10 #include <sys/socket.h>
14 #include <netinet/in.h>
15 #include <arpa/inet.h>
26 #include "crc/crc16.h"
27 #include "lib/ieee754.h"
31 int fio_net_port = FIO_NET_PORT;
35 static int server_fd = -1;
36 static char *fio_server_arg;
37 static char *bind_sock;
38 static struct sockaddr_in saddr_in;
39 static struct sockaddr_in6 saddr_in6;
42 static unsigned int has_zlib = 1;
44 static unsigned int has_zlib = 0;
46 static unsigned int use_zlib;
49 struct fio_fork_item {
50 struct flist_head list;
58 struct fio_mutex lock;
64 static const char *fio_server_ops[FIO_NET_CMD_NR] = {
89 const char *fio_server_op(unsigned int op)
93 if (op < FIO_NET_CMD_NR)
94 return fio_server_ops[op];
96 sprintf(buf, "UNKNOWN/%d", op);
100 static ssize_t iov_total_len(const struct iovec *iov, int count)
112 static int fio_sendv_data(int sk, struct iovec *iov, int count)
114 ssize_t total_len = iov_total_len(iov, count);
118 ret = writev(sk, iov, count);
125 if (ret >= iov->iov_len) {
130 iov->iov_base += ret;
136 else if (errno == EAGAIN || errno == EINTR)
140 } while (!exit_backend);
151 int fio_send_data(int sk, const void *p, unsigned int len)
153 struct iovec iov = { .iov_base = (void *) p, .iov_len = len };
155 assert(len <= sizeof(struct fio_net_cmd) + FIO_SERVER_MAX_FRAGMENT_PDU);
157 return fio_sendv_data(sk, &iov, 1);
160 int fio_recv_data(int sk, void *p, unsigned int len)
163 int ret = recv(sk, p, len, MSG_WAITALL);
173 else if (errno == EAGAIN || errno == EINTR)
177 } while (!exit_backend);
185 static int verify_convert_cmd(struct fio_net_cmd *cmd)
189 cmd->cmd_crc16 = le16_to_cpu(cmd->cmd_crc16);
190 cmd->pdu_crc16 = le16_to_cpu(cmd->pdu_crc16);
192 crc = fio_crc16(cmd, FIO_NET_CMD_CRC_SZ);
193 if (crc != cmd->cmd_crc16) {
194 log_err("fio: server bad crc on command (got %x, wanted %x)\n",
195 cmd->cmd_crc16, crc);
199 cmd->version = le16_to_cpu(cmd->version);
200 cmd->opcode = le16_to_cpu(cmd->opcode);
201 cmd->flags = le32_to_cpu(cmd->flags);
202 cmd->tag = le64_to_cpu(cmd->tag);
203 cmd->pdu_len = le32_to_cpu(cmd->pdu_len);
205 switch (cmd->version) {
209 log_err("fio: bad server cmd version %d\n", cmd->version);
213 if (cmd->pdu_len > FIO_SERVER_MAX_FRAGMENT_PDU) {
214 log_err("fio: command payload too large: %u\n", cmd->pdu_len);
222 * Read (and defragment, if necessary) incoming commands
224 struct fio_net_cmd *fio_net_recv_cmd(int sk)
226 struct fio_net_cmd cmd, *tmp, *cmdret = NULL;
227 size_t cmd_size = 0, pdu_offset = 0;
233 ret = fio_recv_data(sk, &cmd, sizeof(cmd));
237 /* We have a command, verify it and swap if need be */
238 ret = verify_convert_cmd(&cmd);
243 /* if this is text, add room for \0 at the end */
244 cmd_size = sizeof(cmd) + cmd.pdu_len + 1;
247 cmd_size += cmd.pdu_len;
249 if (cmd_size / 1024 > FIO_SERVER_MAX_CMD_MB * 1024) {
250 log_err("fio: cmd+pdu too large (%llu)\n", (unsigned long long) cmd_size);
255 tmp = realloc(cmdret, cmd_size);
257 log_err("fio: server failed allocating cmd\n");
264 memcpy(cmdret, &cmd, sizeof(cmd));
265 else if (cmdret->opcode != cmd.opcode) {
266 log_err("fio: fragment opcode mismatch (%d != %d)\n",
267 cmdret->opcode, cmd.opcode);
275 /* There's payload, get it */
276 pdu = (void *) cmdret->payload + pdu_offset;
277 ret = fio_recv_data(sk, pdu, cmd.pdu_len);
281 /* Verify payload crc */
282 crc = fio_crc16(pdu, cmd.pdu_len);
283 if (crc != cmd.pdu_crc16) {
284 log_err("fio: server bad crc on payload ");
285 log_err("(got %x, wanted %x)\n", cmd.pdu_crc16, crc);
290 pdu_offset += cmd.pdu_len;
292 cmdret->pdu_len += cmd.pdu_len;
294 } while (cmd.flags & FIO_NET_CMD_F_MORE);
300 /* zero-terminate text input */
301 if (cmdret->pdu_len) {
302 if (cmdret->opcode == FIO_NET_CMD_TEXT) {
303 struct cmd_text_pdu *__pdu = (struct cmd_text_pdu *) cmdret->payload;
304 char *buf = (char *) __pdu->buf;
306 buf[__pdu->buf_len] = '\0';
307 } else if (cmdret->opcode == FIO_NET_CMD_JOB) {
308 struct cmd_job_pdu *__pdu = (struct cmd_job_pdu *) cmdret->payload;
309 char *buf = (char *) __pdu->buf;
310 int len = le32_to_cpu(__pdu->buf_len);
316 /* frag flag is internal */
317 cmdret->flags &= ~FIO_NET_CMD_F_MORE;
