client: fix double removal of client on job file open failure
[fio.git] / client.c
index af6621dc63d3e0ce7c196e1bd5daa0e747463a7d..6bc114541062231f24d9bc0215e47102e05ec868 100644 (file)
--- a/client.c
+++ b/client.c
@@ -23,6 +23,7 @@
 #include "server.h"
 #include "flist.h"
 #include "hash.h"
+#include "verify.h"
 
 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
@@ -32,6 +33,8 @@ static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
 
+static void convert_text(struct fio_net_cmd *cmd);
+
 struct client_ops fio_client_ops = {
        .text           = handle_text,
        .disk_util      = handle_du,
@@ -58,15 +61,19 @@ int sum_stat_clients;
 
 static int sum_stat_nr;
 static struct json_object *root = NULL;
+static struct json_object *job_opt_object = NULL;
 static struct json_array *clients_array = NULL;
 static struct json_array *du_array = NULL;
-static int do_output_all_clients;
+
+static int error_clients;
 
 #define FIO_CLIENT_HASH_BITS   7
 #define FIO_CLIENT_HASH_SZ     (1 << FIO_CLIENT_HASH_BITS)
 #define FIO_CLIENT_HASH_MASK   (FIO_CLIENT_HASH_SZ - 1)
 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
 
+static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
+
 static void fio_client_add_hash(struct fio_client *client)
 {
        int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
@@ -89,12 +96,49 @@ static void fio_init fio_client_hash_init(void)
                INIT_FLIST_HEAD(&client_hash[i]);
 }
 
+static int read_data(int fd, void *data, size_t size)
+{
+       ssize_t ret;
+
+       while (size) {
+               ret = read(fd, data, size);
+               if (ret < 0) {
+                       if (errno == EAGAIN || errno == EINTR)
+                               continue;
+                       break;
+               } else if (!ret)
+                       break;
+               else {
+                       data += ret;
+                       size -= ret;
+               }
+       }
+
+       if (size)
+               return EAGAIN;
+
+       return 0;
+}
+
 static void fio_client_json_init(void)
 {
-       if (output_format != FIO_OUTPUT_JSON)
+       char time_buf[32];
+       time_t time_p;
+
+       if (!(output_format & FIO_OUTPUT_JSON))
                return;
+
+       time(&time_p);
+       os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
+       time_buf[strlen(time_buf) - 1] = '\0';
+
        root = json_create_object();
        json_object_add_value_string(root, "fio version", fio_version_string);
+       json_object_add_value_int(root, "timestamp", time_p);
+       json_object_add_value_string(root, "time", time_buf);
+
+       job_opt_object = json_create_object();
+       json_object_add_value_object(root, "global options", job_opt_object);
        clients_array = json_create_array();
        json_object_add_value_array(root, "client_stats", clients_array);
        du_array = json_create_array();
@@ -103,9 +147,11 @@ static void fio_client_json_init(void)
 
 static void fio_client_json_fini(void)
 {
-       if (output_format != FIO_OUTPUT_JSON)
+       if (!(output_format & FIO_OUTPUT_JSON))
                return;
-       json_print_object(root);
+
+       log_info("\n");
+       json_print_object(root, NULL);
        log_info("\n");
        json_free_object(root);
        root = NULL;
@@ -141,23 +187,62 @@ void fio_put_client(struct fio_client *client)
                free(client->argv);
        if (client->name)
                free(client->name);
-       while (client->nr_ini_file)
-               free(client->ini_file[--client->nr_ini_file]);
-       if (client->ini_file)
-               free(client->ini_file);
+       while (client->nr_files) {
+               struct client_file *cf = &client->files[--client->nr_files];
+
+               free(cf->file);
+       }
+       if (client->files)
+               free(client->files);
+       if (client->opt_lists)
+               free(client->opt_lists);
 
        if (!client->did_stat)
-               sum_stat_clients -= client->nr_stat;
+               sum_stat_clients--;
+
+       if (client->error)
+               error_clients++;
 
        free(client);
 }
 
+static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
+{
+       if (!--eta->pending) {
+               eta_fn(&eta->eta);
+               free(eta);
+               return 0;
+       }
+
+       return 1;
+}
+
+static void fio_drain_client_text(struct fio_client *client)
+{
+       do {
+               struct fio_net_cmd *cmd;
+
+               cmd = fio_net_recv_cmd(client->fd, false);
+               if (!cmd)
+                       break;
+
+               if (cmd->opcode == FIO_NET_CMD_TEXT) {
+                       convert_text(cmd);
+                       client->ops->text(client, cmd);
+               }
+
+               free(cmd);
+       } while (1);
+}
+
 static void remove_client(struct fio_client *client)
 {
        assert(client->refs);
 
        dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 
+       fio_drain_client_text(client);
+
        if (!flist_empty(&client->list))
                flist_del_init(&client->list);
 
