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 | ||
74 | /* | |
75 | * Generic stats collected: nevents can be _roughly_ estimated by data_read | |
76 | * (discounting pdu...) | |
77 | * | |
78 | * These fields are updated w/ pdc_dr_update & pdc_nev_update below. | |
79 | */ | |
80 | struct pdc_stats { | |
81 | unsigned long long data_read; | |
82 | unsigned long long nevents; | |
83 | }; | |
84 | ||
85 | struct devpath { | |
86 | struct list_head head; | |
87 | char *path; /* path to device special file */ | |
88 | char *buts_name; /* name returned from bt kernel code */ | |
89 | struct pdc_stats *stats; | |
90 | int fd, idx, ncpus; | |
91 | unsigned long long drops; | |
92 | ||
93 | /* | |
94 | * For piped output only: | |
95 | * | |
96 | * Each tracer will have a tracer_devpath_head that it will add new | |
97 | * data onto. It's list is protected above (tracer_devpath_head.mutex) | |
98 | * and it will signal the processing thread using the dp_cond, | |
99 | * dp_mutex & dp_entries variables above. | |
100 | */ | |
101 | struct tracer_devpath_head *heads; | |
102 | ||
103 | /* | |
104 | * For network server mode only: | |
105 | */ | |
106 | struct cl_host *ch; | |
107 | u32 cl_id; | |
108 | time_t cl_connect_time; | |
109 | struct io_info *ios; | |
110 | }; | |
111 | ||
112 | /* | |
113 | * For piped output to stdout we will have each tracer thread (one per dev) | |
114 | * tack buffers read from the relay queues on a per-device list. | |
115 | * | |
116 | * The main thread will then collect trace buffers from each of lists in turn. | |
117 | * | |
118 | * We will use a mutex to guard each of the trace_buf list. The tracers | |
119 | * can then signal the main thread using <dp_cond,dp_mutex> and | |
120 | * dp_entries. (When dp_entries is 0, and a tracer adds an entry it will | |
121 | * signal. When dp_entries is 0, the main thread will wait for that condition | |
122 | * to be signalled.) | |
123 | * | |
124 | * adb: It may be better just to have a large buffer per tracer per dev, | |
125 | * and then use it as a ring-buffer. This would certainly cut down a lot | |
126 | * of malloc/free thrashing, at the cost of more memory movements (potentially). | |
127 | */ | |
128 | struct trace_buf { | |
129 | struct list_head head; | |
130 | struct devpath *dpp; | |
131 | void *buf; | |
132 | int cpu, len; | |
133 | }; | |
134 | ||
135 | struct tracer_devpath_head { | |
136 | pthread_mutex_t mutex; | |
137 | struct list_head head; | |
138 | struct trace_buf *prev; | |
139 | }; | |
140 | ||
141 | /* | |
142 | * Used to handle the mmap() interfaces for output file (containing traces) | |
143 | */ | |
144 | struct mmap_info { | |
145 | void *fs_buf; | |
146 | unsigned long long fs_size, fs_max_size, fs_off, fs_buf_len; | |
147 | unsigned long buf_size, buf_nr; | |
148 | int pagesize; | |
149 | }; | |
150 | ||
151 | /* | |
152 | * Each thread doing work on a (client) side of blktrace will have one | |
153 | * of these. The ios array contains input/output information, pfds holds | |
154 | * poll() data. The volatile's provide flags to/from the main executing | |
155 | * thread. | |
156 | */ | |
157 | struct tracer { | |
158 | struct list_head head; | |
159 | struct io_info *ios; | |
160 | struct pollfd *pfds; | |
161 | pthread_t thread; | |
162 | pthread_mutex_t mutex; | |
163 | pthread_cond_t cond; | |
164 | int cpu, nios; | |
165 | volatile int running, status, is_done; | |
166 | }; | |
167 | ||
168 | /* | |
169 | * networking stuff follows. we include a magic number so we know whether | |
170 | * to endianness convert or not. | |
171 | * | |
172 | * The len field is overloaded: | |
173 | * 0 - Indicates an "open" - allowing the server to set up for a dev/cpu | |
174 | * 1 - Indicates a "close" - Shut down connection orderly | |
175 | * | |
176 | * The cpu field is overloaded on close: it will contain the number of drops. | |
177 | */ | |
178 | struct blktrace_net_hdr { | |
179 | u32 magic; /* same as trace magic */ | |
180 | char buts_name[32]; /* trace name */ | |
181 | u32 cpu; /* for which cpu */ | |
182 | u32 max_cpus; | |
183 | u32 len; /* length of following trace data */ | |
184 | u32 cl_id; /* id for set of client per-cpu connections */ | |
185 | u32 buf_size; /* client buf_size for this trace */ | |
186 | u32 buf_nr; /* client buf_nr for this trace */ | |
187 | u32 page_size; /* client page_size for this trace */ | |
188 | }; | |
189 | ||
190 | /* | |
191 | * Each host encountered has one of these. The head is used to link this | |
192 | * on to the network server's ch_list. Connections associated with this | |
193 | * host are linked on conn_list, and any devices traced on that host | |
194 | * are connected on the devpaths list. | |
195 | */ | |
196 | struct cl_host { | |
197 | struct list_head head; | |
198 | struct list_head conn_list; | |
199 | struct list_head devpaths; | |
200 | struct net_server_s *ns; | |
201 | char *hostname; | |
202 | struct in_addr cl_in_addr; | |
203 | int connects, ndevs, cl_opens; | |
204 | }; | |
205 | ||
206 | /* | |
207 | * Each connection (client to server socket ('fd')) has one of these. A | |
208 | * back reference to the host ('ch'), and lists headers (for the host | |
209 | * list, and the network server conn_list) are also included. | |
210 | */ | |
211 | struct cl_conn { | |
212 | struct list_head ch_head, ns_head; | |
213 | struct cl_host *ch; | |
214 | int fd, ncpus; | |
215 | time_t connect_time; | |
216 | }; | |
217 | ||
218 | /* | |
219 | * The network server requires some poll structures to be maintained - | |
220 | * one per conection currently on conn_list. The nchs/ch_list values | |
221 | * are for each host connected to this server. The addr field is used | |
222 | * for scratch as new connections are established. | |
223 | */ | |
224 | struct net_server_s { | |
225 | struct list_head conn_list; | |
226 | struct list_head ch_list; | |
227 | struct pollfd *pfds; | |
228 | int listen_fd, connects, nchs; | |
229 | struct sockaddr_in addr; | |
230 | }; | |
231 | ||
232 | /* | |
233 | * This structure is (generically) used to providide information | |
234 | * for a read-to-write set of values. | |
235 | * | |
236 | * ifn & ifd represent input information | |
237 | * | |
238 | * ofn, ofd, ofp, obuf & mmap_info are used for output file (optionally). | |
239 | */ | |
240 | struct io_info { | |
241 | struct devpath *dpp; | |
242 | FILE *ofp; | |
243 | char *obuf; | |
244 | struct cl_conn *nc; /* Server network connection */ | |
245 | ||
246 | /* | |
247 | * mmap controlled output files | |
248 | */ | |
249 | struct mmap_info mmap_info; | |
250 | ||
251 | /* | |
252 | * Client network fields | |
253 | */ | |
254 | unsigned int ready; | |
255 | unsigned long long data_queued; | |
256 | ||
257 | /* | |
258 | * Input/output file descriptors & names | |
259 | */ | |
260 | int ifd, ofd; | |
261 | char ifn[MAXPATHLEN + 64]; | |
262 | char ofn[MAXPATHLEN + 64]; | |
263 | }; | |
264 | ||
265 | static char blktrace_version[] = "2.0.0"; | |
266 | ||
267 | /* | |
268 | * Linkage to blktrace helper routines (trace conversions) | |
269 | */ | |
270 | int data_is_native = -1; | |
271 | ||
272 | static int ncpus; | |
273 | static int pagesize; | |
274 | static int act_mask = ~0U; | |
275 | static char *debugfs_path = "/sys/kernel/debug"; | |
276 | static char *output_name; | |
277 | static char *output_dir; | |
278 | static int kill_running_trace; | |
279 | static int stop_watch; | |
280 | static unsigned long buf_size = BUF_SIZE; | |
281 | static unsigned long buf_nr = BUF_NR; | |
282 | static LIST_HEAD(devpaths); | |
283 | static LIST_HEAD(tracers); | |
284 | static int ndevs; | |
285 | static volatile int done; | |
286 | static FILE *pfp; | |
287 | static int piped_output; | |
288 | static int ntracers; | |
d0ca268b | 289 | |
3fe0b570 AB |
290 | static pthread_cond_t dp_cond = PTHREAD_COND_INITIALIZER; |
291 | static pthread_mutex_t dp_mutex = PTHREAD_MUTEX_INITIALIZER; | |
292 | static volatile int dp_entries; | |
293 | ||
294 | /* | |
295 | * network cmd line params | |
296 | */ | |
e58f3937 | 297 | static struct sockaddr_in hostname_addr; |
3fe0b570 AB |
298 | static char hostname[MAXHOSTNAMELEN]; |
299 | static int net_port = TRACE_NET_PORT; | |
300 | static int net_use_sendfile = 1; | |
301 | static int net_mode; | |
302 | static int *cl_fds; | |
007c233c | 303 | |
3fe0b570 AB |
304 | static int (*handle_pfds)(struct tracer *, int, int); |
305 | static int (*handle_list)(struct tracer_devpath_head *, struct list_head *); | |
e3e74029 | 306 | |
5d4f19d9 | 307 | #define S_OPTS "d:a:A:r:o:kw:vVb:n:D:lh:p:sI:" |
d5396421 | 308 | static struct option l_opts[] = { |
5c86134e | 309 | { |
d39c04ca | 310 | .name = "dev", |
428683db | 311 | .has_arg = required_argument, |
d39c04ca AB |
312 | .flag = NULL, |
313 | .val = 'd' | |
314 | }, | |
cf1edb17 AB |
315 | { |
316 | .name = "input-devs", | |
317 | .has_arg = required_argument, | |
318 | .flag = NULL, | |
319 | .val = 'I' | |
320 | }, | |
5c86134e | 321 | { |
d39c04ca | 322 | .name = "act-mask", |
428683db | 323 | .has_arg = required_argument, |
d39c04ca AB |
324 | .flag = NULL, |
325 | .val = 'a' | |
326 | }, | |
5c86134e | 327 | { |
d39c04ca | 328 | .name = "set-mask", |
428683db | 329 | .has_arg = required_argument, |
d39c04ca AB |
330 | .flag = NULL, |
331 | .val = 'A' | |
332 | }, | |
5c86134e | 333 | { |
5270dddd | 334 | .name = "relay", |
428683db | 335 | .has_arg = required_argument, |
5270dddd JA |
336 | .flag = NULL, |
337 | .val = 'r' | |
338 | }, | |
d5396421 JA |
339 | { |
340 | .name = "output", | |
428683db | 341 | .has_arg = required_argument, |
d5396421 JA |
342 | .flag = NULL, |
343 | .val = 'o' | |
344 | }, | |
bc39777c JA |
345 | { |
346 | .name = "kill", | |
428683db | 347 | .has_arg = no_argument, |
bc39777c JA |
348 | .flag = NULL, |
349 | .val = 'k' | |
350 | }, | |
ece238a6 NS |
351 | { |
352 | .name = "stopwatch", | |
428683db | 353 | .has_arg = required_argument, |
ece238a6 NS |
354 | .flag = NULL, |
355 | .val = 'w' | |
356 | }, | |
5d4f19d9 JA |
357 | { |
358 | .name = "version", | |
359 | .has_arg = no_argument, | |
360 | .flag = NULL, | |
361 | .val = 'v' | |
362 | }, | |
52724a0e JA |
363 | { |
364 | .name = "version", | |
365 | .has_arg = no_argument, | |
