Commit | Line | Data |
---|---|---|
d0ca268b JA |
1 | /* |
2 | * block queue tracing application | |
3 | * | |
d956a2cd | 4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> |
46e37c55 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
d956a2cd | 6 | * |
3fe0b570 AB |
7 | * Rewrite to have a single thread per CPU (managing all devices on that CPU) |
8 | * Alan D. Brunelle <alan.brunelle@hp.com> - January 2009 | |
9 | * | |
d956a2cd JA |
10 | * This program is free software; you can redistribute it and/or modify |
11 | * it under the terms of the GNU General Public License as published by | |
12 | * the Free Software Foundation; either version 2 of the License, or | |
13 | * (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | * GNU General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
23 | * | |
d0ca268b | 24 | */ |
3fe0b570 AB |
25 | |
26 | #include <errno.h> | |
27 | #include <stdarg.h> | |
28 | #include <stdio.h> | |
29 | #include <stdlib.h> | |
30 | #include <string.h> | |
31 | #include <fcntl.h> | |
32 | #include <getopt.h> | |
33 | #include <sched.h> | |
d0ca268b | 34 | #include <unistd.h> |
3fe0b570 | 35 | #include <poll.h> |
d0ca268b | 36 | #include <signal.h> |
3fe0b570 AB |
37 | #include <pthread.h> |
38 | #include <locale.h> | |
d0ca268b | 39 | #include <sys/ioctl.h> |
3fe0b570 AB |
40 | #include <sys/types.h> |
41 | #include <sys/stat.h> | |
42 | #include <sys/vfs.h> | |
b7106311 | 43 | #include <sys/mman.h> |
3fe0b570 AB |
44 | #include <sys/param.h> |
45 | #include <sys/time.h> | |
46 | #include <sys/resource.h> | |
8e86c98a | 47 | #include <sys/socket.h> |
8e86c98a JA |
48 | #include <netinet/in.h> |
49 | #include <arpa/inet.h> | |
50 | #include <netdb.h> | |
32f18c48 | 51 | #include <sys/sendfile.h> |
d0ca268b | 52 | |
3fe0b570 | 53 | #include "btt/list.h" |
d0ca268b | 54 | #include "blktrace.h" |
52724a0e | 55 | |
8f551a39 JA |
56 | /* |
57 | * You may want to increase this even more, if you are logging at a high | |
58 | * rate and see skipped/missed events | |
59 | */ | |
3fe0b570 AB |
60 | #define BUF_SIZE (512 * 1024) |
61 | #define BUF_NR (4) | |
62 | ||
63 | #define FILE_VBUF_SIZE (128 * 1024) | |
64 | ||
65 | #define DEBUGFS_TYPE (0x64626720) | |
66 | #define TRACE_NET_PORT (8462) | |
67 | ||
68 | enum { | |
69 | Net_none = 0, | |
70 | Net_server, | |
71 | Net_client, | |
72 | }; | |
73 | ||
df81fdb5 AB |
74 | enum thread_status { |
75 | Th_running, | |
76 | Th_leaving, | |
77 | Th_error | |
78 | }; | |
79 | ||
3fe0b570 AB |
80 | /* |
81 | * Generic stats collected: nevents can be _roughly_ estimated by data_read | |
82 | * (discounting pdu...) | |
83 | * | |
84 | * These fields are updated w/ pdc_dr_update & pdc_nev_update below. | |
85 | */ | |
86 | struct pdc_stats { | |
87 | unsigned long long data_read; | |
88 | unsigned long long nevents; | |
89 | }; | |
90 | ||
91 | struct devpath { | |
92 | struct list_head head; | |
93 | char *path; /* path to device special file */ | |
94 | char *buts_name; /* name returned from bt kernel code */ | |
95 | struct pdc_stats *stats; | |
70598a36 | 96 | int fd, ncpus; |
3fe0b570 AB |
97 | unsigned long long drops; |
98 | ||
99 | /* | |
100 | * For piped output only: | |
101 | * | |
102 | * Each tracer will have a tracer_devpath_head that it will add new | |
103 | * data onto. It's list is protected above (tracer_devpath_head.mutex) | |
104 | * and it will signal the processing thread using the dp_cond, | |
105 | * dp_mutex & dp_entries variables above. | |
106 | */ | |
107 | struct tracer_devpath_head *heads; | |
108 | ||
109 | /* | |
110 | * For network server mode only: | |
111 | */ | |
112 | struct cl_host *ch; | |
113 | u32 cl_id; | |
114 | time_t cl_connect_time; | |
115 | struct io_info *ios; | |
116 | }; | |
117 | ||
118 | /* | |
119 | * For piped output to stdout we will have each tracer thread (one per dev) | |
120 | * tack buffers read from the relay queues on a per-device list. | |
121 | * | |
122 | * The main thread will then collect trace buffers from each of lists in turn. | |
123 | * | |
124 | * We will use a mutex to guard each of the trace_buf list. The tracers | |
125 | * can then signal the main thread using <dp_cond,dp_mutex> and | |
126 | * dp_entries. (When dp_entries is 0, and a tracer adds an entry it will | |
127 | * signal. When dp_entries is 0, the main thread will wait for that condition | |
128 | * to be signalled.) | |
129 | * | |
130 | * adb: It may be better just to have a large buffer per tracer per dev, | |
131 | * and then use it as a ring-buffer. This would certainly cut down a lot | |
132 | * of malloc/free thrashing, at the cost of more memory movements (potentially). | |
133 | */ | |
134 | struct trace_buf { | |
135 | struct list_head head; | |
136 | struct devpath *dpp; | |
137 | void *buf; | |
138 | int cpu, len; | |
139 | }; | |
140 | ||
141 | struct tracer_devpath_head { | |
142 | pthread_mutex_t mutex; | |
143 | struct list_head head; | |
144 | struct trace_buf *prev; | |
145 | }; | |
146 | ||
147 | /* | |
148 | * Used to handle the mmap() interfaces for output file (containing traces) | |
149 | */ | |
150 | struct mmap_info { | |
151 | void *fs_buf; | |
152 | unsigned long long fs_size, fs_max_size, fs_off, fs_buf_len; | |
153 | unsigned long buf_size, buf_nr; | |
154 | int pagesize; | |
155 | }; | |
156 | ||
157 | /* | |
158 | * Each thread doing work on a (client) side of blktrace will have one | |
159 | * of these. The ios array contains input/output information, pfds holds | |
160 | * poll() data. The volatile's provide flags to/from the main executing | |
161 | * thread. | |
162 | */ | |
163 | struct tracer { | |
164 | struct list_head head; | |
165 | struct io_info *ios; | |
166 | struct pollfd *pfds; | |
167 | pthread_t thread; | |
3fe0b570 | 168 | int cpu, nios; |
df81fdb5 | 169 | volatile int status, is_done; |
3fe0b570 AB |
170 | }; |
171 | ||
172 | /* | |
173 | * networking stuff follows. we include a magic number so we know whether | |
174 | * to endianness convert or not. | |
175 | * | |
176 | * The len field is overloaded: | |
177 | * 0 - Indicates an "open" - allowing the server to set up for a dev/cpu | |
178 | * 1 - Indicates a "close" - Shut down connection orderly | |
179 | * | |
180 | * The cpu field is overloaded on close: it will contain the number of drops. | |
181 | */ | |
182 | struct blktrace_net_hdr { | |
183 | u32 magic; /* same as trace magic */ | |
184 | char buts_name[32]; /* trace name */ | |
185 | u32 cpu; /* for which cpu */ | |
186 | u32 max_cpus; | |
187 | u32 len; /* length of following trace data */ | |
188 | u32 cl_id; /* id for set of client per-cpu connections */ | |
189 | u32 buf_size; /* client buf_size for this trace */ | |
190 | u32 buf_nr; /* client buf_nr for this trace */ | |
191 | u32 page_size; /* client page_size for this trace */ | |
192 | }; | |
193 | ||
194 | /* | |
195 | * Each host encountered has one of these. The head is used to link this | |
196 | * on to the network server's ch_list. Connections associated with this | |
197 | * host are linked on conn_list, and any devices traced on that host | |
198 | * are connected on the devpaths list. | |
199 | */ | |
200 | struct cl_host { | |
201 | struct list_head head; | |
202 | struct list_head conn_list; | |
203 | struct list_head devpaths; | |
204 | struct net_server_s *ns; | |
205 | char *hostname; | |
206 | struct in_addr cl_in_addr; | |
207 | int connects, ndevs, cl_opens; | |
208 | }; | |
209 | ||
210 | /* | |
211 | * Each connection (client to server socket ('fd')) has one of these. A | |
212 | * back reference to the host ('ch'), and lists headers (for the host | |
213 | * list, and the network server conn_list) are also included. | |
214 | */ | |
215 | struct cl_conn { | |
216 | struct list_head ch_head, ns_head; | |
217 | struct cl_host *ch; | |
218 | int fd, ncpus; | |
219 | time_t connect_time; | |
220 | }; | |
221 | ||
222 | /* | |
223 | * The network server requires some poll structures to be maintained - | |
224 | * one per conection currently on conn_list. The nchs/ch_list values | |
225 | * are for each host connected to this server. The addr field is used | |
226 | * for scratch as new connections are established. | |
227 | */ | |
228 | struct net_server_s { | |
229 | struct list_head conn_list; | |
230 | struct list_head ch_list; | |
231 | struct pollfd *pfds; | |
232 | int listen_fd, connects, nchs; | |
233 | struct sockaddr_in addr; | |
234 | }; | |
235 | ||
236 | /* | |
237 | * This structure is (generically) used to providide information | |
238 | * for a read-to-write set of values. | |
239 | * | |
240 | * ifn & ifd represent input information | |
241 | * | |
242 | * ofn, ofd, ofp, obuf & mmap_info are used for output file (optionally). | |
243 | */ | |
244 | struct io_info { | |
245 | struct devpath *dpp; | |
246 | FILE *ofp; | |
247 | char *obuf; | |
248 | struct cl_conn *nc; /* Server network connection */ | |
249 | ||
250 | /* | |
251 | * mmap controlled output files | |
252 | */ | |
253 | struct mmap_info mmap_info; | |
254 | ||
255 | /* | |
256 | * Client network fields | |
257 | */ | |
258 | unsigned int ready; | |
259 | unsigned long long data_queued; | |
260 | ||
261 | /* | |
262 | * Input/output file descriptors & names | |
263 | */ | |
264 | int ifd, ofd; | |
265 | char ifn[MAXPATHLEN + 64]; | |
266 | char ofn[MAXPATHLEN + 64]; | |
267 | }; | |
268 | ||
269 | static char blktrace_version[] = "2.0.0"; | |
270 | ||
271 | /* | |
272 | * Linkage to blktrace helper routines (trace conversions) | |
273 | */ | |
274 | int data_is_native = -1; | |
275 | ||
055cc3e5 | 276 | static int ndevs; |
d045a704 | 277 | static int max_cpus; |
3fe0b570 | 278 | static int ncpus; |
d045a704 | 279 | static cpu_set_t *online_cpus; |
3fe0b570 AB |
280 | static int pagesize; |
281 | static int act_mask = ~0U; | |
055cc3e5 AB |
282 | static int kill_running_trace; |
283 | static int stop_watch; | |
284 | static int piped_output; | |
285 | ||
3fe0b570 AB |
286 | static char *debugfs_path = "/sys/kernel/debug"; |
287 | static char *output_name; | |
288 | static char *output_dir; | |
055cc3e5 | 289 | |
3fe0b570 AB |
290 | static unsigned long buf_size = BUF_SIZE; |
291 | static unsigned long buf_nr = BUF_NR; | |
055cc3e5 AB |
292 | |
293 | static FILE *pfp; | |
294 | ||
3fe0b570 AB |
295 | static LIST_HEAD(devpaths); |
296 | static LIST_HEAD(tracers); | |
055cc3e5 | 297 | |
3fe0b570 | 298 | static volatile int done; |
d0ca268b | 299 | |
6488ca48 AB |
300 | /* |
301 | * tracer threads add entries, the main thread takes them off and processes | |
302 | * them. These protect the dp_entries variable. | |
303 | */ | |
3fe0b570 AB |
304 | static pthread_cond_t dp_cond = PTHREAD_COND_INITIALIZER; |
305 | static pthread_mutex_t dp_mutex = PTHREAD_MUTEX_INITIALIZER; | |
306 | static volatile int dp_entries; | |
307 | ||
6488ca48 | 308 | /* |
df81fdb5 | 309 | * These synchronize master / thread interactions. |
6488ca48 | 310 | */ |
df81fdb5 AB |
311 | static pthread_cond_t mt_cond = PTHREAD_COND_INITIALIZER; |
312 | static pthread_mutex_t mt_mutex = PTHREAD_MUTEX_INITIALIZER; | |
313 | static volatile int nthreads_running; | |
314 | static volatile int nthreads_leaving; | |
315 | static volatile int nthreads_error; | |
316 | static volatile int tracers_run; | |
6488ca48 | 317 | |
3fe0b570 AB |
318 | /* |
319 | * network cmd line params | |
320 | */ | |
e58f3937 | 321 | static struct sockaddr_in hostname_addr; |
3fe0b570 AB |
322 | static char hostname[MAXHOSTNAMELEN]; |
323 | static int net_port = TRACE_NET_PORT; | |
324 | static int net_use_sendfile = 1; | |
325 | static int net_mode; | |
326 | static int *cl_fds; | |
007c233c | 327 | |
3fe0b570 AB |
328 | static int (*handle_pfds)(struct tracer *, int, int); |
329 | static int (*handle_list)(struct tracer_devpath_head *, struct list_head *); | |
