2 * block queue tracing application
4 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 #include <sys/types.h>
29 #include <sys/ioctl.h>
30 #include <sys/param.h>
31 #include <sys/statfs.h>
34 #include <sys/socket.h>
41 #include <netinet/in.h>
42 #include <arpa/inet.h>
44 #include <sys/sendfile.h>
49 static char blktrace_version[] = "0.99";
52 * You may want to increase this even more, if you are logging at a high
53 * rate and see skipped/missed events
55 #define BUF_SIZE (512 * 1024)
58 #define OFILE_BUF (128 * 1024)
60 #define RELAYFS_TYPE 0xF0B4A981
62 #define S_OPTS "d:a:A:r:o:kw:Vb:n:D:lh:p:s"
63 static struct option l_opts[] = {
66 .has_arg = required_argument,
72 .has_arg = required_argument,
78 .has_arg = required_argument,
84 .has_arg = required_argument,
90 .has_arg = required_argument,
96 .has_arg = no_argument,
102 .has_arg = required_argument,
108 .has_arg = no_argument,
113 .name = "buffer-size",
114 .has_arg = required_argument,
119 .name = "num-sub-buffers",
120 .has_arg = required_argument,
125 .name = "output-dir",
126 .has_arg = required_argument,
132 .has_arg = no_argument,
138 .has_arg = required_argument,
144 .has_arg = required_argument,
150 .has_arg = no_argument,
162 unsigned int max_len;
165 #define FIFO_SIZE (1024) /* should be plenty big! */
166 #define CL_SIZE (128) /* cache line, any bigger? */
168 struct tip_subbuf_fifo {
169 int tail __attribute__((aligned(CL_SIZE)));
170 int head __attribute__((aligned(CL_SIZE)));
171 struct tip_subbuf *q[FIFO_SIZE];
174 struct thread_information {
180 char fn[MAXPATHLEN + 64];
182 struct in_addr cl_in_addr;
190 int (*get_subbuf)(struct thread_information *, unsigned int);
191 int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
192 int (*read_data)(struct thread_information *, void *, unsigned int);
194 unsigned long events_processed;
195 unsigned long long data_read;
196 unsigned long long data_queued;
197 struct device_information *device;
204 struct tip_subbuf_fifo fifo;
205 struct tip_subbuf *leftover_ts;
208 * mmap controlled output files
210 unsigned long long fs_size;
211 unsigned long long fs_max_size;
212 unsigned long fs_off;
214 unsigned long fs_buf_len;
217 struct device_information {
221 volatile int trace_started;
222 unsigned long drop_count;
223 struct thread_information *threads;
227 static struct thread_information *thread_information;
229 static struct device_information *device_information;
231 /* command line option globals */
232 static char *relay_path;
233 static char *output_name;
234 static char *output_dir;
235 static int act_mask = ~0U;
236 static int kill_running_trace;
237 static unsigned long buf_size = BUF_SIZE;
238 static unsigned long buf_nr = BUF_NR;
239 static unsigned int page_size;
241 #define is_done() (*(volatile int *)(&done))
242 static volatile int done;
244 #define is_trace_stopped() (*(volatile int *)(&trace_stopped))
245 static volatile int trace_stopped;
247 #define is_stat_shown() (*(volatile int *)(&stat_shown))
248 static volatile int stat_shown;
250 int data_is_native = -1;
252 static void exit_trace(int status);
254 #define dip_tracing(dip) (*(volatile int *)(&(dip)->trace_started))
255 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
257 #define __for_each_dip(__d, __i, __e) \
258 for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
260 #define for_each_dip(__d, __i) __for_each_dip(__d, __i, ndevs)
261 #define for_each_tip(__d, __t, __j) \
262 for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
265 * networking stuff follows. we include a magic number so we know whether
266 * to endianness convert or not
268 struct blktrace_net_hdr {
