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;
166 #define FIFO_SIZE (1024) /* should be plenty big! */
167 #define CL_SIZE (128) /* cache line, any bigger? */
169 struct tip_subbuf_fifo {
170 int tail __attribute__((aligned(CL_SIZE)));
171 int head __attribute__((aligned(CL_SIZE)));
172 struct tip_subbuf *q[FIFO_SIZE];
175 struct thread_information {
181 char fn[MAXPATHLEN + 64];
192 int (*get_subbuf)(struct thread_information *, unsigned int);
193 int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
194 int (*read_data)(struct thread_information *, void *, unsigned int);
196 unsigned long events_processed;
197 unsigned long long data_read;
198 struct device_information *device;
205 struct tip_subbuf_fifo fifo;
206 struct tip_subbuf *leftover_ts;
209 * mmap controlled output files
211 unsigned long long fs_size;
212 unsigned long long fs_max_size;
213 unsigned long fs_off;
215 unsigned long fs_buf_len;
218 struct device_information {
222 volatile int trace_started;
223 unsigned long drop_count;
224 struct thread_information *threads;
228 static struct thread_information *thread_information;
230 static struct device_information *device_information;
232 /* command line option globals */
233 static char *relay_path;
234 static char *output_name;
235 static char *output_dir;
236 static int act_mask = ~0U;
237 static int kill_running_trace;
238 static unsigned long buf_size = BUF_SIZE;
239 static unsigned long buf_nr = BUF_NR;
240 static unsigned int page_size;
242 #define is_done() (*(volatile int *)(&done))
243 static volatile int done;
245 #define is_trace_stopped() (*(volatile int *)(&trace_stopped))
246 static volatile int trace_stopped;
248 #define is_stat_shown() (*(volatile int *)(&stat_shown))
249 static volatile int stat_shown;
251 int data_is_native = -1;
253 static void exit_trace(int status);
255 #define dip_tracing(dip) (*(volatile int *)(&(dip)->trace_started))
256 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
258 #define __for_each_dip(__d, __i, __e) \
259 for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
261 #define for_each_dip(__d, __i) __for_each_dip(__d, __i, ndevs)
262 #define for_each_tip(__d, __t, __j) \
263 for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
266 * networking stuff follows. we include a magic number so we know whether
267 * to endianness convert or not
269 struct blktrace_net_hdr {
270 u32 magic; /* same as trace magic */
271 char buts_name[32]; /* trace name */
272 u32 cpu; /* for which cpu */
274 u32 len; /* length of following trace data */
277 #define TRACE_NET_PORT (8462)
286 * network cmd line params
288 static char hostname[MAXHOSTNAMELEN];
289 static int net_port = TRACE_NET_PORT;
290 static int net_mode = 0;
291 static int net_sendfile;
293 static int net_in_fd = -1;
294 static int net_out_fd = -1;
296 static void handle_sigint(__attribute__((__unused__)) int sig)
301 static int get_dropped_count(const char *buts_name)
304 char tmp[MAXPATHLEN + 64];
306 snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
307 relay_path, buts_name);
309 fd = open(tmp, O_RDONLY);
312 * this may be ok, if the kernel doesn't support dropped counts
317 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
321 if (read(fd, tmp, sizeof(tmp)) < 0) {
332 static size_t get_subbuf_padding(struct thread_information *tip,
335 size_t padding_size = buf_nr * sizeof(size_t);
338 if (read(tip->pfd, tip->pfd_buf, padding_size) < 0) {
339 perror("tip pad read");
342 ret = tip->pfd_buf[subbuf];
347 static int start_trace(struct device_information *dip)
349 struct blk_user_trace_setup buts;
351 memset(&buts, 0, sizeof(buts));
352 buts.buf_size = buf_size;
353 buts.buf_nr = buf_nr;
354 buts.act_mask = act_mask;
356 if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
357 perror("BLKTRACESETUP");
361 if (ioctl(dip->fd, BLKTRACESTART) < 0) {
362 perror("BLKTRACESTART");
366 memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
367 dip_set_tracing(dip, 1);
371 static void stop_trace(struct device_information *dip)
