allow --client to work with shared filesystem
[fio.git] / server.c
index 32c1d7a4b389bd3c3cc50ba4a989dd7fc1a8d50c..7a9b0a447d306b0e1be0928d2635c4e19a14b96a 100644 (file)
--- a/server.c
+++ b/server.c
@@ -4,24 +4,28 @@
 #include <unistd.h>
 #include <limits.h>
 #include <errno.h>
-#include <fcntl.h>
 #include <sys/poll.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <sys/socket.h>
 #include <sys/stat.h>
 #include <sys/un.h>
+#include <sys/uio.h>
 #include <netinet/in.h>
 #include <arpa/inet.h>
 #include <netdb.h>
 #include <syslog.h>
 #include <signal.h>
+#ifdef CONFIG_ZLIB
 #include <zlib.h>
+#endif
 
 #include "fio.h"
 #include "server.h"
 #include "crc/crc16.h"
 #include "lib/ieee754.h"
+#include "verify.h"
+#include "smalloc.h"
 
 int fio_net_port = FIO_NET_PORT;
 
@@ -33,6 +37,13 @@ static char *bind_sock;
 static struct sockaddr_in saddr_in;
 static struct sockaddr_in6 saddr_in6;
 static int use_ipv6;
+#ifdef CONFIG_ZLIB
+static unsigned int has_zlib = 1;
+#else
+static unsigned int has_zlib = 0;
+#endif
+static unsigned int use_zlib;
+static char me[128];
 
 struct fio_fork_item {
        struct flist_head list;
@@ -42,11 +53,12 @@ struct fio_fork_item {
        pid_t pid;
 };
 
-/* Created on fork on new connection */
-static FLIST_HEAD(conn_list);
-
-/* Created on job fork from connection */
-static FLIST_HEAD(job_list);
+struct cmd_reply {
+       struct fio_mutex lock;
+       void *data;
+       size_t size;
+       int error;
+};
 
 static const char *fio_server_ops[FIO_NET_CMD_NR] = {
        "",
@@ -65,8 +77,12 @@ static const char *fio_server_ops[FIO_NET_CMD_NR] = {
        "DISK_UTIL",
        "SERVER_START",
        "ADD_JOB",
-       "CMD_RUN"
-       "CMD_IOLOG",
+       "RUN",
+       "IOLOG",
+       "UPDATE_JOB",
+       "LOAD_FILE",
+       "VTRIGGER",
+       "SENDFILE",
 };
 
 const char *fio_server_op(unsigned int op)
@@ -206,7 +222,7 @@ static int verify_convert_cmd(struct fio_net_cmd *cmd)
  */
 struct fio_net_cmd *fio_net_recv_cmd(int sk)
 {
-       struct fio_net_cmd cmd, *cmdret = NULL;
+       struct fio_net_cmd cmd, *tmp, *cmdret = NULL;
        size_t cmd_size = 0, pdu_offset = 0;
        uint16_t crc;
        int ret, first = 1;
@@ -229,7 +245,19 @@ struct fio_net_cmd *fio_net_recv_cmd(int sk)
                } else
                        cmd_size += cmd.pdu_len;
 
-               cmdret = realloc(cmdret, cmd_size);
+               if (cmd_size / 1024 > FIO_SERVER_MAX_CMD_MB * 1024) {
+                       log_err("fio: cmd+pdu too large (%llu)\n", (unsigned long long) cmd_size);
+                       ret = 1;
+                       break;
+               }
+
+               tmp = realloc(cmdret, cmd_size);
+               if (!tmp) {
+                       log_err("fio: server failed allocating cmd\n");
+                       ret = 1;
+                       break;
+               }
+               cmdret = tmp;
 
                if (first)
                        memcpy(cmdret, &cmd, sizeof(cmd));
@@ -271,14 +299,14 @@ struct fio_net_cmd *fio_net_recv_cmd(int sk)
                /* zero-terminate text input */
                if (cmdret->pdu_len) {
                        if (cmdret->opcode == FIO_NET_CMD_TEXT) {
-                               struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmdret->payload;
-                               char *buf = (char *) pdu->buf;
+                               struct cmd_text_pdu *__pdu = (struct cmd_text_pdu *) cmdret->payload;
+                               char *buf = (char *) __pdu->buf;
 
-                               buf[pdu->buf_len] = '\0';
+                               buf[__pdu->buf_len] = '\0';
                        } else if (cmdret->opcode == FIO_NET_CMD_JOB) {
-                               struct cmd_job_pdu *pdu = (struct cmd_job_pdu *) cmdret->payload;
-                               char *buf = (char *) pdu->buf;
-                               int len = le32_to_cpu(pdu->buf_len);
+                               struct cmd_job_pdu *__pdu = (struct cmd_job_pdu *) cmdret->payload;
+                               char *buf = (char *) __pdu->buf;
+                               int len = le32_to_cpu(__pdu->buf_len);
 
                                buf[len] = '\0';
                        }
@@ -293,8 +321,9 @@ struct fio_net_cmd *fio_net_recv_cmd(int sk)
 
 static void add_reply(uint64_t tag, struct flist_head *list)
 {
-       struct fio_net_cmd_reply *reply = (struct fio_net_cmd_reply *) tag;
+       struct fio_net_cmd_reply *reply;
 