366 | .flag = NULL, | |
57ea8602 | 367 | .val = 'V' |
52724a0e | 368 | }, |
129aa440 | 369 | { |
3f65c585 | 370 | .name = "buffer-size", |
129aa440 JA |
371 | .has_arg = required_argument, |
372 | .flag = NULL, | |
373 | .val = 'b' | |
374 | }, | |
375 | { | |
3f65c585 | 376 | .name = "num-sub-buffers", |
129aa440 JA |
377 | .has_arg = required_argument, |
378 | .flag = NULL, | |
379 | .val = 'n' | |
380 | }, | |
d1d7f15f | 381 | { |
3f65c585 | 382 | .name = "output-dir", |
d1d7f15f JA |
383 | .has_arg = required_argument, |
384 | .flag = NULL, | |
385 | .val = 'D' | |
386 | }, | |
8e86c98a JA |
387 | { |
388 | .name = "listen", | |
389 | .has_arg = no_argument, | |
390 | .flag = NULL, | |
391 | .val = 'l' | |
392 | }, | |
393 | { | |
394 | .name = "host", | |
395 | .has_arg = required_argument, | |
396 | .flag = NULL, | |
397 | .val = 'h' | |
398 | }, | |
399 | { | |
400 | .name = "port", | |
401 | .has_arg = required_argument, | |
402 | .flag = NULL, | |
403 | .val = 'p' | |
404 | }, | |
32f18c48 | 405 | { |
79971f43 | 406 | .name = "no-sendfile", |
32f18c48 JA |
407 | .has_arg = no_argument, |
408 | .flag = NULL, | |
409 | .val = 's' | |
410 | }, | |
71ef8b7c JA |
411 | { |
412 | .name = NULL, | |
413 | } | |
d39c04ca AB |
414 | }; |
415 | ||
3fe0b570 AB |
416 | static char usage_str[] = \ |
417 | "-d <dev> [ -r debugfs path ] [ -o <output> ] [-k ] [ -w time ]\n" \ | |
418 | "[ -a action ] [ -A action mask ] [ -I <devs file> ] [ -v ]\n\n" \ | |
419 | "\t-d Use specified device. May also be given last after options\n" \ | |
420 | "\t-r Path to mounted debugfs, defaults to /sys/kernel/debug\n" \ | |
421 | "\t-o File(s) to send output to\n" \ | |
422 | "\t-D Directory to prepend to output file names\n" \ | |
423 | "\t-k Kill a running trace\n" \ | |
424 | "\t-w Stop after defined time, in seconds\n" \ | |
425 | "\t-a Only trace specified actions. See documentation\n" \ | |
426 | "\t-A Give trace mask as a single value. See documentation\n" \ | |
427 | "\t-b Sub buffer size in KiB\n" \ | |
428 | "\t-n Number of sub buffers\n" \ | |
429 | "\t-l Run in network listen mode (blktrace server)\n" \ | |
430 | "\t-h Run in network client mode, connecting to the given host\n" \ | |
431 | "\t-p Network port to use (default 8462)\n" \ | |
432 | "\t-s Make the network client NOT use sendfile() to transfer data\n" \ | |
433 | "\t-I Add devices found in <devs file>\n" \ | |
434 | "\t-V Print program version info\n\n"; | |
9db17354 | 435 | |
3fe0b570 AB |
436 | static void clear_events(struct pollfd *pfd) |
437 | { | |
438 | pfd->events = 0; | |
439 | pfd->revents = 0; | |
440 | } | |
21f55651 | 441 | |
3fe0b570 AB |
442 | static inline int net_client_use_sendfile(void) |
443 | { | |
444 | return net_mode == Net_client && net_use_sendfile; | |
445 | } | |
21f55651 | 446 | |
3fe0b570 AB |
447 | static inline int net_client_use_send(void) |
448 | { | |
449 | return net_mode == Net_client && !net_use_sendfile; | |
450 | } | |
b9d4294e | 451 | |
3fe0b570 AB |
452 | static inline int use_tracer_devpaths(void) |
453 | { | |
454 | return piped_output || net_client_use_send(); | |
455 | } | |
b9d4294e | 456 | |
3fe0b570 AB |
457 | static inline int in_addr_eq(struct in_addr a, struct in_addr b) |
458 | { | |
459 | return a.s_addr == b.s_addr; | |
460 | } | |
007c233c | 461 | |
3fe0b570 AB |
462 | static inline void pdc_dr_update(struct devpath *dpp, int cpu, int data_read) |
463 | { | |
464 | dpp->stats[cpu].data_read += data_read; | |
465 | } | |
0cc7d25e | 466 | |
3fe0b570 AB |
467 | static inline void pdc_nev_update(struct devpath *dpp, int cpu, int nevents) |
468 | { | |
469 | dpp->stats[cpu].nevents += nevents; | |
470 | } | |
9db17354 | 471 | |
3fe0b570 AB |
472 | static void show_usage(char *prog) |
473 | { | |
474 | fprintf(stderr, "Usage: %s %s %s", prog, blktrace_version, usage_str); | |
475 | } | |
9db17354 | 476 | |
3fe0b570 AB |
477 | static void init_mmap_info(struct mmap_info *mip) |
478 | { | |
479 | mip->buf_size = buf_size; | |
480 | mip->buf_nr = buf_nr; | |
481 | mip->pagesize = pagesize; | |
482 | } | |
b7106311 | 483 | |
3fe0b570 AB |
484 | static void net_close_connection(int *fd) |
485 | { | |
486 | shutdown(*fd, SHUT_RDWR); | |
487 | close(*fd); | |
488 | *fd = -1; | |
489 | } | |
ff11d54c | 490 | |
3fe0b570 AB |
491 | static void dpp_free(struct devpath *dpp) |
492 | { | |
493 | if (dpp->stats) | |
494 | free(dpp->stats); | |
495 | if (dpp->ios) | |
496 | free(dpp->ios); | |
497 | if (dpp->path) | |
498 | free(dpp->path); | |
499 | if (dpp->buts_name) | |
500 | free(dpp->buts_name); | |
501 | free(dpp); | |
502 | } | |
d0ca268b | 503 | |
3fe0b570 AB |
504 | static int lock_on_cpu(int cpu) |
505 | { | |
506 | cpu_set_t cpu_mask; | |
ff11d54c | 507 | |
3fe0b570 AB |
508 | CPU_ZERO(&cpu_mask); |
509 | CPU_SET(cpu, &cpu_mask); | |
510 | if (sched_setaffinity(getpid(), sizeof(cpu_mask), &cpu_mask) < 0) | |
511 | return errno; | |
d0ca268b | 512 | |
3fe0b570 AB |
513 | return 0; |
514 | } | |
e7c9f3ff | 515 | |
3fe0b570 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; | |
d39c04ca | 522 | |
3fe0b570 AB |
523 | gettimeofday(&now, NULL); |
524 | tsp->tv_sec = now.tv_sec; | |
525 | tsp->tv_nsec = 1000L * now.tv_usec; | |
e7c9f3ff | 526 | |
3fe0b570 AB |
527 | tsp->tv_nsec += (delta_msec * 1000000L); |
528 | if (tsp->tv_nsec > 1000000000L) { | |
529 | long secs = tsp->tv_nsec / 1000000000L; | |
eb3c8108 | 530 | |
3fe0b570 AB |
531 | tsp->tv_sec += secs; |
532 | tsp->tv_nsec -= (secs * 1000000000L); | |
533 | } | |
534 | } | |
a3e4d330 | 535 | |
3fe0b570 AB |
536 | static int increase_limit(int resource, rlim_t increase) |
537 | { | |
538 | struct rlimit rlim; | |
539 | int save_errno = errno; | |
8e86c98a | 540 | |
3fe0b570 AB |
541 | if (!getrlimit(resource, &rlim)) { |
542 | rlim.rlim_cur += increase; | |
543 | if (rlim.rlim_cur >= rlim.rlim_max) | |
544 | rlim.rlim_max = rlim.rlim_cur + increase; | |
72ca8801 | 545 | |
3fe0b570 AB |
546 | if (!setrlimit(resource, &rlim)) |
547 | return 1; | |
548 | } | |
99c1f5ab | 549 | |
3fe0b570 AB |
550 | errno = save_errno; |
551 | return 0; | |
552 | } | |
e0a1988b | 553 | |
3fe0b570 AB |
554 | static int handle_open_failure(void) |
555 | { | |
556 | if (errno == ENFILE || errno == EMFILE) | |
557 | return increase_limit(RLIMIT_NOFILE, 16); | |
558 | return 0; | |
559 | } | |
99c1f5ab | 560 | |
3fe0b570 AB |
561 | static int handle_mem_failure(size_t length) |
562 | { | |
563 | if (errno == ENFILE) | |
564 | return handle_open_failure(); | |
565 | else if (errno == ENOMEM) | |
566 | return increase_limit(RLIMIT_MEMLOCK, 2 * length); | |
567 | return 0; | |
568 | } | |
99c1f5ab | 569 | |
3fe0b570 AB |
570 | static FILE *my_fopen(const char *path, const char *mode) |
571 | { | |
572 | FILE *fp; | |
8e86c98a | 573 | |
3fe0b570 AB |
574 | do { |
575 | fp = fopen(path, mode); | |
576 | } while (fp == NULL && handle_open_failure()); | |
8e86c98a | 577 | |
3fe0b570 AB |
578 | return fp; |
579 | } | |
8e86c98a | 580 | |
3fe0b570 AB |
581 | static int my_open(const char *path, int flags) |
582 | { | |
583 | int fd; | |
8e86c98a | 584 | |
3fe0b570 AB |
585 | do { |
586 | fd = open(path, flags); | |
587 | } while (fd < 0 && handle_open_failure()); | |
e0a1988b | 588 | |
3fe0b570 AB |
589 | return fd; |
590 | } | |
ff11d54c | 591 | |
3fe0b570 AB |
592 | static int my_socket(int domain, int type, int protocol) |
593 | { | |
594 | int fd; | |
ff11d54c | 595 | |
3fe0b570 AB |
596 | do { |
597 | fd = socket(domain, type, protocol); | |
598 | } while (fd < 0 && handle_open_failure()); | |
8e86c98a | 599 | |
3fe0b570 AB |
600 | return fd; |
601 | } | |
602 | ||
d5302b03 AB |
603 | static int my_accept(int sockfd, struct sockaddr *addr, socklen_t *addrlen) |
604 | { | |
605 | int fd; | |
606 | ||
607 | do { | |
608 | fd = accept(sockfd, addr, addrlen); | |
609 | } while (fd < 0 && handle_open_failure()); | |
610 | ||
611 | return fd; | |
612 | } | |
613 | ||
3fe0b570 AB |
614 | static void *my_mmap(void *addr, size_t length, int prot, int flags, int fd, |
615 | off_t offset) | |
69dd57c2 | 616 | { |
3fe0b570 | 617 | void *new; |
69dd57c2 | 618 | |
3fe0b570 AB |
619 | do { |
620 | new = mmap(addr, length, prot, flags, fd, offset); | |
621 | } while (new == MAP_FAILED && handle_mem_failure(length)); | |
622 | ||
623 | return new; | |
624 | } | |
625 | ||
626 | static int my_mlock(const void *addr, size_t len) | |
627 | { | |
628 | int ret; | |
629 | ||
630 | do { | |
631 | ret = mlock(addr, len); | |
632 | } while (ret < 0 && handle_mem_failure(len)); | |
633 | ||
634 | return ret; | |
635 | } | |
636 | ||
637 | static int __stop_trace(int fd) | |
638 | { | |
639 | /* | |
640 | * Should be stopped, don't complain if it isn't | |
641 | */ | |
642 | ioctl(fd, BLKTRACESTOP); | |
643 | return ioctl(fd, BLKTRACETEARDOWN); | |
644 | } | |
645 | ||
646 | static int write_data(char *buf, int len) | |
647 | { | |
648 | int ret; | |
649 | ||
650 | rewrite: | |
651 | ret = fwrite(buf, len, 1, pfp); | |
652 | if (ferror(pfp) || ret != 1) { | |
653 | if (errno == EINTR) { | |
654 | clearerr(pfp); | |
655 | goto rewrite; | |
656 | } | |
657 | ||
658 | if (!piped_output || (errno != EPIPE && errno != EBADF)) { | |
659 | fprintf(stderr, "write(%d) failed: %d/%s\n", | |
660 | len, errno, strerror(errno)); | |
69dd57c2 | 661 | } |
3fe0b570 | 662 | goto err; |
69dd57c2 AB |
663 | } |
664 | ||
3fe0b570 | 665 | fflush(pfp); |
69dd57c2 | 666 | return 0; |
3fe0b570 AB |
667 | |
668 | err: | |
669 | clearerr(pfp); | |
670 | return 1; | |
69dd57c2 AB |
671 | } |
672 | ||
673 | /* | |
3fe0b570 | 674 | * Returns the number of bytes read (successfully) |
69dd57c2 | 675 | */ |
3fe0b570 | 676 | static int __net_recv_data(int fd, void *buf, unsigned int len) |
69dd57c2 | 677 | { |
3fe0b570 AB |
678 | unsigned int bytes_left = len; |
679 | ||
680 | while (bytes_left && !done) { | |
681 | int ret = recv(fd, buf, bytes_left, MSG_WAITALL); | |
682 | ||
683 | if (ret == 0) | |
684 | break; | |
685 | else if (ret < 0) { | |
686 | if (errno != EAGAIN) { | |
687 | perror("server: net_recv_data: recv failed"); | |
688 | break; | |
689 | } else | |
690 | break; | |
691 | } else { | |
692 | buf += ret; | |
693 | bytes_left -= ret; | |
694 | } | |
695 | } | |
69dd57c2 | 696 | |
