Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / 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 = 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
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                "SWAP  %15s%15s\n"
200                "      %15llu%15llu\n"
201                "RECLAIM  %12s%15s\n"
202                "      %15llu%15llu\n",
203                "count", "real total", "virtual total", "delay total",
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,
208                "count", "delay total",
209                (unsigned long long)t->blkio_count,
210                (unsigned long long)t->blkio_delay_total,
211                "count", "delay total",
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);
217 }
218
219 void task_context_switch_counts(struct taskstats *t)
220 {
221         printf("\n\nTask   %15s%15s\n"
222                "       %15llu%15llu\n",
223                "voluntary", "nonvoluntary",
224                (unsigned long long)t->nvcsw, (unsigned long long)t->nivcsw);
225 }
226
227 void print_cgroupstats(struct cgroupstats *c)
228 {
229         printf("sleeping %llu, blocked %llu, running %llu, stopped %llu, "
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);
235 }
236
237
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
247 int main(int argc, char *argv[])
248 {
249         int c, rc, rep_len, aggr_len, len2, cmd_type;
250         __u16 id;
251         __u32 mypid;
252
253         struct nlattr *na;
254         int nl_sd = -1;
255         int len = 0;
256         pid_t tid = 0;
257         pid_t rtid = 0;
258
259         int fd = 0;
260         int count = 0;
261         int write_file = 0;
262         int maskset = 0;
263         char *logfile = NULL;
264         int loop = 0;
265         int containerset = 0;
266         char containerpath[1024];
267         int cfd = 0;
268
269         struct msgtemplate msg;
270
271         while (1) {
272                 c = getopt(argc, argv, "qdiw:r:m:t:p:vlC:");
273                 if (c < 0)
274                         break;
275
276                 switch (c) {
277                 case 'd':
278                         printf("print delayacct stats ON\n");
279                         print_delays = 1;
280                         break;
281                 case 'i':
282                         printf("printing IO accounting\n");
283                         print_io_accounting = 1;
284                         break;
285                 case 'q':
286                         printf("printing task/process context switch rates\n");
287                         print_task_context_switch_counts = 1;
288                         break;
289                 case 'C':
290                         containerset = 1;
291                         strncpy(containerpath, optarg, strlen(optarg) + 1);
292                         break;
293                 case 'w':
294                         logfile = strdup(optarg);
295                         printf("write to file %s\n", logfile);
296                         write_file = 1;
297                         break;
298                 case 'r':
299                         rcvbufsz = atoi(optarg);
300                         printf("receive buf size %d\n", rcvbufsz);
301                         if (rcvbufsz < 0)
302                                 err(1, "Invalid rcv buf size\n");
303                         break;
304                 case 'm':
305                         strncpy(cpumask, optarg, sizeof(cpumask));
306                         maskset = 1;
307                         printf("cpumask %s maskset %d\n", cpumask, maskset);
308                         break;
309                 case 't':
310                         tid = atoi(optarg);
311                         if (!tid)
312                                 err(1, "Invalid tgid\n");
313                         cmd_type = TASKSTATS_CMD_ATTR_TGID;
314                         break;
315                 case 'p':
316                         tid = atoi(optarg);
317                         if (!tid)
318                                 err(1, "Invalid pid\n");
319                         cmd_type = TASKSTATS_CMD_ATTR_PID;
320                         break;
321                 case 'v':
322                         printf("debug on\n");
323                         dbg = 1;
324                         break;
325                 case 'l':
326                         printf("listen forever\n");
327                         loop = 1;
328                         break;
329                 default:
330                         usage();
331                         exit(-1);
332                 }
333         }
334
335         if (write_file) {
336                 fd = open(logfile, O_WRONLY | O_CREAT | O_TRUNC,
337                           S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
338                 if (fd == -1) {
339                         perror("Cannot open output file\n");
340                         exit(1);
341                 }
342         }
343
344         if ((nl_sd = create_nl_socket(NETLINK_GENERIC)) < 0)
345                 err(1, "error creating Netlink socket\n");
346
347
348         mypid = getpid();
349         id = get_family_id(nl_sd);
350         if (!id) {
351                 fprintf(stderr, "Error getting family id, errno %d\n", errno);
352                 goto err;
353         }
354         PRINTF("family id %d\n", id);
355
356         if (maskset) {
357                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
358                               TASKSTATS_CMD_ATTR_REGISTER_CPUMASK,
359                               &cpumask, strlen(cpumask) + 1);
360                 PRINTF("Sent register cpumask, retval %d\n", rc);
361                 if (rc < 0) {
362                         fprintf(stderr, "error sending register cpumask\n");
363                         goto err;
364                 }
365         }
366
367         if (tid && containerset) {
368                 fprintf(stderr, "Select either -t or -C, not both\n");
369                 goto err;
370         }
371
372         if (tid) {
373                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
374                               cmd_type, &tid, sizeof(__u32));
375                 PRINTF("Sent pid/tgid, retval %d\n", rc);
376                 if (rc < 0) {
377                         fprintf(stderr, "error sending tid/tgid cmd\n");
