projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
server: rewrite message handling
[fio.git] / server.c
... / ...
CommitLineData
1#include <stdio.h>
2#include <stdlib.h>
3#include <stdarg.h>
4#include <unistd.h>
5#include <limits.h>
6#include <errno.h>
7#include <sys/poll.h>
8#include <sys/types.h>
9#include <sys/wait.h>
10#include <sys/socket.h>
11#include <sys/stat.h>
12#include <sys/un.h>
13#include <sys/uio.h>
14#include <netinet/in.h>
15#include <arpa/inet.h>
16#include <netdb.h>
17#include <syslog.h>
18#include <signal.h>
19#ifdef CONFIG_ZLIB
20#include <zlib.h>
21#endif
22
23#include "fio.h"
24#include "options.h"
25#include "server.h"
26#include "crc/crc16.h"
27#include "lib/ieee754.h"
28#include "verify.h"
29#include "smalloc.h"
30
31int fio_net_port = FIO_NET_PORT;
32
33int exit_backend = 0;
34
35enum {
36 SK_F_FREE = 1,
37 SK_F_COPY = 2,
38 SK_F_SIMPLE = 4,
39};
40
41struct sk_entry {
42 struct flist_head list;
43 int opcode;
44 void *buf;
45 off_t size;
46 uint64_t *tagptr;
47 int flags;
48};
49
50struct sk_out {
51 int sk;
52 struct fio_mutex *lock;
53 struct flist_head list;
54 struct fio_mutex *wait;
55};
56
57static char *fio_server_arg;
58static char *bind_sock;
59static struct sockaddr_in saddr_in;
60static struct sockaddr_in6 saddr_in6;
61static int use_ipv6;
62#ifdef CONFIG_ZLIB
63static unsigned int has_zlib = 1;
64#else
65static unsigned int has_zlib = 0;
66#endif
67static unsigned int use_zlib;
68static char me[128];
69
70static pthread_key_t sk_out_key;
71
72struct fio_fork_item {
73 struct flist_head list;
74 int exitval;
75 int signal;
76 int exited;
77 pid_t pid;
78};
79
80struct cmd_reply {
81 struct fio_mutex lock;
82 void *data;
83 size_t size;
84 int error;
85};
86
87static const char *fio_server_ops[FIO_NET_CMD_NR] = {
88 "",
89 "QUIT",
90 "EXIT",
91 "JOB",
92 "JOBLINE",
93 "TEXT",
94 "TS",
95 "GS",
96 "SEND_ETA",
97 "ETA",
98 "PROBE",
99 "START",
100 "STOP",
101 "DISK_UTIL",
102 "SERVER_START",
103 "ADD_JOB",
104 "RUN",
105 "IOLOG",
106 "UPDATE_JOB",
107 "LOAD_FILE",
108 "VTRIGGER",
109 "SENDFILE",
110};
111
112static void sk_lock(struct sk_out *sk_out)
113{
114 fio_mutex_down(sk_out->lock);
115}
116
117static void sk_unlock(struct sk_out *sk_out)
118{
119 fio_mutex_up(sk_out->lock);
120}
121
122const char *fio_server_op(unsigned int op)
123{
124 static char buf[32];
125
126 if (op < FIO_NET_CMD_NR)
127 return fio_server_ops[op];
128
129 sprintf(buf, "UNKNOWN/%d", op);
130 return buf;
131}
132
133static ssize_t iov_total_len(const struct iovec *iov, int count)
134{
135 ssize_t ret = 0;
136
137 while (count--) {
138 ret += iov->iov_len;
139 iov++;
140 }
141
142 return ret;
143}
144
145static int fio_sendv_data(int sk, struct iovec *iov, int count)
146{
147 ssize_t total_len = iov_total_len(iov, count);
148 ssize_t ret;
149
150 do {
151 ret = writev(sk, iov, count);
152 if (ret > 0) {
153 total_len -= ret;
154 if (!total_len)
155 break;
156
157 while (ret) {
158 if (ret >= iov->iov_len) {
159 ret -= iov->iov_len;
160 iov++;
161 continue;
162 }
163 iov->iov_base += ret;
164 iov->iov_len -= ret;
165 ret = 0;
166 }
167 } else if (!ret)
168 break;
169 else if (errno == EAGAIN || errno == EINTR)
170 continue;
171 else
172 break;
173 } while (!exit_backend);
174
175 if (!total_len)
176 return 0;
177
178 if (errno)
179 return -errno;
180
181 return 1;
182}
183
184int fio_send_data(int sk, const void *p, unsigned int len)
185{
186 struct iovec iov = { .iov_base = (void *) p, .iov_len = len };
187
188 assert(len <= sizeof(struct fio_net_cmd) + FIO_SERVER_MAX_FRAGMENT_PDU);
189
190 return fio_sendv_data(sk, &iov, 1);
191}
192
193int fio_recv_data(int sk, void *p, unsigned int len)
194{
195 do {
196 int ret = recv(sk, p, len, MSG_WAITALL);
197
198 if (ret > 0) {
199 len -= ret;
200 if (!len)
201 break;
202 p += ret;
203 continue;
204 } else if (!ret)
205 break;
206 else if (errno == EAGAIN || errno == EINTR)
207 continue;
208 else
209 break;
210 } while (!exit_backend);
211
212 if (!len)
213 return 0;
214
215 return -1;
216}
217
218static int verify_convert_cmd(struct fio_net_cmd *cmd)
219{
220 uint16_t crc;
221
222 cmd->cmd_crc16 = le16_to_cpu(cmd->cmd_crc16);
223 cmd->pdu_crc16 = le16_to_cpu(cmd->pdu_crc16);
224
225 crc = fio_crc16(cmd, FIO_NET_CMD_CRC_SZ);
226 if (crc != cmd->cmd_crc16) {
227 log_err("fio: server bad crc on command (got %x, wanted %x)\n",
228 cmd->cmd_crc16, crc);
229 return 1;
230 }
231
232 cmd->version = le16_to_cpu(cmd->version);
233 cmd->opcode = le16_to_cpu(cmd->opcode);
234 cmd->flags = le32_to_cpu(cmd->flags);
235 cmd->tag = le64_to_cpu(cmd->tag);
236 cmd->pdu_len = le32_to_cpu(cmd->pdu_len);
237
238 switch (cmd->version) {
239 case FIO_SERVER_VER:
240 break;
241 default:
242 log_err("fio: bad server cmd version %d\n", cmd->version);
243 return 1;
244 }
245
246 if (cmd->pdu_len > FIO_SERVER_MAX_FRAGMENT_PDU) {
247 log_err("fio: command payload too large: %u\n", cmd->pdu_len);
248 return 1;
249 }
250
251 return 0;
252}
253
254/*
255 * Read (and defragment, if necessary) incoming commands
256 */
257struct fio_net_cmd *fio_net_recv_cmd(int sk)
258{
259 struct fio_net_cmd cmd, *tmp, *cmdret = NULL;
260 size_t cmd_size = 0, pdu_offset = 0;
261 uint16_t crc;
262 int ret, first = 1;
263 void *pdu = NULL;
264
265 do {
266 ret = fio_recv_data(sk, &cmd, sizeof(cmd));
267 if (ret)
268 break;
269
270 /* We have a command, verify it and swap if need be */
271 ret = verify_convert_cmd(&cmd);
272 if (ret)
273 break;
274
275 if (first) {
276 /* if this is text, add room for \0 at the end */
277 cmd_size = sizeof(cmd) + cmd.pdu_len + 1;
278 assert(!cmdret);
279 } else
280 cmd_size += cmd.pdu_len;
281
282 if (cmd_size / 1024 > FIO_SERVER_MAX_CMD_MB * 1024) {
283 log_err("fio: cmd+pdu too large (%llu)\n", (unsigned long long) cmd_size);
284 ret = 1;
285 break;
286 }
287
288 tmp = realloc(cmdret, cmd_size);
289 if (!tmp) {
290 log_err("fio: server failed allocating cmd\n");
291 ret = 1;
292 break;
293 }
294 cmdret = tmp;
295
296 if (first)
297 memcpy(cmdret, &cmd, sizeof(cmd));
298 else if (cmdret->opcode != cmd.opcode) {
299 log_err("fio: fragment opcode mismatch (%d != %d)\n",
300 cmdret->opcode, cmd.opcode);
301 ret = 1;
302 break;
303 }
304
305 if (!cmd.pdu_len)
306 break;
307
308 /* There's payload, get it */
309 pdu = (void *) cmdret->payload + pdu_offset;
310 ret = fio_recv_data(sk, pdu, cmd.pdu_len);
311 if (ret)
312 break;
313
314 /* Verify payload crc */
315 crc = fio_crc16(pdu, cmd.pdu_len);
316 if (crc != cmd.pdu_crc16) {
317 log_err("fio: server bad crc on payload ");
318 log_err("(got %x, wanted %x)\n", cmd.pdu_crc16, crc);
319 ret = 1;
320 break;
321 }
322
323 pdu_offset += cmd.pdu_len;
324 if (!first)
325 cmdret->pdu_len += cmd.pdu_len;
326 first = 0;
