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;
44 struct flist_head list; /* link on sk_out->list */
45 int flags; /* SK_F_* */
46 int opcode; /* Actual command fields */
50 struct flist_head next; /* Other sk_entry's, if linked command */
53 static char *fio_server_arg;
54 static char *bind_sock;
55 static struct sockaddr_in saddr_in;
56 static struct sockaddr_in6 saddr_in6;
59 static unsigned int has_zlib = 1;
61 static unsigned int has_zlib = 0;
63 static unsigned int use_zlib;
66 static pthread_key_t sk_out_key;
68 struct fio_fork_item {
69 struct flist_head list;
77 struct fio_mutex lock;
83 static const char *fio_server_ops[FIO_NET_CMD_NR] = {
109 static void sk_lock(struct sk_out *sk_out)
111 fio_mutex_down(&sk_out->lock);
114 static void sk_unlock(struct sk_out *sk_out)
116 fio_mutex_up(&sk_out->lock);
119 void sk_out_assign(struct sk_out *sk_out)
127 pthread_setspecific(sk_out_key, sk_out);
130 static void sk_out_free(struct sk_out *sk_out)
132 __fio_mutex_remove(&sk_out->lock);
133 __fio_mutex_remove(&sk_out->wait);
134 __fio_mutex_remove(&sk_out->xmit);
138 static int __sk_out_drop(struct sk_out *sk_out)
144 assert(sk_out->refs != 0);
145 refs = --sk_out->refs;
150 pthread_setspecific(sk_out_key, NULL);
158 void sk_out_drop(void)
160 struct sk_out *sk_out;
162 sk_out = pthread_getspecific(sk_out_key);
163 __sk_out_drop(sk_out);
166 static void __fio_init_net_cmd(struct fio_net_cmd *cmd, uint16_t opcode,
167 uint32_t pdu_len, uint64_t tag)
169 memset(cmd, 0, sizeof(*cmd));
171 cmd->version = __cpu_to_le16(FIO_SERVER_VER);
172 cmd->opcode = cpu_to_le16(opcode);
173 cmd->tag = cpu_to_le64(tag);
174 cmd->pdu_len = cpu_to_le32(pdu_len);
178 static void fio_init_net_cmd(struct fio_net_cmd *cmd, uint16_t opcode,
179 const void *pdu, uint32_t pdu_len, uint64_t tag)
181 __fio_init_net_cmd(cmd, opcode, pdu_len, tag);
184 memcpy(&cmd->payload, pdu, pdu_len);
187 const char *fio_server_op(unsigned int op)
191 if (op < FIO_NET_CMD_NR)
192 return fio_server_ops[op];
194 sprintf(buf, "UNKNOWN/%d", op);
198 static ssize_t iov_total_len(const struct iovec *iov, int count)
210 static int fio_sendv_data(int sk, struct iovec *iov, int count)
212 ssize_t total_len = iov_total_len(iov, count);
216 ret = writev(sk, iov, count);
223 if (ret >= iov->iov_len) {
228 iov->iov_base += ret;
234 else if (errno == EAGAIN || errno == EINTR)
238 } while (!exit_backend);
246 static int fio_send_data(int sk, const void *p, unsigned int len)
248 struct iovec iov = { .iov_base = (void *) p, .iov_len = len };
250 assert(len <= sizeof(struct fio_net_cmd) + FIO_SERVER_MAX_FRAGMENT_PDU);
252 return fio_sendv_data(sk, &iov, 1);
255 static int fio_recv_data(int sk, void *buf, unsigned int len, bool wait)
263 flags = OS_MSG_DONTWAIT;
266 int ret = recv(sk, p, len, flags);
276 else if (errno == EAGAIN || errno == EINTR) {
282 } while (!exit_backend);
290 static int verify_convert_cmd(struct fio_net_cmd *cmd)
294 cmd->cmd_crc16 = le16_to_cpu(cmd->cmd_crc16);
295 cmd->pdu_crc16 = le16_to_cpu(cmd->pdu_crc16);
297 crc = fio_crc16(cmd, FIO_NET_CMD_CRC_SZ);
298 if (crc != cmd->cmd_crc16) {
299 log_err("fio: server bad crc on command (got %x, wanted %x)\n",
300 cmd->cmd_crc16, crc);
304 cmd->version = le16_to_cpu(cmd->version);
305 cmd->opcode = le16_to_cpu(cmd->opcode);
306 cmd->flags = le32_to_cpu(cmd->flags);
307 cmd->tag = le64_to_cpu(cmd->tag);
308 cmd->pdu_len = le32_to_cpu(cmd->pdu_len);
310 switch (cmd->version) {
314 log_err("fio: bad server cmd version %d\n", cmd->version);
318 if (cmd->pdu_len > FIO_SERVER_MAX_FRAGMENT_PDU) {
319 log_err("fio: command payload too large: %u\n", cmd->pdu_len);
327 * Read (and defragment, if necessary) incoming commands
329 struct fio_net_cmd *fio_net_recv_cmd(int sk, bool wait)
331 struct fio_net_cmd cmd, *tmp, *cmdret = NULL;
332 size_t cmd_size = 0, pdu_offset = 0;
338 ret = fio_recv_data(sk, &cmd, sizeof(cmd), wait);
342 /* We have a command, verify it and swap if need be */
343 ret = verify_convert_cmd(&cmd);
348 /* if this is text, add room for \0 at the end */
349 cmd_size = sizeof(cmd) + cmd.pdu_len + 1;
352 cmd_size += cmd.pdu_len;
354 if (cmd_size / 1024 > FIO_SERVER_MAX_CMD_MB * 1024) {
355 log_err("fio: cmd+pdu too large (%llu)\n", (unsigned long long) cmd_size);
360 tmp = realloc(cmdret, cmd_size);
362 log_err("fio: server failed allocating cmd\n");
369 memcpy(cmdret, &cmd, sizeof(cmd));
370 else if (cmdret->opcode != cmd.opcode) {
371 log_err("fio: fragment opcode mismatch (%d != %d)\n",
372 cmdret->opcode, cmd.opcode);
380 /* There's payload, get it */
381 pdu = (char *) cmdret->payload + pdu_offset;
382 ret = fio_recv_data(sk, pdu, cmd.pdu_len, wait);
386 /* Verify payload crc */
387 crc = fio_crc16(pdu, cmd.pdu_len);
388 if (crc != cmd.pdu_crc16) {
389 log_err("fio: server bad crc on payload ");
390 log_err("(got %x, wanted %x)\n", cmd.pdu_crc16, crc);
395 pdu_offset += cmd.pdu_len;
397 cmdret->pdu_len += cmd.pdu_len;
399 } while (cmd.flags & FIO_NET_CMD_F_MORE);
405 /* zero-terminate text input */
406 if (cmdret->pdu_len) {
407 if (cmdret->opcode == FIO_NET_CMD_TEXT) {
408 struct cmd_text_pdu *__pdu = (struct cmd_text_pdu *) cmdret->payload;
409 char *buf = (char *) __pdu->buf;
411 buf[__pdu->buf_len] = '\0';
412 } else if (cmdret->opcode == FIO_NET_CMD_JOB) {
413 struct cmd_job_pdu *__pdu = (struct cmd_job_pdu *) cmdret->payload;
414 char *buf = (char *) __pdu->buf;