323 static void add_reply(uint64_t tag, struct flist_head *list)
325 struct fio_net_cmd_reply *reply;
327 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
328 flist_add_tail(&reply->list, list);
331 static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
333 struct fio_net_cmd_reply *reply;
335 reply = calloc(1, sizeof(*reply));
336 INIT_FLIST_HEAD(&reply->list);
337 fio_gettime(&reply->tv, NULL);
338 reply->saved_tag = tag;
339 reply->opcode = opcode;
341 return (uintptr_t) reply;
344 static void free_reply(uint64_t tag)
346 struct fio_net_cmd_reply *reply;
348 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
352 void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu)
356 cmd->cmd_crc16 = __cpu_to_le16(fio_crc16(cmd, FIO_NET_CMD_CRC_SZ));
358 pdu_len = le32_to_cpu(cmd->pdu_len);
359 cmd->pdu_crc16 = __cpu_to_le16(fio_crc16(pdu, pdu_len));
362 void fio_net_cmd_crc(struct fio_net_cmd *cmd)
364 fio_net_cmd_crc_pdu(cmd, cmd->payload);
367 int fio_net_send_cmd(int fd, uint16_t opcode, const void *buf, off_t size,
368 uint64_t *tagptr, struct flist_head *list)
370 struct fio_net_cmd *cmd = NULL;
371 size_t this_len, cur_len = 0;
377 tag = *tagptr = alloc_reply(*tagptr, opcode);
379 tag = tagptr ? *tagptr : 0;
383 if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU)
384 this_len = FIO_SERVER_MAX_FRAGMENT_PDU;
386 if (!cmd || cur_len < sizeof(*cmd) + this_len) {
390 cur_len = sizeof(*cmd) + this_len;
391 cmd = malloc(cur_len);
394 fio_init_net_cmd(cmd, opcode, buf, this_len, tag);
397 cmd->flags = __cpu_to_le32(FIO_NET_CMD_F_MORE);
399 fio_net_cmd_crc(cmd);
401 ret = fio_send_data(fd, cmd, sizeof(*cmd) + this_len);
404 } while (!ret && size);
410 add_reply(tag, list);
419 static int fio_net_send_simple_stack_cmd(int sk, uint16_t opcode, uint64_t tag)
421 struct fio_net_cmd cmd;
423 fio_init_net_cmd(&cmd, opcode, NULL, 0, tag);
424 fio_net_cmd_crc(&cmd);
426 return fio_send_data(sk, &cmd, sizeof(cmd));
430 * If 'list' is non-NULL, then allocate and store the sent command for
431 * later verification.
433 int fio_net_send_simple_cmd(int sk, uint16_t opcode, uint64_t tag,
434 struct flist_head *list)
439 tag = alloc_reply(tag, opcode);
441 ret = fio_net_send_simple_stack_cmd(sk, opcode, tag);
450 add_reply(tag, list);
455 int fio_net_send_quit(int sk)
457 dprint(FD_NET, "server: sending quit\n");
459 return fio_net_send_simple_cmd(sk, FIO_NET_CMD_QUIT, 0, NULL);
462 static int fio_net_send_ack(int sk, struct fio_net_cmd *cmd, int error,
465 struct cmd_end_pdu epdu;
471 epdu.error = __cpu_to_le32(error);
472 epdu.signal = __cpu_to_le32(signal);
473 return fio_net_send_cmd(sk, FIO_NET_CMD_STOP, &epdu, sizeof(epdu), &tag, NULL);
476 int fio_net_send_stop(int sk, int error, int signal)
478 dprint(FD_NET, "server: sending stop (%d, %d)\n", error, signal);
479 return fio_net_send_ack(sk, NULL, error, signal);
482 static void fio_server_add_fork_item(pid_t pid, struct flist_head *list)
484 struct fio_fork_item *ffi;
486 ffi = malloc(sizeof(*ffi));
491 flist_add_tail(&ffi->list, list);
494 static void fio_server_add_conn_pid(struct flist_head *conn_list, pid_t pid)
496 dprint(FD_NET, "server: forked off connection job (pid=%u)\n", (int) pid);
497 fio_server_add_fork_item(pid, conn_list);
500 static void fio_server_add_job_pid(struct flist_head *job_list, pid_t pid)
502 dprint(FD_NET, "server: forked off job job (pid=%u)\n", (int) pid);
503 fio_server_add_fork_item(pid, job_list);
506 static void fio_server_check_fork_item(struct fio_fork_item *ffi)
510 ret = waitpid(ffi->pid, &status, WNOHANG);
512 if (errno == ECHILD) {
513 log_err("fio: connection pid %u disappeared\n", (int) ffi->pid);
516 log_err("fio: waitpid: %s\n", strerror(errno));
517 } else if (ret == ffi->pid) {
518 if (WIFSIGNALED(status)) {
519 ffi->signal = WTERMSIG(status);
522 if (WIFEXITED(status)) {
523 if (WEXITSTATUS(status))
524 ffi->exitval = WEXITSTATUS(status);
530 static void fio_server_fork_item_done(struct fio_fork_item *ffi)
532 dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", (int) ffi->pid, ffi->signal, ffi->exitval);
535 * Fold STOP and QUIT...
