Commit | Line | Data |
---|---|---|
d0ca268b JA |
1 | /* |
2 | * block queue tracing application | |
3 | * | |
d956a2cd JA |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> |
5 | * | |
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. | |
10 | * | |
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. | |
15 | * | |
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 | |
19 | * | |
d0ca268b JA |
20 | */ |
21 | #include <pthread.h> | |
22 | #include <sys/types.h> | |
23 | #include <sys/stat.h> | |
24 | #include <unistd.h> | |
25 | #include <locale.h> | |
26 | #include <signal.h> | |
27 | #include <fcntl.h> | |
28 | #include <string.h> | |
29 | #include <sys/ioctl.h> | |
b9d4294e | 30 | #include <sys/param.h> |
e3e74029 | 31 | #include <sys/statfs.h> |
eb3c8108 | 32 | #include <sys/poll.h> |
b7106311 | 33 | #include <sys/mman.h> |
8e86c98a | 34 | #include <sys/socket.h> |
d0ca268b JA |
35 | #include <stdio.h> |
36 | #include <stdlib.h> | |
37 | #include <sched.h> | |
d39c04ca AB |
38 | #include <ctype.h> |
39 | #include <getopt.h> | |
da39451f | 40 | #include <errno.h> |
8e86c98a JA |
41 | #include <netinet/in.h> |
42 | #include <arpa/inet.h> | |
43 | #include <netdb.h> | |
32f18c48 | 44 | #include <sys/sendfile.h> |
d0ca268b JA |
45 | |
46 | #include "blktrace.h" | |
21f55651 | 47 | #include "barrier.h" |
d0ca268b | 48 | |
13d928f0 | 49 | static char blktrace_version[] = "0.99"; |
52724a0e | 50 | |
8f551a39 JA |
51 | /* |
52 | * You may want to increase this even more, if you are logging at a high | |
53 | * rate and see skipped/missed events | |
54 | */ | |
007c233c | 55 | #define BUF_SIZE (512 * 1024) |
d0ca268b JA |
56 | #define BUF_NR (4) |
57 | ||
007c233c JA |
58 | #define OFILE_BUF (128 * 1024) |
59 | ||
e3e74029 NS |
60 | #define RELAYFS_TYPE 0xF0B4A981 |
61 | ||
32f18c48 | 62 | #define S_OPTS "d:a:A:r:o:kw:Vb:n:D:lh:p:s" |
d5396421 | 63 | static struct option l_opts[] = { |
5c86134e | 64 | { |
d39c04ca | 65 | .name = "dev", |
428683db | 66 | .has_arg = required_argument, |
d39c04ca AB |
67 | .flag = NULL, |
68 | .val = 'd' | |
69 | }, | |
5c86134e | 70 | { |
d39c04ca | 71 | .name = "act-mask", |
428683db | 72 | .has_arg = required_argument, |
d39c04ca AB |
73 | .flag = NULL, |
74 | .val = 'a' | |
75 | }, | |
5c86134e | 76 | { |
d39c04ca | 77 | .name = "set-mask", |
428683db | 78 | .has_arg = required_argument, |
d39c04ca AB |
79 | .flag = NULL, |
80 | .val = 'A' | |
81 | }, | |
5c86134e | 82 | { |
5270dddd | 83 | .name = "relay", |
428683db | 84 | .has_arg = required_argument, |
5270dddd JA |
85 | .flag = NULL, |
86 | .val = 'r' | |
87 | }, | |
d5396421 JA |
88 | { |
89 | .name = "output", | |
428683db | 90 | .has_arg = required_argument, |
d5396421 JA |
91 | .flag = NULL, |
92 | .val = 'o' | |
93 | }, | |
bc39777c JA |
94 | { |
95 | .name = "kill", | |
428683db | 96 | .has_arg = no_argument, |
bc39777c JA |
97 | .flag = NULL, |
98 | .val = 'k' | |
99 | }, | |
ece238a6 NS |
100 | { |
101 | .name = "stopwatch", | |
428683db | 102 | .has_arg = required_argument, |
ece238a6 NS |
103 | .flag = NULL, |
104 | .val = 'w' | |
105 | }, | |
52724a0e JA |
106 | { |
107 | .name = "version", | |
108 | .has_arg = no_argument, | |
109 | .flag = NULL, | |
57ea8602 | 110 | .val = 'V' |
52724a0e | 111 | }, |
129aa440 | 112 | { |
3f65c585 | 113 | .name = "buffer-size", |
129aa440 JA |
114 | .has_arg = required_argument, |
115 | .flag = NULL, | |
116 | .val = 'b' | |
117 | }, | |
118 | { | |
3f65c585 | 119 | .name = "num-sub-buffers", |
129aa440 JA |
120 | .has_arg = required_argument, |
121 | .flag = NULL, | |
122 | .val = 'n' | |
123 | }, | |
d1d7f15f | 124 | { |
3f65c585 | 125 | .name = "output-dir", |
d1d7f15f JA |
126 | .has_arg = required_argument, |
127 | .flag = NULL, | |
128 | .val = 'D' | |
129 | }, | |
8e86c98a JA |
130 | { |
131 | .name = "listen", | |
132 | .has_arg = no_argument, | |
133 | .flag = NULL, | |
134 | .val = 'l' | |
135 | }, | |
136 | { | |
137 | .name = "host", | |
138 | .has_arg = required_argument, | |
139 | .flag = NULL, | |
140 | .val = 'h' | |
141 | }, | |
142 | { | |
143 | .name = "port", | |
144 | .has_arg = required_argument, | |
145 | .flag = NULL, | |
146 | .val = 'p' | |
147 | }, | |
32f18c48 JA |
148 | { |
149 | .name = "sendfile", | |
150 | .has_arg = no_argument, | |
151 | .flag = NULL, | |
152 | .val = 's' | |
153 | }, | |
71ef8b7c JA |
154 | { |
155 | .name = NULL, | |
156 | } | |
d39c04ca AB |
157 | }; |
158 | ||
9db17354 | 159 | struct tip_subbuf { |
9db17354 | 160 | void *buf; |
5be4bdaf JA |
161 | unsigned int len; |
162 | unsigned int max_len; | |
32f18c48 | 163 | off_t offset; |
9db17354 JA |
164 | }; |
165 | ||
21f55651 JA |
166 | #define FIFO_SIZE (1024) /* should be plenty big! */ |
167 | #define CL_SIZE (128) /* cache line, any bigger? */ | |
168 | ||
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]; | |
173 | }; | |
174 | ||
d0ca268b JA |
175 | struct thread_information { |
176 | int cpu; | |
177 | pthread_t thread; | |
b9d4294e JA |
178 | |
179 | int fd; | |
a3e4d330 | 180 | void *fd_buf; |
b9d4294e JA |
181 | char fn[MAXPATHLEN + 64]; |
182 | ||
007c233c JA |
183 | FILE *ofile; |
184 | char *ofile_buffer; | |
32f18c48 | 185 | off_t ofile_offset; |
9db17354 | 186 | int ofile_stdout; |
8e86c98a | 187 | int ofile_mmap; |
007c233c | 188 | |
0cc7d25e JA |
189 | int (*get_subbuf)(struct thread_information *, unsigned int); |
190 | int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *); | |
191 | int (*read_data)(struct thread_information *, void *, unsigned int); | |
192 | ||
d0ca268b | 193 | unsigned long events_processed; |
b7106311 | 194 | unsigned long long data_read; |
e7c9f3ff | 195 | struct device_information *device; |
9db17354 JA |
196 | |
197 | int exited; | |
198 | ||
b7106311 JA |
199 | /* |
200 | * piped fifo buffers | |
201 | */ | |
21f55651 | 202 | struct tip_subbuf_fifo fifo; |
7de86b12 | 203 | struct tip_subbuf *leftover_ts; |
b7106311 JA |
204 | |
205 | /* | |
206 | * mmap controlled output files | |
207 | */ | |
208 | unsigned long long fs_size; | |
209 | unsigned long long fs_max_size; | |
210 | unsigned long fs_off; | |
211 | void *fs_buf; | |
212 | unsigned long fs_buf_len; | |
d0ca268b JA |
213 | }; |
214 | ||
e7c9f3ff NS |
215 | struct device_information { |
216 | int fd; | |
