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
48int rcvbufsz;
49char name[100];
50int dbg;
51int print_delays;
52int print_io_accounting;
53int print_task_context_switch_counts;
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 to %d\n",
101				rcvbufsz);
102			goto error;
103		}
104
105	memset(&local, 0, sizeof(local));
106	local.nl_family = AF_NETLINK;
107
108	if (bind(fd, (struct sockaddr *) &local, sizeof(local)) < 0)
109		goto error;
110
111	return fd;
112error:
113	close(fd);
114	return -1;
115}
116
117
118static int send_cmd(int sd, __u16 nlmsg_type, __u32 nlmsg_pid,
119	     __u8 genl_cmd, __u16 nla_type,
120	     void *nla_data, int nla_len)
121{
122	struct nlattr *na;
123	struct sockaddr_nl nladdr;
124	int r, buflen;
125	char *buf;
126
127	struct msgtemplate msg;
128
129	msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN);
130	msg.n.nlmsg_type = nlmsg_type;
131	msg.n.nlmsg_flags = NLM_F_REQUEST;
132	msg.n.nlmsg_seq = 0;
133	msg.n.nlmsg_pid = nlmsg_pid;
134	msg.g.cmd = genl_cmd;
135	msg.g.version = 0x1;
136	na = (struct nlattr *) GENLMSG_DATA(&msg);
137	na->nla_type = nla_type;
138	na->nla_len = nla_len + NLA_HDRLEN;
139	memcpy(NLA_DATA(na), nla_data, nla_len);
140	msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len);
141
142	buf = (char *) &msg;
143	buflen = msg.n.nlmsg_len ;
144	memset(&nladdr, 0, sizeof(nladdr));
145	nladdr.nl_family = AF_NETLINK;
146	while ((r = sendto(sd, buf, buflen, 0, (struct sockaddr *) &nladdr,
147			   sizeof(nladdr))) < buflen) {
148		if (r > 0) {
149			buf += r;
150			buflen -= r;
151		} else if (errno != EAGAIN)
152			return -1;
153	}
154	return 0;
155}
156
157
158/*
159 * Probe the controller in genetlink to find the family id
160 * for the TASKSTATS family
161 */
162static int get_family_id(int sd)
163{
164	struct {
165		struct nlmsghdr n;
166		struct genlmsghdr g;
167		char buf[256];
168	} ans;
169
170	int id = 0, rc;
171	struct nlattr *na;
172	int rep_len;
173
174	strcpy(name, TASKSTATS_GENL_NAME);
175	rc = send_cmd(sd, GENL_ID_CTRL, getpid(), CTRL_CMD_GETFAMILY,
176			CTRL_ATTR_FAMILY_NAME, (void *)name,
177			strlen(TASKSTATS_GENL_NAME)+1);
178	if (rc < 0)
179		return 0;	/* sendto() failure? */
180
181	rep_len = recv(sd, &ans, sizeof(ans), 0);
182	if (ans.n.nlmsg_type == NLMSG_ERROR ||
183	    (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len))
184		return 0;
185
186	na = (struct nlattr *) GENLMSG_DATA(&ans);
187	na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len));
188	if (na->nla_type == CTRL_ATTR_FAMILY_ID) {
189		id = *(__u16 *) NLA_DATA(na);
190	}
191	return id;
192}
193
194#define average_ms(t, c) (t / 1000000ULL / (c ? c : 1))
195
196static void print_delayacct(struct taskstats *t)
197{
198	printf("\n\nCPU   %15s%15s%15s%15s%15s\n"
199	       "      %15llu%15llu%15llu%15llu%15.3fms\n"
200	       "IO    %15s%15s%15s\n"
201	       "      %15llu%15llu%15.3fms\n"
202	       "SWAP  %15s%15s%15s\n"
203	       "      %15llu%15llu%15.3fms\n"
204	       "RECLAIM  %12s%15s%15s\n"
205	       "      %15llu%15llu%15.3fms\n"
206	       "THRASHING%12s%15s%15s\n"
207	       "      %15llu%15llu%15.3fms\n"
208	       "COMPACT  %12s%15s%15s\n"
209	       "      %15llu%15llu%15.3fms\n"
210	       "WPCOPY   %12s%15s%15s\n"
211	       "      %15llu%15llu%15.3fms\n"
212	       "IRQ   %15s%15s%15s\n"
213	       "      %15llu%15llu%15.3fms\n",
214	       "count", "real total", "virtual total",
215	       "delay total", "delay average",
216	       (unsigned long long)t->cpu_count,
