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/delayacct.h>
  22 #include <linux/cpumask.h>
  23 #include <linux/percpu.h>
  24 #include <net/genetlink.h>
  25 #include <asm/atomic.h>
  26
  27 /*
  28  * Maximum length of a cpumask that can be specified in
  29  * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute
  30  */
  31 #define TASKSTATS_CPUMASK_MAXLEN        (100+6*NR_CPUS)
  32
  33 static DEFINE_PER_CPU(__u32, taskstats_seqnum) = { 0 };
  34 static int family_registered;
  35 kmem_cache_t *taskstats_cache;
  36
  37 static struct genl_family family = {
  38         .id             = GENL_ID_GENERATE,
  39         .name           = TASKSTATS_GENL_NAME,
  40         .version        = TASKSTATS_GENL_VERSION,
  41         .maxattr        = TASKSTATS_CMD_ATTR_MAX,
  42 };
  43
  44 static struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1]
  45 __read_mostly = {
  46         [TASKSTATS_CMD_ATTR_PID]  = { .type = NLA_U32 },
  47         [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 },
  48         [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING },
  49         [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },};
  50
  51 struct listener {
  52         struct list_head list;
  53         pid_t pid;
  54         char valid;
  55 };
  56
  57 struct listener_list {
  58         struct rw_semaphore sem;
  59         struct list_head list;
  60 };
  61 static DEFINE_PER_CPU(struct listener_list, listener_array);
  62
  63 enum actions {
  64         REGISTER,
  65         DEREGISTER,
  66         CPU_DONT_CARE
  67 };
  68
  69 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp,
  70                         void **replyp, size_t size)
  71 {
  72         struct sk_buff *skb;
  73         void *reply;
  74
  75         /*
  76          * If new attributes are added, please revisit this allocation
  77          */
  78         skb = nlmsg_new(size);
  79         if (!skb)
  80                 return -ENOMEM;
  81
  82         if (!info) {
  83                 int seq = get_cpu_var(taskstats_seqnum)++;
  84                 put_cpu_var(taskstats_seqnum);
  85
  86                 reply = genlmsg_put(skb, 0, seq,
  87                                 family.id, 0, 0,
  88                                 cmd, family.version);
  89         } else
  90                 reply = genlmsg_put(skb, info->snd_pid, info->snd_seq,
  91                                 family.id, 0, 0,
  92                                 cmd, family.version);
  93         if (reply == NULL) {
  94                 nlmsg_free(skb);
  95                 return -EINVAL;
  96         }
  97
  98         *skbp = skb;
  99         *replyp = reply;
 100         return 0;
 101 }
 102
 103 /*
 104  * Send taskstats data in @skb to listener with nl_pid @pid
 105  */
 106 static int send_reply(struct sk_buff *skb, pid_t pid)
 107 {
 108         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 109         void *reply = genlmsg_data(genlhdr);
 110         int rc;
 111
 112         rc = genlmsg_end(skb, reply);
 113         if (rc < 0) {
 114                 nlmsg_free(skb);
 115                 return rc;
 116         }
 117
 118         return genlmsg_unicast(skb, pid);
 119 }
 120
 121 /*
 122  * Send taskstats data in @skb to listeners registered for @cpu's exit data
 123  */
 124 static void send_cpu_listeners(struct sk_buff *skb, unsigned int cpu)
 125 {
 126         struct genlmsghdr *genlhdr = nlmsg_data((struct nlmsghdr *)skb->data);
 127         struct listener_list *listeners;
 128         struct listener *s, *tmp;
 129         struct sk_buff *skb_next, *skb_cur = skb;
 130         void *reply = genlmsg_data(genlhdr);
 131         int rc, delcount = 0;
 132
 133         rc = genlmsg_end(skb, reply);
 134         if (rc < 0) {
 135                 nlmsg_free(skb);
 136                 return;
 137         }
 138
 139         rc = 0;
 140         listeners = &per_cpu(listener_array, cpu);
 141         down_read(&listeners->sem);
 142         list_for_each_entry(s, &listeners->list, list) {
 143                 skb_next = NULL;
 144                 if (!list_is_last(&s->list, &listeners->list)) {