3fe0b570 | 697 | return len - bytes_left; |
69dd57c2 AB |
698 | } |
699 | ||
3fe0b570 | 700 | static int net_recv_data(int fd, void *buf, unsigned int len) |
8e86c98a | 701 | { |
3fe0b570 AB |
702 | return __net_recv_data(fd, buf, len); |
703 | } | |
7035d92d | 704 | |
3fe0b570 AB |
705 | /* |
706 | * Returns number of bytes written | |
707 | */ | |
708 | static int net_send_data(int fd, void *buf, unsigned int buf_len) | |
709 | { | |
710 | int ret; | |
711 | unsigned int bytes_left = buf_len; | |
712 | ||
713 | while (bytes_left) { | |
714 | ret = send(fd, buf, bytes_left, 0); | |
715 | if (ret < 0) { | |
716 | perror("send"); | |
717 | break; | |
718 | } | |
719 | ||
720 | buf += ret; | |
721 | bytes_left -= ret; | |
7035d92d JA |
722 | } |
723 | ||
3fe0b570 | 724 | return buf_len - bytes_left; |
8e86c98a JA |
725 | } |
726 | ||
3fe0b570 | 727 | static int net_send_header(int fd, int cpu, char *buts_name, int len) |
eb3c8108 | 728 | { |
3fe0b570 | 729 | struct blktrace_net_hdr hdr; |
eb3c8108 | 730 | |
3fe0b570 | 731 | memset(&hdr, 0, sizeof(hdr)); |
eb3c8108 | 732 | |
3fe0b570 AB |
733 | hdr.magic = BLK_IO_TRACE_MAGIC; |
734 | strncpy(hdr.buts_name, buts_name, sizeof(hdr.buts_name)); | |
735 | hdr.buts_name[sizeof(hdr.buts_name)-1] = '\0'; | |
736 | hdr.cpu = cpu; | |
737 | hdr.max_cpus = ncpus; | |
738 | hdr.len = len; | |
739 | hdr.cl_id = getpid(); | |
740 | hdr.buf_size = buf_size; | |
741 | hdr.buf_nr = buf_nr; | |
742 | hdr.page_size = pagesize; | |
eb3c8108 | 743 | |
3fe0b570 AB |
744 | return net_send_data(fd, &hdr, sizeof(hdr)) != sizeof(hdr); |
745 | } | |
eb3c8108 | 746 | |
3fe0b570 AB |
747 | static void net_send_open_close(int fd, int cpu, char *buts_name, int len) |
748 | { | |
749 | struct blktrace_net_hdr ret_hdr; | |
eb3c8108 | 750 | |
3fe0b570 AB |
751 | net_send_header(fd, cpu, buts_name, len); |
752 | net_recv_data(fd, &ret_hdr, sizeof(ret_hdr)); | |
753 | } | |
eb3c8108 | 754 | |
3fe0b570 AB |
755 | static void net_send_open(int fd, int cpu, char *buts_name) |
756 | { | |
757 | net_send_open_close(fd, cpu, buts_name, 0); | |
eb3c8108 JA |
758 | } |
759 | ||
3fe0b570 | 760 | static void net_send_close(int fd, char *buts_name, int drops) |
d0ca268b | 761 | { |
3fe0b570 AB |
762 | /* |
763 | * Overload CPU w/ number of drops | |
764 | * | |
765 | * XXX: Need to clear/set done around call - done=1 (which | |
766 | * is true here) stops reads from happening... :-( | |
767 | */ | |
768 | done = 0; | |
769 | net_send_open_close(fd, drops, buts_name, 1); | |
770 | done = 1; | |
771 | } | |
d0ca268b | 772 | |
3fe0b570 AB |
773 | static void ack_open_close(int fd, char *buts_name) |
774 | { | |
775 | net_send_header(fd, 0, buts_name, 2); | |
776 | } | |
d0ca268b | 777 | |
3fe0b570 AB |
778 | static void net_send_drops(int fd) |
779 | { | |
780 | struct list_head *p; | |
ed71a31e | 781 | |
3fe0b570 AB |
782 | __list_for_each(p, &devpaths) { |
783 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
784 | ||
785 | net_send_close(fd, dpp->buts_name, dpp->drops); | |
d0ca268b | 786 | } |
3fe0b570 | 787 | } |
d0ca268b | 788 | |
3fe0b570 AB |
789 | /* |
790 | * Returns: | |
791 | * 0: "EOF" | |
792 | * 1: OK | |
793 | * -1: Error | |
794 | */ | |
795 | static int net_get_header(struct cl_conn *nc, struct blktrace_net_hdr *bnh) | |
796 | { | |
797 | int bytes_read; | |
798 | int fl = fcntl(nc->fd, F_GETFL); | |
799 | ||
800 | fcntl(nc->fd, F_SETFL, fl | O_NONBLOCK); | |
801 | bytes_read = __net_recv_data(nc->fd, bnh, sizeof(*bnh)); | |
802 | fcntl(nc->fd, F_SETFL, fl & ~O_NONBLOCK); | |
803 | ||
804 | if (bytes_read == sizeof(*bnh)) | |
805 | return 1; | |
806 | else if (bytes_read == 0) | |
807 | return 0; | |
808 | return -1; | |
d0ca268b JA |
809 | } |
810 | ||
e58f3937 | 811 | static int net_setup_addr(void) |
d0ca268b | 812 | { |
e58f3937 | 813 | struct sockaddr_in *addr = &hostname_addr; |
cf9208ea | 814 | |
e58f3937 AB |
815 | memset(addr, 0, sizeof(*addr)); |
816 | addr->sin_family = AF_INET; | |
817 | addr->sin_port = htons(net_port); | |
3fe0b570 | 818 | |
e58f3937 AB |
819 | if (inet_aton(hostname, &addr->sin_addr) != 1) { |
820 | struct hostent *hent; | |
821 | retry: | |
822 | hent = gethostbyname(hostname); | |
3fe0b570 | 823 | if (!hent) { |
e58f3937 AB |
824 | if (h_errno == TRY_AGAIN) { |
825 | usleep(100); | |
826 | goto retry; | |
827 | } else if (h_errno == NO_RECOVERY) { | |
828 | fprintf(stderr, "gethostbyname(%s)" | |
829 | "non-recoverable error encountered\n", | |
830 | hostname); | |
831 | } else { | |
832 | /* | |
833 | * HOST_NOT_FOUND, NO_ADDRESS or NO_DATA | |
834 | */ | |
835 | fprintf(stderr, "Host %s not found\n", | |
836 | hostname); | |
837 | } | |
3fe0b570 AB |
838 | return 1; |
839 | } | |
840 | ||
e58f3937 | 841 | memcpy(&addr->sin_addr, hent->h_addr, 4); |
3fe0b570 AB |
842 | strcpy(hostname, hent->h_name); |
843 | } | |
7035d92d | 844 | |
e58f3937 AB |
845 | return 0; |
846 | } | |
847 | ||
848 | static int net_setup_client(void) | |
849 | { | |
850 | int fd; | |
851 | struct sockaddr_in *addr = &hostname_addr; | |
852 | ||
3fe0b570 AB |
853 | fd = my_socket(AF_INET, SOCK_STREAM, 0); |
854 | if (fd < 0) { | |
855 | perror("client: socket"); | |
856 | return -1; | |
857 | } | |
cf9208ea | 858 | |
e58f3937 | 859 | if (connect(fd, (struct sockaddr *)addr, sizeof(*addr)) < 0) { |
3fe0b570 AB |
860 | if (errno == ECONNREFUSED) |
861 | fprintf(stderr, | |
862 | "\nclient: Connection to %s refused, " | |
863 | "perhaps the server is not started?\n\n", | |
864 | hostname); | |
865 | else | |
866 | perror("client: connect"); | |
867 | close(fd); | |
868 | return -1; | |
707b0914 | 869 | } |
3fe0b570 AB |
870 | |
871 | return fd; | |
d0ca268b JA |
872 | } |
873 | ||
3fe0b570 | 874 | static int open_client_connections(void) |
e7c9f3ff | 875 | { |
3fe0b570 | 876 | int cpu; |
e7c9f3ff | 877 | |
3fe0b570 AB |
878 | cl_fds = calloc(ncpus, sizeof(*cl_fds)); |
879 | for (cpu = 0; cpu < ncpus; cpu++) { | |
880 | cl_fds[cpu] = net_setup_client(); | |
881 | if (cl_fds[cpu] < 0) | |
882 | goto err; | |
eb3c8108 | 883 | } |
3fe0b570 AB |
884 | return 0; |
885 | ||
886 | err: | |
887 | while (cpu > 0) | |
888 | close(cl_fds[cpu--]); | |
889 | free(cl_fds); | |
890 | return 1; | |
e7c9f3ff NS |
891 | } |
892 | ||
3fe0b570 | 893 | static void close_client_connections(void) |
eb3c8108 | 894 | { |
3fe0b570 AB |
895 | if (cl_fds) { |
896 | int cpu, *fdp; | |
eb3c8108 | 897 | |
3fe0b570 AB |
898 | for (cpu = 0, fdp = cl_fds; cpu < ncpus; cpu++, fdp++) { |
899 | if (*fdp >= 0) { | |
900 | net_send_drops(*fdp); | |
901 | net_close_connection(fdp); | |
902 | } | |
7934e668 | 903 | } |
3fe0b570 | 904 | free(cl_fds); |
ff11d54c | 905 | } |
eb3c8108 JA |
906 | } |
907 | ||
3fe0b570 | 908 | static void setup_buts(void) |
d0ca268b | 909 | { |
3fe0b570 | 910 | struct list_head *p; |
bbabf03a | 911 | |
3fe0b570 AB |
912 | __list_for_each(p, &devpaths) { |
913 | struct blk_user_trace_setup buts; | |
914 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
ae9f71b3 | 915 | |
3fe0b570 AB |
916 | memset(&buts, 0, sizeof(buts)); |
917 | buts.buf_size = buf_size; | |
918 | buts.buf_nr = buf_nr; | |
919 | buts.act_mask = act_mask; | |
920 | ||
921 | if (ioctl(dpp->fd, BLKTRACESETUP, &buts) < 0) { | |
922 | fprintf(stderr, "BLKTRACESETUP(2) %s failed: %d/%s\n", | |
923 | dpp->path, errno, strerror(errno)); | |
9db17354 | 924 | continue; |
3fe0b570 AB |
925 | } else if (ioctl(dpp->fd, BLKTRACESTART) < 0) { |
926 | fprintf(stderr, "BLKTRACESTART %s failed: %d/%s\n", | |
927 | dpp->path, errno, strerror(errno)); | |
9db17354 | 928 | continue; |
bbabf03a | 929 | } |
8e86c98a | 930 | |
3fe0b570 AB |
931 | dpp->ncpus = ncpus; |
932 | dpp->buts_name = strdup(buts.name); | |
933 | if (dpp->stats) | |
934 | free(dpp->stats); | |
935 | dpp->stats = calloc(dpp->ncpus, sizeof(*dpp->stats)); | |
936 | memset(dpp->stats, 0, dpp->ncpus * sizeof(*dpp->stats)); | |
937 | } | |
8a43bac5 JA |
938 | } |
939 | ||
3fe0b570 | 940 | static int get_drops(struct devpath *dpp) |
8e86c98a | 941 | { |
3fe0b570 AB |
942 | int fd, drops = 0; |
943 | char fn[MAXPATHLEN + 64], tmp[256]; | |
8e86c98a | 944 | |
3fe0b570 AB |
945 | snprintf(fn, sizeof(fn), "%s/block/%s/dropped", debugfs_path, |
946 | dpp->buts_name); | |
8e86c98a | 947 | |
3fe0b570 AB |
948 | fd = my_open(fn, O_RDONLY); |
949 | if (fd < 0) { | |
950 | /* | |
951 | * This may be ok: the kernel may not support | |
952 | * dropped counts. | |
953 | */ | |
954 | if (errno != ENOENT) | |
955 | fprintf(stderr, "Could not open %s: %d/%s\n", | |
956 | fn, errno, strerror(errno)); | |
957 | return 0; | |
958 | } else if (read(fd, tmp, sizeof(tmp)) < 0) { | |
959 | fprintf(stderr, "Could not read %s: %d/%s\n", | |
960 | fn, errno, strerror(errno)); | |
961 | } else | |
962 | drops = atoi(tmp); | |
963 | close(fd); | |
8e86c98a | 964 | |
3fe0b570 | 965 | return drops; |
8e86c98a JA |
966 | } |
967 | ||
3fe0b570 | 968 | static void get_all_drops(void) |
a3e4d330 | 969 | { |
3fe0b570 | 970 | struct list_head *p; |
21f55651 | 971 | |
3fe0b570 AB |
972 | __list_for_each(p, &devpaths) { |
973 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
974 | dpp->drops = get_drops(dpp); | |
21f55651 | 975 | } |
9db17354 | 976 | } |
eb3c8108 | 977 | |
3fe0b570 | 978 | static inline struct trace_buf *alloc_trace_buf(int cpu, int bufsize) |
9db17354 | 979 | { |
3fe0b570 | 980 | struct trace_buf *tbp; |
21f55651 | 981 | |
3fe0b570 AB |
982 | tbp = malloc(sizeof(*tbp) + bufsize); |
983 | INIT_LIST_HEAD(&tbp->head); | |
984 | tbp->len = 0; | |
985 | tbp->buf = (void *)(tbp + 1); | |
986 | tbp->cpu = cpu; | |
987 | tbp->dpp = NULL; /* Will be set when tbp is added */ | |
21f55651 | 988 | |
3fe0b570 | 989 | return tbp; |
a3e4d330 JA |
990 | } |
991 | ||
3fe0b570 | 992 | static void free_tracer_heads(struct devpath *dpp) |
b7106311 | 993 | { |
3fe0b570 AB |
994 | int cpu; |
995 | struct tracer_devpath_head *hd; | |
b7106311 | 996 | |
3fe0b570 AB |
997 | for (cpu = 0, hd = dpp->heads; cpu < ncpus; cpu++, hd++) { |
998 | if (hd->prev) | |
999 | free(hd->prev); | |
1000 | pthread_mutex_destroy(&hd->mutex); | |
1001 | } | |
1002 | free(dpp->heads); | |
1003 | } | |
b7106311 | 1004 | |
3fe0b570 AB |
1005 | static int setup_tracer_devpaths(void) |
1006 | { | |