217 | char *path; | |
218 | char buts_name[32]; | |
99c1f5ab | 219 | volatile int trace_started; |
eb3c8108 | 220 | unsigned long drop_count; |
e7c9f3ff NS |
221 | struct thread_information *threads; |
222 | }; | |
d0ca268b | 223 | |
e7c9f3ff | 224 | static int ncpus; |
d0ca268b | 225 | static struct thread_information *thread_information; |
e7c9f3ff NS |
226 | static int ndevs; |
227 | static struct device_information *device_information; | |
228 | ||
229 | /* command line option globals */ | |
230 | static char *relay_path; | |
d5396421 | 231 | static char *output_name; |
d1d7f15f | 232 | static char *output_dir; |
5c86134e | 233 | static int act_mask = ~0U; |
bc39777c | 234 | static int kill_running_trace; |
eb3c8108 JA |
235 | static unsigned long buf_size = BUF_SIZE; |
236 | static unsigned long buf_nr = BUF_NR; | |
b7106311 | 237 | static unsigned int page_size; |
d39c04ca | 238 | |
e7c9f3ff NS |
239 | #define is_done() (*(volatile int *)(&done)) |
240 | static volatile int done; | |
241 | ||
eb3c8108 JA |
242 | #define is_trace_stopped() (*(volatile int *)(&trace_stopped)) |
243 | static volatile int trace_stopped; | |
244 | ||
245 | #define is_stat_shown() (*(volatile int *)(&stat_shown)) | |
246 | static volatile int stat_shown; | |
a3e4d330 | 247 | |
8e86c98a JA |
248 | int data_is_native = -1; |
249 | ||
72ca8801 NS |
250 | static void exit_trace(int status); |
251 | ||
99c1f5ab JA |
252 | #define dip_tracing(dip) (*(volatile int *)(&(dip)->trace_started)) |
253 | #define dip_set_tracing(dip, v) ((dip)->trace_started = (v)) | |
254 | ||
255 | #define __for_each_dip(__d, __i, __e) \ | |
256 | for (__i = 0, __d = device_information; __i < __e; __i++, __d++) | |
257 | ||
258 | #define for_each_dip(__d, __i) __for_each_dip(__d, __i, ndevs) | |
9db17354 JA |
259 | #define for_each_tip(__d, __t, __j) \ |
260 | for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++) | |
99c1f5ab | 261 | |
8e86c98a JA |
262 | /* |
263 | * networking stuff follows. we include a magic number so we know whether | |
264 | * to endianness convert or not | |
265 | */ | |
266 | struct blktrace_net_hdr { | |
267 | u32 magic; /* same as trace magic */ | |
22cd0c02 | 268 | char buts_name[32]; /* trace name */ |
8e86c98a | 269 | u32 cpu; /* for which cpu */ |
22cd0c02 | 270 | u32 max_cpus; |
8e86c98a JA |
271 | u32 len; /* length of following trace data */ |
272 | }; | |
273 | ||
274 | #define TRACE_NET_PORT (8462) | |
275 | ||
276 | enum { | |
277 | Net_none = 0, | |
278 | Net_server, | |
279 | Net_client, | |
280 | }; | |
281 | ||
282 | /* | |
283 | * network cmd line params | |
284 | */ | |
285 | static char hostname[MAXHOSTNAMELEN]; | |
286 | static int net_port = TRACE_NET_PORT; | |
287 | static int net_mode = 0; | |
32f18c48 | 288 | static int net_sendfile; |
8e86c98a JA |
289 | |
290 | static int net_in_fd = -1; | |
291 | static int net_out_fd = -1; | |
292 | ||
293 | static void handle_sigint(__attribute__((__unused__)) int sig) | |
294 | { | |
295 | done = 1; | |
296 | } | |
297 | ||
eb3c8108 JA |
298 | static int get_dropped_count(const char *buts_name) |
299 | { | |
300 | int fd; | |
301 | char tmp[MAXPATHLEN + 64]; | |
302 | ||
303 | snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped", | |
304 | relay_path, buts_name); | |
305 | ||
306 | fd = open(tmp, O_RDONLY); | |
307 | if (fd < 0) { | |
308 | /* | |
309 | * this may be ok, if the kernel doesn't support dropped counts | |
310 | */ | |
311 | if (errno == ENOENT) | |
312 | return 0; | |
313 | ||
314 | fprintf(stderr, "Couldn't open dropped file %s\n", tmp); | |
315 | return -1; | |
316 | } | |
317 | ||
318 | if (read(fd, tmp, sizeof(tmp)) < 0) { | |
319 | perror(tmp); | |
320 | close(fd); | |
321 | return -1; | |
322 | } | |
323 | ||
324 | close(fd); | |
325 | ||
326 | return atoi(tmp); | |
327 | } | |
328 | ||
e7c9f3ff | 329 | static int start_trace(struct device_information *dip) |
d0ca268b JA |
330 | { |
331 | struct blk_user_trace_setup buts; | |
332 | ||
1f79c4a0 | 333 | memset(&buts, 0, sizeof(buts)); |
129aa440 JA |
334 | buts.buf_size = buf_size; |
335 | buts.buf_nr = buf_nr; | |
d39c04ca | 336 | buts.act_mask = act_mask; |
d0ca268b | 337 | |
ed71a31e JA |
338 | if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) { |
339 | perror("BLKTRACESETUP"); | |
340 | return 1; | |
341 | } | |
342 | ||
343 | if (ioctl(dip->fd, BLKTRACESTART) < 0) { | |
344 | perror("BLKTRACESTART"); | |
d0ca268b JA |
345 | return 1; |
346 | } | |
347 | ||
e7c9f3ff | 348 | memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name)); |
99c1f5ab | 349 | dip_set_tracing(dip, 1); |
d0ca268b JA |
350 | return 0; |
351 | } | |
352 | ||
e7c9f3ff | 353 | static void stop_trace(struct device_information *dip) |
d0ca268b | 354 | { |
99c1f5ab JA |
355 | if (dip_tracing(dip) || kill_running_trace) { |
356 | dip_set_tracing(dip, 0); | |
cf9208ea | 357 | |
ed71a31e JA |
358 | if (ioctl(dip->fd, BLKTRACESTOP) < 0) |
359 | perror("BLKTRACESTOP"); | |
360 | if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0) | |
361 | perror("BLKTRACETEARDOWN"); | |
cf9208ea | 362 | |
e7c9f3ff | 363 | close(dip->fd); |
cf9208ea | 364 | dip->fd = -1; |
707b0914 | 365 | } |
d0ca268b JA |
366 | } |
367 | ||
e7c9f3ff NS |
368 | static void stop_all_traces(void) |
369 | { | |
370 | struct device_information *dip; | |
371 | int i; | |
372 | ||
eb3c8108 JA |
373 | for_each_dip(dip, i) { |
374 | dip->drop_count = get_dropped_count(dip->buts_name); | |
e7c9f3ff | 375 | stop_trace(dip); |
eb3c8108 | 376 | } |
e7c9f3ff NS |
377 | } |
378 | ||
eb3c8108 JA |
379 | static void wait_for_data(struct thread_information *tip) |
380 | { | |
381 | struct pollfd pfd = { .fd = tip->fd, .events = POLLIN }; | |
382 | ||
9db17354 | 383 | do { |
b4aabcb3 | 384 | poll(&pfd, 1, 100); |
9db17354 JA |
385 | if (pfd.revents & POLLIN) |
386 | break; | |
387 | if (tip->ofile_stdout) | |
388 | break; | |
389 | } while (!is_done()); | |
eb3c8108 JA |
390 | } |
391 | ||
0cc7d25e JA |
392 | static int read_data_file(struct thread_information *tip, void *buf, |
393 | unsigned int len) | |
d0ca268b | 394 | { |
ae9f71b3 | 395 | int ret = 0; |
bbabf03a | 396 | |
9db17354 JA |
397 | do { |
398 | wait_for_data(tip); | |
ae9f71b3 | 399 | |
9db17354 JA |
400 | ret = read(tip->fd, buf, len); |
401 | if (!ret) | |
