3 * Utility to get per-pid and per-tgid delay accounting statistics
4 * Also illustrates usage of the taskstats interface
6 * Copyright (C) Shailabh Nagar, IBM Corp. 2005
7 * Copyright (C) Balbir Singh, IBM Corp. 2006
8 * Copyright (c) Jay Lan, SGI. 2006
11 * gcc -I/usr/src/linux/include getdelays.c -o getdelays
21 #include <sys/types.h>
23 #include <sys/socket.h>
27 #include <linux/genetlink.h>
28 #include <linux/taskstats.h>
29 #include <linux/cgroupstats.h>
32 * Generic macros for dealing with netlink sockets. Might be duplicated
33 * elsewhere. It is recommended that commercial grade applications use
34 * libnl or libnetlink and use the interfaces provided by the library
36 #define GENLMSG_DATA(glh) ((void *)(NLMSG_DATA(glh) + GENL_HDRLEN))
37 #define GENLMSG_PAYLOAD(glh) (NLMSG_PAYLOAD(glh, 0) - GENL_HDRLEN)
38 #define NLA_DATA(na) ((void *)((char*)(na) + NLA_HDRLEN))
39 #define NLA_PAYLOAD(len) (len - NLA_HDRLEN)
41 #define err(code, fmt, arg...) \
43 fprintf(stderr, fmt, ##arg); \
52 int print_io_accounting
;
53 int print_task_context_switch_counts
;
56 #define PRINTF(fmt, arg...) { \
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 */
70 char buf
[MAX_MSG_SIZE
];
73 char cpumask
[100+6*MAX_CPUS
];
75 static void usage(void)
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");
87 * Create a raw netlink socket and bind
89 static int create_nl_socket(int protocol
)
92 struct sockaddr_nl local
;
94 fd
= socket(AF_NETLINK
, SOCK_RAW
, protocol
);
99 if (setsockopt(fd
, SOL_SOCKET
, SO_RCVBUF
,
100 &rcvbufsz
, sizeof(rcvbufsz
)) < 0) {
101 fprintf(stderr
, "Unable to set socket rcv buf size "
107 memset(&local
, 0, sizeof(local
));
108 local
.nl_family
= AF_NETLINK
;
110 if (bind(fd
, (struct sockaddr
*) &local
, sizeof(local
)) < 0)
120 static int send_cmd(int sd
, __u16 nlmsg_type
, __u32 nlmsg_pid
,
121 __u8 genl_cmd
, __u16 nla_type
,
122 void *nla_data
, int nla_len
)
125 struct sockaddr_nl nladdr
;
129 struct msgtemplate msg
;
131 msg
.n
.nlmsg_len
= NLMSG_LENGTH(GENL_HDRLEN
);
132 msg
.n
.nlmsg_type
= nlmsg_type
;
133 msg
.n
.nlmsg_flags
= NLM_F_REQUEST
;
135 msg
.n
.nlmsg_pid
= nlmsg_pid
;
136 msg
.g
.cmd
= genl_cmd
;
138 na
= (struct nlattr
*) GENLMSG_DATA(&msg
);
139 na
->nla_type
= nla_type
;
140 na
->nla_len
= nla_len
+ 1 + NLA_HDRLEN
;
141 memcpy(NLA_DATA(na
), nla_data
, nla_len
);
142 msg
.n
.nlmsg_len
+= NLMSG_ALIGN(na
->nla_len
);
145 buflen
= msg
.n
.nlmsg_len
;
146 memset(&nladdr
, 0, sizeof(nladdr
));
147 nladdr
.nl_family
= AF_NETLINK
;
148 while ((r
= sendto(sd
, buf
, buflen
, 0, (struct sockaddr
*) &nladdr
,
149 sizeof(nladdr
))) < buflen
) {
153 } else if (errno
!= EAGAIN
)
161 * Probe the controller in genetlink to find the family id
162 * for the TASKSTATS family
164 static int get_family_id(int sd
)
176 strcpy(name
, TASKSTATS_GENL_NAME
);
177 rc
= send_cmd(sd
, GENL_ID_CTRL
, getpid(), CTRL_CMD_GETFAMILY
,
178 CTRL_ATTR_FAMILY_NAME
, (void *)name
,
179 strlen(TASKSTATS_GENL_NAME
)+1);
181 return 0; /* sendto() failure? */
183 rep_len
= recv(sd
, &ans
, sizeof(ans
), 0);
184 if (ans
.n
.nlmsg_type
== NLMSG_ERROR
||
185 (rep_len
< 0) || !NLMSG_OK((&ans
.n
), rep_len
))
188 na
= (struct nlattr
*) GENLMSG_DATA(&ans
);
189 na
= (struct nlattr
*) ((char *) na
+ NLA_ALIGN(na
->nla_len
));
190 if (na
->nla_type
== CTRL_ATTR_FAMILY_ID
