1 // SPDX-License-Identifier: GPL-2.0 2 /* getdelays.c 3 * 4 * Utility to get per-pid and per-tgid delay accounting statistics 5 * Also illustrates usage of the taskstats interface 6 * 7 * Copyright (C) Shailabh Nagar, IBM Corp. 2005 8 * Copyright (C) Balbir Singh, IBM Corp. 2006 9 * Copyright (c) Jay Lan, SGI. 2006 10 * 11 * Compile with 12 * gcc -I/usr/src/linux/include getdelays.c -o getdelays 13 */ 14 15 #include <stdio.h> 16 #include <stdlib.h> 17 #include <errno.h> 18 #include <unistd.h> 19 #include <poll.h> 20 #include <string.h> 21 #include <fcntl.h> 22 #include <sys/types.h> 23 #include <sys/stat.h> 24 #include <sys/socket.h> 25 #include <sys/wait.h> 26 #include <signal.h> 27 28 #include <linux/genetlink.h> 29 #include <linux/taskstats.h> 30 #include <linux/cgroupstats.h> 31 32 /* 33 * Generic macros for dealing with netlink sockets. Might be duplicated 34 * elsewhere. It is recommended that commercial grade applications use 35 * libnl or libnetlink and use the interfaces provided by the library 36 */ 37 #define GENLMSG_DATA(glh) ((void *)(NLMSG_DATA(glh) + GENL_HDRLEN)) 38 #define GENLMSG_PAYLOAD(glh) (NLMSG_PAYLOAD(glh, 0) - GENL_HDRLEN) 39 #define NLA_DATA(na) ((void *)((char*)(na) + NLA_HDRLEN)) 40 #define NLA_PAYLOAD(len) (len - NLA_HDRLEN) 41 42 #define err(code, fmt, arg...) \ 43 do { \ 44 fprintf(stderr, fmt, ##arg); \ 45 exit(code); \ 46 } while (0) 47 48 int done; 49 int rcvbufsz; 50 char name[100]; 51 int dbg; 52 int print_delays; 53 int print_io_accounting; 54 int print_task_context_switch_counts; 55 56 #define PRINTF(fmt, arg...) { \ 57 if (dbg) { \ 58 printf(fmt, ##arg); \ 59 } \ 60 } 61 62 /* Maximum size of response requested or message sent */ 63 #define MAX_MSG_SIZE 1024 64 /* Maximum number of cpus expected to be specified in a cpumask */ 65 #define MAX_CPUS 32 66 67 struct msgtemplate { 68 struct nlmsghdr n; 69 struct genlmsghdr g; 70 char buf[MAX_MSG_SIZE]; 71 }; 72 73 char cpumask[100+6*MAX_CPUS]; 74 75 static void usage(void) 76 { 77 fprintf(stderr, "getdelays [-dilv] [-w logfile] [-r bufsize] " 78 "[-m cpumask] [-t tgid] [-p pid]\n"); 79 fprintf(stderr, " -d: print delayacct stats\n"); 80 fprintf(stderr, " -i: print IO accounting (works only with -p)\n"); 81 fprintf(stderr, " -l: listen forever\n"); 82 fprintf(stderr, " -v: debug on\n"); 83 fprintf(stderr, " -C: container path\n"); 84 } 85 86 /* 87 * Create a raw netlink socket and bind 88 */ 89 static int create_nl_socket(int protocol) 90 { 91 int fd; 92 struct sockaddr_nl local; 93 94 fd = socket(AF_NETLINK, SOCK_RAW, protocol); 95 if (fd < 0) 96 return -1; 97 98 if (rcvbufsz) 99 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, 100 &rcvbufsz, sizeof(rcvbufsz)) < 0) { 101 fprintf(stderr, "Unable to set socket rcv buf size to %d\n", 102 rcvbufsz); 103 goto error; 104 } 105 106 memset(&local, 0, sizeof(local)); 107 local.nl_family = AF_NETLINK; 108 109 if (bind(fd, (struct sockaddr *) &local, sizeof(local)) < 0) 110 goto error; 111 112 return fd; 113 error: 114 close(fd); 115 return -1; 116 } 117 118 119 static int send_cmd(int sd, __u16 nlmsg_type, __u32 nlmsg_pid, 120 __u8 genl_cmd, __u16 nla_type, 121 void *nla_data, int nla_len) 122 { 123 struct nlattr *na; 124 struct sockaddr_nl nladdr; 125 int r, buflen; 126 char *buf; 127 128 struct msgtemplate msg; 129 130 msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN); 131 msg.n.nlmsg_type = nlmsg_type; 132 msg.n.nlmsg_flags = NLM_F_REQUEST; 133 msg.n.nlmsg_seq = 0; 134 msg.n.nlmsg_pid = nlmsg_pid; 135 msg.g.cmd = genl_cmd; 136 msg.g.version = 0x1; 137 na = (struct nlattr *) GENLMSG_DATA(&msg); 138 na->nla_type = nla_type; 139 na->nla_len = nla_len + 1 + NLA_HDRLEN; 140 memcpy(NLA_DATA(na), nla_data, nla_len); 141 msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len); 142 143 buf = (char *) &msg; 144 buflen = msg.n.nlmsg_len ; 145 memset(&nladdr, 0, sizeof(nladdr)); 146 nladdr.nl_family = AF_NETLINK; 147 while ((r = sendto(sd, buf, buflen, 0, (struct sockaddr *) &nladdr, 148 sizeof(nladdr))) < buflen) { 149 if (r > 0) { 150 buf += r; 151 buflen -= r; 152 } else if (errno != EAGAIN) 153 return -1; 154 } 155 return 0; 156 } 157 158 159 /* 160 * Probe the controller in genetlink to find the family id 161 * for the TASKSTATS family 162 */ 163 static int get_family_id(int sd) 164 { 165 struct { 166 struct nlmsghdr n; 167 struct genlmsghdr g; 168 char buf[256]; 169 } ans; 170 171 int id = 0, rc; 172 struct nlattr *na; 173 int rep_len; 174 175 strcpy(name, TASKSTATS_GENL_NAME); 176 rc = send_cmd(sd, GENL_ID_CTRL, getpid(), CTRL_CMD_GETFAMILY, 177 CTRL_ATTR_FAMILY_NAME, (void *)name, 178 strlen(TASKSTATS_GENL_NAME)+1); 179 if (rc < 0) 180 return 0; /* sendto() failure? */ 181 182 rep_len = recv(sd, &ans, sizeof(ans), 0); 183 if (ans.n.nlmsg_type == NLMSG_ERROR || 184 (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len)) 185 return 0; 186 187 na = (struct nlattr *) GENLMSG_DATA(&ans); 188 na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len)); 189 if (na->nla_type == CTRL_ATTR_FAMILY_ID) { 190 id = *(__u16 *) NLA_DATA(na); 191 } 192 return id; 193 } 194 195 #define average_ms(t, c) (t / 1000000ULL / (c ? c : 1)) 196 197 static void print_delayacct(struct taskstats *t) 198 { 199 printf("\n\nCPU %15s%15s%15s%15s%15s\n" 200 " %15llu%15llu%15llu%15llu%15.3fms\n" 201 "IO %15s%15s%15s\n" 202 " %15llu%15llu%15llums\n" 203 "SWAP %15s%15s%15s\n" 204 " %15llu%15llu%15llums\n" 205 "RECLAIM %12s%15s%15s\n" 206 " %15llu%15llu%15llums\n", 207 "count", "real total", "virtual total", 208 "delay total", "delay average", 209 (unsigned long long)t->cpu_count, 210 (unsigned long long)t->cpu_run_real_total, 211 (unsigned long long)t->cpu_run_virtual_total, 212 (unsigned long long)t->cpu_delay_total, 213 average_ms((double)t->cpu_delay_total, t->cpu_count), 214 "count", "delay total", "delay average", 215 (unsigned long long)t->blkio_count, 216 (unsigned long long)t->blkio_delay_total, 217 average_ms(t->blkio_delay_total, t->blkio_count), 218 "count", "delay total", "delay average", 219 (unsigned long long)t->swapin_count, 220 (unsigned long long)t->swapin_delay_total, 221 average_ms(t->swapin_delay_total, t->swapin_count), 222 "count", "delay total", "delay average", 223 (unsigned long long)t->freepages_count, 224 (unsigned long long)t->freepages_delay_total, 225 average_ms(t->freepages_delay_total, t->freepages_count)); 226 } 227 228 static void task_context_switch_counts(struct taskstats *t) 229 { 230 printf("\n\nTask %15s%15s\n" 231 " %15llu%15llu\n", 232 "voluntary", "nonvoluntary", 233 (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw); 234 } 235 236 static void print_cgroupstats(struct cgroupstats *c) 237 { 238 printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, " 239 "uninterruptible %llu\n", (unsigned long long)c->nr_sleeping, 240 (unsigned long long)c->nr_io_wait, 241 (unsigned long long)c->nr_running, 242 (unsigned long long)c->nr_stopped, 243 (unsigned long long)c->nr_uninterruptible); 244 } 245 246 247 static void print_ioacct(struct taskstats *t) 248 { 249 printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n", 250 t->ac_comm, 251 (unsigned long long)t->read_bytes, 252 (unsigned long long)t->write_bytes, 253 (unsigned long long)t->cancelled_write_bytes); 254 } 255 256 int main(int