e3e74029 | 330 | |
5d4f19d9 | 331 | #define S_OPTS "d:a:A:r:o:kw:vVb:n:D:lh:p:sI:" |
d5396421 | 332 | static struct option l_opts[] = { |
5c86134e | 333 | { |
d39c04ca | 334 | .name = "dev", |
428683db | 335 | .has_arg = required_argument, |
d39c04ca AB |
336 | .flag = NULL, |
337 | .val = 'd' | |
338 | }, | |
cf1edb17 AB |
339 | { |
340 | .name = "input-devs", | |
341 | .has_arg = required_argument, | |
342 | .flag = NULL, | |
343 | .val = 'I' | |
344 | }, | |
5c86134e | 345 | { |
d39c04ca | 346 | .name = "act-mask", |
428683db | 347 | .has_arg = required_argument, |
d39c04ca AB |
348 | .flag = NULL, |
349 | .val = 'a' | |
350 | }, | |
5c86134e | 351 | { |
d39c04ca | 352 | .name = "set-mask", |
428683db | 353 | .has_arg = required_argument, |
d39c04ca AB |
354 | .flag = NULL, |
355 | .val = 'A' | |
356 | }, | |
5c86134e | 357 | { |
5270dddd | 358 | .name = "relay", |
428683db | 359 | .has_arg = required_argument, |
5270dddd JA |
360 | .flag = NULL, |
361 | .val = 'r' | |
362 | }, | |
d5396421 JA |
363 | { |
364 | .name = "output", | |
428683db | 365 | .has_arg = required_argument, |
d5396421 JA |
366 | .flag = NULL, |
367 | .val = 'o' | |
368 | }, | |
bc39777c JA |
369 | { |
370 | .name = "kill", | |
428683db | 371 | .has_arg = no_argument, |
bc39777c JA |
372 | .flag = NULL, |
373 | .val = 'k' | |
374 | }, | |
ece238a6 NS |
375 | { |
376 | .name = "stopwatch", | |
428683db | 377 | .has_arg = required_argument, |
ece238a6 NS |
378 | .flag = NULL, |
379 | .val = 'w' | |
380 | }, | |
5d4f19d9 JA |
381 | { |
382 | .name = "version", | |
383 | .has_arg = no_argument, | |
384 | .flag = NULL, | |
385 | .val = 'v' | |
386 | }, | |
52724a0e JA |
387 | { |
388 | .name = "version", | |
389 | .has_arg = no_argument, | |
390 | .flag = NULL, | |
57ea8602 | 391 | .val = 'V' |
52724a0e | 392 | }, |
129aa440 | 393 | { |
3f65c585 | 394 | .name = "buffer-size", |
129aa440 JA |
395 | .has_arg = required_argument, |
396 | .flag = NULL, | |
397 | .val = 'b' | |
398 | }, | |
399 | { | |
3f65c585 | 400 | .name = "num-sub-buffers", |
129aa440 JA |
401 | .has_arg = required_argument, |
402 | .flag = NULL, | |
403 | .val = 'n' | |
404 | }, | |
d1d7f15f | 405 | { |
3f65c585 | 406 | .name = "output-dir", |
d1d7f15f JA |
407 | .has_arg = required_argument, |
408 | .flag = NULL, | |
409 | .val = 'D' | |
410 | }, | |
8e86c98a JA |
411 | { |
412 | .name = "listen", | |
413 | .has_arg = no_argument, | |
414 | .flag = NULL, | |
415 | .val = 'l' | |
416 | }, | |
417 | { | |
418 | .name = "host", | |
419 | .has_arg = required_argument, | |
420 | .flag = NULL, | |
421 | .val = 'h' | |
422 | }, | |
423 | { | |
424 | .name = "port", | |
425 | .has_arg = required_argument, | |
426 | .flag = NULL, | |
427 | .val = 'p' | |
428 | }, | |
32f18c48 | 429 | { |
79971f43 | 430 | .name = "no-sendfile", |
32f18c48 JA |
431 | .has_arg = no_argument, |
432 | .flag = NULL, | |
433 | .val = 's' | |
434 | }, | |
71ef8b7c JA |
435 | { |
436 | .name = NULL, | |
437 | } | |
d39c04ca AB |
438 | }; |
439 | ||
b9a7e9fc ES |
440 | static char usage_str[] = "\n\n" \ |
441 | "-d <dev> | --dev=<dev>\n" \ | |
442 | "[ -r <debugfs path> | --relay=<debugfs path> ]\n" \ | |
443 | "[ -o <file> | --output=<file>]\n" \ | |
444 | "[ -D <dir> | --output-dir=<dir>\n" \ | |
445 | "[ -w <time> | --stopwatch=<time>]\n" \ | |
446 | "[ -a <action field> | --act-mask=<action field>]\n" \ | |
447 | "[ -A <action mask> | --set-mask=<action mask>]\n" \ | |
448 | "[ -b <size> | --buffer-size]\n" \ | |
449 | "[ -n <number> | --num-sub-buffers=<number>]\n" \ | |
450 | "[ -l | --listen]\n" \ | |
451 | "[ -h <hostname> | --host=<hostname>]\n" \ | |
452 | "[ -p <port number> | --port=<port number>]\n" \ | |
453 | "[ -s | --no-sendfile]\n" \ | |
454 | "[ -I <devs file> | --input-devs=<devs file>]\n" \ | |
455 | "[ -v <version> | --version]\n" \ | |
456 | "[ -V <version> | --version]\n" \ | |
457 | ||
3fe0b570 AB |
458 | "\t-d Use specified device. May also be given last after options\n" \ |
459 | "\t-r Path to mounted debugfs, defaults to /sys/kernel/debug\n" \ | |
460 | "\t-o File(s) to send output to\n" \ | |
461 | "\t-D Directory to prepend to output file names\n" \ | |
3fe0b570 AB |
462 | "\t-w Stop after defined time, in seconds\n" \ |
463 | "\t-a Only trace specified actions. See documentation\n" \ | |
464 | "\t-A Give trace mask as a single value. See documentation\n" \ | |
4fad2b83 JT |
465 | "\t-b Sub buffer size in KiB (default 512)\n" \ |
466 | "\t-n Number of sub buffers (default 4)\n" \ | |
3fe0b570 AB |
467 | "\t-l Run in network listen mode (blktrace server)\n" \ |
468 | "\t-h Run in network client mode, connecting to the given host\n" \ | |
469 | "\t-p Network port to use (default 8462)\n" \ | |
470 | "\t-s Make the network client NOT use sendfile() to transfer data\n" \ | |
471 | "\t-I Add devices found in <devs file>\n" \ | |
b9a7e9fc | 472 | "\t-v Print program version info\n" \ |
3fe0b570 | 473 | "\t-V Print program version info\n\n"; |
9db17354 | 474 | |
3fe0b570 AB |
475 | static void clear_events(struct pollfd *pfd) |
476 | { | |
477 | pfd->events = 0; | |
478 | pfd->revents = 0; | |
479 | } | |
21f55651 | 480 | |
3fe0b570 AB |
481 | static inline int net_client_use_sendfile(void) |
482 | { | |
483 | return net_mode == Net_client && net_use_sendfile; | |
484 | } | |
21f55651 | 485 | |
3fe0b570 AB |
486 | static inline int net_client_use_send(void) |
487 | { | |
488 | return net_mode == Net_client && !net_use_sendfile; | |
489 | } | |
b9d4294e | 490 | |
3fe0b570 AB |
491 | static inline int use_tracer_devpaths(void) |
492 | { | |
493 | return piped_output || net_client_use_send(); | |
494 | } | |
b9d4294e | 495 | |
3fe0b570 AB |
496 | static inline int in_addr_eq(struct in_addr a, struct in_addr b) |
497 | { | |
498 | return a.s_addr == b.s_addr; | |
499 | } | |
007c233c | 500 | |
3fe0b570 AB |
501 | static inline void pdc_dr_update(struct devpath *dpp, int cpu, int data_read) |
502 | { | |
503 | dpp->stats[cpu].data_read += data_read; | |
504 | } | |
0cc7d25e | 505 | |
3fe0b570 AB |
506 | static inline void pdc_nev_update(struct devpath *dpp, int cpu, int nevents) |
507 | { | |
508 | dpp->stats[cpu].nevents += nevents; | |
509 | } | |
9db17354 | 510 | |
3fe0b570 AB |
511 | static void show_usage(char *prog) |
512 | { | |
bc14c53f | 513 | fprintf(stderr, "Usage: %s %s", prog, usage_str); |
3fe0b570 | 514 | } |
9db17354 | 515 | |
df81fdb5 AB |
516 | /* |
517 | * Create a timespec 'msec' milliseconds into the future | |
518 | */ | |
519 | static inline void make_timespec(struct timespec *tsp, long delta_msec) | |
520 | { | |
521 | struct timeval now; | |
522 | ||
523 | gettimeofday(&now, NULL); | |
524 | tsp->tv_sec = now.tv_sec; | |
525 | tsp->tv_nsec = 1000L * now.tv_usec; | |
526 | ||
527 | tsp->tv_nsec += (delta_msec * 1000000L); | |
528 | if (tsp->tv_nsec > 1000000000L) { | |
529 | long secs = tsp->tv_nsec / 1000000000L; | |
530 | ||
531 | tsp->tv_sec += secs; | |
532 | tsp->tv_nsec -= (secs * 1000000000L); | |
533 | } | |
534 | } | |
535 | ||
536 | /* | |
537 | * Add a timer to ensure wait ends | |
538 | */ | |
539 | static void t_pthread_cond_wait(pthread_cond_t *cond, pthread_mutex_t *mutex) | |
540 | { | |
541 | struct timespec ts; | |
542 | ||
543 | make_timespec(&ts, 50); | |
544 | pthread_cond_timedwait(cond, mutex, &ts); | |
545 | } | |
546 | ||
547 | static void unblock_tracers(void) | |
548 | { | |
549 | pthread_mutex_lock(&mt_mutex); | |
550 | tracers_run = 1; | |
551 | pthread_cond_broadcast(&mt_cond); | |
552 | pthread_mutex_unlock(&mt_mutex); | |
553 | } | |
554 | ||
555 | static void tracer_wait_unblock(struct tracer *tp) | |
556 | { | |
557 | pthread_mutex_lock(&mt_mutex); | |
558 | while (!tp->is_done && !tracers_run) | |
559 | pthread_cond_wait(&mt_cond, &mt_mutex); | |
560 | pthread_mutex_unlock(&mt_mutex); | |
561 | } | |
562 | ||
563 | static void tracer_signal_ready(struct tracer *tp, | |
564 | enum thread_status th_status, | |
565 | int status) | |
566 | { | |
567 | pthread_mutex_lock(&mt_mutex); | |
568 | tp->status = status; | |
569 | ||
570 | if (th_status == Th_running) | |
571 | nthreads_running++; | |
572 | else if (th_status == Th_error) | |
573 | nthreads_error++; | |
574 | else | |
575 | nthreads_leaving++; | |
576 | ||
577 | pthread_cond_signal(&mt_cond); | |
578 | pthread_mutex_unlock(&mt_mutex); | |
579 | } | |
580 | ||
581 | static void wait_tracers_ready(int ncpus_started) | |
582 | { | |
583 | pthread_mutex_lock(&mt_mutex); | |
584 | while ((nthreads_running + nthreads_error) < ncpus_started) | |
585 | t_pthread_cond_wait(&mt_cond, &mt_mutex); | |
586 | pthread_mutex_unlock(&mt_mutex); | |
587 | } | |
588 | ||
589 | static void wait_tracers_leaving(void) | |
590 | { | |
591 | pthread_mutex_lock(&mt_mutex); | |
592 | while (nthreads_leaving < nthreads_running) | |
593 | t_pthread_cond_wait(&mt_cond, &mt_mutex); | |
594 | pthread_mutex_unlock(&mt_mutex); | |
595 | } | |
596 | ||
3fe0b570 AB |
597 | static void init_mmap_info(struct mmap_info *mip) |
598 | { | |
599 | mip->buf_size = buf_size; | |
600 | mip->buf_nr = buf_nr; | |
601 | mip->pagesize = pagesize; | |
602 | } | |
b7106311 | 603 | |
3fe0b570 AB |
604 | static void net_close_connection(int *fd) |
605 | { | |
606 | shutdown(*fd, SHUT_RDWR); | |
607 | close(*fd); | |
608 | *fd = -1; | |
609 | } | |
ff11d54c | 610 | |
3fe0b570 AB |
611 | static void dpp_free(struct devpath *dpp) |
612 | { | |
613 | if (dpp->stats) | |
614 | free(dpp->stats); | |
615 | if (dpp->ios) | |
616 | free(dpp->ios); | |
617 | if (dpp->path) | |
618 | free(dpp->path); | |
619 | if (dpp->buts_name) | |
620 | free(dpp->buts_name); | |
621 | free(dpp); | |
622 | } | |
d0ca268b | 623 | |
3fe0b570 AB |
624 | static int lock_on_cpu(int cpu) |
625 | { | |
0a915aab NZ |
626 | cpu_set_t * cpu_mask; |
627 | size_t size; | |
d045a704 JK |
628 | |
629 | cpu_mask = CPU_ALLOC(max_cpus); | |
630 | size = CPU_ALLOC_SIZE(max_cpus); | |
0a915aab NZ |
631 | |
632 | CPU_ZERO_S(size, cpu_mask); | |
633 | CPU_SET_S(cpu, size, cpu_mask); | |
634 | if (sched_setaffinity(0, size, cpu_mask) < 0) { | |
635 | CPU_FREE(cpu_mask); | |
3fe0b570 | 636 | return errno; |
0a915aab | 637 | } |
d0ca268b | 638 | |
0a915aab | 639 | CPU_FREE(cpu_mask); |
3fe0b570 AB |
640 | return 0; |
641 | } | |
e7c9f3ff | 642 | |
3fe0b570 AB |
643 | static int increase_limit(int resource, rlim_t increase) |
644 | { | |
645 | struct rlimit rlim; | |
646 | int save_errno = errno; | |
8e86c98a | 647 | |
3fe0b570 AB |
648 | if (!getrlimit(resource, &rlim)) { |
649 | rlim.rlim_cur += increase; | |
650 | if (rlim.rlim_cur >= rlim.rlim_max) | |
651 | rlim.rlim_max = rlim.rlim_cur + increase; | |
72ca8801 | 652 | |
3fe0b570 AB |
653 | if (!setrlimit(resource, &rlim)) |
654 | return 1; | |
655 | } | |
99c1f5ab | 656 | |
3fe0b570 AB |
657 | errno = save_errno; |
658 | return 0; | |
659 | } | |
e0a1988b | 660 | |
3fe0b570 AB |
661 | static int handle_open_failure(void) |
662 | { | |
663 | if (errno == ENFILE || errno == EMFILE) | |
664 | return increase_limit(RLIMIT_NOFILE, 16); | |
665 | return 0; | |
666 | } | |
99c1f5ab | 667 | |
3fe0b570 AB |
668 | static int handle_mem_failure(size_t length) |
669 | { | |
670 | if (errno == ENFILE) | |
671 | return handle_open_failure(); | |
672 | else if (errno == ENOMEM) | |
673 | return increase_limit(RLIMIT_MEMLOCK, 2 * length); | |
674 | return 0; | |
675 | } | |
99c1f5ab | 676 | |
3fe0b570 AB |
677 | static FILE *my_fopen(const char *path, const char *mode) |
678 | { | |
679 | FILE *fp; | |
8e86c98a | 680 | |
3fe0b570 AB |
681 | do { |
682 | fp = fopen(path, mode); | |
683 | } while (fp == NULL && handle_open_failure()); | |
8e86c98a | 684 | |
3fe0b570 AB |
685 | return fp; |
686 | } | |
8e86c98a | 687 | |
3fe0b570 AB |
688 | static int my_open(const char *path, int flags) |
