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 1024 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 } 123 124 125 static int send_cmd(int sd, __u16 nlmsg_type, __u32 nlmsg_pid, 126 __u8 genl_cmd, __u16 nla_type, 127 void *nla_data, int nla_len) 128 { 129 struct nlattr *na; 130 struct sockaddr_nl nladdr; 131 int r, buflen; 132 char *buf; 133 134 struct msgtemplate msg; 135 136 msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN); 137 msg.n.nlmsg_type = nlmsg_type; 138 msg.n.nlmsg_flags = NLM_F_REQUEST; 139 msg.n.nlmsg_seq = 0; 140 msg.n.nlmsg_pid = nlmsg_pid; 141 msg.g.cmd = genl_cmd; 142 msg.g.version = 0x1; 143 na = (struct nlattr *) GENLMSG_DATA(&msg); 144 na->nla_type = nla_type; 145 na->nla_len = nla_len + 1 + NLA_HDRLEN; 146 memcpy(NLA_DATA(na), nla_data, nla_len); 147 msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len); 148 149 buf = (char *) &msg; 150 buflen = msg.n.nlmsg_len; 151 memset(&nladdr, 0, sizeof(nladdr)); 152 nladdr.nl_family = AF_NETLINK; 153 while ((r = sendto(sd, buf, buflen, 0, (struct sockaddr *) &nladdr, 154 sizeof(nladdr))) < buflen) { 155 if (r > 0) { 156 buf += r; 157 buflen -= r; 158 } else if (errno != EAGAIN) 159 return -1; 160 } 161 return 0; 162 } 163 164 165 /* 166 * Probe the controller in genetlink to find the family id 167 * for the TASKSTATS family 168 */ 169 static int get_family_id(int sd) 170 { 171 struct { 172 struct nlmsghdr n; 173 struct genlmsghdr g; 174 char buf[256]; 175 } ans; 176 177 int id = 0, rc; 178 struct nlattr *na; 179 int rep_len; 180 181 strcpy(name, TASKSTATS_GENL_NAME); 182 rc = send_cmd(sd, GENL_ID_CTRL, getpid(), CTRL_CMD_GETFAMILY, 183 CTRL_ATTR_FAMILY_NAME, (void *)name, 184 strlen(TASKSTATS_GENL_NAME)+1); 185 if (rc < 0) 186 return 0; /* sendto() failure? */ 187 188 rep_len = recv(sd, &ans, sizeof(ans), 0); 189 if (ans.n.nlmsg_type == NLMSG_ERROR || 190 (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len)) 191 return 0; 192 193 na = (struct nlattr *) GENLMSG_DATA(&ans); 194 na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len)); 195 if (na->nla_type == CTRL_ATTR_FAMILY_ID) 196 id = *(__u16 *) NLA_DATA(na); 197 198 return id; 199 } 200 201 #define average_ms(t, c) (t / 1000000ULL / (c ? c : 1)) 202 203 static void print_procacct(struct taskstats *t) 204 { 205 /* First letter: T is a mere thread, G the last in a group, U unknown. */ 206 printf( 207 "%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" 208 , t->version >= 12 ? (t->ac_flag & AGROUP ? 'P' : 'T') : '?' 209 , (unsigned long)t->ac_pid 210 , (unsigned long)(t->version >= 12 ? t->ac_tgid : 0) 211 , (unsigned long)t->ac_uid 212 , (unsigned long long)t->ac_etime 213 , (unsigned long long)(t->version >= 12 ? t->ac_tgetime : 0) 214 , (unsigned long long)(t->ac_utime+t->ac_stime) 215 , (unsigned long long)t->hiwater_vm 216 , (unsigned long long)t->hiwater_rss 217 , (unsigned long)(t->version >= 12 ? MAJOR(t->ac_exe_dev) : 0) 218 , (unsigned long)(t->version >= 12 ? MINOR(t->ac_exe_dev) : 0) 219 , (unsigned long long)(t->version >= 12 ? t->ac_exe_inode : 0) 220 , t->ac_comm 221 ); 222 } 223 224 void handle_aggr(int mother, struct nlattr *na, int fd) 225 { 226 int aggr_len = NLA_PAYLOAD(na->nla_len); 227 int len2 = 0; 228 pid_t rtid = 0; 229 230 na = (struct nlattr *) NLA_DATA(na); 231 while (len2 < aggr_len) { 232 switch (na->nla_type) { 233 case TASKSTATS_TYPE_PID: 234 rtid = *(int *) NLA_DATA(na); 235 PRINTF("PID\t%d\n", rtid); 236 break; 237 case TASKSTATS_TYPE_TGID: 238 rtid = *(int *) NLA_DATA(na); 239 PRINTF("TGID\t%d\n", rtid); 240 break; 241 case TASKSTATS_TYPE_STATS: 242 if (mother == TASKSTATS_TYPE_AGGR_PID) 243 print_procacct((struct taskstats *) NLA_DATA(na)); 244 if (fd) { 245 if (write(fd, NLA_DATA(na), na->nla_len) < 0) 246 err(1, "write error\n"); 247 } 248 break; 249 case TASKSTATS_TYPE_NULL: 250 break; 251 default: 252 fprintf(stderr, "Unknown nested nla_type %d\n", 253 