 145                         skb_next = skb_clone(skb_cur, GFP_KERNEL);
 146                         if (!skb_next)
 147                                 break;
 148                 }
 149                 rc = genlmsg_unicast(skb_cur, s->pid);
 150                 if (rc == -ECONNREFUSED) {
 151                         s->valid = 0;
 152                         delcount++;
 153                 }
 154                 skb_cur = skb_next;
 155         }
 156         up_read(&listeners->sem);
 157
 158         if (skb_cur)
 159                 nlmsg_free(skb_cur);
 160
 161         if (!delcount)
 162                 return;
 163
 164         /* Delete invalidated entries */
 165         down_write(&listeners->sem);
 166         list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 167                 if (!s->valid) {
 168                         list_del(&s->list);
 169                         kfree(s);
 170                 }
 171         }
 172         up_write(&listeners->sem);
 173 }
 174
 175 static int fill_pid(pid_t pid, struct task_struct *pidtsk,
 176                 struct taskstats *stats)
 177 {
 178         int rc = 0;
 179         struct task_struct *tsk = pidtsk;
 180
 181         if (!pidtsk) {
 182                 read_lock(&tasklist_lock);
 183                 tsk = find_task_by_pid(pid);
 184                 if (!tsk) {
 185                         read_unlock(&tasklist_lock);
 186                         return -ESRCH;
 187                 }
 188                 get_task_struct(tsk);
 189                 read_unlock(&tasklist_lock);
 190         } else
 191                 get_task_struct(tsk);
 192
 193         /*
 194          * Each accounting subsystem adds calls to its functions to
 195          * fill in relevant parts of struct taskstsats as follows
 196          *
 197          *      per-task-foo(stats, tsk);
 198          */
 199
 200         delayacct_add_tsk(stats, tsk);
 201         stats->version = TASKSTATS_VERSION;
 202
 203         /* Define err: label here if needed */
 204         put_task_struct(tsk);
 205         return rc;
 206
 207 }
 208
 209 static int fill_tgid(pid_t tgid, struct task_struct *tgidtsk,
 210                 struct taskstats *stats)
 211 {
 212         struct task_struct *tsk, *first;
 213         unsigned long flags;
 214
 215         /*
 216          * Add additional stats from live tasks except zombie thread group
 217          * leaders who are already counted with the dead tasks
 218          */
 219         first = tgidtsk;
 220         if (!first) {
 221                 read_lock(&tasklist_lock);
 222                 first = find_task_by_pid(tgid);
 223                 if (!first) {
 224                         read_unlock(&tasklist_lock);
 225                         return -ESRCH;
 226                 }
 227                 get_task_struct(first);
 228                 read_unlock(&tasklist_lock);
 229         } else
 230                 get_task_struct(first);
 231
 232
 233         tsk = first;
 234         read_lock(&tasklist_lock);
 235         /* Start with stats from dead tasks */
 236         if (first->signal) {
 237                 spin_lock_irqsave(&first->signal->stats_lock, flags);
 238                 if (first->signal->stats)
 239                         memcpy(stats, first->signal->stats, sizeof(*stats));
 240                 spin_unlock_irqrestore(&first->signal->stats_lock, flags);
 241         }
 242
 243         do {
 244                 if (tsk->exit_state == EXIT_ZOMBIE && thread_group_leader(tsk))
 245                         continue;
 246                 /*
 247                  * Accounting subsystem can call its functions here to
 248                  * fill in relevant parts of struct taskstsats as follows
 249                  *
 250                  *      per-task-foo(stats, tsk);
 251                  */
 252                 delayacct_add_tsk(stats, tsk);
 253
 254         } while_each_thread(first, tsk);
 255         read_unlock(&tasklist_lock);
 256         stats->version = TASKSTATS_VERSION;
 257
 258         /*
 259          * Accounting subsytems can also add calls here to modify
 260          * fields of taskstats.
