]>
Commit | Line | Data |
---|---|---|
a3baf649 SN |
1 | /* getdelays.c |
2 | * | |
3 | * Utility to get per-pid and per-tgid delay accounting statistics | |
4 | * Also illustrates usage of the taskstats interface | |
5 | * | |
6 | * Copyright (C) Shailabh Nagar, IBM Corp. 2005 | |
7 | * Copyright (C) Balbir Singh, IBM Corp. 2006 | |
9e06d3f9 | 8 | * Copyright (c) Jay Lan, SGI. 2006 |
a3baf649 | 9 | * |
d2f7bf13 AM |
10 | * Compile with |
11 | * gcc -I/usr/src/linux/include getdelays.c -o getdelays | |
a3baf649 SN |
12 | */ |
13 | ||
14 | #include <stdio.h> | |
15 | #include <stdlib.h> | |
16 | #include <errno.h> | |
17 | #include <unistd.h> | |
18 | #include <poll.h> | |
19 | #include <string.h> | |
20 | #include <fcntl.h> | |
21 | #include <sys/types.h> | |
22 | #include <sys/stat.h> | |
23 | #include <sys/socket.h> | |
a3baf649 SN |
24 | #include <signal.h> |
25 | ||
26 | #include <linux/genetlink.h> | |
27 | #include <linux/taskstats.h> | |
546040dc | 28 | #include <linux/cgroupstats.h> |
a3baf649 SN |
29 | |
30 | /* | |
31 | * Generic macros for dealing with netlink sockets. Might be duplicated | |
32 | * elsewhere. It is recommended that commercial grade applications use | |
33 | * libnl or libnetlink and use the interfaces provided by the library | |
34 | */ | |
35 | #define GENLMSG_DATA(glh) ((void *)(NLMSG_DATA(glh) + GENL_HDRLEN)) | |
36 | #define GENLMSG_PAYLOAD(glh) (NLMSG_PAYLOAD(glh, 0) - GENL_HDRLEN) | |
37 | #define NLA_DATA(na) ((void *)((char*)(na) + NLA_HDRLEN)) | |
38 | #define NLA_PAYLOAD(len) (len - NLA_HDRLEN) | |
39 | ||
d2f7bf13 AM |
40 | #define err(code, fmt, arg...) \ |
41 | do { \ | |
42 | fprintf(stderr, fmt, ##arg); \ | |
43 | exit(code); \ | |
44 | } while (0) | |
45 | ||
46 | int done; | |
47 | int rcvbufsz; | |
48 | char name[100]; | |
49 | int dbg; | |
50 | int print_delays; | |
cf709844 | 51 | int print_io_accounting; |
b663a79c | 52 | int print_task_context_switch_counts; |
9e06d3f9 | 53 | __u64 stime, utime; |
d2f7bf13 | 54 | |
9e06d3f9 SN |
55 | #define PRINTF(fmt, arg...) { \ |
56 | if (dbg) { \ | |
57 | printf(fmt, ##arg); \ | |
58 | } \ | |
59 | } | |
60 | ||
61 | /* Maximum size of response requested or message sent */ | |
88040230 | 62 | #define MAX_MSG_SIZE 1024 |
9e06d3f9 SN |
63 | /* Maximum number of cpus expected to be specified in a cpumask */ |
64 | #define MAX_CPUS 32 | |
9e06d3f9 SN |
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]; | |
a3baf649 | 73 | |
f16825bb RD |
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"); | |
546040dc | 82 | fprintf(stderr, " -C: container path\n"); |
f16825bb RD |
83 | } |
84 | ||
a3baf649 SN |
85 | /* |
86 | * Create a raw netlink socket and bind | |
87 | */ | |
9e06d3f9 | 88 | static int create_nl_socket(int protocol) |
a3baf649 | 89 | { |
9e06d3f9 SN |
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) { | |
d2f7bf13 AM |
100 | fprintf(stderr, "Unable to set socket rcv buf size " |
101 | "to %d\n", | |
102 | rcvbufsz); | |
9e06d3f9 SN |
103 | return -1; |
104 | } | |
a3baf649 | 105 | |
9e06d3f9 SN |
106 | memset(&local, 0, sizeof(local)); |