327 } while (cmd.flags & FIO_NET_CMD_F_MORE);
328
329 if (ret) {
330 free(cmdret);
331 cmdret = NULL;
332 } else if (cmdret) {
333 /* zero-terminate text input */
334 if (cmdret->pdu_len) {
335 if (cmdret->opcode == FIO_NET_CMD_TEXT) {
336 struct cmd_text_pdu *__pdu = (struct cmd_text_pdu *) cmdret->payload;
337 char *buf = (char *) __pdu->buf;
338
339 buf[__pdu->buf_len] = '\0';
340 } else if (cmdret->opcode == FIO_NET_CMD_JOB) {
341 struct cmd_job_pdu *__pdu = (struct cmd_job_pdu *) cmdret->payload;
342 char *buf = (char *) __pdu->buf;
343 int len = le32_to_cpu(__pdu->buf_len);
344
345 buf[len] = '\0';
346 }
347 }
348
349 /* frag flag is internal */
350 cmdret->flags &= ~FIO_NET_CMD_F_MORE;
351 }
352
353 return cmdret;
354}
355
356static void add_reply(uint64_t tag, struct flist_head *list)
357{
358 struct fio_net_cmd_reply *reply;
359
360 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
361 flist_add_tail(&reply->list, list);
362}
363
364static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
365{
366 struct fio_net_cmd_reply *reply;
367
368 reply = calloc(1, sizeof(*reply));
369 INIT_FLIST_HEAD(&reply->list);
370 fio_gettime(&reply->tv, NULL);
371 reply->saved_tag = tag;
372 reply->opcode = opcode;
373
374 return (uintptr_t) reply;
375}
376
377static void free_reply(uint64_t tag)
378{
379 struct fio_net_cmd_reply *reply;
380
381 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
382 free(reply);
383}
384
385void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu)
386{
387 uint32_t pdu_len;
388
389 cmd->cmd_crc16 = __cpu_to_le16(fio_crc16(cmd, FIO_NET_CMD_CRC_SZ));
390
391 pdu_len = le32_to_cpu(cmd->pdu_len);
392 cmd->pdu_crc16 = __cpu_to_le16(fio_crc16(pdu, pdu_len));
393}
394
395void fio_net_cmd_crc(struct fio_net_cmd *cmd)
396{
397 fio_net_cmd_crc_pdu(cmd, cmd->payload);
398}
399
400int fio_net_send_cmd(int fd, uint16_t opcode, const void *buf, off_t size,
401 uint64_t *tagptr, struct flist_head *list)
402{
403 struct fio_net_cmd *cmd = NULL;
404 size_t this_len, cur_len = 0;
405 uint64_t tag;
406 int ret;
407
408 if (list) {
409 assert(tagptr);
410 tag = *tagptr = alloc_reply(*tagptr, opcode);
411 } else
412 tag = tagptr ? *tagptr : 0;
413
414 do {
415 this_len = size;
416 if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU)
417 this_len = FIO_SERVER_MAX_FRAGMENT_PDU;
418
419 if (!cmd || cur_len < sizeof(*cmd) + this_len) {
420 if (cmd)
421 free(cmd);
422
423 cur_len = sizeof(*cmd) + this_len;
424 cmd = malloc(cur_len);
425 }
426
427 fio_init_net_cmd(cmd, opcode, buf, this_len, tag);
428
429 if (this_len < size)
430 cmd->flags = __cpu_to_le32(FIO_NET_CMD_F_MORE);
431
432 fio_net_cmd_crc(cmd);
433
434 ret = fio_send_data(fd, cmd, sizeof(*cmd) + this_len);
435 size -= this_len;
436 buf += this_len;
437 } while (!ret && size);
438
439 if (list) {
440 if (ret)
441 free_reply(tag);
442 else
443 add_reply(tag, list);
444 }
445
446 if (cmd)
447 free(cmd);
448
449 return ret;
450}
451
452static int fio_net_queue_cmd(uint16_t opcode, void *buf, off_t size,
453 uint64_t *tagptr, int flags)
454{
455 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
456 struct sk_entry *entry;
457
458 entry = smalloc(sizeof(*entry));
459 entry->opcode = opcode;
460 if (flags & SK_F_COPY) {
461 entry->buf = smalloc(size);
462 memcpy(entry->buf, buf, size);
463 } else
464 entry->buf = buf;
465 entry->size = size;
466 entry->tagptr = tagptr;
467 entry->flags = flags;
468
469 sk_lock(sk_out);
470 flist_add_tail(&entry->list, &sk_out->list);
471 sk_unlock(sk_out);
472
473 fio_mutex_up(sk_out->wait);
474
475 return 0;
476}
477
478static int fio_net_send_simple_stack_cmd(int sk, uint16_t opcode, uint64_t tag)
479{
480 struct fio_net_cmd cmd;
481
482 fio_init_net_cmd(&cmd, opcode, NULL, 0, tag);
483 fio_net_cmd_crc(&cmd);
484
485 return fio_send_data(sk, &cmd, sizeof(cmd));
486}
487
488/*
489 * If 'list' is non-NULL, then allocate and store the sent command for
490 * later verification.
491 */
492int fio_net_send_simple_cmd(int sk, uint16_t opcode, uint64_t tag,
493 struct flist_head *list)
494{
495 int ret;
496
497 if (list)
498 tag = alloc_reply(tag, opcode);
499
500 ret = fio_net_send_simple_stack_cmd(sk, opcode, tag);
501 if (ret) {
502 if (list)
503 free_reply(tag);
504
505 return ret;
506 }
507
508 if (list)
509 add_reply(tag, list);
510
511 return 0;
512}
513
514static int fio_net_queue_quit(void)
515{
516 dprint(FD_NET, "server: sending quit\n");
517
518 return fio_net_queue_cmd(FIO_NET_CMD_QUIT, NULL, 0, 0, SK_F_SIMPLE);
519}
520
521int fio_net_send_quit(int sk)
522{
523 dprint(FD_NET, "server: sending quit\n");
524
525 return fio_net_send_simple_cmd(sk, FIO_NET_CMD_QUIT, 0, NULL);
526}
527
528static int fio_net_send_ack(struct fio_net_cmd *cmd, int error, int signal)
529{
530 struct cmd_end_pdu epdu;
531 uint64_t tag = 0;
532
533 if (cmd)
534 tag = cmd->tag;
535
536 epdu.error = __cpu_to_le32(error);
537 epdu.signal = __cpu_to_le32(signal);
538 return fio_net_queue_cmd(FIO_NET_CMD_STOP, &epdu, sizeof(epdu), &tag, SK_F_COPY);
539}
540
541static int fio_net_queue_stop(int error, int signal)
542{
543 dprint(FD_NET, "server: sending stop (%d, %d)\n", error, signal);
544 return fio_net_send_ack(NULL, error, signal);
545}
546
547static void fio_server_add_fork_item(pid_t pid, struct flist_head *list)
548{
549 struct fio_fork_item *ffi;
550
551 ffi = malloc(sizeof(*ffi));
552 ffi->exitval = 0;
553 ffi->signal = 0;
554 ffi->exited = 0;
555 ffi->pid = pid;
556 flist_add_tail(&ffi->list, list);
557}
558
559static void fio_server_add_conn_pid(struct flist_head *conn_list, pid_t pid)
560{
561 dprint(FD_NET, "server: forked off connection job (pid=%u)\n", (int) pid);
562 fio_server_add_fork_item(pid, conn_list);
563}
564
565static void fio_server_add_job_pid(struct flist_head *job_list, pid_t pid)
566{
567 dprint(FD_NET, "server: forked off job job (pid=%u)\n", (int) pid);
568 fio_server_add_fork_item(pid, job_list);
569}
570
571static void fio_server_check_fork_item(struct fio_fork_item *ffi)
572{
573 int ret, status;
574
575 ret = waitpid(ffi->pid, &status, WNOHANG);
576 if (ret < 0) {
577 if (errno == ECHILD) {
578 log_err("fio: connection pid %u disappeared\n", (int) ffi->pid);
579 ffi->exited = 1;
580 } else
581 log_err("fio: waitpid: %s\n", strerror(errno));
582 } else if (ret == ffi->pid) {
583 if (WIFSIGNALED(status)) {
584 ffi->signal = WTERMSIG(status);
585 ffi->exited = 1;
586 }
587 if (WIFEXITED(status)) {
588 if (WEXITSTATUS(status))
589 ffi->exitval = WEXITSTATUS(status);
590 ffi->exited = 1;
591 }
592 }
593}
594
595static void fio_server_fork_item_done(struct fio_fork_item *ffi, bool stop)
596{
597 dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", (int) ffi->pid, ffi->signal, ffi->exitval);
598
599 /*
600 * Fold STOP and QUIT...