373 if (dip_tracing(dip) || kill_running_trace) {
374 dip_set_tracing(dip, 0);
376 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
377 perror("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)
399 struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
403 if (pfd.revents & POLLIN)
405 if (tip->ofile_stdout)
407 } while (!is_done());
410 static int read_data_file(struct thread_information *tip, void *buf,
418 ret = read(tip->fd, buf, len);
424 if (errno != EAGAIN) {
426 fprintf(stderr,"Thread %d failed read of %s\n",
432 } while (!is_done());
438 static int read_data_net(struct thread_information *tip, void *buf,
441 unsigned int bytes_left = len;
445 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
450 if (errno != EAGAIN) {
452 fprintf(stderr, "server: failed read\n");
460 } while (!is_done() && bytes_left);
462 return len - bytes_left;
465 static int read_data(struct thread_information *tip, void *buf,
468 int ret = tip->read_data(tip, buf, len);
471 tip->data_read += ret;
476 static inline struct tip_subbuf *
477 subbuf_fifo_dequeue(struct thread_information *tip)
479 const int head = tip->fifo.head;
480 const int next = (head + 1) & (FIFO_SIZE - 1);
482 if (head != tip->fifo.tail) {
483 struct tip_subbuf *ts = tip->fifo.q[head];
486 tip->fifo.head = next;
493 static inline int subbuf_fifo_queue(struct thread_information *tip,
494 struct tip_subbuf *ts)
496 const int tail = tip->fifo.tail;
497 const int next = (tail + 1) & (FIFO_SIZE - 1);
499 if (next != tip->fifo.head) {
500 tip->fifo.q[tail] = ts;
502 tip->fifo.tail = next;
506 fprintf(stderr, "fifo too small!\n");
511 * For file output, truncate and mmap the file appropriately
513 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
515 int ofd = fileno(tip->ofile);
519 * extend file, if we have to. use chunks of 16 subbuffers.
521 if (tip->fs_off + buf_size > tip->fs_buf_len) {
523 munlock(tip->fs_buf, tip->fs_buf_len);
524 munmap(tip->fs_buf, tip->fs_buf_len);
528 tip->fs_off = tip->fs_size & (page_size - 1);
529 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
530 tip->fs_max_size += tip->fs_buf_len;
532 if (ftruncate(ofd, tip->fs_max_size) < 0) {
537 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
538 MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
539 if (tip->fs_buf == MAP_FAILED) {
543 mlock(tip->fs_buf, tip->fs_buf_len);
546 ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
557 * Use the copy approach for pipes and network
559 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
561 struct tip_subbuf *ts = malloc(sizeof(*ts));
564 ts->buf = malloc(buf_size);
565 ts->max_len = maxlen;
567 ret = read_data(tip, ts->buf, ts->max_len);
570 return subbuf_fifo_queue(tip, ts);
576 static int get_subbuf_sendfile(struct thread_information *tip,
579 struct tip_subbuf *ts;
581 unsigned int ready, this_size;
587 * hack to get last data out, we can't use sendfile for that
590 return get_subbuf(tip, maxlen);
592 if (fstat(tip->fd, &sb) < 0) {
593 perror("trace stat");
597 ready = sb.st_size - tip->ofile_offset;
601 this_size = buf_size;
603 if (this_size > ready)
606 ts = malloc(sizeof(*ts));
608 ts->max_len = maxlen;
612 ts->max_len = ts->len;
613 ts->offset = tip->ofile_offset;
614 tip->ofile_offset += ts->len;
616 err = subbuf_fifo_queue(tip, ts);
626 static void close_thread(struct thread_information *tip)
634 if (tip->ofile_buffer)
635 free(tip->ofile_buffer);
644 tip->ofile_buffer = NULL;
648 static void tip_ftrunc_final(struct thread_information *tip)
651 * truncate to right size and cleanup mmap
653 if (tip->ofile_mmap) {
654 int ofd = fileno(tip->ofile);
657 munmap(tip->fs_buf, tip->fs_buf_len);
659 ftruncate(ofd, tip->fs_size);
663 static void *thread_main(void *arg)
665 struct thread_information *tip = arg;
666 pid_t pid = getpid();
670 CPU_SET((tip->cpu), &cpu_mask);
672 if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
673 perror("sched_setaffinity");
677 snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
