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, unsigned int cpu)
 123 {
 124         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 125         struct listener_list *listeners;
 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         listeners = &per_cpu(listener_array, cpu);
 139         down_read(&listeners->sem);
 140         list_for_each_entry(s, &listeners->list, list) {
 141                 skb_next = NULL;
 142                 if (!list_is_last(&s->list, &listeners->list)) {
 143                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 144                         if (!skb_next)
 145                                 break;
 146                 }
 147                 rc = genlmsg_unicast(skb_cur, s->pid);
 148                 if (rc == -ECONNREFUSED) {
 149                         s->valid = 0;
 150                         delcount++;
 151                 }
 152                 skb_cur = skb_next;
 153         }
 154         up_read(&listeners->sem);
 155
 156         if (skb_cur)
 157                 nlmsg_free(skb_cur);
 158
 159         if (!delcount)
 160                 return;
 161
 162         /* Delete invalidated entries */
 163         down_write(&listeners->sem);
 164         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 165                 if (!s->valid) {
 166                         list_del(&s->list);
 167                         kfree(s);
 168                 }
 169         }
 170         up_write(&listeners->sem);
 171 }
 172
 173 static int fill_pid(pid_t pid, struct task_struct *tsk,
 174                 struct taskstats *stats)
 175 {
 176         int rc = 0;
 177
 178         if (!tsk) {
 179                 rcu_read_lock();
 180                 tsk = find_task_by_pid(pid);
 181                 if (tsk)
 182                         get_task_struct(tsk);
 183                 rcu_read_unlock();
 184                 if (!tsk)
 185                         return -ESRCH;
 186         } else
 187                 get_task_struct(tsk);
 188
 189         /*
 190          * Each accounting subsystem adds calls to its functions to
 191          * fill in relevant parts of struct taskstsats as follows
 192          *
 193          *      per-task-foo(stats, tsk);
 194          */
 195
 196         delayacct_add_tsk(stats, tsk);
 197
 198         /* fill in basic acct fields */
 199         stats->version = TASKSTATS_VERSION;
 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
 232         tsk = first;
 233         do {
 234                 if (tsk->exit_state)
 235                         continue;
 236                 /*
 237                  * Accounting subsystem can call its functions here to
 238                  * fill in relevant parts of struct taskstsats as follows
 239                  *
 240                  *      per-task-foo(stats, tsk);
 241                  */
 242                 delayacct_add_tsk(stats, tsk);
 243
 244         } while_each_thread(first, tsk);
 245
 246         unlock_task_sighand(first, &flags);
 247         rc = 0;
 248 out:
 249         rcu_read_unlock();
 250
 251         stats->version = TASKSTATS_VERSION;
 252         /*
 253          * Accounting subsytems can also add calls here to modify
 254          * fields of taskstats.
 255          */
 256         return rc;
 257 }
 258
 259
 260 static void fill_tgid_exit(struct task_struct *tsk)
 261 {
 262         unsigned long flags;
 263
 264         spin_lock_irqsave(&tsk->sighand->siglock, flags);
 265         if (!tsk->signal->stats)
 266                 goto ret;
 267
 268         /*
 269          * Each accounting subsystem calls its functions here to
 270          * accumalate its per-task stats for tsk, into the per-tgid structure
 271          *
 272          *      per-task-foo(tsk->signal->stats, tsk);
 273          */
 274         delayacct_add_tsk(tsk->signal->stats, tsk);
 275 ret:
 276         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
 277         return;
 278 }
 279
 280 static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
 281 {
 282         struct listener_list *listeners;
 283         struct listener *s, *tmp;
 284         unsigned int cpu;
 285         cpumask_t mask = *maskp;
 286
 287         if (!cpus_subset(mask, cpu_possible_map))
 288                 return -EINVAL;
 289
 290         if (isadd == REGISTER) {
 291                 for_each_cpu_mask(cpu, mask) {
 292                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 293                                          cpu_to_node(cpu));
 294                         if (!s)
 295                                 goto cleanup;
 296                         s->pid = pid;
 297                         INIT_LIST_HEAD(&s->list);
 298                         s->valid = 1;
 299
 300                         listeners = &per_cpu(listener_array, cpu);
 301                         down_write(&listeners->sem);
 302                         list_add(&s->list, &listeners->list);
 303                         up_write(&listeners->sem);
 304                 }
 305                 return 0;
 306         }
 307
 308         /* Deregister or cleanup */
 309 cleanup:
 310         for_each_cpu_mask(cpu, mask) {
 311                 listeners = &per_cpu(listener_array, cpu);
 312                 down_write(&listeners->sem);
 313                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 314                         if (s->pid == pid) {
 315                                 list_del(&s->list);
 316                                 kfree(s);
 317                                 break;
 318                         }
 319                 }
 320                 up_write(&listeners->sem);
 321         }
 322         return 0;
 323 }
 324
 325 static int parse(struct nlattr *na, cpumask_t *mask)
 326 {
 327         char *data;
 328         int len;
 329         int ret;
 330
 331         if (na == NULL)
 332                 return 1;
 333         len = nla_len(na);
 334         if (len > TASKSTATS_CPUMASK_MAXLEN)
 335                 return -E2BIG;
 336         if (len < 1)
 337                 return -EINVAL;
 338         data = kmalloc(len, GFP_KERNEL);
 339         if (!data)
 340                 return -ENOMEM;
 341         nla_strlcpy(data, na, len);
 342         ret = cpulist_parse(data, *mask);
 343         kfree(data);
 344         return ret;
 345 }
 346
 347 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 348 {
 349         int rc = 0;
 350         struct sk_buff *rep_skb;
 351         struct taskstats stats;
 352         void *reply;
 353         size_t size;
 354         struct nlattr *na;
 355         cpumask_t mask;
 356
 357         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 358         if (rc < 0)
