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
48 int rcvbufsz;
49 char name[100];
50 int dbg;
51 int print_delays;
52 int print_io_accounting;
53 int 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
66 struct msgtemplate {
67 struct nlmsghdr n;
68 struct genlmsghdr g;
69 char buf[MAX_MSG_SIZE];
70 };
71
72 char cpumask[100+6*MAX_CPUS];
73
usage(void)74 static 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 */
create_nl_socket(int protocol)88 static 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;
112 error:
113 close(fd);
114 return -1;
115 }
116
117
send_cmd(int sd,__u16 nlmsg_type,__u32 nlmsg_pid,__u8 genl_cmd,__u16 nla_type,void * nla_data,int nla_len)118 static 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 */
get_family_id(int sd)162 static 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 #define delay_ms(t) (t / 1000000ULL)
196
print_delayacct(struct taskstats * t)197 static void print_delayacct(struct taskstats *t)
198 {
199 printf("\n\nCPU %15s%15s%15s%15s%15s%15s%15s\n"
200 " %15llu%15llu%15llu%15llu%15.3fms%13.6fms%13.6fms\n"
201 "IO %15s%15s%15s%15s%15s\n"
202 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
203 "SWAP %15s%15s%15s%15s%15s\n"
204 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
205 "RECLAIM %12s%15s%15s%15s%15s\n"
206 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
207 "THRASHING%12s%15s%15s%15s%15s\n"
208 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
209 "COMPACT %12s%15s%15s%15s%15s\n"
210 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
211 "WPCOPY %12s%15s%15s%15s%15s\n"
212 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n"
213 "IRQ %15s%15s%15s%15s%15s\n"
214 " %15llu%15llu%15.3fms%13.6fms%13.6fms\n",
215 "count", "real total", "virtual total",
216 "delay total", "delay average", "delay max", "delay min",
217 (unsigned long long)t->cpu_count,
218 (unsigned long long)t->cpu_run_real_total,
219 (unsigned long long)t->cpu_run_virtual_total,
220 (unsigned long long)t->cpu_delay_total,
221 average_ms((double)t->cpu_delay_total, t->cpu_count),
222 delay_ms((double)t->cpu_delay_max),
223 delay_ms((double)t->cpu_delay_min),
224 "count", "delay total", "delay average", "delay max", "delay min",
225 (unsigned long long)t->blkio_count,
226 (unsigned long long)t->blkio_delay_total,
227 average_ms((double)t->blkio_delay_total, t->blkio_count),
228 delay_ms((double)t->blkio_delay_max),
229 delay_ms((double)t->blkio_delay_min),
230 "count", "delay total", "delay average", "delay max", "delay min",
231 (unsigned long long)t->swapin_count,
232 (unsigned long long)t->swapin_delay_total,
233 average_ms((double)t->swapin_delay_total, t->swapin_count),
234 delay_ms((double)t->swapin_delay_max),
235 delay_ms((double)t->swapin_delay_min),
236 "count", "delay total", "delay average", "delay max", "delay min",
237 (unsigned long long)t->freepages_count,
238 (unsigned long long)t->freepages_delay_total,
239 average_ms((double)t->freepages_delay_total, t->freepages_count),
240 delay_ms((double)t->freepages_delay_max),
241 delay_ms((double)t->freepages_delay_min),
242 "count", "delay total", "delay average", "delay max", "delay min",
243 (unsigned long long)t->thrashing_count,
244 (unsigned long long)t->thrashing_delay_total,
245 average_ms((double)t->thrashing_delay_total, t->thrashing_count),
246 delay_ms((double)t->thrashing_delay_max),
247 delay_ms((double)t->thrashing_delay_min),
248 "count", "delay total", "delay average", "delay max", "delay min",
249 (unsigned long long)t->compact_count,
250 (unsigned long long)t->compact_delay_total,
251 average_ms((double)t->compact_delay_total, t->compact_count),
252 delay_ms((double)t->compact_delay_max),
253 delay_ms((double)t->compact_delay_min),
254 "count", "delay total", "delay average", "delay max", "delay min",
255 (unsigned long long)t->wpcopy_count,
256 (unsigned long long)t->wpcopy_delay_total,
257 average_ms((double)t->wpcopy_delay_total, t->wpcopy_count),
258 delay_ms((double)t->wpcopy_delay_max),
259 delay_ms((double)t->wpcopy_delay_min),
260 "count", "delay total", "delay average", "delay max", "delay min",
261 (unsigned long long)t->irq_count,
262 (unsigned long long)t->irq_delay_total,
