xref: /linux/tools/accounting/procacct.c (revision 3d0fe49454652117522f60bfbefb978ba0e5300b)
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;
265 	__u16 id;
266 	__u32 mypid;
267 
268 	struct nlattr *na;
269 	int nl_sd = -1;
270 	int len = 0;
271 
272 	int fd = 0;
273 	int write_file = 0;
274 	int maskset = 0;
275 	char *logfile = NULL;
276 	int cfd = 0;
277 	int forking = 0;
278 
279 	struct msgtemplate msg;
280 
281 	while (!forking) {
282 		c = getopt(argc, argv, "m:vr:");
283 		if (c < 0)
284 			break;
285 
286 		switch (c) {
287 		case 'w':
288 			logfile = strdup(optarg);
289 			printf("write to file %s\n", logfile);
290 			write_file = 1;
291 			break;
292 		case 'r':
293 			rcvbufsz = atoi(optarg);
294 			printf("receive buf size %d\n", rcvbufsz);
295 			if (rcvbufsz < 0)
296 				err(1, "Invalid rcv buf size\n");
297 			break;
298 		case 'm':
299 			strncpy(cpumask, optarg, sizeof(cpumask));
300 			cpumask[sizeof(cpumask) - 1] = '\0';
301 			maskset = 1;
302 			break;
303 		case 'v':
304 			printf("debug on\n");
305 			dbg = 1;
306 			break;
307 		default:
308 			usage();
309 			exit(-1);
310 		}
311 	}
312 	if (!maskset) {
313 		maskset = 1;
314 		strncpy(cpumask, "1", sizeof(cpumask));
315 		cpumask[sizeof(cpumask) - 1] = '\0';
316 	}
317 	printf("cpumask %s maskset %d\n", cpumask, maskset);
318 
319 	if (write_file) {
320 		fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
321 		if (fd == -1) {
322 			perror("Cannot open output file\n");
323 			exit(1);
324 		}
325 	}
326 
327 	nl_sd = create_nl_socket(NETLINK_GENERIC);
328 	if (nl_sd < 0)
329 		err(1, "error creating Netlink socket\n");
330 
331 	mypid = getpid();
332 	id = get_family_id(nl_sd);
333 	if (!id) {
334 		fprintf(stderr, "Error getting family id, errno %d\n", errno);
335 		goto err;
336 	}
337 	PRINTF("family id %d\n", id);
338 
339 	if (maskset) {
340 		rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
341 			      TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
342 			      &cpumask, strlen(cpumask) + 1);
343 		PRINTF("Sent register cpumask, retval %d\n", rc);
344 		if (rc < 0) {
345 			fprintf(stderr, "error sending register cpumask\n");
346 			goto err;
347 		}
348 	}
349 
350 	do {
351 		rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
352 		PRINTF("received %d bytes\n", rep_len);
353 
354 		if (rep_len < 0) {
355 			fprintf(stderr, "nonfatal reply error: errno %d\n",
356 				errno);
357 			continue;
358 		}
359 		if (msg.n.nlmsg_type == NLMSG_ERROR ||
360 		    !NLMSG_OK((&msg.n), rep_len)) {
361 			struct nlmsgerr *err = NLMSG_DATA(&msg);
362 
363 			fprintf(stderr, "fatal reply error,  errno %d\n",
364 				err->error);
365 			goto done;
366 		}
367 
368 		PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
369 		       sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
370 
371 
372 		rep_len = GENLMSG_PAYLOAD(&msg.n);
373 
374 		na = (struct nlattr *) GENLMSG_DATA(&msg);
375 		len = 0;
376 		while (len < rep_len) {
377 			len += NLA_ALIGN(na->nla_len);
378 			int mother = na->nla_type;
379 
380 			PRINTF("mother=%i\n", mother);
381 			switch (na->nla_type) {
382 			case TASKSTATS_TYPE_AGGR_PID:
383 			case TASKSTATS_TYPE_AGGR_TGID:
384 				/* For nested attributes, na follows */
385 				handle_aggr(mother, na, fd);
386 				break;
387 			default:
388 				fprintf(stderr, "Unexpected nla_type %d\n",
389 					na->nla_type);
390 			case TASKSTATS_TYPE_NULL:
391 				break;
392 			}
393 			na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
394 		}
395 	} while (1);
396 done:
397 	if (maskset) {
398 		rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
399 			      TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
400 			      &cpumask, strlen(cpumask) + 1);
401 		printf("Sent deregister mask, retval %d\n", rc);
402 		if (rc < 0)
403 			err(rc, "error sending deregister cpumask\n");
404 	}
405 err:
406 	close(nl_sd);
407 	if (fd)
408 		close(fd);
409 	if (cfd)
410 		close(cfd);
411 	return 0;
412 }
413