269 u32 magic; /* same as trace magic */
270 char buts_name[32]; /* trace name */
271 u32 cpu; /* for which cpu */
273 u32 len; /* length of following trace data */
276 #define TRACE_NET_PORT (8462)
285 * network cmd line params
287 static char hostname[MAXHOSTNAMELEN];
288 static int net_port = TRACE_NET_PORT;
289 static int net_mode = 0;
290 static int net_use_sendfile;
292 static int net_in_fd = -1;
293 static int net_out_fd = -1;
295 static void handle_sigint(__attribute__((__unused__)) int sig)
297 struct device_information *dip;
301 * stop trace so we can reap currently produced data
303 for_each_dip(dip, i) {
306 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
307 perror("BLKTRACESTOP");
313 static int get_dropped_count(const char *buts_name)
316 char tmp[MAXPATHLEN + 64];
318 snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
319 relay_path, buts_name);
321 fd = open(tmp, O_RDONLY);
324 * this may be ok, if the kernel doesn't support dropped counts
329 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
333 if (read(fd, tmp, sizeof(tmp)) < 0) {
344 static int start_trace(struct device_information *dip)
346 struct blk_user_trace_setup buts;
348 memset(&buts, 0, sizeof(buts));
349 buts.buf_size = buf_size;
350 buts.buf_nr = buf_nr;
351 buts.act_mask = act_mask;
353 if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
354 perror("BLKTRACESETUP");
358 if (ioctl(dip->fd, BLKTRACESTART) < 0) {
359 perror("BLKTRACESTART");
363 memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
364 dip_set_tracing(dip, 1);
368 static void stop_trace(struct device_information *dip)
370 if (dip_tracing(dip) || kill_running_trace) {
371 dip_set_tracing(dip, 0);
374 * should be stopped, just don't complain if it isn't
376 ioctl(dip->fd, BLKTRACESTOP);
378 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
379 perror("BLKTRACETEARDOWN");
386 static void stop_all_traces(void)
388 struct device_information *dip;
391 for_each_dip(dip, i) {
392 dip->drop_count = get_dropped_count(dip->buts_name);
397 static void wait_for_data(struct thread_information *tip, int events)
399 struct pollfd pfd = { .fd = tip->fd, .events = events };
402 if (poll(&pfd, 1, 100) < 0) {
406 if (pfd.revents & events)
408 if (tip->ofile_stdout)
410 } while (!is_done());
413 static int read_data_file(struct thread_information *tip, void *buf,
419 wait_for_data(tip, POLLIN);
421 ret = read(tip->fd, buf, len);
427 if (errno != EAGAIN) {
429 fprintf(stderr,"Thread %d failed read of %s\n",
435 } while (!is_done());
441 static int read_data_net(struct thread_information *tip, void *buf,
444 unsigned int bytes_left = len;
448 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
453 if (errno != EAGAIN) {
455 fprintf(stderr, "server: failed read\n");
463 } while (!is_done() && bytes_left);
465 return len - bytes_left;
468 static inline struct tip_subbuf *
469 subbuf_fifo_dequeue(struct thread_information *tip)
471 const int head = tip->fifo.head;
472 const int next = (head + 1) & (FIFO_SIZE - 1);
474 if (head != tip->fifo.tail) {
475 struct tip_subbuf *ts = tip->fifo.q[head];
478 tip->fifo.head = next;
485 static inline int subbuf_fifo_queue(struct thread_information *tip,
486 struct tip_subbuf *ts)
488 const int tail = tip->fifo.tail;
489 const int next = (tail + 1) & (FIFO_SIZE - 1);
491 if (next != tip->fifo.head) {
492 tip->fifo.q[tail] = ts;
494 tip->fifo.tail = next;
498 fprintf(stderr, "fifo too small!\n");
503 * For file output, truncate and mmap the file appropriately
505 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
507 int ofd = fileno(tip->ofile);
511 * extend file, if we have to. use chunks of 16 subbuffers.