537 fio_net_send_stop(server_fd, ffi->exitval, ffi->signal);
538 fio_net_send_quit(server_fd);
539 flist_del(&ffi->list);
543 static void fio_server_check_fork_items(struct flist_head *list)
545 struct flist_head *entry, *tmp;
546 struct fio_fork_item *ffi;
548 flist_for_each_safe(entry, tmp, list) {
549 ffi = flist_entry(entry, struct fio_fork_item, list);
551 fio_server_check_fork_item(ffi);
554 fio_server_fork_item_done(ffi);
558 static void fio_server_check_jobs(struct flist_head *job_list)
560 fio_server_check_fork_items(job_list);
563 static void fio_server_check_conns(struct flist_head *conn_list)
565 fio_server_check_fork_items(conn_list);
568 static int handle_load_file_cmd(struct fio_net_cmd *cmd)
570 struct cmd_load_file_pdu *pdu = (struct cmd_load_file_pdu *) cmd->payload;
571 void *file_name = pdu->file;
572 struct cmd_start_pdu spdu;
574 dprint(FD_NET, "server: loading local file %s\n", (char *) file_name);
576 pdu->name_len = le16_to_cpu(pdu->name_len);
577 pdu->client_type = le16_to_cpu(pdu->client_type);
579 if (parse_jobs_ini(file_name, 0, 0, pdu->client_type)) {
580 fio_net_send_quit(server_fd);
584 spdu.jobs = cpu_to_le32(thread_number);
585 spdu.stat_outputs = cpu_to_le32(stat_number);
586 fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
590 static int handle_run_cmd(struct flist_head *job_list, struct fio_net_cmd *cmd)
600 fio_server_add_job_pid(job_list, pid);
609 static int handle_job_cmd(struct fio_net_cmd *cmd)
611 struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmd->payload;
612 void *buf = pdu->buf;
613 struct cmd_start_pdu spdu;
615 pdu->buf_len = le32_to_cpu(pdu->buf_len);
616 pdu->client_type = le32_to_cpu(pdu->client_type);
618 if (parse_jobs_ini(buf, 1, 0, pdu->client_type)) {
619 fio_net_send_quit(server_fd);
623 spdu.jobs = cpu_to_le32(thread_number);
624 spdu.stat_outputs = cpu_to_le32(stat_number);
625 fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
629 static int handle_jobline_cmd(struct fio_net_cmd *cmd)
631 void *pdu = cmd->payload;
632 struct cmd_single_line_pdu *cslp;
633 struct cmd_line_pdu *clp;
634 unsigned long offset;
635 struct cmd_start_pdu spdu;
640 clp->lines = le16_to_cpu(clp->lines);
641 clp->client_type = le16_to_cpu(clp->client_type);
642 argv = malloc(clp->lines * sizeof(char *));
643 offset = sizeof(*clp);
645 dprint(FD_NET, "server: %d command line args\n", clp->lines);
647 for (i = 0; i < clp->lines; i++) {
649 argv[i] = (char *) cslp->text;
651 offset += sizeof(*cslp) + le16_to_cpu(cslp->len);
652 dprint(FD_NET, "server: %d: %s\n", i, argv[i]);
655 if (parse_cmd_line(clp->lines, argv, clp->client_type)) {
656 fio_net_send_quit(server_fd);
663 spdu.jobs = cpu_to_le32(thread_number);
664 spdu.stat_outputs = cpu_to_le32(stat_number);
665 fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
669 static int handle_probe_cmd(struct fio_net_cmd *cmd)
671 struct cmd_client_probe_pdu *pdu = (struct cmd_client_probe_pdu *) cmd->payload;
672 struct cmd_probe_reply_pdu probe;
673 uint64_t tag = cmd->tag;
675 dprint(FD_NET, "server: sending probe reply\n");
677 strcpy(me, (char *) pdu->server);
679 memset(&probe, 0, sizeof(probe));
680 gethostname((char *) probe.hostname, sizeof(probe.hostname));
681 #ifdef CONFIG_BIG_ENDIAN
684 strncpy((char *) probe.fio_version, fio_version_string, sizeof(probe.fio_version));
687 probe.arch = FIO_ARCH;
688 probe.bpp = sizeof(void *);
689 probe.cpus = __cpu_to_le32(cpus_online());
692 * If the client supports compression and we do too, then enable it
694 if (has_zlib && le64_to_cpu(pdu->flags) & FIO_PROBE_FLAG_ZLIB) {
695 probe.flags = __cpu_to_le64(FIO_PROBE_FLAG_ZLIB);
702 return fio_net_send_cmd(server_fd, FIO_NET_CMD_PROBE, &probe, sizeof(probe), &tag, NULL);
705 static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
708 uint64_t tag = cmd->tag;
712 je = get_jobs_eta(1, &size);
716 dprint(FD_NET, "server sending status\n");
718 je->nr_running = cpu_to_le32(je->nr_running);
719 je->nr_ramp = cpu_to_le32(je->nr_ramp);
720 je->nr_pending = cpu_to_le32(je->nr_pending);
721 je->nr_setting_up = cpu_to_le32(je->nr_setting_up);
722 je->files_open = cpu_to_le32(je->files_open);
724 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
725 je->m_rate[i] = cpu_to_le32(je->m_rate[i]);
726 je->t_rate[i] = cpu_to_le32(je->t_rate[i]);
727 je->m_iops[i] = cpu_to_le32(je->m_iops[i]);
728 je->t_iops[i] = cpu_to_le32(je->t_iops[i]);
729 je->rate[i] = cpu_to_le32(je->rate[i]);