+       reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
        flist_add_tail(&reply->list, list);
 }
 
@@ -304,7 +333,7 @@ static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
 
        reply = calloc(1, sizeof(*reply));
        INIT_FLIST_HEAD(&reply->list);
-       gettimeofday(&reply->tv, NULL);
+       fio_gettime(&reply->tv, NULL);
        reply->saved_tag = tag;
        reply->opcode = opcode;
 
@@ -313,8 +342,9 @@ static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
 
 static void free_reply(uint64_t tag)
 {
-       struct fio_net_cmd_reply *reply = (struct fio_net_cmd_reply *) tag;
+       struct fio_net_cmd_reply *reply;
 
+       reply = (struct fio_net_cmd_reply *) (uintptr_t) tag;
        free(reply);
 }
 
@@ -460,16 +490,16 @@ static void fio_server_add_fork_item(pid_t pid, struct flist_head *list)
        flist_add_tail(&ffi->list, list);
 }
 
-static void fio_server_add_conn_pid(pid_t pid)
+static void fio_server_add_conn_pid(struct flist_head *conn_list, pid_t pid)
 {
-       dprint(FD_NET, "server: forked off connection job (pid=%u)\n", pid);
-       fio_server_add_fork_item(pid, &conn_list);
+       dprint(FD_NET, "server: forked off connection job (pid=%u)\n", (int) pid);
+       fio_server_add_fork_item(pid, conn_list);
 }
 
-static void fio_server_add_job_pid(pid_t pid)
+static void fio_server_add_job_pid(struct flist_head *job_list, pid_t pid)
 {
-       dprint(FD_NET, "server: forked off job job (pid=%u)\n", pid);
-       fio_server_add_fork_item(pid, &job_list);
+       dprint(FD_NET, "server: forked off job job (pid=%u)\n", (int) pid);
+       fio_server_add_fork_item(pid, job_list);
 }
 
 static void fio_server_check_fork_item(struct fio_fork_item *ffi)
@@ -479,7 +509,7 @@ static void fio_server_check_fork_item(struct fio_fork_item *ffi)
        ret = waitpid(ffi->pid, &status, WNOHANG);
        if (ret < 0) {
                if (errno == ECHILD) {
-                       log_err("fio: connection pid %u disappeared\n", ffi->pid);
+                       log_err("fio: connection pid %u disappeared\n", (int) ffi->pid);
                        ffi->exited = 1;
                } else
                        log_err("fio: waitpid: %s\n", strerror(errno));
@@ -498,7 +528,7 @@ static void fio_server_check_fork_item(struct fio_fork_item *ffi)
 
 static void fio_server_fork_item_done(struct fio_fork_item *ffi)
 {
-       dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", ffi->pid, ffi->signal, ffi->exitval);
+       dprint(FD_NET, "pid %u exited, sig=%u, exitval=%d\n", (int) ffi->pid, ffi->signal, ffi->exitval);
 
        /*
         * Fold STOP and QUIT...
@@ -524,26 +554,49 @@ static void fio_server_check_fork_items(struct flist_head *list)
        }
 }
 
-static void fio_server_check_jobs(void)
+static void fio_server_check_jobs(struct flist_head *job_list)
+{
+       fio_server_check_fork_items(job_list);
+}
+
+static void fio_server_check_conns(struct flist_head *conn_list)
 {
-       fio_server_check_fork_items(&job_list);
+       fio_server_check_fork_items(conn_list);
 }
 
-static void fio_server_check_conns(void)
+static int handle_load_file_cmd(struct fio_net_cmd *cmd)
 {
-       fio_server_check_fork_items(&conn_list);
+       struct cmd_load_file_pdu *pdu = (struct cmd_load_file_pdu *) cmd->payload;
+       void *file_name = pdu->file;
+       struct cmd_start_pdu spdu;
+
+       dprint(FD_NET, "server: loading local file %s\n", (char *) file_name);
+
+       pdu->name_len = le16_to_cpu(pdu->name_len);
+       pdu->client_type = le16_to_cpu(pdu->client_type);
+
+       if (parse_jobs_ini(file_name, 0, 0, pdu->client_type)) {
+               fio_net_send_quit(server_fd);
+               return -1;
+       }
+
+       spdu.jobs = cpu_to_le32(thread_number);
+       spdu.stat_outputs = cpu_to_le32(stat_number);
+       fio_net_send_cmd(server_fd, FIO_NET_CMD_START, &spdu, sizeof(spdu), NULL, NULL);
+       return 0;
 }
 
-static int handle_run_cmd(struct fio_net_cmd *cmd)
+static int handle_run_cmd(struct flist_head *job_list, struct fio_net_cmd *cmd)
 {
        pid_t pid;
        int ret;
 
+       fio_time_init();
        set_genesis_time();
 
        pid = fork();
        if (pid) {
-               fio_server_add_job_pid(pid);
+               fio_server_add_job_pid(job_list, pid);
                return 0;
        }
 