 261          */
 262         put_task_struct(first);
 263         return 0;
 264 }
 265
 266
 267 static void fill_tgid_exit(struct task_struct *tsk)
 268 {
 269         unsigned long flags;
 270
 271         spin_lock_irqsave(&tsk->signal->stats_lock, flags);
 272         if (!tsk->signal->stats)
 273                 goto ret;
 274
 275         /*
 276          * Each accounting subsystem calls its functions here to
 277          * accumalate its per-task stats for tsk, into the per-tgid structure
 278          *
 279          *      per-task-foo(tsk->signal->stats, tsk);
 280          */
 281         delayacct_add_tsk(tsk->signal->stats, tsk);
 282 ret:
 283         spin_unlock_irqrestore(&tsk->signal->stats_lock, flags);
 284         return;
 285 }
 286
 287 static int add_del_listener(pid_t pid, cpumask_t *maskp, int isadd)
 288 {
 289         struct listener_list *listeners;
 290         struct listener *s, *tmp;
 291         unsigned int cpu;
 292         cpumask_t mask = *maskp;
 293
 294         if (!cpus_subset(mask, cpu_possible_map))
 295                 return -EINVAL;
 296
 297         if (isadd == REGISTER) {
 298                 for_each_cpu_mask(cpu, mask) {
 299                         s = kmalloc_node(sizeof(struct listener), GFP_KERNEL,
 300                                          cpu_to_node(cpu));
 301                         if (!s)
 302                                 goto cleanup;
 303                         s->pid = pid;
 304                         INIT_LIST_HEAD(&s->list);
 305                         s->valid = 1;
 306
 307                         listeners = &per_cpu(listener_array, cpu);
 308                         down_write(&listeners->sem);
 309                         list_add(&s->list, &listeners->list);
 310                         up_write(&listeners->sem);
 311                 }
 312                 return 0;
 313         }
 314
 315         /* Deregister or cleanup */
 316 cleanup:
 317         for_each_cpu_mask(cpu, mask) {
 318                 listeners = &per_cpu(listener_array, cpu);
 319                 down_write(&listeners->sem);
 320                 list_for_each_entry_safe(s, tmp, &listeners->list, list) {
 321                         if (s->pid == pid) {
 322                                 list_del(&s->list);
 323                                 kfree(s);
 324                                 break;
 325                         }
 326                 }
 327                 up_write(&listeners->sem);
 328         }
 329         return 0;
 330 }
 331
 332 static int parse(struct nlattr *na, cpumask_t *mask)
 333 {
 334         char *data;
 335         int len;
 336         int ret;
 337
 338         if (na == NULL)
 339                 return 1;
 340         len = nla_len(na);
 341         if (len > TASKSTATS_CPUMASK_MAXLEN)
 342                 return -E2BIG;
 343         if (len < 1)
 344                 return -EINVAL;
 345         data = kmalloc(len, GFP_KERNEL);
 346         if (!data)
 347                 return -ENOMEM;
 348         nla_strlcpy(data, na, len);
 349         ret = cpulist_parse(data, *mask);
 350         kfree(data);
 351         return ret;
 352 }
 353
 354 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
 355 {
 356         int rc = 0;
 357         struct sk_buff *rep_skb;
 358         struct taskstats stats;
 359         void *reply;
 360         size_t size;
 361         struct nlattr *na;
 362         cpumask_t mask;
 363
 364         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], &mask);
 365         if (rc < 0)
 366                 return rc;
 367         if (rc == 0)
 368                 return add_del_listener(info->snd_pid, &mask, REGISTER);
 369
 370         rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], &mask);
 371         if (rc < 0)
 372                 return rc;
 373         if (rc == 0)
 374                 return add_del_listener(info->snd_pid, &mask, DEREGISTER);
 375
 376         /*
 377          * Size includes space for nested attributes
 378          */
 379         size = nla_total_size(sizeof(u32)) +
 380                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 381
 382         memset(&stats, 0, sizeof(stats));
 383         rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 384         if (rc < 0)
 385                 return rc;
 386
 387         if (info->attrs[TASKSTATS_CMD_ATTR_PID]) {
 388                 u32 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]);
 389                 rc = fill_pid(pid, NULL, &stats);
 390                 if (rc < 0)
 391                         goto err;
 392
 393                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 394                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, pid);
 395                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 396                                 stats);
 397         } else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) {
 398                 u32 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]);
 399                 rc = fill_tgid(tgid, NULL, &stats);
 400                 if (rc < 0)
 401                         goto err;
 402
 403                 na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 404                 NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, tgid);
 405                 NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 406                                 stats);
 407         } else {
 408                 rc = -EINVAL;
 409                 goto err;
 410         }
 411
 412         nla_nest_end(rep_skb, na);
 413
 414         return send_reply(rep_skb, info->snd_pid);
 415
 416 nla_put_failure:
 417         return genlmsg_cancel(rep_skb, reply);
 418 err:
 419         nlmsg_free(rep_skb);
 420         return rc;
 421 }
 422
 423 void taskstats_exit_alloc(struct taskstats **ptidstats, unsigned int *mycpu)
 424 {
 425         struct listener_list *listeners;
 426         struct taskstats *tmp;
 427         /*
 428          * This is the cpu on which the task is exiting currently and will
 429          * be the one for which the exit event is sent, even if the cpu
 430          * on which this function is running changes later.