730 je->iops[i] = cpu_to_le32(je->iops[i]);
733 je->elapsed_sec = cpu_to_le64(je->elapsed_sec);
734 je->eta_sec = cpu_to_le64(je->eta_sec);
735 je->nr_threads = cpu_to_le32(je->nr_threads);
736 je->is_pow2 = cpu_to_le32(je->is_pow2);
737 je->unit_base = cpu_to_le32(je->unit_base);
739 fio_net_send_cmd(server_fd, FIO_NET_CMD_ETA, je, size, &tag, NULL);
744 static int send_update_job_reply(int fd, uint64_t __tag, int error)
746 uint64_t tag = __tag;
749 pdu_error = __cpu_to_le32(error);
750 return fio_net_send_cmd(fd, FIO_NET_CMD_UPDATE_JOB, &pdu_error, sizeof(pdu_error), &tag, NULL);
753 static int handle_update_job_cmd(struct fio_net_cmd *cmd)
755 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
756 struct thread_data *td;
759 tnumber = le32_to_cpu(pdu->thread_number);
761 dprint(FD_NET, "server: updating options for job %u\n", tnumber);
763 if (!tnumber || tnumber > thread_number) {
764 send_update_job_reply(server_fd, cmd->tag, ENODEV);
768 td = &threads[tnumber - 1];
769 convert_thread_options_to_cpu(&td->o, &pdu->top);
770 send_update_job_reply(server_fd, cmd->tag, 0);
774 static int handle_trigger_cmd(struct fio_net_cmd *cmd)
776 struct cmd_vtrigger_pdu *pdu = (struct cmd_vtrigger_pdu *) cmd->payload;
777 char *buf = (char *) pdu->cmd;
778 struct all_io_list *rep;
781 pdu->len = le16_to_cpu(pdu->len);
782 buf[pdu->len] = '\0';
784 rep = get_all_io_list(IO_LIST_ALL, &sz);
786 struct all_io_list state;
788 state.threads = cpu_to_le64((uint64_t) 0);
789 fio_net_send_cmd(server_fd, FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, NULL);
791 fio_net_send_cmd(server_fd, FIO_NET_CMD_VTRIGGER, rep, sz, NULL, NULL);
799 static int handle_command(struct flist_head *job_list, struct fio_net_cmd *cmd)
803 dprint(FD_NET, "server: got op [%s], pdu=%u, tag=%llx\n",
804 fio_server_op(cmd->opcode), cmd->pdu_len,
805 (unsigned long long) cmd->tag);
807 switch (cmd->opcode) {
808 case FIO_NET_CMD_QUIT:
809 fio_terminate_threads(TERMINATE_ALL);
811 case FIO_NET_CMD_EXIT:
814 case FIO_NET_CMD_LOAD_FILE:
815 ret = handle_load_file_cmd(cmd);
817 case FIO_NET_CMD_JOB:
818 ret = handle_job_cmd(cmd);
820 case FIO_NET_CMD_JOBLINE:
821 ret = handle_jobline_cmd(cmd);
823 case FIO_NET_CMD_PROBE:
824 ret = handle_probe_cmd(cmd);
826 case FIO_NET_CMD_SEND_ETA:
827 ret = handle_send_eta_cmd(cmd);
829 case FIO_NET_CMD_RUN:
830 ret = handle_run_cmd(job_list, cmd);
832 case FIO_NET_CMD_UPDATE_JOB:
833 ret = handle_update_job_cmd(cmd);
835 case FIO_NET_CMD_VTRIGGER:
836 ret = handle_trigger_cmd(cmd);
838 case FIO_NET_CMD_SENDFILE: {
839 struct cmd_sendfile_reply *in;
840 struct cmd_reply *rep;
842 rep = (struct cmd_reply *) (uintptr_t) cmd->tag;
844 in = (struct cmd_sendfile_reply *) cmd->payload;
845 in->size = le32_to_cpu(in->size);
846 in->error = le32_to_cpu(in->error);
849 rep->error = in->error;
852 rep->data = smalloc(in->size);
857 rep->size = in->size;
858 memcpy(rep->data, in->data, in->size);
861 fio_mutex_up(&rep->lock);
865 log_err("fio: unknown opcode: %s\n", fio_server_op(cmd->opcode));
872 static int handle_connection(int sk)
874 struct fio_net_cmd *cmd = NULL;
875 FLIST_HEAD(job_list);
882 while (!exit_backend) {
883 struct pollfd pfd = {
892 if (!flist_empty(&job_list))
895 ret = poll(&pfd, 1, timeout);
899 log_err("fio: poll: %s\n", strerror(errno));
902 fio_server_check_jobs(&job_list);
906 if (pfd.revents & POLLIN)
908 if (pfd.revents & (POLLERR|POLLHUP)) {
912 } while (!exit_backend);
914 fio_server_check_jobs(&job_list);
919 cmd = fio_net_recv_cmd(sk);
925 ret = handle_command(&job_list, cmd);
940 /* get the address on this host bound by the input socket,
941 * whether it is ipv6 or ipv4 */
943 int get_my_addr_str( int sk )
946 struct sockaddr * sockaddr_p;
947 struct sockaddr_in myaddr4 = {0};
948 struct sockaddr_in6 myaddr6 = {0};
950 socklen_t len = use_ipv6 ? sizeof(myaddr6) : sizeof(myaddr4);
953 sockaddr_p = (struct sockaddr * )&myaddr6;
955 sockaddr_p = (struct sockaddr * )&myaddr4;
956 ret = getsockname(sk, sockaddr_p, &len);
958 log_err("fio: getsockaddr: %s\n", strerror(errno));
962 net_addr = (char * )&myaddr6.sin6_addr;
964 net_addr = (char * )&myaddr4.sin_addr;
965 if (NULL == inet_ntop(use_ipv6?AF_INET6:AF_INET, net_addr, client_sockaddr_str, INET6_ADDRSTRLEN-1)) {
966 log_err("inet_ntop: failed to convert addr to string\n");
969 dprint(FD_NET, "fio server bound to addr %s\n", client_sockaddr_str);
973 static int accept_loop(int listen_sk)
975 struct sockaddr_in addr;