@@ -614,11 +667,14 @@ static int handle_jobline_cmd(struct fio_net_cmd *cmd)
 
 static int handle_probe_cmd(struct fio_net_cmd *cmd)
 {
-       struct cmd_probe_pdu probe;
+       struct cmd_client_probe_pdu *pdu = (struct cmd_client_probe_pdu *) cmd->payload;
+       struct cmd_probe_reply_pdu probe;
        uint64_t tag = cmd->tag;
 
        dprint(FD_NET, "server: sending probe reply\n");
 
+       strcpy(me, (char *) pdu->server);
+
        memset(&probe, 0, sizeof(probe));
        gethostname((char *) probe.hostname, sizeof(probe.hostname));
 #ifdef CONFIG_BIG_ENDIAN
@@ -630,7 +686,17 @@ static int handle_probe_cmd(struct fio_net_cmd *cmd)
        probe.arch      = FIO_ARCH;
        probe.bpp       = sizeof(void *);
        probe.cpus      = __cpu_to_le32(cpus_online());
-       probe.flags     = 0;
+
+       /*
+        * If the client supports compression and we do too, then enable it
+        */
+       if (has_zlib && le64_to_cpu(pdu->flags) & FIO_PROBE_FLAG_ZLIB) {
+               probe.flags = __cpu_to_le64(FIO_PROBE_FLAG_ZLIB);
+               use_zlib = 1;
+       } else {
+               probe.flags = 0;
+               use_zlib = 0;
+       }
 
        return fio_net_send_cmd(server_fd, FIO_NET_CMD_PROBE, &probe, sizeof(probe), &tag, NULL);
 }
@@ -638,27 +704,20 @@ static int handle_probe_cmd(struct fio_net_cmd *cmd)
 static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
 {
        struct jobs_eta *je;
-       size_t size;
        uint64_t tag = cmd->tag;
+       size_t size;
        int i;
 
-       if (!thread_number)
+       je = get_jobs_eta(1, &size);
+       if (!je)
                return 0;
 
-       size = sizeof(*je) + thread_number * sizeof(char) + 1;
-       je = malloc(size);
-       memset(je, 0, size);
-
-       if (!calc_thread_status(je, 1)) {
-               free(je);
-               return 0;
-       }
-
        dprint(FD_NET, "server sending status\n");
 
        je->nr_running          = cpu_to_le32(je->nr_running);
        je->nr_ramp             = cpu_to_le32(je->nr_ramp);
        je->nr_pending          = cpu_to_le32(je->nr_pending);
+       je->nr_setting_up       = cpu_to_le32(je->nr_setting_up);
        je->files_open          = cpu_to_le32(je->files_open);
 
        for (i = 0; i < DDIR_RWDIR_CNT; i++) {
@@ -666,12 +725,15 @@ static int handle_send_eta_cmd(struct fio_net_cmd *cmd)
                je->t_rate[i]   = cpu_to_le32(je->t_rate[i]);
                je->m_iops[i]   = cpu_to_le32(je->m_iops[i]);
                je->t_iops[i]   = cpu_to_le32(je->t_iops[i]);
+               je->rate[i]     = cpu_to_le32(je->rate[i]);
+               je->iops[i]     = cpu_to_le32(je->iops[i]);
        }
 
        je->elapsed_sec         = cpu_to_le64(je->elapsed_sec);
        je->eta_sec             = cpu_to_le64(je->eta_sec);
        je->nr_threads          = cpu_to_le32(je->nr_threads);
        je->is_pow2             = cpu_to_le32(je->is_pow2);
+       je->unit_base           = cpu_to_le32(je->unit_base);
 
        fio_net_send_cmd(server_fd, FIO_NET_CMD_ETA, je, size, &tag, NULL);
        free(je);
@@ -708,7 +770,32 @@ static int handle_update_job_cmd(struct fio_net_cmd *cmd)
        return 0;
 }
 
-static int handle_command(struct fio_net_cmd *cmd)
+static int handle_trigger_cmd(struct fio_net_cmd *cmd)
+{
+       struct cmd_vtrigger_pdu *pdu = (struct cmd_vtrigger_pdu *) cmd->payload;
+       char *buf = (char *) pdu->cmd;
+       struct all_io_list *rep;
+       size_t sz;
+
+       pdu->len = le16_to_cpu(pdu->len);
+       buf[pdu->len] = '\0';
+
+       rep = get_all_io_list(IO_LIST_ALL, &sz);
+       if (!rep) {
+               struct all_io_list state;
+
+               state.threads = cpu_to_le64((uint64_t) 0);
+               fio_net_send_cmd(server_fd, FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, NULL);
+       } else {
+               fio_net_send_cmd(server_fd, FIO_NET_CMD_VTRIGGER, rep, sz, NULL, NULL);
+               free(rep);
+       }
+
+       exec_trigger(buf);
+       return 0;
+}
+
+static int handle_command(struct flist_head *job_list, struct fio_net_cmd *cmd)
 {
        int ret;
 