415 int len = le32_to_cpu(__pdu->buf_len);
421 /* frag flag is internal */
422 cmdret->flags &= ~FIO_NET_CMD_F_MORE;
428 static void add_reply(uint64_t tag, struct flist_head *list)
430 struct fio_net_cmd_reply *reply;
432 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
433 flist_add_tail(&reply->list, list);
436 static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
438 struct fio_net_cmd_reply *reply;
440 reply = calloc(1, sizeof(*reply));
441 INIT_FLIST_HEAD(&reply->list);
442 fio_gettime(&reply->ts, NULL);
443 reply->saved_tag = tag;
444 reply->opcode = opcode;
446 return (uintptr_t) reply;
449 static void free_reply(uint64_t tag)
451 struct fio_net_cmd_reply *reply;
453 reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
457 static void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu)
461 cmd->cmd_crc16 = __cpu_to_le16(fio_crc16(cmd, FIO_NET_CMD_CRC_SZ));
463 pdu_len = le32_to_cpu(cmd->pdu_len);
464 cmd->pdu_crc16 = __cpu_to_le16(fio_crc16(pdu, pdu_len));
467 static void fio_net_cmd_crc(struct fio_net_cmd *cmd)
469 fio_net_cmd_crc_pdu(cmd, cmd->payload);
472 int fio_net_send_cmd(int fd, uint16_t opcode, const void *buf, off_t size,
473 uint64_t *tagptr, struct flist_head *list)
475 struct fio_net_cmd *cmd = NULL;
476 size_t this_len, cur_len = 0;
482 tag = *tagptr = alloc_reply(*tagptr, opcode);
484 tag = tagptr ? *tagptr : 0;
488 if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU)
489 this_len = FIO_SERVER_MAX_FRAGMENT_PDU;
491 if (!cmd || cur_len < sizeof(*cmd) + this_len) {
495 cur_len = sizeof(*cmd) + this_len;
496 cmd = malloc(cur_len);
499 fio_init_net_cmd(cmd, opcode, buf, this_len, tag);
502 cmd->flags = __cpu_to_le32(FIO_NET_CMD_F_MORE);
504 fio_net_cmd_crc(cmd);
506 ret = fio_send_data(fd, cmd, sizeof(*cmd) + this_len);
509 } while (!ret && size);
515 add_reply(tag, list);
524 static struct sk_entry *fio_net_prep_cmd(uint16_t opcode, void *buf,
525 size_t size, uint64_t *tagptr,
528 struct sk_entry *entry;
530 entry = smalloc(sizeof(*entry));
531 INIT_FLIST_HEAD(&entry->next);
532 entry->opcode = opcode;
533 if (flags & SK_F_COPY) {
534 entry->buf = smalloc(size);
535 memcpy(entry->buf, buf, size);
541 entry->tag = *tagptr;
544 entry->flags = flags;
548 static int handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry);
550 static void fio_net_queue_entry(struct sk_entry *entry)
552 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
554 if (entry->flags & SK_F_INLINE)
555 handle_sk_entry(sk_out, entry);
558 flist_add_tail(&entry->list, &sk_out->list);
561 fio_mutex_up(&sk_out->wait);
565 static int fio_net_queue_cmd(uint16_t opcode, void *buf, off_t size,
566 uint64_t *tagptr, int flags)
568 struct sk_entry *entry;
570 entry = fio_net_prep_cmd(opcode, buf, size, tagptr, flags);
572 fio_net_queue_entry(entry);
579 static int fio_net_send_simple_stack_cmd(int sk, uint16_t opcode, uint64_t tag)
581 struct fio_net_cmd cmd;
583 fio_init_net_cmd(&cmd, opcode, NULL, 0, tag);
584 fio_net_cmd_crc(&cmd);
586 return fio_send_data(sk, &cmd, sizeof(cmd));
590 * If 'list' is non-NULL, then allocate and store the sent command for
591 * later verification.
593 int fio_net_send_simple_cmd(int sk, uint16_t opcode, uint64_t tag,
594 struct flist_head *list)
599 tag = alloc_reply(tag, opcode);
601 ret = fio_net_send_simple_stack_cmd(sk, opcode, tag);
610 add_reply(tag, list);
615 static int fio_net_queue_quit(void)
617 dprint(FD_NET, "server: sending quit\n");
619 return fio_net_queue_cmd(FIO_NET_CMD_QUIT, NULL, 0, NULL, SK_F_SIMPLE | SK_F_INLINE);
622 int fio_net_send_quit(int sk)
624 dprint(FD_NET, "server: sending quit\n");
626 return fio_net_send_simple_cmd(sk, FIO_NET_CMD_QUIT, 0, NULL);
629 static int fio_net_send_ack(struct fio_net_cmd *cmd, int error, int signal)
631 struct cmd_end_pdu epdu;
637 epdu.error = __cpu_to_le32(error);
638 epdu.signal = __cpu_to_le32(signal);
639 return fio_net_queue_cmd(FIO_NET_CMD_STOP, &epdu, sizeof(epdu), &tag, SK_F_COPY | SK_F_INLINE);
642 static int fio_net_queue_stop(int error, int signal)
644 dprint(FD_NET, "server: sending stop (%d, %d)\n", error, signal);
645 return fio_net_send_ack(NULL, error, signal);
648 static void fio_server_add_fork_item(pid_t pid, struct flist_head *list)
650 struct fio_fork_item *ffi;
652 ffi = malloc(sizeof(*ffi));
657 flist_add_tail(&ffi->list, list);
660 static void fio_server_add_conn_pid(struct flist_head *conn_list, pid_t pid)
662 dprint(FD_NET, "server: forked off connection job (pid=%u)\n", (int) pid);
663 fio_server_add_fork_item(pid, conn_list);
666 static void fio_server_add_job_pid(struct flist_head *job_list, pid_t pid)
668 dprint(FD_NET, "server: forked off job job (pid=%u)\n", (int) pid);
669 fio_server_add_fork_item(pid, job_list);
672 static void fio_server_check_fork_item(struct fio_fork_item *ffi)
676 ret = waitpid(ffi->pid, &status, WNOHANG);
678 if (errno == ECHILD) {
679 log_err("fio: connection pid %u disappeared\n", (int) ffi->pid);
682 log_err("fio: waitpid: %s\n", strerror(errno));
683 } else if (ret == ffi->pid) {
684 if (WIFSIGNALED(status)) {
685 ffi->signal = WTERMSIG(status);
688 if (WIFEXITED(status)) {
689 if (WEXITSTATUS(status))
690 ffi->exitval = WEXITSTATUS(status);
696 static void fio_server_fork_item_done(struct fio_fork_item *ffi, bool stop)
698 dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", (int) ffi->pid, ffi->signal, ffi->exitval);
701 * Fold STOP and QUIT...
