1/* getdelays.c 2 * 3 * Utility to get per-pid and per-tgid delay accounting statistics 4 * Also illustrates usage of the taskstats interface 5 * 6 * Copyright (C) Shailabh Nagar, IBM Corp. 2005 7 * Copyright (C) Balbir Singh, IBM Corp. 2006 8 * Copyright (c) Jay Lan, SGI. 2006 9 * 10 * Compile with 11 * gcc -I/usr/src/linux/include getdelays.c -o getdelays 12 */ 13 14#include <stdio.h> 15#include <stdlib.h> 16#include <errno.h> 17#include <unistd.h> 18#include <poll.h> 19#include <string.h> 20#include <fcntl.h> 21#include <sys/types.h> 22#include <sys/stat.h> 23#include <sys/socket.h> 24#include <signal.h> 25 26#include <linux/genetlink.h> 27#include <linux/taskstats.h> 28#include <linux/cgroupstats.h> 29 30/* 31 * Generic macros for dealing with netlink sockets. Might be duplicated 32 * elsewhere. It is recommended that commercial grade applications use 33 * libnl or libnetlink and use the interfaces provided by the library 34 */ 35#define GENLMSG_DATA(glh) ((void *)(NLMSG_DATA(glh) + GENL_HDRLEN)) 36#define GENLMSG_PAYLOAD(glh) (NLMSG_PAYLOAD(glh, 0) - GENL_HDRLEN) 37#define NLA_DATA(na) ((void *)((char*)(na) + NLA_HDRLEN)) 38#define NLA_PAYLOAD(len) (len - NLA_HDRLEN) 39 40#define err(code, fmt, arg...) \ 41 do { \ 42 fprintf(stderr, fmt, ##arg); \ 43 exit(code); \ 44 } while (0) 45 46int done; 47int rcvbufsz; 48char name[100]; 49int dbg; 50int print_delays; 51int print_io_accounting; 52int print_task_context_switch_counts; 53__u64 stime, utime; 54 55#define PRINTF(fmt, arg...) { \ 56 if (dbg) { \ 57 printf(fmt, ##arg); \ 58 } \ 59 } 60 61/* Maximum size of response requested or message sent */ 62#define MAX_MSG_SIZE 1024 63/* Maximum number of cpus expected to be specified in a cpumask */ 64#define MAX_CPUS 32 65 66struct msgtemplate { 67 struct nlmsghdr n; 68 struct genlmsghdr g; 69 char buf[MAX_MSG_SIZE]; 70}; 71 72char cpumask[100+6*MAX_CPUS]; 73 74static void usage(void) 75{ 76 fprintf(stderr, "getdelays [-dilv] [-w logfile] [-r bufsize] " 77 "[-m cpumask] [-t tgid] [-p pid]\n"); 78 fprintf(stderr, " -d: print delayacct stats\n"); 79 fprintf(stderr, " -i: print IO accounting (works only with -p)\n"); 80 fprintf(stderr, " -l: listen forever\n"); 81 fprintf(stderr, " -v: debug on\n"); 82 fprintf(stderr, " -C: container path\n"); 83} 84 85/* 86 * Create a raw netlink socket and bind 87 */ 88static int create_nl_socket(int protocol) 89{ 90 int fd; 91 struct sockaddr_nl local; 92 93 fd = socket(AF_NETLINK, SOCK_RAW, protocol); 94 if (fd < 0) 95 return -1; 96 97 if (rcvbufsz) 98 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, 99 &rcvbufsz, sizeof(rcvbufsz)) < 0) { 100 fprintf(stderr, "Unable to set socket rcv buf size " 101 "to %d\n", 102 rcvbufsz); 103 return -1; 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; 113error: 114 close(fd); 115 return -1; 116} 117 118 119static 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 */ 163static 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); 179 180 rep_len = recv(sd, &ans, sizeof(ans), 0); 181 if (ans.n.nlmsg_type == NLMSG_ERROR || 182 (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len)) 183 return 0; 184 185 na = (struct nlattr *) GENLMSG_DATA(&ans); 186 na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len)); 187 if (na->nla_type == CTRL_ATTR_FAMILY_ID) { 188 id = *(__u16 *) NLA_DATA(na); 189 } 190 return id; 191} 192 193static void print_delayacct(struct taskstats *t) 194{ 195 printf("\n\nCPU %15s%15s%15s%15s\n" 196 " %15llu%15llu%15llu%15llu\n" 197 "IO %15s%15s\n" 198 " %15llu%15llu\n" 199 "SWAP %15s%15s\n" 200 " %15llu%15llu\n" 201 "RECLAIM %12s%15s\n" 202 " %15llu%15llu\n", 203 "count", "real total", "virtual total", "delay total", 204 (unsigned long long)t->cpu_count, 205 (unsigned long long)t->cpu_run_real_total, 206 (unsigned long long)t->cpu_run_virtual_total, 