1007 | struct list_head *p; | |
b7106311 | 1008 | |
3fe0b570 AB |
1009 | if (net_client_use_send()) |
1010 | if (open_client_connections()) | |
1011 | return 1; | |
b7106311 | 1012 | |
3fe0b570 AB |
1013 | __list_for_each(p, &devpaths) { |
1014 | int cpu; | |
1015 | struct tracer_devpath_head *hd; | |
1016 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
b7106311 | 1017 | |
3fe0b570 AB |
1018 | dpp->heads = calloc(ncpus, sizeof(struct tracer_devpath_head)); |
1019 | for (cpu = 0, hd = dpp->heads; cpu < ncpus; cpu++, hd++) { | |
1020 | INIT_LIST_HEAD(&hd->head); | |
1021 | pthread_mutex_init(&hd->mutex, NULL); | |
1022 | hd->prev = NULL; | |
1023 | } | |
b7106311 JA |
1024 | } |
1025 | ||
3fe0b570 | 1026 | return 0; |
b7106311 JA |
1027 | } |
1028 | ||
3fe0b570 AB |
1029 | static inline void add_trace_buf(struct devpath *dpp, int cpu, |
1030 | struct trace_buf **tbpp) | |
18eed2a7 | 1031 | { |
3fe0b570 AB |
1032 | struct trace_buf *tbp = *tbpp; |
1033 | struct tracer_devpath_head *hd = &dpp->heads[cpu]; | |
18eed2a7 | 1034 | |
3fe0b570 | 1035 | tbp->dpp = dpp; |
2f064793 | 1036 | |
3fe0b570 AB |
1037 | pthread_mutex_lock(&hd->mutex); |
1038 | list_add_tail(&tbp->head, &hd->head); | |
1039 | pthread_mutex_unlock(&hd->mutex); | |
18eed2a7 | 1040 | |
3fe0b570 | 1041 | *tbpp = alloc_trace_buf(cpu, buf_size); |
18eed2a7 JA |
1042 | } |
1043 | ||
3fe0b570 | 1044 | static inline void incr_entries(int entries_handled) |
a3e4d330 | 1045 | { |
3fe0b570 AB |
1046 | pthread_mutex_lock(&dp_mutex); |
1047 | if (dp_entries == 0) | |
1048 | pthread_cond_signal(&dp_cond); | |
1049 | dp_entries += entries_handled; | |
1050 | pthread_mutex_unlock(&dp_mutex); | |
a3e4d330 JA |
1051 | } |
1052 | ||
3fe0b570 | 1053 | static int add_devpath(char *path) |
8e86c98a | 1054 | { |
3fe0b570 AB |
1055 | int fd; |
1056 | struct devpath *dpp; | |
1057 | ||
8e86c98a | 1058 | /* |
3fe0b570 | 1059 | * Verify device is valid before going too far |
8e86c98a | 1060 | */ |
3fe0b570 AB |
1061 | fd = my_open(path, O_RDONLY | O_NONBLOCK); |
1062 | if (fd < 0) { | |
1063 | fprintf(stderr, "Invalid path %s specified: %d/%s\n", | |
1064 | path, errno, strerror(errno)); | |
1065 | return 1; | |
1066 | } | |
8e86c98a | 1067 | |
3fe0b570 AB |
1068 | dpp = malloc(sizeof(*dpp)); |
1069 | memset(dpp, 0, sizeof(*dpp)); | |
1070 | dpp->path = strdup(path); | |
1071 | dpp->fd = fd; | |
1072 | dpp->idx = ndevs++; | |
1073 | list_add_tail(&dpp->head, &devpaths); | |
8e86c98a | 1074 | |
3fe0b570 | 1075 | return 0; |
8e86c98a JA |
1076 | } |
1077 | ||
3fe0b570 | 1078 | static void rel_devpaths(void) |
a3e4d330 | 1079 | { |
3fe0b570 | 1080 | struct list_head *p, *q; |
a3e4d330 | 1081 | |
3fe0b570 AB |
1082 | list_for_each_safe(p, q, &devpaths) { |
1083 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
a3e4d330 | 1084 | |
3fe0b570 AB |
1085 | list_del(&dpp->head); |
1086 | __stop_trace(dpp->fd); | |
1087 | close(dpp->fd); | |
a3e4d330 | 1088 | |
3fe0b570 AB |
1089 | if (dpp->heads) |
1090 | free_tracer_heads(dpp); | |
a3e4d330 | 1091 | |
3fe0b570 AB |
1092 | dpp_free(dpp); |
1093 | ndevs--; | |
b7106311 | 1094 | } |
8e86c98a | 1095 | } |
b7106311 | 1096 | |
3fe0b570 | 1097 | static int flush_subbuf_net(struct trace_buf *tbp) |
8e86c98a | 1098 | { |
3fe0b570 AB |
1099 | int fd = cl_fds[tbp->cpu]; |
1100 | struct devpath *dpp = tbp->dpp; | |
b7106311 | 1101 | |
3fe0b570 AB |
1102 | if (net_send_header(fd, tbp->cpu, dpp->buts_name, tbp->len)) |
1103 | return 1; | |
8e86c98a | 1104 | |
3fe0b570 AB |
1105 | if (net_send_data(fd, tbp->buf, tbp->len) != tbp->len) |
1106 | return 1; | |
a3e4d330 | 1107 | |
8e86c98a | 1108 | return 0; |
a3e4d330 JA |
1109 | } |
1110 | ||
3fe0b570 AB |
1111 | static int |
1112 | handle_list_net(__attribute__((__unused__))struct tracer_devpath_head *hd, | |
1113 | struct list_head *list) | |
8e86c98a | 1114 | { |
3fe0b570 AB |
1115 | struct trace_buf *tbp; |
1116 | struct list_head *p, *q; | |
1117 | int entries_handled = 0; | |
8e86c98a | 1118 | |
3fe0b570 AB |
1119 | list_for_each_safe(p, q, list) { |
1120 | tbp = list_entry(p, struct trace_buf, head); | |
8e86c98a | 1121 | |
3fe0b570 AB |
1122 | list_del(&tbp->head); |
1123 | entries_handled++; | |
6a752c90 | 1124 | |
3fe0b570 AB |
1125 | if (cl_fds[tbp->cpu] >= 0) { |
1126 | if (flush_subbuf_net(tbp)) { | |
1127 | close(cl_fds[tbp->cpu]); | |
1128 | cl_fds[tbp->cpu] = -1; | |
1129 | } | |
1130 | } | |
7ab2f837 | 1131 | |
3fe0b570 | 1132 | free(tbp); |
7934e668 JA |
1133 | } |
1134 | ||
3fe0b570 | 1135 | return entries_handled; |
6a752c90 JA |
1136 | } |
1137 | ||
3fe0b570 AB |
1138 | static int handle_list_file(struct tracer_devpath_head *hd, |
1139 | struct list_head *list) | |
f6fead25 | 1140 | { |
3fe0b570 AB |
1141 | int off, t_len, nevents; |
1142 | struct blk_io_trace *t; | |
1143 | struct list_head *p, *q; | |
1144 | int entries_handled = 0; | |
1145 | struct trace_buf *tbp, *prev; | |
11629347 | 1146 | |
3fe0b570 AB |
1147 | prev = hd->prev; |
1148 | list_for_each_safe(p, q, list) { | |
1149 | tbp = list_entry(p, struct trace_buf, head); | |
1150 | list_del(&tbp->head); | |
1151 | entries_handled++; | |
18eed2a7 | 1152 | |
3fe0b570 AB |
1153 | /* |
1154 | * If there was some leftover before, tack this new | |
1155 | * entry onto the tail of the previous one. | |
1156 | */ | |
1157 | if (prev) { | |
1158 | unsigned long tot_len; | |
1159 | struct trace_buf *tmp = tbp; | |
1160 | ||
1161 | tbp = prev; | |
1162 | prev = NULL; | |
1163 | ||
1164 | tot_len = tbp->len + tmp->len; | |
1165 | if (tot_len > buf_size) { | |
1166 | /* | |
1167 | * tbp->head isn't connected (it was 'prev' | |
1168 | * so it had been taken off of the list | |
1169 | * before). Therefore, we can realloc | |
1170 | * the whole structures, as the other fields | |
1171 | * are "static". | |
1172 | */ | |
1173 | tbp = realloc(tbp->buf, sizeof(*tbp) + tot_len); | |
1174 | tbp->buf = (void *)(tbp + 1); | |
1175 | } | |
32f18c48 | 1176 | |
3fe0b570 AB |
1177 | memcpy(tbp->buf + tbp->len, tmp->buf, tmp->len); |
1178 | tbp->len = tot_len; | |
32f18c48 | 1179 | |
3fe0b570 AB |
1180 | free(tmp); |
1181 | } | |
ff11d54c | 1182 | |
3fe0b570 AB |
1183 | /* |
1184 | * See how many whole traces there are - send them | |
1185 | * all out in one go. | |
1186 | */ | |
1187 | off = 0; | |
1188 | nevents = 0; | |
1189 | while (off + (int)sizeof(*t) <= tbp->len) { | |
1190 | t = (struct blk_io_trace *)(tbp->buf + off); | |
1191 | t_len = sizeof(*t) + t->pdu_len; | |
1192 | if (off + t_len > tbp->len) | |
1193 | break; | |
ff11d54c | 1194 | |
3fe0b570 AB |
1195 | off += t_len; |
1196 | nevents++; | |
1197 | } | |
1198 | if (nevents) | |
1199 | pdc_nev_update(tbp->dpp, tbp->cpu, nevents); | |
4aeec019 | 1200 | |
3fe0b570 AB |
1201 | /* |
1202 | * Write any full set of traces, any remaining data is kept | |
1203 | * for the next pass. | |
1204 | */ | |
1205 | if (off) { | |
1206 | if (write_data(tbp->buf, off) || off == tbp->len) | |
1207 | free(tbp); | |
1208 | else { | |
1209 | /* | |
1210 | * Move valid data to beginning of buffer | |
1211 | */ | |
1212 | tbp->len -= off; | |
1213 | memmove(tbp->buf, tbp->buf + off, tbp->len); | |
1214 | prev = tbp; | |
1215 | } | |
1216 | } else | |
1217 | prev = tbp; | |
ff11d54c | 1218 | } |
3fe0b570 | 1219 | hd->prev = prev; |
ff11d54c | 1220 | |
3fe0b570 | 1221 | return entries_handled; |
ff11d54c TZ |
1222 | } |
1223 | ||
3fe0b570 | 1224 | static void __process_trace_bufs(void) |
8a43bac5 | 1225 | { |
3fe0b570 AB |
1226 | int cpu; |
1227 | struct list_head *p; | |
1228 | struct list_head list; | |
1229 | int handled = 0; | |
1230 | ||
1231 | __list_for_each(p, &devpaths) { | |
1232 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1233 | struct tracer_devpath_head *hd = dpp->heads; | |
1234 | ||
1235 | for (cpu = 0; cpu < ncpus; cpu++, hd++) { | |
1236 | pthread_mutex_lock(&hd->mutex); | |
1237 | if (list_empty(&hd->head)) { | |
1238 | pthread_mutex_unlock(&hd->mutex); | |
1239 | continue; | |
1240 | } | |
8a43bac5 | 1241 | |
3fe0b570 AB |
1242 | list_replace_init(&hd->head, &list); |
1243 | pthread_mutex_unlock(&hd->mutex); | |
6480258a | 1244 | |
3fe0b570 AB |
1245 | handled += handle_list(hd, &list); |
1246 | } | |
d0ca268b JA |
1247 | } |
1248 | ||
3fe0b570 AB |
1249 | if (handled) { |
1250 | pthread_mutex_lock(&dp_mutex); | |
1251 | dp_entries -= handled; | |
1252 | pthread_mutex_unlock(&dp_mutex); | |
1253 | } | |
8a43bac5 JA |
1254 | } |
1255 | ||
3fe0b570 | 1256 | static void process_trace_bufs(void) |
8a43bac5 | 1257 | { |
3fe0b570 AB |
1258 | while (!done) { |
1259 | pthread_mutex_lock(&dp_mutex); | |
1260 | while (!done && dp_entries == 0) { | |
1261 | struct timespec ts; | |
d0ca268b | 1262 | |
3fe0b570 AB |
1263 | make_timespec(&ts, 50); |
1264 | pthread_cond_timedwait(&dp_cond, &dp_mutex, &ts); | |
9cfa6c2b | 1265 | } |
3fe0b570 | 1266 | pthread_mutex_unlock(&dp_mutex); |
3a9d6c13 | 1267 | |
3fe0b570 | 1268 | __process_trace_bufs(); |
3a9d6c13 | 1269 | } |
3fe0b570 | 1270 | } |
3a9d6c13 | 1271 | |
3fe0b570 AB |
1272 | static void clean_trace_bufs(void) |
1273 | { | |
3a9d6c13 | 1274 | /* |
3fe0b570 AB |
1275 | * No mutex needed here: we're only reading from the lists, |
1276 | * tracers are done | |
3a9d6c13 | 1277 | */ |
3fe0b570 AB |
1278 | while (dp_entries) |
1279 | __process_trace_bufs(); | |
1280 | } | |
4b5db44a | 1281 | |
3fe0b570 AB |
1282 | static inline void read_err(int cpu, char *ifn) |
1283 | { | |
1284 | if (errno != EAGAIN) | |
1285 | fprintf(stderr, "Thread %d failed read of %s: %d/%s\n", | |
1286 | cpu, ifn, errno, strerror(errno)); | |
4b5db44a JA |
1287 | } |
1288 | ||
3fe0b570 | 1289 | static int net_sendfile(struct io_info *iop) |
d5396421 | 1290 | { |
3fe0b570 | 1291 | int ret; |
d5396421 | 1292 | |
3fe0b570 AB |
1293 | ret = sendfile(iop->ofd, iop->ifd, NULL, iop->ready); |
1294 | if (ret < 0) { | |
1295 | perror("sendfile"); | |
1296 | return 1; | |
1297 | } else if (ret < (int)iop->ready) { | |
1298 | fprintf(stderr, "short sendfile send (%d of %d)\n", | |
1299 | ret, iop->ready); | |
1300 | return 1; | |
1301 | } | |
91816d54 | 1302 | |
9db17354 | 1303 | return 0; |
91816d54 JA |
1304 | } |
1305 | ||
3fe0b570 | 1306 | static inline int net_sendfile_data(struct tracer *tp, struct io_info *iop) |