601 */
602 if (stop) {
603 fio_net_queue_stop(ffi->exitval, ffi->signal);
604 fio_net_queue_quit();
605 }
606
607 flist_del(&ffi->list);
608 free(ffi);
609}
610
611static void fio_server_check_fork_items(struct flist_head *list, bool stop)
612{
613 struct flist_head *entry, *tmp;
614 struct fio_fork_item *ffi;
615
616 flist_for_each_safe(entry, tmp, list) {
617 ffi = flist_entry(entry, struct fio_fork_item, list);
618
619 fio_server_check_fork_item(ffi);
620
621 if (ffi->exited)
622 fio_server_fork_item_done(ffi, stop);
623 }
624}
625
626static void fio_server_check_jobs(struct flist_head *job_list)
627{
628 fio_server_check_fork_items(job_list, true);
629}
630
631static void fio_server_check_conns(struct flist_head *conn_list)
632{
633 fio_server_check_fork_items(conn_list, false);
634}
635
636static int handle_load_file_cmd(struct fio_net_cmd *cmd)
637{
638 struct cmd_load_file_pdu *pdu = (struct cmd_load_file_pdu *) cmd->payload;
639 void *file_name = pdu->file;
640 struct cmd_start_pdu spdu;
641
642 dprint(FD_NET, "server: loading local file %s\n", (char *) file_name);
643
644 pdu->name_len = le16_to_cpu(pdu->name_len);
645 pdu->client_type = le16_to_cpu(pdu->client_type);
646
647 if (parse_jobs_ini(file_name, 0, 0, pdu->client_type)) {
648 fio_net_queue_quit();
649 return -1;
650 }
651
652 spdu.jobs = cpu_to_le32(thread_number);
653 spdu.stat_outputs = cpu_to_le32(stat_number);
654 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
655 return 0;
656}
657
658static int handle_run_cmd(struct sk_out *sk_out, struct flist_head *job_list,
659 struct fio_net_cmd *cmd)
660{
661 struct backend_data data;
662 pid_t pid;
663 int ret;
664
665 fio_time_init();
666 set_genesis_time();
667
668 pid = fork();
669 if (pid) {
670 fio_server_add_job_pid(job_list, pid);
671 return 0;
672 }
673
674 data.key = sk_out_key;
675 data.ptr = sk_out;
676 //pthread_setspecific(sk_out_key, sk_out);
677 ret = fio_backend(&data);
678 free_threads_shm();
679 _exit(ret);
680}
681
682static int handle_job_cmd(struct fio_net_cmd *cmd)
683{
684 struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmd->payload;
685 void *buf = pdu->buf;
686 struct cmd_start_pdu spdu;
687
688 pdu->buf_len = le32_to_cpu(pdu->buf_len);
689 pdu->client_type = le32_to_cpu(pdu->client_type);
690
691 if (parse_jobs_ini(buf, 1, 0, pdu->client_type)) {
692 fio_net_queue_quit();
693 return -1;
694 }
695
696 spdu.jobs = cpu_to_le32(thread_number);
697 spdu.stat_outputs = cpu_to_le32(stat_number);
698
699 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
700 return 0;
701}
702
703static int handle_jobline_cmd(struct fio_net_cmd *cmd)
704{
705 void *pdu = cmd->payload;
706 struct cmd_single_line_pdu *cslp;
707 struct cmd_line_pdu *clp;
708 unsigned long offset;
709 struct cmd_start_pdu spdu;
710 char **argv;
711 int i;
712
713 clp = pdu;
714 clp->lines = le16_to_cpu(clp->lines);
715 clp->client_type = le16_to_cpu(clp->client_type);
716 argv = malloc(clp->lines * sizeof(char *));
717 offset = sizeof(*clp);
718
719 dprint(FD_NET, "server: %d command line args\n", clp->lines);
720
721 for (i = 0; i < clp->lines; i++) {
722 cslp = pdu + offset;
723 argv[i] = (char *) cslp->text;
724
725 offset += sizeof(*cslp) + le16_to_cpu(cslp->len);
726 dprint(FD_NET, "server: %d: %s\n", i, argv[i]);
727 }
728
729 if (parse_cmd_line(clp->lines, argv, clp->client_type)) {
730 fio_net_queue_quit();
731 free(argv);
732 return -1;
733 }
734
735 free(argv);
736
737 spdu.jobs = cpu_to_le32(thread_number);
738 spdu.stat_outputs = cpu_to_le32(stat_number);
739
740 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
741 return 0;
742}
743
744static int handle_probe_cmd(struct fio_net_cmd *cmd)
745{
746 struct cmd_client_probe_pdu *pdu = (struct cmd_client_probe_pdu *) cmd->payload;
747 struct cmd_probe_reply_pdu probe;
748 uint64_t tag = cmd->tag;
749
750 dprint(FD_NET, "server: sending probe reply\n");
751
752 strcpy(me, (char *) pdu->server);
753
754 memset(&probe, 0, sizeof(probe));
755 gethostname((char *) probe.hostname, sizeof(probe.hostname));
756#ifdef CONFIG_BIG_ENDIAN
757 probe.bigendian = 1;
758#endif
759 strncpy((char *) probe.fio_version, fio_version_string, sizeof(probe.fio_version));
760
761 probe.os = FIO_OS;
762 probe.arch = FIO_ARCH;
763 probe.bpp = sizeof(void *);
764 probe.cpus = __cpu_to_le32(cpus_online());
765
766 /*
767 * If the client supports compression and we do too, then enable it
768 */
769 if (has_zlib && le64_to_cpu(pdu->flags) & FIO_PROBE_FLAG_ZLIB) {
770 probe.flags = __cpu_to_le64(FIO_PROBE_FLAG_ZLIB);
771 use_zlib = 1;
772 } else {
773 probe.flags = 0;
774 use_zlib = 0;
775 }
776
777 return fio_net_queue_cmd(FIO_NET_CMD_PROBE, &probe, sizeof(probe), &tag, SK_F_COPY);
778}
779
780static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
781{
782 struct jobs_eta *je;
783 uint64_t tag = cmd->tag;
784 size_t size;
785 int i;
786
787 dprint(FD_NET, "server sending status\n");
788
789 /*
790 * Fake ETA return if we don't have a local one, otherwise the client
791 * will end up timing out waiting for a response to the ETA request
792 */
793 je = get_jobs_eta(true, &size);
794 if (!je) {
795 size = sizeof(*je);
796 je = calloc(1, size);
797 } else {
798 je->nr_running = cpu_to_le32(je->nr_running);
799 je->nr_ramp = cpu_to_le32(je->nr_ramp);
800 je->nr_pending = cpu_to_le32(je->nr_pending);
801 je->nr_setting_up = cpu_to_le32(je->nr_setting_up);
802 je->files_open = cpu_to_le32(je->files_open);
803
804 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
805 je->m_rate[i] = cpu_to_le32(je->m_rate[i]);
806 je->t_rate[i] = cpu_to_le32(je->t_rate[i]);
807 je->m_iops[i] = cpu_to_le32(je->m_iops[i]);
808 je->t_iops[i] = cpu_to_le32(je->t_iops[i]);
809 je->rate[i] = cpu_to_le32(je->rate[i]);
810 je->iops[i] = cpu_to_le32(je->iops[i]);
811 }
812
813 je->elapsed_sec = cpu_to_le64(je->elapsed_sec);
814 je->eta_sec = cpu_to_le64(je->eta_sec);
815 je->nr_threads = cpu_to_le32(je->nr_threads);
816 je->is_pow2 = cpu_to_le32(je->is_pow2);
817 je->unit_base = cpu_to_le32(je->unit_base);
818 }
819
820 fio_net_queue_cmd(FIO_NET_CMD_ETA, je, size, &tag, SK_F_FREE);
821 return 0;
822}
823
824static int send_update_job_reply(uint64_t __tag, int error)
825{
826 uint64_t tag = __tag;
827 uint32_t pdu_error;
828
829 pdu_error = __cpu_to_le32(error);
830 return fio_net_queue_cmd(FIO_NET_CMD_UPDATE_JOB, &pdu_error, sizeof(pdu_error), &tag, SK_F_COPY);
831}
832
833static int handle_update_job_cmd(struct fio_net_cmd *cmd)