976 struct sockaddr_in6 addr6;
977 socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr);
979 int ret = 0, sk, exitval = 0;
980 FLIST_HEAD(conn_list);
982 dprint(FD_NET, "server enter accept loop\n");
984 fio_set_fd_nonblocking(listen_sk, "server");
986 while (!exit_backend) {
996 if (!flist_empty(&conn_list))
999 ret = poll(&pfd, 1, timeout);
1003 log_err("fio: poll: %s\n", strerror(errno));
1006 fio_server_check_conns(&conn_list);
1010 if (pfd.revents & POLLIN)
1012 } while (!exit_backend);
1014 fio_server_check_conns(&conn_list);
1016 if (exit_backend || ret < 0)
1020 sk = accept(listen_sk, (struct sockaddr *) &addr6, &len);
1022 sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
1025 log_err("fio: accept: %s\n", strerror(errno));
1030 from = inet_ntop(AF_INET6, (struct sockaddr *) &addr6.sin6_addr, buf, sizeof(buf));
1032 from = inet_ntop(AF_INET, (struct sockaddr *) &addr.sin_addr, buf, sizeof(buf));
1034 dprint(FD_NET, "server: connect from %s\n", from);
1039 fio_server_add_conn_pid(&conn_list, pid);
1044 get_my_addr_str(sk); /* if error, it's already logged, non-fatal */
1045 handle_connection(sk);
1051 int fio_server_text_output(int level, const char *buf, size_t len)
1053 struct cmd_text_pdu *pdu;
1057 if (server_fd == -1)
1058 return log_local_buf(buf, len);
1060 tlen = sizeof(*pdu) + len;
1063 pdu->level = __cpu_to_le32(level);
1064 pdu->buf_len = __cpu_to_le32(len);
1066 gettimeofday(&tv, NULL);
1067 pdu->log_sec = __cpu_to_le64(tv.tv_sec);
1068 pdu->log_usec = __cpu_to_le64(tv.tv_usec);
1070 memcpy(pdu->buf, buf, len);
1072 fio_net_send_cmd(server_fd, FIO_NET_CMD_TEXT, pdu, tlen, NULL, NULL);
1077 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
1079 dst->max_val = cpu_to_le64(src->max_val);
1080 dst->min_val = cpu_to_le64(src->min_val);
1081 dst->samples = cpu_to_le64(src->samples);
1084 * Encode to IEEE 754 for network transfer
1086 dst->mean.u.i = cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
1087 dst->S.u.i = cpu_to_le64(fio_double_to_uint64(src->S.u.f));
1090 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1094 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1095 dst->max_run[i] = cpu_to_le64(src->max_run[i]);
1096 dst->min_run[i] = cpu_to_le64(src->min_run[i]);
1097 dst->max_bw[i] = cpu_to_le64(src->max_bw[i]);
1098 dst->min_bw[i] = cpu_to_le64(src->min_bw[i]);
1099 dst->io_kb[i] = cpu_to_le64(src->io_kb[i]);
1100 dst->agg[i] = cpu_to_le64(src->agg[i]);
1103 dst->kb_base = cpu_to_le32(src->kb_base);
1104 dst->unit_base = cpu_to_le32(src->unit_base);
1105 dst->groupid = cpu_to_le32(src->groupid);
1106 dst->unified_rw_rep = cpu_to_le32(src->unified_rw_rep);
1110 * Send a CMD_TS, which packs struct thread_stat and group_run_stats
1111 * into a single payload.
1113 void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
1115 struct cmd_ts_pdu p;
1118 dprint(FD_NET, "server sending end stats\n");
1120 memset(&p, 0, sizeof(p));
1122 strncpy(p.ts.name, ts->name, FIO_JOBNAME_SIZE - 1);
1123 strncpy(p.ts.verror, ts->verror, FIO_VERROR_SIZE - 1);
1124 strncpy(p.ts.description, ts->description, FIO_JOBDESC_SIZE - 1);
1126 p.ts.error = cpu_to_le32(ts->error);
1127 p.ts.thread_number = cpu_to_le32(ts->thread_number);
1128 p.ts.groupid = cpu_to_le32(ts->groupid);
1129 p.ts.pid = cpu_to_le32(ts->pid);
1130 p.ts.members = cpu_to_le32(ts->members);
1131 p.ts.unified_rw_rep = cpu_to_le32(ts->unified_rw_rep);
1133 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1134 convert_io_stat(&p.ts.clat_stat[i], &ts->clat_stat[i]);
1135 convert_io_stat(&p.ts.slat_stat[i], &ts->slat_stat[i]);
1136 convert_io_stat(&p.ts.lat_stat[i], &ts->lat_stat[i]);
1137 convert_io_stat(&p.ts.bw_stat[i], &ts->bw_stat[i]);
1140 p.ts.usr_time = cpu_to_le64(ts->usr_time);
1141 p.ts.sys_time = cpu_to_le64(ts->sys_time);
1142 p.ts.ctx = cpu_to_le64(ts->ctx);
1143 p.ts.minf = cpu_to_le64(ts->minf);
1144 p.ts.majf = cpu_to_le64(ts->majf);
1145 p.ts.clat_percentiles = cpu_to_le64(ts->clat_percentiles);
1146 p.ts.percentile_precision = cpu_to_le64(ts->percentile_precision);
1148 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
1149 fio_fp64_t *src = &ts->percentile_list[i];
1150 fio_fp64_t *dst = &p.ts.percentile_list[i];
1152 dst->u.i = cpu_to_le64(fio_double_to_uint64(src->u.f));
1155 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
1156 p.ts.io_u_map[i] = cpu_to_le32(ts->io_u_map[i]);
1157 p.ts.io_u_submit[i] = cpu_to_le32(ts->io_u_submit[i]);
