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 done;
49 int rcvbufsz;
50 char name[100];
51 int dbg;
52 int print_delays;
53 int print_io_accounting;
54 int print_task_context_switch_counts;
55
56 #define PRINTF(fmt, arg...) { \
57 if (dbg) { \
58 printf(fmt, ##arg); \
59 } \
60 }
61
62 /* Maximum size of response requested or message sent */
63 #define MAX_MSG_SIZE 1024
64 /* Maximum number of cpus expected to be specified in a cpumask */
65 #define MAX_CPUS 32
66
67 struct msgtemplate {
68 struct nlmsghdr n;
69 struct genlmsghdr g;
70 char buf[MAX_MSG_SIZE];
71 };
72
73 char cpumask[100+6*MAX_CPUS];
74
usage(void)75 static void usage(void)
76 {
77 fprintf(stderr, "getdelays [-dilv] [-w logfile] [-r bufsize] "
78 "[-m cpumask] [-t tgid] [-p pid]\n");
79 fprintf(stderr, " -d: print delayacct stats\n");
80 fprintf(stderr, " -i: print IO accounting (works only with -p)\n");
81 fprintf(stderr, " -l: listen forever\n");
82 fprintf(stderr, " -v: debug on\n");
83 fprintf(stderr, " -C: container path\n");
84 }
85
86 /*
87 * Create a raw netlink socket and bind
88 */
create_nl_socket(int protocol)89 static int create_nl_socket(int protocol)
90 {
91 int fd;
92 struct sockaddr_nl local;
93
94 fd = socket(AF_NETLINK, SOCK_RAW, protocol);
95 if (fd < 0)
96 return -1;
97
98 if (rcvbufsz)
99 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF,
100 &rcvbufsz, sizeof(rcvbufsz)) < 0) {
101 fprintf(stderr, "Unable to set socket rcv buf size to %d\n",
102 rcvbufsz);
103 goto error;
104 }
105
106 memset(&local, 0, sizeof(local));
107 local.nl_family = AF_NETLINK;
108
109 if (bind(fd, (struct sockaddr *) &local, sizeof(local)) < 0)
110 goto error;
111
112 return fd;
113 error:
114 close(fd);
115 return -1;
116 }
117
118
send_cmd(int sd,__u16 nlmsg_type,__u32 nlmsg_pid,__u8 genl_cmd,__u16 nla_type,void * nla_data,int nla_len)119 static int send_cmd(int sd, __u16 nlmsg_type, __u32 nlmsg_pid,
120 __u8 genl_cmd, __u16 nla_type,
121 void *nla_data, int nla_len)
122 {
123 struct nlattr *na;
124 struct sockaddr_nl nladdr;
125 int r, buflen;
126 char *buf;
127
128 struct msgtemplate msg;
129
130 msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN);
131 msg.n.nlmsg_type = nlmsg_type;
132 msg.n.nlmsg_flags = NLM_F_REQUEST;
133 msg.n.nlmsg_seq = 0;
134 msg.n.nlmsg_pid = nlmsg_pid;
135 msg.g.cmd = genl_cmd;
136 msg.g.version = 0x1;
137 na = (struct nlattr *) GENLMSG_DATA(&msg);
138 na->nla_type = nla_type;
139 na->nla_len = nla_len + NLA_HDRLEN;
140 memcpy(NLA_DATA(na), nla_data, nla_len);
141 msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len);
142
143 buf = (char *) &msg;
144 buflen = msg.n.nlmsg_len ;
145 memset(&nladdr, 0, sizeof(nladdr));
146 nladdr.nl_family = AF_NETLINK;
147 while ((r = sendto(sd, buf, buflen, 0, (struct sockaddr *) &nladdr,
148 sizeof(nladdr))) < buflen) {
149 if (r > 0) {
150 buf += r;
151 buflen -= r;
152 } else if (errno != EAGAIN)
153 return -1;
154 }
155 return 0;
156 }
157
158
159 /*
160 * Probe the controller in genetlink to find the family id
161 * for the TASKSTATS family
162 */
get_family_id(int sd)163 static int get_family_id(int sd)
164 {
165 struct {
166 struct nlmsghdr n;
167 struct genlmsghdr g;
168 char buf[256];
169 } ans;
170
171 int id = 0, rc;
172 struct nlattr *na;
173 int rep_len;
174
175 strcpy(name, TASKSTATS_GENL_NAME);
176 rc = send_cmd(sd, GENL_ID_CTRL, getpid(), CTRL_CMD_GETFAMILY,
