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)
298 struct device_information *dip;
302 * stop trace so we can reap currently produced data
304 for_each_dip(dip, i) {
305 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
306 perror("BLKTRACESTOP");
312 static int get_dropped_count(const char *buts_name)
315 char tmp[MAXPATHLEN + 64];
317 snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
318 relay_path, buts_name);
320 fd = open(tmp, O_RDONLY);
323 * this may be ok, if the kernel doesn't support dropped counts
328 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
332 if (read(fd, tmp, sizeof(tmp)) < 0) {
343 static size_t get_subbuf_padding(struct thread_information *tip,
346 size_t padding_size = buf_nr * sizeof(size_t);
349 if (read(tip->pfd, tip->pfd_buf, padding_size) < 0) {
350 perror("tip pad read");
353 ret = tip->pfd_buf[subbuf];
358 static int start_trace(struct device_information *dip)
360 struct blk_user_trace_setup buts;
362 memset(&buts, 0, sizeof(buts));
363 buts.buf_size = buf_size;
364 buts.buf_nr = buf_nr;
365 buts.act_mask = act_mask;
367 if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
368 perror("BLKTRACESETUP");
372 if (ioctl(dip->fd, BLKTRACESTART) < 0) {
373 perror("BLKTRACESTART");
377 memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
378 dip_set_tracing(dip, 1);
382 static void stop_trace(struct device_information *dip)
384 if (dip_tracing(dip) || kill_running_trace) {
385 dip_set_tracing(dip, 0);
388 * should be stopped, just don't complain if it isn't
390 ioctl(dip->fd, BLKTRACESTOP);
392 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
393 perror("BLKTRACETEARDOWN");
400 static void stop_all_traces(void)
402 struct device_information *dip;
405 for_each_dip(dip, i) {
406 dip->drop_count = get_dropped_count(dip->buts_name);
411 static void wait_for_data(struct thread_information *tip)
413 struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
417 if (pfd.revents & POLLIN)
419 if (tip->ofile_stdout)
421 } while (!is_done());
424 static int read_data_file(struct thread_information *tip, void *buf,
432 ret = read(tip->fd, buf, len);
438 if (errno != EAGAIN) {
440 fprintf(stderr,"Thread %d failed read of %s\n",
446 } while (!is_done());
452 static int read_data_net(struct thread_information *tip, void *buf,
455 unsigned int bytes_left = len;
459 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
464 if (errno != EAGAIN) {
466 fprintf(stderr, "server: failed read\n");
474 } while (!is_done() && bytes_left);
476 return len - bytes_left;
479 static int read_data(struct thread_information *tip, void *buf,
482 return tip->read_data(tip, buf, len);
485 static inline struct tip_subbuf *
486 subbuf_fifo_dequeue(struct thread_information *tip)
488 const int head = tip->fifo.head;
489 const int next = (head + 1) & (FIFO_SIZE - 1);
491 if (head != tip->fifo.tail) {
492 struct tip_subbuf *ts = tip->fifo.q[head];
495 tip->fifo.head = next;
502 static inline int subbuf_fifo_queue(struct thread_information *tip,
503 struct tip_subbuf *ts)
505 const int tail = tip->fifo.tail;
506 const int next = (tail + 1) & (FIFO_SIZE - 1);
508 if (next != tip->fifo.head) {
509 tip->fifo.q[tail] = ts;
511 tip->fifo.tail = next;
515 fprintf(stderr, "fifo too small!\n");
520 * For file output, truncate and mmap the file appropriately
522 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
524 int ofd = fileno(tip->ofile);
528 * extend file, if we have to. use chunks of 16 subbuffers.