689 | { | |
690 | int fd; | |
8e86c98a | 691 | |
3fe0b570 AB |
692 | do { |
693 | fd = open(path, flags); | |
694 | } while (fd < 0 && handle_open_failure()); | |
e0a1988b | 695 | |
3fe0b570 AB |
696 | return fd; |
697 | } | |
ff11d54c | 698 | |
3fe0b570 AB |
699 | static int my_socket(int domain, int type, int protocol) |
700 | { | |
701 | int fd; | |
ff11d54c | 702 | |
3fe0b570 AB |
703 | do { |
704 | fd = socket(domain, type, protocol); | |
705 | } while (fd < 0 && handle_open_failure()); | |
8e86c98a | 706 | |
3fe0b570 AB |
707 | return fd; |
708 | } | |
709 | ||
d5302b03 AB |
710 | static int my_accept(int sockfd, struct sockaddr *addr, socklen_t *addrlen) |
711 | { | |
712 | int fd; | |
713 | ||
714 | do { | |
715 | fd = accept(sockfd, addr, addrlen); | |
716 | } while (fd < 0 && handle_open_failure()); | |
717 | ||
718 | return fd; | |
719 | } | |
720 | ||
3fe0b570 AB |
721 | static void *my_mmap(void *addr, size_t length, int prot, int flags, int fd, |
722 | off_t offset) | |
69dd57c2 | 723 | { |
3fe0b570 | 724 | void *new; |
69dd57c2 | 725 | |
3fe0b570 AB |
726 | do { |
727 | new = mmap(addr, length, prot, flags, fd, offset); | |
728 | } while (new == MAP_FAILED && handle_mem_failure(length)); | |
729 | ||
730 | return new; | |
731 | } | |
732 | ||
ae7c049d TM |
733 | static int my_mlock(struct tracer *tp, |
734 | const void *addr, size_t len) | |
3fe0b570 | 735 | { |
ae7c049d | 736 | int ret, retry = 0; |
3fe0b570 AB |
737 | |
738 | do { | |
739 | ret = mlock(addr, len); | |
ae7c049d TM |
740 | if ((retry >= 10) && tp && tp->is_done) |
741 | break; | |
742 | retry++; | |
3fe0b570 AB |
743 | } while (ret < 0 && handle_mem_failure(len)); |
744 | ||
745 | return ret; | |
746 | } | |
747 | ||
ae7c049d TM |
748 | static int setup_mmap(int fd, unsigned int maxlen, |
749 | struct mmap_info *mip, | |
750 | struct tracer *tp) | |
055cc3e5 AB |
751 | { |
752 | if (mip->fs_off + maxlen > mip->fs_buf_len) { | |
753 | unsigned long nr = max(16, mip->buf_nr); | |
754 | ||
755 | if (mip->fs_buf) { | |
756 | munlock(mip->fs_buf, mip->fs_buf_len); | |
757 | munmap(mip->fs_buf, mip->fs_buf_len); | |
758 | mip->fs_buf = NULL; | |
759 | } | |
760 | ||
761 | mip->fs_off = mip->fs_size & (mip->pagesize - 1); | |
762 | mip->fs_buf_len = (nr * mip->buf_size) - mip->fs_off; | |
763 | mip->fs_max_size += mip->fs_buf_len; | |
764 | ||
765 | if (ftruncate(fd, mip->fs_max_size) < 0) { | |
766 | perror("setup_mmap: ftruncate"); | |
767 | return 1; | |
768 | } | |
769 | ||
770 | mip->fs_buf = my_mmap(NULL, mip->fs_buf_len, PROT_WRITE, | |
771 | MAP_SHARED, fd, | |
772 | mip->fs_size - mip->fs_off); | |
773 | if (mip->fs_buf == MAP_FAILED) { | |
774 | perror("setup_mmap: mmap"); | |
775 | return 1; | |
776 | } | |
ae7c049d TM |
777 | if (my_mlock(tp, mip->fs_buf, mip->fs_buf_len) < 0) { |
778 | perror("setup_mlock: mlock"); | |
779 | return 1; | |
780 | } | |
055cc3e5 AB |
781 | } |
782 | ||
783 | return 0; | |
784 | } | |
785 | ||
3fe0b570 AB |
786 | static int __stop_trace(int fd) |
787 | { | |
788 | /* | |
789 | * Should be stopped, don't complain if it isn't | |
790 | */ | |
791 | ioctl(fd, BLKTRACESTOP); | |
792 | return ioctl(fd, BLKTRACETEARDOWN); | |
793 | } | |
794 | ||
795 | static int write_data(char *buf, int len) | |
796 | { | |
797 | int ret; | |
798 | ||
799 | rewrite: | |
800 | ret = fwrite(buf, len, 1, pfp); | |
801 | if (ferror(pfp) || ret != 1) { | |
802 | if (errno == EINTR) { | |
803 | clearerr(pfp); | |
804 | goto rewrite; | |
805 | } | |
806 | ||
807 | if (!piped_output || (errno != EPIPE && errno != EBADF)) { | |
808 | fprintf(stderr, "write(%d) failed: %d/%s\n", | |
809 | len, errno, strerror(errno)); | |
69dd57c2 | 810 | } |
3fe0b570 | 811 | goto err; |
69dd57c2 AB |
812 | } |
813 | ||
3fe0b570 | 814 | fflush(pfp); |
69dd57c2 | 815 | return 0; |
3fe0b570 AB |
816 | |
817 | err: | |
818 | clearerr(pfp); | |
819 | return 1; | |
69dd57c2 AB |
820 | } |
821 | ||
822 | /* | |
3fe0b570 | 823 | * Returns the number of bytes read (successfully) |
69dd57c2 | 824 | */ |
3fe0b570 | 825 | static int __net_recv_data(int fd, void *buf, unsigned int len) |
69dd57c2 | 826 | { |
3fe0b570 AB |
827 | unsigned int bytes_left = len; |
828 | ||
829 | while (bytes_left && !done) { | |
830 | int ret = recv(fd, buf, bytes_left, MSG_WAITALL); | |
831 | ||
832 | if (ret == 0) | |
833 | break; | |
834 | else if (ret < 0) { | |
5d65b5e6 AB |
835 | if (errno == EAGAIN) { |
836 | usleep(50); | |
837 | continue; | |
838 | } | |
839 | perror("server: net_recv_data: recv failed"); | |
055cc3e5 | 840 | break; |
3fe0b570 AB |
841 | } else { |
842 | buf += ret; | |
843 | bytes_left -= ret; | |
844 | } | |
845 | } | |
69dd57c2 | 846 | |
3fe0b570 | 847 | return len - bytes_left; |
69dd57c2 AB |
848 | } |
849 | ||
3fe0b570 | 850 | static int net_recv_data(int fd, void *buf, unsigned int len) |
8e86c98a | 851 | { |
3fe0b570 AB |
852 | return __net_recv_data(fd, buf, len); |
853 | } | |
7035d92d | 854 | |
3fe0b570 AB |
855 | /* |
856 | * Returns number of bytes written | |
857 | */ | |
858 | static int net_send_data(int fd, void *buf, unsigned int buf_len) | |
859 | { | |
860 | int ret; | |
861 | unsigned int bytes_left = buf_len; | |
862 | ||
863 | while (bytes_left) { | |
864 | ret = send(fd, buf, bytes_left, 0); | |
865 | if (ret < 0) { | |
866 | perror("send"); | |
867 | break; | |
868 | } | |
869 | ||
870 | buf += ret; | |
871 | bytes_left -= ret; | |
7035d92d JA |
872 | } |
873 | ||
3fe0b570 | 874 | return buf_len - bytes_left; |
8e86c98a JA |
875 | } |
876 | ||
3fe0b570 | 877 | static int net_send_header(int fd, int cpu, char *buts_name, int len) |
eb3c8108 | 878 | { |
3fe0b570 | 879 | struct blktrace_net_hdr hdr; |
eb3c8108 | 880 | |
3fe0b570 | 881 | memset(&hdr, 0, sizeof(hdr)); |
eb3c8108 | 882 | |
3fe0b570 | 883 | hdr.magic = BLK_IO_TRACE_MAGIC; |
d324757e | 884 | memset(hdr.buts_name, 0, sizeof(hdr.buts_name)); |
3fe0b570 | 885 | strncpy(hdr.buts_name, buts_name, sizeof(hdr.buts_name)); |
d324757e | 886 | hdr.buts_name[sizeof(hdr.buts_name) - 1] = '\0'; |
3fe0b570 | 887 | hdr.cpu = cpu; |
d045a704 | 888 | hdr.max_cpus = max_cpus; |
3fe0b570 AB |
889 | hdr.len = len; |
890 | hdr.cl_id = getpid(); | |
891 | hdr.buf_size = buf_size; | |
892 | hdr.buf_nr = buf_nr; | |
893 | hdr.page_size = pagesize; | |
eb3c8108 | 894 | |
3fe0b570 AB |
895 | return net_send_data(fd, &hdr, sizeof(hdr)) != sizeof(hdr); |
896 | } | |
eb3c8108 | 897 | |
3fe0b570 AB |
898 | static void net_send_open_close(int fd, int cpu, char *buts_name, int len) |
899 | { | |
900 | struct blktrace_net_hdr ret_hdr; | |
eb3c8108 | 901 | |
3fe0b570 AB |
902 | net_send_header(fd, cpu, buts_name, len); |
903 | net_recv_data(fd, &ret_hdr, sizeof(ret_hdr)); | |
904 | } | |
eb3c8108 | 905 | |
3fe0b570 AB |
906 | static void net_send_open(int fd, int cpu, char *buts_name) |
907 | { | |
908 | net_send_open_close(fd, cpu, buts_name, 0); | |
eb3c8108 JA |
909 | } |
910 | ||
3fe0b570 | 911 | static void net_send_close(int fd, char *buts_name, int drops) |
d0ca268b | 912 | { |
3fe0b570 AB |
913 | /* |
914 | * Overload CPU w/ number of drops | |
915 | * | |
916 | * XXX: Need to clear/set done around call - done=1 (which | |
917 | * is true here) stops reads from happening... :-( | |
918 | */ | |
919 | done = 0; | |
920 | net_send_open_close(fd, drops, buts_name, 1); | |
921 | done = 1; | |
922 | } | |
d0ca268b | 923 | |
3fe0b570 AB |
924 | static void ack_open_close(int fd, char *buts_name) |
925 | { | |
926 | net_send_header(fd, 0, buts_name, 2); | |
927 | } | |
d0ca268b | 928 | |
3fe0b570 AB |
929 | static void net_send_drops(int fd) |
930 | { | |
931 | struct list_head *p; | |
ed71a31e | 932 | |
3fe0b570 AB |
933 | __list_for_each(p, &devpaths) { |
934 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
935 | ||
936 | net_send_close(fd, dpp->buts_name, dpp->drops); | |
d0ca268b | 937 | } |
3fe0b570 | 938 | } |
d0ca268b | 939 | |
3fe0b570 AB |
940 | /* |
941 | * Returns: | |
055cc3e5 AB |
942 | * 0: "EOF" |
943 | * 1: OK | |
944 | * -1: Error | |
3fe0b570 AB |
945 | */ |
946 | static int net_get_header(struct cl_conn *nc, struct blktrace_net_hdr *bnh) | |
947 | { | |
948 | int bytes_read; | |
949 | int fl = fcntl(nc->fd, F_GETFL); | |
950 | ||
951 | fcntl(nc->fd, F_SETFL, fl | O_NONBLOCK); | |
952 | bytes_read = __net_recv_data(nc->fd, bnh, sizeof(*bnh)); | |
953 | fcntl(nc->fd, F_SETFL, fl & ~O_NONBLOCK); | |
954 | ||
955 | if (bytes_read == sizeof(*bnh)) | |
956 | return 1; | |
957 | else if (bytes_read == 0) | |
958 | return 0; | |
055cc3e5 AB |
959 | else |
960 | return -1; | |
d0ca268b JA |
961 | } |
962 | ||
e58f3937 | 963 | static int net_setup_addr(void) |
d0ca268b | 964 | { |
e58f3937 | 965 | struct sockaddr_in *addr = &hostname_addr; |
cf9208ea | 966 | |
e58f3937 AB |
967 | memset(addr, 0, sizeof(*addr)); |
968 | addr->sin_family = AF_INET; | |
969 | addr->sin_port = htons(net_port); | |
3fe0b570 | 970 | |
e58f3937 AB |
971 | if (inet_aton(hostname, &addr->sin_addr) != 1) { |
972 | struct hostent *hent; | |
973 | retry: | |
974 | hent = gethostbyname(hostname); | |
3fe0b570 | 975 | if (!hent) { |
e58f3937 AB |
976 | if (h_errno == TRY_AGAIN) { |
977 | usleep(100); | |
978 | goto retry; | |
979 | } else if (h_errno == NO_RECOVERY) { | |
980 | fprintf(stderr, "gethostbyname(%s)" | |
981 | "non-recoverable error encountered\n", | |
982 | hostname); | |
983 | } else { | |
984 | /* | |
985 | * HOST_NOT_FOUND, NO_ADDRESS or NO_DATA | |
986 | */ | |
987 | fprintf(stderr, "Host %s not found\n", | |
988 | hostname); | |
989 | } | |
3fe0b570 AB |
990 | return 1; |
991 | } | |
992 | ||
e58f3937 | 993 | memcpy(&addr->sin_addr, hent->h_addr, 4); |
d324757e ES |
994 | memset(hostname, 0, sizeof(hostname)); |
995 | strncpy(hostname, hent->h_name, sizeof(hostname)); | |
996 | hostname[sizeof(hostname) - 1] = '\0'; | |
3fe0b570 | 997 | } |
7035d92d | 998 | |
e58f3937 AB |
999 | return 0; |
1000 | } | |
1001 | ||
1002 | static int net_setup_client(void) | |
1003 | { | |
1004 | int fd; | |
1005 | struct sockaddr_in *addr = &hostname_addr; | |
1006 | ||
3fe0b570 AB |
1007 | fd = my_socket(AF_INET, SOCK_STREAM, 0); |
1008 | if (fd < 0) { | |
1009 | perror("client: socket"); | |
1010 | return -1; | |
1011 | } | |
cf9208ea | 1012 | |
e58f3937 | 1013 | if (connect(fd, (struct sockaddr *)addr, sizeof(*addr)) < 0) { |
3fe0b570 AB |
1014 | if (errno == ECONNREFUSED) |
1015 | fprintf(stderr, | |
1016 | "\nclient: Connection to %s refused, " | |
1017 | "perhaps the server is not started?\n\n", | |
1018 | hostname); | |
1019 | else | |
1020 | perror("client: connect"); | |
055cc3e5 | 1021 | |
3fe0b570 AB |
1022 | close(fd); |
1023 | return -1; | |
707b0914 | 1024 | } |
3fe0b570 AB |
1025 | |
1026 | return fd; | |
d0ca268b JA |
1027 | } |
1028 | ||
3fe0b570 | 1029 | static int open_client_connections(void) |
e7c9f3ff | 1030 | { |
3fe0b570 | 1031 | int cpu; |
d045a704 | 1032 | size_t alloc_size = CPU_ALLOC_SIZE(max_cpus); |
e7c9f3ff | 1033 | |
3fe0b570 | 1034 | cl_fds = calloc(ncpus, sizeof(*cl_fds)); |
d045a704 JK |
1035 | for (cpu = 0; cpu < max_cpus; cpu++) { |
1036 | if (!CPU_ISSET_S(cpu, alloc_size, online_cpus)) | |
1037 | continue; | |
3fe0b570 AB |
1038 | cl_fds[cpu] = net_setup_client(); |
1039 | if (cl_fds[cpu] < 0) | |
1040 | goto err; | |
eb3c8108 | 1041 | } |
3fe0b570 AB |
1042 | return 0; |
1043 | ||
1044 | err: | |
1045 | while (cpu > 0) | |
1046 | close(cl_fds[cpu--]); | |
1047 | free(cl_fds); | |
1048 | return 1; | |
e7c9f3ff NS |
1049 | } |
1050 | ||
3fe0b570 | 1051 | static void close_client_connections(void) |
eb3c8108 | 1052 | { |
3fe0b570 AB |
1053 | if (cl_fds) { |
1054 | int cpu, *fdp; | |
d045a704 | 1055 | size_t alloc_size = CPU_ALLOC_SIZE(max_cpus); |
eb3c8108 | 1056 | |
d045a704 JK |
1057 | for (cpu = 0, fdp = cl_fds; cpu < max_cpus; cpu++, fdp++) { |