263 average_ms((double)t->irq_delay_total, t->irq_count),
264 delay_ms((double)t->irq_delay_max),
265 delay_ms((double)t->irq_delay_min));
266 }
267
task_context_switch_counts(struct taskstats * t)268 static void task_context_switch_counts(struct taskstats *t)
269 {
270 printf("\n\nTask %15s%15s\n"
271 " %15llu%15llu\n",
272 "voluntary", "nonvoluntary",
273 (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw);
274 }
275
print_cgroupstats(struct cgroupstats * c)276 static void print_cgroupstats(struct cgroupstats *c)
277 {
278 printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
279 "uninterruptible %llu\n", (unsigned long long)c->nr_sleeping,
280 (unsigned long long)c->nr_io_wait,
281 (unsigned long long)c->nr_running,
282 (unsigned long long)c->nr_stopped,
283 (unsigned long long)c->nr_uninterruptible);
284 }
285
286
print_ioacct(struct taskstats * t)287 static void print_ioacct(struct taskstats *t)
288 {
289 printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n",
290 t->ac_comm,
291 (unsigned long long)t->read_bytes,
292 (unsigned long long)t->write_bytes,
293 (unsigned long long)t->cancelled_write_bytes);
294 }
295
main(int argc,char * argv[])296 int main(int argc, char *argv[])
297 {
298 int c, rc, rep_len, aggr_len, len2;
299 int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC;
300 __u16 id;
301 __u32 mypid;
302
303 struct nlattr *na;
304 int nl_sd = -1;
305 int len = 0;
306 pid_t tid = 0;
307 pid_t rtid = 0;
308
309 int fd = 0;
310 int write_file = 0;
311 int maskset = 0;
312 char *logfile = NULL;
313 int loop = 0;
314 int containerset = 0;
315 char *containerpath = NULL;
316 int cfd = 0;
317 int forking = 0;
318 sigset_t sigset;
319
320 struct msgtemplate msg;
321
322 while (!forking) {
323 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:c:");
324 if (c < 0)
325 break;
326
327 switch (c) {
328 case 'd':
329 printf("print delayacct stats ON\n");
330 print_delays = 1;
331 break;
332 case 'i':
333 printf("printing IO accounting\n");
334 print_io_accounting = 1;
335 break;
336 case 'q':
337 printf("printing task/process context switch rates\n");
338 print_task_context_switch_counts = 1;
339 break;
340 case 'C':
341 containerset = 1;
342 containerpath = optarg;
343 break;
344 case 'w':
345 logfile = strdup(optarg);
346 printf("write to file %s\n", logfile);
347 write_file = 1;
348 break;
349 case 'r':
350 rcvbufsz = atoi(optarg);
351 printf("receive buf size %d\n", rcvbufsz);
352 if (rcvbufsz < 0)
353 err(1, "Invalid rcv buf size\n");
354 break;
355 case 'm':
356 strncpy(cpumask, optarg, sizeof(cpumask));
357 cpumask[sizeof(cpumask) - 1] = '\0';
358 maskset = 1;
359 printf("cpumask %s maskset %d\n", cpumask, maskset);
360 break;
361 case 't':
362 tid = atoi(optarg);
363 if (!tid)
364 err(1, "Invalid tgid\n");
365 cmd_type = TASKSTATS_CMD_ATTR_TGID;
366 break;
367 case 'p':
368 tid = atoi(optarg);
369 if (!tid)
370 err(1, "Invalid pid\n");
371 cmd_type = TASKSTATS_CMD_ATTR_PID;
372 break;
373 case 'c':
374
375 /* Block SIGCHLD for sigwait() later */
376 if (sigemptyset(&sigset) == -1)
377 err(1, "Failed to empty sigset");
378 if (sigaddset(&sigset, SIGCHLD))
379 err(1, "Failed to set sigchld in sigset");
380 sigprocmask(SIG_BLOCK, &sigset, NULL);
381
382 /* fork/exec a child */
383 tid = fork();
384 if (tid < 0)
385 err(1, "Fork failed\n");
386 if (tid == 0)
387 if (execvp(argv[optind - 1],
388 &argv[optind - 1]) < 0)
389 exit(-1);
390
391 /* Set the command type and avoid further processing */
392 cmd_type = TASKSTATS_CMD_ATTR_PID;
393 forking = 1;
394 break;
395 case 'v':
396 printf("debug on\n");
397 dbg = 1;
398 break;
399 case 'l':
400 printf("listen forever\n");
401 loop = 1;
402 break;
403 default:
404 usage();
405 exit(-1);
406 }
407 }
408
409 if (write_file) {
410 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC,
411 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
412 if (fd == -1) {
413 perror("Cannot open output file\n");
414 exit(1);
415 }
416 }
417
418 nl_sd = create_nl_socket(NETLINK_GENERIC);
419 if (nl_sd < 0)
420 err(1, "error creating Netlink socket\n");
421
422
423 mypid = getpid();
424 id = get_family_id(nl_sd);
425 if (!id) {
426 fprintf(stderr, "Error getting family id, errno %d\n", errno);