678 relay_path, tip->device->buts_name, tip->cpu);
679 tip->fd = open(tip->fn, O_RDONLY);
682 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
687 if (net_mode == Net_client && net_sendfile) {
688 char tmp[MAXPATHLEN + 64];
690 snprintf(tmp, sizeof(tmp), "%s/block/%s/trace%d.padding",
691 relay_path, tip->device->buts_name, tip->cpu);
693 tip->pfd = open(tmp, O_RDONLY);
695 fprintf(stderr, "Couldn't open padding file %s\n", tmp);
699 tip->pfd_buf = malloc(buf_nr * sizeof(size_t));
703 if (tip->get_subbuf(tip, buf_size))
707 tip_ftrunc_final(tip);
712 static int write_data_net(int fd, void *buf, unsigned int buf_len)
714 unsigned int bytes_left = buf_len;
718 ret = send(fd, buf, bytes_left, 0);
731 static int net_send_header(struct thread_information *tip, unsigned int len)
733 struct blktrace_net_hdr hdr;
735 hdr.magic = BLK_IO_TRACE_MAGIC;
736 strcpy(hdr.buts_name, tip->device->buts_name);
738 hdr.max_cpus = ncpus;
741 return write_data_net(net_out_fd, &hdr, sizeof(hdr));
745 * send header with 0 length to signal end-of-run
747 static void net_client_send_close(void)
749 struct blktrace_net_hdr hdr;
751 hdr.magic = BLK_IO_TRACE_MAGIC;
753 hdr.max_cpus = ncpus;
756 write_data_net(net_out_fd, &hdr, sizeof(hdr));
759 static int flush_subbuf_net(struct thread_information *tip,
760 struct tip_subbuf *ts)
762 if (net_send_header(tip, ts->len))
764 if (write_data_net(net_out_fd, ts->buf, ts->len))
767 tip->data_read += ts->len;
773 static int flush_subbuf_sendfile(struct thread_information *tip,
774 struct tip_subbuf *ts)
781 * currently we cannot use sendfile() on the last bytes read, as they
782 * may not be a full subbuffer. get_subbuf_sendfile() falls back to
783 * the read approach for those, so use send() to ship them out
786 return flush_subbuf_net(tip, ts);
788 subbuf = (ts->offset / buf_size) % buf_nr;
789 padding = get_subbuf_padding(tip, subbuf);
790 len = ts->len - padding;
792 if (net_send_header(tip, len))
794 if (sendfile(net_out_fd, tip->fd, &ts->offset, len) < 0) {
799 tip->data_read += ts->len;
804 static int write_data(struct thread_information *tip, void *buf,
805 unsigned int buf_len)
813 ret = fwrite(buf, buf_len, 1, tip->ofile);
823 if (tip->ofile_stdout)
829 static int flush_subbuf_file(struct thread_information *tip,
830 struct tip_subbuf *ts)
832 unsigned int offset = 0;
833 struct blk_io_trace *t;
834 int pdu_len, events = 0;
837 * surplus from last run
839 if (tip->leftover_ts) {
840 struct tip_subbuf *prev_ts = tip->leftover_ts;
842 if (prev_ts->len + ts->len > prev_ts->max_len) {
843 prev_ts->max_len += ts->len;
844 prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
847 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
848 prev_ts->len += ts->len;
854 tip->leftover_ts = NULL;
857 while (offset + sizeof(*t) <= ts->len) {
858 t = ts->buf + offset;
860 if (verify_trace(t)) {
861 write_data(tip, ts->buf, offset);
865 pdu_len = t->pdu_len;
867 if (offset + sizeof(*t) + pdu_len > ts->len)
870 offset += sizeof(*t) + pdu_len;
871 tip->events_processed++;
872 tip->data_read += sizeof(*t) + pdu_len;
876 if (write_data(tip, ts->buf, offset))
880 * leftover bytes, save them for next time
882 if (offset != ts->len) {
883 tip->leftover_ts = ts;
885 memmove(ts->buf, ts->buf + offset, ts->len);
894 static int write_tip_events(struct thread_information *tip)
896 struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
899 return tip->flush_subbuf(tip, ts);
905 * scans the tips we know and writes out the subbuffers we accumulate
907 static void get_and_write_events(void)
909 struct device_information *dip;
910 struct thread_information *tip;
911 int i, j, events, ret, tips_running;
916 for_each_dip(dip, i) {
917 for_each_tip(dip, tip, j) {
918 ret = write_tip_events(tip);
934 for_each_dip(dip, i) {
935 for_each_tip(dip, tip, j) {
936 ret = write_tip_events(tip);
939 tips_running += !tip->exited;
943 } while (events || tips_running);
946 static void wait_for_threads(void)
949 * for piped or network output, poll and fetch data for writeout.