1158 p.ts.io_u_complete[i] = cpu_to_le32(ts->io_u_complete[i]);
1161 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
1162 p.ts.io_u_lat_u[i] = cpu_to_le32(ts->io_u_lat_u[i]);
1163 p.ts.io_u_lat_m[i] = cpu_to_le32(ts->io_u_lat_m[i]);
1166 for (i = 0; i < DDIR_RWDIR_CNT; i++)
1167 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1168 p.ts.io_u_plat[i][j] = cpu_to_le32(ts->io_u_plat[i][j]);
1170 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1171 p.ts.total_io_u[i] = cpu_to_le64(ts->total_io_u[i]);
1172 p.ts.short_io_u[i] = cpu_to_le64(ts->short_io_u[i]);
1173 p.ts.drop_io_u[i] = cpu_to_le64(ts->drop_io_u[i]);
1176 p.ts.total_submit = cpu_to_le64(ts->total_submit);
1177 p.ts.total_complete = cpu_to_le64(ts->total_complete);
1179 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1180 p.ts.io_bytes[i] = cpu_to_le64(ts->io_bytes[i]);
1181 p.ts.runtime[i] = cpu_to_le64(ts->runtime[i]);
1184 p.ts.total_run_time = cpu_to_le64(ts->total_run_time);
1185 p.ts.continue_on_error = cpu_to_le16(ts->continue_on_error);
1186 p.ts.total_err_count = cpu_to_le64(ts->total_err_count);
1187 p.ts.first_error = cpu_to_le32(ts->first_error);
1188 p.ts.kb_base = cpu_to_le32(ts->kb_base);
1189 p.ts.unit_base = cpu_to_le32(ts->unit_base);
1191 p.ts.latency_depth = cpu_to_le32(ts->latency_depth);
1192 p.ts.latency_target = cpu_to_le64(ts->latency_target);
1193 p.ts.latency_window = cpu_to_le64(ts->latency_window);
1194 p.ts.latency_percentile.u.i = cpu_to_le64(fio_double_to_uint64(ts->latency_percentile.u.f));
1196 p.ts.nr_block_infos = le64_to_cpu(ts->nr_block_infos);
1197 for (i = 0; i < p.ts.nr_block_infos; i++)
1198 p.ts.block_infos[i] = le32_to_cpu(ts->block_infos[i]);
1200 convert_gs(&p.rs, rs);
1202 fio_net_send_cmd(server_fd, FIO_NET_CMD_TS, &p, sizeof(p), NULL, NULL);
1205 void fio_server_send_gs(struct group_run_stats *rs)
1207 struct group_run_stats gs;
1209 dprint(FD_NET, "server sending group run stats\n");
1211 convert_gs(&gs, rs);
1212 fio_net_send_cmd(server_fd, FIO_NET_CMD_GS, &gs, sizeof(gs), NULL, NULL);
1215 static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src)
1219 for (i = 0; i < 2; i++) {
1220 dst->ios[i] = cpu_to_le64(src->ios[i]);
1221 dst->merges[i] = cpu_to_le64(src->merges[i]);
1222 dst->sectors[i] = cpu_to_le64(src->sectors[i]);
1223 dst->ticks[i] = cpu_to_le64(src->ticks[i]);
1226 dst->io_ticks = cpu_to_le64(src->io_ticks);
1227 dst->time_in_queue = cpu_to_le64(src->time_in_queue);
1228 dst->slavecount = cpu_to_le32(src->slavecount);
1229 dst->max_util.u.i = cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
1232 static void convert_dus(struct disk_util_stat *dst, struct disk_util_stat *src)
1236 dst->name[FIO_DU_NAME_SZ - 1] = '\0';
1237 strncpy((char *) dst->name, (char *) src->name, FIO_DU_NAME_SZ - 1);
1239 for (i = 0; i < 2; i++) {
1240 dst->s.ios[i] = cpu_to_le64(src->s.ios[i]);
1241 dst->s.merges[i] = cpu_to_le64(src->s.merges[i]);
1242 dst->s.sectors[i] = cpu_to_le64(src->s.sectors[i]);
1243 dst->s.ticks[i] = cpu_to_le64(src->s.ticks[i]);
1246 dst->s.io_ticks = cpu_to_le64(src->s.io_ticks);
1247 dst->s.time_in_queue = cpu_to_le64(src->s.time_in_queue);
1248 dst->s.msec = cpu_to_le64(src->s.msec);
1251 void fio_server_send_du(void)
1253 struct disk_util *du;
1254 struct flist_head *entry;
1255 struct cmd_du_pdu pdu;
1257 dprint(FD_NET, "server: sending disk_util %d\n", !flist_empty(&disk_list));
1259 memset(&pdu, 0, sizeof(pdu));
1261 flist_for_each(entry, &disk_list) {
1262 du = flist_entry(entry, struct disk_util, list);
1264 convert_dus(&pdu.dus, &du->dus);
1265 convert_agg(&pdu.agg, &du->agg);
1267 fio_net_send_cmd(server_fd, FIO_NET_CMD_DU, &pdu, sizeof(pdu), NULL, NULL);
1272 * Send a command with a separate PDU, not inlined in the command
1274 static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
1275 off_t size, uint64_t tag, uint32_t flags)
1277 struct fio_net_cmd cmd;
1278 struct iovec iov[2];
1280 iov[0].iov_base = (void *) &cmd;
1281 iov[0].iov_len = sizeof(cmd);
1282 iov[1].iov_base = (void *) buf;
1283 iov[1].iov_len = size;
1285 __fio_init_net_cmd(&cmd, opcode, size, tag);
1286 cmd.flags = __cpu_to_le32(flags);
1287 fio_net_cmd_crc_pdu(&cmd, buf);
1289 return fio_sendv_data(sk, iov, 2);
1292 static int fio_send_iolog_gz(struct cmd_iolog_pdu *pdu, struct io_log *log)
1300 * Dirty - since the log is potentially huge, compress it into
1301 * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving
1302 * side defragment it.