177 CTRL_ATTR_FAMILY_NAME, (void *)name,
178 strlen(TASKSTATS_GENL_NAME)+1);
179 if (rc < 0)
180 return 0; /* sendto() failure? */
181
182 rep_len = recv(sd, &ans, sizeof(ans), 0);
183 if (ans.n.nlmsg_type == NLMSG_ERROR ||
184 (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len))
185 return 0;
186
187 na = (struct nlattr *) GENLMSG_DATA(&ans);
188 na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len));
189 if (na->nla_type == CTRL_ATTR_FAMILY_ID) {
190 id = *(__u16 *) NLA_DATA(na);
191 }
192 return id;
193 }
194
195 #define average_ms(t, c) (t / 1000000ULL / (c ? c : 1))
196
print_delayacct(struct taskstats * t)197 static void print_delayacct(struct taskstats *t)
198 {
199 printf("\n\nCPU %15s%15s%15s%15s%15s\n"
200 " %15llu%15llu%15llu%15llu%15.3fms\n"
201 "IO %15s%15s%15s\n"
202 " %15llu%15llu%15llums\n"
203 "SWAP %15s%15s%15s\n"
204 " %15llu%15llu%15llums\n"
205 "RECLAIM %12s%15s%15s\n"
206 " %15llu%15llu%15llums\n"
207 "THRASHING%12s%15s%15s\n"
208 " %15llu%15llu%15llums\n",
209 "count", "real total", "virtual total",
210 "delay total", "delay average",
211 (unsigned long long)t->cpu_count,
212 (unsigned long long)t->cpu_run_real_total,
213 (unsigned long long)t->cpu_run_virtual_total,
214 (unsigned long long)t->cpu_delay_total,
215 average_ms((double)t->cpu_delay_total, t->cpu_count),
216 "count", "delay total", "delay average",
217 (unsigned long long)t->blkio_count,
218 (unsigned long long)t->blkio_delay_total,
219 average_ms(t->blkio_delay_total, t->blkio_count),
220 "count", "delay total", "delay average",
221 (unsigned long long)t->swapin_count,
222 (unsigned long long)t->swapin_delay_total,
223 average_ms(t->swapin_delay_total, t->swapin_count),
224 "count", "delay total", "delay average",
225 (unsigned long long)t->freepages_count,
226 (unsigned long long)t->freepages_delay_total,
227 average_ms(t->freepages_delay_total, t->freepages_count),
228 "count", "delay total", "delay average",
229 (unsigned long long)t->thrashing_count,
230 (unsigned long long)t->thrashing_delay_total,
231 average_ms(t->thrashing_delay_total, t->thrashing_count));
232 }
233
task_context_switch_counts(struct taskstats * t)234 static void task_context_switch_counts(struct taskstats *t)
235 {
236 printf("\n\nTask %15s%15s\n"
237 " %15llu%15llu\n",
238 "voluntary", "nonvoluntary",
239 (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw);
240 }
241
print_cgroupstats(struct cgroupstats * c)242 static void print_cgroupstats(struct cgroupstats *c)
243 {
244 printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
245 "uninterruptible %llu\n", (unsigned long long)c->nr_sleeping,
246 (unsigned long long)c->nr_io_wait,
247 (unsigned long long)c->nr_running,
248 (unsigned long long)c->nr_stopped,
249 (unsigned long long)c->nr_uninterruptible);
250 }
251
252
print_ioacct(struct taskstats * t)253 static void print_ioacct(struct taskstats *t)
254 {
255 printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n",
256 t->ac_comm,
257 (unsigned long long)t->read_bytes,
258 (unsigned long long)t->write_bytes,
259 (unsigned long long)t->cancelled_write_bytes);
260 }
261
main(int argc,char * argv[])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 pid_t rtid = 0;
274
275 int fd = 0;
276 int count = 0;
277 int write_file = 0;
278 int maskset = 0;
279 char *logfile = NULL;
280 int loop = 0;
281 int containerset = 0;
282 char *containerpath = NULL;
283 int cfd = 0;