d0ca268b | 1307 | { |
3fe0b570 | 1308 | struct devpath *dpp = iop->dpp; |
d0ca268b | 1309 | |
3fe0b570 AB |
1310 | if (net_send_header(iop->ofd, tp->cpu, dpp->buts_name, iop->ready)) |
1311 | return 1; | |
1312 | return net_sendfile(iop); | |
1313 | } | |
d0ca268b | 1314 | |
3fe0b570 AB |
1315 | static int handle_pfds_netclient(struct tracer *tp, int nevs, int force_read) |
1316 | { | |
1317 | struct stat sb; | |
1318 | int i, nentries = 0; | |
1319 | struct pdc_stats *sp; | |
1320 | struct pollfd *pfd = tp->pfds; | |
1321 | struct io_info *iop = tp->ios; | |
1322 | ||
1323 | for (i = 0; nevs > 0 && i < ndevs; i++, pfd++, iop++, sp++) { | |
1324 | if (pfd->revents & POLLIN || force_read) { | |
1325 | if (fstat(iop->ifd, &sb) < 0) { | |
1326 | perror(iop->ifn); | |
1327 | pfd->events = 0; | |
1328 | } else if (sb.st_size > (off_t)iop->data_queued) { | |
1329 | iop->ready = sb.st_size - iop->data_queued; | |
1330 | iop->data_queued = sb.st_size; | |
1331 | if (!net_sendfile_data(tp, iop)) { | |
1332 | pdc_dr_update(iop->dpp, tp->cpu, | |
1333 | iop->ready); | |
1334 | nentries++; | |
1335 | } else | |
1336 | clear_events(pfd); | |
9db17354 | 1337 | } |
3fe0b570 | 1338 | nevs--; |
9db17354 | 1339 | } |
d0ca268b JA |
1340 | } |
1341 | ||
3fe0b570 AB |
1342 | if (nentries) |
1343 | incr_entries(nentries); | |
1344 | ||
1345 | return nentries; | |
d0ca268b JA |
1346 | } |
1347 | ||
3fe0b570 | 1348 | static int handle_pfds_entries(struct tracer *tp, int nevs, int force_read) |
b7106311 | 1349 | { |
3fe0b570 AB |
1350 | int i, nentries = 0; |
1351 | struct trace_buf *tbp; | |
1352 | struct pollfd *pfd = tp->pfds; | |
1353 | struct io_info *iop = tp->ios; | |
1354 | ||
1355 | tbp = alloc_trace_buf(tp->cpu, buf_size); | |
1356 | for (i = 0; i < ndevs; i++, pfd++, iop++) { | |
1357 | if (pfd->revents & POLLIN || force_read) { | |
1358 | tbp->len = read(iop->ifd, tbp->buf, buf_size); | |
1359 | if (tbp->len > 0) { | |
1360 | pdc_dr_update(iop->dpp, tp->cpu, tbp->len); | |
1361 | add_trace_buf(iop->dpp, tp->cpu, &tbp); | |
1362 | nentries++; | |
1363 | } else if (tbp->len == 0) { | |
1364 | /* | |
1365 | * Short reads after we're done stop us | |
1366 | * from trying reads. | |
1367 | */ | |
1368 | if (tp->is_done) | |
1369 | clear_events(pfd); | |
1370 | } else { | |
1371 | read_err(tp->cpu, iop->ifn); | |
1372 | if (errno != EAGAIN || tp->is_done) | |
1373 | clear_events(pfd); | |
1374 | } | |
1375 | if (!piped_output && --nevs == 0) | |
1376 | break; | |
1377 | } | |
b7106311 | 1378 | } |
3fe0b570 | 1379 | free(tbp); |
6a752c90 | 1380 | |
3fe0b570 AB |
1381 | if (nentries) |
1382 | incr_entries(nentries); | |
1383 | ||
1384 | return nentries; | |
b7106311 JA |
1385 | } |
1386 | ||
3fe0b570 | 1387 | static int fill_ofname(struct io_info *iop, int cpu) |
8e86c98a | 1388 | { |
3fe0b570 | 1389 | int len; |
e3bf54d8 | 1390 | struct stat sb; |
3fe0b570 | 1391 | char *dst = iop->ofn; |
8e86c98a JA |
1392 | |
1393 | if (output_dir) | |
3fe0b570 | 1394 | len = snprintf(iop->ofn, sizeof(iop->ofn), "%s/", output_dir); |
dd870ef6 | 1395 | else |
3fe0b570 | 1396 | len = snprintf(iop->ofn, sizeof(iop->ofn), "./"); |
8e86c98a | 1397 | |
e3bf54d8 | 1398 | if (net_mode == Net_server) { |
3fe0b570 | 1399 | struct cl_conn *nc = iop->nc; |
e0a1988b | 1400 | |
3fe0b570 AB |
1401 | len += sprintf(dst + len, "%s-", nc->ch->hostname); |
1402 | len += strftime(dst + len, 64, "%F-%T/", | |
1403 | gmtime(&iop->dpp->cl_connect_time)); | |
e3bf54d8 JA |
1404 | } |
1405 | ||
3fe0b570 | 1406 | if (stat(iop->ofn, &sb) < 0) { |
e3bf54d8 | 1407 | if (errno != ENOENT) { |
3fe0b570 AB |
1408 | fprintf(stderr, |
1409 | "Destination dir %s stat failed: %d/%s\n", | |
1410 | iop->ofn, errno, strerror(errno)); | |
e3bf54d8 JA |
1411 | return 1; |
1412 | } | |
3fe0b570 AB |
1413 | if (mkdir(iop->ofn, 0755) < 0) { |
1414 | fprintf(stderr, | |
1415 | "Destination dir %s can't be made: %d/%s\n", | |
1416 | iop->ofn, errno, strerror(errno)); | |
e3bf54d8 JA |
1417 | return 1; |
1418 | } | |
1419 | } | |
1420 | ||
8e86c98a | 1421 | if (output_name) |
3fe0b570 AB |
1422 | snprintf(iop->ofn + len, sizeof(iop->ofn), "%s.blktrace.%d", |
1423 | output_name, cpu); | |
8e86c98a | 1424 | else |
3fe0b570 AB |
1425 | snprintf(iop->ofn + len, sizeof(iop->ofn), "%s.blktrace.%d", |
1426 | iop->dpp->buts_name, cpu); | |
e3bf54d8 JA |
1427 | |
1428 | return 0; | |
8e86c98a JA |
1429 | } |
1430 | ||
3fe0b570 | 1431 | static int set_vbuf(struct io_info *iop, int mode, size_t size) |
0cc7d25e | 1432 | { |
3fe0b570 AB |
1433 | iop->obuf = malloc(size); |
1434 | if (setvbuf(iop->ofp, iop->obuf, mode, size) < 0) { | |
1435 | fprintf(stderr, "setvbuf(%s, %d) failed: %d/%s\n", | |
1436 | iop->dpp->path, (int)size, errno, | |
1437 | strerror(errno)); | |
1438 | free(iop->obuf); | |
ddf22842 JA |
1439 | return 1; |
1440 | } | |
d5396421 | 1441 | |
ddf22842 JA |
1442 | return 0; |
1443 | } | |
007c233c | 1444 | |
3fe0b570 | 1445 | static int iop_open(struct io_info *iop, int cpu) |
ddf22842 | 1446 | { |
3fe0b570 AB |
1447 | iop->ofd = -1; |
1448 | if (fill_ofname(iop, cpu)) | |
1449 | return 1; | |
0cc7d25e | 1450 | |
3fe0b570 AB |
1451 | iop->ofp = my_fopen(iop->ofn, "w+"); |
1452 | if (iop->ofp == NULL) { | |
1453 | fprintf(stderr, "Open output file %s failed: %d/%s\n", | |
1454 | iop->ofn, errno, strerror(errno)); | |
1455 | return 1; | |
1456 | } | |
1457 | if (set_vbuf(iop, _IOLBF, FILE_VBUF_SIZE)) { | |
1458 | fprintf(stderr, "set_vbuf for file %s failed: %d/%s\n", | |
1459 | iop->ofn, errno, strerror(errno)); | |
1460 | fclose(iop->ofp); | |
1461 | return 1; | |
d0ca268b JA |
1462 | } |
1463 | ||
3fe0b570 | 1464 | iop->ofd = fileno(iop->ofp); |
e7c9f3ff | 1465 | return 0; |
d0ca268b JA |
1466 | } |
1467 | ||
3fe0b570 | 1468 | static int open_ios(struct tracer *tp) |
3aabcd89 | 1469 | { |
3fe0b570 AB |
1470 | struct pollfd *pfd; |
1471 | struct io_info *iop; | |
1472 | struct list_head *p; | |
1473 | ||
1474 | tp->ios = calloc(ndevs, sizeof(struct io_info)); | |
1475 | tp->pfds = calloc(ndevs, sizeof(struct pollfd)); | |
1476 | ||
1477 | memset(tp->ios, 0, ndevs * sizeof(struct io_info)); | |
1478 | memset(tp->pfds, 0, ndevs * sizeof(struct pollfd)); | |
1479 | ||
1480 | tp->nios = 0; | |
1481 | iop = tp->ios; | |
1482 | pfd = tp->pfds; | |
1483 | __list_for_each(p, &devpaths) { | |
1484 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1485 | ||
1486 | iop->dpp = dpp; | |
1487 | iop->ofd = -1; | |
1488 | snprintf(iop->ifn, sizeof(iop->ifn), "%s/block/%s/trace%d", | |
1489 | debugfs_path, dpp->buts_name, tp->cpu); | |
1490 | ||
1491 | iop->ifd = my_open(iop->ifn, O_RDONLY | O_NONBLOCK); | |
1492 | if (iop->ifd < 0) { | |
1493 | fprintf(stderr, "Thread %d failed open %s: %d/%s\n", | |
1494 | tp->cpu, iop->ifn, errno, strerror(errno)); | |
1495 | return 1; | |
1496 | } | |
1497 | ||
1498 | init_mmap_info(&iop->mmap_info); | |
1499 | ||
1500 | pfd->fd = iop->ifd; | |
1501 | pfd->events = POLLIN; | |
1502 | ||
1503 | if (piped_output) | |
1504 | ; | |
1505 | else if (net_client_use_sendfile()) { | |
1506 | iop->ofd = net_setup_client(); | |
1507 | if (iop->ofd < 0) | |
1508 | goto err; | |
1509 | net_send_open(iop->ofd, tp->cpu, dpp->buts_name); | |
1510 | } else if (net_mode == Net_none) { | |
1511 | if (iop_open(iop, tp->cpu)) | |
1512 | goto err; | |
1513 | } else { | |
1514 | /* | |
1515 | * This ensures that the server knows about all | |
1516 | * connections & devices before _any_ closes | |
1517 | */ | |
1518 | net_send_open(cl_fds[tp->cpu], tp->cpu, dpp->buts_name); | |
1519 | } | |
007c233c | 1520 | |
3fe0b570 AB |
1521 | pfd++; |
1522 | iop++; | |
1523 | tp->nios++; | |
9db17354 | 1524 | } |
3aabcd89 | 1525 | |
3fe0b570 | 1526 | return 0; |
72ca8801 | 1527 | |
3fe0b570 AB |
1528 | err: |
1529 | close(iop->ifd); /* tp->nios _not_ bumped */ | |
1530 | return 1; | |
e7c9f3ff NS |
1531 | } |
1532 | ||
3fe0b570 | 1533 | static void close_iop(struct io_info *iop) |
e7c9f3ff | 1534 | { |
3fe0b570 | 1535 | struct mmap_info *mip = &iop->mmap_info; |
007c233c | 1536 | |
3fe0b570 AB |
1537 | if (mip->fs_buf) |
1538 | munmap(mip->fs_buf, mip->fs_buf_len); | |
72ca8801 | 1539 | |
3fe0b570 AB |
1540 | if (!piped_output) { |
1541 | if (ftruncate(fileno(iop->ofp), mip->fs_size) < 0) { | |
1542 | fprintf(stderr, | |
1543 | "Ignoring err: ftruncate(%s): %d/%s\n", | |
1544 | iop->ofn, errno, strerror(errno)); | |
1545 | } | |
eb3c8108 JA |
1546 | } |
1547 | ||
3fe0b570 AB |
1548 | if (iop->ofp) |
1549 | fclose(iop->ofp); | |
1550 | if (iop->obuf) | |
1551 | free(iop->obuf); | |
72ca8801 NS |
1552 | } |
1553 | ||
3fe0b570 | 1554 | static void close_ios(struct tracer *tp) |
e7c9f3ff | 1555 | { |
3fe0b570 AB |
1556 | while (tp->nios > 0) { |
1557 | struct io_info *iop = &tp->ios[--tp->nios]; | |
e7c9f3ff | 1558 | |
3fe0b570 AB |
1559 | iop->dpp->drops = get_drops(iop->dpp); |
1560 | if (iop->ifd >= 0) | |
1561 | close(iop->ifd); | |
e7c9f3ff | 1562 | |
3fe0b570 AB |
1563 | if (iop->ofp) |
1564 | close_iop(iop); | |
1565 | else if (iop->ofd >= 0) { | |
1566 | struct devpath *dpp = iop->dpp; | |
d0ca268b | 1567 | |
3fe0b570 AB |
1568 | net_send_close(iop->ofd, dpp->buts_name, dpp->drops); |
1569 | net_close_connection(&iop->ofd); | |
e7c9f3ff NS |
1570 | } |
1571 | } | |
99c1f5ab | 1572 | |
3fe0b570 AB |
1573 | free(tp->ios); |
1574 | free(tp->pfds); | |
e7c9f3ff NS |
1575 | } |
1576 | ||
3fe0b570 | 1577 | static int setup_mmap(int fd, unsigned int maxlen, struct mmap_info *mip) |
e7c9f3ff | 1578 | { |
3fe0b570 AB |
1579 | if (mip->fs_off + maxlen > mip->fs_buf_len) { |
1580 | unsigned long nr = max(16, mip->buf_nr); | |
e7c9f3ff | 1581 | |
3fe0b570 AB |
1582 | if (mip->fs_buf) { |
1583 | munlock(mip->fs_buf, mip->fs_buf_len); | |
1584 | munmap(mip->fs_buf, mip->fs_buf_len); | |
1585 | mip->fs_buf = NULL; | |
e7c9f3ff | 1586 | } |
99c1f5ab | 1587 | |
3fe0b570 AB |
1588 | mip->fs_off = mip->fs_size & (mip->pagesize - 1); |
1589 | mip->fs_buf_len = (nr * mip->buf_size) - mip->fs_off; | |
1590 | mip->fs_max_size += mip->fs_buf_len; | |
e7c9f3ff | 1591 | |
3fe0b570 AB |
1592 | if (ftruncate(fd, mip->fs_max_size) < 0) { |
1593 | perror("__setup_mmap: ftruncate"); | |
1594 | return 1; | |
e7c9f3ff | 1595 | } |
99c1f5ab | 1596 | |
3fe0b570 AB |