) {
191 id
= *(__u16
*) NLA_DATA(na
);
196 static void print_delayacct(struct taskstats
*t
)
198 printf("\n\nCPU %15s%15s%15s%15s\n"
199 " %15llu%15llu%15llu%15llu\n"
206 "count", "real total", "virtual total", "delay total",
207 (unsigned long long)t
->cpu_count
,
208 (unsigned long long)t
->cpu_run_real_total
,
209 (unsigned long long)t
->cpu_run_virtual_total
,
210 (unsigned long long)t
->cpu_delay_total
,
211 "count", "delay total",
212 (unsigned long long)t
->blkio_count
,
213 (unsigned long long)t
->blkio_delay_total
,
214 "count", "delay total",
215 (unsigned long long)t
->swapin_count
,
216 (unsigned long long)t
->swapin_delay_total
,
217 "count", "delay total",
218 (unsigned long long)t
->freepages_count
,
219 (unsigned long long)t
->freepages_delay_total
);
222 static void task_context_switch_counts(struct taskstats
*t
)
224 printf("\n\nTask %15s%15s\n"
226 "voluntary", "nonvoluntary",
227 (unsigned long long)t
->nvcsw
, (unsigned long long)t
->nivcsw
);
230 static void print_cgroupstats(struct cgroupstats
*c
)
232 printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
233 "uninterruptible %llu\n", (unsigned long long)c
->nr_sleeping
,
234 (unsigned long long)c
->nr_io_wait
,
235 (unsigned long long)c
->nr_running
,
236 (unsigned long long)c
->nr_stopped
,
237 (unsigned long long)c
->nr_uninterruptible
);
241 static void print_ioacct(struct taskstats
*t
)
243 printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n",
245 (unsigned long long)t
->read_bytes
,
246 (unsigned long long)t
->write_bytes
,
247 (unsigned long long)t
->cancelled_write_bytes
);
250 int main(int argc
, char *argv
[])
252 int c
, rc
, rep_len
, aggr_len
, len2
;
253 int cmd_type
= TASKSTATS_CMD_ATTR_UNSPEC
;
267 char *logfile
= NULL
;
269 int containerset
= 0;
270 char containerpath
[1024];
275 struct msgtemplate msg
;
278 c
= getopt(argc
, argv
, "qdiw:r:m:t:p:vlC:c:");
284 printf("print delayacct stats ON\n");
288 printf("printing IO accounting\n");
289 print_io_accounting
= 1;
292 printf("printing task/process context switch rates\n");
293 print_task_context_switch_counts
= 1;
297 strncpy(containerpath
, optarg
, strlen(optarg
) + 1);
300 logfile
= strdup(optarg
);
301 printf("write to file %s\n", logfile
);
305 rcvbufsz
= atoi(optarg
);
306 printf("receive buf size %d\n", rcvbufsz
);
308 err(1, "Invalid rcv buf size\n");
311 strncpy(cpumask
, optarg
, sizeof(cpumask
));
313 printf("cpumask %s maskset %d\n", cpumask
, maskset
);
318 err(1, "Invalid tgid\n");
319 cmd_type
= TASKSTATS_CMD_ATTR_TGID
;
324 err(1, "Invalid pid\n");
325 cmd_type
= TASKSTATS_CMD_ATTR_PID
;
329 /* Block SIGCHLD for sigwait() later */
330 if (sigemptyset(&sigset
) == -1)
331 err(1, "Failed to empty sigset");
332 if (sigaddset(&sigset
, SIGCHLD
))
333 err(1, "Failed to set sigchld in sigset");
334 sigprocmask(SIG_BLOCK
, &sigset
, NULL
);
336 /* fork/exec a child */
339 err(1, "Fork failed\n");
341 if (execvp(argv
[optind
- 1],
342 &argv
[optind
- 1]) < 0)
345 /* Set the command type and avoid further processing */
346 cmd_type
= TASKSTATS_CMD_ATTR_PID
;
350 printf("debug on\n");
354 printf("listen forever\n");
364 fd
= open(logfile
, O_WRONLY
| O_CREAT
| O_TRUNC
,
365 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IROTH
);
367 perror("Cannot open output file\n");
372 if ((nl_sd