217	       (unsigned long long)t->cpu_run_real_total,
218	       (unsigned long long)t->cpu_run_virtual_total,
219	       (unsigned long long)t->cpu_delay_total,
220	       average_ms((double)t->cpu_delay_total, t->cpu_count),
221	       "count", "delay total", "delay average",
222	       (unsigned long long)t->blkio_count,
223	       (unsigned long long)t->blkio_delay_total,
224	       average_ms((double)t->blkio_delay_total, t->blkio_count),
225	       "count", "delay total", "delay average",
226	       (unsigned long long)t->swapin_count,
227	       (unsigned long long)t->swapin_delay_total,
228	       average_ms((double)t->swapin_delay_total, t->swapin_count),
229	       "count", "delay total", "delay average",
230	       (unsigned long long)t->freepages_count,
231	       (unsigned long long)t->freepages_delay_total,
232	       average_ms((double)t->freepages_delay_total, t->freepages_count),
233	       "count", "delay total", "delay average",
234	       (unsigned long long)t->thrashing_count,
235	       (unsigned long long)t->thrashing_delay_total,
236	       average_ms((double)t->thrashing_delay_total, t->thrashing_count),
237	       "count", "delay total", "delay average",
238	       (unsigned long long)t->compact_count,
239	       (unsigned long long)t->compact_delay_total,
240	       average_ms((double)t->compact_delay_total, t->compact_count),
241	       "count", "delay total", "delay average",
242	       (unsigned long long)t->wpcopy_count,
243	       (unsigned long long)t->wpcopy_delay_total,
244	       average_ms((double)t->wpcopy_delay_total, t->wpcopy_count),
245	       "count", "delay total", "delay average",
246	       (unsigned long long)t->irq_count,
247	       (unsigned long long)t->irq_delay_total,
248	       average_ms((double)t->irq_delay_total, t->irq_count));
249}
250
251static void task_context_switch_counts(struct taskstats *t)
252{
253	printf("\n\nTask   %15s%15s\n"
254	       "       %15llu%15llu\n",
255	       "voluntary", "nonvoluntary",
256	       (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw);
257}
258
259static void print_cgroupstats(struct cgroupstats *c)
260{
261	printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
262		"uninterruptible %llu\n", (unsigned long long)c->nr_sleeping,
263		(unsigned long long)c->nr_io_wait,
264		(unsigned long long)c->nr_running,
265		(unsigned long long)c->nr_stopped,
266		(unsigned long long)c->nr_uninterruptible);
267}
268
269
270static void print_ioacct(struct taskstats *t)
271{
272	printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n",
273		t->ac_comm,
274		(unsigned long long)t->read_bytes,
275		(unsigned long long)t->write_bytes,
276		(unsigned long long)t->cancelled_write_bytes);
277}
278
279int main(int argc, char *argv[])
280{
281	int c, rc, rep_len, aggr_len, len2;
282	int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC;
283	__u16 id;
284	__u32 mypid;
285
286	struct nlattr *na;
287	int nl_sd = -1;
288	int len = 0;
289	pid_t tid = 0;
290	pid_t rtid = 0;
291
292	int fd = 0;
293	int write_file = 0;
294	int maskset = 0;
295	char *logfile = NULL;
296	int loop = 0;
297	int containerset = 0;
298	char *containerpath = NULL;
299	int cfd = 0;
300	int forking = 0;
301	sigset_t sigset;
302
303	struct msgtemplate msg;
304
305	while (!forking) {
306		c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:c:");
307		if (c < 0)
308			break;
309
310		switch (c) {
311		case 'd':
312			printf("print delayacct stats ON\n");
313			print_delays = 1;
314			break;
315		case 'i':