1597 | mip->fs_buf = my_mmap(NULL, mip->fs_buf_len, PROT_WRITE, |
1598 | MAP_SHARED, fd, | |
1599 | mip->fs_size - mip->fs_off); | |
1600 | if (mip->fs_buf == MAP_FAILED) { | |
1601 | perror("__setup_mmap: mmap"); | |
1602 | return 1; | |
1603 | } | |
1604 | my_mlock(mip->fs_buf, mip->fs_buf_len); | |
d0ca268b JA |
1605 | } |
1606 | ||
e7c9f3ff | 1607 | return 0; |
d0ca268b JA |
1608 | } |
1609 | ||
3fe0b570 | 1610 | static int handle_pfds_file(struct tracer *tp, int nevs, int force_read) |
e7c9f3ff | 1611 | { |
3fe0b570 AB |
1612 | struct mmap_info *mip; |
1613 | int i, ret, nentries = 0; | |
1614 | struct pollfd *pfd = tp->pfds; | |
1615 | struct io_info *iop = tp->ios; | |
1616 | ||
1617 | for (i = 0; nevs > 0 && i < ndevs; i++, pfd++, iop++) { | |
1618 | if (pfd->revents & POLLIN || force_read) { | |
1619 | mip = &iop->mmap_info; | |
1620 | ||
1621 | ret = setup_mmap(iop->ofd, buf_size, mip); | |
1622 | if (ret < 0) { | |
1623 | pfd->events = 0; | |
1624 | break; | |
1625 | } | |
428683db | 1626 | |
3fe0b570 AB |
1627 | ret = read(iop->ifd, mip->fs_buf + mip->fs_off, |
1628 | buf_size); | |
1629 | if (ret > 0) { | |
1630 | pdc_dr_update(iop->dpp, tp->cpu, ret); | |
1631 | mip->fs_size += ret; | |
1632 | mip->fs_off += ret; | |
1633 | nentries++; | |
1634 | } else if (ret == 0) { | |
1635 | /* | |
1636 | * Short reads after we're done stop us | |
1637 | * from trying reads. | |
1638 | */ | |
1639 | if (tp->is_done) | |
1640 | clear_events(pfd); | |
1641 | } else { | |
1642 | read_err(tp->cpu, iop->ifn); | |
1643 | if (errno != EAGAIN || tp->is_done) | |
1644 | clear_events(pfd); | |
1645 | } | |
1646 | nevs--; | |
e7c9f3ff | 1647 | } |
e7c9f3ff | 1648 | } |
56070ea4 | 1649 | |
3fe0b570 | 1650 | return nentries; |
e7c9f3ff | 1651 | } |
52724a0e | 1652 | |
3fe0b570 | 1653 | static void *thread_main(void *arg) |
8e86c98a | 1654 | { |
3fe0b570 AB |
1655 | int ret, ndone; |
1656 | int to_val; | |
8e86c98a | 1657 | |
3fe0b570 | 1658 | struct tracer *tp = arg; |
8e86c98a | 1659 | |
3fe0b570 AB |
1660 | ret = lock_on_cpu(tp->cpu); |
1661 | if (ret) | |
1662 | goto err; | |
ff11d54c | 1663 | |
3fe0b570 AB |
1664 | ret = open_ios(tp); |
1665 | if (ret) { | |
1666 | close_ios(tp); | |
1667 | goto err; | |
8e86c98a JA |
1668 | } |
1669 | ||
3fe0b570 AB |
1670 | pthread_mutex_lock(&tp->mutex); |
1671 | tp->running = 1; | |
1672 | pthread_cond_signal(&tp->cond); | |
1673 | pthread_mutex_unlock(&tp->mutex); | |
6a6d3f0f | 1674 | |
3fe0b570 AB |
1675 | if (piped_output) |
1676 | to_val = 50; /* Frequent partial handles */ | |
ff11d54c | 1677 | else |
3fe0b570 AB |
1678 | to_val = 500; /* 1/2 second intervals */ |
1679 | ||
1680 | while (!tp->is_done) { | |
1681 | ndone = poll(tp->pfds, ndevs, to_val); | |
1682 | if (ndone || piped_output) | |
1683 | (void)handle_pfds(tp, ndone, piped_output); | |
1684 | else if (ndone < 0 && errno != EINTR) | |
1685 | fprintf(stderr, "Thread %d poll failed: %d/%s\n", | |
1686 | tp->cpu, errno, strerror(errno)); | |
1687 | } | |
22cd0c02 JA |
1688 | |
1689 | /* | |
3fe0b570 | 1690 | * Trace is stopped, pull data until we get a short read |
22cd0c02 | 1691 | */ |
3fe0b570 AB |
1692 | while (handle_pfds(tp, ndevs, 1) > 0) |
1693 | ; | |
ff11d54c | 1694 | |
3fe0b570 | 1695 | close_ios(tp); |
8e86c98a | 1696 | |
3fe0b570 AB |
1697 | err: |
1698 | pthread_mutex_lock(&tp->mutex); | |
1699 | tp->running = 0; | |
1700 | tp->status = ret; | |
1701 | pthread_cond_signal(&tp->cond); | |
1702 | pthread_mutex_unlock(&tp->mutex); | |
1703 | return NULL; | |
22cd0c02 JA |
1704 | } |
1705 | ||
3fe0b570 | 1706 | static int start_tracer(int cpu) |
22cd0c02 | 1707 | { |
3fe0b570 | 1708 | struct tracer *tp; |
22cd0c02 | 1709 | |
3fe0b570 AB |
1710 | tp = malloc(sizeof(*tp)); |
1711 | memset(tp, 0, sizeof(*tp)); | |
7ab2f837 | 1712 | |
3fe0b570 AB |
1713 | INIT_LIST_HEAD(&tp->head); |
1714 | pthread_mutex_init(&tp->mutex, NULL); | |
1715 | pthread_cond_init(&tp->cond, NULL); | |
1716 | tp->running = 0; | |
1717 | tp->status = 0; | |
1718 | tp->cpu = cpu; | |
8e86c98a | 1719 | |
3fe0b570 AB |
1720 | if (pthread_create(&tp->thread, NULL, thread_main, tp)) { |
1721 | fprintf(stderr, "FAILED to start thread on CPU %d: %d/%s\n", | |
1722 | cpu, errno, strerror(errno)); | |
1723 | goto err; | |
1724 | } | |
8e86c98a | 1725 | |
3fe0b570 AB |
1726 | pthread_mutex_lock(&tp->mutex); |
1727 | while (!tp->running && (tp->status == 0)) | |
1728 | pthread_cond_wait(&tp->cond, &tp->mutex); | |
1729 | pthread_mutex_unlock(&tp->mutex); | |
1730 | ||
1731 | if (tp->status == 0) { | |
1732 | list_add_tail(&tp->head, &tracers); | |
1733 | return 0; | |
8e86c98a | 1734 | } |
3fe0b570 AB |
1735 | |
1736 | fprintf(stderr, "FAILED to start thread on CPU %d\n", cpu); | |
1737 | ||
1738 | err: | |
1739 | pthread_mutex_destroy(&tp->mutex); | |
1740 | pthread_cond_destroy(&tp->cond); | |
1741 | free(tp); | |
1742 | return 1; | |
8e86c98a JA |
1743 | } |
1744 | ||
3fe0b570 | 1745 | static int start_tracers(void) |
e0a1988b | 1746 | { |
3fe0b570 AB |
1747 | int cpu; |
1748 | ||
1749 | for (cpu = 0; cpu < ncpus; cpu++) | |
1750 | if (start_tracer(cpu)) | |
1751 | break; | |
e0a1988b | 1752 | |
3fe0b570 AB |
1753 | return cpu; |
1754 | } | |
e0a1988b | 1755 | |
3fe0b570 AB |
1756 | static void stop_tracers(void) |
1757 | { | |
1758 | struct list_head *p; | |
e0a1988b JA |
1759 | |
1760 | /* | |
3fe0b570 | 1761 | * Stop the tracing - makes the tracer threads clean up quicker. |
e0a1988b | 1762 | */ |
3fe0b570 AB |
1763 | __list_for_each(p, &devpaths) { |
1764 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
1765 | (void)ioctl(dpp->fd, BLKTRACESTOP); | |
e0a1988b JA |
1766 | } |
1767 | ||
3fe0b570 AB |
1768 | /* |
1769 | * Tell each tracer to quit | |
1770 | */ | |
1771 | __list_for_each(p, &tracers) { | |
1772 | struct tracer *tp = list_entry(p, struct tracer, head); | |
1773 | tp->is_done = 1; | |
1774 | } | |
ff11d54c | 1775 | } |
e0a1988b | 1776 | |
3fe0b570 | 1777 | static void del_tracers(void) |
ff11d54c | 1778 | { |
3fe0b570 | 1779 | struct list_head *p, *q; |
ff11d54c | 1780 | |
3fe0b570 AB |
1781 | list_for_each_safe(p, q, &tracers) { |
1782 | struct tracer *tp = list_entry(p, struct tracer, head); | |
ff11d54c | 1783 | |
3fe0b570 AB |
1784 | list_del(&tp->head); |
1785 | free(tp); | |
e0a1988b | 1786 | } |
3fe0b570 | 1787 | ntracers = 0; |
ff11d54c | 1788 | } |
e0a1988b | 1789 | |
3fe0b570 | 1790 | static void wait_tracers(void) |
ff11d54c | 1791 | { |
3fe0b570 | 1792 | struct list_head *p; |
ff11d54c | 1793 | |
3fe0b570 AB |
1794 | if (use_tracer_devpaths()) |
1795 | process_trace_bufs(); | |
1796 | ||
1797 | __list_for_each(p, &tracers) { | |
1798 | int ret; | |
1799 | struct tracer *tp = list_entry(p, struct tracer, head); | |
1800 | ||
1801 | pthread_mutex_lock(&tp->mutex); | |
1802 | while (tp->running) | |
1803 | pthread_cond_wait(&tp->cond, &tp->mutex); | |
1804 | pthread_mutex_unlock(&tp->mutex); | |
1805 | ||
1806 | ret = pthread_join(tp->thread, NULL); | |
1807 | if (ret) | |
1808 | fprintf(stderr, "Thread join %d failed %d\n", | |
1809 | tp->cpu, ret); | |
ff11d54c TZ |
1810 | } |
1811 | ||
3fe0b570 AB |
1812 | if (use_tracer_devpaths()) |
1813 | clean_trace_bufs(); | |
1814 | ||
1815 | get_all_drops(); | |
ff11d54c TZ |
1816 | } |
1817 | ||
3fe0b570 | 1818 | static void exit_tracing(void) |
ff11d54c | 1819 | { |
3fe0b570 AB |
1820 | signal(SIGINT, SIG_IGN); |
1821 | signal(SIGHUP, SIG_IGN); | |
1822 | signal(SIGTERM, SIG_IGN); | |
1823 | signal(SIGALRM, SIG_IGN); | |
1824 | ||
1825 | stop_tracers(); | |
1826 | wait_tracers(); | |
1827 | del_tracers(); | |
1828 | rel_devpaths(); | |
e0a1988b JA |
1829 | } |
1830 | ||
3fe0b570 | 1831 | static void handle_sigint(__attribute__((__unused__)) int sig) |
8e86c98a | 1832 | { |
3fe0b570 AB |
1833 | done = 1; |
1834 | stop_tracers(); | |
8e86c98a JA |
1835 | } |
1836 | ||
3fe0b570 | 1837 | static void show_stats(struct list_head *devpaths) |
659bcc3f | 1838 | { |
3fe0b570 AB |
1839 | FILE *ofp; |
1840 | struct list_head *p; | |
1841 | unsigned long long nevents, data_read; | |
1842 | unsigned long long total_drops = 0; | |
1843 | unsigned long long total_events = 0; | |
1844 | ||
1845 | if (piped_output) | |
1846 | ofp = my_fopen("/dev/null", "w"); | |
1847 | else | |
1848 | ofp = stdout; | |
ff11d54c | 1849 | |
3fe0b570 AB |
1850 | __list_for_each(p, devpaths) { |
1851 | int cpu; | |
1852 | struct pdc_stats *sp; | |
1853 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
e0a1988b | 1854 | |
3fe0b570 AB |
1855 | if (net_mode == Net_server) |
1856 | printf("server: end of run for %s:%s\n", | |
1857 | dpp->ch->hostname, dpp->buts_name); | |
e0a1988b | 1858 | |
3fe0b570 AB |
1859 | data_read = 0; |
1860 | nevents = 0; | |
1861 | ||
1862 | fprintf(ofp, "=== %s ===\n", dpp->buts_name); | |
1863 | for (cpu = 0, sp = dpp->stats; cpu < dpp->ncpus; cpu++, sp++) { | |
1864 | /* | |
1865 | * Estimate events if not known... | |
1866 | */ | |
1867 | if (sp->nevents == 0) { | |
1868 | sp->nevents = sp->data_read / | |
1869 | sizeof(struct blk_io_trace); | |
ff11d54c | 1870 | } |
e0a1988b | 1871 | |
3fe0b570 AB |
1872 | fprintf(ofp, |
1873 | " CPU%3d: %20llu events, %8llu KiB data\n", | |
1874 | cpu, sp->nevents, (sp->data_read + 1023) >> 10); | |
e0a1988b | 1875 | |
3fe0b570 AB |
1876 | data_read += sp->data_read; |
1877 | nevents += sp->nevents; | |
e0a1988b JA |
1878 | } |
1879 | ||
3fe0b570 AB |
1880 | fprintf(ofp, " Total: %20llu events (dropped %llu)," |
1881 | " %8llu KiB data\n", nevents, | |
1882 | dpp->drops, (data_read + 1024) >> 10); | |
8e86c98a | 1883 | |
3fe0b570 AB |
1884 | total_drops += dpp->drops; |
1885 | total_events += (nevents + dpp->drops); | |
8e86c98a JA |
1886 | } |
1887 | ||
3fe0b570 AB |
1888 | fflush(ofp); |
1889 | if (piped_output) | |
1890 | fclose(ofp); | |
8e86c98a | 1891 | |
3fe0b570 AB |
1892 | if (total_drops) { |
1893 | double drops_ratio = 1.0; | |
8e86c98a | 1894 | |
3fe0b570 AB |
1895 | if (total_events) |
1896 | drops_ratio = (double)total_drops/(double)total_events; | |
8e86c98a | 1897 | |
3fe0b570 AB |
1898 | fprintf(stderr, "\nYou have %llu (%5.1lf%%) dropped events\n" |
1899 | "Consider using a larger buffer size (-b) " | |
1900 | "and/or more buffers (-n)\n", | |
1901 | total_drops, 100.0 * drops_ratio); | |
