1 // SPDX-License-Identifier: GPL-2.0 2 /* procacct.c 3 * 4 * Demonstrator of fetching resource data on task exit, as a way 5 * to accumulate accurate program resource usage statistics, without 6 * prior identification of the programs. For that, the fields for 7 * device and inode of the program executable binary file are also 8 * extracted in addition to the command string. 9 * 10 * The TGID together with the PID and the AGROUP flag allow 11 * identification of threads in a process and single-threaded processes. 12 * The ac_tgetime field gives proper whole-process walltime. 13 * 14 * Written (changed) by Thomas Orgis, University of Hamburg in 2022 15 * 16 * This is a cheap derivation (inheriting the style) of getdelays.c: 17 * 18 * Utility to get per-pid and per-tgid delay accounting statistics 19 * Also illustrates usage of the taskstats interface 20 * 21 * Copyright (C) Shailabh Nagar, IBM Corp. 2005 22 * Copyright (C) Balbir Singh, IBM Corp. 2006 23 * Copyright (c) Jay Lan, SGI. 2006 24 */ 25 26 #include <stdio.h> 27 #include <stdlib.h> 28 #include <errno.h> 29 #include <unistd.h> 30 #include <poll.h> 31 #include <string.h> 32 #include <fcntl.h> 33 #include <sys/types.h> 34 #include <sys/stat.h> 35 #include <sys/socket.h> 36 #include <sys/wait.h> 37 #include <signal.h> 38 39 #include <linux/genetlink.h> 40 #include <linux/acct.h> 41 #include <linux/taskstats.h> 42 #include <linux/kdev_t.h> 43 44 /* 45 * Generic macros for dealing with netlink sockets. Might be duplicated 46 * elsewhere. It is recommended that commercial grade applications use 47 * libnl or libnetlink and use the interfaces provided by the library 48 */ 49 #define GENLMSG_DATA(glh) ((void *)(NLMSG_DATA(glh) + GENL_HDRLEN)) 50 #define GENLMSG_PAYLOAD(glh) (NLMSG_PAYLOAD(glh, 0) - GENL_HDRLEN) 51 #define NLA_DATA(na) ((void *)((char *)(na) + NLA_HDRLEN)) 52 #define NLA_PAYLOAD(len) (len - NLA_HDRLEN) 53 54 #define err(code, fmt, arg...) \ 55 do { \ 56 fprintf(stderr, fmt, ##arg); \ 57 exit(code); \ 58 } while (0) 59 60 int rcvbufsz; 61 char name[100]; 62 int dbg; 63 int print_delays; 64 int print_io_accounting; 65 int print_task_context_switch_counts; 66 67 #define PRINTF(fmt, arg...) { \ 68 if (dbg) { \ 69 printf(fmt, ##arg); \ 70 } \ 71 } 72 73 /* Maximum size of response requested or message sent */ 74 #define MAX_MSG_SIZE 2048 75 /* Maximum number of cpus expected to be specified in a cpumask */ 76 #define MAX_CPUS 32 77 78 struct msgtemplate { 79 struct nlmsghdr n; 80 struct genlmsghdr g; 81 char buf[MAX_MSG_SIZE]; 82 }; 83 84 char cpumask[100+6*MAX_CPUS]; 85 86 static void usage(void) 87 { 88 fprintf(stderr, "procacct [-v] [-w logfile] [-r bufsize] [-m cpumask]\n"); 89 fprintf(stderr, " -v: debug on\n"); 90 } 91 92 /* 93 * Create a raw netlink socket and bind 94 */ 95 static int create_nl_socket(int protocol) 96 { 97 int fd; 98 struct sockaddr_nl local; 99 100 fd = socket(AF_NETLINK, SOCK_RAW, protocol); 