834{
835 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
836 struct thread_data *td;
837 uint32_t tnumber;
838
839 tnumber = le32_to_cpu(pdu->thread_number);
840
841 dprint(FD_NET, "server: updating options for job %u\n", tnumber);
842
843 if (!tnumber || tnumber > thread_number) {
844 send_update_job_reply(cmd->tag, ENODEV);
845 return 0;
846 }
847
848 td = &threads[tnumber - 1];
849 convert_thread_options_to_cpu(&td->o, &pdu->top);
850 send_update_job_reply(cmd->tag, 0);
851 return 0;
852}
853
854static int handle_trigger_cmd(struct fio_net_cmd *cmd)
855{
856 struct cmd_vtrigger_pdu *pdu = (struct cmd_vtrigger_pdu *) cmd->payload;
857 char *buf = (char *) pdu->cmd;
858 struct all_io_list *rep;
859 size_t sz;
860
861 pdu->len = le16_to_cpu(pdu->len);
862 buf[pdu->len] = '\0';
863
864 rep = get_all_io_list(IO_LIST_ALL, &sz);
865 if (!rep) {
866 struct all_io_list state;
867
868 state.threads = cpu_to_le64((uint64_t) 0);
869 fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, SK_F_COPY);
870 } else
871 fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, rep, sz, NULL, SK_F_FREE);
872
873 exec_trigger(buf);
874 return 0;
875}
876
877static int handle_command(struct sk_out *sk_out, struct flist_head *job_list,
878 struct fio_net_cmd *cmd)
879{
880 int ret;
881
882 dprint(FD_NET, "server: got op [%s], pdu=%u, tag=%llx\n",
883 fio_server_op(cmd->opcode), cmd->pdu_len,
884 (unsigned long long) cmd->tag);
885
886 switch (cmd->opcode) {
887 case FIO_NET_CMD_QUIT:
888 fio_terminate_threads(TERMINATE_ALL);
889 ret = 0;
890 break;
891 case FIO_NET_CMD_EXIT:
892 exit_backend = 1;
893 return -1;
894 case FIO_NET_CMD_LOAD_FILE:
895 ret = handle_load_file_cmd(cmd);
896 break;
897 case FIO_NET_CMD_JOB:
898 ret = handle_job_cmd(cmd);
899 break;
900 case FIO_NET_CMD_JOBLINE:
901 ret = handle_jobline_cmd(cmd);
902 break;
903 case FIO_NET_CMD_PROBE:
904 ret = handle_probe_cmd(cmd);
905 break;
906 case FIO_NET_CMD_SEND_ETA:
907 ret = handle_send_eta_cmd(cmd);
908 break;
909 case FIO_NET_CMD_RUN:
910 ret = handle_run_cmd(sk_out, job_list, cmd);
911 break;
912 case FIO_NET_CMD_UPDATE_JOB:
913 ret = handle_update_job_cmd(cmd);
914 break;
915 case FIO_NET_CMD_VTRIGGER:
916 ret = handle_trigger_cmd(cmd);
917 break;
918 case FIO_NET_CMD_SENDFILE: {
919 struct cmd_sendfile_reply *in;
920 struct cmd_reply *rep;
921
922 rep = (struct cmd_reply *) (uintptr_t) cmd->tag;
923
924 in = (struct cmd_sendfile_reply *) cmd->payload;
925 in->size = le32_to_cpu(in->size);
926 in->error = le32_to_cpu(in->error);
927 if (in->error) {
928 ret = 1;
929 rep->error = in->error;
930 } else {
931 ret = 0;
932 rep->data = smalloc(in->size);
933 if (!rep->data) {
934 ret = 1;
935 rep->error = ENOMEM;
936 } else {
937 rep->size = in->size;
938 memcpy(rep->data, in->data, in->size);
939 }
940 }
941 fio_mutex_up(&rep->lock);
942 break;
943 }
944 default:
945 log_err("fio: unknown opcode: %s\n", fio_server_op(cmd->opcode));
946 ret = 1;
947 }
948
949 return ret;
950}
951
952static void handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry)
953{
954 if (entry->flags & SK_F_SIMPLE) {
955 uint64_t tag = 0;
956
957 if (entry->tagptr)
958 tag = *entry->tagptr;
959
960 fio_net_send_simple_cmd(sk_out->sk, entry->opcode, tag, NULL);
961 } else
962 fio_net_send_cmd(sk_out->sk, entry->opcode, entry->buf, entry->size, entry->tagptr, NULL);
963
964 if (entry->flags & SK_F_FREE)
965 free(entry->buf);
966 else if (entry->flags & SK_F_COPY)
967 sfree(entry->buf);
968
969 sfree(entry);
970}
971
972static void handle_xmits(struct sk_out *sk_out)
973{
974 struct sk_entry *entry;
975 FLIST_HEAD(list);
976
977 sk_lock(sk_out);
978 if (flist_empty(&sk_out->list)) {
979 sk_unlock(sk_out);
980 return;
981 }
982
983 flist_splice_init(&sk_out->list, &list);
984 sk_unlock(sk_out);
985
986 while (!flist_empty(&list)) {
987 entry = flist_entry(list.next, struct sk_entry, list);
988 flist_del(&entry->list);
989 handle_sk_entry(sk_out, entry);
990 }
991}
992
993static int handle_connection(struct sk_out *sk_out)
994{
995 struct fio_net_cmd *cmd = NULL;
996 FLIST_HEAD(job_list);
997 int ret = 0;
998
999 reset_fio_state();
1000
1001 /* read forever */
1002 while (!exit_backend) {
1003 struct pollfd pfd = {
1004 .fd = sk_out->sk,
1005 .events = POLLIN,
1006 };
1007
1008 ret = 0;
1009 do {
1010 int timeout = 1000;
1011
1012 if (!flist_empty(&job_list))
1013 timeout = 100;
1014
1015 handle_xmits(sk_out);
1016
1017 ret = poll(&pfd, 1, 0);
1018 if (ret < 0) {
1019 if (errno == EINTR)
1020 break;
1021 log_err("fio: poll: %s\n", strerror(errno));
1022 break;
1023 } else if (!ret) {
1024 fio_server_check_jobs(&job_list);
1025 fio_mutex_down_timeout(sk_out->wait, timeout);
1026 continue;
1027 }
1028
1029 if (pfd.revents & POLLIN)
1030 break;
1031 if (pfd.revents & (POLLERR|POLLHUP)) {
1032 ret = 1;
1033 break;
1034 }
1035 } while (!exit_backend);
1036
1037 fio_server_check_jobs(&job_list);
1038
1039 if (ret < 0)
1040 break;
1041
1042 cmd = fio_net_recv_cmd(sk_out->sk);
1043 if (!cmd) {
1044 ret = -1;
1045 break;
1046 }
1047
1048 ret = handle_command(sk_out, &job_list, cmd);
1049 if (ret)
1050 break;
1051
1052 free(cmd);
1053 cmd = NULL;
1054 }
1055
1056 if (cmd)
1057 free(cmd);
1058
1059 handle_xmits(sk_out);
1060
1061 close(sk_out->sk);
1062 _exit(ret);
1063}
1064
1065/* get the address on this host bound by the input socket,
1066 * whether it is ipv6 or ipv4 */
1067
1068int get_my_addr_str(int sk)
1069{
1070 struct sockaddr_in6 myaddr6 = { 0, };
1071 struct sockaddr_in myaddr4 = { 0, };
1072 struct sockaddr *sockaddr_p;
1073 char *net_addr;
1074 socklen_t len;
1075 int ret;
1076
1077 if (use_ipv6) {
1078 len = sizeof(myaddr6);
1079 sockaddr_p = (struct sockaddr * )&myaddr6;
1080 net_addr = (char * )&myaddr6.sin6_addr;
1081 } else {
1082 len = sizeof(myaddr4);
1083 sockaddr_p = (struct sockaddr * )&myaddr4;
1084 net_addr = (char * )&myaddr4.sin_addr;
1085 }
1086
1087 ret = getsockname(sk, sockaddr_p, &len);
1088 if (ret) {
1089 log_err("fio: getsockaddr: %s\n", strerror(errno));
1090 return -1;
1091 }
1092
1093 if (!inet_ntop(use_ipv6?AF_INET6:AF_INET, net_addr, client_sockaddr_str, INET6_ADDRSTRLEN - 1)) {
1094 log_err("inet_ntop: failed to convert addr to string\n");
1095 return -1;
1096 }
1097
1098 dprint(FD_NET, "fio server bound to addr %s\n", client_sockaddr_str);
1099 return 0;
1100}
1101
1102static int accept_loop(int listen_sk)
1103{
1104 struct sockaddr_in addr;
