10 #include <sys/socket.h>
14 #include <netinet/in.h>
15 #include <arpa/inet.h>
25 #include "crc/crc16.h"
26 #include "lib/ieee754.h"
28 int fio_net_port = FIO_NET_PORT;
32 static int server_fd = -1;
33 static char *fio_server_arg;
34 static char *bind_sock;
35 static struct sockaddr_in saddr_in;
36 static struct sockaddr_in6 saddr_in6;
39 static unsigned int has_zlib = 1;
41 static unsigned int has_zlib = 0;
43 static unsigned int use_zlib;
45 struct fio_fork_item {
46 struct flist_head list;
53 static const char *fio_server_ops[FIO_NET_CMD_NR] = {
74 const char *fio_server_op(unsigned int op)
78 if (op < FIO_NET_CMD_NR)
79 return fio_server_ops[op];
81 sprintf(buf, "UNKNOWN/%d", op);
85 static ssize_t iov_total_len(const struct iovec *iov, int count)
97 static int fio_sendv_data(int sk, struct iovec *iov, int count)
99 ssize_t total_len = iov_total_len(iov, count);
103 ret = writev(sk, iov, count);
110 if (ret >= iov->iov_len) {
115 iov->iov_base += ret;
121 else if (errno == EAGAIN || errno == EINTR)
125 } while (!exit_backend);
136 int fio_send_data(int sk, const void *p, unsigned int len)
138 struct iovec iov = { .iov_base = (void *) p, .iov_len = len };
140 assert(len <= sizeof(struct fio_net_cmd) + FIO_SERVER_MAX_FRAGMENT_PDU);
142 return fio_sendv_data(sk, &iov, 1);
145 int fio_recv_data(int sk, void *p, unsigned int len)
148 int ret = recv(sk, p, len, MSG_WAITALL);
158 else if (errno == EAGAIN || errno == EINTR)
162 } while (!exit_backend);
170 static int verify_convert_cmd(struct fio_net_cmd *cmd)
174 cmd->cmd_crc16 = le16_to_cpu(cmd->cmd_crc16);
175 cmd->pdu_crc16 = le16_to_cpu(cmd->pdu_crc16);
177 crc = fio_crc16(cmd, FIO_NET_CMD_CRC_SZ);
178 if (crc != cmd->cmd_crc16) {
179 log_err("fio: server bad crc on command (got %x, wanted %x)\n",
180 cmd->cmd_crc16, crc);
184 cmd->version = le16_to_cpu(cmd->version);
185 cmd->opcode = le16_to_cpu(cmd->opcode);
186 cmd->flags = le32_to_cpu(cmd->flags);
187 cmd->tag = le64_to_cpu(cmd->tag);
188 cmd->pdu_len = le32_to_cpu(cmd->pdu_len);
190 switch (cmd->version) {
194 log_err("fio: bad server cmd version %d\n", cmd->version);
198 if (cmd->pdu_len > FIO_SERVER_MAX_FRAGMENT_PDU) {
199 log_err("fio: command payload too large: %u\n", cmd->pdu_len);
207 * Read (and defragment, if necessary) incoming commands
209 struct fio_net_cmd *fio_net_recv_cmd(int sk)
211 struct fio_net_cmd cmd, *tmp, *cmdret = NULL;
212 size_t cmd_size = 0, pdu_offset = 0;
218 ret = fio_recv_data(sk, &cmd, sizeof(cmd));
222 /* We have a command, verify it and swap if need be */
223 ret = verify_convert_cmd(&cmd);
228 /* if this is text, add room for \0 at the end */
229 cmd_size = sizeof(cmd) + cmd.pdu_len + 1;
232 cmd_size += cmd.pdu_len;
234 if (cmd_size / 1024 > FIO_SERVER_MAX_CMD_MB * 1024) {
235 log_err("fio: cmd+pdu too large (%llu)\n", (unsigned long long) cmd_size);
240 tmp = realloc(cmdret, cmd_size);
242 log_err("fio: server failed allocating cmd\n");
249 memcpy(cmdret, &cmd, sizeof(cmd));
250 else if (cmdret->opcode != cmd.opcode) {
251 log_err("fio: fragment opcode mismatch (%d != %d)\n",
252 cmdret->opcode, cmd.opcode);
260 /* There's payload, get it */
261 pdu = (void *) cmdret->payload + pdu_offset;
262 ret = fio_recv_data(sk, pdu, cmd.pdu_len);
266 /* Verify payload crc */
267 crc = fio_crc16(pdu, cmd.pdu_len);
268 if (crc != cmd.pdu_crc16) {
269 log_err("fio: server bad crc on payload ");
270 log_err("(got %x, wanted %x)\n", cmd.pdu_crc16, crc);
275 pdu_offset += cmd.pdu_len;
277 cmdret->pdu_len += cmd.pdu_len;
279 } while (cmd.flags & FIO_NET_CMD_F_MORE);
285 /* zero-terminate text input */
286 if (cmdret->pdu_len) {
287 if (cmdret->opcode == FIO_NET_CMD_TEXT) {
288 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmdret->payload;
289 char *buf = (char *) pdu->buf;
291 buf[pdu->buf_len] = '\0';
292 } else if (cmdret->opcode == FIO_NET_CMD_JOB) {
293 struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmdret->payload;
294 char *buf = (char *) pdu->buf;
295 int len = le32_to_cpu(pdu->buf_len);
301 /* frag flag is internal */
302 cmdret->flags &= ~FIO_NET_CMD_F_MORE;