402 | continue; | |
403 | else if (ret > 0) | |
404 | return ret; | |
405 | else { | |
bbabf03a | 406 | if (errno != EAGAIN) { |
a3e4d330 JA |
407 | perror(tip->fn); |
408 | fprintf(stderr,"Thread %d failed read of %s\n", | |
409 | tip->cpu, tip->fn); | |
410 | break; | |
411 | } | |
9db17354 | 412 | continue; |
bbabf03a | 413 | } |
9db17354 | 414 | } while (!is_done()); |
8a43bac5 | 415 | |
bbabf03a | 416 | return ret; |
8e86c98a | 417 | |
8a43bac5 JA |
418 | } |
419 | ||
0cc7d25e JA |
420 | static int read_data_net(struct thread_information *tip, void *buf, |
421 | unsigned int len) | |
8e86c98a JA |
422 | { |
423 | unsigned int bytes_left = len; | |
424 | int ret = 0; | |
425 | ||
426 | do { | |
427 | ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL); | |
428 | ||
429 | if (!ret) | |
430 | continue; | |
431 | else if (ret < 0) { | |
432 | if (errno != EAGAIN) { | |
433 | perror(tip->fn); | |
434 | fprintf(stderr, "server: failed read\n"); | |
435 | return 0; | |
436 | } | |
437 | continue; | |
438 | } else { | |
439 | buf += ret; | |
440 | bytes_left -= ret; | |
441 | } | |
442 | } while (!is_done() && bytes_left); | |
443 | ||
410d7c62 | 444 | return len - bytes_left; |
8e86c98a JA |
445 | } |
446 | ||
0cc7d25e JA |
447 | static int read_data(struct thread_information *tip, void *buf, |
448 | unsigned int len) | |
8e86c98a | 449 | { |
0cc7d25e | 450 | int ret = tip->read_data(tip, buf, len); |
8e86c98a | 451 | |
410d7c62 JA |
452 | if (ret > 0) |
453 | tip->data_read += ret; | |
454 | ||
455 | return ret; | |
8e86c98a JA |
456 | } |
457 | ||
458 | static inline struct tip_subbuf * | |
459 | subbuf_fifo_dequeue(struct thread_information *tip) | |
a3e4d330 | 460 | { |
21f55651 JA |
461 | const int head = tip->fifo.head; |
462 | const int next = (head + 1) & (FIFO_SIZE - 1); | |
463 | ||
464 | if (head != tip->fifo.tail) { | |
465 | struct tip_subbuf *ts = tip->fifo.q[head]; | |
466 | ||
467 | store_barrier(); | |
468 | tip->fifo.head = next; | |
469 | return ts; | |
470 | } | |
471 | ||
472 | return NULL; | |
9db17354 | 473 | } |
eb3c8108 | 474 | |
21f55651 JA |
475 | static inline int subbuf_fifo_queue(struct thread_information *tip, |
476 | struct tip_subbuf *ts) | |
9db17354 | 477 | { |
21f55651 JA |
478 | const int tail = tip->fifo.tail; |
479 | const int next = (tail + 1) & (FIFO_SIZE - 1); | |
480 | ||
481 | if (next != tip->fifo.head) { | |
482 | tip->fifo.q[tail] = ts; | |
483 | store_barrier(); | |
484 | tip->fifo.tail = next; | |
485 | return 0; | |
486 | } | |
487 | ||
488 | fprintf(stderr, "fifo too small!\n"); | |
489 | return 1; | |
a3e4d330 JA |
490 | } |
491 | ||
b7106311 JA |
492 | /* |
493 | * For file output, truncate and mmap the file appropriately | |
494 | */ | |
8e86c98a | 495 | static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen) |
b7106311 JA |
496 | { |
497 | int ofd = fileno(tip->ofile); | |
498 | int ret; | |
499 | ||
500 | /* | |
501 | * extend file, if we have to. use chunks of 16 subbuffers. | |
502 | */ | |
503 | if (tip->fs_off + buf_size > tip->fs_buf_len) { | |
504 | if (tip->fs_buf) { | |
5975d309 | 505 | munlock(tip->fs_buf, tip->fs_buf_len); |
b7106311 JA |
506 | munmap(tip->fs_buf, tip->fs_buf_len); |
507 | tip->fs_buf = NULL; | |
508 | } | |
509 | ||
510 | tip->fs_off = tip->fs_size & (page_size - 1); | |
511 | tip->fs_buf_len = (16 * buf_size) - tip->fs_off; | |
512 | tip->fs_max_size += tip->fs_buf_len; | |
513 | ||
514 | if (ftruncate(ofd, tip->fs_max_size) < 0) { | |
515 | perror("ftruncate"); | |
516 | return -1; | |
517 | } | |
518 | ||
519 | tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE, | |
520 | MAP_SHARED, ofd, tip->fs_size - tip->fs_off); | |
521 | if (tip->fs_buf == MAP_FAILED) { | |
522 | perror("mmap"); | |
523 | return -1; | |
524 | } | |
5975d309 | 525 | mlock(tip->fs_buf, tip->fs_buf_len); |
b7106311 JA |
526 | } |
527 | ||
8e86c98a | 528 | ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen); |
b7106311 | 529 | if (ret >= 0) { |
b7106311 JA |
530 | tip->fs_size += ret; |
531 | tip->fs_off += ret; | |
532 | return 0; | |
533 | } | |
534 | ||
535 | return -1; | |
536 | } | |
537 | ||
32f18c48 JA |
538 | static int get_subbuf_sendfile(struct thread_information *tip, |
539 | unsigned int maxlen) | |
540 | { | |
541 | struct tip_subbuf *ts = malloc(sizeof(*ts)); | |
542 | struct stat sb; | |
543 | ||
544 | ts->buf = malloc(buf_size); | |
545 | ts->max_len = maxlen; | |
546 | ts->buf = NULL; | |
547 | ||
548 | if (fstat(tip->fd, &sb) < 0) { | |
549 | perror("trace stat"); | |
550 | return 1; | |
551 | } | |
552 | ||
553 | ts->len = sb.st_size - tip->ofile_offset; | |
554 | ts->max_len = ts->len; | |
555 | ts->offset = tip->ofile_offset; | |
556 | tip->ofile_offset += ts->len; | |
557 | return subbuf_fifo_queue(tip, ts); | |
558 | } | |
559 | ||
b7106311 | 560 | /* |
22cd0c02 | 561 | * Use the copy approach for pipes and network |
b7106311 | 562 | */ |
0cc7d25e | 563 | static int get_subbuf(struct thread_information *tip, unsigned int maxlen) |
a3e4d330 | 564 | { |
8e86c98a | 565 | struct tip_subbuf *ts = malloc(sizeof(*ts)); |
22cd0c02 | 566 | int ret; |
8e86c98a | 567 | |
22cd0c02 | 568 | ts->buf = malloc(buf_size); |
0cc7d25e | 569 | ts->max_len = maxlen; |
8e86c98a | 570 | |
22cd0c02 JA |
571 | ret = read_data(tip, ts->buf, ts->max_len); |
572 | if (ret > 0) { | |
573 | ts->len = ret; | |
574 | return subbuf_fifo_queue(tip, ts); | |
9db17354 | 575 | } |
a3e4d330 | 576 | |
b7106311 | 577 | return ret; |
a3e4d330 JA |
578 | } |
579 | ||
9db17354 | 580 | static void close_thread(struct thread_information *tip) |
a3e4d330 | 581 | { |
9db17354 JA |
582 | if (tip->fd != -1) |
583 | close(tip->fd); | |
584 | if (tip->ofile) | |
585 | fclose(tip->ofile); | |
586 | if (tip->ofile_buffer) | |
587 | free(tip->ofile_buffer); | |
588 | if (tip->fd_buf) | |
589 | free(tip->fd_buf); | |
1c99bc21 | 590 | |
9db17354 JA |
591 | tip->fd = -1; |
592 | tip->ofile = NULL; | |
593 | tip->ofile_buffer = NULL; | |
594 | tip->fd_buf = NULL; | |
a3e4d330 JA |
595 | } |
596 | ||
8e86c98a JA |
597 | static void tip_ftrunc_final(struct thread_information *tip) |
598 | { | |
599 | /* | |
600 | * truncate to right size and cleanup mmap | |
601 | */ | |
602 | if (tip->ofile_mmap) { | |
603 | int ofd = fileno(tip->ofile); | |