107 | local.nl_family = AF_NETLINK; | |
a3baf649 | 108 | |
9e06d3f9 SN |
109 | if (bind(fd, (struct sockaddr *) &local, sizeof(local)) < 0) |
110 | goto error; | |
a3baf649 | 111 | |
9e06d3f9 SN |
112 | return fd; |
113 | error: | |
114 | close(fd); | |
115 | return -1; | |
a3baf649 SN |
116 | } |
117 | ||
9e06d3f9 SN |
118 | |
119 | 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) | |
a3baf649 | 122 | { |
9e06d3f9 SN |
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 + 1 + 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; | |
a3baf649 SN |
156 | } |
157 | ||
9e06d3f9 | 158 | |
a3baf649 SN |
159 | /* |
160 | * Probe the controller in genetlink to find the family id | |
161 | * for the TASKSTATS family | |
162 | */ | |
163 | int get_family_id(int sd) | |
164 | { | |
9e06d3f9 SN |
165 | struct { |
166 | struct nlmsghdr n; | |
167 | struct genlmsghdr g; | |
168 | char buf[256]; | |
169 | } ans; | |
170 | ||
10e6f32b | 171 | int id = 0, rc; |
9e06d3f9 SN |
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 | ||
180 | rep_len = recv(sd, &ans, sizeof(ans), 0); | |
181 | if (ans.n.nlmsg_type == NLMSG_ERROR || | |
182 | (rep_len < 0) || !NLMSG_OK((&ans.n), rep_len)) | |
183 | return 0; | |
a3baf649 | 184 | |
9e06d3f9 SN |
185 | na = (struct nlattr *) GENLMSG_DATA(&ans); |
186 | na = (struct nlattr *) ((char *) na + NLA_ALIGN(na->nla_len)); | |
187 | if (na->nla_type == CTRL_ATTR_FAMILY_ID) { | |
188 | id = *(__u16 *) NLA_DATA(na); | |
189 | } | |
190 | return id; | |
a3baf649 SN |
191 | } |
192 | ||
9e06d3f9 | 193 | void print_delayacct(struct taskstats *t) |
a3baf649 | 194 | { |
9e06d3f9 SN |
195 | printf("\n\nCPU %15s%15s%15s%15s\n" |
196 | " %15llu%15llu%15llu%15llu\n" | |
197 | "IO %15s%15s\n" | |
198 | " %15llu%15llu\n" | |
9b0975a2 KK |
199 | "SWAP %15s%15s\n" |
200 | " %15llu%15llu\n" | |
201 | "RECLAIM %12s%15s\n" | |
efbee7f1 | 202 | " %15llu%15llu\n", |
9e06d3f9 | 203 | "count", "real total", "virtual total", "delay total", |
66659313 RD |
204 | (unsigned long long)t->cpu_count, |
205 | (unsigned long long)t->cpu_run_real_total, | |
206 | (unsigned long long)t->cpu_run_virtual_total, | |
207 | (unsigned long long)t->cpu_delay_total, | |
9e06d3f9 | 208 | "count", "delay total", |
66659313 RD |
209 | (unsigned long long)t->blkio_count, |
210 | (unsigned long long)t->blkio_delay_total, | |
9b0975a2 | 211 | "count", "delay total", |
66659313 RD |
212 | (unsigned long long)t->swapin_count, |
213 | (unsigned long long)t->swapin_delay_total, | |
214 | "count", "delay total", | |
215 | (unsigned long long)t->freepages_count, | |
216 | (unsigned long long)t->freepages_delay_total); | |
a3baf649 SN |
217 | } |
218 | ||
b663a79c MU |
219 | void task_context_switch_counts(struct taskstats *t) |
220 | { | |
221 | printf("\n\nTask %15s%15s\n" | |
10e6f32b | 222 | " %15llu%15llu\n", |
b663a79c | 223 | "voluntary", "nonvoluntary", |
66659313 | 224 | (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw); |
b663a79c MU |
225 | } |
226 | ||
546040dc BS |
227 | void print_cgroupstats(struct cgroupstats *c) |
228 | { | |
229 | printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, " | |