argc, char *argv[]) 257 { 258 int c, rc, rep_len, aggr_len, len2; 259 int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC; 260 __u16 id; 261 __u32 mypid; 262 263 struct nlattr *na; 264 int nl_sd = -1; 265 int len = 0; 266 pid_t tid = 0; 267 pid_t rtid = 0; 268 269 int fd = 0; 270 int count = 0; 271 int write_file = 0; 272 int maskset = 0; 273 char *logfile = NULL; 274 int loop = 0; 275 int containerset = 0; 276 char *containerpath = NULL; 277 int cfd = 0; 278 int forking = 0; 279 sigset_t sigset; 280 281 struct msgtemplate msg; 282 283 while (!forking) { 284 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:c:"); 285 if (c < 0) 286 break; 287 288 switch (c) { 289 case 'd': 290 printf("print delayacct stats ON\n"); 291 print_delays = 1; 292 break; 293 case 'i': 294 printf("printing IO accounting\n"); 295 print_io_accounting = 1; 296 break; 297 case 'q': 298 printf("printing task/process context switch rates\n"); 299 print_task_context_switch_counts = 1; 300 break; 301 case 'C': 302 containerset = 1; 303 containerpath = optarg; 304 break; 305 case 'w': 306 logfile = strdup(optarg); 307 printf("write to file %s\n", logfile); 308 write_file = 1; 309 break; 310 case 'r': 311 rcvbufsz = atoi(optarg); 312 printf("receive buf size %d\n", rcvbufsz); 313 if (rcvbufsz < 0) 314 err(1, "Invalid rcv buf size\n"); 315 break; 316 case 'm': 317 strncpy(cpumask, optarg, sizeof(cpumask)); 318 cpumask[sizeof(cpumask) - 1] = '\0'; 319 maskset = 1; 320 printf("cpumask %s maskset %d\n", cpumask, maskset); 321 break; 322 case 't': 323 tid = atoi(optarg); 324 if (!tid) 325 err(1, "Invalid tgid\n"); 326 cmd_type = TASKSTATS_CMD_ATTR_TGID; 327 break; 328 case 'p': 329 tid = atoi(optarg); 330 if (!tid) 331 err(1, "Invalid pid\n"); 332 cmd_type = TASKSTATS_CMD_ATTR_PID; 333 break; 334 case 'c': 335 336 /* Block SIGCHLD for sigwait() later */ 337 if (sigemptyset(&sigset) == -1) 338 err(1, "Failed to empty sigset"); 339 if (sigaddset(&sigset, SIGCHLD)) 340 err(1, "Failed to set sigchld in sigset"); 341 sigprocmask(SIG_BLOCK, &sigset, NULL); 342 343 /* fork/exec a child */ 344 tid = fork(); 345 if (tid < 0) 346 err(1, "Fork failed\n"); 347 if (tid == 0) 348 if (execvp(argv[optind - 1], 349 &argv[optind - 1]) < 0) 350 exit(-1); 351 352 /* Set the command type and avoid further processing */ 353 cmd_type = TASKSTATS_CMD_ATTR_PID; 354 forking = 1; 355 break; 356 case 'v': 357 printf("debug on\n"); 358 dbg = 1; 359 break; 360 case 'l': 361 printf("listen forever\n"); 362 loop = 1; 363 break; 364 default: 365 usage(); 366 exit(-1); 367 } 368 } 369 370 if (write_file) { 371 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, 372 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH); 373 if (fd == -1) { 374 perror("Cannot open output file\n"); 375 exit(1); 376 } 377 } 378 379 nl_sd = create_nl_socket(NETLINK_GENERIC); 380 if (nl_sd < 0) 381 err(1, "error creating Netlink socket\n"); 382 383 384 mypid = getpid(); 385 id = get_family_id(nl_sd); 386 if (!id) { 387 fprintf(stderr, "Error getting family id, errno %d\n", errno); 388 goto err; 389 } 390 PRINTF("family id %d\n", id); 391 392 if (maskset) { 393 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 394 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK, 395 &cpumask, strlen(cpumask) + 1); 396 PRINTF("Sent register cpumask, retval %d\n", rc); 397 if (rc < 0) { 398 fprintf(stderr, "error sending register cpumask\n"); 399 goto err; 400 } 401 } 402 403 if (tid && containerset) { 404 fprintf(stderr, "Select either -t or -C, not both\n"); 405 goto err; 406 } 407 408 /* 409 * If we forked a child, wait for it to exit. Cannot use waitpid() 410 * as