101 if (fd < 0) 102 return -1; 103 104 if (rcvbufsz) 105 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, 106 &rcvbufsz, sizeof(rcvbufsz)) < 0) { 107 fprintf(stderr, "Unable to set socket rcv buf size to %d\n", 108 rcvbufsz); 109 goto error; 110 } 111 112 memset(&local, 0, sizeof(local)); 113 local.nl_family = AF_NETLINK; 114 115 if (bind(fd, (struct sockaddr *) &local, sizeof(local)) < 0) 116 goto error; 117 118 return fd; 119 error: 120 close(fd); 121 return -1; 122 } 123 124 static int recv_taskstats_msg(int sd, struct msgtemplate *msg) 125 { 126 struct sockaddr_nl nladdr; 127 struct iovec iov = { 128 .iov_base = msg, 129 .iov_len = sizeof(*msg), 130 }; 131 struct msghdr hdr = { 132 .msg_name = &nladdr, 133 .msg_namelen = sizeof(nladdr), 134 .msg_iov = &iov, 135 .msg_iovlen = 1, 136 }; 137 int ret; 138 139 ret = recvmsg(sd, &hdr, 0); 140 if (ret < 0) 141 return -1; 142 if (hdr.msg_flags & MSG_TRUNC) { 143 errno = EMSGSIZE; 144 return -1; 145 } 146 147 return ret; 148 } 149 150 151 static int send_cmd(int sd, __u16 nlmsg_type, __u32 nlmsg_pid, 152 __u8 genl_cmd, __u16 nla_type, 153 void *nla_data, int nla_len) 154 { 155 struct nlattr *na; 156 struct sockaddr_nl nladdr; 157 int r, buflen; 158 char *buf; 159 160 struct msgtemplate msg; 161 162 msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN); 163 msg.n.nlmsg_type = nlmsg_type; 164 msg.n.nlmsg_flags = NLM_F_REQUEST; 165 msg.n.nlmsg_seq = 0; 166 msg.n.nlmsg_pid = nlmsg_pid; 167 msg.g.cmd = genl_cmd; 168 msg.g.version = 0x1; 169 na = (struct nlattr *) GENLMSG_DATA(&msg); 170 na->nla_type = nla_type; 171 na->nla_len = nla_len + 1 + NLA_HDRLEN; 172 memcpy(NLA_DATA(na), nla_data, nla_len); 173 msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len); 174 175 buf = (char *) &msg; 176 buflen = msg.n.nlmsg_len; 177 memset(&nladdr, 0, sizeof(nladdr)); 178 nladdr.nl_family = AF_NETLINK; 179 while ((r = sendto(sd, buf, buflen, 0, (struct sockaddr *) &nladdr, 180 sizeof(nladdr))) < buflen) { 181 if (r > 0) { 182 buf += r; 183 buflen -= r; 184 } else if (errno != EAGAIN) 185 return -1; 186 } 187 return 0; 188 } 189 190 191 /* 192 * Probe the controller in genetlink to find the family id 193 * for the TASKSTATS family 194 */ 195 static int get_family_id(int sd) 196 { 197 struct { 198 struct nlmsghdr n; 199 struct genlmsghdr g; 200 char buf[256]; 201 } ans; 202 203 int id = 0, rc; 204 struct nlattr *na; 205 int rep_len; 206 207 strcpy(name, TASKSTATS_GENL_NAME); 208 rc = send_cmd(sd, GENL_ID_CTRL, getpid(), CTRL_CMD_GETFAMILY, 209 CTRL_ATTR_FAMILY_NAME, (void *)name, 210 strlen(TASKSTATS_GENL_NAME)+1); 211 if (rc < 0) 212 return 0; /* sendto() failure? */ 213 214 rep_len = recv(sd, &ans, sizeof(ans), 0); 215 if (ans.n.nlmsg_type == NLMSG_ERROR || 216 (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len)) 217 return 0; 218 219 na = (struct nlattr *) GENLMSG_DATA(&ans); 220 na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len)); 