378                         goto done;
379                 }
380         }
381
382         if (containerset) {
383                 cfd = open(containerpath, O_RDONLY);
384                 if (cfd < 0) {
385                         perror("error opening container file");
386                         goto err;
387                 }
388                 rc = send_cmd(nl_sd, id, mypid, CGROUPSTATS_CMD_GET,
389                               CGROUPSTATS_CMD_ATTR_FD, &cfd, sizeof(__u32));
390                 if (rc < 0) {
391                         perror("error sending cgroupstats command");
392                         goto err;
393                 }
394         }
395
396         do {
397                 int i;
398
399                 rep_len = recv(nl_sd, &msg, sizeof(msg), 0);
400                 PRINTF("received %d bytes\n", rep_len);
401
402                 if (rep_len < 0) {
403                         fprintf(stderr, "nonfatal reply error: errno %d\n",
404                                 errno);
405                         continue;
406                 }
407                 if (msg.n.nlmsg_type == NLMSG_ERROR ||
408                     !NLMSG_OK((&msg.n), rep_len)) {
409                         struct nlmsgerr *err = NLMSG_DATA(&msg);
410                         fprintf(stderr, "fatal reply error,  errno %d\n",
411                                 err->error);
412                         goto done;
413                 }
414
415                 PRINTF("nlmsghdr size=%zu, nlmsg_len=%d, rep_len=%d\n",
416                        sizeof(struct nlmsghdr), msg.n.nlmsg_len, rep_len);
417
418
419                 rep_len = GENLMSG_PAYLOAD(&msg.n);
420
421                 na = (struct nlattr *) GENLMSG_DATA(&msg);
422                 len = 0;
423                 i = 0;
424                 while (len < rep_len) {
425                         len += NLA_ALIGN(na->nla_len);
426                         switch (na->nla_type) {
427                         case TASKSTATS_TYPE_AGGR_TGID:
428                                 /* Fall through */
429                         case TASKSTATS_TYPE_AGGR_PID:
430                                 aggr_len = NLA_PAYLOAD(na->nla_len);
431                                 len2 = 0;
432                                 /* For nested attributes, na follows */
433                                 na = (struct nlattr *) NLA_DATA(na);
434                                 done = 0;
435                                 while (len2 < aggr_len) {
436                                         switch (na->nla_type) {
437                                         case TASKSTATS_TYPE_PID:
438                                                 rtid = *(int *) NLA_DATA(na);
439                                                 if (print_delays)
440                                                         printf("PID\t%d\n", rtid);
441                                                 break;
442                                         case TASKSTATS_TYPE_TGID:
443                                                 rtid = *(int *) NLA_DATA(na);
444                                                 if (print_delays)
445                                                         printf("TGID\t%d\n", rtid);
446                                                 break;
447                                         case TASKSTATS_TYPE_STATS:
448                                                 count++;
449                                                 if (print_delays)
450                                                         print_delayacct((struct taskstats *) NLA_DATA(na));
451                                                 if (print_io_accounting)
452                                                         print_ioacct((struct taskstats *) NLA_DATA(na));
453                                                 if (print_task_context_switch_counts)
454                                                         task_context_switch_counts((struct taskstats *) NLA_DATA(na));
455                                                 if (fd) {
456                                                         if (write(fd, NLA_DATA(na), na->nla_len) < 0) {
457                                                                 err(1,"write error\n");
458                                                         }
459                                                 }
460                                                 if (!loop)
461                                                         goto done;
462                                                 break;
463                                         default:
464                                                 fprintf(stderr, "Unknown nested"
465                                                         " nla_type %d\n",
466                                                         na->nla_type);
467                                                 break;
468                                         }
469                                         len2 += NLA_ALIGN(na->nla_len);
470                                         na = (struct nlattr *) ((char *) na + len2);
471                                 }
472                                 break;
473
474                         case CGROUPSTATS_TYPE_CGROUP_STATS:
475                                 print_cgroupstats(NLA_DATA(na));
476                                 break;
477                         default:
478                                 fprintf(stderr, "Unknown nla_type %d\n",
479                                         na->nla_type);
480                                 break;
481                         }
482                         na = (struct nlattr *) (GENLMSG_DATA(&msg) + len);
483                 }
484         } while (loop);
485 done:
486         if (maskset) {
487                 rc = send_cmd(nl_sd, id, mypid, TASKSTATS_CMD_GET,
488                               TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK,
489                               &cpumask, strlen(cpumask) + 1);
490                 printf("Sent deregister mask, retval %d\n", rc);
491                 if (rc < 0)
492                         err(rc, "error sending deregister cpumask\n");
493         }
494 err:
495         close(nl_sd);
496         if (fd)
497                 close(fd);
498         if (cfd)
499                 close(cfd);
500         return 0;
501 }