1058 | if (!CPU_ISSET_S(cpu, alloc_size, online_cpus)) | |
1059 | continue; | |
3fe0b570 AB |
1060 | if (*fdp >= 0) { |
1061 | net_send_drops(*fdp); | |
1062 | net_close_connection(fdp); | |
1063 | } | |
7934e668 | 1064 | } |
3fe0b570 | 1065 | free(cl_fds); |
ff11d54c | 1066 | } |
eb3c8108 JA |
1067 | } |
1068 | ||
ab6809de | 1069 | static int setup_buts(void) |
d0ca268b | 1070 | { |
3fe0b570 | 1071 | struct list_head *p; |
ab6809de | 1072 | int ret = 0; |
bbabf03a | 1073 | |
3fe0b570 AB |
1074 | __list_for_each(p, &devpaths) { |
1075 | struct blk_user_trace_setup buts; | |
1076 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
ae9f71b3 | 1077 | |
3fe0b570 AB |
1078 | memset(&buts, 0, sizeof(buts)); |
1079 | buts.buf_size = buf_size; | |
1080 | buts.buf_nr = buf_nr; | |
1081 | buts.act_mask = act_mask; | |
1082 | ||
055cc3e5 | 1083 | if (ioctl(dpp->fd, BLKTRACESETUP, &buts) >= 0) { |
d045a704 | 1084 | dpp->ncpus = max_cpus; |
3b552a2d AB |
1085 | dpp->buts_name = strdup(buts.name); |
1086 | if (dpp->stats) | |
1087 | free(dpp->stats); | |
1088 | dpp->stats = calloc(dpp->ncpus, sizeof(*dpp->stats)); | |
1089 | memset(dpp->stats, 0, dpp->ncpus * sizeof(*dpp->stats)); | |
ab6809de | 1090 | } else { |
055cc3e5 AB |
1091 | fprintf(stderr, "BLKTRACESETUP(2) %s failed: %d/%s\n", |
1092 | dpp->path, errno, strerror(errno)); | |
ab6809de JA |
1093 | ret++; |
1094 | } | |
3b552a2d | 1095 | } |
ab6809de JA |
1096 | |
1097 | return ret; | |
3b552a2d AB |
1098 | } |
1099 | ||
1100 | static void start_buts(void) | |
1101 | { | |
1102 | struct list_head *p; | |
1103 | ||
1104 | __list_for_each(p, &devpaths) { | |
1105 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1106 | ||
1107 | if (ioctl(dpp->fd, BLKTRACESTART) < 0) { | |
3fe0b570 AB |
1108 | fprintf(stderr, "BLKTRACESTART %s failed: %d/%s\n", |
1109 | dpp->path, errno, strerror(errno)); | |
bbabf03a | 1110 | } |
3fe0b570 | 1111 | } |
8a43bac5 JA |
1112 | } |
1113 | ||
3fe0b570 | 1114 | static int get_drops(struct devpath *dpp) |
8e86c98a | 1115 | { |
3fe0b570 AB |
1116 | int fd, drops = 0; |
1117 | char fn[MAXPATHLEN + 64], tmp[256]; | |
8e86c98a | 1118 | |
3fe0b570 AB |
1119 | snprintf(fn, sizeof(fn), "%s/block/%s/dropped", debugfs_path, |
1120 | dpp->buts_name); | |
8e86c98a | 1121 | |
3fe0b570 AB |
1122 | fd = my_open(fn, O_RDONLY); |
1123 | if (fd < 0) { | |
1124 | /* | |
1125 | * This may be ok: the kernel may not support | |
1126 | * dropped counts. | |
1127 | */ | |
1128 | if (errno != ENOENT) | |
1129 | fprintf(stderr, "Could not open %s: %d/%s\n", | |
1130 | fn, errno, strerror(errno)); | |
1131 | return 0; | |
1132 | } else if (read(fd, tmp, sizeof(tmp)) < 0) { | |
1133 | fprintf(stderr, "Could not read %s: %d/%s\n", | |
1134 | fn, errno, strerror(errno)); | |
1135 | } else | |
1136 | drops = atoi(tmp); | |
1137 | close(fd); | |
8e86c98a | 1138 | |
3fe0b570 | 1139 | return drops; |
8e86c98a JA |
1140 | } |
1141 | ||
3fe0b570 | 1142 | static void get_all_drops(void) |
a3e4d330 | 1143 | { |
3fe0b570 | 1144 | struct list_head *p; |
21f55651 | 1145 | |
3fe0b570 AB |
1146 | __list_for_each(p, &devpaths) { |
1147 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
055cc3e5 | 1148 | |
3fe0b570 | 1149 | dpp->drops = get_drops(dpp); |
21f55651 | 1150 | } |
9db17354 | 1151 | } |
eb3c8108 | 1152 | |
3fe0b570 | 1153 | static inline struct trace_buf *alloc_trace_buf(int cpu, int bufsize) |
9db17354 | 1154 | { |
3fe0b570 | 1155 | struct trace_buf *tbp; |
21f55651 | 1156 | |
3fe0b570 AB |
1157 | tbp = malloc(sizeof(*tbp) + bufsize); |
1158 | INIT_LIST_HEAD(&tbp->head); | |
1159 | tbp->len = 0; | |
1160 | tbp->buf = (void *)(tbp + 1); | |
1161 | tbp->cpu = cpu; | |
1162 | tbp->dpp = NULL; /* Will be set when tbp is added */ | |
21f55651 | 1163 | |
3fe0b570 | 1164 | return tbp; |
a3e4d330 JA |
1165 | } |
1166 | ||
3fe0b570 | 1167 | static void free_tracer_heads(struct devpath *dpp) |
b7106311 | 1168 | { |
3fe0b570 AB |
1169 | int cpu; |
1170 | struct tracer_devpath_head *hd; | |
b7106311 | 1171 | |
d045a704 | 1172 | for (cpu = 0, hd = dpp->heads; cpu < max_cpus; cpu++, hd++) { |
3fe0b570 AB |
1173 | if (hd->prev) |
1174 | free(hd->prev); | |
055cc3e5 | 1175 | |
3fe0b570 AB |
1176 | pthread_mutex_destroy(&hd->mutex); |
1177 | } | |
1178 | free(dpp->heads); | |
1179 | } | |
b7106311 | 1180 | |
3fe0b570 AB |
1181 | static int setup_tracer_devpaths(void) |
1182 | { | |
1183 | struct list_head *p; | |
b7106311 | 1184 | |
3fe0b570 AB |
1185 | if (net_client_use_send()) |
1186 | if (open_client_connections()) | |
1187 | return 1; | |
b7106311 | 1188 | |
3fe0b570 AB |
1189 | __list_for_each(p, &devpaths) { |
1190 | int cpu; | |
1191 | struct tracer_devpath_head *hd; | |
1192 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
b7106311 | 1193 | |
d045a704 JK |
1194 | dpp->heads = calloc(max_cpus, sizeof(struct tracer_devpath_head)); |
1195 | for (cpu = 0, hd = dpp->heads; cpu < max_cpus; cpu++, hd++) { | |
3fe0b570 AB |
1196 | INIT_LIST_HEAD(&hd->head); |
1197 | pthread_mutex_init(&hd->mutex, NULL); | |
1198 | hd->prev = NULL; | |
1199 | } | |
b7106311 JA |
1200 | } |
1201 | ||
3fe0b570 | 1202 | return 0; |
b7106311 JA |
1203 | } |
1204 | ||
3fe0b570 AB |
1205 | static inline void add_trace_buf(struct devpath *dpp, int cpu, |
1206 | struct trace_buf **tbpp) | |
18eed2a7 | 1207 | { |
3fe0b570 AB |
1208 | struct trace_buf *tbp = *tbpp; |
1209 | struct tracer_devpath_head *hd = &dpp->heads[cpu]; | |
18eed2a7 | 1210 | |
3fe0b570 | 1211 | tbp->dpp = dpp; |
2f064793 | 1212 | |
3fe0b570 AB |
1213 | pthread_mutex_lock(&hd->mutex); |
1214 | list_add_tail(&tbp->head, &hd->head); | |
1215 | pthread_mutex_unlock(&hd->mutex); | |
18eed2a7 | 1216 | |
3fe0b570 | 1217 | *tbpp = alloc_trace_buf(cpu, buf_size); |
18eed2a7 JA |
1218 | } |
1219 | ||
3fe0b570 | 1220 | static inline void incr_entries(int entries_handled) |
a3e4d330 | 1221 | { |
3fe0b570 AB |
1222 | pthread_mutex_lock(&dp_mutex); |
1223 | if (dp_entries == 0) | |
1224 | pthread_cond_signal(&dp_cond); | |
1225 | dp_entries += entries_handled; | |
1226 | pthread_mutex_unlock(&dp_mutex); | |
a3e4d330 JA |
1227 | } |
1228 | ||
055cc3e5 AB |
1229 | static void decr_entries(int handled) |
1230 | { | |
1231 | pthread_mutex_lock(&dp_mutex); | |
1232 | dp_entries -= handled; | |
1233 | pthread_mutex_unlock(&dp_mutex); | |
1234 | } | |
1235 | ||
1236 | static int wait_empty_entries(void) | |
1237 | { | |
1238 | pthread_mutex_lock(&dp_mutex); | |
1239 | while (!done && dp_entries == 0) | |
1240 | t_pthread_cond_wait(&dp_cond, &dp_mutex); | |
1241 | pthread_mutex_unlock(&dp_mutex); | |
1242 | ||
1243 | return !done; | |
1244 | } | |
1245 | ||
3fe0b570 | 1246 | static int add_devpath(char *path) |
8e86c98a | 1247 | { |
3fe0b570 AB |
1248 | int fd; |
1249 | struct devpath *dpp; | |
4b747a40 | 1250 | struct list_head *p; |
3fe0b570 | 1251 | |
4b747a40 ES |
1252 | /* |
1253 | * Verify device is not duplicated | |
1254 | */ | |
1255 | __list_for_each(p, &devpaths) { | |
1256 | struct devpath *tmp = list_entry(p, struct devpath, head); | |
1257 | if (!strcmp(tmp->path, path)) | |
1258 | return 0; | |
1259 | } | |
8e86c98a | 1260 | /* |
3fe0b570 | 1261 | * Verify device is valid before going too far |
8e86c98a | 1262 | */ |
3fe0b570 AB |
1263 | fd = my_open(path, O_RDONLY | O_NONBLOCK); |
1264 | if (fd < 0) { | |
1265 | fprintf(stderr, "Invalid path %s specified: %d/%s\n", | |
1266 | path, errno, strerror(errno)); | |
1267 | return 1; | |
1268 | } | |
8e86c98a | 1269 | |
3fe0b570 AB |
1270 | dpp = malloc(sizeof(*dpp)); |
1271 | memset(dpp, 0, sizeof(*dpp)); | |
1272 | dpp->path = strdup(path); | |
1273 | dpp->fd = fd; | |
70598a36 | 1274 | ndevs++; |
3fe0b570 | 1275 | list_add_tail(&dpp->head, &devpaths); |
8e86c98a | 1276 | |
3fe0b570 | 1277 | return 0; |
8e86c98a JA |
1278 | } |
1279 | ||
3fe0b570 | 1280 | static void rel_devpaths(void) |
a3e4d330 | 1281 | { |
3fe0b570 | 1282 | struct list_head *p, *q; |
a3e4d330 | 1283 | |
3fe0b570 AB |
1284 | list_for_each_safe(p, q, &devpaths) { |
1285 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
a3e4d330 | 1286 | |
3fe0b570 AB |
1287 | list_del(&dpp->head); |
1288 | __stop_trace(dpp->fd); | |
1289 | close(dpp->fd); | |
a3e4d330 | 1290 | |
3fe0b570 AB |
1291 | if (dpp->heads) |
1292 | free_tracer_heads(dpp); | |
a3e4d330 | 1293 | |
3fe0b570 AB |
1294 | dpp_free(dpp); |
1295 | ndevs--; | |
b7106311 | 1296 | } |
8e86c98a | 1297 | } |
b7106311 | 1298 | |
3fe0b570 | 1299 | static int flush_subbuf_net(struct trace_buf *tbp) |
8e86c98a | 1300 | { |
3fe0b570 AB |
1301 | int fd = cl_fds[tbp->cpu]; |
1302 | struct devpath *dpp = tbp->dpp; | |
b7106311 | 1303 | |
3fe0b570 AB |
1304 | if (net_send_header(fd, tbp->cpu, dpp->buts_name, tbp->len)) |
1305 | return 1; | |
055cc3e5 | 1306 | else if (net_send_data(fd, tbp->buf, tbp->len) != tbp->len) |
3fe0b570 | 1307 | return 1; |
a3e4d330 | 1308 | |
8e86c98a | 1309 | return 0; |
a3e4d330 JA |
1310 | } |
1311 | ||
3fe0b570 AB |
1312 | static int |
1313 | handle_list_net(__attribute__((__unused__))struct tracer_devpath_head *hd, | |
1314 | struct list_head *list) | |
8e86c98a | 1315 | { |
3fe0b570 AB |
1316 | struct trace_buf *tbp; |
1317 | struct list_head *p, *q; | |
1318 | int entries_handled = 0; | |
8e86c98a | 1319 | |
3fe0b570 AB |
1320 | list_for_each_safe(p, q, list) { |
1321 | tbp = list_entry(p, struct trace_buf, head); | |
8e86c98a | 1322 | |
3fe0b570 AB |
1323 | list_del(&tbp->head); |
1324 | entries_handled++; | |
6a752c90 | 1325 | |
3fe0b570 AB |
1326 | if (cl_fds[tbp->cpu] >= 0) { |
1327 | if (flush_subbuf_net(tbp)) { | |
1328 | close(cl_fds[tbp->cpu]); | |
1329 | cl_fds[tbp->cpu] = -1; | |
1330 | } | |
1331 | } | |
7ab2f837 | 1332 | |
3fe0b570 | 1333 | free(tbp); |
7934e668 JA |
1334 | } |
1335 | ||
3fe0b570 | 1336 | return entries_handled; |
6a752c90 JA |
1337 | } |
1338 | ||
055cc3e5 AB |
1339 | /* |
1340 | * Tack 'tbp's buf onto the tail of 'prev's buf | |
1341 | */ | |
1342 | static struct trace_buf *tb_combine(struct trace_buf *prev, | |
1343 | struct trace_buf *tbp) | |
1344 | { | |
1345 | unsigned long tot_len; | |
1346 | ||
1347 | tot_len = prev->len + tbp->len; | |
1348 | if (tot_len > buf_size) { | |
1349 | /* | |
1350 | * tbp->head isn't connected (it was 'prev' | |
1351 | * so it had been taken off of the list | |
1352 | * before). Therefore, we can realloc | |
1353 | * the whole structures, as the other fields | |
1354 | * are "static". | |
1355 | */ | |
d8365957 | 1356 | prev = realloc(prev, sizeof(*prev) + tot_len); |
055cc3e5 AB |
1357 | prev->buf = (void *)(prev + 1); |
1358 | } | |
1359 | ||
1360 | memcpy(prev->buf + prev->len, tbp->buf, tbp->len); | |
1361 | prev->len = tot_len; | |
1362 | ||
1363 | free(tbp); | |
1364 | return prev; | |
1365 | } | |
1366 | ||
3fe0b570 AB |
1367 | static int handle_list_file(struct tracer_devpath_head *hd, |
1368 | struct list_head *list) | |
f6fead25 | 1369 | { |
3fe0b570 AB |
1370 | int off, t_len, nevents; |
1371 | struct blk_io_trace *t; | |
1372 | struct list_head *p, *q; | |
1373 | int entries_handled = 0; | |
1374 | struct trace_buf *tbp, *prev; | |
11629347 | 1375 | |
3fe0b570 AB |
1376 | prev = hd->prev; |
1377 | list_for_each_safe(p, q, list) { | |
1378 | tbp = list_entry(p, struct trace_buf, head); | |
1379 | list_del(&tbp->head); | |
1380 | entries_handled++; | |
18eed2a7 | 1381 | |
3fe0b570 AB |
1382 | /* |
1383 | * If there was some leftover before, tack this new | |
1384 | * entry onto the tail of the previous one. | |
1385 | */ | |
055cc3e5 AB |
1386 | if (prev) |
1387 | tbp = tb_combine(prev, tbp); | |
ff11d54c | 1388 | |
3fe0b570 AB |
1389 | /* |
1390 | * See how many whole traces there are - send them | |
1391 | * all out in one go. | |
1392 | */ | |
1393 | off = 0; | |
1394 | nevents = 0; | |
1395 | while (off + (int)sizeof(*t) <= tbp->len) { | |