1105 struct sockaddr_in6 addr6;
1106 socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr);
1107 struct pollfd pfd;
1108 int ret = 0, sk, exitval = 0;
1109 struct sk_out *sk_out;
1110 FLIST_HEAD(conn_list);
1111
1112 dprint(FD_NET, "server enter accept loop\n");
1113
1114 fio_set_fd_nonblocking(listen_sk, "server");
1115
1116 sk_out = smalloc(sizeof(*sk_out));
1117 INIT_FLIST_HEAD(&sk_out->list);
1118 sk_out->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
1119 sk_out->wait = fio_mutex_init(FIO_MUTEX_LOCKED);
1120
1121 pthread_setspecific(sk_out_key, sk_out);
1122
1123 while (!exit_backend) {
1124 const char *from;
1125 char buf[64];
1126 pid_t pid;
1127
1128 pfd.fd = listen_sk;
1129 pfd.events = POLLIN;
1130 do {
1131 int timeout = 1000;
1132
1133 if (!flist_empty(&conn_list))
1134 timeout = 100;
1135
1136 ret = poll(&pfd, 1, timeout);
1137 if (ret < 0) {
1138 if (errno == EINTR)
1139 break;
1140 log_err("fio: poll: %s\n", strerror(errno));
1141 break;
1142 } else if (!ret) {
1143 fio_server_check_conns(&conn_list);
1144 continue;
1145 }
1146
1147 if (pfd.revents & POLLIN)
1148 break;
1149 } while (!exit_backend);
1150
1151 fio_server_check_conns(&conn_list);
1152
1153 if (exit_backend || ret < 0)
1154 break;
1155
1156 if (use_ipv6)
1157 sk = accept(listen_sk, (struct sockaddr *) &addr6, &len);
1158 else
1159 sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
1160
1161 if (sk < 0) {
1162 log_err("fio: accept: %s\n", strerror(errno));
1163 return -1;
1164 }
1165
1166 if (use_ipv6)
1167 from = inet_ntop(AF_INET6, (struct sockaddr *) &addr6.sin6_addr, buf, sizeof(buf));
1168 else
1169 from = inet_ntop(AF_INET, (struct sockaddr *) &addr.sin_addr, buf, sizeof(buf));
1170
1171 dprint(FD_NET, "server: connect from %s\n", from);
1172
1173 sk_out->sk = sk;
1174
1175 pid = fork();
1176 if (pid) {
1177 close(sk);
1178 fio_server_add_conn_pid(&conn_list, pid);
1179 pthread_setspecific(sk_out_key, sk_out);
1180 continue;
1181 }
1182
1183 /* exits */
1184 get_my_addr_str(sk); /* if error, it's already logged, non-fatal */
1185 handle_connection(sk_out);
1186 }
1187
1188#if 0
1189 fio_mutex_remove(sk_out->lock);
1190 fio_mutex_remove(sk_out->wait);
1191 sfree(sk_out);
1192 pthread_setspecific(sk_out_key, NULL);
1193#endif
1194
1195 return exitval;
1196}
1197
1198int fio_server_text_output(int level, const char *buf, size_t len)
1199{
1200 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
1201 struct cmd_text_pdu *pdu;
1202 unsigned int tlen;
1203 struct timeval tv;
1204
1205 if (!sk_out || sk_out->sk == -1)
1206 return -1;
1207
1208 tlen = sizeof(*pdu) + len;
1209 pdu = malloc(tlen);
1210
1211 pdu->level = __cpu_to_le32(level);
1212 pdu->buf_len = __cpu_to_le32(len);
1213
1214 gettimeofday(&tv, NULL);
1215 pdu->log_sec = __cpu_to_le64(tv.tv_sec);
1216 pdu->log_usec = __cpu_to_le64(tv.tv_usec);
1217
1218 memcpy(pdu->buf, buf, len);
1219
1220 fio_net_queue_cmd(FIO_NET_CMD_TEXT, pdu, tlen, NULL, SK_F_COPY);
1221 free(pdu);
1222 return len;
1223}
1224
1225static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
1226{
1227 dst->max_val = cpu_to_le64(src->max_val);
1228 dst->min_val = cpu_to_le64(src->min_val);
1229 dst->samples = cpu_to_le64(src->samples);
1230
1231 /*
1232 * Encode to IEEE 754 for network transfer
1233 */
1234 dst->mean.u.i = cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
1235 dst->S.u.i = cpu_to_le64(fio_double_to_uint64(src->S.u.f));
1236}
1237
1238static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1239{
1240 int i;
1241
1242 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1243 dst->max_run[i] = cpu_to_le64(src->max_run[i]);
1244 dst->min_run[i] = cpu_to_le64(src->min_run[i]);
1245 dst->max_bw[i] = cpu_to_le64(src->max_bw[i]);
1246 dst->min_bw[i] = cpu_to_le64(src->min_bw[i]);
1247 dst->io_kb[i] = cpu_to_le64(src->io_kb[i]);
1248 dst->agg[i] = cpu_to_le64(src->agg[i]);
1249 }
1250
1251 dst->kb_base = cpu_to_le32(src->kb_base);
1252 dst->unit_base = cpu_to_le32(src->unit_base);
1253 dst->groupid = cpu_to_le32(src->groupid);
1254 dst->unified_rw_rep = cpu_to_le32(src->unified_rw_rep);
1255}
1256
1257/*
1258 * Send a CMD_TS, which packs struct thread_stat and group_run_stats
1259 * into a single payload.
1260 */
1261void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
1262{
1263 struct cmd_ts_pdu p;
1264 int i, j;
1265
1266 dprint(FD_NET, "server sending end stats\n");
1267
1268 memset(&p, 0, sizeof(p));
1269
1270 strncpy(p.ts.name, ts->name, FIO_JOBNAME_SIZE - 1);
1271 strncpy(p.ts.verror, ts->verror, FIO_VERROR_SIZE - 1);
1272 strncpy(p.ts.description, ts->description, FIO_JOBDESC_SIZE - 1);
1273
1274 p.ts.error = cpu_to_le32(ts->error);
1275 p.ts.thread_number = cpu_to_le32(ts->thread_number);
1276 p.ts.groupid = cpu_to_le32(ts->groupid);
1277 p.ts.pid = cpu_to_le32(ts->pid);
1278 p.ts.members = cpu_to_le32(ts->members);
1279 p.ts.unified_rw_rep = cpu_to_le32(ts->unified_rw_rep);
1280
1281 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1282 convert_io_stat(&p.ts.clat_stat[i], &ts->clat_stat[i]);
1283 convert_io_stat(&p.ts.slat_stat[i], &ts->slat_stat[i]);
1284 convert_io_stat(&p.ts.lat_stat[i], &ts->lat_stat[i]);
1285 convert_io_stat(&p.ts.bw_stat[i], &ts->bw_stat[i]);
1286 }
1287
1288 p.ts.usr_time = cpu_to_le64(ts->usr_time);
1289 p.ts.sys_time = cpu_to_le64(ts->sys_time);
1290 p.ts.ctx = cpu_to_le64(ts->ctx);
1291 p.ts.minf = cpu_to_le64(ts->minf);
1292 p.ts.majf = cpu_to_le64(ts->majf);
1293 p.ts.clat_percentiles = cpu_to_le64(ts->clat_percentiles);
1294 p.ts.percentile_precision = cpu_to_le64(ts->percentile_precision);
1295
1296 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
1297 fio_fp64_t *src = &ts->percentile_list[i];
1298 fio_fp64_t *dst = &p.ts.percentile_list[i];
1299
1300 dst->u.i = cpu_to_le64(fio_double_to_uint64(src->u.f));
1301 }
1302
1303 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
1304 p.ts.io_u_map[i] = cpu_to_le32(ts->io_u_map[i]);
1305 p.ts.io_u_submit[i] = cpu_to_le32(ts->io_u_submit[i]);
1306 p.ts.io_u_complete[i] = cpu_to_le32(ts->io_u_complete[i]);
1307 }
1308
1309 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
1310 p.ts.io_u_lat_u[i] = cpu_to_le32(ts->io_u_lat_u[i]);
1311 p.ts.io_u_lat_m[i] = cpu_to_le32(ts->io_u_lat_m[i]);
1312 }
1313
1314 for (i = 0; i < DDIR_RWDIR_CNT; i++)
1315 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1316 p.ts.io_u_plat[i][j] = cpu_to_le32(ts->io_u_plat[i][j]);
1317