@@ -723,6 +810,9 @@ static int handle_command(struct fio_net_cmd *cmd)
        case FIO_NET_CMD_EXIT:
                exit_backend = 1;
                return -1;
+       case FIO_NET_CMD_LOAD_FILE:
+               ret = handle_load_file_cmd(cmd);
+               break;
        case FIO_NET_CMD_JOB:
                ret = handle_job_cmd(cmd);
                break;
@@ -736,11 +826,40 @@ static int handle_command(struct fio_net_cmd *cmd)
                ret = handle_send_eta_cmd(cmd);
                break;
        case FIO_NET_CMD_RUN:
-               ret = handle_run_cmd(cmd);
+               ret = handle_run_cmd(job_list, cmd);
                break;
        case FIO_NET_CMD_UPDATE_JOB:
                ret = handle_update_job_cmd(cmd);
                break;
+       case FIO_NET_CMD_VTRIGGER:
+               ret = handle_trigger_cmd(cmd);
+               break;
+       case FIO_NET_CMD_SENDFILE: {
+               struct cmd_sendfile_reply *in;
+               struct cmd_reply *rep;
+
+               rep = (struct cmd_reply *) (uintptr_t) cmd->tag;
+
+               in = (struct cmd_sendfile_reply *) cmd->payload;
+               in->size = le32_to_cpu(in->size);
+               in->error = le32_to_cpu(in->error);
+               if (in->error) {
+                       ret = 1;
+                       rep->error = in->error;
+               } else {
+                       ret = 0;
+                       rep->data = smalloc(in->size);
+                       if (!rep->data) {
+                               ret = 1;
+                               rep->error = ENOMEM;
+                       } else {
+                               rep->size = in->size;
+                               memcpy(rep->data, in->data, in->size);
+                       }
+               }
+               fio_mutex_up(&rep->lock);
+               break;
+               }
        default:
                log_err("fio: unknown opcode: %s\n", fio_server_op(cmd->opcode));
                ret = 1;
@@ -752,10 +871,10 @@ static int handle_command(struct fio_net_cmd *cmd)
 static int handle_connection(int sk)
 {
        struct fio_net_cmd *cmd = NULL;
+       FLIST_HEAD(job_list);
        int ret = 0;
 
        reset_fio_state();
-       INIT_FLIST_HEAD(&job_list);
        server_fd = sk;
 
        /* read forever */
@@ -779,7 +898,7 @@ static int handle_connection(int sk)
                                log_err("fio: poll: %s\n", strerror(errno));
                                break;
                        } else if (!ret) {
-                               fio_server_check_jobs();
+                               fio_server_check_jobs(&job_list);
                                continue;
                        }
 
@@ -791,7 +910,7 @@ static int handle_connection(int sk)
                        }
                } while (!exit_backend);
 
-               fio_server_check_jobs();
+               fio_server_check_jobs(&job_list);
 
                if (ret < 0)
                        break;
@@ -802,7 +921,7 @@ static int handle_connection(int sk)
                        break;
                }
 
-               ret = handle_command(cmd);
+               ret = handle_command(&job_list, cmd);
                if (ret)
                        break;
 
@@ -820,17 +939,19 @@ static int handle_connection(int sk)
 static int accept_loop(int listen_sk)
 {
        struct sockaddr_in addr;
-       socklen_t len = sizeof(addr);
+       struct sockaddr_in6 addr6;
+       socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr);
        struct pollfd pfd;
-       int ret = 0, sk, flags, exitval = 0;
+       int ret = 0, sk, exitval = 0;
+       FLIST_HEAD(conn_list);
 
        dprint(FD_NET, "server enter accept loop\n");
 
-       flags = fcntl(listen_sk, F_GETFL);
-       flags |= O_NONBLOCK;
-       fcntl(listen_sk, F_SETFL, flags);
+       fio_set_fd_nonblocking(listen_sk, "server");
 
        while (!exit_backend) {
+               const char *from;
+               char buf[64];
                pid_t pid;
 
                pfd.fd = listen_sk;
@@ -848,7 +969,7 @@ static int accept_loop(int listen_sk)
                                log_err("fio: poll: %s\n", strerror(errno));
                                break;
                        } else if (!ret) {
-                               fio_server_check_conns();
+                               fio_server_check_conns(&conn_list);
                                continue;
                        }
 
@@ -856,27 +977,37 @@ static int accept_loop(int listen_sk)
                                break;
                } while (!exit_backend);
 
-               fio_server_check_conns();
+               fio_server_check_conns(&conn_list);
 
                if (exit_backend || ret < 0)
                        break;
 
-               sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
+               if (use_ipv6)
+                       sk = accept(listen_sk, (struct sockaddr *) &addr6, &len);
+               else
+                       sk = accept(listen_sk, (struct sockaddr *) &addr, &len);
+
                if (sk < 0) {
                        log_err("fio: accept: %s\n", strerror(errno));
                        return -1;
                }
 