704 fio_net_queue_stop(ffi->exitval, ffi->signal);
705 fio_net_queue_quit();
708 flist_del(&ffi->list);
712 static void fio_server_check_fork_items(struct flist_head *list, bool stop)
714 struct flist_head *entry, *tmp;
715 struct fio_fork_item *ffi;
717 flist_for_each_safe(entry, tmp, list) {
718 ffi = flist_entry(entry, struct fio_fork_item, list);
720 fio_server_check_fork_item(ffi);
723 fio_server_fork_item_done(ffi, stop);
727 static void fio_server_check_jobs(struct flist_head *job_list)
729 fio_server_check_fork_items(job_list, true);
732 static void fio_server_check_conns(struct flist_head *conn_list)
734 fio_server_check_fork_items(conn_list, false);
737 static int handle_load_file_cmd(struct fio_net_cmd *cmd)
739 struct cmd_load_file_pdu *pdu = (struct cmd_load_file_pdu *) cmd->payload;
740 void *file_name = pdu->file;
741 struct cmd_start_pdu spdu;
743 dprint(FD_NET, "server: loading local file %s\n", (char *) file_name);
745 pdu->name_len = le16_to_cpu(pdu->name_len);
746 pdu->client_type = le16_to_cpu(pdu->client_type);
748 if (parse_jobs_ini(file_name, 0, 0, pdu->client_type)) {
749 fio_net_queue_quit();
753 spdu.jobs = cpu_to_le32(thread_number);
754 spdu.stat_outputs = cpu_to_le32(stat_number);
755 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
759 static int handle_run_cmd(struct sk_out *sk_out, struct flist_head *job_list,
760 struct fio_net_cmd *cmd)
765 sk_out_assign(sk_out);
772 fio_server_add_job_pid(job_list, pid);
776 ret = fio_backend(sk_out);
782 static int handle_job_cmd(struct fio_net_cmd *cmd)
784 struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmd->payload;
785 void *buf = pdu->buf;
786 struct cmd_start_pdu spdu;
788 pdu->buf_len = le32_to_cpu(pdu->buf_len);
789 pdu->client_type = le32_to_cpu(pdu->client_type);
791 if (parse_jobs_ini(buf, 1, 0, pdu->client_type)) {
792 fio_net_queue_quit();
796 spdu.jobs = cpu_to_le32(thread_number);
797 spdu.stat_outputs = cpu_to_le32(stat_number);
799 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
803 static int handle_jobline_cmd(struct fio_net_cmd *cmd)
805 void *pdu = cmd->payload;
806 struct cmd_single_line_pdu *cslp;
807 struct cmd_line_pdu *clp;
808 unsigned long offset;
809 struct cmd_start_pdu spdu;
814 clp->lines = le16_to_cpu(clp->lines);
815 clp->client_type = le16_to_cpu(clp->client_type);
816 argv = malloc(clp->lines * sizeof(char *));
817 offset = sizeof(*clp);
819 dprint(FD_NET, "server: %d command line args\n", clp->lines);
821 for (i = 0; i < clp->lines; i++) {
823 argv[i] = (char *) cslp->text;
825 offset += sizeof(*cslp) + le16_to_cpu(cslp->len);
826 dprint(FD_NET, "server: %d: %s\n", i, argv[i]);
829 if (parse_cmd_line(clp->lines, argv, clp->client_type)) {
830 fio_net_queue_quit();
837 spdu.jobs = cpu_to_le32(thread_number);
838 spdu.stat_outputs = cpu_to_le32(stat_number);
840 fio_net_queue_cmd(FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, SK_F_COPY);
844 static int handle_probe_cmd(struct fio_net_cmd *cmd)
846 struct cmd_client_probe_pdu *pdu = (struct cmd_client_probe_pdu *) cmd->payload;
847 struct cmd_probe_reply_pdu probe;
848 uint64_t tag = cmd->tag;
850 dprint(FD_NET, "server: sending probe reply\n");
852 strcpy(me, (char *) pdu->server);
854 memset(&probe, 0, sizeof(probe));
855 gethostname((char *) probe.hostname, sizeof(probe.hostname));
856 #ifdef CONFIG_BIG_ENDIAN
859 strncpy((char *) probe.fio_version, fio_version_string, sizeof(probe.fio_version) - 1);
862 probe.arch = FIO_ARCH;
863 probe.bpp = sizeof(void *);
864 probe.cpus = __cpu_to_le32(cpus_online());
867 * If the client supports compression and we do too, then enable it
869 if (has_zlib && le64_to_cpu(pdu->flags) & FIO_PROBE_FLAG_ZLIB) {
870 probe.flags = __cpu_to_le64(FIO_PROBE_FLAG_ZLIB);
877 return fio_net_queue_cmd(FIO_NET_CMD_PROBE, &probe, sizeof(probe), &tag, SK_F_COPY);
880 static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
883 uint64_t tag = cmd->tag;
887 dprint(FD_NET, "server sending status\n");
890 * Fake ETA return if we don't have a local one, otherwise the client
891 * will end up timing out waiting for a response to the ETA request
893 je = get_jobs_eta(true, &size);
896 je = calloc(1, size);
898 je->nr_running = cpu_to_le32(je->nr_running);
899 je->nr_ramp = cpu_to_le32(je->nr_ramp);
900 je->nr_pending = cpu_to_le32(je->nr_pending);
901 je->nr_setting_up = cpu_to_le32(je->nr_setting_up);
902 je->files_open = cpu_to_le32(je->files_open);
904 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
905 je->m_rate[i] = cpu_to_le64(je->m_rate[i]);
906 je->t_rate[i] = cpu_to_le64(je->t_rate[i]);
907 je->m_iops[i] = cpu_to_le32(je->m_iops[i]);
908 je->t_iops[i] = cpu_to_le32(je->t_iops[i]);
909 je->rate[i] = cpu_to_le64(je->rate[i]);
910 je->iops[i] = cpu_to_le32(je->iops[i]);
913 je->elapsed_sec = cpu_to_le64(je->elapsed_sec);
914 je->eta_sec = cpu_to_le64(je->eta_sec);
915 je->nr_threads = cpu_to_le32(je->nr_threads);
916 je->is_pow2 = cpu_to_le32(je->is_pow2);
917 je->unit_base = cpu_to_le32(je->unit_base);
920 fio_net_queue_cmd(FIO_NET_CMD_ETA, je, size, &tag, SK_F_FREE);
924 static int send_update_job_reply(uint64_t __tag, int error)
926 uint64_t tag = __tag;
929 pdu_error = __cpu_to_le32(error);
930 return fio_net_queue_cmd(FIO_NET_CMD_UPDATE_JOB, &pdu_error, sizeof(pdu_error), &tag, SK_F_COPY);
933 static int handle_update_job_cmd(struct fio_net_cmd *cmd)
935 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
936 struct thread_data *td;
939 tnumber = le32_to_cpu(pdu->thread_number);
941 dprint(FD_NET, "server: updating options for job %u\n", tnumber);
943 if (!tnumber || tnumber > thread_number) {
944 send_update_job_reply(cmd->tag, ENODEV);
948 td = &threads[tnumber - 1];
949 convert_thread_options_to_cpu(&td->o, &pdu->top);
950 send_update_job_reply(cmd->tag, 0);
954 static int handle_trigger_cmd(struct fio_net_cmd *cmd, struct flist_head *job_list)
956 struct cmd_vtrigger_pdu *pdu = (struct cmd_vtrigger_pdu *) cmd->payload;
957 char *buf = (char *) pdu->cmd;
958 struct all_io_list *rep;
961 pdu->len = le16_to_cpu(pdu->len);
962 buf[pdu->len] = '\0';
964 rep = get_all_io_list(IO_LIST_ALL, &sz);
966 struct all_io_list state;
968 state.threads = cpu_to_le64((uint64_t) 0);
969 fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, SK_F_COPY | SK_F_INLINE);
971 fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, rep, sz, NULL, SK_F_FREE | SK_F_INLINE);
973 fio_terminate_threads(TERMINATE_ALL);
974 fio_server_check_jobs(job_list);
979 static int handle_command(struct sk_out *sk_out, struct flist_head *job_list,
980 struct fio_net_cmd *cmd)
984 dprint(FD_NET, "server: got op [%s], pdu=%u, tag=%llx\n",
985 fio_server_op(cmd->opcode), cmd->pdu_len,
986 (unsigned long long) cmd->tag);
988 switch (cmd->opcode) {
989 case FIO_NET_CMD_QUIT:
990 fio_terminate_threads(TERMINATE_ALL);
993 case FIO_NET_CMD_EXIT:
996 case FIO_NET_CMD_LOAD_FILE:
997 ret = handle_load_file_cmd(cmd);
999 case FIO_NET_CMD_JOB:
1000 ret = handle_job_cmd(cmd);
1002 case FIO_NET_CMD_JOBLINE:
1003 ret = handle_jobline_cmd(cmd);
1005 case FIO_NET_CMD_PROBE:
1006 ret = handle_probe_cmd(cmd);
1008 case FIO_NET_CMD_SEND_ETA:
1009 ret = handle_send_eta_cmd(cmd);
1011 case FIO_NET_CMD_RUN:
1012 ret = handle_run_cmd(sk_out, job_list, cmd);
1014 case FIO_NET_CMD_UPDATE_JOB:
1015 ret = handle_update_job_cmd(cmd);
1017 case FIO_NET_CMD_VTRIGGER:
1018 ret = handle_trigger_cmd(cmd, job_list);
1020 case FIO_NET_CMD_SENDFILE: {
1021 struct cmd_sendfile_reply *in;
1022 struct cmd_reply *rep;
1024 rep = (struct cmd_reply *) (uintptr_t) cmd->tag;
1026 in = (struct cmd_sendfile_reply *) cmd->payload;
1027 in->size = le32_to_cpu(in->size);
1028 in->error = le32_to_cpu(in->error);