207 (unsigned long long)t->cpu_delay_total, 208 "count", "delay total", 209 (unsigned long long)t->blkio_count, 210 (unsigned long long)t->blkio_delay_total, 211 "count", "delay total", 212 (unsigned long long)t->swapin_count, 213 (unsigned long long)t->swapin_delay_total, 214 "count", "delay total", 215 (unsigned long long)t->freepages_count, 216 (unsigned long long)t->freepages_delay_total); 217} 218 219static void task_context_switch_counts(struct taskstats *t) 220{ 221 printf("\n\nTask %15s%15s\n" 222 " %15llu%15llu\n", 223 "voluntary", "nonvoluntary", 224 (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw); 225} 226 227static void print_cgroupstats(struct cgroupstats *c) 228{ 229 printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, " 230 "uninterruptible %llu\n", (unsigned long long)c->nr_sleeping, 231 (unsigned long long)c->nr_io_wait, 232 (unsigned long long)c->nr_running, 233 (unsigned long long)c->nr_stopped, 234 (unsigned long long)c->nr_uninterruptible); 235} 236 237 238static void print_ioacct(struct taskstats *t) 239{ 240 printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n", 241 t->ac_comm, 242 (unsigned long long)t->read_bytes, 243 (unsigned long long)t->write_bytes, 244 (unsigned long long)t->cancelled_write_bytes); 245} 246 247int main(int argc, char *argv[]) 248{ 249 int c, rc, rep_len, aggr_len, len2; 250 int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC; 251 __u16 id; 252 __u32 mypid; 253 254 struct nlattr *na; 255 int nl_sd = -1; 256 int len = 0; 257 pid_t tid = 0; 258 pid_t rtid = 0; 259 260 int fd = 0; 261 int count = 0; 262 int write_file = 0; 263 int maskset = 0; 264 char *logfile = NULL; 265 int loop = 0; 266 int containerset = 0; 267 char containerpath[1024]; 268 int cfd = 0; 269 270 struct msgtemplate msg; 271 272 while (1) { 273 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:"); 274 if (c < 0) 275 break; 276 277 switch (c) { 278 case 'd': 279 printf("print delayacct stats ON\n"); 280 print_delays = 1; 281 break; 282 case 'i': 283 printf("printing IO accounting\n"); 284 print_io_accounting = 1; 285 break; 286 case 'q': 287 printf("printing task/process context switch rates\n"); 288 print_task_context_switch_counts = 1; 289 break; 290 case 'C': 291 containerset = 1; 292 strncpy(containerpath, optarg, strlen(optarg) + 1); 293 break; 294 case 'w': 295 logfile = strdup(optarg); 296 printf("write to file %s\n", logfile); 297 write_file = 1; 298 break; 299 case 'r': 300 rcvbufsz = atoi(optarg); 301 printf("receive buf size %d\n", rcvbufsz); 302 if (rcvbufsz < 0) 303 err(1, "Invalid rcv buf size\n"); 304 break; 305 case 'm': 306 strncpy(cpumask, optarg, sizeof(cpumask)); 307 maskset = 1; 308 printf("cpumask %s maskset %d\n", cpumask, maskset); 309 break; 310 case 't': 311 tid = atoi(optarg); 312 if (!tid) 313 err(1, "Invalid tgid\n"); 314 cmd_type = TASKSTATS_CMD_ATTR_TGID; 315 break; 316 case 'p': 317 tid = atoi(optarg); 318 if (!tid) 319 err(1, "Invalid pid\n"); 320 cmd_type = TASKSTATS_CMD_ATTR_PID; 321 break; 322 case 'v': 323 printf("debug on\n"); 324 dbg = 1; 325 break; 326 case 'l': 327 printf("listen forever\n"); 328 loop = 1; 329 break; 330 default: 331 usage(); 332 exit(-1); 333 } 334 } 335 336 if (write_file) { 337 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, 338 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH); 339 if (fd == -1) { 340 perror("Cannot open output file\n"); 341 exit(1); 342 } 343 } 344 345 if ((nl_sd = create_nl_socket(NETLINK_GENERIC)) < 0) 346 err(1, "error creating Netlink socket\n"); 347 348 349 mypid = getpid(); 350 id = get_family_id(nl_sd); 351 if (!id) { 352 fprintf(stderr, "Error getting family id, errno %d\n", errno); 353 goto err; 354 } 355 PRINTF("family id %d\n", id); 356 357 if (maskset) { 358 