= create_nl_socket(NETLINK_GENERIC
)) < 0)
373 err(1, "error creating Netlink socket\n");
377 id
= get_family_id(nl_sd
);
379 fprintf(stderr
, "Error getting family id, errno %d\n", errno
);
382 PRINTF("family id %d\n", id
);
385 rc
= send_cmd(nl_sd
, id
, mypid
, TASKSTATS_CMD_GET
,
386 TASKSTATS_CMD_ATTR_REGISTER_CPUMASK
,
387 &cpumask
, strlen(cpumask
) + 1);
388 PRINTF("Sent register cpumask, retval %d\n", rc
);
390 fprintf(stderr
, "error sending register cpumask\n");
395 if (tid
&& containerset
) {
396 fprintf(stderr
, "Select either -t or -C, not both\n");
401 * If we forked a child, wait for it to exit. Cannot use waitpid()
402 * as all the delicious data would be reaped as part of the wait
404 if (tid
&& forking
) {
406 sigwait(&sigset
, &sig_received
);
410 rc
= send_cmd(nl_sd
, id
, mypid
, TASKSTATS_CMD_GET
,
411 cmd_type
, &tid
, sizeof(__u32
));
412 PRINTF("Sent pid/tgid, retval %d\n", rc
);
414 fprintf(stderr
, "error sending tid/tgid cmd\n");
420 cfd
= open(containerpath
, O_RDONLY
);
422 perror("error opening container file");
425 rc
= send_cmd(nl_sd
, id
, mypid
, CGROUPSTATS_CMD_GET
,
426 CGROUPSTATS_CMD_ATTR_FD
, &cfd
, sizeof(__u32
));
428 perror("error sending cgroupstats command");
432 if (!maskset
&& !tid
&& !containerset
) {
438 rep_len
= recv(nl_sd
, &msg
, sizeof(msg
), 0);
439 PRINTF("received %d bytes\n", rep_len
);
442 fprintf(stderr
, "nonfatal reply error: errno %d\n",
446 if (msg
.n
.nlmsg_type
== NLMSG_ERROR
||
447 !NLMSG_OK((&msg
.n
), rep_len
)) {
448 struct nlmsgerr
*err
= NLMSG_DATA(&msg
);
449 fprintf(stderr
, "fatal reply error, errno %d\n",
454 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
455 sizeof(struct nlmsghdr
), msg
.n
.nlmsg_len
, rep_len
);
458 rep_len
= GENLMSG_PAYLOAD(&msg
.n
);
460 na
= (struct nlattr
*) GENLMSG_DATA(&msg
);
462 while (len
< rep_len
) {
463 len
+= NLA_ALIGN(na
->nla_len
);
464 switch (na
->nla_type
) {
465 case TASKSTATS_TYPE_AGGR_TGID
:
467 case TASKSTATS_TYPE_AGGR_PID
:
468 aggr_len
= NLA_PAYLOAD(na
->nla_len
);
470 /* For nested attributes, na follows */
471 na
= (struct nlattr
*) NLA_DATA(na
);
473 while (len2
< aggr_len
) {
474 switch (na
->nla_type
) {
475 case TASKSTATS_TYPE_PID
:
476 rtid
= *(int *) NLA_DATA(na
);
478 printf("PID\t%d\n", rtid
);
480 case TASKSTATS_TYPE_TGID
:
481 rtid
= *(int *) NLA_DATA(na
);
483 printf("TGID\t%d\n", rtid
);
485 case TASKSTATS_TYPE_STATS
:
488 print_delayacct((struct taskstats
*) NLA_DATA(na
));
489 if (print_io_accounting
)
490 print_ioacct((struct taskstats
*) NLA_DATA(na
));
491 if (print_task_context_switch_counts
)
492 task_context_switch_counts((struct taskstats
*) NLA_DATA(na
));
494 if (write(fd
, NLA_DATA(na
), na
->nla_len
) < 0) {
495 err(1,"write error\n");
502 fprintf(stderr
, "Unknown nested"
507 len2
+= NLA_ALIGN(na
->nla_len
);
508 na
= (struct nlattr
*) ((char *) na
+ len2
);
512 case CGROUPSTATS_TYPE_CGROUP_STATS
:
513 print_cgroupstats(NLA_DATA(na
));
516 fprintf(stderr
, "Unknown nla_type %d\n",
518 case TASKSTATS_TYPE_NULL
:
521 na
= (struct nlattr
*) (GENLMSG_DATA(&msg
) + len
);
526 rc
= send_cmd(nl_sd
, id
, mypid
, TASKSTATS_CMD_GET
,
527 TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK
,
528 &cpumask
, strlen(cpumask
) + 1);
529 printf("Sent deregister mask, retval %d\n", rc
);
531 err(rc
, "error sending deregister cpumask\n");