308 static void add_reply(uint64_t tag, struct flist_head *list)
310 struct fio_net_cmd_reply *reply;
312 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
313 flist_add_tail(&reply->list, list);
316 static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
318 struct fio_net_cmd_reply *reply;
320 reply = calloc(1, sizeof(*reply));
321 INIT_FLIST_HEAD(&reply->list);
322 gettimeofday(&reply->tv, NULL);
323 reply->saved_tag = tag;
324 reply->opcode = opcode;
326 return (uintptr_t) reply;
329 static void free_reply(uint64_t tag)
331 struct fio_net_cmd_reply *reply;
333 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
337 void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu)
341 cmd->cmd_crc16 = __cpu_to_le16(fio_crc16(cmd, FIO_NET_CMD_CRC_SZ));
343 pdu_len = le32_to_cpu(cmd->pdu_len);
344 cmd->pdu_crc16 = __cpu_to_le16(fio_crc16(pdu, pdu_len));
347 void fio_net_cmd_crc(struct fio_net_cmd *cmd)
349 fio_net_cmd_crc_pdu(cmd, cmd->payload);
352 int fio_net_send_cmd(int fd, uint16_t opcode, const void *buf, off_t size,
353 uint64_t *tagptr, struct flist_head *list)
355 struct fio_net_cmd *cmd = NULL;
356 size_t this_len, cur_len = 0;
362 tag = *tagptr = alloc_reply(*tagptr, opcode);
364 tag = tagptr ? *tagptr : 0;
368 if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU)
369 this_len = FIO_SERVER_MAX_FRAGMENT_PDU;
371 if (!cmd || cur_len < sizeof(*cmd) + this_len) {
375 cur_len = sizeof(*cmd) + this_len;
376 cmd = malloc(cur_len);
379 fio_init_net_cmd(cmd, opcode, buf, this_len, tag);
382 cmd->flags = __cpu_to_le32(FIO_NET_CMD_F_MORE);
384 fio_net_cmd_crc(cmd);
386 ret = fio_send_data(fd, cmd, sizeof(*cmd) + this_len);
389 } while (!ret && size);
395 add_reply(tag, list);
404 static int fio_net_send_simple_stack_cmd(int sk, uint16_t opcode, uint64_t tag)
406 struct fio_net_cmd cmd;
408 fio_init_net_cmd(&cmd, opcode, NULL, 0, tag);
409 fio_net_cmd_crc(&cmd);
411 return fio_send_data(sk, &cmd, sizeof(cmd));
415 * If 'list' is non-NULL, then allocate and store the sent command for
416 * later verification.
418 int fio_net_send_simple_cmd(int sk, uint16_t opcode, uint64_t tag,
419 struct flist_head *list)
424 tag = alloc_reply(tag, opcode);
426 ret = fio_net_send_simple_stack_cmd(sk, opcode, tag);
435 add_reply(tag, list);
440 int fio_net_send_quit(int sk)
442 dprint(FD_NET, "server: sending quit\n");
444 return fio_net_send_simple_cmd(sk, FIO_NET_CMD_QUIT, 0, NULL);
447 static int fio_net_send_ack(int sk, struct fio_net_cmd *cmd, int error,
450 struct cmd_end_pdu epdu;
456 epdu.error = __cpu_to_le32(error);
457 epdu.signal = __cpu_to_le32(signal);
458 return fio_net_send_cmd(sk, FIO_NET_CMD_STOP, &epdu, sizeof(epdu), &tag, NULL);
461 int fio_net_send_stop(int sk, int error, int signal)
463 dprint(FD_NET, "server: sending stop (%d, %d)\n", error, signal);
464 return fio_net_send_ack(sk, NULL, error, signal);
467 static void fio_server_add_fork_item(pid_t pid, struct flist_head *list)
469 struct fio_fork_item *ffi;
471 ffi = malloc(sizeof(*ffi));
476 flist_add_tail(&ffi->list, list);
479 static void fio_server_add_conn_pid(struct flist_head *conn_list, pid_t pid)
481 dprint(FD_NET, "server: forked off connection job (pid=%u)\n", (int) pid);
482 fio_server_add_fork_item(pid, conn_list);
485 static void fio_server_add_job_pid(struct flist_head *job_list, pid_t pid)
487 dprint(FD_NET, "server: forked off job job (pid=%u)\n", (int) pid);
488 fio_server_add_fork_item(pid, job_list);
491 static void fio_server_check_fork_item(struct fio_fork_item *ffi)
495 ret = waitpid(ffi->pid, &status, WNOHANG);
497 if (errno == ECHILD) {
498 log_err("fio: connection pid %u disappeared\n", (int) ffi->pid);
501 log_err("fio: waitpid: %s\n", strerror(errno));
502 } else if (ret == ffi->pid) {
503 if (WIFSIGNALED(status)) {
504 ffi->signal = WTERMSIG(status);
507 if (WIFEXITED(status)) {
508 if (WEXITSTATUS(status))
509 ffi->exitval = WEXITSTATUS(status);
515 static void fio_server_fork_item_done(struct fio_fork_item *ffi)
517 dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", (int) ffi->pid, ffi->signal, ffi->exitval);
520 * Fold STOP and QUIT...