513 if (tip->fs_off + buf_size > tip->fs_buf_len) {
515 munlock(tip->fs_buf, tip->fs_buf_len);
516 munmap(tip->fs_buf, tip->fs_buf_len);
520 tip->fs_off = tip->fs_size & (page_size - 1);
521 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
522 tip->fs_max_size += tip->fs_buf_len;
524 if (ftruncate(ofd, tip->fs_max_size) < 0) {
529 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
530 MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
531 if (tip->fs_buf == MAP_FAILED) {
535 mlock(tip->fs_buf, tip->fs_buf_len);
538 ret = tip->read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
540 tip->data_read += ret;
550 * Use the copy approach for pipes and network
552 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
554 struct tip_subbuf *ts = malloc(sizeof(*ts));
557 ts->buf = malloc(buf_size);
558 ts->max_len = maxlen;
560 ret = tip->read_data(tip, ts->buf, ts->max_len);
563 tip->data_read += ret;
564 if (subbuf_fifo_queue(tip, ts))
571 static int get_subbuf_sendfile(struct thread_information *tip,
574 struct tip_subbuf *ts;
578 wait_for_data(tip, POLLMSG);
581 * hack to get last data out, we can't use sendfile for that
584 return get_subbuf(tip, maxlen);
586 if (fstat(tip->fd, &sb) < 0) {
587 perror("trace stat");
590 ready = sb.st_size - tip->data_queued;
596 ts = malloc(sizeof(*ts));
600 tip->data_queued += ready;
602 if (subbuf_fifo_queue(tip, ts))
608 static void close_thread(struct thread_information *tip)
614 if (tip->ofile_buffer)
615 free(tip->ofile_buffer);
621 tip->ofile_buffer = NULL;
625 static void tip_ftrunc_final(struct thread_information *tip)
628 * truncate to right size and cleanup mmap
630 if (tip->ofile_mmap) {
631 int ofd = fileno(tip->ofile);
634 munmap(tip->fs_buf, tip->fs_buf_len);
636 ftruncate(ofd, tip->fs_size);
640 static void *thread_main(void *arg)
642 struct thread_information *tip = arg;
643 pid_t pid = getpid();
647 CPU_SET((tip->cpu), &cpu_mask);
649 if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
650 perror("sched_setaffinity");
654 snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
655 relay_path, tip->device->buts_name, tip->cpu);
656 tip->fd = open(tip->fn, O_RDONLY);
659 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
665 if (tip->get_subbuf(tip, buf_size) < 0)
670 * trace is stopped, pull data until we get a short read
672 while (tip->get_subbuf(tip, buf_size) > 0)
675 tip_ftrunc_final(tip);
680 static int write_data_net(int fd, void *buf, unsigned int buf_len)
682 unsigned int bytes_left = buf_len;
686 ret = send(fd, buf, bytes_left, 0);
699 static int net_send_header(struct thread_information *tip, unsigned int len)
701 struct blktrace_net_hdr hdr;
703 hdr.magic = BLK_IO_TRACE_MAGIC;
704 strcpy(hdr.buts_name, tip->device->buts_name);
706 hdr.max_cpus = ncpus;
709 return write_data_net(net_out_fd, &hdr, sizeof(hdr));
713 * send header with 0 length to signal end-of-run
715 static void net_client_send_close(void)
717 struct device_information *dip;
718 struct blktrace_net_hdr hdr;
721 hdr.magic = BLK_IO_TRACE_MAGIC;
723 hdr.max_cpus = ncpus;
726 for_each_dip(dip, i) {
727 strcpy(hdr.buts_name, dip->buts_name);
728 hdr.cpu += get_dropped_count(dip->buts_name);
731 write_data_net(net_out_fd, &hdr, sizeof(hdr));
734 static int flush_subbuf_net(struct thread_information *tip,
735 struct tip_subbuf *ts)
737 if (net_send_header(tip, ts->len))
739 if (write_data_net(net_out_fd, ts->buf, ts->len))
747 static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
749 int ret = sendfile(net_out_fd, tip->fd, NULL, ts->len);
754 } else if (ret < (int) ts->len) {
755 fprintf(stderr, "short sendfile send (%d of %d)\n", ret, ts->len);
762 static int flush_subbuf_sendfile(struct thread_information *tip,
763 struct tip_subbuf *ts)