950 * for files, we just wait around for trace threads to exit
952 if ((output_name && !strcmp(output_name, "-")) ||
953 net_mode == Net_client)
954 get_and_write_events();
956 struct device_information *dip;
957 struct thread_information *tip;
958 int i, j, tips_running;
965 for_each_tip(dip, tip, j)
966 tips_running += !tip->exited;
967 } while (tips_running);
970 if (net_mode == Net_client)
971 net_client_send_close();
974 static void fill_ofname(char *dst, char *buts_name, int cpu)
979 len = sprintf(dst, "%s/", output_dir);
982 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
984 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
987 static void fill_ops(struct thread_information *tip)
992 if (net_mode == Net_client) {
994 tip->get_subbuf = get_subbuf_sendfile;
995 tip->flush_subbuf = flush_subbuf_sendfile;
997 tip->get_subbuf = get_subbuf;
998 tip->flush_subbuf = flush_subbuf_net;
1001 if (tip->ofile_mmap)
1002 tip->get_subbuf = mmap_subbuf;
1004 tip->get_subbuf = get_subbuf;
1006 tip->flush_subbuf = flush_subbuf_file;
1009 if (net_mode == Net_server)
1010 tip->read_data = read_data_net;
1012 tip->read_data = read_data_file;
1015 static int tip_open_output(struct device_information *dip,
1016 struct thread_information *tip)
1018 int pipeline = output_name && !strcmp(output_name, "-");
1019 int mode, vbuf_size;
1022 if (net_mode == Net_client) {
1024 tip->ofile_stdout = 0;
1025 tip->ofile_mmap = 0;
1027 mode = 0; /* gcc 4.x issues a bogus warning */
1028 } else if (pipeline) {
1029 tip->ofile = fdopen(STDOUT_FILENO, "w");
1030 tip->ofile_stdout = 1;
1031 tip->ofile_mmap = 0;
1035 fill_ofname(op, dip->buts_name, tip->cpu);
1036 tip->ofile = fopen(op, "w+");
1037 tip->ofile_stdout = 0;
1038 tip->ofile_mmap = 1;
1040 vbuf_size = OFILE_BUF;
1043 if (net_mode != Net_client && tip->ofile == NULL) {
1049 tip->ofile_buffer = malloc(vbuf_size);
1050 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1061 static int start_threads(struct device_information *dip)
1063 struct thread_information *tip;
1066 for_each_tip(dip, tip, j) {
1069 tip->events_processed = 0;
1072 memset(&tip->fifo, 0, sizeof(tip->fifo));
1073 tip->leftover_ts = NULL;
1075 if (tip_open_output(dip, tip))
1078 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1079 perror("pthread_create");
1088 static void stop_threads(struct device_information *dip)
1090 struct thread_information *tip;
1094 for_each_tip(dip, tip, i) {
1095 (void) pthread_join(tip->thread, (void *) &ret);
1100 static void stop_all_threads(void)
1102 struct device_information *dip;
1105 for_each_dip(dip, i)
1109 static void stop_all_tracing(void)
1111 struct device_information *dip;
1114 for_each_dip(dip, i)
1118 static void exit_trace(int status)
1120 if (!is_trace_stopped()) {
1129 static int resize_devices(char *path)
1131 int size = (ndevs + 1) * sizeof(struct device_information);
1133 device_information = realloc(device_information, size);
1134 if (!device_information) {
1135 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1138 device_information[ndevs].path = path;
1143 static int open_devices(void)
1145 struct device_information *dip;
1148 for_each_dip(dip, i) {
1149 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1159 static int start_devices(void)
1161 struct device_information *dip;
1164 size = ncpus * sizeof(struct thread_information);
1165 thread_information = malloc(size * ndevs);
1166 if (!thread_information) {
1167 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1171 for_each_dip(dip, i) {
1172 if (start_trace(dip)) {
1174 fprintf(stderr, "Failed to start trace on %s\n",
1181 __for_each_dip(dip, j, i)
1187 for_each_dip(dip, i) {
1188 dip->threads = thread_information + (i * ncpus);
1189 if (start_threads(dip)) {
1190 fprintf(stderr, "Failed to start worker threads\n");
1196 __for_each_dip(dip, j, i)
1198 for_each_dip(dip, i)
1207 static void show_stats(void)
1209 struct device_information *dip;
