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 <linux/slab.h> 26 #include <linux/cgroupstats.h> 27 #include <linux/cgroup.h> 28 #include <linux/fs.h> 29 #include <linux/file.h> 30 #include <linux/pid_namespace.h> 31 #include <net/genetlink.h> 32 #include <linux/atomic.h> 33 34 /* 35 * Maximum length of a cpumask that can be specified in 36 * the TASKSTATS_CMD_ATTR_REGISTER/DEREGISTER_CPUMASK attribute 37 */ 38 #define TASKSTATS_CPUMASK_MAXLEN (100+6*NR_CPUS) 39 40 static DEFINE_PER_CPU(__u32, taskstats_seqnum); 41 static int family_registered; 42 struct kmem_cache *taskstats_cache; 43 44 static struct genl_family family = { 45 .id = GENL_ID_GENERATE, 46 .name = TASKSTATS_GENL_NAME, 47 .version = TASKSTATS_GENL_VERSION, 48 .maxattr = TASKSTATS_CMD_ATTR_MAX, 49 }; 50 51 static const struct nla_policy taskstats_cmd_get_policy[TASKSTATS_CMD_ATTR_MAX+1] = { 52 [TASKSTATS_CMD_ATTR_PID] = { .type = NLA_U32 }, 53 [TASKSTATS_CMD_ATTR_TGID] = { .type = NLA_U32 }, 54 [TASKSTATS_CMD_ATTR_REGISTER_CPUMASK] = { .type = NLA_STRING }, 55 [TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK] = { .type = NLA_STRING },}; 56 57 static const struct nla_policy cgroupstats_cmd_get_policy[CGROUPSTATS_CMD_ATTR_MAX+1] = { 58 [CGROUPSTATS_CMD_ATTR_FD] = { .type = NLA_U32 }, 59 }; 60 61 struct listener { 62 struct list_head list; 63 pid_t pid; 64 char valid; 65 }; 66 67 struct listener_list { 68 struct rw_semaphore sem; 69 struct list_head list; 70 }; 71 static DEFINE_PER_CPU(struct listener_list, listener_array); 72 73 enum actions { 74 REGISTER, 75 DEREGISTER, 76 CPU_DONT_CARE 77 }; 78 79 static int prepare_reply(struct genl_info *info, u8 cmd, struct sk_buff **skbp, 80 size_t size) 81 { 82 struct sk_buff *skb; 83 void *reply; 84 85 /* 86 * If new attributes are added, please revisit this allocation 87 */ 88 skb = genlmsg_new(size, GFP_KERNEL); 89 if (!skb) 90 return -ENOMEM; 91 92 if (!info) { 93 int seq = this_cpu_inc_return(taskstats_seqnum) - 1; 94 95 reply = genlmsg_put(skb, 0, seq, &family, 0, cmd); 96 } else 97 reply = genlmsg_put_reply(skb, info, &family, 0, cmd); 98 if (reply == NULL) { 99 nlmsg_free(skb); 100 return -EINVAL; 101 } 102 103 *skbp = skb; 104 return 0; 105 } 106 107 /* 108 * Send taskstats data in @skb to listener with nl_pid @pid 109 */ 110 static int send_reply(struct sk_buff *skb, struct genl_info *info) 111 { 112 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb)); 113 void *reply = genlmsg_data(genlhdr); 114 int rc; 115 116 rc = genlmsg_end(skb, reply); 117 if (rc < 0) { 118 nlmsg_free(skb); 119 return rc; 120 } 121 122 return genlmsg_reply(skb, info); 123 } 124 125 /* 126 * Send taskstats data in @skb to listeners registered for @cpu's exit data 127 */ 128 static void send_cpu_listeners(struct sk_buff *skb, 129 struct listener_list *listeners) 130 { 131 struct genlmsghdr *genlhdr = nlmsg_data(nlmsg_hdr(skb)); 132 struct listener *s, *tmp; 133 struct sk_buff *skb_next, *skb_cur = skb; 134 void *reply = genlmsg_data(genlhdr); 135 int rc, delcount = 0; 136 137 rc = genlmsg_end(skb, reply); 138 if (rc < 0) { 139 nlmsg_free(skb); 140 return; 141 } 142 143 rc = 0; 144 down_read(&listeners->sem); 145 list_for_each_entry(s, &listeners->list, list) { 146 skb_next = NULL; 147 if (!list_is_last(&s->list, &listeners->list)) { 148 skb_next = skb_clone(skb_cur, GFP_KERNEL); 149 if (!skb_next) 150 break; 151 } 152 rc = genlmsg_unicast(&init_net, skb_cur, s->pid); 153 if (rc == -ECONNREFUSED) { 154 s->valid = 0; 155 delcount++; 156 } 157 skb_cur = skb_next; 158 } 159 up_read(&listeners->sem); 160 161 if (skb_cur) 