all the delicious data would be reaped as part of the wait 411 */ 412 if (tid && forking) { 413 int sig_received; 414 sigwait(&sigset, &sig_received); 415 } 416 417 if (tid) { 418 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 419 cmd_type, &tid, sizeof(__u32)); 420 PRINTF("Sent pid/tgid, retval %d\n", rc); 421 if (rc < 0) { 422 fprintf(stderr, "error sending tid/tgid cmd\n"); 423 goto done; 424 } 425 } 426 427 if (containerset) { 428 cfd = open(containerpath, O_RDONLY); 429 if (cfd < 0) { 430 perror("error opening container file"); 431 goto err; 432 } 433 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET, 434 CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32)); 435 if (rc < 0) { 436 perror("error sending cgroupstats command"); 437 goto err; 438 } 439 } 440 if (!maskset && !tid && !containerset) { 441 usage(); 442 goto err; 443 } 444 445 do { 446 rep_len = recv(nl_sd, &msg, sizeof(msg), 0); 447 PRINTF("received %d bytes\n", rep_len); 448 449 if (rep_len < 0) { 450 fprintf(stderr, "nonfatal reply error: errno %d\n", 451 errno); 452 continue; 453 } 454 if (msg.n.nlmsg_type == NLMSG_ERROR || 455 !NLMSG_OK((&msg.n), rep_len)) { 456 struct nlmsgerr *err = NLMSG_DATA(&msg); 457 fprintf(stderr, "fatal reply error, errno %d\n", 458 err->error); 459 goto done; 460 } 461 462 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n", 463 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len); 464 465 466 rep_len = GENLMSG_PAYLOAD(&msg.n); 467 468 na = (struct nlattr *) GENLMSG_DATA(&msg); 469 len = 0; 470 while (len < rep_len) { 471 len += NLA_ALIGN(na->nla_len); 472 switch (na->nla_type) { 473 case TASKSTATS_TYPE_AGGR_TGID: 474 /* Fall through */ 475 case TASKSTATS_TYPE_AGGR_PID: 476 aggr_len = NLA_PAYLOAD(na->nla_len); 477 len2 = 0; 478 /* For nested attributes, na follows */ 479 na = (struct nlattr *) NLA_DATA(na); 480 done = 0; 481 while (len2 < aggr_len) { 482 switch (na->nla_type) { 483 case TASKSTATS_TYPE_PID: 484 rtid = *(int *) NLA_DATA(na); 485 if (print_delays) 486 printf("PID\t%d\n", rtid); 487 break; 488 case TASKSTATS_TYPE_TGID: 489 rtid = *(int *) NLA_DATA(na); 490 if (print_delays) 491 printf("TGID\t%d\n", rtid); 492 break; 493 case TASKSTATS_TYPE_STATS: 494 count++; 495 if (print_delays) 496 print_delayacct((struct taskstats *) NLA_DATA(na)); 497 if (print_io_accounting) 498 print_ioacct((struct taskstats *) NLA_DATA(na)); 499 if (print_task_context_switch_counts) 500 task_context_switch_counts((struct taskstats *) NLA_DATA(na)); 501 if (fd) { 502 if (write(fd, NLA_DATA(na), na->nla_len) < 0) { 503 err(1,"write error\n"); 504 } 505 } 506 if (!loop) 507 goto done; 508 break; 509 case TASKSTATS_TYPE_NULL: 510 break; 511 default: 512 fprintf(stderr, "Unknown nested" 513 " nla_type %d\n", 514 na->nla_type); 515 break; 516 } 517 len2 += NLA_ALIGN(na->nla_len); 518 na = (struct nlattr *)((char *)na + 519 NLA_ALIGN(na->nla_len)); 520 } 521 break; 522 523 case CGROUPSTATS_TYPE_CGROUP_STATS: 524 print_cgroupstats(NLA_DATA(na)); 525 break; 526 default: 527 fprintf(stderr, "Unknown nla_type %d\n", 528 na->nla_type); 529 case TASKSTATS_TYPE_NULL: 530 break; 531 } 532 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len); 533 } 534 } while (loop); 535 done: 536 if (maskset) { 537 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 538 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK, 539 &cpumask, strlen(cpumask) + 1); 540 printf("Sent deregister mask, retval %d\n", rc); 541 if (rc < 0) 542 err(rc, "error sending deregister cpumask\n"); 543 } 544 err: 545 close(nl_sd); 546 if (fd) 547 close(fd); 548 if (cfd) 549 close(cfd); 550 return 0; 551 } 552