1031 rep->error = in->error;
1034 rep->data = smalloc(in->size);
1037 rep->error = ENOMEM;
1039 rep->size = in->size;
1040 memcpy(rep->data, in->data, in->size);
1043 fio_mutex_up(&rep->lock);
1047 log_err("fio: unknown opcode: %s\n", fio_server_op(cmd->opcode));
1055 * Send a command with a separate PDU, not inlined in the command
1057 static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
1058 off_t size, uint64_t tag, uint32_t flags)
1060 struct fio_net_cmd cmd;
1061 struct iovec iov[2];
1065 iov[0].iov_base = (void *) &cmd;
1066 iov[0].iov_len = sizeof(cmd);
1069 uint32_t this_flags = flags;
1072 if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU)
1073 this_len = FIO_SERVER_MAX_FRAGMENT_PDU;
1075 if (this_len < size)
1076 this_flags |= FIO_NET_CMD_F_MORE;
1078 __fio_init_net_cmd(&cmd, opcode, this_len, tag);
1079 cmd.flags = __cpu_to_le32(this_flags);
1080 fio_net_cmd_crc_pdu(&cmd, buf);
1082 iov[1].iov_base = (void *) buf;
1083 iov[1].iov_len = this_len;
1085 ret = fio_sendv_data(sk, iov, 2);
1088 } while (!ret && size);
1093 static void finish_entry(struct sk_entry *entry)
1095 if (entry->flags & SK_F_FREE)
1097 else if (entry->flags & SK_F_COPY)
1103 static void entry_set_flags(struct sk_entry *entry, struct flist_head *list,
1104 unsigned int *flags)
1106 if (!flist_empty(list))
1107 *flags = FIO_NET_CMD_F_MORE;
1112 static int send_vec_entry(struct sk_out *sk_out, struct sk_entry *first)
1117 entry_set_flags(first, &first->next, &flags);
1119 ret = fio_send_cmd_ext_pdu(sk_out->sk, first->opcode, first->buf,
1120 first->size, first->tag, flags);
1122 while (!flist_empty(&first->next)) {
1123 struct sk_entry *next;
1125 next = flist_first_entry(&first->next, struct sk_entry, list);
1126 flist_del_init(&next->list);
1128 entry_set_flags(next, &first->next, &flags);
1130 ret += fio_send_cmd_ext_pdu(sk_out->sk, next->opcode, next->buf,
1131 next->size, next->tag, flags);
1138 static int handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry)
1142 fio_mutex_down(&sk_out->xmit);
1144 if (entry->flags & SK_F_VEC)
1145 ret = send_vec_entry(sk_out, entry);
1146 else if (entry->flags & SK_F_SIMPLE) {
1147 ret = fio_net_send_simple_cmd(sk_out->sk, entry->opcode,
1150 ret = fio_net_send_cmd(sk_out->sk, entry->opcode, entry->buf,
1151 entry->size, &entry->tag, NULL);
1154 fio_mutex_up(&sk_out->xmit);
1157 log_err("fio: failed handling cmd %s\n", fio_server_op(entry->opcode));
1159 finish_entry(entry);
1163 static int handle_xmits(struct sk_out *sk_out)
1165 struct sk_entry *entry;
1170 if (flist_empty(&sk_out->list)) {
1175 flist_splice_init(&sk_out->list, &list);
1178 while (!flist_empty(&list)) {
1179 entry = flist_entry(list.next, struct sk_entry, list);
1180 flist_del(&entry->list);
1181 ret += handle_sk_entry(sk_out, entry);
1187 static int handle_connection(struct sk_out *sk_out)
1189 struct fio_net_cmd *cmd = NULL;
1190 FLIST_HEAD(job_list);
1196 while (!exit_backend) {
1197 struct pollfd pfd = {
1206 if (!flist_empty(&job_list))
1209 handle_xmits(sk_out);
1211 ret = poll(&pfd, 1, 0);
1215 log_err("fio: poll: %s\n", strerror(errno));
1218 fio_server_check_jobs(&job_list);
1219 fio_mutex_down_timeout(&sk_out->wait, timeout);
1223 if (pfd.revents & POLLIN)
1225 if (pfd.revents & (POLLERR|POLLHUP)) {
1229 } while (!exit_backend);
1231 fio_server_check_jobs(&job_list);
1236 cmd = fio_net_recv_cmd(sk_out->sk, true);
1242 ret = handle_command(sk_out, &job_list, cmd);
1253 handle_xmits(sk_out);
1257 __sk_out_drop(sk_out);
1261 /* get the address on this host bound by the input socket,
1262 * whether it is ipv6 or ipv4 */
1264 static int get_my_addr_str(int sk)
1266 struct sockaddr_in6 myaddr6 = { 0, };
1267 struct sockaddr_in myaddr4 = { 0, };
1268 struct sockaddr *sockaddr_p;
1274 len = sizeof(myaddr6);
1275 sockaddr_p = (struct sockaddr * )&myaddr6;
1276 net_addr = (char * )&myaddr6.sin6_addr;
1278 len = sizeof(myaddr4);
1279 sockaddr_p = (struct sockaddr * )&myaddr4;
1280 net_addr = (char * )&myaddr4.sin_addr;
1283 ret = getsockname(sk, sockaddr_p, &len);
1285 log_err("fio: getsockname: %s\n", strerror(errno));
1289 if (!inet_ntop(use_ipv6?AF_INET6:AF_INET, net_addr, client_sockaddr_str, INET6_ADDRSTRLEN - 1)) {
1290 log_err("inet_ntop: failed to convert addr to string\n");
1294 dprint(FD_NET, "fio server bound to addr %s\n", client_sockaddr_str);
1298 static int accept_loop(int listen_sk)
1300 struct sockaddr_in addr;
1301 struct sockaddr_in6 addr6;
1302 socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr);
1304 int ret = 0, sk, exitval = 0;
1305 FLIST_HEAD(conn_list);
1307 dprint(FD_NET, "server enter accept loop\n");
1309 fio_set_fd_nonblocking(listen_sk, "server");
1311 while (!exit_backend) {
1312 struct sk_out *sk_out;
1318 pfd.events = POLLIN;
1322 if (!flist_empty(&conn_list))
1325 ret = poll(&pfd, 1, timeout);
1329 log_err("fio: poll: %s\n", strerror(errno));
1332 fio_server_check_conns(&conn_list);
1336 if (pfd.revents & POLLIN)
1338 } while (!exit_backend);
1340 fio_server_check_conns(&conn_list);
1342 if (exit_backend || ret < 0)
1346 sk = accept(listen_sk, (struct sockaddr *) &addr6, &len);
1348 sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
1351 log_err("fio: accept: %s\n", strerror(errno));
1356 from = inet_ntop(AF_INET6, (struct sockaddr *) &addr6.sin6_addr, buf, sizeof(buf));
1358 from = inet_ntop(AF_INET, (struct sockaddr *) &addr.sin_addr, buf, sizeof(buf));
1360 dprint(FD_NET, "server: connect from %s\n", from);
1362 sk_out = smalloc(sizeof(*sk_out));
1364 INIT_FLIST_HEAD(&sk_out->list);
1365 __fio_mutex_init(&sk_out->lock, FIO_MUTEX_UNLOCKED);
1366 __fio_mutex_init(&sk_out->wait, FIO_MUTEX_LOCKED);
1367 __fio_mutex_init(&sk_out->xmit, FIO_MUTEX_UNLOCKED);
1372 fio_server_add_conn_pid(&conn_list, pid);
1376 /* if error, it's already logged, non-fatal */
1377 get_my_addr_str(sk);
1380 * Assign sk_out here, it'll be dropped in handle_connection()
1381 * since that function calls _exit() when done
1383 sk_out_assign(sk_out);
1384 handle_connection(sk_out);
1390 int fio_server_text_output(int level, const char *buf, size_t len)
1392 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
1393 struct cmd_text_pdu *pdu;
1397 if (!sk_out || sk_out->sk == -1)
1400 tlen = sizeof(*pdu) + len;
1403 pdu->level = __cpu_to_le32(level);
1404 pdu->buf_len = __cpu_to_le32(len);
1406 gettimeofday(&tv, NULL);
1407 pdu->log_sec = __cpu_to_le64(tv.tv_sec);
1408 pdu->log_usec = __cpu_to_le64(tv.tv_usec);
1410 memcpy(pdu->buf, buf, len);
1412 fio_net_queue_cmd(FIO_NET_CMD_TEXT, pdu, tlen, NULL, SK_F_COPY);
1417 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
1419 dst->max_val = cpu_to_le64(src->max_val);
1420 dst->min_val = cpu_to_le64(src->min_val);
1421 dst->samples = cpu_to_le64(src->samples);
1424 * Encode to IEEE 754 for network transfer
1426 dst->mean.u.i = cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
1427 dst->S.u.i = cpu_to_le64(fio_double_to_uint64(src->S.u.f));
1430 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1434 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1435 dst->max_run[i] = cpu_to_le64(src->max_run[i]);
1436 dst->min_run[i] = cpu_to_le64(src->min_run[i]);
1437 dst->max_bw[i] = cpu_to_le64(src->max_bw[i]);
1438 dst->min_bw[i] = cpu_to_le64(src->min_bw[i]);
1439 dst->iobytes[i] = cpu_to_le64(src->iobytes[i]);
1440 dst->agg[i] = cpu_to_le64(src->agg[i]);
1443 dst->kb_base = cpu_to_le32(src->kb_base);
1444 dst->unit_base = cpu_to_le32(src->unit_base);
1445 dst->groupid = cpu_to_le32(src->groupid);
1446 dst->unified_rw_rep = cpu_to_le32(src->unified_rw_rep);
1450 * Send a CMD_TS, which packs struct thread_stat and group_run_stats
1451 * into a single payload.