66659313 RD |
230 | "uninterruptible %llu\n", (unsigned long long)c->nr_sleeping, |
231 | (unsigned long long)c->nr_io_wait, | |
232 | (unsigned long long)c->nr_running, | |
233 | (unsigned long long)c->nr_stopped, | |
234 | (unsigned long long)c->nr_uninterruptible); | |
546040dc BS |
235 | } |
236 | ||
237 | ||
cf709844 AM |
238 | void print_ioacct(struct taskstats *t) |
239 | { | |
240 | printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n", | |
241 | t->ac_comm, | |
242 | (unsigned long long)t->read_bytes, | |
243 | (unsigned long long)t->write_bytes, | |
244 | (unsigned long long)t->cancelled_write_bytes); | |
245 | } | |
246 | ||
a3baf649 SN |
247 | int main(int argc, char *argv[]) |
248 | { | |
b8d9a865 JSR |
249 | int c, rc, rep_len, aggr_len, len2; |
250 | int cmd_type = TASKSTATS_CMD_ATTR_UNSPEC; | |
9e06d3f9 SN |
251 | __u16 id; |
252 | __u32 mypid; | |
253 | ||
254 | struct nlattr *na; | |
255 | int nl_sd = -1; | |
256 | int len = 0; | |
257 | pid_t tid = 0; | |
258 | pid_t rtid = 0; | |
259 | ||
260 | int fd = 0; | |
261 | int count = 0; | |
262 | int write_file = 0; | |
263 | int maskset = 0; | |
7f76c403 | 264 | char *logfile = NULL; |
9e06d3f9 | 265 | int loop = 0; |
546040dc BS |
266 | int containerset = 0; |
267 | char containerpath[1024]; | |
268 | int cfd = 0; | |
9e06d3f9 SN |
269 | |
270 | struct msgtemplate msg; | |
271 | ||
272 | while (1) { | |
546040dc | 273 | c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:"); |
9e06d3f9 SN |
274 | if (c < 0) |
275 | break; | |
a3baf649 | 276 | |
9e06d3f9 SN |
277 | switch (c) { |
278 | case 'd': | |
279 | printf("print delayacct stats ON\n"); | |
280 | print_delays = 1; | |
281 | break; | |
cf709844 AM |
282 | case 'i': |
283 | printf("printing IO accounting\n"); | |
284 | print_io_accounting = 1; | |
285 | break; | |
b663a79c MU |
286 | case 'q': |
287 | printf("printing task/process context switch rates\n"); | |
288 | print_task_context_switch_counts = 1; | |
289 | break; | |
546040dc BS |
290 | case 'C': |
291 | containerset = 1; | |
292 | strncpy(containerpath, optarg, strlen(optarg) + 1); | |
293 | break; | |
9e06d3f9 | 294 | case 'w': |
7f76c403 | 295 | logfile = strdup(optarg); |
9e06d3f9 SN |
296 | printf("write to file %s\n", logfile); |
297 | write_file = 1; | |
298 | break; | |
299 | case 'r': | |
300 | rcvbufsz = atoi(optarg); | |
301 | printf("receive buf size %d\n", rcvbufsz); | |
302 | if (rcvbufsz < 0) | |
303 | err(1, "Invalid rcv buf size\n"); | |
304 | break; | |
305 | case 'm': | |
306 | strncpy(cpumask, optarg, sizeof(cpumask)); | |
307 | maskset = 1; | |
308 | printf("cpumask %s maskset %d\n", cpumask, maskset); | |
309 | break; | |
310 | case 't': | |
311 | tid = atoi(optarg); | |
312 | if (!tid) | |
313 | err(1, "Invalid tgid\n"); | |
314 | cmd_type = TASKSTATS_CMD_ATTR_TGID; | |
9e06d3f9 SN |
315 | break; |
316 | case 'p': | |
317 | tid = atoi(optarg); | |
318 | if (!tid) | |
319 | err(1, "Invalid pid\n"); | |
320 | cmd_type = TASKSTATS_CMD_ATTR_PID; | |
9e06d3f9 SN |
321 | break; |
322 | case 'v': | |
323 | printf("debug on\n"); | |
324 | dbg = 1; | |