530 if (tip->fs_off + buf_size > tip->fs_buf_len) {
532 munlock(tip->fs_buf, tip->fs_buf_len);
533 munmap(tip->fs_buf, tip->fs_buf_len);
537 tip->fs_off = tip->fs_size & (page_size - 1);
538 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
539 tip->fs_max_size += tip->fs_buf_len;
541 if (ftruncate(ofd, tip->fs_max_size) < 0) {
546 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
547 MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
548 if (tip->fs_buf == MAP_FAILED) {
552 mlock(tip->fs_buf, tip->fs_buf_len);
555 ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
557 tip->data_read += ret;
567 * Use the copy approach for pipes and network
569 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
571 struct tip_subbuf *ts = malloc(sizeof(*ts));
574 ts->buf = malloc(buf_size);
575 ts->max_len = maxlen;
577 ret = read_data(tip, ts->buf, ts->max_len);
580 tip->data_read += ret;
581 if (subbuf_fifo_queue(tip, ts))
588 static int get_subbuf_sendfile(struct thread_information *tip,
591 struct tip_subbuf *ts;
593 unsigned int ready, this_size, total;
598 * hack to get last data out, we can't use sendfile for that
601 return get_subbuf(tip, maxlen);
603 if (fstat(tip->fd, &sb) < 0) {
604 perror("trace stat");
608 ready = sb.st_size - tip->ofile_offset;
611 * delay a little, since poll() will return data available
612 * until sendfile() is run
618 this_size = buf_size;
621 if (this_size > ready)
624 ts = malloc(sizeof(*ts));
630 ts->offset = tip->ofile_offset;
631 tip->ofile_offset += ts->len;
633 if (subbuf_fifo_queue(tip, ts))
642 static void close_thread(struct thread_information *tip)
650 if (tip->ofile_buffer)
651 free(tip->ofile_buffer);
660 tip->ofile_buffer = NULL;
664 static void tip_ftrunc_final(struct thread_information *tip)
667 * truncate to right size and cleanup mmap
669 if (tip->ofile_mmap) {
670 int ofd = fileno(tip->ofile);
673 munmap(tip->fs_buf, tip->fs_buf_len);
675 ftruncate(ofd, tip->fs_size);
679 static void *thread_main(void *arg)
681 struct thread_information *tip = arg;
682 pid_t pid = getpid();
686 CPU_SET((tip->cpu), &cpu_mask);
688 if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
689 perror("sched_setaffinity");
693 snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
694 relay_path, tip->device->buts_name, tip->cpu);
695 tip->fd = open(tip->fn, O_RDONLY);
698 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
703 if (net_mode == Net_client && net_sendfile) {
704 char tmp[MAXPATHLEN + 64];
706 snprintf(tmp, sizeof(tmp), "%s/block/%s/trace%d.padding",
707 relay_path, tip->device->buts_name, tip->cpu);
709 tip->pfd = open(tmp, O_RDONLY);
711 fprintf(stderr, "Couldn't open padding file %s\n", tmp);
715 tip->pfd_buf = malloc(buf_nr * sizeof(size_t));
719 if (tip->get_subbuf(tip, buf_size) < 0)
724 * trace is stopped, pull data until we get a short read
726 while (tip->get_subbuf(tip, buf_size) > 0)
729 tip_ftrunc_final(tip);
734 static int write_data_net(int fd, void *buf, unsigned int buf_len)
736 unsigned int bytes_left = buf_len;
740 ret = send(fd, buf, bytes_left, 0);
753 static int net_send_header(struct thread_information *tip, unsigned int len)
755 struct blktrace_net_hdr hdr;