284 int forking = 0;
285 sigset_t sigset;
286
287 struct msgtemplate msg;
288
289 while (!forking) {
290 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:c:");
291 if (c < 0)
292 break;
293
294 switch (c) {
295 case 'd':
296 printf("print delayacct stats ON\n");
297 print_delays = 1;
298 break;
299 case 'i':
300 printf("printing IO accounting\n");
301 print_io_accounting = 1;
302 break;
303 case 'q':
304 printf("printing task/process context switch rates\n");
305 print_task_context_switch_counts = 1;
306 break;
307 case 'C':
308 containerset = 1;
309 containerpath = optarg;
310 break;
311 case 'w':
312 logfile = strdup(optarg);
313 printf("write to file %s\n", logfile);
314 write_file = 1;
315 break;
316 case 'r':
317 rcvbufsz = atoi(optarg);
318 printf("receive buf size %d\n", rcvbufsz);
319 if (rcvbufsz < 0)
320 err(1, "Invalid rcv buf size\n");
321 break;
322 case 'm':
323 strncpy(cpumask, optarg, sizeof(cpumask));
324 cpumask[sizeof(cpumask) - 1] = '\0';
325 maskset = 1;
326 printf("cpumask %s maskset %d\n", cpumask, maskset);
327 break;
328 case 't':
329 tid = atoi(optarg);
330 if (!tid)
331 err(1, "Invalid tgid\n");
332 cmd_type = TASKSTATS_CMD_ATTR_TGID;
333 break;
334 case 'p':
335 tid = atoi(optarg);
336 if (!tid)
337 err(1, "Invalid pid\n");
338 cmd_type = TASKSTATS_CMD_ATTR_PID;
339 break;
340 case 'c':
341
342 /* Block SIGCHLD for sigwait() later */
343 if (sigemptyset(&sigset) == -1)
344 err(1, "Failed to empty sigset");
345 if (sigaddset(&sigset, SIGCHLD))
346 err(1, "Failed to set sigchld in sigset");
347 sigprocmask(SIG_BLOCK, &sigset, NULL);
348
349 /* fork/exec a child */
350 tid = fork();
351 if (tid < 0)
352 err(1, "Fork failed\n");
353 if (tid == 0)
354 if (execvp(argv[optind - 1],
355 &argv[optind - 1]) < 0)
356 exit(-1);
357
358 /* Set the command type and avoid further processing */
359 cmd_type = TASKSTATS_CMD_ATTR_PID;
360 forking = 1;
361 break;
362 case 'v':
363 printf("debug on\n");
364 dbg = 1;
365 break;
366 case 'l':
367 printf("listen forever\n");
368 loop = 1;
369 break;
370 default:
371 usage();
372 exit(-1);
373 }
374 }
375
376 if (write_file) {
377 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC,
378 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
379 if (fd == -1) {
380 perror("Cannot open output file\n");
381 exit(1);
382 }
383 }
384
385 nl_sd = create_nl_socket(NETLINK_GENERIC);
386 if (nl_sd < 0)
387 err(1, "error creating Netlink socket\n");
388
389
390 mypid = getpid();
391 id = get_family_id(nl_sd);
392 if (!id) {
393 fprintf(stderr, "Error getting family id, errno %d\n", errno);
394 goto err;
395 }
396 PRINTF("family id %d\n", id);
397
398 if (maskset) {
399 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
400 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
401 &cpumask, strlen(cpumask) + 1);
402 PRINTF("Sent register cpumask, retval %d\n", rc);
403 if (rc < 0) {
404 fprintf(stderr, "error sending register cpumask\n");
405 goto err;
406 }
407 }
408
409 if (tid && containerset) {
410 fprintf(stderr, "Select either -t or -C, not both\n");
411 goto err;
412 }
413
414 /*
415 * If we forked a child, wait for it to exit. Cannot use waitpid()
416 * as all the delicious data would be reaped as part of the wait
417 */
418 if (tid && forking) {
419 int sig_received;
420 sigwait(&sigset, &sig_received);
421 }
422
423 if (tid) {
424 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