8e86c98a | 1902 | } |
8e86c98a JA |
1903 | } |
1904 | ||
3fe0b570 | 1905 | static int handle_args(int argc, char *argv[]) |
8e86c98a | 1906 | { |
3fe0b570 | 1907 | int c, i; |
e3e74029 | 1908 | struct statfs st; |
d39c04ca AB |
1909 | int act_mask_tmp = 0; |
1910 | ||
1911 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) { | |
1912 | switch (c) { | |
1913 | case 'a': | |
1914 | i = find_mask_map(optarg); | |
1915 | if (i < 0) { | |
3fe0b570 | 1916 | fprintf(stderr, "Invalid action mask %s\n", |
d39c04ca | 1917 | optarg); |
7425d456 | 1918 | return 1; |
d39c04ca AB |
1919 | } |
1920 | act_mask_tmp |= i; | |
1921 | break; | |
1922 | ||
1923 | case 'A': | |
3fe0b570 | 1924 | if ((sscanf(optarg, "%x", &i) != 1) || |
98f8386b | 1925 | !valid_act_opt(i)) { |
d39c04ca | 1926 | fprintf(stderr, |
ab197ca7 | 1927 | "Invalid set action mask %s/0x%x\n", |
d39c04ca | 1928 | optarg, i); |
7425d456 | 1929 | return 1; |
d39c04ca AB |
1930 | } |
1931 | act_mask_tmp = i; | |
1932 | break; | |
d0ca268b | 1933 | |
d39c04ca | 1934 | case 'd': |
3fe0b570 | 1935 | if (add_devpath(optarg) != 0) |
e7c9f3ff | 1936 | return 1; |
d39c04ca AB |
1937 | break; |
1938 | ||
cf1edb17 AB |
1939 | case 'I': { |
1940 | char dev_line[256]; | |
3fe0b570 | 1941 | FILE *ifp = my_fopen(optarg, "r"); |
cf1edb17 AB |
1942 | |
1943 | if (!ifp) { | |
3fe0b570 AB |
1944 | fprintf(stderr, |
1945 | "Invalid file for devices %s\n", | |
cf1edb17 AB |
1946 | optarg); |
1947 | return 1; | |
1948 | } | |
1949 | ||
1950 | while (fscanf(ifp, "%s\n", dev_line) == 1) | |
3fe0b570 | 1951 | if (add_devpath(dev_line) != 0) |
cf1edb17 AB |
1952 | return 1; |
1953 | break; | |
1954 | } | |
cf1edb17 | 1955 | |
5270dddd | 1956 | case 'r': |
3d06efea | 1957 | debugfs_path = optarg; |
5270dddd JA |
1958 | break; |
1959 | ||
d5396421 | 1960 | case 'o': |
66efebf8 | 1961 | output_name = optarg; |
d5396421 | 1962 | break; |
bc39777c JA |
1963 | case 'k': |
1964 | kill_running_trace = 1; | |
1965 | break; | |
ece238a6 NS |
1966 | case 'w': |
1967 | stop_watch = atoi(optarg); | |
1968 | if (stop_watch <= 0) { | |
1969 | fprintf(stderr, | |
1970 | "Invalid stopwatch value (%d secs)\n", | |
1971 | stop_watch); | |
1972 | return 1; | |
1973 | } | |
1974 | break; | |
57ea8602 | 1975 | case 'V': |
5d4f19d9 | 1976 | case 'v': |
52724a0e | 1977 | printf("%s version %s\n", argv[0], blktrace_version); |
3fe0b570 AB |
1978 | exit(0); |
1979 | /*NOTREACHED*/ | |
129aa440 | 1980 | case 'b': |
eb3c8108 | 1981 | buf_size = strtoul(optarg, NULL, 10); |
183a0855 | 1982 | if (buf_size <= 0 || buf_size > 16*1024) { |
3fe0b570 AB |
1983 | fprintf(stderr, "Invalid buffer size (%lu)\n", |
1984 | buf_size); | |
129aa440 JA |
1985 | return 1; |
1986 | } | |
1987 | buf_size <<= 10; | |
1988 | break; | |
1989 | case 'n': | |
eb3c8108 | 1990 | buf_nr = strtoul(optarg, NULL, 10); |
129aa440 JA |
1991 | if (buf_nr <= 0) { |
1992 | fprintf(stderr, | |
eb3c8108 | 1993 | "Invalid buffer nr (%lu)\n", buf_nr); |
129aa440 JA |
1994 | return 1; |
1995 | } | |
1996 | break; | |
d1d7f15f JA |
1997 | case 'D': |
1998 | output_dir = optarg; | |
1999 | break; | |
8e86c98a JA |
2000 | case 'h': |
2001 | net_mode = Net_client; | |
2002 | strcpy(hostname, optarg); | |
2003 | break; | |
2004 | case 'l': | |
2005 | net_mode = Net_server; | |
2006 | break; | |
2007 | case 'p': | |
2008 | net_port = atoi(optarg); | |
2009 | break; | |
32f18c48 | 2010 | case 's': |
79971f43 | 2011 | net_use_sendfile = 0; |
32f18c48 | 2012 | break; |
d39c04ca | 2013 | default: |
ee1f4158 | 2014 | show_usage(argv[0]); |
3fe0b570 AB |
2015 | exit(1); |
2016 | /*NOTREACHED*/ | |
d39c04ca AB |
2017 | } |
2018 | } | |
2019 | ||
3fe0b570 AB |
2020 | while (optind < argc) |
2021 | if (add_devpath(argv[optind++]) != 0) | |
2022 | return 1; | |
8e86c98a | 2023 | |
3fe0b570 AB |
2024 | if (net_mode != Net_server && ndevs == 0) { |
2025 | show_usage(argv[0]); | |
2026 | return 1; | |
2027 | } | |
8e86c98a | 2028 | |
3fe0b570 AB |
2029 | if (statfs(debugfs_path, &st) < 0 || st.f_type != (long)DEBUGFS_TYPE) { |
2030 | fprintf(stderr, "Invalid debug path %s: %d/%s\n", | |
2031 | debugfs_path, errno, strerror(errno)); | |
2032 | return 1; | |
2033 | } | |
2034 | ||
2035 | if (act_mask_tmp != 0) | |
2036 | act_mask = act_mask_tmp; | |
2037 | ||
e58f3937 AB |
2038 | if (net_mode == Net_client && net_setup_addr()) |
2039 | return 1; | |
2040 | ||
3fe0b570 AB |
2041 | /* |
2042 | * Set up for appropriate PFD handler based upon output name. | |
2043 | */ | |
2044 | if (net_client_use_sendfile()) | |
2045 | handle_pfds = handle_pfds_netclient; | |
2046 | else if (net_client_use_send()) | |
2047 | handle_pfds = handle_pfds_entries; | |
2048 | else if (output_name && (strcmp(output_name, "-") == 0)) { | |
2049 | piped_output = 1; | |
2050 | handle_pfds = handle_pfds_entries; | |
2051 | pfp = stdout; | |
2052 | setvbuf(pfp, NULL, _IONBF, 0); | |
2053 | } else | |
2054 | handle_pfds = handle_pfds_file; | |
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | static void ch_add_connection(struct net_server_s *ns, struct cl_host *ch, | |
2059 | int fd) | |
2060 | { | |
2061 | struct cl_conn *nc; | |
2062 | ||
2063 | nc = malloc(sizeof(*nc)); | |
2064 | memset(nc, 0, sizeof(*nc)); | |
2065 | ||
2066 | time(&nc->connect_time); | |
2067 | nc->ch = ch; | |
2068 | nc->fd = fd; | |
2069 | nc->ncpus = -1; | |
2070 | ||
2071 | list_add_tail(&nc->ch_head, &ch->conn_list); | |
2072 | ch->connects++; | |
2073 | ||
2074 | list_add_tail(&nc->ns_head, &ns->conn_list); | |
2075 | ns->connects++; | |
2076 | ns->pfds = realloc(ns->pfds, (ns->connects+1) * sizeof(struct pollfd)); | |
2077 | } | |
2078 | ||
2079 | static void ch_rem_connection(struct net_server_s *ns, struct cl_host *ch, | |
2080 | struct cl_conn *nc) | |
2081 | { | |
2082 | net_close_connection(&nc->fd); | |
2083 | ||
2084 | list_del(&nc->ch_head); | |
2085 | ch->connects--; | |
2086 | ||
2087 | list_del(&nc->ns_head); | |
2088 | ns->connects--; | |
2089 | ns->pfds = realloc(ns->pfds, (ns->connects+1) * sizeof(struct pollfd)); | |
2090 | ||
2091 | free(nc); | |
2092 | } | |
2093 | ||
2094 | static struct cl_host *net_find_client_host(struct net_server_s *ns, | |
2095 | struct in_addr cl_in_addr) | |
2096 | { | |
2097 | struct list_head *p; | |
2098 | ||
2099 | __list_for_each(p, &ns->ch_list) { | |
2100 | struct cl_host *ch = list_entry(p, struct cl_host, head); | |
2101 | ||
2102 | if (in_addr_eq(ch->cl_in_addr, cl_in_addr)) | |
2103 | return ch; | |
2104 | } | |
2105 | ||
2106 | return NULL; | |
2107 | } | |
2108 | ||
2109 | static struct cl_host *net_add_client_host(struct net_server_s *ns, | |
2110 | struct sockaddr_in *addr) | |
2111 | { | |
2112 | struct cl_host *ch; | |
2113 | ||
2114 | ch = malloc(sizeof(*ch)); | |
2115 | memset(ch, 0, sizeof(*ch)); | |
2116 | ||
2117 | ch->ns = ns; | |
2118 | ch->cl_in_addr = addr->sin_addr; | |
2119 | list_add_tail(&ch->head, &ns->ch_list); | |
2120 | ns->nchs++; | |
ec685dd2 | 2121 | |
3fe0b570 AB |
2122 | ch->hostname = strdup(inet_ntoa(addr->sin_addr)); |
2123 | printf("server: connection from %s\n", ch->hostname); | |
2124 | ||
2125 | INIT_LIST_HEAD(&ch->conn_list); | |
2126 | INIT_LIST_HEAD(&ch->devpaths); | |
2127 | ||
2128 | return ch; | |
2129 | } | |
2130 | ||
2131 | static void device_done(struct devpath *dpp, int ncpus) | |
2132 | { | |
2133 | int cpu; | |
2134 | struct io_info *iop; | |
2135 | ||
2136 | for (cpu = 0, iop = dpp->ios; cpu < ncpus; cpu++, iop++) | |
2137 | close_iop(iop); | |
2138 | ||
2139 | list_del(&dpp->head); | |
2140 | dpp_free(dpp); | |
2141 | } | |
2142 | ||
2143 | static void net_ch_remove(struct cl_host *ch, int ncpus) | |
2144 | { | |
2145 | struct list_head *p, *q; | |
2146 | struct net_server_s *ns = ch->ns; | |
2147 | ||
2148 | list_for_each_safe(p, q, &ch->devpaths) { | |
2149 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
2150 | device_done(dpp, ncpus); | |
ec685dd2 | 2151 | } |
8e86c98a | 2152 | |
3fe0b570 AB |
2153 | list_for_each_safe(p, q, &ch->conn_list) { |
2154 | struct cl_conn *nc = list_entry(p, struct cl_conn, ch_head); | |
2155 | ||
2156 | ch_rem_connection(ns, ch, nc); | |
22cd0c02 JA |
2157 | } |
2158 | ||
3fe0b570 AB |
2159 | list_del(&ch->head); |
2160 | ns->nchs--; | |
2161 | ||
2162 | if (ch->hostname) | |
2163 | free(ch->hostname); | |
2164 | free(ch); | |
2165 | } | |
2166 | ||
2167 | static void net_add_connection(struct net_server_s *ns) | |
2168 | { | |
2169 | int fd; | |
2170 | struct cl_host *ch; | |
2171 | socklen_t socklen = sizeof(ns->addr); | |
2172 | ||
d5302b03 | 2173 | fd = my_accept(ns->listen_fd, (struct sockaddr *)&ns->addr, &socklen); |
3fe0b570 AB |
2174 | if (fd < 0) { |
2175 | /* | |
2176 | * This is OK: we just won't accept this connection, | |
2177 | * nothing fatal. | |
2178 | */ | |
2179 | perror("accept"); | |
2180 | } else { | |
2181 | ch = net_find_client_host(ns, ns->addr.sin_addr); | |
2182 | if (!ch) | |
2183 | ch = net_add_client_host(ns, &ns->addr); | |
2184 | ||
2185 | ch_add_connection(ns, ch, fd); | |
d39c04ca | 2186 | } |
3fe0b570 | 2187 | } |
d39c04ca | 2188 | |
3fe0b570 AB |
2189 | static struct devpath *nc_add_dpp(struct cl_conn *nc, |
2190 | struct blktrace_net_hdr *bnh, | |
2191 | time_t connect_time) | |
2192 | { | |
2193 | int cpu; | |
2194 | struct io_info *iop; | |
2195 | struct devpath *dpp; | |
2196 | ||
2197 | dpp = malloc(sizeof(*dpp)); | |
2198 | memset(dpp, 0, sizeof(*dpp)); | |
2199 | ||
2200 | dpp->buts_name = strdup(bnh->buts_name); | |
2201 | dpp->path = strdup(bnh->buts_name); | |
2202 | dpp->fd = -1; | |
2203 | dpp->ch = nc->ch; | |
2204 | dpp->cl_id = bnh->cl_id; | |
2205 | dpp->cl_connect_time = connect_time; | |
2206 | dpp->ncpus = nc->ncpus; | |
2207 | dpp->stats = calloc(dpp->ncpus, sizeof(*dpp->stats)); | |
2208 | memset(dpp->stats, 0, dpp->ncpus * sizeof(*dpp->stats)); | |
2209 | ||
2210 | list_add_tail(&dpp->head, &nc->ch->devpaths); | |
2211 | nc->ch->ndevs++; | |
2212 | ||
2213 | dpp->ios = calloc(nc->ncpus, sizeof(*iop)); | |
2214 | memset(dpp->ios, 0, ndevs * sizeof(*iop)); | |
2215 | ||
2216 | for (cpu = 0, iop = dpp->ios; cpu < nc->ncpus; cpu++, iop++) { | |
2217 | iop->dpp = dpp; | |
2218 | iop->nc = nc; | |
2219 | init_mmap_info(&iop->mmap_info); | |
2220 | ||
2221 | if (iop_open(iop, cpu)) | |