162 nlmsg_free(skb_cur); 163 164 if (!delcount) 165 return; 166 167 /* Delete invalidated entries */ 168 down_write(&listeners->sem); 169 list_for_each_entry_safe(s, tmp, &listeners->list, list) { 170 if (!s->valid) { 171 list_del(&s->list); 172 kfree(s); 173 } 174 } 175 up_write(&listeners->sem); 176 } 177 178 static void fill_stats(struct user_namespace *user_ns, 179 struct pid_namespace *pid_ns, 180 struct task_struct *tsk, struct taskstats *stats) 181 { 182 memset(stats, 0, sizeof(*stats)); 183 /* 184 * Each accounting subsystem adds calls to its functions to 185 * fill in relevant parts of struct taskstsats as follows 186 * 187 * per-task-foo(stats, tsk); 188 */ 189 190 delayacct_add_tsk(stats, tsk); 191 192 /* fill in basic acct fields */ 193 stats->version = TASKSTATS_VERSION; 194 stats->nvcsw = tsk->nvcsw; 195 stats->nivcsw = tsk->nivcsw; 196 bacct_add_tsk(user_ns, pid_ns, stats, tsk); 197 198 /* fill in extended acct fields */ 199 xacct_add_tsk(stats, tsk); 200 } 201 202 static int fill_stats_for_pid(pid_t pid, struct taskstats *stats) 203 { 204 struct task_struct *tsk; 205 206 rcu_read_lock(); 207 tsk = find_task_by_vpid(pid); 208 if (tsk) 209 get_task_struct(tsk); 210 rcu_read_unlock(); 211 if (!tsk) 212 return -ESRCH; 213 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats); 214 put_task_struct(tsk); 215 return 0; 216 } 217 218 static int fill_stats_for_tgid(pid_t tgid, struct taskstats *stats) 219 { 220 struct task_struct *tsk, *first; 221 unsigned long flags; 222 int rc = -ESRCH; 223 224 /* 225 * Add additional stats from live tasks except zombie thread group 226 * leaders who are already counted with the dead tasks 227 */ 228 rcu_read_lock(); 229 first = find_task_by_vpid(tgid); 230 231 if (!first || !lock_task_sighand(first, &flags)) 232 goto out; 233 234 if (first->signal->stats) 235 memcpy(stats, first->signal->stats, sizeof(*stats)); 236 else 237 memset(stats, 0, sizeof(*stats)); 238 239 tsk = first; 240 do { 241 if (tsk->exit_state) 242 continue; 243 /* 244 * Accounting subsystem can call its functions here to 245 * fill in relevant parts of struct taskstsats as follows 246 * 247 * per-task-foo(stats, tsk); 248 */ 249 delayacct_add_tsk(stats, tsk); 250 251 stats->nvcsw += tsk->nvcsw; 252 stats->nivcsw += tsk->nivcsw; 253 } while_each_thread(first, tsk); 254 255 unlock_task_sighand(first, &flags); 256 rc = 0; 257 out: 258 rcu_read_unlock(); 259 260 stats->version = TASKSTATS_VERSION; 261 /* 262 * Accounting subsystems can also add calls here to modify 263 * fields of taskstats. 264 */ 265 return rc; 266 } 267 268 static void fill_tgid_exit(struct task_struct *tsk) 269 { 270 unsigned long flags; 271 272 spin_lock_irqsave(&tsk->sighand->siglock, flags); 273 if (!tsk->signal->stats) 274 goto ret; 275 276 /* 277 * Each accounting subsystem calls its functions here to 278 * accumalate its per-task stats for tsk, into the per-tgid structure 279 * 280 * per-task-foo(tsk->signal->stats, tsk); 281 */ 282 delayacct_add_tsk(tsk->signal->stats, tsk); 283 ret: 284 spin_unlock_irqrestore(&tsk->sighand->siglock, flags); 285 return; 286 } 287 288 static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd) 289 { 290 struct listener_list *listeners; 291 struct listener *s, *tmp, *s2; 292 unsigned int cpu; 293 294 if (!cpumask_subset(mask, cpu_possible_mask)) 295 return -EINVAL; 296 297 if (current_user_ns() != &init_user_ns) 298 return -EINVAL; 299 300 if (task_active_pid_ns(current) != &init_pid_ns) 301 return -EINVAL; 302 303 if (isadd == REGISTER) { 304 for_each_cpu(cpu, mask) { 305 s = kmalloc_node(sizeof(struct