325 | break; | |
326 | case 'l': | |
327 | printf("listen forever\n"); | |
328 | loop = 1; | |
329 | break; | |
330 | default: | |
f16825bb | 331 | usage(); |
9e06d3f9 | 332 | exit(-1); |
a3baf649 | 333 | } |
a3baf649 | 334 | } |
a3baf649 | 335 | |
9e06d3f9 SN |
336 | if (write_file) { |
337 | fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC, | |
338 | S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH); | |
339 | if (fd == -1) { | |
340 | perror("Cannot open output file\n"); | |
341 | exit(1); | |
342 | } | |
343 | } | |
a3baf649 | 344 | |
9e06d3f9 SN |
345 | if ((nl_sd = create_nl_socket(NETLINK_GENERIC)) < 0) |
346 | err(1, "error creating Netlink socket\n"); | |
a3baf649 | 347 | |
a3baf649 | 348 | |
9e06d3f9 SN |
349 | mypid = getpid(); |
350 | id = get_family_id(nl_sd); | |
351 | if (!id) { | |
d2f7bf13 | 352 | fprintf(stderr, "Error getting family id, errno %d\n", errno); |
9e06d3f9 | 353 | goto err; |
a3baf649 | 354 | } |
9e06d3f9 SN |
355 | PRINTF("family id %d\n", id); |
356 | ||
357 | if (maskset) { | |
358 | rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, | |
359 | TASKSTATS_CMD_ATTR_REGISTER_CPUMASK, | |
7d1bdca9 | 360 | &cpumask, strlen(cpumask) + 1); |
9e06d3f9 SN |
361 | PRINTF("Sent register cpumask, retval %d\n", rc); |
362 | if (rc < 0) { | |
d2f7bf13 | 363 | fprintf(stderr, "error sending register cpumask\n"); |
9e06d3f9 SN |
364 | goto err; |
365 | } | |
a3baf649 SN |
366 | } |
367 | ||
546040dc BS |
368 | if (tid && containerset) { |
369 | fprintf(stderr, "Select either -t or -C, not both\n"); | |
370 | goto err; | |
371 | } | |
372 | ||
9e06d3f9 SN |
373 | if (tid) { |
374 | rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, | |
375 | cmd_type, &tid, sizeof(__u32)); | |
376 | PRINTF("Sent pid/tgid, retval %d\n", rc); | |
377 | if (rc < 0) { | |
d2f7bf13 | 378 | fprintf(stderr, "error sending tid/tgid cmd\n"); |
9e06d3f9 SN |
379 | goto done; |
380 | } | |
a3baf649 SN |
381 | } |
382 | ||
546040dc BS |
383 | if (containerset) { |
384 | cfd = open(containerpath, O_RDONLY); | |
385 | if (cfd < 0) { | |
386 | perror("error opening container file"); | |
387 | goto err; | |
388 | } | |
389 | rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET, | |
390 | CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32)); | |
391 | if (rc < 0) { | |
392 | perror("error sending cgroupstats command"); | |
393 | goto err; | |
394 | } | |
395 | } | |
65a67bd2 MM |
396 | if (!maskset && !tid && !containerset) { |
397 | usage(); | |
398 | goto err; | |
399 | } | |
546040dc | 400 | |
9e06d3f9 SN |
401 | do { |
402 | int i; | |
a3baf649 | 403 | |
9e06d3f9 SN |
404 | rep_len = recv(nl_sd, &msg, sizeof(msg), 0); |
405 | PRINTF("received %d bytes\n", rep_len); | |
a3baf649 | 406 | |
9e06d3f9 | 407 | if (rep_len < 0) { |
d2f7bf13 AM |
408 | fprintf(stderr, "nonfatal reply error: errno %d\n", |
409 | errno); | |
9e06d3f9 SN |
410 | continue; |
411 | } | |
412 | if (msg.n.nlmsg_type == NLMSG_ERROR || | |
413 | !NLMSG_OK((&msg.n), rep_len)) { | |
7d1bdca9 | 414 | struct nlmsgerr *err = NLMSG_DATA(&msg); |
d2f7bf13 AM |