1304 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1306 stream.zalloc = Z_NULL;
1307 stream.zfree = Z_NULL;
1308 stream.opaque = Z_NULL;
1310 if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK) {
1315 stream.next_in = (void *) log->log;
1316 stream.avail_in = log->nr_samples * log_entry_sz(log);
1319 unsigned int this_len, flags = 0;
1321 stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1322 stream.next_out = out_pdu;
1323 ret = deflate(&stream, Z_FINISH);
1324 /* may be Z_OK, or Z_STREAM_END */
1328 this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream.avail_out;
1330 if (stream.avail_in)
1331 flags = FIO_NET_CMD_F_MORE;
1333 ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG,
1334 out_pdu, this_len, 0, flags);
1337 } while (stream.avail_in);
1340 deflateEnd(&stream);
1347 int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
1349 struct cmd_iolog_pdu pdu;
1352 pdu.nr_samples = cpu_to_le64(log->nr_samples);
1353 pdu.thread_number = cpu_to_le32(td->thread_number);
1354 pdu.log_type = cpu_to_le32(log->log_type);
1355 pdu.compressed = cpu_to_le32(use_zlib);
1357 strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX);
1358 pdu.name[FIO_NET_NAME_MAX - 1] = '\0';
1360 for (i = 0; i < log->nr_samples; i++) {
1361 struct io_sample *s = get_sample(log, i);
1363 s->time = cpu_to_le64(s->time);
1364 s->val = cpu_to_le64(s->val);
1365 s->__ddir = cpu_to_le32(s->__ddir);
1366 s->bs = cpu_to_le32(s->bs);
1368 if (log->log_offset) {
1369 struct io_sample_offset *so = (void *) s;
1371 so->offset = cpu_to_le64(so->offset);
1376 * Send header first, it's not compressed.
1378 ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, &pdu,
1379 sizeof(pdu), 0, FIO_NET_CMD_F_MORE);
1384 * Now send actual log, compress if we can, otherwise just plain
1387 return fio_send_iolog_gz(&pdu, log);
1389 return fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, log->log,
1390 log->nr_samples * log_entry_sz(log), 0, 0);
1393 void fio_server_send_add_job(struct thread_data *td)
1395 struct cmd_add_job_pdu pdu;
1397 memset(&pdu, 0, sizeof(pdu));
1398 pdu.thread_number = cpu_to_le32(td->thread_number);
1399 pdu.groupid = cpu_to_le32(td->groupid);
1400 convert_thread_options_to_net(&pdu.top, &td->o);
1402 fio_net_send_cmd(server_fd, FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL, NULL);
1405 void fio_server_send_start(struct thread_data *td)
1407 assert(server_fd != -1);
1409 fio_net_send_simple_cmd(server_fd, FIO_NET_CMD_SERVER_START, 0, NULL);
1412 int fio_server_get_verify_state(const char *name, int threadnumber,
1413 void **datap, int *version)
1415 struct thread_io_list *s;
1416 struct cmd_sendfile out;
1417 struct cmd_reply *rep;
1421 dprint(FD_NET, "server: request verify state\n");
1423 rep = smalloc(sizeof(*rep));
1425 log_err("fio: smalloc pool too small\n");
1429 __fio_mutex_init(&rep->lock, FIO_MUTEX_LOCKED);
1433 verify_state_gen_name((char *) out.path, sizeof(out.path), name, me,
1435 tag = (uint64_t) (uintptr_t) rep;
1436 fio_net_send_cmd(server_fd, FIO_NET_CMD_SENDFILE, &out, sizeof(out),
1440 * Wait for the backend to receive the reply
1442 if (fio_mutex_down_timeout(&rep->lock, 10)) {
1443 log_err("fio: timed out waiting for reply\n");
1448 log_err("fio: failure on receiving state file: %s\n", strerror(rep->error));
1452 fio_net_send_quit(server_fd);
1457 * The format is verify_state_hdr, then thread_io_list. Verify
1458 * the header, and the thread_io_list checksum
1460 s = rep->data + sizeof(struct verify_state_hdr);
1461 if (verify_state_hdr(rep->data, s, version))
1465 * Don't need the header from now, copy just the thread_io_list
1467 rep->size -= sizeof(struct verify_state_hdr);
1468 data = malloc(rep->size);
1469 memcpy(data, s, rep->size);
1473 __fio_mutex_remove(&rep->lock);
1478 static int fio_init_server_ip(void)
1480 struct sockaddr *addr;
1487 sk = socket(AF_INET6, SOCK_STREAM, 0);
1489 sk = socket(AF_INET, SOCK_STREAM, 0);
1492 log_err("fio: socket: %s\n", strerror(errno));
1497 if (setsockopt(sk, SOL_SOCKET, SO_REUSEADDR, (void *)&opt, sizeof(opt)) < 0) {
1498 log_err("fio: setsockopt(REUSEADDR): %s\n", strerror(errno));
1503 if (setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) {
1504 log_err("fio: setsockopt(REUSEPORT): %s\n", strerror(errno));
1511 const void *src = &saddr_in6.sin6_addr;
1513 addr = (struct sockaddr *) &saddr_in6;
1514 socklen = sizeof(saddr_in6);
1515 saddr_in6.sin6_family = AF_INET6;
1516 str = inet_ntop(AF_INET6, src, buf, sizeof(buf));
1518 const void *src = &saddr_in.sin_addr;
1520 addr = (struct sockaddr *) &saddr_in;
1521 socklen = sizeof(saddr_in);
1522 saddr_in.sin_family = AF_INET;
1523 str = inet_ntop(AF_INET, src, buf, sizeof(buf));
1526 if (bind(sk, addr, socklen) < 0) {
1527 log_err("fio: bind: %s\n", strerror(errno));
1528 log_info("fio: failed with IPv%c %s\n", use_ipv6 ? '6' : '4', str);
1536 static int fio_init_server_sock(void)
1538 struct sockaddr_un addr;
1543 sk = socket(AF_UNIX, SOCK_STREAM, 0);
1545 log_err("fio: socket: %s\n", strerror(errno));
1551 memset(&addr, 0, sizeof(addr));
1552 addr.sun_family = AF_UNIX;
1553 strncpy(addr.sun_path, bind_sock, sizeof(addr.sun_path) - 1);
1555 len = sizeof(addr.sun_family) + strlen(bind_sock) + 1;
1557 if (bind(sk, (struct sockaddr *) &addr, len) < 0) {
1558 log_err("fio: bind: %s\n", strerror(errno));
1567 static int fio_init_server_connection(void)
1572 dprint(FD_NET, "starting server\n");
1575 sk = fio_init_server_ip();
1577 sk = fio_init_server_sock();
1582 memset(bind_str, 0, sizeof(bind_str));
1591 src = &saddr_in6.sin6_addr;
1594 src = &saddr_in.sin_addr;
1597 p = (char *) inet_ntop(af, src, bind_str, sizeof(bind_str));
1599 sprintf(port, ",%u", fio_net_port);
1603 strncpy(bind_str, port, sizeof(bind_str) - 1);
1605 strncpy(bind_str, bind_sock, sizeof(bind_str) - 1);
1607 log_info("fio: server listening on %s\n", bind_str);
1609 if (listen(sk, 0) < 0) {
1610 log_err("fio: listen: %s\n", strerror(errno));
1618 int fio_server_parse_host(const char *host, int ipv6, struct in_addr *inp,
1619 struct in6_addr *inp6)
1625 ret = inet_pton(AF_INET6, host, inp6);
1627 ret = inet_pton(AF_INET, host, inp);
1630 struct addrinfo hints, *res;
1632 memset(&hints, 0, sizeof(hints));
1633 hints.ai_family = ipv6 ? AF_INET6 : AF_INET;
1634 hints.ai_socktype = SOCK_STREAM;
1636 ret = getaddrinfo(host, NULL, &hints, &res);
1638 log_err("fio: failed to resolve <%s> (%s)\n", host,
1644 memcpy(inp6, &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr, sizeof(*inp6));
1646 memcpy(inp, &((struct sockaddr_in *) res->ai_addr)->sin_addr, sizeof(*inp));
1656 * Parse a host/ip/port string. Reads from 'str'.