221 if (na->nla_type == CTRL_ATTR_FAMILY_ID) 222 id = *(__u16 *) NLA_DATA(na); 223 224 return id; 225 } 226 227 #define average_ms(t, c) (t / 1000000ULL / (c ? c : 1)) 228 229 static void print_procacct(struct taskstats *t) 230 { 231 /* First letter: T is a mere thread, G the last in a group, U unknown. */ 232 printf( 233 "%c pid=%lu tgid=%lu uid=%lu wall=%llu gwall=%llu cpu=%llu vmpeak=%llu rsspeak=%llu dev=%lu:%lu inode=%llu comm=%s\n" 234 , t->version >= 12 ? (t->ac_flag & AGROUP ? 'P' : 'T') : '?' 235 , (unsigned long)t->ac_pid 236 , (unsigned long)(t->version >= 12 ? t->ac_tgid : 0) 237 , (unsigned long)t->ac_uid 238 , (unsigned long long)t->ac_etime 239 , (unsigned long long)(t->version >= 12 ? t->ac_tgetime : 0) 240 , (unsigned long long)(t->ac_utime+t->ac_stime) 241 , (unsigned long long)t->hiwater_vm 242 , (unsigned long long)t->hiwater_rss 243 , (unsigned long)(t->version >= 12 ? MAJOR(t->ac_exe_dev) : 0) 244 , (unsigned long)(t->version >= 12 ? MINOR(t->ac_exe_dev) : 0) 245 , (unsigned long long)(t->version >= 12 ? t->ac_exe_inode : 0) 246 , t->ac_comm 247 ); 248 } 249 250 void handle_aggr(int mother, struct nlattr *na, int fd) 251 { 252 int aggr_len = NLA_PAYLOAD(na->nla_len); 253 int len2 = 0; 254 pid_t rtid = 0; 255 256 na = (struct nlattr *) NLA_DATA(na); 257 while (len2 < aggr_len) { 258 switch (na->nla_type) { 259 case TASKSTATS_TYPE_PID: 260 rtid = *(int *) NLA_DATA(na); 261 PRINTF("PID\t%d\n", rtid); 262 break; 263 case TASKSTATS_TYPE_TGID: 264 rtid = *(int *) NLA_DATA(na); 265 PRINTF("TGID\t%d\n", rtid); 266 break; 267 case TASKSTATS_TYPE_STATS: 268 PRINTF("version %u\n", 269 ((struct taskstats *)NLA_DATA(na))->version); 270 if (mother == TASKSTATS_TYPE_AGGR_PID) 271 print_procacct((struct taskstats *) NLA_DATA(na)); 272 if (fd) { 273 if (write(fd, NLA_DATA(na), na->nla_len) < 0) 274 err(1, "write error\n"); 275 } 276 break; 277 case TASKSTATS_TYPE_NULL: 278 break; 279 default: 280 fprintf(stderr, "Unknown nested nla_type %d\n", 281 na->nla_type); 282 break; 283 } 284 len2 += NLA_ALIGN(na->nla_len); 285 na = (struct nlattr *)((char *)na + 286 NLA_ALIGN(na->nla_len)); 287 } 288 } 289 290 int main(int argc, char *argv[]) 291 { 292 int c, rc, rep_len; 293 __u16 id; 294 __u32 mypid; 295 296 struct nlattr *na; 297 int nl_sd = -1; 298 int len = 0; 299 300 int fd = 0; 301 int write_file = 0; 302 int maskset = 0; 303 char *logfile = NULL; 304 int cfd = 0; 305 306 struct msgtemplate msg; 307 308 while (1) { 309 c = getopt(argc, argv, "m:vr:w:"); 310 if (c < 0) 311 break; 312 313 switch (c) { 314 case 'w': 315 logfile = strdup(optarg); 316 printf("write to file %s\n", logfile); 317 write_file = 1; 318 break; 319 case 'r': 320 rcvbufsz = atoi(optarg); 321 printf("receive buf size %d\n", rcvbufsz); 322 if (rcvbufsz < 0) 323 err(1, "Invalid rcv buf size\n"); 324 break; 325 case 'm': 326 strncpy(cpumask, optarg, sizeof(cpumask)); 327 cpumask[sizeof(cpumask) - 