522 fio_net_send_stop(server_fd, ffi->exitval, ffi->signal);
523 fio_net_send_quit(server_fd);
524 flist_del(&ffi->list);
528 static void fio_server_check_fork_items(struct flist_head *list)
530 struct flist_head *entry, *tmp;
531 struct fio_fork_item *ffi;
533 flist_for_each_safe(entry, tmp, list) {
534 ffi = flist_entry(entry, struct fio_fork_item, list);
536 fio_server_check_fork_item(ffi);
539 fio_server_fork_item_done(ffi);
543 static void fio_server_check_jobs(struct flist_head *job_list)
545 fio_server_check_fork_items(job_list);
548 static void fio_server_check_conns(struct flist_head *conn_list)
550 fio_server_check_fork_items(conn_list);
553 static int handle_run_cmd(struct flist_head *job_list, struct fio_net_cmd *cmd)
562 fio_server_add_job_pid(job_list, pid);
571 static int handle_job_cmd(struct fio_net_cmd *cmd)
573 struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmd->payload;
574 void *buf = pdu->buf;
575 struct cmd_start_pdu spdu;
577 pdu->buf_len = le32_to_cpu(pdu->buf_len);
578 pdu->client_type = le32_to_cpu(pdu->client_type);
580 if (parse_jobs_ini(buf, 1, 0, pdu->client_type)) {
581 fio_net_send_quit(server_fd);
585 spdu.jobs = cpu_to_le32(thread_number);
586 spdu.stat_outputs = cpu_to_le32(stat_number);
587 fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
591 static int handle_jobline_cmd(struct fio_net_cmd *cmd)
593 void *pdu = cmd->payload;
594 struct cmd_single_line_pdu *cslp;
595 struct cmd_line_pdu *clp;
596 unsigned long offset;
597 struct cmd_start_pdu spdu;
602 clp->lines = le16_to_cpu(clp->lines);
603 clp->client_type = le16_to_cpu(clp->client_type);
604 argv = malloc(clp->lines * sizeof(char *));
605 offset = sizeof(*clp);
607 dprint(FD_NET, "server: %d command line args\n", clp->lines);
609 for (i = 0; i < clp->lines; i++) {
611 argv[i] = (char *) cslp->text;
613 offset += sizeof(*cslp) + le16_to_cpu(cslp->len);
614 dprint(FD_NET, "server: %d: %s\n", i, argv[i]);
617 if (parse_cmd_line(clp->lines, argv, clp->client_type)) {
618 fio_net_send_quit(server_fd);
625 spdu.jobs = cpu_to_le32(thread_number);
626 spdu.stat_outputs = cpu_to_le32(stat_number);
627 fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
631 static int handle_probe_cmd(struct fio_net_cmd *cmd)
633 struct cmd_client_probe_pdu *pdu = (struct cmd_client_probe_pdu *) cmd->payload;
634 struct cmd_probe_reply_pdu probe;
635 uint64_t tag = cmd->tag;
637 dprint(FD_NET, "server: sending probe reply\n");
639 memset(&probe, 0, sizeof(probe));
640 gethostname((char *) probe.hostname, sizeof(probe.hostname));
641 #ifdef CONFIG_BIG_ENDIAN
644 strncpy((char *) probe.fio_version, fio_version_string, sizeof(probe.fio_version));
647 probe.arch = FIO_ARCH;
648 probe.bpp = sizeof(void *);
649 probe.cpus = __cpu_to_le32(cpus_online());
652 * If the client supports compression and we do too, then enable it
654 if (has_zlib && le64_to_cpu(pdu->flags) & FIO_PROBE_FLAG_ZLIB) {
655 probe.flags = __cpu_to_le64(FIO_PROBE_FLAG_ZLIB);
662 return fio_net_send_cmd(server_fd, FIO_NET_CMD_PROBE, &probe, sizeof(probe), &tag, NULL);
665 static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
669 uint64_t tag = cmd->tag;
675 size = sizeof(*je) + thread_number * sizeof(char) + 1;
679 if (!calc_thread_status(je, 1)) {
684 dprint(FD_NET, "server sending status\n");
686 je->nr_running = cpu_to_le32(je->nr_running);
687 je->nr_ramp = cpu_to_le32(je->nr_ramp);
688 je->nr_pending = cpu_to_le32(je->nr_pending);
689 je->nr_setting_up = cpu_to_le32(je->nr_setting_up);
690 je->files_open = cpu_to_le32(je->files_open);
692 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
693 je->m_rate[i] = cpu_to_le32(je->m_rate[i]);
694 je->t_rate[i] = cpu_to_le32(je->t_rate[i]);