-               dprint(FD_NET, "server: connect from %s\n", inet_ntoa(addr.sin_addr));
+               if (use_ipv6)
+                       from = inet_ntop(AF_INET6, (struct sockaddr *) &addr6.sin6_addr, buf, sizeof(buf));
+               else
+                       from = inet_ntop(AF_INET, (struct sockaddr *) &addr.sin_addr, buf, sizeof(buf));
+
+               dprint(FD_NET, "server: connect from %s\n", from);
 
                pid = fork();
                if (pid) {
                        close(sk);
-                       fio_server_add_conn_pid(pid);
+                       fio_server_add_conn_pid(&conn_list, pid);
                        continue;
                }
 
                /* exits */
+               strncpy(client_sockaddr_str, from, INET6_ADDRSTRLEN);
                handle_connection(sk);
        }
 
@@ -918,8 +1049,8 @@ static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
        /*
         * Encode to IEEE 754 for network transfer
         */
-       dst->mean.u.i   = __cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
-       dst->S.u.i      = __cpu_to_le64(fio_double_to_uint64(src->S.u.f));
+       dst->mean.u.i   = cpu_to_le64(fio_double_to_uint64(src->mean.u.f));
+       dst->S.u.i      = cpu_to_le64(fio_double_to_uint64(src->S.u.f));
 }
 
 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
@@ -954,9 +1085,9 @@ void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
 
        memset(&p, 0, sizeof(p));
 
-       strcpy(p.ts.name, ts->name);
-       strcpy(p.ts.verror, ts->verror);
-       strcpy(p.ts.description, ts->description);
+       strncpy(p.ts.name, ts->name, FIO_JOBNAME_SIZE - 1);
+       strncpy(p.ts.verror, ts->verror, FIO_VERROR_SIZE - 1);
+       strncpy(p.ts.description, ts->description, FIO_JOBDESC_SIZE - 1);
 
        p.ts.error              = cpu_to_le32(ts->error);
        p.ts.thread_number      = cpu_to_le32(ts->thread_number);
@@ -978,12 +1109,13 @@ void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
        p.ts.minf               = cpu_to_le64(ts->minf);
        p.ts.majf               = cpu_to_le64(ts->majf);
        p.ts.clat_percentiles   = cpu_to_le64(ts->clat_percentiles);
+       p.ts.percentile_precision = cpu_to_le64(ts->percentile_precision);
 
        for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
                fio_fp64_t *src = &ts->percentile_list[i];
                fio_fp64_t *dst = &p.ts.percentile_list[i];
 
-               dst->u.i = __cpu_to_le64(fio_double_to_uint64(src->u.f));
+               dst->u.i = cpu_to_le64(fio_double_to_uint64(src->u.f));
        }
 
        for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
@@ -1004,6 +1136,7 @@ void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
        for (i = 0; i < DDIR_RWDIR_CNT; i++) {
                p.ts.total_io_u[i]      = cpu_to_le64(ts->total_io_u[i]);
                p.ts.short_io_u[i]      = cpu_to_le64(ts->short_io_u[i]);
+               p.ts.drop_io_u[i]       = cpu_to_le64(ts->drop_io_u[i]);
        }
 
        p.ts.total_submit       = cpu_to_le64(ts->total_submit);
@@ -1021,6 +1154,15 @@ void fio_server_send_ts(struct thread_stat *ts, struct group_run_stats *rs)
        p.ts.kb_base            = cpu_to_le32(ts->kb_base);
        p.ts.unit_base          = cpu_to_le32(ts->unit_base);
 
+       p.ts.latency_depth      = cpu_to_le32(ts->latency_depth);
+       p.ts.latency_target     = cpu_to_le64(ts->latency_target);
+       p.ts.latency_window     = cpu_to_le64(ts->latency_window);
+       p.ts.latency_percentile.u.i = cpu_to_le64(fio_double_to_uint64(ts->latency_percentile.u.f));
+
+       p.ts.nr_block_infos     = le64_to_cpu(ts->nr_block_infos);
+       for (i = 0; i < p.ts.nr_block_infos; i++)
+               p.ts.block_infos[i] = le32_to_cpu(ts->block_infos[i]);
+
        convert_gs(&p.rs, rs);
 
        fio_net_send_cmd(server_fd, FIO_NET_CMD_TS, &p, sizeof(p), NULL, NULL);
@@ -1041,34 +1183,35 @@ static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src)
        int i;
 
        for (i = 0; i < 2; i++) {
-               dst->ios[i]     = cpu_to_le32(src->ios[i]);
-               dst->merges[i]  = cpu_to_le32(src->merges[i]);
+               dst->ios[i]     = cpu_to_le64(src->ios[i]);
+               dst->merges[i]  = cpu_to_le64(src->merges[i]);
                dst->sectors[i] = cpu_to_le64(src->sectors[i]);
-               dst->ticks[i]   = cpu_to_le32(src->ticks[i]);
+               dst->ticks[i]   = cpu_to_le64(src->ticks[i]);
        }
 
-       dst->io_ticks           = cpu_to_le32(src->io_ticks);
-       dst->time_in_queue      = cpu_to_le32(src->time_in_queue);
+       dst->io_ticks           = cpu_to_le64(src->io_ticks);
+       dst->time_in_queue      = cpu_to_le64(src->time_in_queue);
        dst->slavecount         = cpu_to_le32(src->slavecount);
-       dst->max_util.u.i       = __cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
+       dst->max_util.u.i       = cpu_to_le64(fio_double_to_uint64(src->max_util.u.f));
 }
 
 static void convert_dus(struct disk_util_stat *dst, struct disk_util_stat *src)
 {
        int i;
 