768 * currently we cannot use sendfile() on the last bytes read, as they
769 * may not be a full subbuffer. get_subbuf_sendfile() falls back to
770 * the read approach for those, so use send() to ship them out
773 return flush_subbuf_net(tip, ts);
775 if (net_send_header(tip, ts->len))
777 if (net_sendfile(tip, ts))
780 tip->data_read += ts->len;
781 tip->ofile_offset += buf_size;
788 static int write_data(struct thread_information *tip, void *buf,
789 unsigned int buf_len)
797 ret = fwrite(buf, buf_len, 1, tip->ofile);
807 if (tip->ofile_stdout)
813 static int flush_subbuf_file(struct thread_information *tip,
814 struct tip_subbuf *ts)
816 unsigned int offset = 0;
817 struct blk_io_trace *t;
818 int pdu_len, events = 0;
821 * surplus from last run
823 if (tip->leftover_ts) {
824 struct tip_subbuf *prev_ts = tip->leftover_ts;
826 if (prev_ts->len + ts->len > prev_ts->max_len) {
827 prev_ts->max_len += ts->len;
828 prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
831 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
832 prev_ts->len += ts->len;
838 tip->leftover_ts = NULL;
841 while (offset + sizeof(*t) <= ts->len) {
842 t = ts->buf + offset;
844 if (verify_trace(t)) {
845 write_data(tip, ts->buf, offset);
849 pdu_len = t->pdu_len;
851 if (offset + sizeof(*t) + pdu_len > ts->len)
854 offset += sizeof(*t) + pdu_len;
855 tip->events_processed++;
856 tip->data_read += sizeof(*t) + pdu_len;
860 if (write_data(tip, ts->buf, offset))
864 * leftover bytes, save them for next time
866 if (offset != ts->len) {
867 tip->leftover_ts = ts;
869 memmove(ts->buf, ts->buf + offset, ts->len);
878 static int write_tip_events(struct thread_information *tip)
880 struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
883 return tip->flush_subbuf(tip, ts);
889 * scans the tips we know and writes out the subbuffers we accumulate
891 static void get_and_write_events(void)
893 struct device_information *dip;
894 struct thread_information *tip;
895 int i, j, events, ret, tips_running;
900 for_each_dip(dip, i) {
901 for_each_tip(dip, tip, j) {
902 ret = write_tip_events(tip);
918 for_each_dip(dip, i) {
919 for_each_tip(dip, tip, j) {
920 ret = write_tip_events(tip);
923 tips_running += !tip->exited;
927 } while (events || tips_running);
930 static void wait_for_threads(void)
933 * for piped or network output, poll and fetch data for writeout.
934 * for files, we just wait around for trace threads to exit
936 if ((output_name && !strcmp(output_name, "-")) ||
937 net_mode == Net_client)
938 get_and_write_events();
940 struct device_information *dip;
941 struct thread_information *tip;
942 int i, j, tips_running;
949 for_each_tip(dip, tip, j)
950 tips_running += !tip->exited;
951 } while (tips_running);
954 if (net_mode == Net_client)
955 net_client_send_close();
958 static int fill_ofname(struct thread_information *tip, char *dst,
966 len = sprintf(dst, "%s/", output_dir);
968 if (net_mode == Net_server) {
969 len += sprintf(dst + len, "%s-", inet_ntoa(tip->cl_in_addr));
971 len += strftime(dst + len, 64, "%F-%T/", gmtime(&t));
974 if (stat(dst, &sb) < 0) {
975 if (errno != ENOENT) {
979 if (mkdir(dst, 0755) < 0) {
981 fprintf(stderr, "Can't make output dir\n");
987 sprintf(dst + len, "%s.blktrace.%d", output_name, tip->cpu);
989 sprintf(dst + len, "%s.blktrace.%d", buts_name, tip->cpu);
994 static void fill_ops(struct thread_information *tip)
999 if (net_mode == Net_client) {
1000 if (net_use_sendfile) {
1001 tip->get_subbuf = get_subbuf_sendfile;
1002 tip->flush_subbuf = flush_subbuf_sendfile;
1004 tip->get_subbuf = get_subbuf;
1005 tip->flush_subbuf = flush_subbuf_net;
1008 if (tip->ofile_mmap)
1009 tip->get_subbuf = mmap_subbuf;
1011 tip->get_subbuf = get_subbuf;
1013 tip->flush_subbuf = flush_subbuf_file;