604 | ||
605 | if (tip->fs_buf) | |
606 | munmap(tip->fs_buf, tip->fs_buf_len); | |
607 | ||
608 | ftruncate(ofd, tip->fs_size); | |
609 | } | |
610 | } | |
611 | ||
9db17354 | 612 | static void *thread_main(void *arg) |
a3e4d330 | 613 | { |
9db17354 JA |
614 | struct thread_information *tip = arg; |
615 | pid_t pid = getpid(); | |
616 | cpu_set_t cpu_mask; | |
a3e4d330 | 617 | |
9db17354 JA |
618 | CPU_ZERO(&cpu_mask); |
619 | CPU_SET((tip->cpu), &cpu_mask); | |
a3e4d330 | 620 | |
9db17354 JA |
621 | if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) { |
622 | perror("sched_setaffinity"); | |
623 | exit_trace(1); | |
624 | } | |
a3e4d330 | 625 | |
9db17354 JA |
626 | snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d", |
627 | relay_path, tip->device->buts_name, tip->cpu); | |
628 | tip->fd = open(tip->fn, O_RDONLY); | |
629 | if (tip->fd < 0) { | |
630 | perror(tip->fn); | |
631 | fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu, | |
632 | tip->fn); | |
633 | exit_trace(1); | |
a3e4d330 JA |
634 | } |
635 | ||
b7106311 | 636 | while (!is_done()) { |
0cc7d25e JA |
637 | if (tip->get_subbuf(tip, buf_size)) |
638 | break; | |
b7106311 JA |
639 | } |
640 | ||
8e86c98a JA |
641 | tip_ftrunc_final(tip); |
642 | tip->exited = 1; | |
643 | return NULL; | |
644 | } | |
b7106311 | 645 | |
8e86c98a JA |
646 | static int write_data_net(int fd, void *buf, unsigned int buf_len) |
647 | { | |
648 | unsigned int bytes_left = buf_len; | |
649 | int ret; | |
b7106311 | 650 | |
8e86c98a JA |
651 | while (bytes_left) { |
652 | ret = send(fd, buf, bytes_left, 0); | |
653 | if (ret < 0) { | |
654 | perror("send"); | |
655 | return 1; | |
656 | } | |
657 | ||
658 | buf += ret; | |
659 | bytes_left -= ret; | |
9db17354 | 660 | } |
a3e4d330 | 661 | |
8e86c98a | 662 | return 0; |
a3e4d330 JA |
663 | } |
664 | ||
32f18c48 | 665 | static int net_send_header(struct thread_information *tip, unsigned int len) |
8e86c98a JA |
666 | { |
667 | struct blktrace_net_hdr hdr; | |
8e86c98a JA |
668 | |
669 | hdr.magic = BLK_IO_TRACE_MAGIC; | |
22cd0c02 | 670 | strcpy(hdr.buts_name, tip->device->buts_name); |
8e86c98a | 671 | hdr.cpu = tip->cpu; |
22cd0c02 | 672 | hdr.max_cpus = ncpus; |
32f18c48 | 673 | hdr.len = len; |
8e86c98a | 674 | |
32f18c48 JA |
675 | return write_data_net(net_out_fd, &hdr, sizeof(hdr)); |
676 | } | |
8e86c98a | 677 | |
6a752c90 JA |
678 | /* |
679 | * send header with 0 length to signal end-of-run | |
680 | */ | |
681 | static void net_client_send_close(void) | |
682 | { | |
683 | struct blktrace_net_hdr hdr; | |
684 | ||
685 | hdr.magic = BLK_IO_TRACE_MAGIC; | |
686 | hdr.cpu = 0; | |
687 | hdr.max_cpus = ncpus; | |
688 | hdr.len = 0; | |
689 | ||
690 | write_data_net(net_out_fd, &hdr, sizeof(hdr)); | |
691 | } | |
692 | ||
32f18c48 JA |
693 | static int flush_subbuf_net(struct thread_information *tip, |
694 | struct tip_subbuf *ts) | |
695 | { | |
696 | if (net_send_header(tip, ts->len)) | |
697 | return 1; | |
22cd0c02 JA |
698 | if (write_data_net(net_out_fd, ts->buf, ts->len)) |
699 | return 1; | |
8e86c98a | 700 | |
f0597a7e | 701 | free(ts->buf); |
8e86c98a JA |
702 | free(ts); |
703 | return 0; | |
704 | } | |
705 | ||
32f18c48 JA |
706 | static int flush_subbuf_sendfile(struct thread_information *tip, |
707 | struct tip_subbuf *ts) | |
708 | { | |
709 | if (net_send_header(tip, ts->len)) | |
710 | return 1; | |
711 | if (sendfile(net_out_fd, tip->fd, &ts->offset, ts->len) < 0) { | |
712 | perror("sendfile"); | |
713 | return 1; | |
714 | } | |
715 | ||
716 | free(ts); | |
717 | return 0; | |
718 | } | |
719 | ||
8e86c98a JA |
720 | static int write_data(struct thread_information *tip, void *buf, |
721 | unsigned int buf_len) | |
8a43bac5 | 722 | { |
7126171a | 723 | int ret; |
8a43bac5 | 724 | |
6480258a JA |
725 | if (!buf_len) |
726 | return 0; | |
727 | ||
7126171a JA |
728 | while (1) { |
729 | ret = fwrite(buf, buf_len, 1, tip->ofile); | |
007c233c | 730 | if (ret == 1) |
8a43bac5 JA |
731 | break; |
732 | ||
db6fe5bc JA |
733 | if (ret < 0) { |
734 | perror("write"); | |
735 | return 1; | |
8a43bac5 | 736 | } |
d0ca268b JA |
737 | } |
738 | ||
9db17354 | 739 | if (tip->ofile_stdout) |
7126171a JA |
740 | fflush(tip->ofile); |
741 | ||
8a43bac5 JA |
742 | return 0; |
743 | } | |
744 | ||
8e86c98a JA |
745 | static int flush_subbuf_file(struct thread_information *tip, |
746 | struct tip_subbuf *ts) | |
8a43bac5 | 747 | { |
9db17354 JA |
748 | unsigned int offset = 0; |
749 | struct blk_io_trace *t; | |
750 | int pdu_len, events = 0; | |
8a43bac5 | 751 | |
9db17354 | 752 | /* |
7de86b12 | 753 | * surplus from last run |
9db17354 | 754 | */ |
7de86b12 AB |
755 | if (tip->leftover_ts) { |
756 | struct tip_subbuf *prev_ts = tip->leftover_ts; | |
757 | ||
9e8b753c | 758 | if (prev_ts->len + ts->len > prev_ts->max_len) { |
7de86b12 AB |
759 | prev_ts->max_len += ts->len; |
760 | prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len); | |
761 | } | |
762 | ||
9e8b753c | 763 | memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len); |
7de86b12 AB |
764 | prev_ts->len += ts->len; |
765 | ||
766 | free(ts->buf); | |
767 | free(ts); | |
768 | ||
769 | ts = prev_ts; | |
770 | tip->leftover_ts = NULL; | |
9db17354 | 771 | } |
d0ca268b | 772 | |
9db17354 JA |
773 | while (offset + sizeof(*t) <= ts->len) { |
774 | t = ts->buf + offset; | |
3a9d6c13 | 775 | |
9cfa6c2b AB |
776 | if (verify_trace(t)) { |
777 | write_data(tip, ts->buf, offset); | |
9db17354 | 778 | return -1; |
9cfa6c2b | 779 | } |
3a9d6c13 | 780 | |
9db17354 | 781 | pdu_len = t->pdu_len; |
3a9d6c13 | 782 | |
9db17354 | 783 | if (offset + sizeof(*t) + pdu_len > ts->len) |
3a9d6c13 | 784 | break; |
4b5db44a | 785 | |
9db17354 JA |
786 | offset += sizeof(*t) + pdu_len; |
787 | tip->events_processed++; | |
b7106311 | 788 | tip->data_read += sizeof(*t) + pdu_len; |
9db17354 | 789 | events++; |
3a9d6c13 JA |
790 | } |
791 | ||
9cfa6c2b AB |
792 | if (write_data(tip, ts->buf, offset)) |
793 | return -1; | |
794 | ||
3a9d6c13 | 795 | /* |
9db17354 | 796 | * leftover bytes, save them for next time |
3a9d6c13 | 797 | */ |
9db17354 | 798 | if (offset != ts->len) { |
7de86b12 | 799 | tip->leftover_ts = ts; |
9e8b753c JA |
800 | ts->len -= offset; |
801 | memmove(ts->buf, ts->buf + offset, ts->len); | |