695 je->m_iops[i] = cpu_to_le32(je->m_iops[i]);
696 je->t_iops[i] = cpu_to_le32(je->t_iops[i]);
697 je->rate[i] = cpu_to_le32(je->rate[i]);
698 je->iops[i] = cpu_to_le32(je->iops[i]);
701 je->elapsed_sec = cpu_to_le64(je->elapsed_sec);
702 je->eta_sec = cpu_to_le64(je->eta_sec);
703 je->nr_threads = cpu_to_le32(je->nr_threads);
704 je->is_pow2 = cpu_to_le32(je->is_pow2);
705 je->unit_base = cpu_to_le32(je->unit_base);
707 fio_net_send_cmd(server_fd, FIO_NET_CMD_ETA, je, size, &tag, NULL);
712 static int send_update_job_reply(int fd, uint64_t __tag, int error)
714 uint64_t tag = __tag;
717 pdu_error = __cpu_to_le32(error);
718 return fio_net_send_cmd(fd, FIO_NET_CMD_UPDATE_JOB, &pdu_error, sizeof(pdu_error), &tag, NULL);
721 static int handle_update_job_cmd(struct fio_net_cmd *cmd)
723 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
724 struct thread_data *td;
727 tnumber = le32_to_cpu(pdu->thread_number);
729 dprint(FD_NET, "server: updating options for job %u\n", tnumber);
731 if (!tnumber || tnumber > thread_number) {
732 send_update_job_reply(server_fd, cmd->tag, ENODEV);
736 td = &threads[tnumber - 1];
737 convert_thread_options_to_cpu(&td->o, &pdu->top);
738 send_update_job_reply(server_fd, cmd->tag, 0);
742 static int handle_command(struct flist_head *job_list, struct fio_net_cmd *cmd)
746 dprint(FD_NET, "server: got op [%s], pdu=%u, tag=%llx\n",
747 fio_server_op(cmd->opcode), cmd->pdu_len,
748 (unsigned long long) cmd->tag);
750 switch (cmd->opcode) {
751 case FIO_NET_CMD_QUIT:
752 fio_terminate_threads(TERMINATE_ALL);
754 case FIO_NET_CMD_EXIT:
757 case FIO_NET_CMD_JOB:
758 ret = handle_job_cmd(cmd);
760 case FIO_NET_CMD_JOBLINE:
761 ret = handle_jobline_cmd(cmd);
763 case FIO_NET_CMD_PROBE:
764 ret = handle_probe_cmd(cmd);
766 case FIO_NET_CMD_SEND_ETA:
767 ret = handle_send_eta_cmd(cmd);
769 case FIO_NET_CMD_RUN:
770 ret = handle_run_cmd(job_list, cmd);
772 case FIO_NET_CMD_UPDATE_JOB:
773 ret = handle_update_job_cmd(cmd);
776 log_err("fio: unknown opcode: %s\n", fio_server_op(cmd->opcode));
783 static int handle_connection(int sk)
785 struct fio_net_cmd *cmd = NULL;
786 FLIST_HEAD(job_list);
793 while (!exit_backend) {
794 struct pollfd pfd = {
803 if (!flist_empty(&job_list))
806 ret = poll(&pfd, 1, timeout);
810 log_err("fio: poll: %s\n", strerror(errno));
813 fio_server_check_jobs(&job_list);
817 if (pfd.revents & POLLIN)
819 if (pfd.revents & (POLLERR|POLLHUP)) {
823 } while (!exit_backend);
825 fio_server_check_jobs(&job_list);
830 cmd = fio_net_recv_cmd(sk);
836 ret = handle_command(&job_list, cmd);
851 static int accept_loop(int listen_sk)
853 struct sockaddr_in addr;
854 struct sockaddr_in6 addr6;
855 socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr);
857 int ret = 0, sk, exitval = 0;
858 FLIST_HEAD(conn_list);
860 dprint(FD_NET, "server enter accept loop\n");
862 fio_set_fd_nonblocking(listen_sk, "server");
864 while (!exit_backend) {
874 if (!flist_empty(&conn_list))
877 ret = poll(&pfd, 1, timeout);
881 log_err("fio: poll: %s\n", strerror(errno));
884 fio_server_check_conns(&conn_list);
888 if (pfd.revents & POLLIN)
890 } while (!exit_backend);
892 fio_server_check_conns(&conn_list);
894 if (exit_backend || ret < 0)
898 sk = accept(listen_sk, (struct sockaddr *) &addr6, &len);
900 sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
903 log_err("fio: accept: %s\n", strerror(errno));
908 from = inet_ntop(AF_INET6, (struct sockaddr *) &addr6.sin6_addr, buf, sizeof(buf));
910 from = inet_ntop(AF_INET, (struct sockaddr *) &addr.sin_addr, buf, sizeof(buf));
912 dprint(FD_NET, "server: connect from %s\n", from);
917 fio_server_add_conn_pid(&conn_list, pid);
922 handle_connection(sk);