1016 if (net_mode == Net_server)
1017 tip->read_data = read_data_net;
1019 tip->read_data = read_data_file;
1022 static int tip_open_output(struct device_information *dip,
1023 struct thread_information *tip)
1025 int pipeline = output_name && !strcmp(output_name, "-");
1026 int mode, vbuf_size;
1029 if (net_mode == Net_client) {
1031 tip->ofile_stdout = 0;
1032 tip->ofile_mmap = 0;
1034 } else if (pipeline) {
1035 tip->ofile = fdopen(STDOUT_FILENO, "w");
1036 tip->ofile_stdout = 1;
1037 tip->ofile_mmap = 0;
1041 if (fill_ofname(tip, op, dip->buts_name))
1043 tip->ofile = fopen(op, "w+");
1044 tip->ofile_stdout = 0;
1045 tip->ofile_mmap = 1;
1047 vbuf_size = OFILE_BUF;
1050 if (tip->ofile == NULL) {
1055 tip->ofile_buffer = malloc(vbuf_size);
1056 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1067 static int start_threads(struct device_information *dip)
1069 struct thread_information *tip;
1072 for_each_tip(dip, tip, j) {
1075 tip->events_processed = 0;
1077 memset(&tip->fifo, 0, sizeof(tip->fifo));
1078 tip->leftover_ts = NULL;
1080 if (tip_open_output(dip, tip))
1083 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1084 perror("pthread_create");
1093 static void stop_threads(struct device_information *dip)
1095 struct thread_information *tip;
1099 for_each_tip(dip, tip, i) {
1100 (void) pthread_join(tip->thread, (void *) &ret);
1105 static void stop_all_threads(void)
1107 struct device_information *dip;
1110 for_each_dip(dip, i)
1114 static void stop_all_tracing(void)
1116 struct device_information *dip;
1119 for_each_dip(dip, i)
1123 static void exit_trace(int status)
1125 if (!is_trace_stopped()) {
1134 static int resize_devices(char *path)
1136 int size = (ndevs + 1) * sizeof(struct device_information);
1138 device_information = realloc(device_information, size);
1139 if (!device_information) {
1140 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1143 device_information[ndevs].path = path;
1148 static int open_devices(void)
1150 struct device_information *dip;
1153 for_each_dip(dip, i) {
1154 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1164 static int start_devices(void)
1166 struct device_information *dip;
1169 size = ncpus * sizeof(struct thread_information);
1170 thread_information = malloc(size * ndevs);
1171 if (!thread_information) {
1172 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1176 for_each_dip(dip, i) {
1177 if (start_trace(dip)) {
1179 fprintf(stderr, "Failed to start trace on %s\n",
1186 __for_each_dip(dip, j, i)
1192 for_each_dip(dip, i) {
1193 dip->threads = thread_information + (i * ncpus);
1194 if (start_threads(dip)) {
1195 fprintf(stderr, "Failed to start worker threads\n");
1201 __for_each_dip(dip, j, i)
1203 for_each_dip(dip, i)
1212 static void show_stats(void)
1214 struct device_information *dip;
1215 struct thread_information *tip;
1216 unsigned long long events_processed, data_read;
1217 unsigned long total_drops;
1218 int i, j, no_stdout = 0;
1220 if (is_stat_shown())
1223 if (output_name && !strcmp(output_name, "-"))
1229 for_each_dip(dip, i) {
1231 printf("Device: %s\n", dip->path);
1232 events_processed = 0;
1234 for_each_tip(dip, tip, j) {
1236 printf(" CPU%3d: %20lu events, %8llu KiB data\n",
1237 tip->cpu, tip->events_processed,
1238 (tip->data_read + 1023) >> 10);
1239 events_processed += tip->events_processed;
1240 data_read += tip->data_read;
1242 total_drops += dip->drop_count;
1244 printf(" Total: %20llu events (dropped %lu), %8llu KiB data\n",
1245 events_processed, dip->drop_count,
1246 (data_read + 1023) >> 10);
1250 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1253 static struct device_information *net_get_dip(char *buts_name,
1254 struct in_addr *cl_in_addr)
1256 struct device_information *dip;