listener), 306 GFP_KERNEL, cpu_to_node(cpu)); 307 if (!s) 308 goto cleanup; 309 310 s->pid = pid; 311 s->valid = 1; 312 313 listeners = &per_cpu(listener_array, cpu); 314 down_write(&listeners->sem); 315 list_for_each_entry(s2, &listeners->list, list) { 316 if (s2->pid == pid && s2->valid) 317 goto exists; 318 } 319 list_add(&s->list, &listeners->list); 320 s = NULL; 321 exists: 322 up_write(&listeners->sem); 323 kfree(s); /* nop if NULL */ 324 } 325 return 0; 326 } 327 328 /* Deregister or cleanup */ 329 cleanup: 330 for_each_cpu(cpu, mask) { 331 listeners = &per_cpu(listener_array, cpu); 332 down_write(&listeners->sem); 333 list_for_each_entry_safe(s, tmp, &listeners->list, list) { 334 if (s->pid == pid) { 335 list_del(&s->list); 336 kfree(s); 337 break; 338 } 339 } 340 up_write(&listeners->sem); 341 } 342 return 0; 343 } 344 345 static int parse(struct nlattr *na, struct cpumask *mask) 346 { 347 char *data; 348 int len; 349 int ret; 350 351 if (na == NULL) 352 return 1; 353 len = nla_len(na); 354 if (len > TASKSTATS_CPUMASK_MAXLEN) 355 return -E2BIG; 356 if (len < 1) 357 return -EINVAL; 358 data = kmalloc(len, GFP_KERNEL); 359 if (!data) 360 return -ENOMEM; 361 nla_strlcpy(data, na, len); 362 ret = cpulist_parse(data, mask); 363 kfree(data); 364 return ret; 365 } 366 367 #if defined(CONFIG_64BIT) && !defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) 368 #define TASKSTATS_NEEDS_PADDING 1 369 #endif 370 371 static struct taskstats *mk_reply(struct sk_buff *skb, int type, u32 pid) 372 { 373 struct nlattr *na, *ret; 374 int aggr; 375 376 aggr = (type == TASKSTATS_TYPE_PID) 377 ? TASKSTATS_TYPE_AGGR_PID 378 : TASKSTATS_TYPE_AGGR_TGID; 379 380 /* 381 * The taskstats structure is internally aligned on 8 byte 382 * boundaries but the layout of the aggregrate reply, with 383 * two NLA headers and the pid (each 4 bytes), actually 384 * force the entire structure to be unaligned. This causes 385 * the kernel to issue unaligned access warnings on some 386 * architectures like ia64. Unfortunately, some software out there 387 * doesn't properly unroll the NLA packet and assumes that the start 388 * of the taskstats structure will always be 20 bytes from the start 389 * of the netlink payload. Aligning the start of the taskstats 390 * structure breaks this software, which we don't want. So, for now 391 * the alignment only happens on architectures that require it 392 * and those users will have to update to fixed versions of those 393 * packages. Space is reserved in the packet only when needed. 394 * This ifdef should be removed in several years e.g. 2012 once 395 * we can be confident that fixed versions are installed on most 396 * systems. We add the padding before the aggregate since the 397 * aggregate is already a defined type. 398 */ 399 #ifdef TASKSTATS_NEEDS_PADDING 400 if (nla_put(skb, TASKSTATS_TYPE_NULL, 0, NULL) < 0) 401 goto err; 402 #endif 403 na = nla_nest_start(skb, aggr); 404 if (!na) 405 goto err; 406 407 if (nla_put(skb, type, sizeof(pid), &pid) < 0) 408 goto err; 409 ret = nla_reserve(skb, TASKSTATS_TYPE_STATS, sizeof(struct taskstats)); 410 if (!ret) 411 goto err; 412 nla_nest_end(skb, na); 413 414 return nla_data(ret); 415 err: 416 return NULL; 417 } 418 419 static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info) 420 { 421 int rc = 0; 422 struct sk_buff *rep_skb; 423 struct cgroupstats *stats; 424 struct nlattr *na; 425 size_t size; 426 u32 fd; 427 struct fd f; 428 429 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD]; 430 if (!na) 431 return -EINVAL; 432 433 