-       strcpy((char *) dst->name, (char *) src->name);
+       dst->name[FIO_DU_NAME_SZ - 1] = '\0';
+       strncpy((char *) dst->name, (char *) src->name, FIO_DU_NAME_SZ - 1);
 
        for (i = 0; i < 2; i++) {
-               dst->ios[i]     = cpu_to_le32(src->ios[i]);
-               dst->merges[i]  = cpu_to_le32(src->merges[i]);
-               dst->sectors[i] = cpu_to_le64(src->sectors[i]);
-               dst->ticks[i]   = cpu_to_le32(src->ticks[i]);
+               dst->s.ios[i]           = cpu_to_le64(src->s.ios[i]);
+               dst->s.merges[i]        = cpu_to_le64(src->s.merges[i]);
+               dst->s.sectors[i]       = cpu_to_le64(src->s.sectors[i]);
+               dst->s.ticks[i]         = cpu_to_le64(src->s.ticks[i]);
        }
 
-       dst->io_ticks           = cpu_to_le32(src->io_ticks);
-       dst->time_in_queue      = cpu_to_le32(src->time_in_queue);
-       dst->msec               = cpu_to_le64(src->msec);
+       dst->s.io_ticks         = cpu_to_le64(src->s.io_ticks);
+       dst->s.time_in_queue    = cpu_to_le64(src->s.time_in_queue);
+       dst->s.msec             = cpu_to_le64(src->s.msec);
 }
 
 void fio_server_send_du(void)
@@ -1100,7 +1243,7 @@ static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
        struct fio_net_cmd cmd;
        struct iovec iov[2];
 
-       iov[0].iov_base = &cmd;
+       iov[0].iov_base = (void *) &cmd;
        iov[0].iov_len = sizeof(cmd);
        iov[1].iov_base = (void *) buf;
        iov[1].iov_len = size;
@@ -1112,26 +1255,12 @@ static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf,
        return fio_sendv_data(sk, iov, 2);
 }
 
-int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
+static int fio_send_iolog_gz(struct cmd_iolog_pdu *pdu, struct io_log *log)
 {
-       struct cmd_iolog_pdu pdu;
+       int ret = 0;
+#ifdef CONFIG_ZLIB
        z_stream stream;
        void *out_pdu;
-       int i, ret = 0;
-
-       pdu.thread_number = cpu_to_le32(td->thread_number);
-       pdu.nr_samples = __cpu_to_le32(log->nr_samples);
-       pdu.log_type = cpu_to_le32(log->log_type);
-       strcpy((char *) pdu.name, name);
-
-       for (i = 0; i < log->nr_samples; i++) {
-               struct io_sample *s = &log->log[i];
-
-               s->time = cpu_to_le64(s->time);
-               s->val  = cpu_to_le64(s->val);
-               s->ddir = cpu_to_le32(s->ddir);
-               s->bs   = cpu_to_le32(s->bs);
-       }
 
        /*
         * Dirty - since the log is potentially huge, compress it into
@@ -1149,20 +1278,11 @@ int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
                goto err;
        }
 
-       /*
-        * Send header first, it's not compressed.
-        */
-       ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, &pdu,
-                                       sizeof(pdu), 0, FIO_NET_CMD_F_MORE);
-       if (ret)
-               goto err_zlib;
-
        stream.next_in = (void *) log->log;
-       stream.avail_in = log->nr_samples * sizeof(struct io_sample);
+       stream.avail_in = log->nr_samples * log_entry_sz(log);
 
        do {
                unsigned int this_len, flags = 0;
-               int ret;
 
                stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU;
                stream.next_out = out_pdu;
@@ -1186,9 +1306,56 @@ err_zlib:
        deflateEnd(&stream);
 err:
        free(out_pdu);
+#endif
        return ret;
 }
 
+int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name)
+{
+       struct cmd_iolog_pdu pdu;
+       int i, ret = 0;
+
+       pdu.nr_samples = cpu_to_le64(log->nr_samples);
+       pdu.thread_number = cpu_to_le32(td->thread_number);
+       pdu.log_type = cpu_to_le32(log->log_type);
+       pdu.compressed = cpu_to_le32(use_zlib);
+
+       strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX);
+       pdu.name[FIO_NET_NAME_MAX - 1] = '\0';
+
+       for (i = 0; i < log->nr_samples; i++) {
+               struct io_sample *s = get_sample(log, i);
+
+               s->time         = cpu_to_le64(s->time);
+               s->val          = cpu_to_le64(s->val);
+               s->__ddir       = cpu_to_le32(s->__ddir);
+               s->bs           = cpu_to_le32(s->bs);
+
+               if (log->log_offset) {
+                       struct io_sample_offset *so = (void *) s;
+
+                       so->offset = cpu_to_le64(so->offset);
+               }
+       }
+
+       /*
+        * Send header first, it's not compressed.
+        */
+       ret = fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, &pdu,
+                                       sizeof(pdu), 0, FIO_NET_CMD_F_MORE);
+       if (ret)
+               return ret;
+
+       /*
+        * Now send actual log, compress if we can, otherwise just plain
+        */
+       if (use_zlib)
+               return fio_send_iolog_gz(&pdu, log);
+
+       return fio_send_cmd_ext_pdu(server_fd, FIO_NET_CMD_IOLOG, log->log,
+                       log->nr_samples * log_entry_sz(log), 0, 0);
+}
+
 void fio_server_send_add_job(struct thread_data *td)
 {
        struct cmd_add_job_pdu pdu;
@@ -1208,10 +1375,78 @@ void fio_server_send_start(struct thread_data *td)
        fio_net_send_simple_cmd(server_fd, FIO_NET_CMD_SERVER_START, 0, NULL);
 }
 