7de86b12 AB |
802 | } else { |
803 | free(ts->buf); | |
804 | free(ts); | |
9db17354 | 805 | } |
4b5db44a | 806 | |
9db17354 | 807 | return events; |
4b5db44a JA |
808 | } |
809 | ||
9db17354 | 810 | static int write_tip_events(struct thread_information *tip) |
d5396421 | 811 | { |
21f55651 | 812 | struct tip_subbuf *ts = subbuf_fifo_dequeue(tip); |
d5396421 | 813 | |
0cc7d25e JA |
814 | if (ts) |
815 | return tip->flush_subbuf(tip, ts); | |
91816d54 | 816 | |
9db17354 | 817 | return 0; |
91816d54 JA |
818 | } |
819 | ||
9db17354 JA |
820 | /* |
821 | * scans the tips we know and writes out the subbuffers we accumulate | |
822 | */ | |
823 | static void get_and_write_events(void) | |
d0ca268b | 824 | { |
9db17354 JA |
825 | struct device_information *dip; |
826 | struct thread_information *tip; | |
27223f19 | 827 | int i, j, events, ret, tips_running; |
d0ca268b | 828 | |
9db17354 JA |
829 | while (!is_done()) { |
830 | events = 0; | |
d0ca268b | 831 | |
9db17354 JA |
832 | for_each_dip(dip, i) { |
833 | for_each_tip(dip, tip, j) { | |
834 | ret = write_tip_events(tip); | |
835 | if (ret > 0) | |
836 | events += ret; | |
837 | } | |
838 | } | |
d0ca268b | 839 | |
9db17354 JA |
840 | if (!events) |
841 | usleep(10); | |
d0ca268b JA |
842 | } |
843 | ||
a3e4d330 | 844 | /* |
9db17354 | 845 | * reap stored events |
a3e4d330 | 846 | */ |
9db17354 JA |
847 | do { |
848 | events = 0; | |
27223f19 | 849 | tips_running = 0; |
9db17354 JA |
850 | for_each_dip(dip, i) { |
851 | for_each_tip(dip, tip, j) { | |
852 | ret = write_tip_events(tip); | |
853 | if (ret > 0) | |
854 | events += ret; | |
27223f19 | 855 | tips_running += !tip->exited; |
9db17354 | 856 | } |
69e65a9e | 857 | } |
9db17354 | 858 | usleep(10); |
27223f19 | 859 | } while (events || tips_running); |
d0ca268b JA |
860 | } |
861 | ||
b7106311 JA |
862 | static void wait_for_threads(void) |
863 | { | |
864 | /* | |
8e86c98a JA |
865 | * for piped or network output, poll and fetch data for writeout. |
866 | * for files, we just wait around for trace threads to exit | |
b7106311 | 867 | */ |
8e86c98a JA |
868 | if ((output_name && !strcmp(output_name, "-")) || |
869 | net_mode == Net_client) | |
b7106311 JA |
870 | get_and_write_events(); |
871 | else { | |
872 | struct device_information *dip; | |
873 | struct thread_information *tip; | |
874 | int i, j, tips_running; | |
875 | ||
876 | do { | |
877 | tips_running = 0; | |
878 | usleep(1000); | |
879 | ||
880 | for_each_dip(dip, i) | |
881 | for_each_tip(dip, tip, j) | |
882 | tips_running += !tip->exited; | |
883 | } while (tips_running); | |
884 | } | |
6a752c90 JA |
885 | |
886 | if (net_mode == Net_client) | |
887 | net_client_send_close(); | |
b7106311 JA |
888 | } |
889 | ||
8e86c98a JA |
890 | static void fill_ofname(char *dst, char *buts_name, int cpu) |
891 | { | |
892 | int len = 0; | |
893 | ||
894 | if (output_dir) | |
895 | len = sprintf(dst, "%s/", output_dir); | |
896 | ||
897 | if (output_name) | |
898 | sprintf(dst + len, "%s.blktrace.%d", output_name, cpu); | |
899 | else | |
900 | sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu); | |
901 | } | |
902 | ||
0cc7d25e JA |
903 | static void fill_ops(struct thread_information *tip) |
904 | { | |
905 | /* | |
906 | * setup ops | |
907 | */ | |
32f18c48 JA |
908 | if (net_mode == Net_client) { |
909 | if (net_sendfile) { | |
910 | tip->get_subbuf = get_subbuf_sendfile; | |
911 | tip->flush_subbuf = flush_subbuf_sendfile; | |
912 | } else { | |
913 | tip->get_subbuf = get_subbuf; | |
914 | tip->flush_subbuf = flush_subbuf_net; | |
915 | } | |
916 | } else { | |
917 | if (tip->ofile_mmap) | |
918 | tip->get_subbuf = mmap_subbuf; | |
919 | else | |
920 | tip->get_subbuf = get_subbuf; | |
0cc7d25e | 921 | |
0cc7d25e | 922 | tip->flush_subbuf = flush_subbuf_file; |
32f18c48 JA |
923 | } |
924 | ||
0cc7d25e JA |
925 | if (net_mode == Net_server) |
926 | tip->read_data = read_data_net; | |
927 | else | |
928 | tip->read_data = read_data_file; | |
929 | } | |
930 | ||
e7c9f3ff | 931 | static int start_threads(struct device_information *dip) |
d0ca268b JA |
932 | { |
933 | struct thread_information *tip; | |
e7c9f3ff | 934 | int j, pipeline = output_name && !strcmp(output_name, "-"); |
8e86c98a | 935 | int mode, vbuf_size; |
22cd0c02 | 936 | char op[64]; |
d0ca268b | 937 | |
99c1f5ab | 938 | for_each_tip(dip, tip, j) { |
e7c9f3ff NS |
939 | tip->cpu = j; |
940 | tip->device = dip; | |
d0ca268b | 941 | tip->events_processed = 0; |
21f55651 | 942 | memset(&tip->fifo, 0, sizeof(tip->fifo)); |
9db17354 | 943 | tip->leftover_ts = NULL; |
d0ca268b | 944 | |
e7c9f3ff | 945 | if (pipeline) { |
007c233c | 946 | tip->ofile = fdopen(STDOUT_FILENO, "w"); |
9db17354 | 947 | tip->ofile_stdout = 1; |
8e86c98a | 948 | tip->ofile_mmap = 0; |
007c233c | 949 | mode = _IOLBF; |
57e8a2ad | 950 | vbuf_size = 512; |
d5396421 | 951 | } else { |
22cd0c02 JA |
952 | fill_ofname(op, dip->buts_name, tip->cpu); |
953 | tip->ofile = fopen(op, "w+"); | |
9db17354 | 954 | tip->ofile_stdout = 0; |
8e86c98a | 955 | tip->ofile_mmap = 1; |
007c233c | 956 | mode = _IOFBF; |
57e8a2ad | 957 | vbuf_size = OFILE_BUF; |
d5396421 JA |
958 | } |
959 | ||
007c233c | 960 | if (tip->ofile == NULL) { |
22cd0c02 | 961 | perror(op); |
e7c9f3ff | 962 | return 1; |
d5396421 JA |
963 | } |
964 | ||
57e8a2ad TZ |
965 | tip->ofile_buffer = malloc(vbuf_size); |
966 | if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) { | |
007c233c JA |
967 | perror("setvbuf"); |
968 | close_thread(tip); | |
969 | return 1; | |
970 | } | |
971 | ||
0cc7d25e JA |
972 | fill_ops(tip); |
973 | ||
9db17354 | 974 | if (pthread_create(&tip->thread, NULL, thread_main, tip)) { |
e7c9f3ff | 975 | perror("pthread_create"); |
007c233c | 976 | close_thread(tip); |
e7c9f3ff | 977 | return 1; |
d0ca268b JA |
978 | } |
979 | } | |
980 | ||
e7c9f3ff | 981 | return 0; |
d0ca268b JA |
982 | } |
983 | ||
e7c9f3ff | 984 | static void stop_threads(struct device_information *dip) |
3aabcd89 | 985 | { |
e7c9f3ff | 986 | struct thread_information *tip; |
91816d54 | 987 | unsigned long ret; |
007c233c JA |
988 | int i; |
989 | ||
9db17354 | 990 | for_each_tip(dip, tip, i) { |
91816d54 | 991 | (void) pthread_join(tip->thread, (void *) &ret); |
9db17354 JA |
992 | close_thread(tip); |
993 | } | |