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]); 434 f = fdget(fd); 435 if (!f.file) 436 return 0; 437 438 size = nla_total_size(sizeof(struct cgroupstats)); 439 440 rc = prepare_reply(info, CGROUPSTATS_CMD_NEW, &rep_skb, 441 size); 442 if (rc < 0) 443 goto err; 444 445 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS, 446 sizeof(struct cgroupstats)); 447 if (na == NULL) { 448 rc = -EMSGSIZE; 449 goto err; 450 } 451 452 stats = nla_data(na); 453 memset(stats, 0, sizeof(*stats)); 454 455 rc = cgroupstats_build(stats, f.file->f_dentry); 456 if (rc < 0) { 457 nlmsg_free(rep_skb); 458 goto err; 459 } 460 461 rc = send_reply(rep_skb, info); 462 463 err: 464 fdput(f); 465 return rc; 466 } 467 468 static int cmd_attr_register_cpumask(struct genl_info *info) 469 { 470 cpumask_var_t mask; 471 int rc; 472 473 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 474 return -ENOMEM; 475 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask); 476 if (rc < 0) 477 goto out; 478 rc = add_del_listener(info->snd_portid, mask, REGISTER); 479 out: 480 free_cpumask_var(mask); 481 return rc; 482 } 483 484 static int cmd_attr_deregister_cpumask(struct genl_info *info) 485 { 486 cpumask_var_t mask; 487 int rc; 488 489 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 490 return -ENOMEM; 491 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask); 492 if (rc < 0) 493 goto out; 494 rc = add_del_listener(info->snd_portid, mask, DEREGISTER); 495 out: 496 free_cpumask_var(mask); 497 return rc; 498 } 499 500 static size_t taskstats_packet_size(void) 501 { 502 size_t size; 503 504 size = nla_total_size(sizeof(u32)) + 505 nla_total_size(sizeof(struct taskstats)) + nla_total_size(0); 506 #ifdef TASKSTATS_NEEDS_PADDING 507 size += nla_total_size(0); /* Padding for alignment */ 508 #endif 509 return size; 510 } 511 512 static int cmd_attr_pid(struct genl_info *info) 513 { 514 struct taskstats *stats; 515 struct sk_buff *rep_skb; 516 size_t size; 517 u32 pid; 518 int rc; 519 520 size = taskstats_packet_size(); 521 522 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size); 523 if (rc < 0) 524 return rc; 525 526 rc = -EINVAL; 527 pid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_PID]); 528 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, pid); 529 if (!stats) 530 goto err; 531 532 rc = fill_stats_for_pid(pid, stats); 533 if (rc < 0) 534 goto err; 535 return send_reply(rep_skb, info); 536 err: 537 nlmsg_free(rep_skb); 538 return rc; 539 } 540 541 static int cmd_attr_tgid(struct genl_info *info) 542 { 543 struct taskstats *stats; 544 struct sk_buff *rep_skb; 545 size_t size; 546 u32 tgid; 547 int rc; 548 549 size = taskstats_packet_size(); 550 551 rc = prepare_reply(info, TASKSTATS_CMD_NEW, &rep_skb, size); 552 if (rc < 0) 553 return rc; 554 555 rc = -EINVAL; 556 tgid = nla_get_u32(info->attrs[TASKSTATS_CMD_ATTR_TGID]); 557 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tgid); 558 if (!stats) 559 goto err; 560 561 rc = fill_stats_for_tgid(tgid, stats); 562 if (rc < 0) 563 goto err; 564 return send_reply(rep_skb, info); 565 err: 566 nlmsg_free(rep_skb); 567 return rc; 568 } 569 570 static int taskstats_user_cmd(struct sk_buff *skb, struct genl_info *info) 571 { 572 if (info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK]) 573 return cmd_attr_register_cpumask(info); 574 else if (info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK]) 575 return cmd_attr_deregister_cpumask(info); 576 else if (info->attrs[TASKSTATS_CMD_ATTR_PID]) 577 return cmd_attr_pid(info); 578 else if (info->attrs[TASKSTATS_CMD_ATTR_TGID]) 579 return