@@ -262,17 +347,29 @@ err:
        return NULL;
 }
 
-void fio_client_add_ini_file(void *cookie, const char *ini_file)
+int fio_client_add_ini_file(void *cookie, const char *ini_file, int remote)
 {
        struct fio_client *client = cookie;
+       struct client_file *cf;
        size_t new_size;
+       void *new_files;
+
+       if (!client)
+               return 1;
 
        dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 
-       new_size = (client->nr_ini_file + 1) * sizeof(char *);
-       client->ini_file = realloc(client->ini_file, new_size);
-       client->ini_file[client->nr_ini_file] = strdup(ini_file);
-       client->nr_ini_file++;
+       new_size = (client->nr_files + 1) * sizeof(struct client_file);
+       new_files = realloc(client->files, new_size);
+       if (!new_files)
+               return 1;
+
+       client->files = new_files;
+       cf = &client->files[client->nr_files];
+       cf->file = strdup(ini_file);
+       cf->remote = remote;
+       client->nr_files++;
+       return 0;
 }
 
 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
@@ -323,10 +420,27 @@ int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
        return 0;
 }
 
+static const char *server_name(struct fio_client *client, char *buf,
+                              size_t bufsize)
+{
+       const char *from;
+
+       if (client->ipv6)
+               from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
+       else if (client->is_sock)
+               from = "sock";
+       else
+               from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
+
+       return from;
+}
+
 static void probe_client(struct fio_client *client)
 {
        struct cmd_client_probe_pdu pdu;
+       const char *sname;
        uint64_t tag;
+       char buf[64];
 
        dprint(FD_NET, "client: send probe\n");
 
@@ -336,6 +450,10 @@ static void probe_client(struct fio_client *client)
        pdu.flags = 0;
 #endif
 
+       sname = server_name(client, buf, sizeof(buf));
+       memset(pdu.server, 0, sizeof(pdu.server));
+       strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
+
        fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
 }
 
@@ -458,11 +576,6 @@ static void sig_int(int sig)
        fio_clients_terminate();
 }
 
-static void sig_show_status(int sig)
-{
-       show_running_run_stats();
-}
-
 static void client_signal_handler(void)
 {
        struct sigaction act;
@@ -599,11 +712,34 @@ int fio_start_all_clients(void)
        return flist_empty(&client_list);
 }
 
+static int __fio_client_send_remote_ini(struct fio_client *client,
+                                       const char *filename)
+{
+       struct cmd_load_file_pdu *pdu;
+       size_t p_size;
+       int ret;
+
+       dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
+
+       p_size = sizeof(*pdu) + strlen(filename) + 1;
+       pdu = malloc(p_size);
+       memset(pdu, 0, p_size);
+       pdu->name_len = strlen(filename);
+       strcpy((char *) pdu->file, filename);
+       pdu->client_type = cpu_to_le16((uint16_t) client->type);
+
+       client->sent_job = 1;
+       ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
+       free(pdu);
+       return ret;
+}
+
 /*
  * Send file contents to server backend. We could use sendfile(), but to remain
  * more portable lets just read/write the darn thing.
  */
-static int __fio_client_send_ini(struct fio_client *client, const char *filename)
+static int __fio_client_send_local_ini(struct fio_client *client,
+                                      const char *filename)
 {
        struct cmd_job_pdu *pdu;
        size_t p_size;
@@ -617,15 +753,13 @@ static int __fio_client_send_ini(struct fio_client *client, const char *filename
 
        fd = open(filename, O_RDONLY);
        if (fd < 0) {
-               int ret = -errno;
-
+               ret = -errno;
                log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
                return ret;
        }
 
        if (fstat(fd, &sb) < 0) {
-               int ret = -errno;
-
+               ret = -errno;
                log_err("fio: job file stat: %s\n", strerror(errno));
                close(fd);
                return ret;
@@ -637,21 +771,7 @@ static int __fio_client_send_ini(struct fio_client *client, const char *filename
 
        len = sb.st_size;
        p = buf;
-       do {
-               ret = read(fd, p, len);
-               if (ret > 0) {
-                       len -= ret;
-                       if (!len)
-                               break;
-                       p += ret;
-                       continue;
-               } else if (!ret)
-                       break;
-               else if (errno == EAGAIN || errno == EINTR)
-                       continue;
-       } while (1);
-
-       if (len) {
+       if (read_data(fd, p, len)) {
                log_err("fio: failed reading job file %s\n", filename);
                close(fd);
                free(pdu);
@@ -668,37 +788,56 @@ static int __fio_client_send_ini(struct fio_client *client, const char *filename
        return ret;
 }
 