427 goto err;
428 }
429 PRINTF("family id %d\n", id);
430
431 if (maskset) {
432 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
433 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
434 &cpumask, strlen(cpumask) + 1);
435 PRINTF("Sent register cpumask, retval %d\n", rc);
436 if (rc < 0) {
437 fprintf(stderr, "error sending register cpumask\n");
438 goto err;
439 }
440 }
441
442 if (tid && containerset) {
443 fprintf(stderr, "Select either -t or -C, not both\n");
444 goto err;
445 }
446
447 /*
448 * If we forked a child, wait for it to exit. Cannot use waitpid()
449 * as all the delicious data would be reaped as part of the wait
450 */
451 if (tid && forking) {
452 int sig_received;
453 sigwait(&sigset, &sig_received);
454 }
455
456 if (tid) {
457 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
458 cmd_type, &tid, sizeof(__u32));
459 PRINTF("Sent pid/tgid, retval %d\n", rc);
460 if (rc < 0) {
461 fprintf(stderr, "error sending tid/tgid cmd\n");
462 goto done;
463 }
464 }
465
466 if (containerset) {
467 cfd = open(containerpath, O_RDONLY);
468 if (cfd < 0) {
469 perror("error opening container file");
470 goto err;
471 }
472 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET,
473 CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32));
474 if (rc < 0) {
475 perror("error sending cgroupstats command");
476 goto err;
477 }
478 }
479 if (!maskset && !tid && !containerset) {
480 usage();
481 goto err;
482 }
483
484 do {
485 rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
486 PRINTF("received %d bytes\n", rep_len);
487
488 if (rep_len < 0) {
489 fprintf(stderr, "nonfatal reply error: errno %d\n",
490 errno);
491 continue;
492 }
493 if (msg.n.nlmsg_type == NLMSG_ERROR ||
494 !NLMSG_OK((&msg.n), rep_len)) {
495 struct nlmsgerr *err = NLMSG_DATA(&msg);
496 fprintf(stderr, "fatal reply error, errno %d\n",
497 err->error);
498 goto done;
499 }
500
501 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
502 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
503
504
505 rep_len = GENLMSG_PAYLOAD(&msg.n);
506
507 na = (struct nlattr *) GENLMSG_DATA(&msg);
508 len = 0;
509 while (len < rep_len) {
510 len += NLA_ALIGN(na->nla_len);
511 switch (na->nla_type) {
512 case TASKSTATS_TYPE_AGGR_TGID:
513 /* Fall through */
514 case TASKSTATS_TYPE_AGGR_PID:
515 aggr_len = NLA_PAYLOAD(na->nla_len);
516 len2 = 0;
517 /* For nested attributes, na follows */
518 na = (struct nlattr *) NLA_DATA(na);
519 while (len2 < aggr_len) {
520 switch (na->nla_type) {
521 case TASKSTATS_TYPE_PID:
522 rtid = *(int *) NLA_DATA(na);
523 if (print_delays)
524 printf("PID\t%d\n", rtid);
525 break;
526 case TASKSTATS_TYPE_TGID:
527 rtid = *(int *) NLA_DATA(na);
528 if (print_delays)
529 printf("TGID\t%d\n", rtid);
530 break;
531 case TASKSTATS_TYPE_STATS:
532 if (print_delays)
533 print_delayacct((struct taskstats *) NLA_DATA(na));
534 if (print_io_accounting)
535 print_ioacct((struct taskstats *) NLA_DATA(na));
536 if (print_task_context_switch_counts)
537 task_context_switch_counts((struct taskstats *) NLA_DATA(na));
538 if (fd) {
539 if (write(fd, NLA_DATA(na), na->nla_len) < 0) {
540 err(1,"write error\n");
541 }
542 }
543 if (!loop)
544 goto done;
545 break;
546 case TASKSTATS_TYPE_NULL:
547 break;
548 default:
549 fprintf(stderr, "Unknown nested"
550 " nla_type %d\n",
551 na->nla_type);
552 break;
553 }
554 len2 += NLA_ALIGN(na->nla_len);
555 na = (struct nlattr *)((char *)na +
556 NLA_ALIGN(na->nla_len));
557 }
558 break;
559
560 case CGROUPSTATS_TYPE_CGROUP_STATS:
561 print_cgroupstats(NLA_DATA(na));
562 break;
563 default:
564 fprintf(stderr, "Unknown nla_type %d\n",
565 na->nla_type);
566 case TASKSTATS_TYPE_NULL:
567 break;
568 }
569 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
570 }
571 } while (loop);
572 done:
573 if (maskset) {
574 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
575 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
576 &cpumask, strlen(cpumask) + 1);
577 printf("Sent deregister mask, retval %d\n", rc);
578 if (rc < 0)
579 err(rc, "error sending deregister cpumask\n");
580 }
581 err:
582 close(nl_sd);
583 if (fd)
584 close(fd);
585 if (cfd)
586 close(cfd);
587 return 0;
588 }
589