cmd_attr_tgid(info); 580 else 581 return -EINVAL; 582 } 583 584 static struct taskstats *taskstats_tgid_alloc(struct task_struct *tsk) 585 { 586 struct signal_struct *sig = tsk->signal; 587 struct taskstats *stats; 588 589 if (sig->stats || thread_group_empty(tsk)) 590 goto ret; 591 592 /* No problem if kmem_cache_zalloc() fails */ 593 stats = kmem_cache_zalloc(taskstats_cache, GFP_KERNEL); 594 595 spin_lock_irq(&tsk->sighand->siglock); 596 if (!sig->stats) { 597 sig->stats = stats; 598 stats = NULL; 599 } 600 spin_unlock_irq(&tsk->sighand->siglock); 601 602 if (stats) 603 kmem_cache_free(taskstats_cache, stats); 604 ret: 605 return sig->stats; 606 } 607 608 /* Send pid data out on exit */ 609 void taskstats_exit(struct task_struct *tsk, int group_dead) 610 { 611 int rc; 612 struct listener_list *listeners; 613 struct taskstats *stats; 614 struct sk_buff *rep_skb; 615 size_t size; 616 int is_thread_group; 617 618 if (!family_registered) 619 return; 620 621 /* 622 * Size includes space for nested attributes 623 */ 624 size = taskstats_packet_size(); 625 626 is_thread_group = !!taskstats_tgid_alloc(tsk); 627 if (is_thread_group) { 628 /* PID + STATS + TGID + STATS */ 629 size = 2 * size; 630 /* fill the tsk->signal->stats structure */ 631 fill_tgid_exit(tsk); 632 } 633 634 listeners = __this_cpu_ptr(&listener_array); 635 if (list_empty(&listeners->list)) 636 return; 637 638 rc = prepare_reply(NULL, TASKSTATS_CMD_NEW, &rep_skb, size); 639 if (rc < 0) 640 return; 641 642 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, 643 task_pid_nr_ns(tsk, &init_pid_ns)); 644 if (!stats) 645 goto err; 646 647 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats); 648 649 /* 650 * Doesn't matter if tsk is the leader or the last group member leaving 651 */ 652 if (!is_thread_group || !group_dead) 653 goto send; 654 655 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, 656 task_tgid_nr_ns(tsk, &init_pid_ns)); 657 if (!stats) 658 goto err; 659 660 memcpy(stats, tsk->signal->stats, sizeof(*stats)); 661 662 send: 663 send_cpu_listeners(rep_skb, listeners); 664 return; 665 err: 666 nlmsg_free(rep_skb); 667 } 668 669 static struct genl_ops taskstats_ops = { 670 .cmd = TASKSTATS_CMD_GET, 671 .doit = taskstats_user_cmd, 672 .policy = taskstats_cmd_get_policy, 673 .flags = GENL_ADMIN_PERM, 674 }; 675 676 static struct genl_ops cgroupstats_ops = { 677 .cmd = CGROUPSTATS_CMD_GET, 678 .doit = cgroupstats_user_cmd, 679 .policy = cgroupstats_cmd_get_policy, 680 }; 681 682 /* Needed early in initialization */ 683 void __init taskstats_init_early(void) 684 { 685 unsigned int i; 686 687 taskstats_cache = KMEM_CACHE(taskstats, SLAB_PANIC); 688 for_each_possible_cpu(i) { 689 INIT_LIST_HEAD(&(per_cpu(listener_array, i).list)); 690 init_rwsem(&(per_cpu(listener_array, i).sem)); 691 } 692 } 693 694 static int __init taskstats_init(void) 695 { 696 int rc; 697 698 rc = genl_register_family(&family); 699 if (rc) 700 return rc; 701 702 rc = genl_register_ops(&family, &taskstats_ops); 703 if (rc < 0) 704 goto err; 705 706 rc = genl_register_ops(&family, &cgroupstats_ops); 707 if (rc < 0) 708 goto err_cgroup_ops; 709 710 family_registered = 1; 711 pr_info("registered taskstats version %d\n", TASKSTATS_GENL_VERSION); 712 return 0; 713 err_cgroup_ops: 714 genl_unregister_ops(&family, &taskstats_ops); 715 err: 716 genl_unregister_family(&family); 717 return rc; 718 } 719 720 /* 721 * late initcall ensures initialization of statistics collection 722 * mechanisms precedes initialization of the taskstats interface 723 */ 724 late_initcall(taskstats_init); 725