3aabcd89 JA |
994 | } |
995 | ||
e7c9f3ff | 996 | static void stop_all_threads(void) |
72ca8801 | 997 | { |
e7c9f3ff | 998 | struct device_information *dip; |
72ca8801 NS |
999 | int i; |
1000 | ||
99c1f5ab | 1001 | for_each_dip(dip, i) |
e7c9f3ff NS |
1002 | stop_threads(dip); |
1003 | } | |
1004 | ||
1005 | static void stop_all_tracing(void) | |
1006 | { | |
1007 | struct device_information *dip; | |
91816d54 | 1008 | int i; |
007c233c | 1009 | |
91816d54 | 1010 | for_each_dip(dip, i) |
e7c9f3ff | 1011 | stop_trace(dip); |
72ca8801 NS |
1012 | } |
1013 | ||
1014 | static void exit_trace(int status) | |
1015 | { | |
eb3c8108 JA |
1016 | if (!is_trace_stopped()) { |
1017 | trace_stopped = 1; | |
1018 | stop_all_threads(); | |
1019 | stop_all_tracing(); | |
1020 | } | |
1021 | ||
72ca8801 NS |
1022 | exit(status); |
1023 | } | |
1024 | ||
e7c9f3ff NS |
1025 | static int resize_devices(char *path) |
1026 | { | |
1027 | int size = (ndevs + 1) * sizeof(struct device_information); | |
1028 | ||
1029 | device_information = realloc(device_information, size); | |
1030 | if (!device_information) { | |
1031 | fprintf(stderr, "Out of memory, device %s (%d)\n", path, size); | |
1032 | return 1; | |
1033 | } | |
1034 | device_information[ndevs].path = path; | |
1035 | ndevs++; | |
1036 | return 0; | |
1037 | } | |
1038 | ||
1039 | static int open_devices(void) | |
d0ca268b | 1040 | { |
e7c9f3ff | 1041 | struct device_information *dip; |
d0ca268b | 1042 | int i; |
d0ca268b | 1043 | |
99c1f5ab | 1044 | for_each_dip(dip, i) { |
cf9208ea | 1045 | dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK); |
e7c9f3ff NS |
1046 | if (dip->fd < 0) { |
1047 | perror(dip->path); | |
1048 | return 1; | |
1049 | } | |
1050 | } | |
99c1f5ab | 1051 | |
e7c9f3ff NS |
1052 | return 0; |
1053 | } | |
1054 | ||
1055 | static int start_devices(void) | |
1056 | { | |
1057 | struct device_information *dip; | |
1058 | int i, j, size; | |
1059 | ||
1060 | size = ncpus * sizeof(struct thread_information); | |
1061 | thread_information = malloc(size * ndevs); | |
1062 | if (!thread_information) { | |
1063 | fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs); | |
1064 | return 1; | |
1065 | } | |
d5396421 | 1066 | |
99c1f5ab | 1067 | for_each_dip(dip, i) { |
e7c9f3ff NS |
1068 | if (start_trace(dip)) { |
1069 | close(dip->fd); | |
1070 | fprintf(stderr, "Failed to start trace on %s\n", | |
1071 | dip->path); | |
1072 | break; | |
1073 | } | |
1074 | } | |
99c1f5ab | 1075 | |
e7c9f3ff | 1076 | if (i != ndevs) { |
99c1f5ab | 1077 | __for_each_dip(dip, j, i) |
e7c9f3ff | 1078 | stop_trace(dip); |
99c1f5ab | 1079 | |
e7c9f3ff NS |
1080 | return 1; |
1081 | } | |
1082 | ||
99c1f5ab | 1083 | for_each_dip(dip, i) { |
e7c9f3ff NS |
1084 | dip->threads = thread_information + (i * ncpus); |
1085 | if (start_threads(dip)) { | |
1086 | fprintf(stderr, "Failed to start worker threads\n"); | |
1087 | break; | |
1088 | } | |
1089 | } | |
99c1f5ab | 1090 | |
e7c9f3ff | 1091 | if (i != ndevs) { |
99c1f5ab | 1092 | __for_each_dip(dip, j, i) |
e7c9f3ff | 1093 | stop_threads(dip); |
99c1f5ab | 1094 | for_each_dip(dip, i) |
e7c9f3ff | 1095 | stop_trace(dip); |
99c1f5ab | 1096 | |
e7c9f3ff | 1097 | return 1; |
d0ca268b JA |
1098 | } |
1099 | ||
e7c9f3ff | 1100 | return 0; |
d0ca268b JA |
1101 | } |
1102 | ||
e7c9f3ff NS |
1103 | static void show_stats(void) |
1104 | { | |
e7c9f3ff NS |
1105 | struct device_information *dip; |
1106 | struct thread_information *tip; | |
b7106311 | 1107 | unsigned long long events_processed, data_read; |
eb3c8108 | 1108 | unsigned long total_drops; |
2f903295 | 1109 | int i, j, no_stdout = 0; |
eb3c8108 JA |
1110 | |
1111 | if (is_stat_shown()) | |
1112 | return; | |
1113 | ||
2f903295 JA |
1114 | if (output_name && !strcmp(output_name, "-")) |
1115 | no_stdout = 1; | |
1116 | ||
eb3c8108 | 1117 | stat_shown = 1; |
428683db | 1118 | |
56070ea4 | 1119 | total_drops = 0; |
99c1f5ab | 1120 | for_each_dip(dip, i) { |
2f903295 | 1121 | if (!no_stdout) |
56070ea4 | 1122 | printf("Device: %s\n", dip->path); |
e7c9f3ff | 1123 | events_processed = 0; |
b7106311 | 1124 | data_read = 0; |
99c1f5ab | 1125 | for_each_tip(dip, tip, j) { |
2f903295 | 1126 | if (!no_stdout) |
b7106311 JA |
1127 | printf(" CPU%3d: %20lu events, %8llu KiB data\n", |
1128 | tip->cpu, tip->events_processed, | |
1129 | tip->data_read >> 10); | |
e7c9f3ff | 1130 | events_processed += tip->events_processed; |
b7106311 | 1131 | data_read += tip->data_read; |
e7c9f3ff | 1132 | } |
eb3c8108 | 1133 | total_drops += dip->drop_count; |
2f903295 | 1134 | if (!no_stdout) |
b7106311 JA |
1135 | printf(" Total: %20llu events (dropped %lu), %8llu KiB data\n", |
1136 | events_processed, dip->drop_count, | |
1137 | data_read >> 10); | |
e7c9f3ff | 1138 | } |
56070ea4 JA |
1139 | |
1140 | if (total_drops) | |
1141 | fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n"); | |
e7c9f3ff | 1142 | } |
52724a0e | 1143 | |
22cd0c02 | 1144 | static struct device_information *net_get_dip(char *buts_name) |
8e86c98a | 1145 | { |
22cd0c02 | 1146 | struct device_information *dip; |
8e86c98a JA |
1147 | int i; |
1148 | ||
22cd0c02 JA |
1149 | for (i = 0; i < ndevs; i++) { |
1150 | dip = &device_information[i]; | |
8e86c98a | 1151 | |
22cd0c02 JA |
1152 | if (!strcmp(dip->buts_name, buts_name)) |
1153 | return dip; | |
8e86c98a JA |
1154 | } |
1155 | ||
22cd0c02 JA |
1156 | device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip)); |
1157 | dip = &device_information[ndevs]; | |
1158 | strcpy(dip->buts_name, buts_name); | |
410d7c62 | 1159 | strcpy(dip->path, buts_name); |
22cd0c02 JA |
1160 | ndevs++; |
1161 | dip->threads = malloc(ncpus * sizeof(struct thread_information)); | |
1162 | memset(dip->threads, 0, ncpus * sizeof(struct thread_information)); | |
1163 | ||
1164 | /* | |
1165 | * open all files | |
1166 | */ | |
1167 | for (i = 0; i < ncpus; i++) { | |
1168 | struct thread_information *tip = &dip->threads[i]; | |
1169 | char op[64]; | |
8e86c98a | 1170 | |
22cd0c02 JA |
1171 | tip->cpu = i; |
1172 | tip->ofile_stdout = 0; | |
1173 | tip->ofile_mmap = 1; | |
1174 | tip->device = dip; | |
8e86c98a | 1175 | |
22cd0c02 | 1176 | fill_ofname(op, dip->buts_name, tip->cpu); |
8e86c98a | 1177 | |
22cd0c02 JA |
1178 | tip->ofile = fopen(op, "w+"); |