928 int fio_server_text_output(int level, const char *buf, size_t len)
930 struct cmd_text_pdu *pdu;
935 return log_local_buf(buf, len);
937 tlen = sizeof(*pdu) + len;
940 pdu->level = __cpu_to_le32(level);
941 pdu->buf_len = __cpu_to_le32(len);
943 gettimeofday(&tv, NULL);
944 pdu->log_sec = __cpu_to_le64(tv.tv_sec);
945 pdu->log_usec = __cpu_to_le64(tv.tv_usec);
947 memcpy(pdu->buf, buf, len);
949 fio_net_send_cmd(server_fd, FIO_NET_CMD_TEXT, pdu, tlen, NULL, NULL);
954 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
956 dst->max_val = cpu_to_le64(src->max_val);
957 dst->min_val = cpu_to_le64(src->min_val);
958 dst->samples = cpu_to_le64(src->samples);
961 * Encode to IEEE 754 for network transfer
963 dst->mean.u.i = __cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
964 dst->S.u.i = __cpu_to_le64(fio_double_to_uint64(src->S.u.f));
967 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
971 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
972 dst->max_run[i] = cpu_to_le64(src->max_run[i]);
973 dst->min_run[i] = cpu_to_le64(src->min_run[i]);
974 dst->max_bw[i] = cpu_to_le64(src->max_bw[i]);
975 dst->min_bw[i] = cpu_to_le64(src->min_bw[i]);
976 dst->io_kb[i] = cpu_to_le64(src->io_kb[i]);
977 dst->agg[i] = cpu_to_le64(src->agg[i]);
980 dst->kb_base = cpu_to_le32(src->kb_base);
981 dst->unit_base = cpu_to_le32(src->unit_base);
982 dst->groupid = cpu_to_le32(src->groupid);
983 dst->unified_rw_rep = cpu_to_le32(src->unified_rw_rep);
987 * Send a CMD_TS, which packs struct thread_stat and group_run_stats
988 * into a single payload.
990 void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
995 dprint(FD_NET, "server sending end stats\n");
997 memset(&p, 0, sizeof(p));
999 strncpy(p.ts.name, ts->name, FIO_JOBNAME_SIZE - 1);
1000 strncpy(p.ts.verror, ts->verror, FIO_VERROR_SIZE - 1);
1001 strncpy(p.ts.description, ts->description, FIO_JOBDESC_SIZE - 1);
1003 p.ts.error = cpu_to_le32(ts->error);
1004 p.ts.thread_number = cpu_to_le32(ts->thread_number);
1005 p.ts.groupid = cpu_to_le32(ts->groupid);
1006 p.ts.pid = cpu_to_le32(ts->pid);
1007 p.ts.members = cpu_to_le32(ts->members);
1008 p.ts.unified_rw_rep = cpu_to_le32(ts->unified_rw_rep);
1010 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1011 convert_io_stat(&p.ts.clat_stat[i], &ts->clat_stat[i]);
1012 convert_io_stat(&p.ts.slat_stat[i], &ts->slat_stat[i]);
1013 convert_io_stat(&p.ts.lat_stat[i], &ts->lat_stat[i]);
1014 convert_io_stat(&p.ts.bw_stat[i], &ts->bw_stat[i]);
1017 p.ts.usr_time = cpu_to_le64(ts->usr_time);
1018 p.ts.sys_time = cpu_to_le64(ts->sys_time);
1019 p.ts.ctx = cpu_to_le64(ts->ctx);
1020 p.ts.minf = cpu_to_le64(ts->minf);
1021 p.ts.majf = cpu_to_le64(ts->majf);
1022 p.ts.clat_percentiles = cpu_to_le64(ts->clat_percentiles);
1024 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
1025 fio_fp64_t *src = &ts->percentile_list[i];
1026 fio_fp64_t *dst = &p.ts.percentile_list[i];
1028 dst->u.i = __cpu_to_le64(fio_double_to_uint64(src->u.f));
1031 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
1032 p.ts.io_u_map[i] = cpu_to_le32(ts->io_u_map[i]);
1033 p.ts.io_u_submit[i] = cpu_to_le32(ts->io_u_submit[i]);
1034 p.ts.io_u_complete[i] = cpu_to_le32(ts->io_u_complete[i]);
1037 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
1038 p.ts.io_u_lat_u[i] = cpu_to_le32(ts->io_u_lat_u[i]);
1039 p.ts.io_u_lat_m[i] = cpu_to_le32(ts->io_u_lat_m[i]);
1042 for (i = 0; i < DDIR_RWDIR_CNT; i++)
1043 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1044 p.ts.io_u_plat[i][j] = cpu_to_le32(ts->io_u_plat[i][j]);
1046 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1047 p.ts.total_io_u[i] = cpu_to_le64(ts->total_io_u[i]);