 431          */
 432         *mycpu = raw_smp_processor_id();
 433
 434         *ptidstats = NULL;
 435         tmp = kmem_cache_zalloc(taskstats_cache, SLAB_KERNEL);
 436         if (!tmp)
 437                 return;
 438
 439         listeners = &per_cpu(listener_array, *mycpu);
 440         down_read(&listeners->sem);
 441         if (!list_empty(&listeners->list)) {
 442                 *ptidstats = tmp;
 443                 tmp = NULL;
 444         }
 445         up_read(&listeners->sem);
 446         kfree(tmp);
 447 }
 448
 449 /* Send pid data out on exit */
 450 void taskstats_exit_send(struct task_struct *tsk, struct taskstats *tidstats,
 451                         int group_dead, unsigned int mycpu)
 452 {
 453         int rc;
 454         struct sk_buff *rep_skb;
 455         void *reply;
 456         size_t size;
 457         int is_thread_group;
 458         struct nlattr *na;
 459         unsigned long flags;
 460
 461         if (!family_registered || !tidstats)
 462                 return;
 463
 464         spin_lock_irqsave(&tsk->signal->stats_lock, flags);
 465         is_thread_group = tsk->signal->stats ? 1 : 0;
 466         spin_unlock_irqrestore(&tsk->signal->stats_lock, flags);
 467
 468         rc = 0;
 469         /*
 470          * Size includes space for nested attributes
 471          */
 472         size = nla_total_size(sizeof(u32)) +
 473                 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0);
 474
 475         if (is_thread_group)
 476                 size = 2 * size;        /* PID + STATS + TGID + STATS */
 477
 478         rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, &reply, size);
 479         if (rc < 0)
 480                 goto ret;
 481
 482         rc = fill_pid(tsk->pid, tsk, tidstats);
 483         if (rc < 0)
 484                 goto err_skb;
 485
 486         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_PID);
 487         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_PID, (u32)tsk->pid);
 488         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 489                         *tidstats);
 490         nla_nest_end(rep_skb, na);
 491
 492         if (!is_thread_group)
 493                 goto send;
 494
 495         /*
 496          * tsk has/had a thread group so fill the tsk->signal->stats structure
 497          * Doesn't matter if tsk is the leader or the last group member leaving
 498          */
 499
 500         fill_tgid_exit(tsk);
 501         if (!group_dead)
 502                 goto send;
 503
 504         na = nla_nest_start(rep_skb, TASKSTATS_TYPE_AGGR_TGID);
 505         NLA_PUT_U32(rep_skb, TASKSTATS_TYPE_TGID, (u32)tsk->tgid);
 506         /* No locking needed for tsk->signal->stats since group is dead */
 507         NLA_PUT_TYPE(rep_skb, struct taskstats, TASKSTATS_TYPE_STATS,
 508                         *tsk->signal->stats);
 509         nla_nest_end(rep_skb, na);
 510
 511 send:
 512         send_cpu_listeners(rep_skb, mycpu);
 513         return;
 514
 515 nla_put_failure:
 516         genlmsg_cancel(rep_skb, reply);
 517         goto ret;
 518 err_skb:
 519         nlmsg_free(rep_skb);
 520 ret:
 521         return;
 522 }
 523
 524 static struct genl_ops taskstats_ops = {
 525         .cmd            = TASKSTATS_CMD_GET,
 526         .doit           = taskstats_user_cmd,
 527         .policy         = taskstats_cmd_get_policy,
 528 };
 529
 530 /* Needed early in initialization */
 531 void __init taskstats_init_early(void)
 532 {
 533         unsigned int i;
 534
 535         taskstats_cache = kmem_cache_create("taskstats_cache",
 536                                                 sizeof(struct taskstats),
 537                                                 0, SLAB_PANIC, NULL, NULL);
 538         for_each_possible_cpu(i) {
 539                 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list));
 540                 init_rwsem(&(per_cpu(listener_array, i).sem));
 541         }
 542 }
 543
 544 static int __init taskstats_init(void)
 545 {
 546         int rc;
 547
 548         rc = genl_register_family(&family);
 549         if (rc)
 550                 return rc;
 551
 552         rc = genl_register_ops(&family, &taskstats_ops);
 553         if (rc < 0)
 554                 goto err;
 555
 556         family_registered = 1;
 557         return 0;
 558 err:
 559         genl_unregister_family(&family);
 560         return rc;
 561 }
 562
 563 /*
 564  * late initcall ensures initialization of statistics collection
 565  * mechanisms precedes initialization of the taskstats interface
 566  */
 567 late_initcall(taskstats_init);