1661 * *ptr is the host, *port is the port, inp is the destination.
1663 * *ptr is the host, *port is the port, inp6 is the dest, and *ipv6 is 1.
1664 * For local domain sockets:
1665 * *ptr is the filename, *is_sock is 1.
1667 int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
1668 int *port, struct in_addr *inp,
1669 struct in6_addr *inp6, int *ipv6)
1671 const char *host = str;
1677 *port = fio_net_port;
1680 if (!strncmp(str, "sock:", 5)) {
1681 *ptr = strdup(str + 5);
1688 * Is it ip:<ip or host>:port
1690 if (!strncmp(host, "ip:", 3))
1692 else if (!strncmp(host, "ip4:", 4))
1694 else if (!strncmp(host, "ip6:", 4)) {
1697 } else if (host[0] == ':') {
1698 /* String is :port */
1701 if (!lport || lport > 65535) {
1702 log_err("fio: bad server port %u\n", lport);
1705 /* no hostname given, we are done */
1711 * If no port seen yet, check if there's a last ',' at the end
1714 portp = strchr(host, ',');
1718 lport = atoi(portp);
1719 if (!lport || lport > 65535) {
1720 log_err("fio: bad server port %u\n", lport);
1732 *ptr = strdup(host);
1734 if (fio_server_parse_host(*ptr, *ipv6, inp, inp6)) {
1741 *port = fio_net_port;
1747 * Server arg should be one of:
1749 * sock:/path/to/socket
1753 * Where sock uses unix domain sockets, and ip binds the server to
1754 * a specific interface. If no arguments are given to the server, it
1755 * uses IP and binds to 0.0.0.0.
1758 static int fio_handle_server_arg(void)
1760 int port = fio_net_port;
1761 int is_sock, ret = 0;
1763 saddr_in.sin_addr.s_addr = htonl(INADDR_ANY);
1765 if (!fio_server_arg)
1768 ret = fio_server_parse_string(fio_server_arg, &bind_sock, &is_sock,
1769 &port, &saddr_in.sin_addr,
1770 &saddr_in6.sin6_addr, &use_ipv6);
1772 if (!is_sock && bind_sock) {
1778 fio_net_port = port;
1779 saddr_in.sin_port = htons(port);
1780 saddr_in6.sin6_port = htons(port);
1784 static void sig_int(int sig)
1790 static void set_sig_handlers(void)
1792 struct sigaction act;
1794 memset(&act, 0, sizeof(act));
1795 act.sa_handler = sig_int;
1796 act.sa_flags = SA_RESTART;
1797 sigaction(SIGINT, &act, NULL);
1800 static int fio_server(void)
1804 dprint(FD_NET, "starting server\n");
1806 if (fio_handle_server_arg())
1809 sk = fio_init_server_connection();
1815 ret = accept_loop(sk);
1819 if (fio_server_arg) {
1820 free(fio_server_arg);
1821 fio_server_arg = NULL;
1829 void fio_server_got_signal(int signal)
1831 if (signal == SIGPIPE)
1834 log_info("\nfio: terminating on signal %d\n", signal);
1839 static int check_existing_pidfile(const char *pidfile)
1846 if (stat(pidfile, &sb))
1849 f = fopen(pidfile, "r");
1853 if (fread(buf, sb.st_size, 1, f) <= 0) {
1860 if (kill(pid, SIGCONT) < 0)
1861 return errno != ESRCH;
1866 static int write_pid(pid_t pid, const char *pidfile)
1870 fpid = fopen(pidfile, "w");
1872 log_err("fio: failed opening pid file %s\n", pidfile);
1876 fprintf(fpid, "%u\n", (unsigned int) pid);
1882 * If pidfile is specified, background us.
1884 int fio_start_server(char *pidfile)
1891 WSAStartup(MAKEWORD(2, 2), &wsd);
1895 return fio_server();
1897 if (check_existing_pidfile(pidfile)) {
1898 log_err("fio: pidfile %s exists and server appears alive\n",
1906 log_err("fio: failed server fork: %s", strerror(errno));
1910 ret = write_pid(pid, pidfile);
1916 openlog("fio", LOG_NDELAY|LOG_NOWAIT|LOG_PID, LOG_USER);
1918 close(STDIN_FILENO);
1919 close(STDOUT_FILENO);
1920 close(STDERR_FILENO);
1932 void fio_server_set_arg(const char *arg)
1934 fio_server_arg = strdup(arg);