425 cmd_type, &tid, sizeof(__u32));
426 PRINTF("Sent pid/tgid, retval %d\n", rc);
427 if (rc < 0) {
428 fprintf(stderr, "error sending tid/tgid cmd\n");
429 goto done;
430 }
431 }
432
433 if (containerset) {
434 cfd = open(containerpath, O_RDONLY);
435 if (cfd < 0) {
436 perror("error opening container file");
437 goto err;
438 }
439 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET,
440 CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32));
441 if (rc < 0) {
442 perror("error sending cgroupstats command");
443 goto err;
444 }
445 }
446 if (!maskset && !tid && !containerset) {
447 usage();
448 goto err;
449 }
450
451 do {
452 rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
453 PRINTF("received %d bytes\n", rep_len);
454
455 if (rep_len < 0) {
456 fprintf(stderr, "nonfatal reply error: errno %d\n",
457 errno);
458 continue;
459 }
460 if (msg.n.nlmsg_type == NLMSG_ERROR ||
461 !NLMSG_OK((&msg.n), rep_len)) {
462 struct nlmsgerr *err = NLMSG_DATA(&msg);
463 fprintf(stderr, "fatal reply error, errno %d\n",
464 err->error);
465 goto done;
466 }
467
468 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
469 sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
470
471
472 rep_len = GENLMSG_PAYLOAD(&msg.n);
473
474 na = (struct nlattr *) GENLMSG_DATA(&msg);
475 len = 0;
476 while (len < rep_len) {
477 len += NLA_ALIGN(na->nla_len);
478 switch (na->nla_type) {
479 case TASKSTATS_TYPE_AGGR_TGID:
480 /* Fall through */
481 case TASKSTATS_TYPE_AGGR_PID:
482 aggr_len = NLA_PAYLOAD(na->nla_len);
483 len2 = 0;
484 /* For nested attributes, na follows */
485 na = (struct nlattr *) NLA_DATA(na);
486 done = 0;
487 while (len2 < aggr_len) {
488 switch (na->nla_type) {
489 case TASKSTATS_TYPE_PID:
490 rtid = *(int *) NLA_DATA(na);
491 if (print_delays)
492 printf("PID\t%d\n", rtid);
493 break;
494 case TASKSTATS_TYPE_TGID:
495 rtid = *(int *) NLA_DATA(na);
496 if (print_delays)
497 printf("TGID\t%d\n", rtid);
498 break;
499 case TASKSTATS_TYPE_STATS:
500 count++;
501 if (print_delays)
502 print_delayacct((struct taskstats *) NLA_DATA(na));
503 if (print_io_accounting)
504 print_ioacct((struct taskstats *) NLA_DATA(na));
505 if (print_task_context_switch_counts)
506 task_context_switch_counts((struct taskstats *) NLA_DATA(na));
507 if (fd) {
508 if (write(fd, NLA_DATA(na), na->nla_len) < 0) {
509 err(1,"write error\n");
510 }
511 }
512 if (!loop)
513 goto done;
514 break;
515 case TASKSTATS_TYPE_NULL:
516 break;
517 default:
518 fprintf(stderr, "Unknown nested"
519 " nla_type %d\n",
520 na->nla_type);
521 break;
522 }
523 len2 += NLA_ALIGN(na->nla_len);
524 na = (struct nlattr *)((char *)na +
525 NLA_ALIGN(na->nla_len));
526 }
527 break;
528
529 case CGROUPSTATS_TYPE_CGROUP_STATS:
530 print_cgroupstats(NLA_DATA(na));
531 break;
532 default:
533 fprintf(stderr, "Unknown nla_type %d\n",
534 na->nla_type);
535 case TASKSTATS_TYPE_NULL:
536 break;
537 }
538 na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
539 }
540 } while (loop);
541 done:
542 if (maskset) {
543 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
544 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
545 &cpumask, strlen(cpumask) + 1);
546 printf("Sent deregister mask, retval %d\n", rc);
547 if (rc < 0)
548 err(rc, "error sending deregister cpumask\n");
549 }
550 err:
551 close(nl_sd);
552 if (fd)
553 close(fd);
554 if (cfd)
555 close(cfd);
556 return 0;
557 }
558