1318 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1319 p.ts.total_io_u[i] = cpu_to_le64(ts->total_io_u[i]);
1320 p.ts.short_io_u[i] = cpu_to_le64(ts->short_io_u[i]);
1321 p.ts.drop_io_u[i] = cpu_to_le64(ts->drop_io_u[i]);
1322 }
1323
1324 p.ts.total_submit = cpu_to_le64(ts->total_submit);
1325 p.ts.total_complete = cpu_to_le64(ts->total_complete);
1326
1327 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1328 p.ts.io_bytes[i] = cpu_to_le64(ts->io_bytes[i]);
1329 p.ts.runtime[i] = cpu_to_le64(ts->runtime[i]);
1330 }
1331
1332 p.ts.total_run_time = cpu_to_le64(ts->total_run_time);
1333 p.ts.continue_on_error = cpu_to_le16(ts->continue_on_error);
1334 p.ts.total_err_count = cpu_to_le64(ts->total_err_count);
1335 p.ts.first_error = cpu_to_le32(ts->first_error);
1336 p.ts.kb_base = cpu_to_le32(ts->kb_base);
1337 p.ts.unit_base = cpu_to_le32(ts->unit_base);
1338
1339 p.ts.latency_depth = cpu_to_le32(ts->latency_depth);
1340 p.ts.latency_target = cpu_to_le64(ts->latency_target);
1341 p.ts.latency_window = cpu_to_le64(ts->latency_window);
1342 p.ts.latency_percentile.u.i = cpu_to_le64(fio_double_to_uint64(ts->latency_percentile.u.f));
1343
1344 p.ts.nr_block_infos = le64_to_cpu(ts->nr_block_infos);
1345 for (i = 0; i < p.ts.nr_block_infos; i++)
1346 p.ts.block_infos[i] = le32_to_cpu(ts->block_infos[i]);
1347
1348 convert_gs(&p.rs, rs);
1349
1350 fio_net_queue_cmd(FIO_NET_CMD_TS, &p, sizeof(p), NULL, SK_F_COPY);
1351}
1352
1353void fio_server_send_gs(struct group_run_stats *rs)
1354{
1355 struct group_run_stats gs;
1356
1357 dprint(FD_NET, "server sending group run stats\n");
1358
1359 convert_gs(&gs, rs);
1360 fio_net_queue_cmd(FIO_NET_CMD_GS, &gs, sizeof(gs), NULL, SK_F_COPY);
1361}
1362
1363static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src)
1364{
1365 int i;
1366
1367 for (i = 0; i < 2; i++) {
1368 dst->ios[i] = cpu_to_le64(src->ios[i]);
1369 dst->merges[i] = cpu_to_le64(src->merges[i]);
1370 dst->sectors[i] = cpu_to_le64(src->sectors[i]);
1371 dst->ticks[i] = cpu_to_le64(src->ticks[i]);
1372 }
1373
1374 dst->io_ticks = cpu_to_le64(src->io_ticks);
1375 dst->time_in_queue = cpu_to_le64(src->time_in_queue);
1376 dst->slavecount = cpu_to_le32(src->slavecount);
1377 dst->max_util.u.i = cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
1378}
1379
1380static void convert_dus(struct disk_util_stat *dst, struct disk_util_stat *src)
1381{
1382 int i;
1383
1384 dst->name[FIO_DU_NAME_SZ - 1] = '\0';
1385 strncpy((char *) dst->name, (char *) src->name, FIO_DU_NAME_SZ - 1);
1386
1387 for (i = 0; i < 2; i++) {
1388 dst->s.ios[i] = cpu_to_le64(src->s.ios[i]);
1389 dst->s.merges[i] = cpu_to_le64(src->s.merges[i]);
1390 dst->s.sectors[i] = cpu_to_le64(src->s.sectors[i]);
1391 dst->s.ticks[i] = cpu_to_le64(src->s.ticks[i]);
1392 }
1393
1394 dst->s.io_ticks = cpu_to_le64(src->s.io_ticks);
1395 dst->s.time_in_queue = cpu_to_le64(src->s.time_in_queue);
1396 dst->s.msec = cpu_to_le64(src->s.msec);
1397}
1398
1399void fio_server_send_du(void)
1400{
1401 struct disk_util *du;
1402 struct flist_head *entry;
1403 struct cmd_du_pdu pdu;
1404
1405 dprint(FD_NET, "server: sending disk_util %d\n", !flist_empty(&disk_list));
1406
1407 memset(&pdu, 0, sizeof(pdu));
1408
1409 flist_for_each(entry, &disk_list) {
1410 du = flist_entry(entry, struct disk_util, list);
1411
1412 convert_dus(&pdu.dus, &du->dus);
1413 convert_agg(&pdu.agg, &du->agg);
1414
1415 fio_net_queue_cmd(FIO_NET_CMD_DU, &pdu, sizeof(pdu), NULL, SK_F_COPY);
1416 }
1417}
1418
1419/*
1420 * Send a command with a separate PDU, not inlined in the command
1421 */
1422static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
1423 off_t size, uint64_t tag, uint32_t flags)
1424{
1425 struct fio_net_cmd cmd;
1426 struct iovec iov[2];
1427
1428 iov[0].iov_base = (void *) &cmd;
1429 iov[0].iov_len = sizeof(cmd);
1430 iov[1].iov_base = (void *) buf;
1431 iov[1].iov_len = size;
1432
1433 __fio_init_net_cmd(&cmd, opcode, size, tag);
1434 cmd.flags = __cpu_to_le32(flags);
1435 fio_net_cmd_crc_pdu(&cmd, buf);
1436
1437 return fio_sendv_data(sk, iov, 2);
1438}
1439
1440static int fio_send_iolog_gz(struct sk_out *sk_out, struct cmd_iolog_pdu *pdu,
1441 struct io_log *log)
1442{
1443 int ret = 0;
1444#ifdef CONFIG_ZLIB
1445 z_stream stream;
1446 void *out_pdu;
1447
1448 /*
1449 * Dirty - since the log is potentially huge, compress it into
1450 * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving
1451 * side defragment it.
1452 */
1453 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1454
1455 stream.zalloc = Z_NULL;
1456 stream.zfree = Z_NULL;
1457 stream.opaque = Z_NULL;
1458
1459 if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK) {
1460 ret = 1;
1461 goto err;
1462 }
1463
1464 stream.next_in = (void *) log->log;
1465 stream.avail_in = log->nr_samples * log_entry_sz(log);
1466
1467 do {
1468 unsigned int this_len, flags = 0;
1469
1470 stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1471 stream.next_out = out_pdu;
1472 ret = deflate(&stream, Z_FINISH);
1473 /* may be Z_OK, or Z_STREAM_END */
1474 if (ret < 0)
1475 goto err_zlib;
1476
1477 this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream.avail_out;
1478
1479 if (stream.avail_in)
1480 flags = FIO_NET_CMD_F_MORE;
1481
1482 ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG,
1483 out_pdu, this_len, 0, flags);
1484 if (ret)
1485 goto err_zlib;
1486 } while (stream.avail_in);
1487
1488err_zlib:
1489 deflateEnd(&stream);
1490err:
1491 free(out_pdu);
1492#endif
1493 return ret;
1494}
1495
1496int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
1497{
1498 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
1499 struct cmd_iolog_pdu pdu;
1500 int i, ret = 0;
1501
1502 pdu.nr_samples = cpu_to_le64(log->nr_samples);
1503 pdu.thread_number = cpu_to_le32(td->thread_number);
1504 pdu.log_type = cpu_to_le32(log->log_type);
1505 pdu.compressed = cpu_to_le32(use_zlib);
1506
1507 strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX);
1508 pdu.name[FIO_NET_NAME_MAX - 1] = '\0';
1509
1510 for (i = 0; i < log->nr_samples; i++) {
1511 struct io_sample *s = get_sample(log, i);
1512
1513 s->time = cpu_to_le64(s->time);
1514 s->val = cpu_to_le64(s->val);
1515 s->__ddir = cpu_to_le32(s->__ddir);
1516 s->bs = cpu_to_le32(s->bs);
1517
1518 if (log->log_offset) {
1519 struct io_sample_offset *so = (void *) s;
1520
1521 so->offset = cpu_to_le64(so->offset);
1522 }
1523 }
1524
1525 /*
1526 * Send header first, it's not compressed.