na->nla_type); 254 break; 255 } 256 len2 += NLA_ALIGN(na->nla_len); 257 na = (struct nlattr *)((char *)na + 258 NLA_ALIGN(na->nla_len)); 259 } 260 } 261 262 int main(int argc, char *argv[]) 263 { 264 int c, rc, rep_len, aggr_len, len2; 265 int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC; 266 __u16 id; 267 __u32 mypid; 268 269 struct nlattr *na; 270 int nl_sd = -1; 271 int len = 0; 272 pid_t tid = 0; 273 274 int fd = 0; 275 int write_file = 0; 276 int maskset = 0; 277 char *logfile = NULL; 278 int containerset = 0; 279 char *containerpath = NULL; 280 int cfd = 0; 281 int forking = 0; 282 sigset_t sigset; 283 284 struct msgtemplate msg; 285 286 while (!forking) { 287 c = getopt(argc, argv, "m:vr:"); 288 if (c < 0) 289 break; 290 291 switch (c) { 292 case 'w': 293 logfile = strdup(optarg); 294 printf("write to file %s\n", logfile); 295 write_file = 1; 296 break; 297 case 'r': 298 rcvbufsz = atoi(optarg); 299 printf("receive buf size %d\n", rcvbufsz); 300 if (rcvbufsz < 0) 301 err(1, "Invalid rcv buf size\n"); 302 break; 303 case 'm': 304 strncpy(cpumask, optarg, sizeof(cpumask)); 305 cpumask[sizeof(cpumask) - 1] = '\0'; 306 maskset = 1; 307 break; 308 case 'v': 309 printf("debug on\n"); 310 dbg = 1; 311 break; 312 default: 313 usage(); 314 exit(-1); 315 } 316 } 317 if (!maskset) { 318 maskset = 1; 319 strncpy(cpumask, "1", sizeof(cpumask)); 320 cpumask[sizeof(cpumask) - 1] = '\0'; 321 } 322 printf("cpumask %s maskset %d\n", cpumask, maskset); 323 324 if (write_file) { 325 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, 0644); 326 if (fd == -1) { 327 perror("Cannot open output file\n"); 328 exit(1); 329 } 330 } 331 332 nl_sd = create_nl_socket(NETLINK_GENERIC); 333 if (nl_sd < 0) 334 err(1, "error creating Netlink socket\n"); 335 336 mypid = getpid(); 337 id = get_family_id(nl_sd); 338 if (!id) { 339 fprintf(stderr, "Error getting family id, errno %d\n", errno); 340 goto err; 341 } 342 PRINTF("family id %d\n", id); 343 344 if (maskset) { 345 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 346 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK, 347 &cpumask, strlen(cpumask) + 1); 348 PRINTF("Sent register cpumask, retval %d\n", rc); 349 if (rc < 0) { 350 fprintf(stderr, "error sending register cpumask\n"); 351 goto err; 352 } 353 } 354 355 do { 356 rep_len = recv(nl_sd, &msg, sizeof(msg), 0); 357 PRINTF("received %d bytes\n", rep_len); 358 359 if (rep_len < 0) { 360 fprintf(stderr, "nonfatal reply error: errno %d\n", 361 errno); 362 continue; 363 } 364 if (msg.n.nlmsg_type == NLMSG_ERROR || 365 !NLMSG_OK((&msg.n), rep_len)) { 366 struct nlmsgerr *err = NLMSG_DATA(&msg); 367 368 fprintf(stderr, "fatal reply error, errno %d\n", 369 err->error); 370 goto done; 371 } 372 373 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n", 374 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len); 375 376 377 rep_len = GENLMSG_PAYLOAD(&msg.n); 378 379 na = (struct nlattr *) GENLMSG_DATA(&msg); 380 len = 0; 381 while (len < rep_len) { 382 len += NLA_ALIGN(na->nla_len); 383 int mother = na->nla_type; 384 385 PRINTF("mother=%i\n", mother); 386 switch (na->nla_type) { 387 case TASKSTATS_TYPE_AGGR_PID: 388 case TASKSTATS_TYPE_AGGR_TGID: 389 /* For nested attributes, na follows */ 390 handle_aggr(mother, na, fd); 391 break; 392 default: 393 fprintf(stderr, "Unexpected nla_type %d\n", 394 na->nla_type); 395 case TASKSTATS_TYPE_NULL: 396 break; 397 } 398 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len); 399 } 400 } while (1); 401 done: 402 if (maskset) { 403 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, 404 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK, 405 &cpumask, strlen(cpumask) + 1); 406 printf("Sent deregister mask, retval %d\n", rc); 407 if (rc < 0) 408 err(rc, "error sending deregister cpumask\n"); 409 } 410 err: 411 close(nl_sd); 412 if (fd) 413 close(fd); 414 if (cfd) 415 close(cfd); 416 return 0; 417 } 418