1453 void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
1455 struct cmd_ts_pdu p;
1458 uint64_t *ss_iops, *ss_bw;
1460 dprint(FD_NET, "server sending end stats\n");
1462 memset(&p, 0, sizeof(p));
1464 strncpy(p.ts.name, ts->name, FIO_JOBNAME_SIZE - 1);
1465 strncpy(p.ts.verror, ts->verror, FIO_VERROR_SIZE - 1);
1466 strncpy(p.ts.description, ts->description, FIO_JOBDESC_SIZE - 1);
1468 p.ts.error = cpu_to_le32(ts->error);
1469 p.ts.thread_number = cpu_to_le32(ts->thread_number);
1470 p.ts.groupid = cpu_to_le32(ts->groupid);
1471 p.ts.pid = cpu_to_le32(ts->pid);
1472 p.ts.members = cpu_to_le32(ts->members);
1473 p.ts.unified_rw_rep = cpu_to_le32(ts->unified_rw_rep);
1475 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1476 convert_io_stat(&p.ts.clat_stat[i], &ts->clat_stat[i]);
1477 convert_io_stat(&p.ts.slat_stat[i], &ts->slat_stat[i]);
1478 convert_io_stat(&p.ts.lat_stat[i], &ts->lat_stat[i]);
1479 convert_io_stat(&p.ts.bw_stat[i], &ts->bw_stat[i]);
1480 convert_io_stat(&p.ts.iops_stat[i], &ts->iops_stat[i]);
1483 p.ts.usr_time = cpu_to_le64(ts->usr_time);
1484 p.ts.sys_time = cpu_to_le64(ts->sys_time);
1485 p.ts.ctx = cpu_to_le64(ts->ctx);
1486 p.ts.minf = cpu_to_le64(ts->minf);
1487 p.ts.majf = cpu_to_le64(ts->majf);
1488 p.ts.clat_percentiles = cpu_to_le32(ts->clat_percentiles);
1489 p.ts.lat_percentiles = cpu_to_le32(ts->lat_percentiles);
1490 p.ts.percentile_precision = cpu_to_le64(ts->percentile_precision);
1492 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
1493 fio_fp64_t *src = &ts->percentile_list[i];
1494 fio_fp64_t *dst = &p.ts.percentile_list[i];
1496 dst->u.i = cpu_to_le64(fio_double_to_uint64(src->u.f));
1499 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
1500 p.ts.io_u_map[i] = cpu_to_le32(ts->io_u_map[i]);
1501 p.ts.io_u_submit[i] = cpu_to_le32(ts->io_u_submit[i]);
1502 p.ts.io_u_complete[i] = cpu_to_le32(ts->io_u_complete[i]);
1505 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
1506 p.ts.io_u_lat_n[i] = cpu_to_le32(ts->io_u_lat_n[i]);
1507 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
1508 p.ts.io_u_lat_u[i] = cpu_to_le32(ts->io_u_lat_u[i]);
1509 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
1510 p.ts.io_u_lat_m[i] = cpu_to_le32(ts->io_u_lat_m[i]);
1512 for (i = 0; i < DDIR_RWDIR_CNT; i++)
1513 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1514 p.ts.io_u_plat[i][j] = cpu_to_le32(ts->io_u_plat[i][j]);
1516 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1517 p.ts.total_io_u[i] = cpu_to_le64(ts->total_io_u[i]);
1518 p.ts.short_io_u[i] = cpu_to_le64(ts->short_io_u[i]);
1519 p.ts.drop_io_u[i] = cpu_to_le64(ts->drop_io_u[i]);
1522 p.ts.total_submit = cpu_to_le64(ts->total_submit);
1523 p.ts.total_complete = cpu_to_le64(ts->total_complete);
1525 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1526 p.ts.io_bytes[i] = cpu_to_le64(ts->io_bytes[i]);
1527 p.ts.runtime[i] = cpu_to_le64(ts->runtime[i]);
1530 p.ts.total_run_time = cpu_to_le64(ts->total_run_time);
1531 p.ts.continue_on_error = cpu_to_le16(ts->continue_on_error);
1532 p.ts.total_err_count = cpu_to_le64(ts->total_err_count);
1533 p.ts.first_error = cpu_to_le32(ts->first_error);
1534 p.ts.kb_base = cpu_to_le32(ts->kb_base);
1535 p.ts.unit_base = cpu_to_le32(ts->unit_base);
1537 p.ts.latency_depth = cpu_to_le32(ts->latency_depth);
1538 p.ts.latency_target = cpu_to_le64(ts->latency_target);
1539 p.ts.latency_window = cpu_to_le64(ts->latency_window);
1540 p.ts.latency_percentile.u.i = cpu_to_le64(fio_double_to_uint64(ts->latency_percentile.u.f));
1542 p.ts.sig_figs = cpu_to_le32(ts->sig_figs);
1544 p.ts.nr_block_infos = cpu_to_le64(ts->nr_block_infos);
1545 for (i = 0; i < p.ts.nr_block_infos; i++)
1546 p.ts.block_infos[i] = cpu_to_le32(ts->block_infos[i]);
1548 p.ts.ss_dur = cpu_to_le64(ts->ss_dur);
1549 p.ts.ss_state = cpu_to_le32(ts->ss_state);
1550 p.ts.ss_head = cpu_to_le32(ts->ss_head);
1551 p.ts.ss_limit.u.i = cpu_to_le64(fio_double_to_uint64(ts->ss_limit.u.f));
1552 p.ts.ss_slope.u.i = cpu_to_le64(fio_double_to_uint64(ts->ss_slope.u.f));
1553 p.ts.ss_deviation.u.i = cpu_to_le64(fio_double_to_uint64(ts->ss_deviation.u.f));
1554 p.ts.ss_criterion.u.i = cpu_to_le64(fio_double_to_uint64(ts->ss_criterion.u.f));
1556 convert_gs(&p.rs, rs);
1558 dprint(FD_NET, "ts->ss_state = %d\n", ts->ss_state);
1559 if (ts->ss_state & FIO_SS_DATA) {
1560 dprint(FD_NET, "server sending steadystate ring buffers\n");
1562 ss_buf = malloc(sizeof(p) + 2*ts->ss_dur*sizeof(uint64_t));
1564 memcpy(ss_buf, &p, sizeof(p));
1566 ss_iops = (uint64_t *) ((struct cmd_ts_pdu *)ss_buf + 1);
1567 ss_bw = ss_iops + (int) ts->ss_dur;
1568 for (i = 0; i < ts->ss_dur; i++) {
1569 ss_iops[i] = cpu_to_le64(ts->ss_iops_data[i]);
1570 ss_bw[i] = cpu_to_le64(ts->ss_bw_data[i]);
1573 fio_net_queue_cmd(FIO_NET_CMD_TS, ss_buf, sizeof(p) + 2*ts->ss_dur*sizeof(uint64_t), NULL, SK_F_COPY);
1578 fio_net_queue_cmd(FIO_NET_CMD_TS, &p, sizeof(p), NULL, SK_F_COPY);
1581 void fio_server_send_gs(struct group_run_stats *rs)
1583 struct group_run_stats gs;
1585 dprint(FD_NET, "server sending group run stats\n");
1587 convert_gs(&gs, rs);
1588 fio_net_queue_cmd(FIO_NET_CMD_GS, &gs, sizeof(gs), NULL, SK_F_COPY);
1591 void fio_server_send_job_options(struct flist_head *opt_list,
1592 unsigned int groupid)
1594 struct cmd_job_option pdu;
1595 struct flist_head *entry;