1527 */
1528 ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG, &pdu,
1529 sizeof(pdu), 0, FIO_NET_CMD_F_MORE);
1530 if (ret)
1531 return ret;
1532
1533 /*
1534 * Now send actual log, compress if we can, otherwise just plain
1535 */
1536 if (use_zlib)
1537 ret = fio_send_iolog_gz(sk_out, &pdu, log);
1538 else {
1539 ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG,
1540 log->log, log->nr_samples * log_entry_sz(log),
1541 0, 0);
1542 }
1543
1544 return ret;
1545}
1546
1547void fio_server_send_add_job(struct thread_data *td)
1548{
1549 struct cmd_add_job_pdu pdu;
1550
1551 memset(&pdu, 0, sizeof(pdu));
1552 pdu.thread_number = cpu_to_le32(td->thread_number);
1553 pdu.groupid = cpu_to_le32(td->groupid);
1554 convert_thread_options_to_net(&pdu.top, &td->o);
1555
1556 fio_net_queue_cmd(FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL, SK_F_COPY);
1557}
1558
1559void fio_server_send_start(struct thread_data *td)
1560{
1561 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
1562
1563 assert(sk_out->sk != -1);
1564
1565 fio_net_queue_cmd(FIO_NET_CMD_SERVER_START, NULL, 0, 0, SK_F_SIMPLE);
1566}
1567
1568int fio_server_get_verify_state(const char *name, int threadnumber,
1569 void **datap, int *version)
1570{
1571 struct thread_io_list *s;
1572 struct cmd_sendfile out;
1573 struct cmd_reply *rep;
1574 uint64_t tag;
1575 void *data;
1576
1577 dprint(FD_NET, "server: request verify state\n");
1578
1579 rep = smalloc(sizeof(*rep));
1580 if (!rep) {
1581 log_err("fio: smalloc pool too small\n");
1582 return 1;
1583 }
1584
1585 __fio_mutex_init(&rep->lock, FIO_MUTEX_LOCKED);
1586 rep->data = NULL;
1587 rep->error = 0;
1588
1589 verify_state_gen_name((char *) out.path, sizeof(out.path), name, me,
1590 threadnumber);
1591 tag = (uint64_t) (uintptr_t) rep;
1592 fio_net_queue_cmd(FIO_NET_CMD_SENDFILE, &out, sizeof(out), &tag, SK_F_COPY);
1593
1594 /*
1595 * Wait for the backend to receive the reply
1596 */
1597 if (fio_mutex_down_timeout(&rep->lock, 10000)) {
1598 log_err("fio: timed out waiting for reply\n");
1599 goto fail;
1600 }
1601
1602 if (rep->error) {
1603 log_err("fio: failure on receiving state file: %s\n", strerror(rep->error));
1604fail:
1605 *datap = NULL;
1606 sfree(rep);
1607 fio_net_queue_quit();
1608 return 1;
1609 }
1610
1611 /*
1612 * The format is verify_state_hdr, then thread_io_list. Verify
1613 * the header, and the thread_io_list checksum
1614 */
1615 s = rep->data + sizeof(struct verify_state_hdr);
1616 if (verify_state_hdr(rep->data, s, version))
1617 goto fail;
1618
1619 /*
1620 * Don't need the header from now, copy just the thread_io_list
1621 */
1622 rep->size -= sizeof(struct verify_state_hdr);
1623 data = malloc(rep->size);
1624 memcpy(data, s, rep->size);
1625 *datap = data;
1626
1627 sfree(rep->data);
1628 __fio_mutex_remove(&rep->lock);
1629 sfree(rep);
1630 return 0;
1631}
1632
1633static int fio_init_server_ip(void)
1634{
1635 struct sockaddr *addr;
1636 socklen_t socklen;
1637 char buf[80];
1638 const char *str;
1639 int sk, opt;
1640
1641 if (use_ipv6)
1642 sk = socket(AF_INET6, SOCK_STREAM, 0);
1643 else
1644 sk = socket(AF_INET, SOCK_STREAM, 0);
1645
1646 if (sk < 0) {
1647 log_err("fio: socket: %s\n", strerror(errno));
1648 return -1;
1649 }
1650
1651 opt = 1;
1652 if (setsockopt(sk, SOL_SOCKET, SO_REUSEADDR, (void *)&opt, sizeof(opt)) < 0) {
1653 log_err("fio: setsockopt(REUSEADDR): %s\n", strerror(errno));
1654 close(sk);
1655 return -1;
1656 }
1657#ifdef SO_REUSEPORT
1658 if (setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
1659 log_err("fio: setsockopt(REUSEPORT): %s\n", strerror(errno));
1660 close(sk);
1661 return -1;
1662 }
1663#endif
1664
1665 if (use_ipv6) {
1666 const void *src = &saddr_in6.sin6_addr;
1667
1668 addr = (struct sockaddr *) &saddr_in6;
1669 socklen = sizeof(saddr_in6);
1670 saddr_in6.sin6_family = AF_INET6;
1671 str = inet_ntop(AF_INET6, src, buf, sizeof(buf));
1672 } else {
1673 const void *src = &saddr_in.sin_addr;
1674
1675 addr = (struct sockaddr *) &saddr_in;
1676 socklen = sizeof(saddr_in);
1677 saddr_in.sin_family = AF_INET;
1678 str = inet_ntop(AF_INET, src, buf, sizeof(buf));
1679 }
1680
1681 if (bind(sk, addr, socklen) < 0) {
1682 log_err("fio: bind: %s\n", strerror(errno));
1683 log_info("fio: failed with IPv%c %s\n", use_ipv6 ? '6' : '4', str);
1684 close(sk);
1685 return -1;
1686 }
1687
1688 return sk;
1689}
1690
1691static int fio_init_server_sock(void)
1692{
1693 struct sockaddr_un addr;
1694 socklen_t len;
1695 mode_t mode;
1696 int sk;
1697
1698 sk = socket(AF_UNIX, SOCK_STREAM, 0);
1699 if (sk < 0) {
1700 log_err("fio: socket: %s\n", strerror(errno));
1701 return -1;
1702 }
1703
1704 mode = umask(000);
1705
1706 memset(&addr, 0, sizeof(addr));
1707 addr.sun_family = AF_UNIX;
1708 strncpy(addr.sun_path, bind_sock, sizeof(addr.sun_path) - 1);
1709
1710 len = sizeof(addr.sun_family) + strlen(bind_sock) + 1;
1711
1712 if (bind(sk, (struct sockaddr *) &addr, len) < 0) {
1713 log_err("fio: bind: %s\n", strerror(errno));
1714 close(sk);
1715 return -1;
1716 }
1717
1718 umask(mode);
1719 return sk;
1720}
1721
1722static int fio_init_server_connection(void)
1723{
1724 char bind_str[128];
1725 int sk;
1726
1727 dprint(FD_NET, "starting server\n");
1728
1729 if (!bind_sock)
1730 sk = fio_init_server_ip();
1731 else
1732 sk = fio_init_server_sock();
1733
1734 if (sk < 0)
1735 return sk;
1736
1737 memset(bind_str, 0, sizeof(bind_str));
1738
1739 if (!bind_sock) {
1740 char *p, port[16];
1741 const void *src;
1742 int af;
1743
1744 if (use_ipv6) {
1745 af = AF_INET6;
1746 src = &saddr_in6.sin6_addr;
1747 } else {
1748 af = AF_INET;
1749 src = &saddr_in.sin_addr;
1750 }
1751
1752 p = (char *) inet_ntop(af, src, bind_str, sizeof(bind_str));
1753
1754 sprintf(port, ",%u", fio_net_port);
1755 if (p)
1756 strcat(p, port);
1757 else
1758 strncpy(bind_str, port, sizeof(bind_str) - 1);
1759 } else
1760 strncpy(bind_str, bind_sock, sizeof(bind_str) - 1);
1761
1762 log_info("fio: server listening on %s\n", bind_str);
1763
1764 if (listen(sk, 0) < 0) {
1765 log_err("fio: listen: %s\n", strerror(errno));
1766 close(sk);
1767 return -1;
1768 }
1769
1770 return sk;
1771}
1772
1773int fio_server_parse_host(const char *host, int ipv6, struct in_addr *inp,
1774 struct in6_addr *inp6)
1775
1776{
1777 int ret = 0;
1778
1779 if (ipv6)
1780 ret = inet_pton(AF_INET6, host, inp6);
1781 else
1782 ret = inet_pton(AF_INET, host, inp);
1783
1784 if (ret != 1) {
1785 struct addrinfo hints, *res;
1786
1787 memset(&hints, 0, sizeof(hints));
1788 hints.ai_family = ipv6 ? AF_INET6 : AF_INET;
1789 hints.ai_socktype = SOCK_STREAM;
1790
1791 ret = getaddrinfo(host, NULL, &hints, &res);
1792 if (ret) {
1793 log_err("fio: failed to resolve <%s> (%s)\n", host,
1794 gai_strerror(ret));
1795 return 1;
1796 }
1797
1798 if (ipv6)
1799 memcpy(inp6, &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr, sizeof(*inp6));
1800 else
1801 memcpy(inp, &((struct sockaddr_in *) res->ai_addr)->sin_addr, sizeof(*inp));
1802
1803 ret = 1;
1804 freeaddrinfo(res);
1805 }
1806
1807 return !(ret == 1);
1808}
1809
1810/*
1811 * Parse a host/ip/port string. Reads from 'str'.
