438
 439 /* Send pid data out on exit */
 440 void taskstats_exit(struct task_struct *tsk, int group_dead)
 441 {
 442         int rc;
 443         struct listener_list *listeners;
 444         struct taskstats *tidstats;
 445         struct sk_buff *rep_skb;
 446         void *reply;
 447         size_t size;
 448         int is_thread_group;
 449         struct nlattr *na;
 450
 451         if (!family_registered)
 452                 return;
 453
 454         /*
 455          * Size includes space for nested attributes
 456          */
 457         size = nla_total_size(sizeof(u32)) +
 458                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 459
 460         is_thread_group = !!taskstats_tgid_alloc(tsk);
 461         if (is_thread_group) {
 462                 /* PID + STATS + TGID + STATS */
 463                 size = 2 * size;
 464                 /* fill the tsk->signal->stats structure */
 465                 fill_tgid_exit(tsk);
 466         }
 467
 468         listeners = &__raw_get_cpu_var(listener_array);
 469         if (list_empty(&listeners->list))
 470                 return;
 471
 472         tidstats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 473         if (!tidstats)
 474                 return;
 475
 476         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 477         if (rc < 0)
 478                 goto free_stats;
 479
 480         rc = fill_pid(tsk->pid, tsk, tidstats);
 481         if (rc < 0)
 482                 goto err_skb;
 483
 484         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 485         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
 486         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 487                         *tidstats);
 488         nla_nest_end(rep_skb, na);
 489
 490         if (!is_thread_group)
 491                 goto send;
 492
 493         /*
 494          * Doesn't matter if tsk is the leader or the last group member leaving
 495          */
 496         if (!group_dead)
 497                 goto send;
 498
 499         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 500         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
 501         /* No locking needed for tsk->signal->stats since group is dead */
 502         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 503                         *tsk->signal->stats);
 504         nla_nest_end(rep_skb, na);
 505
 506 send:
 507         send_cpu_listeners(rep_skb, listeners);
 508 free_stats:
 509         kmem_cache_free(taskstats_cache, tidstats);
 510         return;
 511
 512 nla_put_failure:
 513         genlmsg_cancel(rep_skb, reply);
 514 err_skb:
 515         nlmsg_free(rep_skb);
 516         goto free_stats;
 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_create("taskstats_cache",
 531                                                 sizeof(struct taskstats),
 532                                                 0, SLAB_PANIC, NULL, NULL);
 533         for_each_possible_cpu(i) {
 534                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 535                 init_rwsem(&(per_cpu(listener_array, i).sem));
 536         }
 537 }
 538
 539 static int __init taskstats_init(void)
 540 {
 541         int rc;
 542
 543         rc = genl_register_family(&family);
 544         if (rc)
 545                 return rc;
 546
 547         rc = genl_register_ops(&family, &taskstats_ops);
 548         if (rc < 0)
 549                 goto err;
 550
 551         family_registered = 1;
 552         return 0;
 553 err:
 554         genl_unregister_family(&family);
 555         return rc;
 556 }
 557
 558 /*
 559  * late initcall ensures initialization of statistics collection
 560  * mechanisms precedes initialization of the taskstats interface
 561  */
 562 late_initcall(taskstats_init);