1179 | if (!tip->ofile) { | |
1180 | perror("fopen"); | |
1181 | return NULL; | |
1182 | } | |
8e86c98a JA |
1183 | } |
1184 | ||
22cd0c02 JA |
1185 | return dip; |
1186 | } | |
1187 | ||
1188 | static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh) | |
1189 | { | |
1190 | struct device_information *dip; | |
1191 | ||
1192 | ncpus = bnh->max_cpus; | |
1193 | dip = net_get_dip(bnh->buts_name); | |
1194 | return &dip->threads[bnh->cpu]; | |
8e86c98a JA |
1195 | } |
1196 | ||
1197 | static int net_get_header(struct blktrace_net_hdr *bnh) | |
1198 | { | |
1199 | int fl = fcntl(net_in_fd, F_GETFL); | |
1200 | int bytes_left, ret; | |
1201 | void *p = bnh; | |
1202 | ||
1203 | fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK); | |
1204 | bytes_left = sizeof(*bnh); | |
1205 | while (bytes_left && !is_done()) { | |
1206 | ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL); | |
1207 | if (ret < 0) { | |
1208 | if (errno != EAGAIN) { | |
1209 | perror("recv header"); | |
1210 | return 1; | |
1211 | } | |
1212 | usleep(100); | |
1213 | continue; | |
1214 | } else if (!ret) { | |
1215 | usleep(100); | |
1216 | continue; | |
1217 | } else { | |
1218 | p += ret; | |
1219 | bytes_left -= ret; | |
1220 | } | |
1221 | } | |
1222 | fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK); | |
1223 | return 0; | |
1224 | } | |
1225 | ||
1226 | static int net_server_loop(void) | |
1227 | { | |
1228 | struct thread_information *tip; | |
1229 | struct blktrace_net_hdr bnh; | |
1230 | ||
1231 | if (net_get_header(&bnh)) | |
1232 | return 1; | |
1233 | ||
1234 | if (data_is_native == -1 && check_data_endianness(bnh.magic)) { | |
1235 | fprintf(stderr, "server: received data is bad\n"); | |
1236 | return 1; | |
1237 | } | |
1238 | ||
1239 | if (!data_is_native) { | |
1240 | bnh.cpu = be32_to_cpu(bnh.cpu); | |
1241 | bnh.len = be32_to_cpu(bnh.len); | |
1242 | } | |
1243 | ||
6a752c90 JA |
1244 | /* |
1245 | * len == 0 means that the other end signalled end-of-run | |
1246 | */ | |
1247 | if (!bnh.len) { | |
1248 | fprintf(stderr, "server: end of run\n"); | |
1249 | return 1; | |
1250 | } | |
1251 | ||
8e86c98a JA |
1252 | tip = net_get_tip(&bnh); |
1253 | if (!tip) | |
1254 | return 1; | |
1255 | ||
1256 | if (mmap_subbuf(tip, bnh.len)) | |
1257 | return 1; | |
1258 | ||
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Start here when we are in server mode - just fetch data from the network | |
1264 | * and dump to files | |
1265 | */ | |
1266 | static int net_server(void) | |
1267 | { | |
1268 | struct sockaddr_in addr; | |
1269 | socklen_t socklen; | |
22cd0c02 | 1270 | int fd, opt, i, j; |
8e86c98a JA |
1271 | |
1272 | fd = socket(AF_INET, SOCK_STREAM, 0); | |
1273 | if (fd < 0) { | |
1274 | perror("server: socket"); | |
1275 | return 1; | |
1276 | } | |
1277 | ||
1278 | opt = 1; | |
1279 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) { | |
1280 | perror("setsockopt"); | |
1281 | return 1; | |
1282 | } | |
1283 | ||
1284 | memset(&addr, 0, sizeof(addr)); | |
1285 | addr.sin_family = AF_INET; | |
1286 | addr.sin_addr.s_addr = htonl(INADDR_ANY); | |
1287 | addr.sin_port = htons(net_port); | |
1288 | ||
1289 | if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) { | |
1290 | perror("bind"); | |
1291 | return 1; | |
1292 | } | |
1293 | ||
1294 | if (listen(fd, 1) < 0) { | |
1295 | perror("listen"); | |
1296 | return 1; | |
1297 | } | |
1298 | ||
6a752c90 | 1299 | repeat: |
8e86c98a JA |
1300 | printf("blktrace: waiting for incoming connection...\n"); |
1301 | ||
1302 | socklen = sizeof(addr); | |
1303 | net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen); | |
1304 | if (net_in_fd < 0) { | |
1305 | perror("accept"); | |
1306 | return 1; | |
1307 | } | |
1308 | ||
1309 | signal(SIGINT, handle_sigint); | |
1310 | signal(SIGHUP, handle_sigint); | |
1311 | signal(SIGTERM, handle_sigint); | |
1312 | signal(SIGALRM, handle_sigint); | |
1313 | ||
1314 | printf("blktrace: connected!\n"); | |
1315 | ||
1316 | while (!is_done()) { | |
1317 | if (net_server_loop()) | |
1318 | break; | |
1319 | } | |
1320 | ||
22cd0c02 JA |
1321 | for (i = 0; i < ndevs; i++) { |
1322 | struct device_information *dip = &device_information[i]; | |
1323 | ||
1324 | for (j = 0; j < ncpus; j++) | |
1325 | tip_ftrunc_final(&dip->threads[j]); | |
1326 | } | |
8e86c98a | 1327 | |
410d7c62 | 1328 | show_stats(); |
6a752c90 JA |
1329 | |
1330 | if (is_done()) | |
1331 | return 0; | |
1332 | ||
1333 | goto repeat; | |
8e86c98a JA |
1334 | } |
1335 | ||
1336 | /* | |
1337 | * Setup outgoing network connection where we will transmit data | |
1338 | */ | |
1339 | static int net_setup_client(void) | |
1340 | { | |
1341 | struct sockaddr_in addr; | |
1342 | int fd; | |
1343 | ||
1344 | fd = socket(AF_INET, SOCK_STREAM, 0); | |
1345 | if (fd < 0) { | |
1346 | perror("client: socket"); | |
1347 | return 1; | |
1348 | } | |
1349 | ||
1350 | memset(&addr, 0, sizeof(addr)); | |
1351 | addr.sin_family = AF_INET; | |
1352 | addr.sin_port = htons(net_port); | |
1353 | ||
1354 | if (inet_aton(hostname, &addr.sin_addr) != 1) { | |
1355 | struct hostent *hent = gethostbyname(hostname); | |
1356 | if (!hent) { | |
1357 | perror("gethostbyname"); | |
1358 | return 1; | |
1359 | } | |
1360 | ||
1361 | memcpy(&addr.sin_addr, hent->h_addr, 4); | |
1362 | strcpy(hostname, hent->h_name); | |
1363 | } | |
1364 | ||
1365 | printf("blktrace: connecting to %s\n", hostname); | |
1366 | ||
1367 | if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) { | |
1368 | perror("client: connect"); | |
1369 | return 1; | |
1370 | } | |
1371 | ||
1372 | printf("blktrace: connected!\n"); | |
1373 | net_out_fd = fd; | |
1374 | return 0; | |
1375 | } | |
1376 | ||
52724a0e JA |
1377 | static char usage_str[] = \ |
1378 | "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \ | |
1379 | "[ -a action ] [ -A action mask ] [ -v ]\n\n" \ | |
1380 | "\t-d Use specified device. May also be given last after options\n" \ | |
1381 | "\t-r Path to mounted relayfs, defaults to /relay\n" \ | |
1382 | "\t-o File(s) to send output to\n" \ | |
d1d7f15f | 1383 | "\t-D Directory to prepend to output file names\n" \ |
52724a0e JA |
1384 | "\t-k Kill a running trace\n" \ |
1385 | "\t-w Stop after defined time, in seconds\n" \ | |