1812 *
1813 * Outputs:
1814 *
1815 * For IPv4:
1816 * *ptr is the host, *port is the port, inp is the destination.
1817 * For IPv6:
1818 * *ptr is the host, *port is the port, inp6 is the dest, and *ipv6 is 1.
1819 * For local domain sockets:
1820 * *ptr is the filename, *is_sock is 1.
1821 */
1822int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
1823 int *port, struct in_addr *inp,
1824 struct in6_addr *inp6, int *ipv6)
1825{
1826 const char *host = str;
1827 char *portp;
1828 int lport = 0;
1829
1830 *ptr = NULL;
1831 *is_sock = 0;
1832 *port = fio_net_port;
1833 *ipv6 = 0;
1834
1835 if (!strncmp(str, "sock:", 5)) {
1836 *ptr = strdup(str + 5);
1837 *is_sock = 1;
1838
1839 return 0;
1840 }
1841
1842 /*
1843 * Is it ip:<ip or host>:port
1844 */
1845 if (!strncmp(host, "ip:", 3))
1846 host += 3;
1847 else if (!strncmp(host, "ip4:", 4))
1848 host += 4;
1849 else if (!strncmp(host, "ip6:", 4)) {
1850 host += 4;
1851 *ipv6 = 1;
1852 } else if (host[0] == ':') {
1853 /* String is :port */
1854 host++;
1855 lport = atoi(host);
1856 if (!lport || lport > 65535) {
1857 log_err("fio: bad server port %u\n", lport);
1858 return 1;
1859 }
1860 /* no hostname given, we are done */
1861 *port = lport;
1862 return 0;
1863 }
1864
1865 /*
1866 * If no port seen yet, check if there's a last ',' at the end
1867 */
1868 if (!lport) {
1869 portp = strchr(host, ',');
1870 if (portp) {
1871 *portp = '\0';
1872 portp++;
1873 lport = atoi(portp);
1874 if (!lport || lport > 65535) {
1875 log_err("fio: bad server port %u\n", lport);
1876 return 1;
1877 }
1878 }
1879 }
1880
1881 if (lport)
1882 *port = lport;
1883
1884 if (!strlen(host))
1885 return 0;
1886
1887 *ptr = strdup(host);
1888
1889 if (fio_server_parse_host(*ptr, *ipv6, inp, inp6)) {
1890 free(*ptr);
1891 *ptr = NULL;
1892 return 1;
1893 }
1894
1895 if (*port == 0)
1896 *port = fio_net_port;
1897
1898 return 0;
1899}
1900
1901/*
1902 * Server arg should be one of:
1903 *
1904 * sock:/path/to/socket
1905 * ip:1.2.3.4
1906 * 1.2.3.4
1907 *
1908 * Where sock uses unix domain sockets, and ip binds the server to
1909 * a specific interface. If no arguments are given to the server, it
1910 * uses IP and binds to 0.0.0.0.
1911 *
1912 */
1913static int fio_handle_server_arg(void)
1914{
1915 int port = fio_net_port;
1916 int is_sock, ret = 0;
1917
1918 saddr_in.sin_addr.s_addr = htonl(INADDR_ANY);
1919
1920 if (!fio_server_arg)
1921 goto out;
1922
1923 ret = fio_server_parse_string(fio_server_arg, &bind_sock, &is_sock,
1924 &port, &saddr_in.sin_addr,
1925 &saddr_in6.sin6_addr, &use_ipv6);
1926
1927 if (!is_sock && bind_sock) {
1928 free(bind_sock);
1929 bind_sock = NULL;
1930 }
1931
1932out:
1933 fio_net_port = port;
1934 saddr_in.sin_port = htons(port);
1935 saddr_in6.sin6_port = htons(port);
1936 return ret;
1937}
1938
1939static void sig_int(int sig)
1940{
1941 if (bind_sock)
1942 unlink(bind_sock);
1943}
1944
1945static void set_sig_handlers(void)
1946{
1947 struct sigaction act;
1948
1949 memset(&act, 0, sizeof(act));
1950 act.sa_handler = sig_int;
1951 act.sa_flags = SA_RESTART;
1952 sigaction(SIGINT, &act, NULL);
1953}
1954
1955static int fio_server(void)
1956{
1957 int sk, ret;
1958
1959 dprint(FD_NET, "starting server\n");
1960
1961 if (fio_handle_server_arg())
1962 return -1;
1963
1964 sk = fio_init_server_connection();
1965 if (sk < 0)
1966 return -1;
1967
1968 set_sig_handlers();
1969
1970 if (pthread_key_create(&sk_out_key, NULL))
1971 log_err("fio: can't create sk_out backend key\n");
1972
1973 ret = accept_loop(sk);
1974
1975 close(sk);
1976
1977 if (fio_server_arg) {
1978 free(fio_server_arg);
1979 fio_server_arg = NULL;
1980 }
1981 if (bind_sock)
1982 free(bind_sock);
1983
1984 return ret;
1985}
1986
1987void fio_server_got_signal(int signal)
1988{
1989 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
1990
1991 assert(sk_out);
1992
1993 if (signal == SIGPIPE)
1994 sk_out->sk = -1;
1995 else {
1996 log_info("\nfio: terminating on signal %d\n", signal);
1997 exit_backend = 1;
1998 }
1999}
2000
2001static int check_existing_pidfile(const char *pidfile)
2002{
2003 struct stat sb;
2004 char buf[16];
2005 pid_t pid;
2006 FILE *f;
2007
2008 if (stat(pidfile, &sb))
2009 return 0;
2010
2011 f = fopen(pidfile, "r");
2012 if (!f)
2013 return 0;
2014
2015 if (fread(buf, sb.st_size, 1, f) <= 0) {
2016 fclose(f);
2017 return 1;
2018 }
2019 fclose(f);
2020
2021 pid = atoi(buf);
2022 if (kill(pid, SIGCONT) < 0)
2023 return errno != ESRCH;
2024
2025 return 1;
2026}
2027
2028static int write_pid(pid_t pid, const char *pidfile)
2029{
2030 FILE *fpid;
2031
2032 fpid = fopen(pidfile, "w");
2033 if (!fpid) {
2034 log_err("fio: failed opening pid file %s\n", pidfile);
2035 return 1;
2036 }
2037
2038 fprintf(fpid, "%u\n", (unsigned int) pid);
2039 fclose(fpid);
2040 return 0;
2041}
2042
2043/*
2044 * If pidfile is specified, background us.
2045 */
2046int fio_start_server(char *pidfile)
2047{
2048 pid_t pid;
2049 int ret;
2050
2051#if defined(WIN32)
2052 WSADATA wsd;
2053 WSAStartup(MAKEWORD(2, 2), &wsd);
2054#endif
2055
2056 if (!pidfile)
2057 return fio_server();
2058
2059 if (check_existing_pidfile(pidfile)) {
2060 log_err("fio: pidfile %s exists and server appears alive\n",
2061 pidfile);
2062 free(pidfile);
2063 return -1;
2064 }
2065
2066 pid = fork();
2067 if (pid < 0) {
2068 log_err("fio: failed server fork: %s", strerror(errno));
2069 free(pidfile);
2070 return -1;
2071 } else if (pid) {
2072 ret = write_pid(pid, pidfile);
2073 free(pidfile);
2074 _exit(ret);
2075 }
2076
2077 setsid();
2078 openlog("fio", LOG_NDELAY|LOG_NOWAIT|LOG_PID, LOG_USER);
2079 log_syslog = 1;
2080 close(STDIN_FILENO);
2081 close(STDOUT_FILENO);
2082 close(STDERR_FILENO);
2083 f_out = NULL;
2084 f_err = NULL;
2085
2086 ret = fio_server();
2087
2088 closelog();
2089 unlink(pidfile);
2090 free(pidfile);
2091 return ret;
2092}
2093
2094void fio_server_set_arg(const char *arg)
2095{
2096 fio_server_arg = strdup(arg);
2097}
fio - Flexible IO Tester