415 | fprintf(stderr, "fatal reply error, errno %d\n", |
416 | err->error); | |
9e06d3f9 SN |
417 | goto done; |
418 | } | |
419 | ||
10e6f32b | 420 | PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n", |
9e06d3f9 SN |
421 | sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len); |
422 | ||
423 | ||
424 | rep_len = GENLMSG_PAYLOAD(&msg.n); | |
425 | ||
426 | na = (struct nlattr *) GENLMSG_DATA(&msg); | |
427 | len = 0; | |
428 | i = 0; | |
429 | while (len < rep_len) { | |
430 | len += NLA_ALIGN(na->nla_len); | |
431 | switch (na->nla_type) { | |
432 | case TASKSTATS_TYPE_AGGR_TGID: | |
433 | /* Fall through */ | |
434 | case TASKSTATS_TYPE_AGGR_PID: | |
435 | aggr_len = NLA_PAYLOAD(na->nla_len); | |
436 | len2 = 0; | |
437 | /* For nested attributes, na follows */ | |
438 | na = (struct nlattr *) NLA_DATA(na); | |
439 | done = 0; | |
440 | while (len2 < aggr_len) { | |
441 | switch (na->nla_type) { | |
442 | case TASKSTATS_TYPE_PID: | |
443 | rtid = *(int *) NLA_DATA(na); | |
444 | if (print_delays) | |
445 | printf("PID\t%d\n", rtid); | |
446 | break; | |
447 | case TASKSTATS_TYPE_TGID: | |
448 | rtid = *(int *) NLA_DATA(na); | |
449 | if (print_delays) | |
450 | printf("TGID\t%d\n", rtid); | |
451 | break; | |
452 | case TASKSTATS_TYPE_STATS: | |
453 | count++; | |
454 | if (print_delays) | |
455 | print_delayacct((struct taskstats *) NLA_DATA(na)); | |
cf709844 AM |
456 | if (print_io_accounting) |
457 | print_ioacct((struct taskstats *) NLA_DATA(na)); | |
b663a79c MU |
458 | if (print_task_context_switch_counts) |
459 | task_context_switch_counts((struct taskstats *) NLA_DATA(na)); | |
9e06d3f9 SN |
460 | if (fd) { |
461 | if (write(fd, NLA_DATA(na), na->nla_len) < 0) { | |
462 | err(1,"write error\n"); | |
463 | } | |
464 | } | |
465 | if (!loop) | |
466 | goto done; | |
467 | break; | |
468 | default: | |
d2f7bf13 AM |
469 | fprintf(stderr, "Unknown nested" |
470 | " nla_type %d\n", | |
471 | na->nla_type); | |
9e06d3f9 SN |
472 | break; |
473 | } | |
474 | len2 += NLA_ALIGN(na->nla_len); | |
475 | na = (struct nlattr *) ((char *) na + len2); | |
476 | } | |
477 | break; | |
478 | ||
546040dc BS |
479 | case CGROUPSTATS_TYPE_CGROUP_STATS: |
480 | print_cgroupstats(NLA_DATA(na)); | |
481 | break; | |
9e06d3f9 | 482 | default: |
d2f7bf13 AM |
483 | fprintf(stderr, "Unknown nla_type %d\n", |
484 | na->nla_type); | |
9e06d3f9 | 485 | break; |
a3baf649 | 486 | } |
9e06d3f9 | 487 | na = (struct nlattr *) (GENLMSG_DATA(&msg) + len); |
a3baf649 | 488 | } |
9e06d3f9 SN |
489 | } while (loop); |
490 | done: | |
491 | if (maskset) { | |
492 | rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET, | |
493 | TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK, | |
7d1bdca9 | 494 | &cpumask, strlen(cpumask) + 1); |
9e06d3f9 SN |
495 | printf("Sent deregister mask, retval %d\n", rc); |
496 | if (rc < 0) | |
497 | err(rc, "error sending deregister cpumask\n"); | |
a3baf649 | 498 | } |
9e06d3f9 SN |
499 | err: |
500 | close(nl_sd); | |
501 | if (fd) | |
502 | close(fd); | |
546040dc BS |
503 | if (cfd) |
504 | close(cfd); | |
9e06d3f9 | 505 | return 0; |
a3baf649 | 506 | } |