1259 for (i = 0; i < ndevs; i++) {
1260 dip = &device_information[i];
1262 if (!strcmp(dip->buts_name, buts_name))
1266 device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1267 dip = &device_information[ndevs];
1268 memset(dip, 0, sizeof(*dip));
1270 strcpy(dip->buts_name, buts_name);
1271 dip->path = strdup(buts_name);
1273 dip->threads = malloc(ncpus * sizeof(struct thread_information));
1274 memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1279 for (i = 0; i < ncpus; i++) {
1280 struct thread_information *tip = &dip->threads[i];
1285 tip->cl_in_addr = *cl_in_addr;
1287 if (tip_open_output(dip, tip))
1294 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh,
1295 struct in_addr *cl_in_addr)
1297 struct device_information *dip;
1299 ncpus = bnh->max_cpus;
1300 dip = net_get_dip(bnh->buts_name, cl_in_addr);
1301 return &dip->threads[bnh->cpu];
1304 static int net_get_header(struct blktrace_net_hdr *bnh)
1306 int fl = fcntl(net_in_fd, F_GETFL);
1307 int bytes_left, ret;
1310 fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1311 bytes_left = sizeof(*bnh);
1312 while (bytes_left && !is_done()) {
1313 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1315 if (errno != EAGAIN) {
1316 perror("recv header");
1329 fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1333 static int net_server_loop(struct in_addr *cl_in_addr)
1335 struct thread_information *tip;
1336 struct blktrace_net_hdr bnh;
1338 if (net_get_header(&bnh))
1341 if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1342 fprintf(stderr, "server: received data is bad\n");
1346 if (!data_is_native) {
1347 bnh.magic = be32_to_cpu(bnh.magic);
1348 bnh.cpu = be32_to_cpu(bnh.cpu);
1349 bnh.len = be32_to_cpu(bnh.len);
1352 if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1353 fprintf(stderr, "server: bad data magic\n");
1358 * len == 0 means that the other end signalled end-of-run
1362 * overload cpu count with dropped events
1364 struct device_information *dip;
1366 dip = net_get_dip(bnh.buts_name, cl_in_addr);
1367 dip->drop_count = bnh.cpu;
1369 fprintf(stderr, "server: end of run\n");
1373 tip = net_get_tip(&bnh, cl_in_addr);
1377 if (mmap_subbuf(tip, bnh.len))
1383 static int get_connection(int fd, struct sockaddr_in *addr)
1385 struct pollfd pfd = { .fd = fd, .events = POLLIN };
1388 printf("blktrace: waiting for incoming connection...\n");
1390 if (poll(&pfd, 1, -1) < 0) {
1391 perror("poll for connection");
1394 if ((pfd.revents & POLLIN) == 0)
1397 socklen = sizeof(*addr);
1398 net_in_fd = accept(fd, (struct sockaddr *) addr, &socklen);
1399 if (net_in_fd < 0) {
1404 printf("blktrace: connection from %s\n", inet_ntoa(addr->sin_addr));
1409 * Start here when we are in server mode - just fetch data from the network
1412 static int net_server(void)
1414 struct device_information *dip;
1415 struct thread_information *tip;
1416 struct sockaddr_in addr;
1419 fd = socket(AF_INET, SOCK_STREAM, 0);
1421 perror("server: socket");
1426 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1427 perror("setsockopt");
1431 memset(&addr, 0, sizeof(addr));
1432 addr.sin_family = AF_INET;
1433 addr.sin_addr.s_addr = htonl(INADDR_ANY);
1434 addr.sin_port = htons(net_port);
1436 if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1441 if (listen(fd, 1) < 0) {
1447 if (get_connection(fd, &addr))
1450 while (!is_done()) {
1451 if (net_server_loop(&addr.sin_addr))
1455 for_each_dip(dip, i)
1456 for_each_tip(dip, tip, j)
1457 tip_ftrunc_final(tip);
1465 * cleanup for next run
1467 for_each_dip(dip, i) {
1468 for_each_tip(dip, tip, j)
1475 free(device_information);
1476 device_information = NULL;
1486 * Setup outgoing network connection where we will transmit data
1488 static int net_setup_client(void)
1490 struct sockaddr_in addr;