1396 | t = (struct blk_io_trace *)(tbp->buf + off); | |
1397 | t_len = sizeof(*t) + t->pdu_len; | |
1398 | if (off + t_len > tbp->len) | |
1399 | break; | |
ff11d54c | 1400 | |
3fe0b570 AB |
1401 | off += t_len; |
1402 | nevents++; | |
1403 | } | |
1404 | if (nevents) | |
1405 | pdc_nev_update(tbp->dpp, tbp->cpu, nevents); | |
4aeec019 | 1406 | |
3fe0b570 AB |
1407 | /* |
1408 | * Write any full set of traces, any remaining data is kept | |
1409 | * for the next pass. | |
1410 | */ | |
1411 | if (off) { | |
055cc3e5 | 1412 | if (write_data(tbp->buf, off) || off == tbp->len) { |
3fe0b570 | 1413 | free(tbp); |
055cc3e5 AB |
1414 | prev = NULL; |
1415 | } | |
3fe0b570 AB |
1416 | else { |
1417 | /* | |
1418 | * Move valid data to beginning of buffer | |
1419 | */ | |
1420 | tbp->len -= off; | |
1421 | memmove(tbp->buf, tbp->buf + off, tbp->len); | |
1422 | prev = tbp; | |
1423 | } | |
1424 | } else | |
1425 | prev = tbp; | |
ff11d54c | 1426 | } |
3fe0b570 | 1427 | hd->prev = prev; |
ff11d54c | 1428 | |
3fe0b570 | 1429 | return entries_handled; |
ff11d54c TZ |
1430 | } |
1431 | ||
3fe0b570 | 1432 | static void __process_trace_bufs(void) |
8a43bac5 | 1433 | { |
3fe0b570 AB |
1434 | int cpu; |
1435 | struct list_head *p; | |
1436 | struct list_head list; | |
1437 | int handled = 0; | |
1438 | ||
1439 | __list_for_each(p, &devpaths) { | |
1440 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1441 | struct tracer_devpath_head *hd = dpp->heads; | |
1442 | ||
d045a704 | 1443 | for (cpu = 0; cpu < max_cpus; cpu++, hd++) { |
3fe0b570 AB |
1444 | pthread_mutex_lock(&hd->mutex); |
1445 | if (list_empty(&hd->head)) { | |
1446 | pthread_mutex_unlock(&hd->mutex); | |
1447 | continue; | |
1448 | } | |
8a43bac5 | 1449 | |
3fe0b570 AB |
1450 | list_replace_init(&hd->head, &list); |
1451 | pthread_mutex_unlock(&hd->mutex); | |
6480258a | 1452 | |
3fe0b570 AB |
1453 | handled += handle_list(hd, &list); |
1454 | } | |
d0ca268b JA |
1455 | } |
1456 | ||
055cc3e5 AB |
1457 | if (handled) |
1458 | decr_entries(handled); | |
8a43bac5 JA |
1459 | } |
1460 | ||
3fe0b570 | 1461 | static void process_trace_bufs(void) |
8a43bac5 | 1462 | { |
055cc3e5 | 1463 | while (wait_empty_entries()) |
3fe0b570 | 1464 | __process_trace_bufs(); |
3fe0b570 | 1465 | } |
3a9d6c13 | 1466 | |
3fe0b570 AB |
1467 | static void clean_trace_bufs(void) |
1468 | { | |
3a9d6c13 | 1469 | /* |
3fe0b570 AB |
1470 | * No mutex needed here: we're only reading from the lists, |
1471 | * tracers are done | |
3a9d6c13 | 1472 | */ |
3fe0b570 AB |
1473 | while (dp_entries) |
1474 | __process_trace_bufs(); | |
1475 | } | |
4b5db44a | 1476 | |
3fe0b570 AB |
1477 | static inline void read_err(int cpu, char *ifn) |
1478 | { | |
1479 | if (errno != EAGAIN) | |
1480 | fprintf(stderr, "Thread %d failed read of %s: %d/%s\n", | |
1481 | cpu, ifn, errno, strerror(errno)); | |
4b5db44a JA |
1482 | } |
1483 | ||
3fe0b570 | 1484 | static int net_sendfile(struct io_info *iop) |
d5396421 | 1485 | { |
3fe0b570 | 1486 | int ret; |
d5396421 | 1487 | |
3fe0b570 AB |
1488 | ret = sendfile(iop->ofd, iop->ifd, NULL, iop->ready); |
1489 | if (ret < 0) { | |
1490 | perror("sendfile"); | |
1491 | return 1; | |
1492 | } else if (ret < (int)iop->ready) { | |
1493 | fprintf(stderr, "short sendfile send (%d of %d)\n", | |
1494 | ret, iop->ready); | |
1495 | return 1; | |
1496 | } | |
91816d54 | 1497 | |
9db17354 | 1498 | return 0; |
91816d54 JA |
1499 | } |
1500 | ||
3fe0b570 | 1501 | static inline int net_sendfile_data(struct tracer *tp, struct io_info *iop) |
d0ca268b | 1502 | { |
3fe0b570 | 1503 | struct devpath *dpp = iop->dpp; |
d0ca268b | 1504 | |
3fe0b570 AB |
1505 | if (net_send_header(iop->ofd, tp->cpu, dpp->buts_name, iop->ready)) |
1506 | return 1; | |
1507 | return net_sendfile(iop); | |
1508 | } | |
d0ca268b | 1509 | |
d7a1f726 JK |
1510 | static int fill_ofname(char *dst, int dstlen, char *subdir, char *buts_name, |
1511 | int cpu) | |
8e86c98a | 1512 | { |
3fe0b570 | 1513 | int len; |
e3bf54d8 | 1514 | struct stat sb; |
8e86c98a JA |
1515 | |
1516 | if (output_dir) | |
d7a1f726 | 1517 | len = snprintf(dst, dstlen, "%s/", output_dir); |
dd870ef6 | 1518 | else |
d7a1f726 | 1519 | len = snprintf(dst, dstlen, "./"); |
8e86c98a | 1520 | |
d7a1f726 JK |
1521 | if (subdir) |
1522 | len += snprintf(dst + len, dstlen - len, "%s", subdir); | |
e0a1988b | 1523 | |
d7a1f726 | 1524 | if (stat(dst, &sb) < 0) { |
e3bf54d8 | 1525 | if (errno != ENOENT) { |
3fe0b570 AB |
1526 | fprintf(stderr, |
1527 | "Destination dir %s stat failed: %d/%s\n", | |
d7a1f726 | 1528 | dst, errno, strerror(errno)); |
e3bf54d8 JA |
1529 | return 1; |
1530 | } | |
60886290 JM |
1531 | /* |
1532 | * There is no synchronization between multiple threads | |
1533 | * trying to create the directory at once. It's harmless | |
1534 | * to let them try, so just detect the problem and move on. | |
1535 | */ | |
d7a1f726 | 1536 | if (mkdir(dst, 0755) < 0 && errno != EEXIST) { |
3fe0b570 AB |
1537 | fprintf(stderr, |
1538 | "Destination dir %s can't be made: %d/%s\n", | |
d7a1f726 | 1539 | dst, errno, strerror(errno)); |
e3bf54d8 JA |
1540 | return 1; |
1541 | } | |
1542 | } | |
1543 | ||
8e86c98a | 1544 | if (output_name) |
d7a1f726 | 1545 | snprintf(dst + len, dstlen - len, "%s.blktrace.%d", |
3fe0b570 | 1546 | output_name, cpu); |
8e86c98a | 1547 | else |
d7a1f726 JK |
1548 | snprintf(dst + len, dstlen - len, "%s.blktrace.%d", |
1549 | buts_name, cpu); | |
e3bf54d8 JA |
1550 | |
1551 | return 0; | |
8e86c98a JA |
1552 | } |
1553 | ||
3fe0b570 | 1554 | static int set_vbuf(struct io_info *iop, int mode, size_t size) |
0cc7d25e | 1555 | { |
3fe0b570 AB |
1556 | iop->obuf = malloc(size); |
1557 | if (setvbuf(iop->ofp, iop->obuf, mode, size) < 0) { | |
1558 | fprintf(stderr, "setvbuf(%s, %d) failed: %d/%s\n", | |
1559 | iop->dpp->path, (int)size, errno, | |
1560 | strerror(errno)); | |
1561 | free(iop->obuf); | |
ddf22842 JA |
1562 | return 1; |
1563 | } | |
d5396421 | 1564 | |
ddf22842 JA |
1565 | return 0; |
1566 | } | |
007c233c | 1567 | |
3fe0b570 | 1568 | static int iop_open(struct io_info *iop, int cpu) |
ddf22842 | 1569 | { |
d7a1f726 JK |
1570 | char hostdir[MAXPATHLEN + 64]; |
1571 | ||
3fe0b570 | 1572 | iop->ofd = -1; |
d7a1f726 JK |
1573 | if (net_mode == Net_server) { |
1574 | struct cl_conn *nc = iop->nc; | |
1575 | int len; | |
1576 | ||
1577 | len = snprintf(hostdir, sizeof(hostdir), "%s-", | |
1578 | nc->ch->hostname); | |
1579 | len += strftime(hostdir + len, sizeof(hostdir) - len, "%F-%T/", | |
1580 | gmtime(&iop->dpp->cl_connect_time)); | |
1581 | } else { | |
1582 | hostdir[0] = 0; | |
1583 | } | |
1584 | ||
1585 | if (fill_ofname(iop->ofn, sizeof(iop->ofn), hostdir, | |
1586 | iop->dpp->buts_name, cpu)) | |
3fe0b570 | 1587 | return 1; |
0cc7d25e | 1588 | |
3fe0b570 AB |
1589 | iop->ofp = my_fopen(iop->ofn, "w+"); |
1590 | if (iop->ofp == NULL) { | |
1591 | fprintf(stderr, "Open output file %s failed: %d/%s\n", | |
1592 | iop->ofn, errno, strerror(errno)); | |
1593 | return 1; | |
1594 | } | |
055cc3e5 | 1595 | |
3fe0b570 AB |
1596 | if (set_vbuf(iop, _IOLBF, FILE_VBUF_SIZE)) { |
1597 | fprintf(stderr, "set_vbuf for file %s failed: %d/%s\n", | |
1598 | iop->ofn, errno, strerror(errno)); | |
1599 | fclose(iop->ofp); | |
1600 | return 1; | |
d0ca268b JA |
1601 | } |
1602 | ||
3fe0b570 | 1603 | iop->ofd = fileno(iop->ofp); |
e7c9f3ff | 1604 | return 0; |
d0ca268b JA |
1605 | } |
1606 | ||
df81fdb5 AB |
1607 | static void close_iop(struct io_info *iop) |
1608 | { | |
1609 | struct mmap_info *mip = &iop->mmap_info; | |
1610 | ||
1611 | if (mip->fs_buf) | |
1612 | munmap(mip->fs_buf, mip->fs_buf_len); | |
1613 | ||
1614 | if (!piped_output) { | |
1615 | if (ftruncate(fileno(iop->ofp), mip->fs_size) < 0) { | |
1616 | fprintf(stderr, | |
1617 | "Ignoring err: ftruncate(%s): %d/%s\n", | |
1618 | iop->ofn, errno, strerror(errno)); | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | if (iop->ofp) | |
1623 | fclose(iop->ofp); | |
1624 | if (iop->obuf) | |
1625 | free(iop->obuf); | |
1626 | } | |
1627 | ||
1628 | static void close_ios(struct tracer *tp) | |
1629 | { | |
1630 | while (tp->nios > 0) { | |
1631 | struct io_info *iop = &tp->ios[--tp->nios]; | |
1632 | ||
1633 | iop->dpp->drops = get_drops(iop->dpp); | |
1634 | if (iop->ifd >= 0) | |
1635 | close(iop->ifd); | |
1636 | ||
1637 | if (iop->ofp) | |
1638 | close_iop(iop); | |
1639 | else if (iop->ofd >= 0) { | |
1640 | struct devpath *dpp = iop->dpp; | |
1641 | ||
1642 | net_send_close(iop->ofd, dpp->buts_name, dpp->drops); | |
1643 | net_close_connection(&iop->ofd); | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | free(tp->ios); | |
1648 | free(tp->pfds); | |
1649 | } | |
1650 | ||
3fe0b570 | 1651 | static int open_ios(struct tracer *tp) |
3aabcd89 | 1652 | { |
3fe0b570 AB |
1653 | struct pollfd *pfd; |
1654 | struct io_info *iop; | |
1655 | struct list_head *p; | |
1656 | ||
1657 | tp->ios = calloc(ndevs, sizeof(struct io_info)); | |
3fe0b570 | 1658 | memset(tp->ios, 0, ndevs * sizeof(struct io_info)); |
055cc3e5 AB |
1659 | |
1660 | tp->pfds = calloc(ndevs, sizeof(struct pollfd)); | |
3fe0b570 AB |
1661 | memset(tp->pfds, 0, ndevs * sizeof(struct pollfd)); |
1662 | ||
1663 | tp->nios = 0; | |
1664 | iop = tp->ios; | |
1665 | pfd = tp->pfds; | |
1666 | __list_for_each(p, &devpaths) { | |
1667 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1668 | ||
1669 | iop->dpp = dpp; | |
1670 | iop->ofd = -1; | |
1671 | snprintf(iop->ifn, sizeof(iop->ifn), "%s/block/%s/trace%d", | |
1672 | debugfs_path, dpp->buts_name, tp->cpu); | |
1673 | ||
1674 | iop->ifd = my_open(iop->ifn, O_RDONLY | O_NONBLOCK); | |
1675 | if (iop->ifd < 0) { | |
1676 | fprintf(stderr, "Thread %d failed open %s: %d/%s\n", | |
1677 | tp->cpu, iop->ifn, errno, strerror(errno)); | |
1678 | return 1; | |
1679 | } | |
1680 | ||
1681 | init_mmap_info(&iop->mmap_info); | |
1682 | ||
1683 | pfd->fd = iop->ifd; | |
1684 | pfd->events = POLLIN; | |
1685 | ||
1686 | if (piped_output) | |
1687 | ; | |
1688 | else if (net_client_use_sendfile()) { | |
1689 | iop->ofd = net_setup_client(); | |
1690 | if (iop->ofd < 0) | |
1691 | goto err; | |
1692 | net_send_open(iop->ofd, tp->cpu, dpp->buts_name); | |
1693 | } else if (net_mode == Net_none) { | |
1694 | if (iop_open(iop, tp->cpu)) | |
1695 | goto err; | |
1696 | } else { | |
1697 | /* | |
1698 | * This ensures that the server knows about all | |
1699 | * connections & devices before _any_ closes | |
1700 | */ | |
1701 | net_send_open(cl_fds[tp->cpu], tp->cpu, dpp->buts_name); | |
1702 | } | |
007c233c | 1703 | |
3fe0b570 AB |
1704 | pfd++; |
1705 | iop++; | |
1706 | tp->nios++; | |
9db17354 | 1707 | } |
3aabcd89 | 1708 | |
3fe0b570 | 1709 | return 0; |
72ca8801 | 1710 | |
3fe0b570 AB |
1711 | err: |
1712 | close(iop->ifd); /* tp->nios _not_ bumped */ | |
df81fdb5 | 1713 | close_ios(tp); |
3fe0b570 | 1714 | return 1; |
e7c9f3ff NS |
1715 | } |
1716 | ||
3fe0b570 | 1717 | static int handle_pfds_file(struct tracer *tp, int nevs, int force_read) |
e7c9f3ff | 1718 | { |
3fe0b570 AB |
1719 | struct mmap_info *mip; |
1720 | int i, ret, nentries = 0; | |
1721 | struct pollfd *pfd = tp->pfds; | |
1722 | struct io_info *iop = tp->ios; | |
1723 | ||
1724 | for (i = 0; nevs > 0 && i < ndevs; i++, pfd++, iop++) { | |
1725 | if (pfd->revents & POLLIN || force_read) { | |
1726 | mip = &iop->mmap_info; | |
1727 | ||
ae7c049d | 1728 | ret = setup_mmap(iop->ofd, buf_size, mip, tp); |
3fe0b570 AB |
1729 | if (ret < 0) { |
1730 | pfd->events = 0; | |
1731 | break; | |
1732 | } | |
428683db | 1733 | |
3fe0b570 AB |
1734 | ret = read(iop->ifd, mip->fs_buf + mip->fs_off, |
1735 | buf_size); | |
1736 | if (ret > 0) { | |
1737 | pdc_dr_update(iop->dpp, tp->cpu, ret); | |
1738 | mip->fs_size += ret; | |
1739 | mip->fs_off += ret; | |
1740 | nentries++; | |