1386 | "\t-a Only trace specified actions. See documentation\n" \ | |
1387 | "\t-A Give trace mask as a single value. See documentation\n" \ | |
129aa440 JA |
1388 | "\t-b Sub buffer size in KiB\n" \ |
1389 | "\t-n Number of sub buffers\n" \ | |
f531b94d JA |
1390 | "\t-l Run in network listen mode (blktrace server)\n" \ |
1391 | "\t-h Run in network client mode, connecting to the given host\n" \ | |
1392 | "\t-p Network port to use (default 8462)\n" \ | |
1393 | "\t-s Make the network client use sendfile() to transfer data\n" \ | |
1394 | "\t-V Print program version info\n\n"; | |
52724a0e | 1395 | |
ee1f4158 NS |
1396 | static void show_usage(char *program) |
1397 | { | |
52724a0e | 1398 | fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str); |
ee1f4158 | 1399 | } |
d0ca268b JA |
1400 | |
1401 | int main(int argc, char *argv[]) | |
1402 | { | |
5270dddd | 1403 | static char default_relay_path[] = "/relay"; |
e3e74029 | 1404 | struct statfs st; |
d39c04ca | 1405 | int i, c; |
ece238a6 | 1406 | int stop_watch = 0; |
d39c04ca AB |
1407 | int act_mask_tmp = 0; |
1408 | ||
1409 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) { | |
1410 | switch (c) { | |
1411 | case 'a': | |
1412 | i = find_mask_map(optarg); | |
1413 | if (i < 0) { | |
ab197ca7 | 1414 | fprintf(stderr,"Invalid action mask %s\n", |
d39c04ca | 1415 | optarg); |
7425d456 | 1416 | return 1; |
d39c04ca AB |
1417 | } |
1418 | act_mask_tmp |= i; | |
1419 | break; | |
1420 | ||
1421 | case 'A': | |
98f8386b AB |
1422 | if ((sscanf(optarg, "%x", &i) != 1) || |
1423 | !valid_act_opt(i)) { | |
d39c04ca | 1424 | fprintf(stderr, |
ab197ca7 | 1425 | "Invalid set action mask %s/0x%x\n", |
d39c04ca | 1426 | optarg, i); |
7425d456 | 1427 | return 1; |
d39c04ca AB |
1428 | } |
1429 | act_mask_tmp = i; | |
1430 | break; | |
d0ca268b | 1431 | |
d39c04ca | 1432 | case 'd': |
e7c9f3ff NS |
1433 | if (resize_devices(optarg) != 0) |
1434 | return 1; | |
d39c04ca AB |
1435 | break; |
1436 | ||
5270dddd JA |
1437 | case 'r': |
1438 | relay_path = optarg; | |
1439 | break; | |
1440 | ||
d5396421 | 1441 | case 'o': |
66efebf8 | 1442 | output_name = optarg; |
d5396421 | 1443 | break; |
bc39777c JA |
1444 | case 'k': |
1445 | kill_running_trace = 1; | |
1446 | break; | |
ece238a6 NS |
1447 | case 'w': |
1448 | stop_watch = atoi(optarg); | |
1449 | if (stop_watch <= 0) { | |
1450 | fprintf(stderr, | |
1451 | "Invalid stopwatch value (%d secs)\n", | |
1452 | stop_watch); | |
1453 | return 1; | |
1454 | } | |
1455 | break; | |
57ea8602 | 1456 | case 'V': |
52724a0e JA |
1457 | printf("%s version %s\n", argv[0], blktrace_version); |
1458 | return 0; | |
129aa440 | 1459 | case 'b': |
eb3c8108 | 1460 | buf_size = strtoul(optarg, NULL, 10); |
183a0855 | 1461 | if (buf_size <= 0 || buf_size > 16*1024) { |
129aa440 | 1462 | fprintf(stderr, |
eb3c8108 | 1463 | "Invalid buffer size (%lu)\n",buf_size); |
129aa440 JA |
1464 | return 1; |
1465 | } | |
1466 | buf_size <<= 10; | |
1467 | break; | |
1468 | case 'n': | |
eb3c8108 | 1469 | buf_nr = strtoul(optarg, NULL, 10); |
129aa440 JA |
1470 | if (buf_nr <= 0) { |
1471 | fprintf(stderr, | |
eb3c8108 | 1472 | "Invalid buffer nr (%lu)\n", buf_nr); |
129aa440 JA |
1473 | return 1; |
1474 | } | |
1475 | break; | |
d1d7f15f JA |
1476 | case 'D': |
1477 | output_dir = optarg; | |
1478 | break; | |
8e86c98a JA |
1479 | case 'h': |
1480 | net_mode = Net_client; | |
1481 | strcpy(hostname, optarg); | |
1482 | break; | |
1483 | case 'l': | |
1484 | net_mode = Net_server; | |
1485 | break; | |
1486 | case 'p': | |
1487 | net_port = atoi(optarg); | |
1488 | break; | |
32f18c48 JA |
1489 | case 's': |
1490 | net_sendfile = 1; | |
1491 | break; | |
d39c04ca | 1492 | default: |
ee1f4158 | 1493 | show_usage(argv[0]); |
7425d456 | 1494 | return 1; |
d39c04ca AB |
1495 | } |
1496 | } | |
1497 | ||
8e86c98a JA |
1498 | setlocale(LC_NUMERIC, "en_US"); |
1499 | ||
1500 | page_size = getpagesize(); | |
1501 | ||
1502 | if (net_mode == Net_server) | |
1503 | return net_server(); | |
1504 | ||
22cd0c02 JA |
1505 | while (optind < argc) { |
1506 | if (resize_devices(argv[optind++]) != 0) | |
1507 | return 1; | |
1508 | } | |
1509 | ||
e7c9f3ff | 1510 | if (ndevs == 0) { |
ee1f4158 | 1511 | show_usage(argv[0]); |
7425d456 | 1512 | return 1; |
d39c04ca AB |
1513 | } |
1514 | ||
5270dddd JA |
1515 | if (!relay_path) |
1516 | relay_path = default_relay_path; | |
1517 | ||
d5396421 | 1518 | if (act_mask_tmp != 0) |
d39c04ca | 1519 | act_mask = act_mask_tmp; |
d0ca268b | 1520 | |
e3e74029 NS |
1521 | if (statfs(relay_path, &st) < 0) { |
1522 | perror("statfs"); | |
1523 | fprintf(stderr,"%s does not appear to be a valid path\n", | |
1524 | relay_path); | |
1525 | return 1; | |
64acacae | 1526 | } else if (st.f_type != (long) RELAYFS_TYPE) { |
e3e74029 | 1527 | fprintf(stderr,"%s does not appear to be a relay filesystem\n", |
d0ca268b | 1528 | relay_path); |
7425d456 | 1529 | return 1; |
d0ca268b JA |
1530 | } |
1531 | ||
e7c9f3ff | 1532 | if (open_devices() != 0) |
7425d456 | 1533 | return 1; |
bc39777c JA |
1534 | |
1535 | if (kill_running_trace) { | |
e7c9f3ff | 1536 | stop_all_traces(); |
7425d456 | 1537 | return 0; |
bc39777c JA |
1538 | } |
1539 | ||
e7c9f3ff NS |
1540 | ncpus = sysconf(_SC_NPROCESSORS_ONLN); |
1541 | if (ncpus < 0) { | |
1542 | fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n"); | |
7425d456 | 1543 | return 1; |
d0ca268b JA |
1544 | } |
1545 | ||
d0ca268b JA |
1546 | signal(SIGINT, handle_sigint); |
1547 | signal(SIGHUP, handle_sigint); | |
1548 | signal(SIGTERM, handle_sigint); | |
ece238a6 | 1549 | signal(SIGALRM, handle_sigint); |
d0ca268b | 1550 | |
8e86c98a JA |
1551 | if (net_mode == Net_client && net_setup_client()) |
1552 | return 1; | |
1553 | ||
1554 | if (start_devices() != 0) | |
1555 | return 1; | |
1556 | ||
e7c9f3ff | 1557 | atexit(stop_all_tracing); |
830fd65c | 1558 | |
ece238a6 NS |
1559 | if (stop_watch) |
1560 | alarm(stop_watch); | |
1561 | ||
b7106311 | 1562 | wait_for_threads(); |
d0ca268b | 1563 | |
eb3c8108 JA |
1564 | if (!is_trace_stopped()) { |
1565 | trace_stopped = 1; | |
91816d54 JA |
1566 | stop_all_threads(); |
1567 | stop_all_traces(); | |
91816d54 | 1568 | } |
d0ca268b | 1569 | |
eb3c8108 JA |
1570 | show_stats(); |
1571 | ||
d0ca268b JA |
1572 | return 0; |
1573 | } | |
1574 |