1] = '\0'; 328 maskset = 1; 329 break; 330 case 'v': 331 printf("debug on\n"); 332 dbg = 1; 333 break; 334 default: 335 usage(); 336 exit(-1); 337 } 338 } 339 if (!maskset) { 340 maskset = 1; 341 strncpy(cpumask, "1", sizeof(cpumask)); 342 cpumask[sizeof(cpumask) - 1] = '\0'; 343 } 344 printf("cpumask %s maskset %d\n", cpumask, maskset); 345 346 if (write_file) { 347 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, 0644); 348 if (fd == -1) { 349 perror("Cannot open output file\n"); 350 exit(1); 351 } 352 } 353 354 nl_sd = create_nl_socket(NETLINK_GENERIC); 355 if (nl_sd < 0) 356 err(1, "error creating Netlink socket\n"); 357 358 mypid = getpid(); 359 id = get_family_id(nl_sd); 360 if (!id) { 361 fprintf(stderr, "Error getting family id, errno %d\n", errno); 362 goto err; 363 } 364 PRINTF("family id %d\n", id); 365 366 if (maskset) { 367 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 368 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK, 369 &cpumask, strlen(cpumask) + 1); 370 PRINTF("Sent register cpumask, retval %d\n", rc); 371 if (rc < 0) { 372 fprintf(stderr, "error sending register cpumask\n"); 373 goto err; 374 } 375 } 376 377 do { 378 rep_len = recv_taskstats_msg(nl_sd, &msg); 379 PRINTF("received %d bytes\n", rep_len); 380 381 if (rep_len < 0) { 382 if (errno == EMSGSIZE) 383 fprintf(stderr, 384 "dropped truncated taskstats netlink message, please increase MAX_MSG_SIZE\n"); 385 else 386 fprintf(stderr, "nonfatal reply error: errno %d\n", 387 errno); 388 continue; 389 } 390 if (msg.n.nlmsg_type == NLMSG_ERROR || 391 !NLMSG_OK((&msg.n), rep_len)) { 392 struct nlmsgerr *err = NLMSG_DATA(&msg); 393 394 fprintf(stderr, "fatal reply error, errno %d\n", 395 err->error); 396 goto done; 397 } 398 399 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n", 400 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len); 401 402 403 rep_len = GENLMSG_PAYLOAD(&msg.n); 404 405 na = (struct nlattr *) GENLMSG_DATA(&msg); 406 len = 0; 407 while (len < rep_len) { 408 len += NLA_ALIGN(na->nla_len); 409 int mother = na->nla_type; 410 411 PRINTF("mother=%i\n", mother); 412 switch (na->nla_type) { 413 case TASKSTATS_TYPE_AGGR_PID: 414 case TASKSTATS_TYPE_AGGR_TGID: 415 /* For nested attributes, na follows */ 416 handle_aggr(mother, na, fd); 417 break; 418 default: 419 fprintf(stderr, "Unexpected nla_type %d\n", 420 na->nla_type); 421 case TASKSTATS_TYPE_NULL: 422 break; 423 } 424 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len); 425 } 426 } while (1); 427 done: 428 if (maskset) { 429 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 430 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK, 431 &cpumask, strlen(cpumask) + 1); 432 printf("Sent deregister mask, retval %d\n", rc); 433 if (rc < 0) 434 err(rc, "error sending deregister cpumask\n"); 435 } 436 err: 437 close(nl_sd); 438 if (fd) 439 close(fd); 440 if (cfd) 441 close(cfd); 442 return 0; 443 } 444