make getdelays cgroupstats aware
[linux-2.6.git] / Documentation / accounting / getdelays.c
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
8  * Copyright (c) Jay Lan, SGI. 2006
9  *
10  * Compile with
11  *      gcc -I/usr/src/linux/include getdelays.c -o getdelays
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>
24 #include <signal.h>
25
26 #include <linux/genetlink.h>
27 #include <linux/taskstats.h>
28 #include <linux/cgroupstats.h>
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
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;
51 int print_io_accounting;
52 int print_task_context_switch_counts;
53 __u64 stime, utime;
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
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  */
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 "
101                                         "to %d\n",
102                                 rcvbufsz);
103                         return -1;
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
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)
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 + 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;
156 }
157
158
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 {
165         struct {
166                 struct nlmsghdr n;
167                 struct genlmsghdr g;
168                 char buf[256];
169         } ans;
170
171         int id, 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
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;
184
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;
191 }
192
193 void print_delayacct(struct taskstats *t)
194 {
195         printf("\n\nCPU   %15s%15s%15s%15s\n"
196                "      %15llu%15llu%15llu%15llu\n"
197                "IO    %15s%15s\n"
198                "      %15llu%15llu\n"
199                "MEM   %15s%15s\n"
200                "      %15llu%15llu\n",
201                "count", "real total", "virtual total", "delay total",
202                t->cpu_count, t->cpu_run_real_total, t->cpu_run_virtual_total,
203                t->cpu_delay_total,
204                "count", "delay total",
205                t->blkio_count, t->blkio_delay_total,
206                "count", "delay total", t->swapin_count, t->swapin_delay_total);
207 }
208
209 void task_context_switch_counts(struct taskstats *t)
210 {
211         printf("\n\nTask   %15s%15s\n"
212                "       %15lu%15lu\n",
213                "voluntary", "nonvoluntary",
214                t->nvcsw, t->nivcsw);
215 }
216
217 void print_cgroupstats(struct cgroupstats *c)
218 {
219         printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
220                 "uninterruptible %llu\n", c->nr_sleeping, c->nr_io_wait,
221                 c->nr_running, c->nr_stopped, c->nr_uninterruptible);
222 }
223
224
225 void print_ioacct(struct taskstats *t)
226 {
227         printf("%s: read=%llu, write=%llu, cancelled_write=%llu\n",
228                 t->ac_comm,
229                 (unsigned long long)t->read_bytes,
230                 (unsigned long long)t->write_bytes,
231                 (unsigned long long)t->cancelled_write_bytes);
232 }
233
234 int main(int argc, char *argv[])
235 {
236         int c, rc, rep_len, aggr_len, len2, cmd_type;
237         __u16 id;
238         __u32 mypid;
239
240         struct nlattr *na;
241         int nl_sd = -1;
242         int len = 0;
243         pid_t tid = 0;
244         pid_t rtid = 0;
245
246         int fd = 0;
247         int count = 0;
248         int write_file = 0;
249         int maskset = 0;
250         char *logfile = NULL;
251         int loop = 0;
252         int containerset = 0;
253         char containerpath[1024];
254         int cfd = 0;
255
256         struct msgtemplate msg;
257
258         while (1) {
259                 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:");
260                 if (c < 0)
261                         break;
262
263                 switch (c) {
264                 case 'd':
265                         printf("print delayacct stats ON\n");
266                         print_delays = 1;
267                         break;
268                 case 'i':
269                         printf("printing IO accounting\n");
270                         print_io_accounting = 1;
271                         break;
272                 case 'q':
273                         printf("printing task/process context switch rates\n");
274                         print_task_context_switch_counts = 1;
275                         break;
276                 case 'C':
277                         containerset = 1;
278                         strncpy(containerpath, optarg, strlen(optarg) + 1);
279                         break;
280                 case 'w':
281                         logfile = strdup(optarg);
282                         printf("write to file %s\n", logfile);
283                         write_file = 1;
284                         break;
285                 case 'r':
286                         rcvbufsz = atoi(optarg);
287                         printf("receive buf size %d\n", rcvbufsz);
288                         if (rcvbufsz < 0)
289                                 err(1, "Invalid rcv buf size\n");
290                         break;
291                 case 'm':
292                         strncpy(cpumask, optarg, sizeof(cpumask));
293                         maskset = 1;
294                         printf("cpumask %s maskset %d\n", cpumask, maskset);
295                         break;
296                 case 't':
297                         tid = atoi(optarg);
298                         if (!tid)
299                                 err(1, "Invalid tgid\n");
300                         cmd_type = TASKSTATS_CMD_ATTR_TGID;
301                         break;
302                 case 'p':
303                         tid = atoi(optarg);
304                         if (!tid)
305                                 err(1, "Invalid pid\n");
306                         cmd_type = TASKSTATS_CMD_ATTR_PID;
307                         break;
308                 case 'v':
309                         printf("debug on\n");
310                         dbg = 1;
311                         break;
312                 case 'l':
313                         printf("listen forever\n");
314                         loop = 1;
315                         break;
316                 default:
317                         usage();
318                         exit(-1);
319                 }
320         }
321
322         if (write_file) {
323                 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC,
324                           S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
325                 if (fd == -1) {
326                         perror("Cannot open output file\n");
327                         exit(1);
328                 }
329         }
330
331         if ((nl_sd = create_nl_socket(NETLINK_GENERIC)) < 0)
332                 err(1, "error creating Netlink socket\n");
333
334
335         mypid = getpid();
336         id = get_family_id(nl_sd);
337         if (!id) {
338                 fprintf(stderr, "Error getting family id, errno %d\n", errno);
339                 goto err;
340         }
341         PRINTF("family id %d\n", id);
342
343         if (maskset) {