1597 if (flist_empty(opt_list))
1600 flist_for_each(entry, opt_list) {
1601 struct print_option *p;
1604 p = flist_entry(entry, struct print_option, list);
1605 memset(&pdu, 0, sizeof(pdu));
1607 if (groupid == -1U) {
1608 pdu.global = __cpu_to_le16(1);
1612 pdu.groupid = cpu_to_le32(groupid);
1614 len = strlen(p->name);
1615 if (len >= sizeof(pdu.name)) {
1616 len = sizeof(pdu.name) - 1;
1617 pdu.truncated = __cpu_to_le16(1);
1619 memcpy(pdu.name, p->name, len);
1621 len = strlen(p->value);
1622 if (len >= sizeof(pdu.value)) {
1623 len = sizeof(pdu.value) - 1;
1624 pdu.truncated = __cpu_to_le16(1);
1626 memcpy(pdu.value, p->value, len);
1628 fio_net_queue_cmd(FIO_NET_CMD_JOB_OPT, &pdu, sizeof(pdu), NULL, SK_F_COPY);
1632 static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src)
1636 for (i = 0; i < 2; i++) {
1637 dst->ios[i] = cpu_to_le64(src->ios[i]);
1638 dst->merges[i] = cpu_to_le64(src->merges[i]);
1639 dst->sectors[i] = cpu_to_le64(src->sectors[i]);
1640 dst->ticks[i] = cpu_to_le64(src->ticks[i]);
1643 dst->io_ticks = cpu_to_le64(src->io_ticks);
1644 dst->time_in_queue = cpu_to_le64(src->time_in_queue);
1645 dst->slavecount = cpu_to_le32(src->slavecount);
1646 dst->max_util.u.i = cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
1649 static void convert_dus(struct disk_util_stat *dst, struct disk_util_stat *src)
1653 dst->name[FIO_DU_NAME_SZ - 1] = '\0';
1654 strncpy((char *) dst->name, (char *) src->name, FIO_DU_NAME_SZ - 1);
1656 for (i = 0; i < 2; i++) {
1657 dst->s.ios[i] = cpu_to_le64(src->s.ios[i]);
1658 dst->s.merges[i] = cpu_to_le64(src->s.merges[i]);
1659 dst->s.sectors[i] = cpu_to_le64(src->s.sectors[i]);
1660 dst->s.ticks[i] = cpu_to_le64(src->s.ticks[i]);
1663 dst->s.io_ticks = cpu_to_le64(src->s.io_ticks);
1664 dst->s.time_in_queue = cpu_to_le64(src->s.time_in_queue);
1665 dst->s.msec = cpu_to_le64(src->s.msec);
1668 void fio_server_send_du(void)
1670 struct disk_util *du;
1671 struct flist_head *entry;
1672 struct cmd_du_pdu pdu;
1674 dprint(FD_NET, "server: sending disk_util %d\n", !flist_empty(&disk_list));
1676 memset(&pdu, 0, sizeof(pdu));
1678 flist_for_each(entry, &disk_list) {
1679 du = flist_entry(entry, struct disk_util, list);
1681 convert_dus(&pdu.dus, &du->dus);
1682 convert_agg(&pdu.agg, &du->agg);
1684 fio_net_queue_cmd(FIO_NET_CMD_DU, &pdu, sizeof(pdu), NULL, SK_F_COPY);
1690 static inline void __fio_net_prep_tail(z_stream *stream, void *out_pdu,
1691 struct sk_entry **last_entry,
1692 struct sk_entry *first)
1694 unsigned int this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream->avail_out;
1696 *last_entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len,
1697 NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE);
1698 flist_add_tail(&(*last_entry)->list, &first->next);
1703 * Deflates the next input given, creating as many new packets in the
1704 * linked list as necessary.
1706 static int __deflate_pdu_buffer(void *next_in, unsigned int next_sz, void **out_pdu,
1707 struct sk_entry **last_entry, z_stream *stream,
1708 struct sk_entry *first)
1712 stream->next_in = next_in;
1713 stream->avail_in = next_sz;
1715 if (! stream->avail_out) {
1717 __fio_net_prep_tail(stream, *out_pdu, last_entry, first);
1719 *out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1721 stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1722 stream->next_out = *out_pdu;
1725 ret = deflate(stream, Z_BLOCK);
1731 } while (stream->avail_in);
1736 static int __fio_append_iolog_gz_hist(struct sk_entry *first, struct io_log *log,
1737 struct io_logs *cur_log, z_stream *stream)
1739 struct sk_entry *entry;
1742 int sample_sz = log_entry_sz(log);
1744 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1745 stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1746 stream->next_out = out_pdu;
1748 for (i = 0; i < cur_log->nr_samples; i++) {
1749 struct io_sample *s;
1750 struct io_u_plat_entry *cur_plat_entry, *prev_plat_entry;
1751 unsigned int *cur_plat, *prev_plat;
1753 s = get_sample(log, cur_log, i);
1754 ret = __deflate_pdu_buffer(s, sample_sz, &out_pdu, &entry, stream, first);
1758 /* Do the subtraction on server side so that client doesn't have to
1759 * reconstruct our linked list from packets.
1761 cur_plat_entry = s->data.plat_entry;
1762 prev_plat_entry = flist_first_entry(&cur_plat_entry->list, struct io_u_plat_entry, list);
1763 cur_plat = cur_plat_entry->io_u_plat;
1764 prev_plat = prev_plat_entry->io_u_plat;
1766 for (j = 0; j < FIO_IO_U_PLAT_NR; j++) {
1767 cur_plat[j] -= prev_plat[j];
1770 flist_del(&prev_plat_entry->list);
1771 free(prev_plat_entry);
1773 ret = __deflate_pdu_buffer(cur_plat_entry, sizeof(*cur_plat_entry),
1774 &out_pdu, &entry, stream, first);
1780 __fio_net_prep_tail(stream, out_pdu, &entry, first);
1785 static int __fio_append_iolog_gz(struct sk_entry *first, struct io_log *log,
1786 struct io_logs *cur_log, z_stream *stream)
1788 unsigned int this_len;
1792 if (log->log_type == IO_LOG_TYPE_HIST)
1793 return __fio_append_iolog_gz_hist(first, log, cur_log, stream);
1795 stream->next_in = (void *) cur_log->log;
1796 stream->avail_in = cur_log->nr_samples * log_entry_sz(log);
1799 struct sk_entry *entry;
1802 * Dirty - since the log is potentially huge, compress it into
1803 * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving
1804 * side defragment it.