1048 p.ts.short_io_u[i] = cpu_to_le64(ts->short_io_u[i]);
1051 p.ts.total_submit = cpu_to_le64(ts->total_submit);
1052 p.ts.total_complete = cpu_to_le64(ts->total_complete);
1054 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1055 p.ts.io_bytes[i] = cpu_to_le64(ts->io_bytes[i]);
1056 p.ts.runtime[i] = cpu_to_le64(ts->runtime[i]);
1059 p.ts.total_run_time = cpu_to_le64(ts->total_run_time);
1060 p.ts.continue_on_error = cpu_to_le16(ts->continue_on_error);
1061 p.ts.total_err_count = cpu_to_le64(ts->total_err_count);
1062 p.ts.first_error = cpu_to_le32(ts->first_error);
1063 p.ts.kb_base = cpu_to_le32(ts->kb_base);
1064 p.ts.unit_base = cpu_to_le32(ts->unit_base);
1066 p.ts.latency_depth = cpu_to_le32(ts->latency_depth);
1067 p.ts.latency_target = cpu_to_le64(ts->latency_target);
1068 p.ts.latency_window = cpu_to_le64(ts->latency_window);
1069 p.ts.latency_percentile.u.i = __cpu_to_le64(fio_double_to_uint64(ts->latency_percentile.u.f));
1071 convert_gs(&p.rs, rs);
1073 fio_net_send_cmd(server_fd, FIO_NET_CMD_TS, &p, sizeof(p), NULL, NULL);
1076 void fio_server_send_gs(struct group_run_stats *rs)
1078 struct group_run_stats gs;
1080 dprint(FD_NET, "server sending group run stats\n");
1082 convert_gs(&gs, rs);
1083 fio_net_send_cmd(server_fd, FIO_NET_CMD_GS, &gs, sizeof(gs), NULL, NULL);
1086 static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src)
1090 for (i = 0; i < 2; i++) {
1091 dst->ios[i] = cpu_to_le32(src->ios[i]);
1092 dst->merges[i] = cpu_to_le32(src->merges[i]);
1093 dst->sectors[i] = cpu_to_le64(src->sectors[i]);
1094 dst->ticks[i] = cpu_to_le32(src->ticks[i]);
1097 dst->io_ticks = cpu_to_le32(src->io_ticks);
1098 dst->time_in_queue = cpu_to_le32(src->time_in_queue);
1099 dst->slavecount = cpu_to_le32(src->slavecount);
1100 dst->max_util.u.i = __cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
1103 static void convert_dus(struct disk_util_stat *dst, struct disk_util_stat *src)
1107 dst->name[FIO_DU_NAME_SZ - 1] = '\0';
1108 strncpy((char *) dst->name, (char *) src->name, FIO_DU_NAME_SZ - 1);
1110 for (i = 0; i < 2; i++) {
1111 dst->s.ios[i] = cpu_to_le32(src->s.ios[i]);
1112 dst->s.merges[i] = cpu_to_le32(src->s.merges[i]);
1113 dst->s.sectors[i] = cpu_to_le64(src->s.sectors[i]);
1114 dst->s.ticks[i] = cpu_to_le32(src->s.ticks[i]);
1117 dst->s.io_ticks = cpu_to_le32(src->s.io_ticks);
1118 dst->s.time_in_queue = cpu_to_le32(src->s.time_in_queue);
1119 dst->s.msec = cpu_to_le64(src->s.msec);
1122 void fio_server_send_du(void)
1124 struct disk_util *du;
1125 struct flist_head *entry;
1126 struct cmd_du_pdu pdu;
1128 dprint(FD_NET, "server: sending disk_util %d\n", !flist_empty(&disk_list));
1130 memset(&pdu, 0, sizeof(pdu));
1132 flist_for_each(entry, &disk_list) {
1133 du = flist_entry(entry, struct disk_util, list);
1135 convert_dus(&pdu.dus, &du->dus);
1136 convert_agg(&pdu.agg, &du->agg);
1138 fio_net_send_cmd(server_fd, FIO_NET_CMD_DU, &pdu, sizeof(pdu), NULL, NULL);
1143 * Send a command with a separate PDU, not inlined in the command
1145 static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
1146 off_t size, uint64_t tag, uint32_t flags)
1148 struct fio_net_cmd cmd;
1149 struct iovec iov[2];
1151 iov[0].iov_base = (void *) &cmd;
1152 iov[0].iov_len = sizeof(cmd);
1153 iov[1].iov_base = (void *) buf;
1154 iov[1].iov_len = size;
1156 __fio_init_net_cmd(&cmd, opcode, size, tag);
1157 cmd.flags = __cpu_to_le32(flags);
1158 fio_net_cmd_crc_pdu(&cmd, buf);
1160 return fio_sendv_data(sk, iov, 2);
1163 static int fio_send_iolog_gz(struct cmd_iolog_pdu *pdu, struct io_log *log)
1171 * Dirty - since the log is potentially huge, compress it into
1172 * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving
1173 * side defragment it.