757 hdr.magic = BLK_IO_TRACE_MAGIC;
758 strcpy(hdr.buts_name, tip->device->buts_name);
760 hdr.max_cpus = ncpus;
763 return write_data_net(net_out_fd, &hdr, sizeof(hdr));
767 * send header with 0 length to signal end-of-run
769 static void net_client_send_close(void)
771 struct blktrace_net_hdr hdr;
773 hdr.magic = BLK_IO_TRACE_MAGIC;
775 hdr.max_cpus = ncpus;
778 write_data_net(net_out_fd, &hdr, sizeof(hdr));
781 static int flush_subbuf_net(struct thread_information *tip,
782 struct tip_subbuf *ts)
784 if (net_send_header(tip, ts->len))
786 if (write_data_net(net_out_fd, ts->buf, ts->len))
794 static int flush_subbuf_sendfile(struct thread_information *tip,
795 struct tip_subbuf *ts)
802 * currently we cannot use sendfile() on the last bytes read, as they
803 * may not be a full subbuffer. get_subbuf_sendfile() falls back to
804 * the read approach for those, so use send() to ship them out
807 return flush_subbuf_net(tip, ts);
809 subbuf = (ts->offset / buf_size) % buf_nr;
810 padding = get_subbuf_padding(tip, subbuf);
811 len = ts->len - padding;
813 if (net_send_header(tip, len))
815 if (sendfile(net_out_fd, tip->fd, &ts->offset, len) < 0) {
820 tip->data_read += len;
825 static int write_data(struct thread_information *tip, void *buf,
826 unsigned int buf_len)
834 ret = fwrite(buf, buf_len, 1, tip->ofile);
844 if (tip->ofile_stdout)
850 static int flush_subbuf_file(struct thread_information *tip,
851 struct tip_subbuf *ts)
853 unsigned int offset = 0;
854 struct blk_io_trace *t;
855 int pdu_len, events = 0;
858 * surplus from last run
860 if (tip->leftover_ts) {
861 struct tip_subbuf *prev_ts = tip->leftover_ts;
863 if (prev_ts->len + ts->len > prev_ts->max_len) {
864 prev_ts->max_len += ts->len;
865 prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
868 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
869 prev_ts->len += ts->len;
875 tip->leftover_ts = NULL;
878 while (offset + sizeof(*t) <= ts->len) {
879 t = ts->buf + offset;
881 if (verify_trace(t)) {
882 write_data(tip, ts->buf, offset);
886 pdu_len = t->pdu_len;
888 if (offset + sizeof(*t) + pdu_len > ts->len)
891 offset += sizeof(*t) + pdu_len;
892 tip->events_processed++;
893 tip->data_read += sizeof(*t) + pdu_len;
897 if (write_data(tip, ts->buf, offset))
901 * leftover bytes, save them for next time
903 if (offset != ts->len) {
904 tip->leftover_ts = ts;
906 memmove(ts->buf, ts->buf + offset, ts->len);
915 static int write_tip_events(struct thread_information *tip)
917 struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
920 return tip->flush_subbuf(tip, ts);
926 * scans the tips we know and writes out the subbuffers we accumulate
928 static void get_and_write_events(void)
930 struct device_information *dip;
931 struct thread_information *tip;
932 int i, j, events, ret, tips_running;
937 for_each_dip(dip, i) {
938 for_each_tip(dip, tip, j) {
939 ret = write_tip_events(tip);
955 for_each_dip(dip, i) {
956 for_each_tip(dip, tip, j) {
957 ret = write_tip_events(tip);
960 tips_running += !tip->exited;
964 } while (events || tips_running);
967 static void wait_for_threads(void)
970 * for piped or network output, poll and fetch data for writeout.