1806 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1808 stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1809 stream->next_out = out_pdu;
1810 ret = deflate(stream, Z_BLOCK);
1811 /* may be Z_OK, or Z_STREAM_END */
1817 this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream->avail_out;
1819 entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len,
1820 NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE);
1821 flist_add_tail(&entry->list, &first->next);
1822 } while (stream->avail_in);
1827 static int fio_append_iolog_gz(struct sk_entry *first, struct io_log *log)
1832 memset(&stream, 0, sizeof(stream));
1833 stream.zalloc = Z_NULL;
1834 stream.zfree = Z_NULL;
1835 stream.opaque = Z_NULL;
1837 if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK)
1840 while (!flist_empty(&log->io_logs)) {
1841 struct io_logs *cur_log;
1843 cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
1844 flist_del_init(&cur_log->list);
1846 ret = __fio_append_iolog_gz(first, log, cur_log, &stream);
1851 ret = deflate(&stream, Z_FINISH);
1853 while (ret != Z_STREAM_END) {
1854 struct sk_entry *entry;
1855 unsigned int this_len;
1858 out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU);
1859 stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
1860 stream.next_out = out_pdu;
1862 ret = deflate(&stream, Z_FINISH);
1863 /* may be Z_OK, or Z_STREAM_END */
1869 this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream.avail_out;
1871 entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len,
1872 NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE);
1873 flist_add_tail(&entry->list, &first->next);
1874 } while (ret != Z_STREAM_END);
1876 ret = deflateEnd(&stream);
1883 static int fio_append_iolog_gz(struct sk_entry *first, struct io_log *log)
1889 static int fio_append_gz_chunks(struct sk_entry *first, struct io_log *log)
1891 struct sk_entry *entry;
1892 struct flist_head *node;
1894 pthread_mutex_lock(&log->chunk_lock);
1895 flist_for_each(node, &log->chunk_list) {
1896 struct iolog_compress *c;
1898 c = flist_entry(node, struct iolog_compress, list);
1899 entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, c->buf, c->len,
1900 NULL, SK_F_VEC | SK_F_INLINE);
1901 flist_add_tail(&entry->list, &first->next);
1903 pthread_mutex_unlock(&log->chunk_lock);
1908 static int fio_append_text_log(struct sk_entry *first, struct io_log *log)
1910 struct sk_entry *entry;
1912 while (!flist_empty(&log->io_logs)) {
1913 struct io_logs *cur_log;
1916 cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
1917 flist_del_init(&cur_log->list);
1919 size = cur_log->nr_samples * log_entry_sz(log);
1921 entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, cur_log->log, size,
1922 NULL, SK_F_VEC | SK_F_INLINE);
1923 flist_add_tail(&entry->list, &first->next);
1929 int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
1931 struct cmd_iolog_pdu pdu;
1932 struct sk_entry *first;
1933 struct flist_head *entry;
1936 pdu.nr_samples = cpu_to_le64(iolog_nr_samples(log));
1937 pdu.thread_number = cpu_to_le32(td->thread_number);
1938 pdu.log_type = cpu_to_le32(log->log_type);
1939 pdu.log_hist_coarseness = cpu_to_le32(log->hist_coarseness);
1941 if (!flist_empty(&log->chunk_list))
1942 pdu.compressed = __cpu_to_le32(STORE_COMPRESSED);
1944 pdu.compressed = __cpu_to_le32(XMIT_COMPRESSED);
1948 strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX);
1949 pdu.name[FIO_NET_NAME_MAX - 1] = '\0';
1952 * We can't do this for a pre-compressed log, but for that case,
1953 * log->nr_samples is zero anyway.
1955 flist_for_each(entry, &log->io_logs) {
1956 struct io_logs *cur_log;
1959 cur_log = flist_entry(entry, struct io_logs, list);
1961 for (i = 0; i < cur_log->nr_samples; i++) {
1962 struct io_sample *s = get_sample(log, cur_log, i);
1964 s->time = cpu_to_le64(s->time);
1965 s->data.val = cpu_to_le64(s->data.val);
1966 s->__ddir = cpu_to_le32(s->__ddir);
1967 s->bs = cpu_to_le32(s->bs);
1969 if (log->log_offset) {
1970 struct io_sample_offset *so = (void *) s;
1972 so->offset = cpu_to_le64(so->offset);
1978 * Assemble header entry first
1980 first = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, &pdu, sizeof(pdu), NULL, SK_F_VEC | SK_F_INLINE | SK_F_COPY);
1983 * Now append actual log entries. If log compression was enabled on
1984 * the job, just send out the compressed chunks directly. If we
1985 * have a plain log, compress if we can, then send. Otherwise, send
1986 * the plain text output.
1988 if (!flist_empty(&log->chunk_list))
1989 ret = fio_append_gz_chunks(first, log);
1991 ret = fio_append_iolog_gz(first, log);
1993 ret = fio_append_text_log(first, log);
1995 fio_net_queue_entry(first);
1999 void fio_server_send_add_job(struct thread_data *td)
2001 struct cmd_add_job_pdu pdu;
2003 memset(&pdu, 0, sizeof(pdu));
2004 pdu.thread_number = cpu_to_le32(td->thread_number);
2005 pdu.groupid = cpu_to_le32(td->groupid);
2006 convert_thread_options_to_net(&pdu.top, &td->o);
2008 fio_net_queue_cmd(FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL,
2012 void fio_server_send_start(struct thread_data *td)
2014 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
2016 assert(sk_out->sk != -1);
2018 fio_net_queue_cmd(FIO_NET_CMD_SERVER_START, NULL, 0, NULL, SK_F_SIMPLE);
2021 int fio_server_get_verify_state(const char *name, int threadnumber,
2024 struct thread_io_list *s;
2025 struct cmd_sendfile out;
2026 struct cmd_reply *rep;
2031 dprint(FD_NET, "server: request verify state\n");
2033 rep = smalloc(sizeof(*rep));
2037 __fio_mutex_init(&rep->lock, FIO_MUTEX_LOCKED);
2041 verify_state_gen_name((char *) out.path, sizeof(out.path), name, me,
2043 tag = (uint64_t) (uintptr_t) rep;
2044 fio_net_queue_cmd(FIO_NET_CMD_SENDFILE, &out, sizeof(out), &tag,
2048 * Wait for the backend to receive the reply
2050 if (fio_mutex_down_timeout(&rep->lock, 10000)) {
2051 log_err("fio: timed out waiting for reply\n");
2057 log_err("fio: failure on receiving state file %s: %s\n",
2058 out.path, strerror(rep->error));
2063 fio_net_queue_quit();
2068 * The format is verify_state_hdr, then thread_io_list. Verify
2069 * the header, and the thread_io_list checksum
2071 s = rep->data + sizeof(struct verify_state_hdr);
2072 if (verify_state_hdr(rep->data, s)) {
2078 * Don't need the header from now, copy just the thread_io_list
2081 rep->size -= sizeof(struct verify_state_hdr);
2082 data = malloc(rep->size);
2083 memcpy(data, s, rep->size);
2087 __fio_mutex_remove(&rep->lock);
2092 static int fio_init_server_ip(void)
2094 struct sockaddr *addr;