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 359 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK, 360 &cpumask, strlen(cpumask) + 1); 361 PRINTF("Sent register cpumask, retval %d\n", rc); 362 if (rc < 0) { 363 fprintf(stderr, "error sending register cpumask\n"); 364 goto err; 365 } 366 } 367 368 if (tid && containerset) { 369 fprintf(stderr, "Select either -t or -C, not both\n"); 370 goto err; 371 } 372 373 if (tid) { 374 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 375 cmd_type, &tid, sizeof(__u32)); 376 PRINTF("Sent pid/tgid, retval %d\n", rc); 377 if (rc < 0) { 378 fprintf(stderr, "error sending tid/tgid cmd\n"); 379 goto done; 380 } 381 } 382 383 if (containerset) { 384 cfd = open(containerpath, O_RDONLY); 385 if (cfd < 0) { 386 perror("error opening container file"); 387 goto err; 388 } 389 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET, 390 CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32)); 391 if (rc < 0) { 392 perror("error sending cgroupstats command"); 393 goto err; 394 } 395 } 396 if (!maskset && !tid && !containerset) { 397 usage(); 398 goto err; 399 } 400 401 do { 402 int i; 403 404 rep_len = recv(nl_sd, &msg, sizeof(msg), 0); 405 PRINTF("received %d bytes\n", rep_len); 406 407 if (rep_len < 0) { 408 fprintf(stderr, "nonfatal reply error: errno %d\n", 409 errno); 410 continue; 411 } 412 if (msg.n.nlmsg_type == NLMSG_ERROR || 413 !NLMSG_OK((&msg.n), rep_len)) { 414 struct nlmsgerr *err = NLMSG_DATA(&msg); 415 fprintf(stderr, "fatal reply error, errno %d\n", 416 err->error); 417 goto done; 418 } 419 420 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n", 421 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len); 422 423 424 rep_len = GENLMSG_PAYLOAD(&msg.n); 425 426 na = (struct nlattr *) GENLMSG_DATA(&msg); 427 len = 0; 428 i = 0; 429 while (len < rep_len) { 430 len += NLA_ALIGN(na->nla_len); 431 switch (na->nla_type) { 432 case TASKSTATS_TYPE_AGGR_TGID: 433 /* Fall through */ 434 case TASKSTATS_TYPE_AGGR_PID: 435 aggr_len = NLA_PAYLOAD(na->nla_len); 436 len2 = 0; 437 /* For nested attributes, na follows */ 438 na = (struct nlattr *) NLA_DATA(na); 439 done = 0; 440 while (len2 < aggr_len) { 441 switch (na->nla_type) { 442 case TASKSTATS_TYPE_PID: 443 rtid = *(int *) NLA_DATA(na); 444 if (print_delays) 445 printf("PID\t%d\n", rtid); 446 break; 447 case TASKSTATS_TYPE_TGID: 448 rtid = *(int *) NLA_DATA(na); 449 if (print_delays) 450 printf("TGID\t%d\n", rtid); 451 break; 452 case TASKSTATS_TYPE_STATS: 453 count++; 454 if (print_delays) 455 print_delayacct((struct taskstats *) NLA_DATA(na)); 456 if (print_io_accounting) 457 print_ioacct((struct taskstats *) NLA_DATA(na)); 458 if (print_task_context_switch_counts) 459 task_context_switch_counts((struct taskstats *) NLA_DATA(na)); 460 if (fd) { 461 if (write(fd, NLA_DATA(na), na->nla_len) < 0) { 462 err(1,"write error\n"); 463 } 464 } 465 if (!loop) 466 goto done; 467 break; 468 default: 469 fprintf(stderr, "Unknown nested" 470 " nla_type %d\n", 471 na->nla_type); 472 break; 473 } 474 len2 += NLA_ALIGN(na->nla_len); 475 na = (struct nlattr *) ((char *) na + len2); 476 } 477 break; 478 479 case CGROUPSTATS_TYPE_CGROUP_STATS: 480 print_cgroupstats(NLA_DATA(na)); 481 break; 482 default: 483 fprintf(stderr, "Unknown nla_type %d\n", 484 na->nla_type); 485 break; 486 } 487 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len); 488 } 489 } while (loop); 490done: 491 if (maskset) { 492 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 493 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK, 494 &cpumask, strlen(cpumask) + 1); 495 printf("Sent deregister mask, retval %d\n", rc); 496 if (rc < 0) 497 err(rc, "error sending deregister cpumask\n"); 498 } 499err: 500 close(nl_sd); 501 if (fd) 502 close(fd); 503 if (cfd) 504 close(cfd); 505 return 0; 506} 507