1741 | } else if (ret == 0) { | |
1742 | /* | |
1743 | * Short reads after we're done stop us | |
1744 | * from trying reads. | |
1745 | */ | |
1746 | if (tp->is_done) | |
1747 | clear_events(pfd); | |
1748 | } else { | |
1749 | read_err(tp->cpu, iop->ifn); | |
1750 | if (errno != EAGAIN || tp->is_done) | |
1751 | clear_events(pfd); | |
1752 | } | |
1753 | nevs--; | |
e7c9f3ff | 1754 | } |
e7c9f3ff | 1755 | } |
56070ea4 | 1756 | |
3fe0b570 | 1757 | return nentries; |
e7c9f3ff | 1758 | } |
52724a0e | 1759 | |
055cc3e5 AB |
1760 | static int handle_pfds_netclient(struct tracer *tp, int nevs, int force_read) |
1761 | { | |
1762 | struct stat sb; | |
1763 | int i, nentries = 0; | |
055cc3e5 AB |
1764 | struct pollfd *pfd = tp->pfds; |
1765 | struct io_info *iop = tp->ios; | |
1766 | ||
62d712a7 | 1767 | for (i = 0; i < ndevs; i++, pfd++, iop++) { |
055cc3e5 AB |
1768 | if (pfd->revents & POLLIN || force_read) { |
1769 | if (fstat(iop->ifd, &sb) < 0) { | |
1770 | perror(iop->ifn); | |
1771 | pfd->events = 0; | |
1772 | } else if (sb.st_size > (off_t)iop->data_queued) { | |
1773 | iop->ready = sb.st_size - iop->data_queued; | |
1774 | iop->data_queued = sb.st_size; | |
1775 | ||
1776 | if (!net_sendfile_data(tp, iop)) { | |
1777 | pdc_dr_update(iop->dpp, tp->cpu, | |
1778 | iop->ready); | |
1779 | nentries++; | |
1780 | } else | |
1781 | clear_events(pfd); | |
1782 | } | |
1783 | if (--nevs == 0) | |
1784 | break; | |
1785 | } | |
1786 | } | |
1787 | ||
1788 | if (nentries) | |
1789 | incr_entries(nentries); | |
1790 | ||
1791 | return nentries; | |
1792 | } | |
1793 | ||
1794 | static int handle_pfds_entries(struct tracer *tp, int nevs, int force_read) | |
1795 | { | |
1796 | int i, nentries = 0; | |
1797 | struct trace_buf *tbp; | |
1798 | struct pollfd *pfd = tp->pfds; | |
1799 | struct io_info *iop = tp->ios; | |
1800 | ||
1801 | tbp = alloc_trace_buf(tp->cpu, buf_size); | |
1802 | for (i = 0; i < ndevs; i++, pfd++, iop++) { | |
1803 | if (pfd->revents & POLLIN || force_read) { | |
1804 | tbp->len = read(iop->ifd, tbp->buf, buf_size); | |
1805 | if (tbp->len > 0) { | |
1806 | pdc_dr_update(iop->dpp, tp->cpu, tbp->len); | |
1807 | add_trace_buf(iop->dpp, tp->cpu, &tbp); | |
1808 | nentries++; | |
1809 | } else if (tbp->len == 0) { | |
1810 | /* | |
1811 | * Short reads after we're done stop us | |
1812 | * from trying reads. | |
1813 | */ | |
1814 | if (tp->is_done) | |
1815 | clear_events(pfd); | |
1816 | } else { | |
1817 | read_err(tp->cpu, iop->ifn); | |
1818 | if (errno != EAGAIN || tp->is_done) | |
1819 | clear_events(pfd); | |
1820 | } | |
1821 | if (!piped_output && --nevs == 0) | |
1822 | break; | |
1823 | } | |
1824 | } | |
1825 | free(tbp); | |
1826 | ||
1827 | if (nentries) | |
1828 | incr_entries(nentries); | |
1829 | ||
1830 | return nentries; | |
1831 | } | |
1832 | ||
3fe0b570 | 1833 | static void *thread_main(void *arg) |
8e86c98a | 1834 | { |
df81fdb5 | 1835 | int ret, ndone, to_val; |
3fe0b570 | 1836 | struct tracer *tp = arg; |
8e86c98a | 1837 | |
3fe0b570 AB |
1838 | ret = lock_on_cpu(tp->cpu); |
1839 | if (ret) | |
1840 | goto err; | |
ff11d54c | 1841 | |
3fe0b570 | 1842 | ret = open_ios(tp); |
df81fdb5 | 1843 | if (ret) |
3fe0b570 | 1844 | goto err; |
6a6d3f0f | 1845 | |
3fe0b570 AB |
1846 | if (piped_output) |
1847 | to_val = 50; /* Frequent partial handles */ | |
ff11d54c | 1848 | else |
3fe0b570 AB |
1849 | to_val = 500; /* 1/2 second intervals */ |
1850 | ||
df81fdb5 AB |
1851 | |
1852 | tracer_signal_ready(tp, Th_running, 0); | |
1853 | tracer_wait_unblock(tp); | |
6488ca48 | 1854 | |
3fe0b570 AB |
1855 | while (!tp->is_done) { |
1856 | ndone = poll(tp->pfds, ndevs, to_val); | |
1857 | if (ndone || piped_output) | |
1858 | (void)handle_pfds(tp, ndone, piped_output); | |
1859 | else if (ndone < 0 && errno != EINTR) | |
1860 | fprintf(stderr, "Thread %d poll failed: %d/%s\n", | |
1861 | tp->cpu, errno, strerror(errno)); | |
1862 | } | |
22cd0c02 JA |
1863 | |
1864 | /* | |
3fe0b570 | 1865 | * Trace is stopped, pull data until we get a short read |
22cd0c02 | 1866 | */ |
3fe0b570 AB |
1867 | while (handle_pfds(tp, ndevs, 1) > 0) |
1868 | ; | |
055cc3e5 | 1869 | |
3fe0b570 | 1870 | close_ios(tp); |
df81fdb5 AB |
1871 | tracer_signal_ready(tp, Th_leaving, 0); |
1872 | return NULL; | |
8e86c98a | 1873 | |
3fe0b570 | 1874 | err: |
df81fdb5 | 1875 | tracer_signal_ready(tp, Th_error, ret); |
3fe0b570 | 1876 | return NULL; |
22cd0c02 JA |
1877 | } |
1878 | ||
3fe0b570 | 1879 | static int start_tracer(int cpu) |
22cd0c02 | 1880 | { |
3fe0b570 | 1881 | struct tracer *tp; |
22cd0c02 | 1882 | |
3fe0b570 AB |
1883 | tp = malloc(sizeof(*tp)); |
1884 | memset(tp, 0, sizeof(*tp)); | |
7ab2f837 | 1885 | |
3fe0b570 | 1886 | INIT_LIST_HEAD(&tp->head); |
3fe0b570 AB |
1887 | tp->status = 0; |
1888 | tp->cpu = cpu; | |
8e86c98a | 1889 | |
3fe0b570 AB |
1890 | if (pthread_create(&tp->thread, NULL, thread_main, tp)) { |
1891 | fprintf(stderr, "FAILED to start thread on CPU %d: %d/%s\n", | |
1892 | cpu, errno, strerror(errno)); | |
df81fdb5 AB |
1893 | free(tp); |
1894 | return 1; | |
8e86c98a | 1895 | } |
3fe0b570 | 1896 | |
df81fdb5 AB |
1897 | list_add_tail(&tp->head, &tracers); |
1898 | return 0; | |
8e86c98a JA |
1899 | } |
1900 | ||
8772bc4f JK |
1901 | static int create_output_files(int cpu) |
1902 | { | |
1903 | char fname[MAXPATHLEN + 64]; | |
1904 | struct list_head *p; | |
1905 | FILE *f; | |
1906 | ||
1907 | __list_for_each(p, &devpaths) { | |
1908 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1909 | ||
1910 | if (fill_ofname(fname, sizeof(fname), NULL, dpp->buts_name, | |
1911 | cpu)) | |
1912 | return 1; | |
1913 | f = my_fopen(fname, "w+"); | |
1914 | if (!f) | |
1915 | return 1; | |
1916 | fclose(f); | |
1917 | } | |
1918 | return 0; | |
1919 | } | |
1920 | ||
df81fdb5 | 1921 | static void start_tracers(void) |
e0a1988b | 1922 | { |
d045a704 | 1923 | int cpu, started = 0; |
df81fdb5 | 1924 | struct list_head *p; |
d045a704 | 1925 | size_t alloc_size = CPU_ALLOC_SIZE(max_cpus); |
3fe0b570 | 1926 | |
d045a704 | 1927 | for (cpu = 0; cpu < max_cpus; cpu++) { |
8772bc4f JK |
1928 | if (!CPU_ISSET_S(cpu, alloc_size, online_cpus)) { |
1929 | /* | |
1930 | * Create fake empty output files so that other tools | |
1931 | * like blkparse don't have to bother with sparse CPU | |
1932 | * number space. | |
1933 | */ | |
1934 | if (create_output_files(cpu)) | |
1935 | break; | |
d045a704 | 1936 | continue; |
8772bc4f | 1937 | } |
3fe0b570 AB |
1938 | if (start_tracer(cpu)) |
1939 | break; | |
d045a704 JK |
1940 | started++; |
1941 | } | |
e0a1988b | 1942 | |
d045a704 | 1943 | wait_tracers_ready(started); |
df81fdb5 AB |
1944 | |
1945 | __list_for_each(p, &tracers) { | |
1946 | struct tracer *tp = list_entry(p, struct tracer, head); | |
1947 | if (tp->status) | |
1948 | fprintf(stderr, | |
1949 | "FAILED to start thread on CPU %d: %d/%s\n", | |
1950 | tp->cpu, tp->status, strerror(tp->status)); | |
1951 | } | |
3fe0b570 | 1952 | } |
e0a1988b | 1953 | |
3fe0b570 AB |
1954 | static void stop_tracers(void) |
1955 | { | |
1956 | struct list_head *p; | |
e0a1988b JA |
1957 | |
1958 | /* | |
3fe0b570 | 1959 | * Stop the tracing - makes the tracer threads clean up quicker. |
e0a1988b | 1960 | */ |
3fe0b570 AB |
1961 | __list_for_each(p, &devpaths) { |
1962 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1963 | (void)ioctl(dpp->fd, BLKTRACESTOP); | |
e0a1988b JA |
1964 | } |
1965 | ||
3fe0b570 AB |
1966 | /* |
1967 | * Tell each tracer to quit | |
1968 | */ | |
1969 | __list_for_each(p, &tracers) { | |
1970 | struct tracer *tp = list_entry(p, struct tracer, head); | |
1971 | tp->is_done = 1; | |
1972 | } | |
838361c6 | 1973 | pthread_cond_broadcast(&mt_cond); |
ff11d54c | 1974 | } |
e0a1988b | 1975 | |
3fe0b570 | 1976 | static void del_tracers(void) |
ff11d54c | 1977 | { |
3fe0b570 | 1978 | struct list_head *p, *q; |
ff11d54c | 1979 | |
3fe0b570 AB |
1980 | list_for_each_safe(p, q, &tracers) { |
1981 | struct tracer *tp = list_entry(p, struct tracer, head); | |
ff11d54c | 1982 | |
3fe0b570 AB |
1983 | list_del(&tp->head); |
1984 | free(tp); | |
e0a1988b | 1985 | } |
ff11d54c | 1986 | } |
e0a1988b | 1987 | |
3fe0b570 | 1988 | static void wait_tracers(void) |
ff11d54c | 1989 | { |
3fe0b570 | 1990 | struct list_head *p; |
ff11d54c | 1991 | |
3fe0b570 AB |
1992 | if (use_tracer_devpaths()) |
1993 | process_trace_bufs(); | |
1994 | ||
df81fdb5 AB |
1995 | wait_tracers_leaving(); |
1996 | ||
3fe0b570 AB |
1997 | __list_for_each(p, &tracers) { |
1998 | int ret; | |
1999 | struct tracer *tp = list_entry(p, struct tracer, head); | |
2000 | ||
3fe0b570 AB |
2001 | ret = pthread_join(tp->thread, NULL); |
2002 | if (ret) | |
2003 | fprintf(stderr, "Thread join %d failed %d\n", | |
2004 | tp->cpu, ret); | |
ff11d54c TZ |
2005 | } |
2006 | ||
3fe0b570 AB |
2007 | if (use_tracer_devpaths()) |
2008 | clean_trace_bufs(); | |
2009 | ||
2010 | get_all_drops(); | |
ff11d54c TZ |
2011 | } |
2012 | ||
3fe0b570 | 2013 | static void exit_tracing(void) |
ff11d54c | 2014 | { |
3fe0b570 AB |
2015 | signal(SIGINT, SIG_IGN); |
2016 | signal(SIGHUP, SIG_IGN); | |
2017 | signal(SIGTERM, SIG_IGN); | |
2018 | signal(SIGALRM, SIG_IGN); | |
2019 | ||
2020 | stop_tracers(); | |
2021 | wait_tracers(); | |
2022 | del_tracers(); | |
2023 | rel_devpaths(); | |
e0a1988b JA |
2024 | } |
2025 | ||
3fe0b570 | 2026 | static void handle_sigint(__attribute__((__unused__)) int sig) |
8e86c98a | 2027 | { |
3fe0b570 AB |
2028 | done = 1; |
2029 | stop_tracers(); | |
8e86c98a JA |
2030 | } |
2031 | ||
3fe0b570 | 2032 | static void show_stats(struct list_head *devpaths) |
659bcc3f | 2033 | { |
3fe0b570 AB |
2034 | FILE *ofp; |
2035 | struct list_head *p; | |
2036 | unsigned long long nevents, data_read; | |
2037 | unsigned long long total_drops = 0; | |
2038 | unsigned long long total_events = 0; | |
2039 | ||
2040 | if (piped_output) | |
2041 | ofp = my_fopen("/dev/null", "w"); | |
2042 | else | |
2043 | ofp = stdout; | |
ff11d54c | 2044 | |
3fe0b570 AB |
2045 | __list_for_each(p, devpaths) { |
2046 | int cpu; | |
2047 | struct pdc_stats *sp; | |
2048 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
e0a1988b | 2049 | |
3fe0b570 AB |
2050 | if (net_mode == Net_server) |
2051 | printf("server: end of run for %s:%s\n", | |
2052 | dpp->ch->hostname, dpp->buts_name); | |
e0a1988b | 2053 | |
3fe0b570 AB |
2054 | data_read = 0; |
2055 | nevents = 0; | |
2056 | ||
2057 | fprintf(ofp, "=== %s ===\n", dpp->buts_name); | |
2058 | for (cpu = 0, sp = dpp->stats; cpu < dpp->ncpus; cpu++, sp++) { | |
2059 | /* | |
2060 | * Estimate events if not known... | |
2061 | */ | |
2062 | if (sp->nevents == 0) { | |
2063 | sp->nevents = sp->data_read / | |
2064 | sizeof(struct blk_io_trace); | |
ff11d54c | 2065 | } |
e0a1988b | 2066 | |
3fe0b570 AB |
2067 | fprintf(ofp, |
2068 | " CPU%3d: %20llu events, %8llu KiB data\n", | |
2069 | cpu, sp->nevents, (sp->data_read + 1023) >> 10); | |
e0a1988b | 2070 | |
3fe0b570 AB |
2071 | data_read += sp->data_read; |
2072 | nevents += sp->nevents; | |
e0a1988b JA |
2073 | } |
2074 | ||
3fe0b570 AB |
2075 | fprintf(ofp, " Total: %20llu events (dropped %llu)," |
2076 | " %8llu KiB data\n", nevents, | |
2077 | dpp->drops, (data_read + 1024) >> 10); | |
8e86c98a | 2078 | |
3fe0b570 AB |
2079 | total_drops += dpp->drops; |
2080 | total_events += (nevents + dpp->drops); | |
8e86c98a JA |
2081 | } |
2082 | ||
3fe0b570 AB |
2083 | fflush(ofp); |
2084 | if (piped_output) | |
2085 | fclose(ofp); | |
8e86c98a | 2086 | |
3fe0b570 AB |
2087 | if (total_drops) { |
2088 | double drops_ratio = 1.0; | |
8e86c98a | 2089 | |
3fe0b570 AB |
2090 | if (total_events) |
2091 | drops_ratio = (double)total_drops/(double)total_events; | |
8e86c98a | 2092 | |
3fe0b570 AB |
2093 | fprintf(stderr, "\nYou have %llu (%5.1lf%%) dropped events\n" |