1175 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1177 stream.zalloc = Z_NULL;
1178 stream.zfree = Z_NULL;
1179 stream.opaque = Z_NULL;
1181 if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK) {
1186 stream.next_in = (void *) log->log;
1187 stream.avail_in = log->nr_samples * sizeof(struct io_sample);
1190 unsigned int this_len, flags = 0;
1193 stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1194 stream.next_out = out_pdu;
1195 ret = deflate(&stream, Z_FINISH);
1196 /* may be Z_OK, or Z_STREAM_END */
1200 this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream.avail_out;
1202 if (stream.avail_in)
1203 flags = FIO_NET_CMD_F_MORE;
1205 ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG,
1206 out_pdu, this_len, 0, flags);
1209 } while (stream.avail_in);
1212 deflateEnd(&stream);
1219 int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
1221 struct cmd_iolog_pdu pdu;
1224 pdu.thread_number = cpu_to_le32(td->thread_number);
1225 pdu.nr_samples = __cpu_to_le32(log->nr_samples);
1226 pdu.log_type = cpu_to_le32(log->log_type);
1227 pdu.compressed = cpu_to_le32(use_zlib);
1229 strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX);
1230 pdu.name[FIO_NET_NAME_MAX - 1] = '\0';
1232 for (i = 0; i < log->nr_samples; i++) {
1233 struct io_sample *s = &log->log[i];
1235 s->time = cpu_to_le64(s->time);
1236 s->val = cpu_to_le64(s->val);
1237 s->ddir = cpu_to_le32(s->ddir);
1238 s->bs = cpu_to_le32(s->bs);
1242 * Send header first, it's not compressed.
1244 ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, &pdu,
1245 sizeof(pdu), 0, FIO_NET_CMD_F_MORE);
1250 * Now send actual log, compress if we can, otherwise just plain
1253 return fio_send_iolog_gz(&pdu, log);
1255 return fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, log->log,
1256 log->nr_samples * sizeof(struct io_sample), 0, 0);
1259 void fio_server_send_add_job(struct thread_data *td)
1261 struct cmd_add_job_pdu pdu;
1263 memset(&pdu, 0, sizeof(pdu));
1264 pdu.thread_number = cpu_to_le32(td->thread_number);
1265 pdu.groupid = cpu_to_le32(td->groupid);
1266 convert_thread_options_to_net(&pdu.top, &td->o);
1268 fio_net_send_cmd(server_fd, FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL, NULL);
1271 void fio_server_send_start(struct thread_data *td)
1273 assert(server_fd != -1);
1275 fio_net_send_simple_cmd(server_fd, FIO_NET_CMD_SERVER_START, 0, NULL);
1278 static int fio_init_server_ip(void)
1280 struct sockaddr *addr;
1287 sk = socket(AF_INET6, SOCK_STREAM, 0);
1289 sk = socket(AF_INET, SOCK_STREAM, 0);
1292 log_err("fio: socket: %s\n", strerror(errno));
1297 if (setsockopt(sk, SOL_SOCKET, SO_REUSEADDR, (void *)&opt, sizeof(opt)) < 0) {
1298 log_err("fio: setsockopt: %s\n", strerror(errno));
1303 if (setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
1304 log_err("fio: setsockopt: %s\n", strerror(errno));
1311 const void *src = &saddr_in6.sin6_addr;
1313 addr = (struct sockaddr *) &saddr_in6;
1314 socklen = sizeof(saddr_in6);
1315 saddr_in6.sin6_family = AF_INET6;
1316 str = inet_ntop(AF_INET6, src, buf, sizeof(buf));
1318 const void *src = &saddr_in.sin_addr;
1320 addr = (struct sockaddr *) &saddr_in;
1321 socklen = sizeof(saddr_in);
1322 saddr_in.sin_family = AF_INET;
1323 str = inet_ntop(AF_INET, src, buf, sizeof(buf));
1326 if (bind(sk, addr, socklen) < 0) {
1327 log_err("fio: bind: %s\n", strerror(errno));
1328 log_info("fio: failed with IPv%c %s\n", use_ipv6 ? '6' : '4', str);
1336 static int fio_init_server_sock(void)
1338 struct sockaddr_un addr;
1343 sk = socket(AF_UNIX, SOCK_STREAM, 0);
1345 log_err("fio: socket: %s\n", strerror(errno));
1351 memset(&addr, 0, sizeof(addr));
1352 addr.sun_family = AF_UNIX;
1353 strncpy(addr.sun_path, bind_sock, sizeof(addr.sun_path) - 1);
1355 len = sizeof(addr.sun_family) + strlen(bind_sock) + 1;
1357 if (bind(sk, (struct sockaddr *) &addr, len) < 0) {
1358 log_err("fio: bind: %s\n", strerror(errno));
1367 static int fio_init_server_connection(void)
1372 dprint(FD_NET, "starting server\n");
1375 sk = fio_init_server_ip();
1377 sk = fio_init_server_sock();
1382 memset(bind_str, 0, sizeof(bind_str));
1391 src = &saddr_in6.sin6_addr;
1394 src = &saddr_in.sin_addr;
1397 p = (char *) inet_ntop(af, src, bind_str, sizeof(bind_str));
1399 sprintf(port, ",%u", fio_net_port);
1403 strncpy(bind_str, port, sizeof(bind_str) - 1);
1405 strncpy(bind_str, bind_sock, sizeof(bind_str) - 1);
1407 log_info("fio: server listening on %s\n", bind_str);
1409 if (listen(sk, 0) < 0) {
1410 log_err("fio: listen: %s\n", strerror(errno));
1418 int fio_server_parse_host(const char *host, int ipv6, struct in_addr *inp,
1419 struct in6_addr *inp6)
1425 ret = inet_pton(AF_INET6, host, inp6);
1427 ret = inet_pton(AF_INET, host, inp);
1430 struct addrinfo hints, *res;
1432 memset(&hints, 0, sizeof(hints));
1433 hints.ai_family = ipv6 ? AF_INET6 : AF_INET;
1434 hints.ai_socktype = SOCK_STREAM;
1436 ret = getaddrinfo(host, NULL, &hints, &res);
1438 log_err("fio: failed to resolve <%s> (%s)\n", host,
1444 memcpy(inp6, &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr, sizeof(*inp6));
1446 memcpy(inp, &((struct sockaddr_in *) res->ai_addr)->sin_addr, sizeof(*inp));
1456 * Parse a host/ip/port string. Reads from 'str'.