971 * for files, we just wait around for trace threads to exit
973 if ((output_name && !strcmp(output_name, "-")) ||
974 net_mode == Net_client)
975 get_and_write_events();
977 struct device_information *dip;
978 struct thread_information *tip;
979 int i, j, tips_running;
986 for_each_tip(dip, tip, j)
987 tips_running += !tip->exited;
988 } while (tips_running);
991 if (net_mode == Net_client)
992 net_client_send_close();
995 static void fill_ofname(char *dst, char *buts_name, int cpu)
1000 len = sprintf(dst, "%s/", output_dir);
1003 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
1005 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
1008 static void fill_ops(struct thread_information *tip)
1013 if (net_mode == Net_client) {
1015 tip->get_subbuf = get_subbuf_sendfile;
1016 tip->flush_subbuf = flush_subbuf_sendfile;
1018 tip->get_subbuf = get_subbuf;
1019 tip->flush_subbuf = flush_subbuf_net;
1022 if (tip->ofile_mmap)
1023 tip->get_subbuf = mmap_subbuf;
1025 tip->get_subbuf = get_subbuf;
1027 tip->flush_subbuf = flush_subbuf_file;
1030 if (net_mode == Net_server)
1031 tip->read_data = read_data_net;
1033 tip->read_data = read_data_file;
1036 static int tip_open_output(struct device_information *dip,
1037 struct thread_information *tip)
1039 int pipeline = output_name && !strcmp(output_name, "-");
1040 int mode, vbuf_size;
1043 if (net_mode == Net_client) {
1045 tip->ofile_stdout = 0;
1046 tip->ofile_mmap = 0;
1048 mode = 0; /* gcc 4.x issues a bogus warning */
1049 } else if (pipeline) {
1050 tip->ofile = fdopen(STDOUT_FILENO, "w");
1051 tip->ofile_stdout = 1;
1052 tip->ofile_mmap = 0;
1056 fill_ofname(op, dip->buts_name, tip->cpu);
1057 tip->ofile = fopen(op, "w+");
1058 tip->ofile_stdout = 0;
1059 tip->ofile_mmap = 1;
1061 vbuf_size = OFILE_BUF;
1064 if (net_mode != Net_client && tip->ofile == NULL) {
1070 tip->ofile_buffer = malloc(vbuf_size);
1071 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1082 static int start_threads(struct device_information *dip)
1084 struct thread_information *tip;
1087 for_each_tip(dip, tip, j) {
1090 tip->events_processed = 0;
1093 memset(&tip->fifo, 0, sizeof(tip->fifo));
1094 tip->leftover_ts = NULL;
1096 if (tip_open_output(dip, tip))
1099 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1100 perror("pthread_create");
1109 static void stop_threads(struct device_information *dip)
1111 struct thread_information *tip;
1115 for_each_tip(dip, tip, i) {
1116 (void) pthread_join(tip->thread, (void *) &ret);
1121 static void stop_all_threads(void)
1123 struct device_information *dip;
1126 for_each_dip(dip, i)
1130 static void stop_all_tracing(void)
1132 struct device_information *dip;
1135 for_each_dip(dip, i)
1139 static void exit_trace(int status)
1141 if (!is_trace_stopped()) {
1150 static int resize_devices(char *path)
1152 int size = (ndevs + 1) * sizeof(struct device_information);
1154 device_information = realloc(device_information, size);
1155 if (!device_information) {
1156 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1159 device_information[ndevs].path = path;
1164 static int open_devices(void)
1166 struct device_information *dip;
1169 for_each_dip(dip, i) {
1170 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1180 static int start_devices(void)
1182 struct device_information *dip;
1185 size = ncpus * sizeof(struct thread_information);
1186 thread_information = malloc(size * ndevs);
1187 if (!thread_information) {
1188 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1192 for_each_dip(dip, i) {
1193 if (start_trace(dip)) {
1195 fprintf(stderr, "Failed to start trace on %s\n",
1202 __for_each_dip(dip, j, i)
1208 for_each_dip(dip, i) {
1209 dip->threads = thread_information + (i * ncpus);
1210 if (start_threads(dip)) {
1211 fprintf(stderr, "Failed to start worker threads\n");
1217 __for_each_dip(dip, j, i)
1219 for_each_dip(dip, i)
1228 static void show_stats(void)
1230 struct device_information *dip;