1210 struct thread_information *tip;
1211 unsigned long long events_processed, data_read;
1212 unsigned long total_drops;
1213 int i, j, no_stdout = 0;
1215 if (is_stat_shown())
1218 if (output_name && !strcmp(output_name, "-"))
1224 for_each_dip(dip, i) {
1226 printf("Device: %s\n", dip->path);
1227 events_processed = 0;
1229 for_each_tip(dip, tip, j) {
1231 printf(" CPU%3d: %20lu events, %8llu KiB data\n",
1232 tip->cpu, tip->events_processed,
1233 (tip->data_read + 1023) >> 10);
1234 events_processed += tip->events_processed;
1235 data_read += tip->data_read;
1237 total_drops += dip->drop_count;
1239 printf(" Total: %20llu events (dropped %lu), %8llu KiB data\n",
1240 events_processed, dip->drop_count,
1241 (data_read + 1023) >> 10);
1245 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1248 static struct device_information *net_get_dip(char *buts_name)
1250 struct device_information *dip;
1253 for (i = 0; i < ndevs; i++) {
1254 dip = &device_information[i];
1256 if (!strcmp(dip->buts_name, buts_name))
1260 device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1261 dip = &device_information[ndevs];
1262 strcpy(dip->buts_name, buts_name);
1263 strcpy(dip->path, buts_name);
1265 dip->threads = malloc(ncpus * sizeof(struct thread_information));
1266 memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1271 for (i = 0; i < ncpus; i++) {
1272 struct thread_information *tip = &dip->threads[i];
1277 if (tip_open_output(dip, tip))
1284 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1286 struct device_information *dip;
1288 ncpus = bnh->max_cpus;
1289 dip = net_get_dip(bnh->buts_name);
1290 return &dip->threads[bnh->cpu];
1293 static int net_get_header(struct blktrace_net_hdr *bnh)
1295 int fl = fcntl(net_in_fd, F_GETFL);
1296 int bytes_left, ret;
1299 fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1300 bytes_left = sizeof(*bnh);
1301 while (bytes_left && !is_done()) {
1302 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1304 if (errno != EAGAIN) {
1305 perror("recv header");
1318 fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1322 static int net_server_loop(void)
1324 struct thread_information *tip;
1325 struct blktrace_net_hdr bnh;
1327 if (net_get_header(&bnh))
1330 if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1331 fprintf(stderr, "server: received data is bad\n");
1335 if (!data_is_native) {
1336 bnh.cpu = be32_to_cpu(bnh.cpu);
1337 bnh.len = be32_to_cpu(bnh.len);
1341 * len == 0 means that the other end signalled end-of-run
1344 fprintf(stderr, "server: end of run\n");
1348 tip = net_get_tip(&bnh);
1352 if (mmap_subbuf(tip, bnh.len))
1359 * Start here when we are in server mode - just fetch data from the network
1362 static int net_server(void)
1364 struct device_information *dip;
1365 struct thread_information *tip;
1366 struct sockaddr_in addr;
1370 fd = socket(AF_INET, SOCK_STREAM, 0);
1372 perror("server: socket");
1377 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1378 perror("setsockopt");
1382 memset(&addr, 0, sizeof(addr));
1383 addr.sin_family = AF_INET;
1384 addr.sin_addr.s_addr = htonl(INADDR_ANY);
1385 addr.sin_port = htons(net_port);
1387 if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1392 if (listen(fd, 1) < 0) {
1398 signal(SIGINT, NULL);
1399 signal(SIGHUP, NULL);
1400 signal(SIGTERM, NULL);
1401 signal(SIGALRM, NULL);
1403 printf("blktrace: waiting for incoming connection...\n");
1405 socklen = sizeof(addr);
1406 net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1407 if (net_in_fd < 0) {
1412 signal(SIGINT, handle_sigint);
1413 signal(SIGHUP, handle_sigint);
1414 signal(SIGTERM, handle_sigint);
1415 signal(SIGALRM, handle_sigint);
1417 printf("blktrace: connection from %s\n", inet_ntoa(addr.sin_addr));
1419 while (!is_done()) {
1420 if (net_server_loop())
1424 for_each_dip(dip, i)
1425 for_each_tip(dip, tip, j)
1426 tip_ftrunc_final(tip);
1434 * cleanup for next run
1436 for_each_dip(dip, i) {