2094 | "Consider using a larger buffer size (-b) " | |
2095 | "and/or more buffers (-n)\n", | |
2096 | total_drops, 100.0 * drops_ratio); | |
8e86c98a | 2097 | } |
8e86c98a JA |
2098 | } |
2099 | ||
3fe0b570 | 2100 | static int handle_args(int argc, char *argv[]) |
8e86c98a | 2101 | { |
3fe0b570 | 2102 | int c, i; |
e3e74029 | 2103 | struct statfs st; |
d39c04ca AB |
2104 | int act_mask_tmp = 0; |
2105 | ||
2106 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) { | |
2107 | switch (c) { | |
2108 | case 'a': | |
2109 | i = find_mask_map(optarg); | |
2110 | if (i < 0) { | |
3fe0b570 | 2111 | fprintf(stderr, "Invalid action mask %s\n", |
d39c04ca | 2112 | optarg); |
7425d456 | 2113 | return 1; |
d39c04ca AB |
2114 | } |
2115 | act_mask_tmp |= i; | |
2116 | break; | |
2117 | ||
2118 | case 'A': | |
3fe0b570 | 2119 | if ((sscanf(optarg, "%x", &i) != 1) || |
98f8386b | 2120 | !valid_act_opt(i)) { |
d39c04ca | 2121 | fprintf(stderr, |
ab197ca7 | 2122 | "Invalid set action mask %s/0x%x\n", |
d39c04ca | 2123 | optarg, i); |
7425d456 | 2124 | return 1; |
d39c04ca AB |
2125 | } |
2126 | act_mask_tmp = i; | |
2127 | break; | |
d0ca268b | 2128 | |
d39c04ca | 2129 | case 'd': |
3fe0b570 | 2130 | if (add_devpath(optarg) != 0) |
e7c9f3ff | 2131 | return 1; |
d39c04ca AB |
2132 | break; |
2133 | ||
cf1edb17 AB |
2134 | case 'I': { |
2135 | char dev_line[256]; | |
3fe0b570 | 2136 | FILE *ifp = my_fopen(optarg, "r"); |
cf1edb17 AB |
2137 | |
2138 | if (!ifp) { | |
3fe0b570 AB |
2139 | fprintf(stderr, |
2140 | "Invalid file for devices %s\n", | |
cf1edb17 AB |
2141 | optarg); |
2142 | return 1; | |
2143 | } | |
2144 | ||
f9a89a6f ES |
2145 | while (fscanf(ifp, "%s\n", dev_line) == 1) { |
2146 | if (add_devpath(dev_line) != 0) { | |
2147 | fclose(ifp); | |
cf1edb17 | 2148 | return 1; |
f9a89a6f ES |
2149 | } |
2150 | } | |
2151 | fclose(ifp); | |
cf1edb17 AB |
2152 | break; |
2153 | } | |
cf1edb17 | 2154 | |
5270dddd | 2155 | case 'r': |
3d06efea | 2156 | debugfs_path = optarg; |
5270dddd JA |
2157 | break; |
2158 | ||
d5396421 | 2159 | case 'o': |
66efebf8 | 2160 | output_name = optarg; |
d5396421 | 2161 | break; |
bc39777c JA |
2162 | case 'k': |
2163 | kill_running_trace = 1; | |
2164 | break; | |
ece238a6 NS |
2165 | case 'w': |
2166 | stop_watch = atoi(optarg); | |
2167 | if (stop_watch <= 0) { | |
2168 | fprintf(stderr, | |
2169 | "Invalid stopwatch value (%d secs)\n", | |
2170 | stop_watch); | |
2171 | return 1; | |
2172 | } | |
2173 | break; | |
57ea8602 | 2174 | case 'V': |
5d4f19d9 | 2175 | case 'v': |
52724a0e | 2176 | printf("%s version %s\n", argv[0], blktrace_version); |
3fe0b570 AB |
2177 | exit(0); |
2178 | /*NOTREACHED*/ | |
129aa440 | 2179 | case 'b': |
eb3c8108 | 2180 | buf_size = strtoul(optarg, NULL, 10); |
183a0855 | 2181 | if (buf_size <= 0 || buf_size > 16*1024) { |
3fe0b570 AB |
2182 | fprintf(stderr, "Invalid buffer size (%lu)\n", |
2183 | buf_size); | |
129aa440 JA |
2184 | return 1; |
2185 | } | |
2186 | buf_size <<= 10; | |
2187 | break; | |
2188 | case 'n': | |
eb3c8108 | 2189 | buf_nr = strtoul(optarg, NULL, 10); |
129aa440 JA |
2190 | if (buf_nr <= 0) { |
2191 | fprintf(stderr, | |
eb3c8108 | 2192 | "Invalid buffer nr (%lu)\n", buf_nr); |
129aa440 JA |
2193 | return 1; |
2194 | } | |
2195 | break; | |
d1d7f15f JA |
2196 | case 'D': |
2197 | output_dir = optarg; | |
2198 | break; | |
8e86c98a JA |
2199 | case 'h': |
2200 | net_mode = Net_client; | |
d324757e ES |
2201 | memset(hostname, 0, sizeof(hostname)); |
2202 | strncpy(hostname, optarg, sizeof(hostname)); | |
2203 | hostname[sizeof(hostname) - 1] = '\0'; | |
8e86c98a JA |
2204 | break; |
2205 | case 'l': | |
2206 | net_mode = Net_server; | |
2207 | break; | |
2208 | case 'p': | |
2209 | net_port = atoi(optarg); | |
2210 | break; | |
32f18c48 | 2211 | case 's': |
79971f43 | 2212 | net_use_sendfile = 0; |
32f18c48 | 2213 | break; |
d39c04ca | 2214 | default: |
ee1f4158 | 2215 | show_usage(argv[0]); |
3fe0b570 AB |
2216 | exit(1); |
2217 | /*NOTREACHED*/ | |
d39c04ca AB |
2218 | } |
2219 | } | |
2220 | ||
3fe0b570 AB |
2221 | while (optind < argc) |
2222 | if (add_devpath(argv[optind++]) != 0) | |
2223 | return 1; | |
8e86c98a | 2224 | |
3fe0b570 AB |
2225 | if (net_mode != Net_server && ndevs == 0) { |
2226 | show_usage(argv[0]); | |
2227 | return 1; | |
2228 | } | |
8e86c98a | 2229 | |
d8365957 | 2230 | if (statfs(debugfs_path, &st) < 0) { |
3fe0b570 AB |
2231 | fprintf(stderr, "Invalid debug path %s: %d/%s\n", |
2232 | debugfs_path, errno, strerror(errno)); | |
2233 | return 1; | |
2234 | } | |
2235 | ||
d8365957 MP |
2236 | if (st.f_type != (long)DEBUGFS_TYPE) { |
2237 | fprintf(stderr, "Debugfs is not mounted at %s\n", debugfs_path); | |
2238 | return 1; | |
2239 | } | |
2240 | ||
3fe0b570 AB |
2241 | if (act_mask_tmp != 0) |
2242 | act_mask = act_mask_tmp; | |
2243 | ||
e58f3937 AB |
2244 | if (net_mode == Net_client && net_setup_addr()) |
2245 | return 1; | |
2246 | ||
3fe0b570 AB |
2247 | /* |
2248 | * Set up for appropriate PFD handler based upon output name. | |
2249 | */ | |
2250 | if (net_client_use_sendfile()) | |
2251 | handle_pfds = handle_pfds_netclient; | |
2252 | else if (net_client_use_send()) | |
2253 | handle_pfds = handle_pfds_entries; | |
2254 | else if (output_name && (strcmp(output_name, "-") == 0)) { | |
2255 | piped_output = 1; | |
2256 | handle_pfds = handle_pfds_entries; | |
2257 | pfp = stdout; | |
ae2dc05e ES |
2258 | if (setvbuf(pfp, NULL, _IONBF, 0)) { |
2259 | perror("setvbuf stdout"); | |
2260 | return 1; | |
2261 | } | |
3fe0b570 AB |
2262 | } else |
2263 | handle_pfds = handle_pfds_file; | |
2264 | return 0; | |
2265 | } | |
2266 | ||
2267 | static void ch_add_connection(struct net_server_s *ns, struct cl_host *ch, | |
2268 | int fd) | |
2269 | { | |
2270 | struct cl_conn *nc; | |
2271 | ||
2272 | nc = malloc(sizeof(*nc)); | |
2273 | memset(nc, 0, sizeof(*nc)); | |
2274 | ||
2275 | time(&nc->connect_time); | |
2276 | nc->ch = ch; | |
2277 | nc->fd = fd; | |
2278 | nc->ncpus = -1; | |
2279 | ||
2280 | list_add_tail(&nc->ch_head, &ch->conn_list); | |
2281 | ch->connects++; | |
2282 | ||
2283 | list_add_tail(&nc->ns_head, &ns->conn_list); | |
2284 | ns->connects++; | |
2285 | ns->pfds = realloc(ns->pfds, (ns->connects+1) * sizeof(struct pollfd)); | |
2286 | } | |
2287 | ||
2288 | static void ch_rem_connection(struct net_server_s *ns, struct cl_host *ch, | |
2289 | struct cl_conn *nc) | |
2290 | { | |
2291 | net_close_connection(&nc->fd); | |
2292 | ||
2293 | list_del(&nc->ch_head); | |
2294 | ch->connects--; | |
2295 | ||
2296 | list_del(&nc->ns_head); | |
2297 | ns->connects--; | |
2298 | ns->pfds = realloc(ns->pfds, (ns->connects+1) * sizeof(struct pollfd)); | |
2299 | ||
2300 | free(nc); | |
2301 | } | |
2302 | ||
2303 | static struct cl_host *net_find_client_host(struct net_server_s *ns, | |
2304 | struct in_addr cl_in_addr) | |
2305 | { | |
2306 | struct list_head *p; | |
2307 | ||
2308 | __list_for_each(p, &ns->ch_list) { | |
2309 | struct cl_host *ch = list_entry(p, struct cl_host, head); | |
2310 | ||
2311 | if (in_addr_eq(ch->cl_in_addr, cl_in_addr)) | |
2312 | return ch; | |
2313 | } | |
2314 | ||
2315 | return NULL; | |
2316 | } | |
2317 | ||
2318 | static struct cl_host *net_add_client_host(struct net_server_s *ns, | |
2319 | struct sockaddr_in *addr) | |
2320 | { | |
2321 | struct cl_host *ch; | |
2322 | ||
2323 | ch = malloc(sizeof(*ch)); | |
2324 | memset(ch, 0, sizeof(*ch)); | |
2325 | ||
2326 | ch->ns = ns; | |
2327 | ch->cl_in_addr = addr->sin_addr; | |
2328 | list_add_tail(&ch->head, &ns->ch_list); | |
2329 | ns->nchs++; | |
ec685dd2 | 2330 | |
3fe0b570 AB |
2331 | ch->hostname = strdup(inet_ntoa(addr->sin_addr)); |
2332 | printf("server: connection from %s\n", ch->hostname); | |
2333 | ||
2334 | INIT_LIST_HEAD(&ch->conn_list); | |
2335 | INIT_LIST_HEAD(&ch->devpaths); | |
2336 | ||
2337 | return ch; | |
2338 | } | |
2339 | ||
2340 | static void device_done(struct devpath *dpp, int ncpus) | |
2341 | { | |
2342 | int cpu; | |
2343 | struct io_info *iop; | |
2344 | ||
2345 | for (cpu = 0, iop = dpp->ios; cpu < ncpus; cpu++, iop++) | |
2346 | close_iop(iop); | |
2347 | ||
2348 | list_del(&dpp->head); | |
2349 | dpp_free(dpp); | |
2350 | } | |
2351 | ||
2352 | static void net_ch_remove(struct cl_host *ch, int ncpus) | |
2353 | { | |
2354 | struct list_head *p, *q; | |
2355 | struct net_server_s *ns = ch->ns; | |
2356 | ||
2357 | list_for_each_safe(p, q, &ch->devpaths) { | |
2358 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
2359 | device_done(dpp, ncpus); | |
ec685dd2 | 2360 | } |
8e86c98a | 2361 | |
3fe0b570 AB |
2362 | list_for_each_safe(p, q, &ch->conn_list) { |
2363 | struct cl_conn *nc = list_entry(p, struct cl_conn, ch_head); | |
2364 | ||
2365 | ch_rem_connection(ns, ch, nc); | |
22cd0c02 JA |
2366 | } |
2367 | ||
3fe0b570 AB |
2368 | list_del(&ch->head); |
2369 | ns->nchs--; | |
2370 | ||
2371 | if (ch->hostname) | |
2372 | free(ch->hostname); | |
2373 | free(ch); | |
2374 | } | |
2375 | ||
2376 | static void net_add_connection(struct net_server_s *ns) | |
2377 | { | |
2378 | int fd; | |
2379 | struct cl_host *ch; | |
2380 | socklen_t socklen = sizeof(ns->addr); | |
2381 | ||
d5302b03 | 2382 | fd = my_accept(ns->listen_fd, (struct sockaddr *)&ns->addr, &socklen); |
3fe0b570 AB |
2383 | if (fd < 0) { |
2384 | /* | |
2385 | * This is OK: we just won't accept this connection, | |
2386 | * nothing fatal. | |
2387 | */ | |
2388 | perror("accept"); | |
2389 | } else { | |
2390 | ch = net_find_client_host(ns, ns->addr.sin_addr); | |
2391 | if (!ch) | |
2392 | ch = net_add_client_host(ns, &ns->addr); | |
2393 | ||
2394 | ch_add_connection(ns, ch, fd); | |
d39c04ca | 2395 | } |
3fe0b570 | 2396 | } |
d39c04ca | 2397 | |
3fe0b570 AB |
2398 | static struct devpath *nc_add_dpp(struct cl_conn *nc, |
2399 | struct blktrace_net_hdr *bnh, | |
2400 | time_t connect_time) | |
2401 | { | |
2402 | int cpu; | |
2403 | struct io_info *iop; | |
2404 | struct devpath *dpp; | |
2405 | ||
2406 | dpp = malloc(sizeof(*dpp)); | |
2407 | memset(dpp, 0, sizeof(*dpp)); | |
2408 | ||
2409 | dpp->buts_name = strdup(bnh->buts_name); | |
2410 | dpp->path = strdup(bnh->buts_name); | |
2411 | dpp->fd = -1; | |
2412 | dpp->ch = nc->ch; | |
2413 | dpp->cl_id = bnh->cl_id; | |
2414 | dpp->cl_connect_time = connect_time; | |
2415 | dpp->ncpus = nc->ncpus; | |
2416 | dpp->stats = calloc(dpp->ncpus, sizeof(*dpp->stats)); | |
2417 | memset(dpp->stats, 0, dpp->ncpus * sizeof(*dpp->stats)); | |
2418 | ||
2419 | list_add_tail(&dpp->head, &nc->ch->devpaths); | |
2420 | nc->ch->ndevs++; | |
2421 | ||
2422 | dpp->ios = calloc(nc->ncpus, sizeof(*iop)); | |
2423 | memset(dpp->ios, 0, ndevs * sizeof(*iop)); | |
2424 | ||
2425 | for (cpu = 0, iop = dpp->ios; cpu < nc->ncpus; cpu++, iop++) { | |
2426 | iop->dpp = dpp; | |
2427 | iop->nc = nc; | |
2428 | init_mmap_info(&iop->mmap_info); | |
2429 | ||
2430 | if (iop_open(iop, cpu)) | |
2431 | goto err; | |
69dd57c2 AB |
2432 | } |
2433 | ||
3fe0b570 | 2434 | return dpp; |
69dd57c2 | 2435 | |
3fe0b570 AB |
2436 | err: |
2437 | /* | |
2438 | * Need to unravel what's been done... | |
2439 | */ | |
2440 | while (cpu >= 0) | |
2441 | close_iop(&dpp->ios[cpu--]); | |
2442 | dpp_free(dpp); | |
2443 | ||
2444 | return NULL; | |
2445 | } | |
d0ca268b | 2446 | |
3fe0b570 AB |
2447 | static struct devpath *nc_find_dpp(struct cl_conn *nc, |
2448 | struct blktrace_net_hdr *bnh) | |
2449 | { | |
2450 | struct list_head *p; | |
2451 | time_t connect_time = nc->connect_time; | |
3d06efea | 2452 | |
3fe0b570 AB |
2453 | __list_for_each(p, &nc->ch->devpaths) { |
2454 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
2455 | ||
2456 | if (!strcmp(dpp->buts_name, bnh->buts_name)) | |
2457 | return dpp; | |
2458 | ||
2459 | if (dpp->cl_id == bnh->cl_id) | |