2101 sk = socket(AF_INET6, SOCK_STREAM, 0);
2103 sk = socket(AF_INET, SOCK_STREAM, 0);
2106 log_err("fio: socket: %s\n", strerror(errno));
2111 if (setsockopt(sk, SOL_SOCKET, SO_REUSEADDR, (void *)&opt, sizeof(opt)) < 0) {
2112 log_err("fio: setsockopt(REUSEADDR): %s\n", strerror(errno));
2118 * Not fatal if fails, so just ignore it if that happens
2120 setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
2124 const void *src = &saddr_in6.sin6_addr;
2126 addr = (struct sockaddr *) &saddr_in6;
2127 socklen = sizeof(saddr_in6);
2128 saddr_in6.sin6_family = AF_INET6;
2129 str = inet_ntop(AF_INET6, src, buf, sizeof(buf));
2131 const void *src = &saddr_in.sin_addr;
2133 addr = (struct sockaddr *) &saddr_in;
2134 socklen = sizeof(saddr_in);
2135 saddr_in.sin_family = AF_INET;
2136 str = inet_ntop(AF_INET, src, buf, sizeof(buf));
2139 if (bind(sk, addr, socklen) < 0) {
2140 log_err("fio: bind: %s\n", strerror(errno));
2141 log_info("fio: failed with IPv%c %s\n", use_ipv6 ? '6' : '4', str);
2149 static int fio_init_server_sock(void)
2151 struct sockaddr_un addr;
2156 sk = socket(AF_UNIX, SOCK_STREAM, 0);
2158 log_err("fio: socket: %s\n", strerror(errno));
2164 memset(&addr, 0, sizeof(addr));
2165 addr.sun_family = AF_UNIX;
2166 strncpy(addr.sun_path, bind_sock, sizeof(addr.sun_path) - 1);
2168 len = sizeof(addr.sun_family) + strlen(bind_sock) + 1;
2170 if (bind(sk, (struct sockaddr *) &addr, len) < 0) {
2171 log_err("fio: bind: %s\n", strerror(errno));
2180 static int fio_init_server_connection(void)
2185 dprint(FD_NET, "starting server\n");
2188 sk = fio_init_server_ip();
2190 sk = fio_init_server_sock();
2195 memset(bind_str, 0, sizeof(bind_str));
2204 src = &saddr_in6.sin6_addr;
2207 src = &saddr_in.sin_addr;
2210 p = (char *) inet_ntop(af, src, bind_str, sizeof(bind_str));
2212 sprintf(port, ",%u", fio_net_port);
2216 strncpy(bind_str, port, sizeof(bind_str) - 1);
2218 strncpy(bind_str, bind_sock, sizeof(bind_str) - 1);
2220 log_info("fio: server listening on %s\n", bind_str);
2222 if (listen(sk, 4) < 0) {
2223 log_err("fio: listen: %s\n", strerror(errno));
2231 int fio_server_parse_host(const char *host, int ipv6, struct in_addr *inp,
2232 struct in6_addr *inp6)
2238 ret = inet_pton(AF_INET6, host, inp6);
2240 ret = inet_pton(AF_INET, host, inp);
2243 struct addrinfo hints, *res;
2245 memset(&hints, 0, sizeof(hints));
2246 hints.ai_family = ipv6 ? AF_INET6 : AF_INET;
2247 hints.ai_socktype = SOCK_STREAM;
2249 ret = getaddrinfo(host, NULL, &hints, &res);
2251 log_err("fio: failed to resolve <%s> (%s)\n", host,
2257 memcpy(inp6, &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr, sizeof(*inp6));
2259 memcpy(inp, &((struct sockaddr_in *) res->ai_addr)->sin_addr, sizeof(*inp));
2269 * Parse a host/ip/port string. Reads from 'str'.
2274 * *ptr is the host, *port is the port, inp is the destination.
2276 * *ptr is the host, *port is the port, inp6 is the dest, and *ipv6 is 1.
2277 * For local domain sockets:
2278 * *ptr is the filename, *is_sock is 1.
2280 int fio_server_parse_string(const char *str, char **ptr, bool *is_sock,
2281 int *port, struct in_addr *inp,
2282 struct in6_addr *inp6, int *ipv6)
2284 const char *host = str;
2290 *port = fio_net_port;
2293 if (!strncmp(str, "sock:", 5)) {
2294 *ptr = strdup(str + 5);
2301 * Is it ip:<ip or host>:port
2303 if (!strncmp(host, "ip:", 3))
2305 else if (!strncmp(host, "ip4:", 4))
2307 else if (!strncmp(host, "ip6:", 4)) {
2310 } else if (host[0] == ':') {
2311 /* String is :port */
2314 if (!lport || lport > 65535) {
2315 log_err("fio: bad server port %u\n", lport);
2318 /* no hostname given, we are done */
2324 * If no port seen yet, check if there's a last ',' at the end
2327 portp = strchr(host, ',');
2331 lport = atoi(portp);
2332 if (!lport || lport > 65535) {
2333 log_err("fio: bad server port %u\n", lport);
2345 *ptr = strdup(host);
2347 if (fio_server_parse_host(*ptr, *ipv6, inp, inp6)) {
2354 *port = fio_net_port;
2360 * Server arg should be one of:
2362 * sock:/path/to/socket
2366 * Where sock uses unix domain sockets, and ip binds the server to
2367 * a specific interface. If no arguments are given to the server, it
2368 * uses IP and binds to 0.0.0.0.
2371 static int fio_handle_server_arg(void)
2373 int port = fio_net_port;
2377 saddr_in.sin_addr.s_addr = htonl(INADDR_ANY);
2379 if (!fio_server_arg)
2382 ret = fio_server_parse_string(fio_server_arg, &bind_sock, &is_sock,
2383 &port, &saddr_in.sin_addr,
2384 &saddr_in6.sin6_addr, &use_ipv6);
2386 if (!is_sock && bind_sock) {
2392 fio_net_port = port;
2393 saddr_in.sin_port = htons(port);
2394 saddr_in6.sin6_port = htons(port);
2398 static void sig_int(int sig)
2404 static void set_sig_handlers(void)
2406 struct sigaction act;
2408 memset(&act, 0, sizeof(act));
2409 act.sa_handler = sig_int;
2410 act.sa_flags = SA_RESTART;
2411 sigaction(SIGINT, &act, NULL);
2414 void fio_server_destroy_sk_key(void)
2416 pthread_key_delete(sk_out_key);
2419 int fio_server_create_sk_key(void)
2421 if (pthread_key_create(&sk_out_key, NULL)) {
2422 log_err("fio: can't create sk_out backend key\n");
2426 pthread_setspecific(sk_out_key, NULL);
2430 static int fio_server(void)
2434 dprint(FD_NET, "starting server\n");
2436 if (fio_handle_server_arg())
2439 sk = fio_init_server_connection();
2445 ret = accept_loop(sk);
2449 if (fio_server_arg) {
2450 free(fio_server_arg);
2451 fio_server_arg = NULL;
2459 void fio_server_got_signal(int signal)
2461 struct sk_out *sk_out = pthread_getspecific(sk_out_key);
2465 if (signal == SIGPIPE)
2468 log_info("\nfio: terminating on signal %d\n", signal);
2473 static int check_existing_pidfile(const char *pidfile)
2480 if (stat(pidfile, &sb))
2483 f = fopen(pidfile, "r");
2487 if (fread(buf, sb.st_size, 1, f) <= 0) {
2494 if (kill(pid, SIGCONT) < 0)
2495 return errno != ESRCH;
2500 static int write_pid(pid_t pid, const char *pidfile)
2504 fpid = fopen(pidfile, "w");
2506 log_err("fio: failed opening pid file %s\n", pidfile);
2510 fprintf(fpid, "%u\n", (unsigned int) pid);
2516 * If pidfile is specified, background us.
2518 int fio_start_server(char *pidfile)
2525 WSAStartup(MAKEWORD(2, 2), &wsd);
2529 return fio_server();
2531 if (check_existing_pidfile(pidfile)) {
2532 log_err("fio: pidfile %s exists and server appears alive\n",
2540 log_err("fio: failed server fork: %s\n", strerror(errno));
2544 ret = write_pid(pid, pidfile);
2550 openlog("fio", LOG_NDELAY|LOG_NOWAIT|LOG_PID, LOG_USER);
2552 close(STDIN_FILENO);
2553 close(STDOUT_FILENO);
2554 close(STDERR_FILENO);
2566 void fio_server_set_arg(const char *arg)
2568 fio_server_arg = strdup(arg);