+int fio_server_get_verify_state(const char *name, int threadnumber,
+                               void **datap)
+{
+       struct thread_io_list *s;
+       struct cmd_sendfile out;
+       struct cmd_reply *rep;
+       uint64_t tag;
+       void *data;
+
+       dprint(FD_NET, "server: request verify state\n");
+
+       rep = smalloc(sizeof(*rep));
+       if (!rep) {
+               log_err("fio: smalloc pool too small\n");
+               return 1;
+       }
+
+       __fio_mutex_init(&rep->lock, FIO_MUTEX_LOCKED);
+       rep->data = NULL;
+       rep->error = 0;
+
+       verify_state_gen_name((char *) out.path, sizeof(out.path), name, me,
+                               threadnumber);
+       tag = (uint64_t) (uintptr_t) rep;
+       fio_net_send_cmd(server_fd, FIO_NET_CMD_SENDFILE, &out, sizeof(out),
+                               &tag, NULL);
+
+       /*
+        * Wait for the backend to receive the reply
+        */
+       if (fio_mutex_down_timeout(&rep->lock, 10)) {
+               log_err("fio: timed out waiting for reply\n");
+               goto fail;
+       }
+
+       if (rep->error) {
+               log_err("fio: failure on receiving state file: %s\n", strerror(rep->error));
+fail:
+               *datap = NULL;
+               sfree(rep);
+               fio_net_send_quit(server_fd);
+               return 1;
+       }
+
+       /*
+        * The format is verify_state_hdr, then thread_io_list. Verify
+        * the header, and the thread_io_list checksum
+        */
+       s = rep->data + sizeof(struct verify_state_hdr);
+       if (verify_state_hdr(rep->data, s))
+               goto fail;
+
+       /*
+        * Don't need the header from now, copy just the thread_io_list
+        */
+       rep->size -= sizeof(struct verify_state_hdr);
+       data = malloc(rep->size);
+       memcpy(data, s, rep->size);
+       *datap = data;
+
+       sfree(rep->data);
+       __fio_mutex_remove(&rep->lock);
+       sfree(rep);
+       return 0;
+}
+
 static int fio_init_server_ip(void)
 {
        struct sockaddr *addr;
        socklen_t socklen;
+       char buf[80];
+       const char *str;
        int sk, opt;
 
        if (use_ipv6)
@@ -1239,17 +1474,24 @@ static int fio_init_server_ip(void)
 #endif
 
        if (use_ipv6) {
+               const void *src = &saddr_in6.sin6_addr;
+
                addr = (struct sockaddr *) &saddr_in6;
                socklen = sizeof(saddr_in6);
                saddr_in6.sin6_family = AF_INET6;
+               str = inet_ntop(AF_INET6, src, buf, sizeof(buf));
        } else {
+               const void *src = &saddr_in.sin_addr;
+
                addr = (struct sockaddr *) &saddr_in;
                socklen = sizeof(saddr_in);
                saddr_in.sin_family = AF_INET;
+               str = inet_ntop(AF_INET, src, buf, sizeof(buf));
        }
 
        if (bind(sk, addr, socklen) < 0) {
                log_err("fio: bind: %s\n", strerror(errno));
+               log_info("fio: failed with IPv%c %s\n", use_ipv6 ? '6' : '4', str);
                close(sk);
                return -1;
        }
@@ -1274,8 +1516,7 @@ static int fio_init_server_sock(void)
 
        memset(&addr, 0, sizeof(addr));
        addr.sun_family = AF_UNIX;
-       strcpy(addr.sun_path, bind_sock);
-       unlink(bind_sock);
+       strncpy(addr.sun_path, bind_sock, sizeof(addr.sun_path) - 1);
 
        len = sizeof(addr.sun_family) + strlen(bind_sock) + 1;
 
@@ -1304,6 +1545,8 @@ static int fio_init_server_connection(void)
        if (sk < 0)
                return sk;
 