344                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
345                               TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
346                               &cpumask, strlen(cpumask) + 1);
347                 PRINTF("Sent register cpumask, retval %d\n", rc);
348                 if (rc < 0) {
349                         fprintf(stderr, "error sending register cpumask\n");
350                         goto err;
351                 }
352         }
353
354         if (tid && containerset) {
355                 fprintf(stderr, "Select either -t or -C, not both\n");
356                 goto err;
357         }
358
359         if (tid) {
360                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
361                               cmd_type, &tid, sizeof(__u32));
362                 PRINTF("Sent pid/tgid, retval %d\n", rc);
363                 if (rc < 0) {
364                         fprintf(stderr, "error sending tid/tgid cmd\n");
365                         goto done;
366                 }
367         }
368
369         if (containerset) {
370                 cfd = open(containerpath, O_RDONLY);
371                 if (cfd < 0) {
372                         perror("error opening container file");
373                         goto err;
374                 }
375                 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET,
376                               CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32));
377                 if (rc < 0) {
378                         perror("error sending cgroupstats command");
379                         goto err;
380                 }
381         }
382
383         do {
384                 int i;
385
386                 rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
387                 PRINTF("received %d bytes\n", rep_len);
388
389                 if (rep_len < 0) {
390                         fprintf(stderr, "nonfatal reply error: errno %d\n",
391                                 errno);
392                         continue;
393                 }
394                 if (msg.n.nlmsg_type == NLMSG_ERROR ||
395                     !NLMSG_OK((&msg.n), rep_len)) {
396                         struct nlmsgerr *err = NLMSG_DATA(&msg);
397                         fprintf(stderr, "fatal reply error,  errno %d\n",
398                                 err->error);
399                         goto done;
400                 }
401
402                 PRINTF("nlmsghdr size=%d, nlmsg_len=%d, rep_len=%d\n",
403                        sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
404
405
406                 rep_len = GENLMSG_PAYLOAD(&msg.n);
407
408                 na = (struct nlattr *) GENLMSG_DATA(&msg);
409                 len = 0;
410                 i = 0;
411                 while (len < rep_len) {
412                         len += NLA_ALIGN(na->nla_len);
413                         switch (na->nla_type) {
414                         case TASKSTATS_TYPE_AGGR_TGID:
415                                 /* Fall through */
416                         case TASKSTATS_TYPE_AGGR_PID:
417                                 aggr_len = NLA_PAYLOAD(na->nla_len);
418                                 len2 = 0;
419                                 /* For nested attributes, na follows */
420                                 na = (struct nlattr *) NLA_DATA(na);
421                                 done = 0;
422                                 while (len2 < aggr_len) {
423                                         switch (na->nla_type) {
424                                         case TASKSTATS_TYPE_PID:
425                                                 rtid = *(int *) NLA_DATA(na);
426                                                 if (print_delays)
427                                                         printf("PID\t%d\n", rtid);
428                                                 break;
429                                         case TASKSTATS_TYPE_TGID:
430                                                 rtid = *(int *) NLA_DATA(na);
431                                                 if (print_delays)
432                                                         printf("TGID\t%d\n", rtid);
433                                                 break;
434                                         case TASKSTATS_TYPE_STATS:
435                                                 count++;
436                                                 if (print_delays)
437                                                         print_delayacct((struct taskstats *) NLA_DATA(na));
438                                                 if (print_io_accounting)
439                                                         print_ioacct((struct taskstats *) NLA_DATA(na));
440                                                 if (print_task_context_switch_counts)
441                                                         task_context_switch_counts((struct taskstats *) NLA_DATA(na));
442                                                 if (fd) {
443                                                         if (write(fd, NLA_DATA(na), na->nla_len) < 0) {
444                                                                 err(1,"write error\n");
445                                                         }
446                                                 }
447                                                 if (!loop)
448                                                         goto done;
449                                                 break;
450                                         default:
451                                                 fprintf(stderr, "Unknown nested"
452                                                         " nla_type %d\n",
453                                                         na->nla_type);
454                                                 break;
455                                         }
456                                         len2 += NLA_ALIGN(na->nla_len);
457                                         na = (struct nlattr *) ((char *) na + len2);
458                                 }
459                                 break;
460
461                         case CGROUPSTATS_TYPE_CGROUP_STATS:
462                                 print_cgroupstats(NLA_DATA(na));
463                                 break;
464                         default:
465                                 fprintf(stderr, "Unknown nla_type %d\n",
466                                         na->nla_type);
467                                 break;
468                         }
469                         na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
470                 }
471         } while (loop);
472 done:
473         if (maskset) {
474                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
475                               TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
476                               &cpumask, strlen(cpumask) + 1);
477                 printf("Sent deregister mask, retval %d\n", rc);
478                 if (rc < 0)
479                         err(rc, "error sending deregister cpumask\n");
480         }
481 err:
482         close(nl_sd);
483         if (fd)
484                 close(fd);
485         if (cfd)
486                 close(cfd);
487         return 0;
488 }