1493 fd = socket(AF_INET, SOCK_STREAM, 0);
1495 perror("client: socket");
1499 memset(&addr, 0, sizeof(addr));
1500 addr.sin_family = AF_INET;
1501 addr.sin_port = htons(net_port);
1503 if (inet_aton(hostname, &addr.sin_addr) != 1) {
1504 struct hostent *hent = gethostbyname(hostname);
1506 perror("gethostbyname");
1510 memcpy(&addr.sin_addr, hent->h_addr, 4);
1511 strcpy(hostname, hent->h_name);
1514 printf("blktrace: connecting to %s\n", hostname);
1516 if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1517 perror("client: connect");
1521 printf("blktrace: connected!\n");
1526 static char usage_str[] = \
1527 "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1528 "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1529 "\t-d Use specified device. May also be given last after options\n" \
1530 "\t-r Path to mounted relayfs, defaults to /relay\n" \
1531 "\t-o File(s) to send output to\n" \
1532 "\t-D Directory to prepend to output file names\n" \
1533 "\t-k Kill a running trace\n" \
1534 "\t-w Stop after defined time, in seconds\n" \
1535 "\t-a Only trace specified actions. See documentation\n" \
1536 "\t-A Give trace mask as a single value. See documentation\n" \
1537 "\t-b Sub buffer size in KiB\n" \
1538 "\t-n Number of sub buffers\n" \
1539 "\t-l Run in network listen mode (blktrace server)\n" \
1540 "\t-h Run in network client mode, connecting to the given host\n" \
1541 "\t-p Network port to use (default 8462)\n" \
1542 "\t-s Make the network client use sendfile() to transfer data\n" \
1543 "\t-V Print program version info\n\n";
1545 static void show_usage(char *program)
1547 fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1550 int main(int argc, char *argv[])
1552 static char default_relay_path[] = "/relay";
1556 int act_mask_tmp = 0;
1558 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1561 i = find_mask_map(optarg);
1563 fprintf(stderr,"Invalid action mask %s\n",
1571 if ((sscanf(optarg, "%x", &i) != 1) ||
1572 !valid_act_opt(i)) {
1574 "Invalid set action mask %s/0x%x\n",
1582 if (resize_devices(optarg) != 0)
1587 relay_path = optarg;
1591 output_name = optarg;
1594 kill_running_trace = 1;
1597 stop_watch = atoi(optarg);
1598 if (stop_watch <= 0) {
1600 "Invalid stopwatch value (%d secs)\n",
1606 printf("%s version %s\n", argv[0], blktrace_version);
1609 buf_size = strtoul(optarg, NULL, 10);
1610 if (buf_size <= 0 || buf_size > 16*1024) {
1612 "Invalid buffer size (%lu)\n",buf_size);
1618 buf_nr = strtoul(optarg, NULL, 10);
1621 "Invalid buffer nr (%lu)\n", buf_nr);
1626 output_dir = optarg;
1629 net_mode = Net_client;
1630 strcpy(hostname, optarg);
1633 net_mode = Net_server;
1636 net_port = atoi(optarg);
1639 net_use_sendfile = 1;
1642 show_usage(argv[0]);
1647 setlocale(LC_NUMERIC, "en_US");
1649 page_size = getpagesize();
1651 if (net_mode == Net_server)
1652 return net_server();
1654 while (optind < argc) {
1655 if (resize_devices(argv[optind++]) != 0)
1660 show_usage(argv[0]);
1665 relay_path = default_relay_path;
1667 if (act_mask_tmp != 0)
1668 act_mask = act_mask_tmp;
1670 if (statfs(relay_path, &st) < 0) {
1672 fprintf(stderr,"%s does not appear to be a valid path\n",
1675 } else if (st.f_type != (long) RELAYFS_TYPE) {
1676 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1681 if (open_devices() != 0)
1684 if (kill_running_trace) {
1689 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1691 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1695 signal(SIGINT, handle_sigint);
1696 signal(SIGHUP, handle_sigint);
1697 signal(SIGTERM, handle_sigint);
1698 signal(SIGALRM, handle_sigint);
1700 if (net_mode == Net_client && net_setup_client())
1703 if (start_devices() != 0)
1706 atexit(stop_all_tracing);
1713 if (!is_trace_stopped()) {