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