+       memset(bind_str, 0, sizeof(bind_str));
+
        if (!bind_sock) {
                char *p, port[16];
                const void *src;
@@ -1323,55 +1566,53 @@ static int fio_init_server_connection(void)
                if (p)
                        strcat(p, port);
                else
-                       strcpy(bind_str, port);
+                       strncpy(bind_str, port, sizeof(bind_str) - 1);
        } else
-               strcpy(bind_str, bind_sock);
+               strncpy(bind_str, bind_sock, sizeof(bind_str) - 1);
 
        log_info("fio: server listening on %s\n", bind_str);
 
        if (listen(sk, 0) < 0) {
                log_err("fio: listen: %s\n", strerror(errno));
+               close(sk);
                return -1;
        }
 
        return sk;
 }
 
-int fio_server_parse_host(const char *host, int *ipv6, struct in_addr *inp,
+int fio_server_parse_host(const char *host, int ipv6, struct in_addr *inp,
                          struct in6_addr *inp6)
 
 {
        int ret = 0;
 
-       if (*ipv6)
+       if (ipv6)
                ret = inet_pton(AF_INET6, host, inp6);
        else
                ret = inet_pton(AF_INET, host, inp);
 
        if (ret != 1) {
-               struct hostent *hent;
+               struct addrinfo hints, *res;
 
-               hent = gethostbyname(host);
-               if (!hent) {
-                       log_err("fio: failed to resolve <%s>\n", host);
-                       return 0;
-               }
+               memset(&hints, 0, sizeof(hints));
+               hints.ai_family = ipv6 ? AF_INET6 : AF_INET;
+               hints.ai_socktype = SOCK_STREAM;
 
-               if (*ipv6) {
-                       if (hent->h_addrtype != AF_INET6) {
-                               log_info("fio: falling back to IPv4\n");
-                               *ipv6 = 0;
-                       } else
-                               memcpy(inp6, hent->h_addr_list[0], 16);
-               }
-               if (!*ipv6) {
-                       if (hent->h_addrtype != AF_INET) {
-                               log_err("fio: lookup type mismatch\n");
-                               return 0;
-                       }
-                       memcpy(inp, hent->h_addr_list[0], 4);
+               ret = getaddrinfo(host, NULL, &hints, &res);
+               if (ret) {
+                       log_err("fio: failed to resolve <%s> (%s)\n", host,
+                                       gai_strerror(ret));
+                       return 1;
                }
+
+               if (ipv6)
+                       memcpy(inp6, &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr, sizeof(*inp6));
+               else
+                       memcpy(inp, &((struct sockaddr_in *) res->ai_addr)->sin_addr, sizeof(*inp));
+
                ret = 1;
+               freeaddrinfo(res);
        }
 
        return !(ret == 1);
@@ -1424,7 +1665,7 @@ int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
                host++;
                lport = atoi(host);
                if (!lport || lport > 65535) {
-                       log_err("fio: bad server port %u\n", port);
+                       log_err("fio: bad server port %u\n", lport);
                        return 1;
                }
                /* no hostname given, we are done */
@@ -1433,7 +1674,7 @@ int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
        }
 
        /*
-        * If no port seen yet, check if there's a last ':' at the end
+        * If no port seen yet, check if there's a last ',' at the end
         */
        if (!lport) {
                portp = strchr(host, ',');
@@ -1442,7 +1683,7 @@ int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
                        portp++;
                        lport = atoi(portp);
                        if (!lport || lport > 65535) {
-                               log_err("fio: bad server port %u\n", port);
+                               log_err("fio: bad server port %u\n", lport);
                                return 1;
                        }
                }
@@ -1456,7 +1697,7 @@ int fio_server_parse_string(const char *str, char **ptr, int *is_sock,
 
        *ptr = strdup(host);
 
-       if (fio_server_parse_host(*ptr, ipv6, inp, inp6)) {
+       if (fio_server_parse_host(*ptr, *ipv6, inp, inp6)) {
                free(*ptr);
                *ptr = NULL;
                return 1;
@@ -1506,6 +1747,22 @@ out:
        return ret;
 }
 
+static void sig_int(int sig)
+{
+       if (bind_sock)
+               unlink(bind_sock);
+}
+
+static void set_sig_handlers(void)
+{
+       struct sigaction act;
+
+       memset(&act, 0, sizeof(act));
+       act.sa_handler = sig_int;
+       act.sa_flags = SA_RESTART;
+       sigaction(SIGINT, &act, NULL);
+}
+
 static int fio_server(void)
 {
        int sk, ret;
@@ -1519,6 +1776,8 @@ static int fio_server(void)
        if (sk < 0)
                return -1;
 
+       set_sig_handlers();
+
        ret = accept_loop(sk);
 
        close(sk);
@@ -1604,6 +1863,7 @@ int fio_start_server(char *pidfile)
        if (check_existing_pidfile(pidfile)) {
                log_err("fio: pidfile %s exists and server appears alive\n",
                                                                pidfile);
+               free(pidfile);
                return -1;
        }
 
@@ -1613,9 +1873,9 @@ int fio_start_server(char *pidfile)
                free(pidfile);
                return -1;
        } else if (pid) {
-               int ret = write_pid(pid, pidfile);
-
-               exit(ret);
+               ret = write_pid(pid, pidfile);
+               free(pidfile);
+               _exit(ret);
        }
 
        setsid();