 359                 return rc;
 360         if (rc == 0)
 361                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 362
 363         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 364         if (rc < 0)
 365                 return rc;
 366         if (rc == 0)
 367                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 368
 369         /*
 370          * Size includes space for nested attributes
 371          */
 372         size = nla_total_size(sizeof(u32)) +
 373                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 374
 375         memset(&stats, 0, sizeof(stats));
 376         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 377         if (rc < 0)
 378                 return rc;
 379
 380         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 381                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 382                 rc = fill_pid(pid, NULL, &stats);
 383                 if (rc < 0)
 384                         goto err;
 385
 386                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 387                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
 388                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 389                                 stats);
 390         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 391                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 392                 rc = fill_tgid(tgid, NULL, &stats);
 393                 if (rc < 0)
 394                         goto err;
 395
 396                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 397                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 398                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 399                                 stats);
 400         } else {
 401                 rc = -EINVAL;
 402                 goto err;
 403         }
 404
 405         nla_nest_end(rep_skb, na);
 406
 407         return send_reply(rep_skb, info->snd_pid);
 408
 409 nla_put_failure:
 410         rc = genlmsg_cancel(rep_skb, reply);
 411 err:
 412         nlmsg_free(rep_skb);
 413         return rc;
 414 }
 415
 416 void taskstats_exit_alloc(struct taskstats **ptidstats, unsigned int *mycpu)
 417 {
 418         struct listener_list *listeners;
 419         /*
 420          * This is the cpu on which the task is exiting currently and will
 421          * be the one for which the exit event is sent, even if the cpu
 422          * on which this function is running changes later.
 423          */
 424         *mycpu = raw_smp_processor_id();
 425
 426         listeners = &per_cpu(listener_array, *mycpu);
 427
 428         *ptidstats = NULL;
 429         if (!list_empty(&listeners->list))
 430                 *ptidstats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL);
 431 }
 432
 433 /* Send pid data out on exit */
 434 void taskstats_exit_send(struct task_struct *tsk, struct taskstats *tidstats,
 435                         int group_dead, unsigned int mycpu)
 436 {
 437         int rc;
 438         struct sk_buff *rep_skb;
 439         void *reply;
 440         size_t size;
 441         int is_thread_group;
 442         struct nlattr *na;
 443
 444         if (!family_registered)
 445                 return;
 446
 447         /*
 448          * Size includes space for nested attributes
 449          */
 450         size = nla_total_size(sizeof(u32)) +
 451                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 452
 453         is_thread_group = (tsk->signal->stats != NULL);
 454         if (is_thread_group) {
 455                 /* PID + STATS + TGID + STATS */
 456                 size = 2 * size;
 457                 /* fill the tsk->signal->stats structure */
 458                 fill_tgid_exit(tsk);
 459         }
 460
 461         if (!tidstats)
 462                 return;
 463
 464         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 465         if (rc < 0)
 466                 goto ret;
 467
 468         rc = fill_pid(tsk->pid, tsk, tidstats);
 469         if (rc < 0)
 470                 goto err_skb;
 471
 472         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 473         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
 474         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 475                         *tidstats);
 476         nla_nest_end(rep_skb, na);
 477
 478         if (!is_thread_group)
 479                 goto send;
 480
 481         /*
 482          * Doesn't matter if tsk is the leader or the last group member leaving
 483          */
 484         if (!group_dead)
 485                 goto send;
 486
 487         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 488         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
 489         /* No locking needed for tsk->signal->stats since group is dead */
 490         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 491                         *tsk->signal->stats);
 492         nla_nest_end(rep_skb, na);
 493
 494 send:
 495         send_cpu_listeners(rep_skb, mycpu);
 496         return;
 497
 498 nla_put_failure:
 499         genlmsg_cancel(rep_skb, reply);
 500 err_skb:
 501         nlmsg_free(rep_skb);
 502 ret:
 503         return;
 504 }
 505
 506 static struct genl_ops taskstats_ops = {
 507         .cmd            = TASKSTATS_CMD_GET,
 508         .doit           = taskstats_user_cmd,
 509         .policy         = taskstats_cmd_get_policy,
 510 };
 511
 512 /* Needed early in initialization */
 513 void __init taskstats_init_early(void)
 514 {
 515         unsigned int i;
 516
 517         taskstats_cache = kmem_cache_create("taskstats_cache",
 518                                                 sizeof(struct taskstats),
 519                                                 0, SLAB_PANIC, NULL, NULL);
 520         for_each_possible_cpu(i) {
 521                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 522                 init_rwsem(&(per_cpu(listener_array, i).sem));
 523         }
 524 }
 525
 526 static int __init taskstats_init(void)
 527 {
 528         int rc;
 529
 530         rc = genl_register_family(&family);
 531         if (rc)
 532                 return rc;
 533
 534         rc = genl_register_ops(&family, &taskstats_ops);
 535         if (rc < 0)
 536                 goto err;
 537
 538         family_registered = 1;
 539         return 0;
 540 err:
 541         genl_unregister_family(&family);
 542         return rc;
 543 }
 544
 545 /*
 546  * late initcall ensures initialization of statistics collection
 547  * mechanisms precedes initialization of the taskstats interface
 548  */
 549 late_initcall(taskstats_init);