2222 | goto err; | |
69dd57c2 AB |
2223 | } |
2224 | ||
3fe0b570 | 2225 | return dpp; |
69dd57c2 | 2226 | |
3fe0b570 AB |
2227 | err: |
2228 | /* | |
2229 | * Need to unravel what's been done... | |
2230 | */ | |
2231 | while (cpu >= 0) | |
2232 | close_iop(&dpp->ios[cpu--]); | |
2233 | dpp_free(dpp); | |
2234 | ||
2235 | return NULL; | |
2236 | } | |
d0ca268b | 2237 | |
3fe0b570 AB |
2238 | static struct devpath *nc_find_dpp(struct cl_conn *nc, |
2239 | struct blktrace_net_hdr *bnh) | |
2240 | { | |
2241 | struct list_head *p; | |
2242 | time_t connect_time = nc->connect_time; | |
3d06efea | 2243 | |
3fe0b570 AB |
2244 | __list_for_each(p, &nc->ch->devpaths) { |
2245 | struct devpath *dpp = list_entry(p, struct devpath, head); | |
2246 | ||
2247 | if (!strcmp(dpp->buts_name, bnh->buts_name)) | |
2248 | return dpp; | |
2249 | ||
2250 | if (dpp->cl_id == bnh->cl_id) | |
2251 | connect_time = dpp->cl_connect_time; | |
d0ca268b JA |
2252 | } |
2253 | ||
3fe0b570 AB |
2254 | return nc_add_dpp(nc, bnh, connect_time); |
2255 | } | |
bc39777c | 2256 | |
3fe0b570 AB |
2257 | static void net_client_read_data(struct cl_conn *nc, struct devpath *dpp, |
2258 | struct blktrace_net_hdr *bnh) | |
2259 | { | |
2260 | int ret; | |
2261 | struct io_info *iop = &dpp->ios[bnh->cpu]; | |
2262 | struct mmap_info *mip = &iop->mmap_info; | |
2263 | ||
2264 | if (setup_mmap(iop->ofd, bnh->len, &iop->mmap_info)) { | |
2265 | fprintf(stderr, "ncd(%s:%d): mmap failed\n", | |
2266 | nc->ch->hostname, nc->fd); | |
2267 | exit(1); | |
2268 | } | |
2269 | ||
2270 | ret = net_recv_data(nc->fd, mip->fs_buf + mip->fs_off, bnh->len); | |
2271 | if (ret > 0) { | |
2272 | pdc_dr_update(dpp, bnh->cpu, ret); | |
2273 | mip->fs_size += ret; | |
2274 | mip->fs_off += ret; | |
2275 | } else if (ret < 0) | |
2276 | exit(1); | |
2277 | } | |
2278 | ||
2279 | /* | |
2280 | * Returns 1 if we closed a host - invalidates other polling information | |
2281 | * that may be present. | |
2282 | */ | |
2283 | static int net_client_data(struct cl_conn *nc) | |
2284 | { | |
2285 | int ret; | |
2286 | struct devpath *dpp; | |
2287 | struct blktrace_net_hdr bnh; | |
2288 | ||
2289 | ret = net_get_header(nc, &bnh); | |
2290 | if (ret == 0) | |
7425d456 | 2291 | return 0; |
3fe0b570 AB |
2292 | |
2293 | if (ret < 0) { | |
2294 | fprintf(stderr, "ncd(%d): header read failed\n", nc->fd); | |
2295 | exit(1); | |
2296 | } | |
2297 | ||
2298 | if (data_is_native == -1 && check_data_endianness(bnh.magic)) { | |
2299 | fprintf(stderr, "ncd(%d): received data is bad\n", nc->fd); | |
2300 | exit(1); | |
2301 | } | |
2302 | ||
2303 | if (!data_is_native) { | |
2304 | bnh.magic = be32_to_cpu(bnh.magic); | |
2305 | bnh.cpu = be32_to_cpu(bnh.cpu); | |
2306 | bnh.max_cpus = be32_to_cpu(bnh.max_cpus); | |
2307 | bnh.len = be32_to_cpu(bnh.len); | |
2308 | bnh.cl_id = be32_to_cpu(bnh.cl_id); | |
2309 | bnh.buf_size = be32_to_cpu(bnh.buf_size); | |
2310 | bnh.buf_nr = be32_to_cpu(bnh.buf_nr); | |
2311 | bnh.page_size = be32_to_cpu(bnh.page_size); | |
2312 | } | |
2313 | ||
2314 | if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
2315 | fprintf(stderr, "ncd(%s:%d): bad data magic\n", | |
2316 | nc->ch->hostname, nc->fd); | |
2317 | exit(1); | |
2318 | } | |
2319 | ||
2320 | if (nc->ncpus == -1) | |
2321 | nc->ncpus = bnh.max_cpus; | |
2322 | ||
2323 | /* | |
2324 | * len == 0 means the other end is sending us a new connection/dpp | |
2325 | * len == 1 means that the other end signalled end-of-run | |
2326 | */ | |
2327 | dpp = nc_find_dpp(nc, &bnh); | |
2328 | if (bnh.len == 0) { | |
2329 | /* | |
2330 | * Just adding in the dpp above is enough | |
2331 | */ | |
2332 | ack_open_close(nc->fd, dpp->buts_name); | |
2333 | nc->ch->cl_opens++; | |
2334 | } else if (bnh.len == 1) { | |
2335 | /* | |
2336 | * overload cpu count with dropped events | |
2337 | */ | |
2338 | dpp->drops = bnh.cpu; | |
2339 | ||
2340 | ack_open_close(nc->fd, dpp->buts_name); | |
2341 | if (--nc->ch->cl_opens == 0) { | |
2342 | show_stats(&nc->ch->devpaths); | |
2343 | net_ch_remove(nc->ch, nc->ncpus); | |
2344 | return 1; | |
2345 | } | |
2346 | } else | |
2347 | net_client_read_data(nc, dpp, &bnh); | |
2348 | ||
2349 | return 0; | |
2350 | } | |
2351 | ||
2352 | static void handle_client_data(struct net_server_s *ns, int events) | |
2353 | { | |
2354 | struct cl_conn *nc; | |
2355 | struct pollfd *pfd; | |
2356 | struct list_head *p, *q; | |
2357 | ||
2358 | pfd = &ns->pfds[1]; | |
2359 | list_for_each_safe(p, q, &ns->conn_list) { | |
2360 | if (pfd->revents & POLLIN) { | |
2361 | nc = list_entry(p, struct cl_conn, ns_head); | |
2362 | ||
2363 | if (net_client_data(nc) || --events == 0) | |
2364 | break; | |
2365 | } | |
2366 | pfd++; | |
2367 | } | |
2368 | } | |
2369 | ||
2370 | static void net_setup_pfds(struct net_server_s *ns) | |
2371 | { | |
2372 | struct pollfd *pfd; | |
2373 | struct list_head *p; | |
2374 | ||
2375 | ns->pfds[0].fd = ns->listen_fd; | |
2376 | ns->pfds[0].events = POLLIN; | |
2377 | ||
2378 | pfd = &ns->pfds[1]; | |
2379 | __list_for_each(p, &ns->conn_list) { | |
2380 | struct cl_conn *nc = list_entry(p, struct cl_conn, ns_head); | |
2381 | ||
2382 | pfd->fd = nc->fd; | |
2383 | pfd->events = POLLIN; | |
2384 | pfd++; | |
2385 | } | |
2386 | } | |
2387 | ||
2388 | static int net_server_handle_connections(struct net_server_s *ns) | |
2389 | { | |
2390 | int events; | |
2391 | ||
2392 | printf("server: waiting for connections...\n"); | |
2393 | ||
2394 | while (!done) { | |
2395 | net_setup_pfds(ns); | |
2396 | events = poll(ns->pfds, ns->connects + 1, -1); | |
2397 | if (events < 0) { | |
2398 | if (errno != EINTR) { | |
2399 | perror("FATAL: poll error"); | |
2400 | return 1; | |
2401 | } | |
2402 | } else if (events > 0) { | |
2403 | if (ns->pfds[0].revents & POLLIN) { | |
2404 | net_add_connection(ns); | |
2405 | events--; | |
2406 | } | |
2407 | ||
2408 | if (events) | |
2409 | handle_client_data(ns, events); | |
2410 | } | |
2411 | } | |
2412 | ||
2413 | return 0; | |
2414 | } | |
2415 | ||
2416 | static int net_server(void) | |
2417 | { | |
2418 | int fd, opt; | |
2419 | int ret = 1; | |
2420 | struct net_server_s net_server; | |
2421 | struct net_server_s *ns = &net_server; | |
2422 | ||
2423 | memset(ns, 0, sizeof(*ns)); | |
2424 | INIT_LIST_HEAD(&ns->ch_list); | |
2425 | INIT_LIST_HEAD(&ns->conn_list); | |
2426 | ns->pfds = malloc(sizeof(struct pollfd)); | |
2427 | ||
2428 | fd = my_socket(AF_INET, SOCK_STREAM, 0); | |
2429 | if (fd < 0) { | |
2430 | perror("server: socket"); | |
2431 | goto out; | |
2432 | } | |
2433 | ||
2434 | opt = 1; | |
2435 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) { | |
2436 | perror("setsockopt"); | |
2437 | goto out; | |
2438 | } | |
2439 | ||
2440 | memset(&ns->addr, 0, sizeof(ns->addr)); | |
2441 | ns->addr.sin_family = AF_INET; | |
2442 | ns->addr.sin_addr.s_addr = htonl(INADDR_ANY); | |
2443 | ns->addr.sin_port = htons(net_port); | |
2444 | ||
2445 | if (bind(fd, (struct sockaddr *) &ns->addr, sizeof(ns->addr)) < 0) { | |
2446 | perror("bind"); | |
2447 | goto out; | |
2448 | } | |
2449 | ||
2450 | if (listen(fd, 1) < 0) { | |
2451 | perror("listen"); | |
2452 | goto out; | |
2453 | } | |
2454 | ||
2455 | /* | |
2456 | * The actual server looping is done here: | |
2457 | */ | |
2458 | ns->listen_fd = fd; | |
2459 | ret = net_server_handle_connections(ns); | |
2460 | ||
2461 | /* | |
2462 | * Clean up and return... | |
2463 | */ | |
2464 | out: | |
2465 | free(ns->pfds); | |
2466 | return ret; | |
2467 | } | |
2468 | ||
2469 | int main(int argc, char *argv[]) | |
2470 | { | |
2471 | int ret = 0; | |
2472 | ||
2473 | setlocale(LC_NUMERIC, "en_US"); | |
2474 | pagesize = getpagesize(); | |
2475 | ncpus = sysconf(_SC_NPROCESSORS_ONLN); | |
2476 | if (ncpus < 0) { | |
2477 | fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed %d/%s\n", | |
2478 | errno, strerror(errno)); | |
2479 | ret = 1; | |
2480 | goto out; | |
2481 | } | |
2482 | ||
2483 | if (handle_args(argc, argv)) { | |
2484 | ret = 1; | |
2485 | goto out; | |
bc39777c JA |
2486 | } |
2487 | ||
d0ca268b JA |
2488 | signal(SIGINT, handle_sigint); |
2489 | signal(SIGHUP, handle_sigint); | |
2490 | signal(SIGTERM, handle_sigint); | |
ece238a6 | 2491 | signal(SIGALRM, handle_sigint); |
38e1f0c6 | 2492 | signal(SIGPIPE, SIG_IGN); |
d0ca268b | 2493 | |
3fe0b570 AB |
2494 | if (kill_running_trace) { |
2495 | struct devpath *dpp; | |
2496 | struct list_head *p; | |
8e86c98a | 2497 | |
3fe0b570 AB |
2498 | __list_for_each(p, &devpaths) { |
2499 | dpp = list_entry(p, struct devpath, head); | |
2500 | if (__stop_trace(dpp->fd)) { | |
2501 | fprintf(stderr, | |
2502 | "BLKTRACETEARDOWN %s failed: %d/%s\n", | |
2503 | dpp->path, errno, strerror(errno)); | |
2504 | } | |
2505 | } | |
2506 | } else if (net_mode == Net_server) { | |
2507 | if (output_name) { | |
2508 | fprintf(stderr, "-o ignored in server mode\n"); | |
2509 | output_name = NULL; | |
2510 | } | |
8e86c98a | 2511 | |
3fe0b570 AB |
2512 | ret = net_server(); |
2513 | } else { | |
2514 | atexit(exit_tracing); | |
830fd65c | 2515 | |
3fe0b570 AB |
2516 | if (net_mode == Net_client) |
2517 | printf("blktrace: connecting to %s\n", hostname); | |
ece238a6 | 2518 | |
3fe0b570 | 2519 | setup_buts(); |
d0ca268b | 2520 | |
3fe0b570 AB |
2521 | if (use_tracer_devpaths()) { |
2522 | if (setup_tracer_devpaths()) | |
2523 | goto out; | |
d0ca268b | 2524 | |
3fe0b570 AB |
2525 | if (piped_output) |
2526 | handle_list = handle_list_file; | |
2527 | else | |
2528 | handle_list = handle_list_net; | |
2529 | } | |
eb3c8108 | 2530 | |
3fe0b570 AB |
2531 | ntracers = start_tracers(); |
2532 | if (ntracers != ncpus) | |
2533 | stop_tracers(); | |
2534 | else { | |
2535 | if (net_mode == Net_client) | |
2536 | printf("blktrace: connected!\n"); | |
2537 | if (stop_watch) | |
2538 | alarm(stop_watch); | |
2539 | } | |
2540 | ||
2541 | wait_tracers(); | |
2542 | if (ntracers == ncpus) | |
2543 | show_stats(&devpaths); | |
2544 | ||
2545 | if (net_client_use_send()) | |
2546 | close_client_connections(); | |
2547 | del_tracers(); | |
2548 | } | |
d0ca268b | 2549 | |
3fe0b570 AB |
2550 | out: |
2551 | if (pfp) | |
2552 | fclose(pfp); | |
2553 | rel_devpaths(); | |
2554 | return ret; | |
2555 | } |