316			printf("printing IO accounting\n");
317			print_io_accounting = 1;
318			break;
319		case 'q':
320			printf("printing task/process context switch rates\n");
321			print_task_context_switch_counts = 1;
322			break;
323		case 'C':
324			containerset = 1;
325			containerpath = optarg;
326			break;
327		case 'w':
328			logfile = strdup(optarg);
329			printf("write to file %s\n", logfile);
330			write_file = 1;
331			break;
332		case 'r':
333			rcvbufsz = atoi(optarg);
334			printf("receive buf size %d\n", rcvbufsz);
335			if (rcvbufsz < 0)
336				err(1, "Invalid rcv buf size\n");
337			break;
338		case 'm':
339			strncpy(cpumask, optarg, sizeof(cpumask));
340			cpumask[sizeof(cpumask) - 1] = '\0';
341			maskset = 1;
342			printf("cpumask %s maskset %d\n", cpumask, maskset);
343			break;
344		case 't':
345			tid = atoi(optarg);
346			if (!tid)
347				err(1, "Invalid tgid\n");
348			cmd_type = TASKSTATS_CMD_ATTR_TGID;
349			break;
350		case 'p':
351			tid = atoi(optarg);
352			if (!tid)
353				err(1, "Invalid pid\n");
354			cmd_type = TASKSTATS_CMD_ATTR_PID;
355			break;
356		case 'c':
357
358			/* Block SIGCHLD for sigwait() later */
359			if (sigemptyset(&sigset) == -1)
360				err(1, "Failed to empty sigset");
361			if (sigaddset(&sigset, SIGCHLD))
362				err(1, "Failed to set sigchld in sigset");
363			sigprocmask(SIG_BLOCK, &sigset, NULL);
364
365			/* fork/exec a child */
366			tid = fork();
367			if (tid < 0)
368				err(1, "Fork failed\n");
369			if (tid == 0)
370				if (execvp(argv[optind - 1],
371				    &argv[optind - 1]) < 0)
372					exit(-1);
373
374			/* Set the command type and avoid further processing */
375			cmd_type = TASKSTATS_CMD_ATTR_PID;
376			forking = 1;
377			break;
378		case 'v':
379			printf("debug on\n");
380			dbg = 1;
381			break;
382		case 'l':
383			printf("listen forever\n");
384			loop = 1;
385			break;
386		default:
387			usage();
388			exit(-1);
389		}
390	}
391
392	if (write_file) {
393		fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC,
394			  S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
395		if (fd == -1) {
396			perror("Cannot open output file\n");
397			exit(1);
398		}
399	}
400
401	nl_sd = create_nl_socket(NETLINK_GENERIC);
402	if (nl_sd < 0)
403		err(1, "error creating Netlink socket\n");
404
405
406	mypid = getpid();
407	id = get_family_id(nl_sd);
408	if (!id) {
409		fprintf(stderr, "Error getting family id, errno %d\n", errno);
410		goto err;
411	}
412	PRINTF("family id %d\n", id);
413
414	if (maskset) {
415		rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
416			      TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
417			      &cpumask, strlen(cpumask) + 1);
418		PRINTF("Sent register cpumask, retval %d\n", rc);
419		if (rc < 0) {
420			fprintf(stderr, "error sending register cpumask\n");
421			goto err;
422		}
423	}
424
425	if (tid && containerset) {
426		fprintf(stderr, "Select either -t or -C, not both\n");
427		goto err;
428	}
429
430	/*
431	 * If we forked a child, wait for it to exit. Cannot use waitpid()
432	 * as all the delicious data would be reaped as part of the wait
433	 */
434	if (tid && forking) {
435		int sig_received;
436		sigwait(&sigset, &sig_received);
437	}
438
439	if (tid) {
440		rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
441			      cmd_type, &tid, sizeof(__u32));
442		PRINTF("Sent pid/tgid, retval %d\n", rc);
443		if (rc < 0) {
444			fprintf(stderr, "error sending tid/tgid cmd\n");
445			goto done;
446		}
447	}
448
449	if (containerset) {