2460 | connect_time = dpp->cl_connect_time; | |
d0ca268b JA |
2461 | } |
2462 | ||
3fe0b570 AB |
2463 | return nc_add_dpp(nc, bnh, connect_time); |
2464 | } | |
bc39777c | 2465 | |
3fe0b570 AB |
2466 | static void net_client_read_data(struct cl_conn *nc, struct devpath *dpp, |
2467 | struct blktrace_net_hdr *bnh) | |
2468 | { | |
2469 | int ret; | |
2470 | struct io_info *iop = &dpp->ios[bnh->cpu]; | |
2471 | struct mmap_info *mip = &iop->mmap_info; | |
2472 | ||
ae7c049d | 2473 | if (setup_mmap(iop->ofd, bnh->len, &iop->mmap_info, NULL)) { |
3fe0b570 AB |
2474 | fprintf(stderr, "ncd(%s:%d): mmap failed\n", |
2475 | nc->ch->hostname, nc->fd); | |
2476 | exit(1); | |
2477 | } | |
2478 | ||
2479 | ret = net_recv_data(nc->fd, mip->fs_buf + mip->fs_off, bnh->len); | |
2480 | if (ret > 0) { | |
2481 | pdc_dr_update(dpp, bnh->cpu, ret); | |
2482 | mip->fs_size += ret; | |
2483 | mip->fs_off += ret; | |
2484 | } else if (ret < 0) | |
2485 | exit(1); | |
2486 | } | |
2487 | ||
2488 | /* | |
2489 | * Returns 1 if we closed a host - invalidates other polling information | |
2490 | * that may be present. | |
2491 | */ | |
2492 | static int net_client_data(struct cl_conn *nc) | |
2493 | { | |
2494 | int ret; | |
2495 | struct devpath *dpp; | |
2496 | struct blktrace_net_hdr bnh; | |
2497 | ||
2498 | ret = net_get_header(nc, &bnh); | |
2499 | if (ret == 0) | |
7425d456 | 2500 | return 0; |
3fe0b570 AB |
2501 | |
2502 | if (ret < 0) { | |
2503 | fprintf(stderr, "ncd(%d): header read failed\n", nc->fd); | |
2504 | exit(1); | |
2505 | } | |
2506 | ||
2507 | if (data_is_native == -1 && check_data_endianness(bnh.magic)) { | |
2508 | fprintf(stderr, "ncd(%d): received data is bad\n", nc->fd); | |
2509 | exit(1); | |
2510 | } | |
2511 | ||
2512 | if (!data_is_native) { | |
2513 | bnh.magic = be32_to_cpu(bnh.magic); | |
2514 | bnh.cpu = be32_to_cpu(bnh.cpu); | |
2515 | bnh.max_cpus = be32_to_cpu(bnh.max_cpus); | |
2516 | bnh.len = be32_to_cpu(bnh.len); | |
2517 | bnh.cl_id = be32_to_cpu(bnh.cl_id); | |
2518 | bnh.buf_size = be32_to_cpu(bnh.buf_size); | |
2519 | bnh.buf_nr = be32_to_cpu(bnh.buf_nr); | |
2520 | bnh.page_size = be32_to_cpu(bnh.page_size); | |
2521 | } | |
2522 | ||
2523 | if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
2524 | fprintf(stderr, "ncd(%s:%d): bad data magic\n", | |
2525 | nc->ch->hostname, nc->fd); | |
2526 | exit(1); | |
2527 | } | |
2528 | ||
2529 | if (nc->ncpus == -1) | |
2530 | nc->ncpus = bnh.max_cpus; | |
2531 | ||
2532 | /* | |
2533 | * len == 0 means the other end is sending us a new connection/dpp | |
2534 | * len == 1 means that the other end signalled end-of-run | |
2535 | */ | |
2536 | dpp = nc_find_dpp(nc, &bnh); | |
2537 | if (bnh.len == 0) { | |
2538 | /* | |
2539 | * Just adding in the dpp above is enough | |
2540 | */ | |
2541 | ack_open_close(nc->fd, dpp->buts_name); | |
2542 | nc->ch->cl_opens++; | |
2543 | } else if (bnh.len == 1) { | |
2544 | /* | |
2545 | * overload cpu count with dropped events | |
2546 | */ | |
2547 | dpp->drops = bnh.cpu; | |
2548 | ||
2549 | ack_open_close(nc->fd, dpp->buts_name); | |
2550 | if (--nc->ch->cl_opens == 0) { | |
2551 | show_stats(&nc->ch->devpaths); | |
2552 | net_ch_remove(nc->ch, nc->ncpus); | |
2553 | return 1; | |
2554 | } | |
2555 | } else | |
2556 | net_client_read_data(nc, dpp, &bnh); | |
2557 | ||
2558 | return 0; | |
2559 | } | |
2560 | ||
2561 | static void handle_client_data(struct net_server_s *ns, int events) | |
2562 | { | |
2563 | struct cl_conn *nc; | |
2564 | struct pollfd *pfd; | |
2565 | struct list_head *p, *q; | |
2566 | ||
2567 | pfd = &ns->pfds[1]; | |
2568 | list_for_each_safe(p, q, &ns->conn_list) { | |
2569 | if (pfd->revents & POLLIN) { | |
2570 | nc = list_entry(p, struct cl_conn, ns_head); | |
2571 | ||
2572 | if (net_client_data(nc) || --events == 0) | |
2573 | break; | |
2574 | } | |
2575 | pfd++; | |
2576 | } | |
2577 | } | |
2578 | ||
2579 | static void net_setup_pfds(struct net_server_s *ns) | |
2580 | { | |
2581 | struct pollfd *pfd; | |
2582 | struct list_head *p; | |
2583 | ||
2584 | ns->pfds[0].fd = ns->listen_fd; | |
2585 | ns->pfds[0].events = POLLIN; | |
2586 | ||
2587 | pfd = &ns->pfds[1]; | |
2588 | __list_for_each(p, &ns->conn_list) { | |
2589 | struct cl_conn *nc = list_entry(p, struct cl_conn, ns_head); | |
2590 | ||
2591 | pfd->fd = nc->fd; | |
2592 | pfd->events = POLLIN; | |
2593 | pfd++; | |
2594 | } | |
2595 | } | |
2596 | ||
2597 | static int net_server_handle_connections(struct net_server_s *ns) | |
2598 | { | |
2599 | int events; | |
2600 | ||
2601 | printf("server: waiting for connections...\n"); | |
2602 | ||
2603 | while (!done) { | |
2604 | net_setup_pfds(ns); | |
2605 | events = poll(ns->pfds, ns->connects + 1, -1); | |
2606 | if (events < 0) { | |
2607 | if (errno != EINTR) { | |
2608 | perror("FATAL: poll error"); | |
2609 | return 1; | |
2610 | } | |
2611 | } else if (events > 0) { | |
2612 | if (ns->pfds[0].revents & POLLIN) { | |
2613 | net_add_connection(ns); | |
2614 | events--; | |
2615 | } | |
2616 | ||
2617 | if (events) | |
2618 | handle_client_data(ns, events); | |
2619 | } | |
2620 | } | |
2621 | ||
2622 | return 0; | |
2623 | } | |
2624 | ||
2625 | static int net_server(void) | |
2626 | { | |
2627 | int fd, opt; | |
2628 | int ret = 1; | |
2629 | struct net_server_s net_server; | |
2630 | struct net_server_s *ns = &net_server; | |
2631 | ||
2632 | memset(ns, 0, sizeof(*ns)); | |
2633 | INIT_LIST_HEAD(&ns->ch_list); | |
2634 | INIT_LIST_HEAD(&ns->conn_list); | |
2635 | ns->pfds = malloc(sizeof(struct pollfd)); | |
2636 | ||
2637 | fd = my_socket(AF_INET, SOCK_STREAM, 0); | |
2638 | if (fd < 0) { | |
2639 | perror("server: socket"); | |
2640 | goto out; | |
2641 | } | |
2642 | ||
2643 | opt = 1; | |
2644 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) { | |
2645 | perror("setsockopt"); | |
2646 | goto out; | |
2647 | } | |
2648 | ||
2649 | memset(&ns->addr, 0, sizeof(ns->addr)); | |
2650 | ns->addr.sin_family = AF_INET; | |
2651 | ns->addr.sin_addr.s_addr = htonl(INADDR_ANY); | |
2652 | ns->addr.sin_port = htons(net_port); | |
2653 | ||
2654 | if (bind(fd, (struct sockaddr *) &ns->addr, sizeof(ns->addr)) < 0) { | |
2655 | perror("bind"); | |
2656 | goto out; | |
2657 | } | |
2658 | ||
2659 | if (listen(fd, 1) < 0) { | |
2660 | perror("listen"); | |
2661 | goto out; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * The actual server looping is done here: | |
2666 | */ | |
2667 | ns->listen_fd = fd; | |
2668 | ret = net_server_handle_connections(ns); | |
2669 | ||
2670 | /* | |
2671 | * Clean up and return... | |
2672 | */ | |
2673 | out: | |
2674 | free(ns->pfds); | |
2675 | return ret; | |
2676 | } | |
2677 | ||
055cc3e5 AB |
2678 | static int run_tracers(void) |
2679 | { | |
2680 | atexit(exit_tracing); | |
2681 | if (net_mode == Net_client) | |
2682 | printf("blktrace: connecting to %s\n", hostname); | |
2683 | ||
ab6809de JA |
2684 | if (setup_buts()) |
2685 | return 1; | |
055cc3e5 AB |
2686 | |
2687 | if (use_tracer_devpaths()) { | |
2688 | if (setup_tracer_devpaths()) | |
2689 | return 1; | |
2690 | ||
2691 | if (piped_output) | |
2692 | handle_list = handle_list_file; | |
2693 | else | |
2694 | handle_list = handle_list_net; | |
2695 | } | |
2696 | ||
2697 | start_tracers(); | |
2698 | if (nthreads_running == ncpus) { | |
2699 | unblock_tracers(); | |
2700 | start_buts(); | |
2701 | if (net_mode == Net_client) | |
2702 | printf("blktrace: connected!\n"); | |
2703 | if (stop_watch) | |
2704 | alarm(stop_watch); | |
2705 | } else | |
2706 | stop_tracers(); | |
2707 | ||
2708 | wait_tracers(); | |
2709 | if (nthreads_running == ncpus) | |
2710 | show_stats(&devpaths); | |
2711 | if (net_client_use_send()) | |
2712 | close_client_connections(); | |
2713 | del_tracers(); | |
2714 | ||
2715 | return 0; | |
2716 | } | |
2717 | ||
d045a704 JK |
2718 | static cpu_set_t *get_online_cpus(void) |
2719 | { | |
2720 | FILE *cpus; | |
2721 | cpu_set_t *set; | |
2722 | size_t alloc_size; | |
2723 | int cpuid, prevcpuid = -1; | |
2724 | char nextch; | |
2725 | int n, ncpu, curcpu = 0; | |
2726 | int *cpu_nums; | |
2727 | ||
2728 | ncpu = sysconf(_SC_NPROCESSORS_CONF); | |
2729 | if (ncpu < 0) | |
2730 | return NULL; | |
2731 | ||
2732 | cpu_nums = malloc(sizeof(int)*ncpu); | |
2733 | if (!cpu_nums) { | |
2734 | errno = ENOMEM; | |
2735 | return NULL; | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * There is no way to easily get maximum CPU number. So we have to | |
2740 | * parse the file first to find it out and then create appropriate | |
2741 | * cpuset | |
2742 | */ | |
2743 | cpus = my_fopen("/sys/devices/system/cpu/online", "r"); | |
2744 | for (;;) { | |
2745 | n = fscanf(cpus, "%d%c", &cpuid, &nextch); | |
2746 | if (n <= 0) | |
2747 | break; | |
2748 | if (n == 2 && nextch == '-') { | |
2749 | prevcpuid = cpuid; | |
2750 | continue; | |
2751 | } | |
2752 | if (prevcpuid == -1) | |
2753 | prevcpuid = cpuid; | |
2754 | while (prevcpuid <= cpuid) { | |
2755 | /* More CPUs listed than configured? */ | |
2756 | if (curcpu >= ncpu) { | |
2757 | errno = EINVAL; | |
2758 | return NULL; | |
2759 | } | |
2760 | cpu_nums[curcpu++] = prevcpuid++; | |
2761 | } | |
2762 | prevcpuid = -1; | |
2763 | } | |
2764 | fclose(cpus); | |
2765 | ||
2766 | ncpu = curcpu; | |
2767 | max_cpus = cpu_nums[ncpu - 1] + 1; | |
2768 | ||
2769 | /* Now that we have maximum cpu number, create a cpuset */ | |
2770 | set = CPU_ALLOC(max_cpus); | |
2771 | if (!set) { | |
2772 | errno = ENOMEM; | |
2773 | return NULL; | |
2774 | } | |
2775 | alloc_size = CPU_ALLOC_SIZE(max_cpus); | |
2776 | CPU_ZERO_S(alloc_size, set); | |
2777 | ||
2778 | for (curcpu = 0; curcpu < ncpu; curcpu++) | |
2779 | CPU_SET_S(cpu_nums[curcpu], alloc_size, set); | |
2780 | ||
2781 | free(cpu_nums); | |
2782 | ||
2783 | return set; | |
2784 | } | |
2785 | ||
3fe0b570 AB |
2786 | int main(int argc, char *argv[]) |
2787 | { | |
2788 | int ret = 0; | |
2789 | ||
2790 | setlocale(LC_NUMERIC, "en_US"); | |
2791 | pagesize = getpagesize(); | |
d045a704 JK |
2792 | online_cpus = get_online_cpus(); |
2793 | if (!online_cpus) { | |
2794 | fprintf(stderr, "cannot get online cpus %d/%s\n", | |
3fe0b570 AB |
2795 | errno, strerror(errno)); |
2796 | ret = 1; | |
2797 | goto out; | |
055cc3e5 | 2798 | } else if (handle_args(argc, argv)) { |
3fe0b570 AB |
2799 | ret = 1; |
2800 | goto out; | |
bc39777c JA |
2801 | } |
2802 | ||
d045a704 | 2803 | ncpus = CPU_COUNT_S(CPU_ALLOC_SIZE(max_cpus), online_cpus); |
ce2151eb AB |
2804 | if (ndevs > 1 && output_name && strcmp(output_name, "-") != 0) { |
2805 | fprintf(stderr, "-o not supported with multiple devices\n"); | |
2806 | ret = 1; | |
2807 | goto out; | |
2808 | } | |
2809 | ||
d0ca268b JA |
2810 | signal(SIGINT, handle_sigint); |
2811 | signal(SIGHUP, handle_sigint); | |
2812 | signal(SIGTERM, handle_sigint); | |
ece238a6 | 2813 | signal(SIGALRM, handle_sigint); |
38e1f0c6 | 2814 | signal(SIGPIPE, SIG_IGN); |
d0ca268b | 2815 | |
3fe0b570 AB |
2816 | if (kill_running_trace) { |
2817 | struct devpath *dpp; | |
2818 | struct list_head *p; | |
8e86c98a | 2819 | |
3fe0b570 AB |
2820 | __list_for_each(p, &devpaths) { |
2821 | dpp = list_entry(p, struct devpath, head); | |
2822 | if (__stop_trace(dpp->fd)) { | |
2823 | fprintf(stderr, | |
2824 | "BLKTRACETEARDOWN %s failed: %d/%s\n", | |
2825 | dpp->path, errno, strerror(errno)); | |
2826 | } | |
2827 | } | |
2828 | } else if (net_mode == Net_server) { | |
2829 | if (output_name) { | |
2830 | fprintf(stderr, "-o ignored in server mode\n"); | |
2831 | output_name = NULL; | |
2832 | } | |
3fe0b570 | 2833 | ret = net_server(); |
055cc3e5 AB |
2834 | } else |
2835 | ret = run_tracers(); | |
d0ca268b | 2836 | |
3fe0b570 AB |
2837 | out: |
2838 | if (pfp) | |
2839 | fclose(pfp); | |
2840 | rel_devpaths(); | |
2841 | return ret; | |
2842 | } |