1231 struct thread_information *tip;
1232 unsigned long long events_processed, data_read;
1233 unsigned long total_drops;
1234 int i, j, no_stdout = 0;
1236 if (is_stat_shown())
1239 if (output_name && !strcmp(output_name, "-"))
1245 for_each_dip(dip, i) {
1247 printf("Device: %s\n", dip->path);
1248 events_processed = 0;
1250 for_each_tip(dip, tip, j) {
1252 printf(" CPU%3d: %20lu events, %8llu KiB data\n",
1253 tip->cpu, tip->events_processed,
1254 (tip->data_read + 1023) >> 10);
1255 events_processed += tip->events_processed;
1256 data_read += tip->data_read;
1258 total_drops += dip->drop_count;
1260 printf(" Total: %20llu events (dropped %lu), %8llu KiB data\n",
1261 events_processed, dip->drop_count,
1262 (data_read + 1023) >> 10);
1266 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1269 static struct device_information *net_get_dip(char *buts_name)
1271 struct device_information *dip;
1274 for (i = 0; i < ndevs; i++) {
1275 dip = &device_information[i];
1277 if (!strcmp(dip->buts_name, buts_name))
1281 device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1282 dip = &device_information[ndevs];
1283 strcpy(dip->buts_name, buts_name);
1284 strcpy(dip->path, buts_name);
1286 dip->threads = malloc(ncpus * sizeof(struct thread_information));
1287 memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1292 for (i = 0; i < ncpus; i++) {
1293 struct thread_information *tip = &dip->threads[i];
1300 if (tip_open_output(dip, tip))
1307 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1309 struct device_information *dip;
1311 ncpus = bnh->max_cpus;
1312 dip = net_get_dip(bnh->buts_name);
1313 return &dip->threads[bnh->cpu];
1316 static int net_get_header(struct blktrace_net_hdr *bnh)
1318 int fl = fcntl(net_in_fd, F_GETFL);
1319 int bytes_left, ret;
1322 fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1323 bytes_left = sizeof(*bnh);
1324 while (bytes_left && !is_done()) {
1325 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1327 if (errno != EAGAIN) {
1328 perror("recv header");
1341 fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1345 static int net_server_loop(void)
1347 struct thread_information *tip;
1348 struct blktrace_net_hdr bnh;
1350 if (net_get_header(&bnh))
1353 if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1354 fprintf(stderr, "server: received data is bad\n");
1358 if (!data_is_native) {
1359 bnh.cpu = be32_to_cpu(bnh.cpu);
1360 bnh.len = be32_to_cpu(bnh.len);
1364 * len == 0 means that the other end signalled end-of-run
1367 fprintf(stderr, "server: end of run\n");
1371 tip = net_get_tip(&bnh);
1375 if (mmap_subbuf(tip, bnh.len))
1382 * Start here when we are in server mode - just fetch data from the network
1385 static int net_server(void)
1387 struct device_information *dip;
1388 struct thread_information *tip;
1389 struct sockaddr_in addr;
1393 fd = socket(AF_INET, SOCK_STREAM, 0);
1395 perror("server: socket");
1400 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1401 perror("setsockopt");
1405 memset(&addr, 0, sizeof(addr));
1406 addr.sin_family = AF_INET;
1407 addr.sin_addr.s_addr = htonl(INADDR_ANY);
1408 addr.sin_port = htons(net_port);
1410 if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1415 if (listen(fd, 1) < 0) {
1421 signal(SIGINT, NULL);
1422 signal(SIGHUP, NULL);
1423 signal(SIGTERM, NULL);
1424 signal(SIGALRM, NULL);
1426 printf("blktrace: waiting for incoming connection...\n");
1428 socklen = sizeof(addr);
1429 net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1430 if (net_in_fd < 0) {
1435 signal(SIGINT, handle_sigint);
1436 signal(SIGHUP, handle_sigint);
1437 signal(SIGTERM, handle_sigint);
1438 signal(SIGALRM, handle_sigint);
1440 printf("blktrace: connection from %s\n", inet_ntoa(addr.sin_addr));
1442 while (!is_done()) {
1443 if (net_server_loop())
1447 for_each_dip(dip, i)
1448 for_each_tip(dip, tip, j)
1449 tip_ftrunc_final(tip);
1457 * cleanup for next run
1459 for_each_dip(dip, i) {