1437 for_each_tip(dip, tip, j)
1443 free(device_information);
1444 device_information = NULL;
1453 * Setup outgoing network connection where we will transmit data
1455 static int net_setup_client(void)
1457 struct sockaddr_in addr;
1460 fd = socket(AF_INET, SOCK_STREAM, 0);
1462 perror("client: socket");
1466 memset(&addr, 0, sizeof(addr));
1467 addr.sin_family = AF_INET;
1468 addr.sin_port = htons(net_port);
1470 if (inet_aton(hostname, &addr.sin_addr) != 1) {
1471 struct hostent *hent = gethostbyname(hostname);
1473 perror("gethostbyname");
1477 memcpy(&addr.sin_addr, hent->h_addr, 4);
1478 strcpy(hostname, hent->h_name);
1481 printf("blktrace: connecting to %s\n", hostname);
1483 if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1484 perror("client: connect");
1488 printf("blktrace: connected!\n");
1493 static char usage_str[] = \
1494 "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1495 "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1496 "\t-d Use specified device. May also be given last after options\n" \
1497 "\t-r Path to mounted relayfs, defaults to /relay\n" \
1498 "\t-o File(s) to send output to\n" \
1499 "\t-D Directory to prepend to output file names\n" \
1500 "\t-k Kill a running trace\n" \
1501 "\t-w Stop after defined time, in seconds\n" \
1502 "\t-a Only trace specified actions. See documentation\n" \
1503 "\t-A Give trace mask as a single value. See documentation\n" \
1504 "\t-b Sub buffer size in KiB\n" \
1505 "\t-n Number of sub buffers\n" \
1506 "\t-l Run in network listen mode (blktrace server)\n" \
1507 "\t-h Run in network client mode, connecting to the given host\n" \
1508 "\t-p Network port to use (default 8462)\n" \
1509 "\t-s Make the network client use sendfile() to transfer data\n" \
1510 "\t-V Print program version info\n\n";
1512 static void show_usage(char *program)
1514 fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1517 int main(int argc, char *argv[])
1519 static char default_relay_path[] = "/relay";
1523 int act_mask_tmp = 0;
1525 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1528 i = find_mask_map(optarg);
1530 fprintf(stderr,"Invalid action mask %s\n",
1538 if ((sscanf(optarg, "%x", &i) != 1) ||
1539 !valid_act_opt(i)) {
1541 "Invalid set action mask %s/0x%x\n",
1549 if (resize_devices(optarg) != 0)
1554 relay_path = optarg;
1558 output_name = optarg;
1561 kill_running_trace = 1;
1564 stop_watch = atoi(optarg);
1565 if (stop_watch <= 0) {
1567 "Invalid stopwatch value (%d secs)\n",
1573 printf("%s version %s\n", argv[0], blktrace_version);
1576 buf_size = strtoul(optarg, NULL, 10);
1577 if (buf_size <= 0 || buf_size > 16*1024) {
1579 "Invalid buffer size (%lu)\n",buf_size);
1585 buf_nr = strtoul(optarg, NULL, 10);
1588 "Invalid buffer nr (%lu)\n", buf_nr);
1593 output_dir = optarg;
1596 net_mode = Net_client;
1597 strcpy(hostname, optarg);
1600 net_mode = Net_server;
1603 net_port = atoi(optarg);
1609 show_usage(argv[0]);
1614 setlocale(LC_NUMERIC, "en_US");
1616 page_size = getpagesize();
1618 if (net_mode == Net_server)
1619 return net_server();
1621 while (optind < argc) {
1622 if (resize_devices(argv[optind++]) != 0)
1627 show_usage(argv[0]);
1632 relay_path = default_relay_path;
1634 if (act_mask_tmp != 0)
1635 act_mask = act_mask_tmp;
1637 if (statfs(relay_path, &st) < 0) {
1639 fprintf(stderr,"%s does not appear to be a valid path\n",
1642 } else if (st.f_type != (long) RELAYFS_TYPE) {
1643 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1648 if (open_devices() != 0)
1651 if (kill_running_trace) {
1656 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1658 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1662 signal(SIGINT, handle_sigint);
1663 signal(SIGHUP, handle_sigint);
1664 signal(SIGTERM, handle_sigint);
1665 signal(SIGALRM, handle_sigint);
1667 if (net_mode == Net_client && net_setup_client())
1670 if (start_devices() != 0)
1673 atexit(stop_all_tracing);
1680 if (!is_trace_stopped()) {