450		cfd = open(containerpath, O_RDONLY);
451		if (cfd < 0) {
452			perror("error opening container file");
453			goto err;
454		}
455		rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET,
456			      CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32));
457		if (rc < 0) {
458			perror("error sending cgroupstats command");
459			goto err;
460		}
461	}
462	if (!maskset && !tid && !containerset) {
463		usage();
464		goto err;
465	}
466
467	do {
468		rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
469		PRINTF("received %d bytes\n", rep_len);
470
471		if (rep_len < 0) {
472			fprintf(stderr, "nonfatal reply error: errno %d\n",
473				errno);
474			continue;
475		}
476		if (msg.n.nlmsg_type == NLMSG_ERROR ||
477		    !NLMSG_OK((&msg.n), rep_len)) {
478			struct nlmsgerr *err = NLMSG_DATA(&msg);
479			fprintf(stderr, "fatal reply error,  errno %d\n",
480				err->error);
481			goto done;
482		}
483
484		PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
485		       sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
486
487
488		rep_len = GENLMSG_PAYLOAD(&msg.n);
489
490		na = (struct nlattr *) GENLMSG_DATA(&msg);
491		len = 0;
492		while (len < rep_len) {
493			len += NLA_ALIGN(na->nla_len);
494			switch (na->nla_type) {
495			case TASKSTATS_TYPE_AGGR_TGID:
496				/* Fall through */
497			case TASKSTATS_TYPE_AGGR_PID:
498				aggr_len = NLA_PAYLOAD(na->nla_len);
499				len2 = 0;
500				/* For nested attributes, na follows */
501				na = (struct nlattr *) NLA_DATA(na);
502				while (len2 < aggr_len) {
503					switch (na->nla_type) {
504					case TASKSTATS_TYPE_PID:
505						rtid = *(int *) NLA_DATA(na);
506						if (print_delays)
507							printf("PID\t%d\n", rtid);
508						break;
509					case TASKSTATS_TYPE_TGID:
510						rtid = *(int *) NLA_DATA(na);
511						if (print_delays)
512							printf("TGID\t%d\n", rtid);
513						break;
514					case TASKSTATS_TYPE_STATS:
515						if (print_delays)
516							print_delayacct((struct taskstats *) NLA_DATA(na));
517						if (print_io_accounting)
518							print_ioacct((struct taskstats *) NLA_DATA(na));
519						if (print_task_context_switch_counts)
520							task_context_switch_counts((struct taskstats *) NLA_DATA(na));
521						if (fd) {
522							if (write(fd, NLA_DATA(na), na->nla_len) < 0) {
523								err(1,"write error\n");
524							}
525						}
526						if (!loop)
527							goto done;
528						break;
529					case TASKSTATS_TYPE_NULL:
530						break;
531					default:
532						fprintf(stderr, "Unknown nested"
533							" nla_type %d\n",
534							na->nla_type);
535						break;
536					}
537					len2 += NLA_ALIGN(na->nla_len);
538					na = (struct nlattr *)((char *)na +
539							       NLA_ALIGN(na->nla_len));
540				}
541				break;
542
543			case CGROUPSTATS_TYPE_CGROUP_STATS:
544				print_cgroupstats(NLA_DATA(na));
545				break;
546			default:
547				fprintf(stderr, "Unknown nla_type %d\n",
548					na->nla_type);
549			case TASKSTATS_TYPE_NULL:
550				break;
551			}
552			na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
553		}
554	} while (loop);
555done:
556	if (maskset) {
557		rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
558			      TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
559			      &cpumask, strlen(cpumask) + 1);
560		printf("Sent deregister mask, retval %d\n", rc);
561		if (rc < 0)
562			err(rc, "error sending deregister cpumask\n");
563	}
564err:
565	close(nl_sd);
566	if (fd)
567		close(fd);
568	if (cfd)
569		close(cfd);
570	return 0;
571}
572