1460 for_each_tip(dip, tip, j)
1466 free(device_information);
1467 device_information = NULL;
1477 * Setup outgoing network connection where we will transmit data
1479 static int net_setup_client(void)
1481 struct sockaddr_in addr;
1484 fd = socket(AF_INET, SOCK_STREAM, 0);
1486 perror("client: socket");
1490 memset(&addr, 0, sizeof(addr));
1491 addr.sin_family = AF_INET;
1492 addr.sin_port = htons(net_port);
1494 if (inet_aton(hostname, &addr.sin_addr) != 1) {
1495 struct hostent *hent = gethostbyname(hostname);
1497 perror("gethostbyname");
1501 memcpy(&addr.sin_addr, hent->h_addr, 4);
1502 strcpy(hostname, hent->h_name);
1505 printf("blktrace: connecting to %s\n", hostname);
1507 if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1508 perror("client: connect");
1512 printf("blktrace: connected!\n");
1517 static char usage_str[] = \
1518 "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1519 "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1520 "\t-d Use specified device. May also be given last after options\n" \
1521 "\t-r Path to mounted relayfs, defaults to /relay\n" \
1522 "\t-o File(s) to send output to\n" \
1523 "\t-D Directory to prepend to output file names\n" \
1524 "\t-k Kill a running trace\n" \
1525 "\t-w Stop after defined time, in seconds\n" \
1526 "\t-a Only trace specified actions. See documentation\n" \
1527 "\t-A Give trace mask as a single value. See documentation\n" \
1528 "\t-b Sub buffer size in KiB\n" \
1529 "\t-n Number of sub buffers\n" \
1530 "\t-l Run in network listen mode (blktrace server)\n" \
1531 "\t-h Run in network client mode, connecting to the given host\n" \
1532 "\t-p Network port to use (default 8462)\n" \
1533 "\t-s Make the network client use sendfile() to transfer data\n" \
1534 "\t-V Print program version info\n\n";
1536 static void show_usage(char *program)
1538 fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1541 int main(int argc, char *argv[])
1543 static char default_relay_path[] = "/relay";
1547 int act_mask_tmp = 0;
1549 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1552 i = find_mask_map(optarg);
1554 fprintf(stderr,"Invalid action mask %s\n",
1562 if ((sscanf(optarg, "%x", &i) != 1) ||
1563 !valid_act_opt(i)) {
1565 "Invalid set action mask %s/0x%x\n",
1573 if (resize_devices(optarg) != 0)
1578 relay_path = optarg;
1582 output_name = optarg;
1585 kill_running_trace = 1;
1588 stop_watch = atoi(optarg);
1589 if (stop_watch <= 0) {
1591 "Invalid stopwatch value (%d secs)\n",
1597 printf("%s version %s\n", argv[0], blktrace_version);
1600 buf_size = strtoul(optarg, NULL, 10);
1601 if (buf_size <= 0 || buf_size > 16*1024) {
1603 "Invalid buffer size (%lu)\n",buf_size);
1609 buf_nr = strtoul(optarg, NULL, 10);
1612 "Invalid buffer nr (%lu)\n", buf_nr);
1617 output_dir = optarg;
1620 net_mode = Net_client;
1621 strcpy(hostname, optarg);
1624 net_mode = Net_server;
1627 net_port = atoi(optarg);
1633 show_usage(argv[0]);
1638 setlocale(LC_NUMERIC, "en_US");
1640 page_size = getpagesize();
1642 if (net_mode == Net_server)
1643 return net_server();
1645 while (optind < argc) {
1646 if (resize_devices(argv[optind++]) != 0)
1651 show_usage(argv[0]);
1656 relay_path = default_relay_path;
1658 if (act_mask_tmp != 0)
1659 act_mask = act_mask_tmp;
1661 if (statfs(relay_path, &st) < 0) {
1663 fprintf(stderr,"%s does not appear to be a valid path\n",
1666 } else if (st.f_type != (long) RELAYFS_TYPE) {
1667 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1672 if (open_devices() != 0)
1675 if (kill_running_trace) {
1680 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1682 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1686 signal(SIGINT, handle_sigint);
1687 signal(SIGHUP, handle_sigint);
1688 signal(SIGTERM, handle_sigint);
1689 signal(SIGALRM, handle_sigint);
1691 if (net_mode == Net_client && net_setup_client())
1694 if (start_devices() != 0)
1697 atexit(stop_all_tracing);
1704 if (!is_trace_stopped()) {