1461 * *ptr is the host, *port is the port, inp is the destination.
1463 * *ptr is the host, *port is the port, inp6 is the dest, and *ipv6 is 1.
1464 * For local domain sockets:
1465 * *ptr is the filename, *is_sock is 1.
1467 int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
1468 int *port, struct in_addr *inp,
1469 struct in6_addr *inp6, int *ipv6)
1471 const char *host = str;
1477 *port = fio_net_port;
1480 if (!strncmp(str, "sock:", 5)) {
1481 *ptr = strdup(str + 5);
1488 * Is it ip:<ip or host>:port
1490 if (!strncmp(host, "ip:", 3))
1492 else if (!strncmp(host, "ip4:", 4))
1494 else if (!strncmp(host, "ip6:", 4)) {
1497 } else if (host[0] == ':') {
1498 /* String is :port */
1501 if (!lport || lport > 65535) {
1502 log_err("fio: bad server port %u\n", lport);
1505 /* no hostname given, we are done */
1511 * If no port seen yet, check if there's a last ',' at the end
1514 portp = strchr(host, ',');
1518 lport = atoi(portp);
1519 if (!lport || lport > 65535) {
1520 log_err("fio: bad server port %u\n", lport);
1532 *ptr = strdup(host);
1534 if (fio_server_parse_host(*ptr, *ipv6, inp, inp6)) {
1541 *port = fio_net_port;
1547 * Server arg should be one of:
1549 * sock:/path/to/socket
1553 * Where sock uses unix domain sockets, and ip binds the server to
1554 * a specific interface. If no arguments are given to the server, it
1555 * uses IP and binds to 0.0.0.0.
1558 static int fio_handle_server_arg(void)
1560 int port = fio_net_port;
1561 int is_sock, ret = 0;
1563 saddr_in.sin_addr.s_addr = htonl(INADDR_ANY);
1565 if (!fio_server_arg)
1568 ret = fio_server_parse_string(fio_server_arg, &bind_sock, &is_sock,
1569 &port, &saddr_in.sin_addr,
1570 &saddr_in6.sin6_addr, &use_ipv6);
1572 if (!is_sock && bind_sock) {
1578 fio_net_port = port;
1579 saddr_in.sin_port = htons(port);
1580 saddr_in6.sin6_port = htons(port);
1584 static void sig_int(int sig)
1590 static void set_sig_handlers(void)
1592 struct sigaction act;
1594 memset(&act, 0, sizeof(act));
1595 act.sa_handler = sig_int;
1596 act.sa_flags = SA_RESTART;
1597 sigaction(SIGINT, &act, NULL);
1600 static int fio_server(void)
1604 dprint(FD_NET, "starting server\n");
1606 if (fio_handle_server_arg())
1609 sk = fio_init_server_connection();
1615 ret = accept_loop(sk);
1619 if (fio_server_arg) {
1620 free(fio_server_arg);
1621 fio_server_arg = NULL;
1629 void fio_server_got_signal(int signal)
1631 if (signal == SIGPIPE)
1634 log_info("\nfio: terminating on signal %d\n", signal);
1639 static int check_existing_pidfile(const char *pidfile)
1646 if (stat(pidfile, &sb))
1649 f = fopen(pidfile, "r");
1653 if (fread(buf, sb.st_size, 1, f) <= 0) {
1660 if (kill(pid, SIGCONT) < 0)
1661 return errno != ESRCH;
1666 static int write_pid(pid_t pid, const char *pidfile)
1670 fpid = fopen(pidfile, "w");
1672 log_err("fio: failed opening pid file %s\n", pidfile);
1676 fprintf(fpid, "%u\n", (unsigned int) pid);
1682 * If pidfile is specified, background us.
1684 int fio_start_server(char *pidfile)
1691 WSAStartup(MAKEWORD(2, 2), &wsd);
1695 return fio_server();
1697 if (check_existing_pidfile(pidfile)) {
1698 log_err("fio: pidfile %s exists and server appears alive\n",
1706 log_err("fio: failed server fork: %s", strerror(errno));
1710 int ret = write_pid(pid, pidfile);
1717 openlog("fio", LOG_NDELAY|LOG_NOWAIT|LOG_PID, LOG_USER);
1719 close(STDIN_FILENO);
1720 close(STDOUT_FILENO);
1721 close(STDERR_FILENO);
1733 void fio_server_set_arg(const char *arg)
1735 fio_server_arg = strdup(arg);