-int fio_client_send_ini(struct fio_client *client, const char *filename)
+int fio_client_send_ini(struct fio_client *client, const char *filename,
+                       int remote)
 {
        int ret;
 
-       ret = __fio_client_send_ini(client, filename);
+       if (!remote)
+               ret = __fio_client_send_local_ini(client, filename);
+       else
+               ret = __fio_client_send_remote_ini(client, filename);
+
        if (!ret)
                client->sent_job = 1;
 
        return ret;
 }
 
+static int fio_client_send_cf(struct fio_client *client,
+                             struct client_file *cf)
+{
+       return fio_client_send_ini(client, cf->file, cf->remote);
+}
+
 int fio_clients_send_ini(const char *filename)
 {
        struct fio_client *client;
        struct flist_head *entry, *tmp;
 
        flist_for_each_safe(entry, tmp, &client_list) {
+               bool failed = false;
+
                client = flist_entry(entry, struct fio_client, list);
 
-               if (client->nr_ini_file) {
+               if (client->nr_files) {
                        int i;
 
-                       for (i = 0; i < client->nr_ini_file; i++) {
-                               const char *ini = client->ini_file[i];
+                       for (i = 0; i < client->nr_files; i++) {
+                               struct client_file *cf;
 
-                               if (fio_client_send_ini(client, ini)) {
+                               cf = &client->files[i];
+
+                               if (fio_client_send_cf(client, cf)) {
+                                       failed = true;
                                        remove_client(client);
                                        break;
                                }
                        }
-               } else if (!filename || fio_client_send_ini(client, filename))
+               }
+               if (client->sent_job || failed)
+                       continue;
+               if (!filename || fio_client_send_ini(client, filename, 0))
                        remove_client(client);
        }
 
@@ -754,6 +893,7 @@ static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
        dst->minf               = le64_to_cpu(src->minf);
        dst->majf               = le64_to_cpu(src->majf);
        dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
+       dst->percentile_precision = le64_to_cpu(src->percentile_precision);
 
        for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
                fio_fp64_t *fps = &src->percentile_list[i];
@@ -780,6 +920,7 @@ static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
        for (i = 0; i < DDIR_RWDIR_CNT; i++) {
                dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
                dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
+               dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
        }
 
        dst->total_submit       = le64_to_cpu(src->total_submit);
@@ -801,6 +942,10 @@ static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
        dst->latency_target     = le64_to_cpu(src->latency_target);
        dst->latency_window     = le64_to_cpu(src->latency_window);
        dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
+
+       dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
+       for (i = 0; i < dst->nr_block_infos; i++)
+               dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
 }
 
 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
@@ -823,28 +968,34 @@ static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 }
 
 static void json_object_add_client_info(struct json_object *obj,
-struct fio_client *client)
+                                       struct fio_client *client)
 {
-       json_object_add_value_string(obj, "hostname", client->hostname);
+       const char *hostname = client->hostname ? client->hostname : "";
+
+       json_object_add_value_string(obj, "hostname", hostname);
        json_object_add_value_int(obj, "port", client->port);
 }
 
 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
+       struct flist_head *opt_list = NULL;
        struct json_object *tsobj;
 
-       tsobj = show_thread_status(&p->ts, &p->rs);
+       if (client->opt_lists && p->ts.thread_number <= client->jobs)
+               opt_list = &client->opt_lists[p->ts.thread_number - 1];
+
+       tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
        client->did_stat = 1;
        if (tsobj) {
                json_object_add_client_info(tsobj, client);
                json_array_add_value_object(clients_array, tsobj);
        }
 
-       if (!do_output_all_clients)
+       if (sum_stat_clients <= 1)
                return;
 
-       sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
+       sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
        sum_group_stats(&client_gs, &p->rs);
 
        client_ts.members++;
@@ -854,7 +1005,7 @@ static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 
        if (++sum_stat_nr == sum_stat_clients) {
                strcpy(client_ts.name, "All clients");
-               tsobj = show_thread_status(&client_ts, &client_gs);
+               tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
                if (tsobj) {
                        json_object_add_client_info(tsobj, client);
                        json_array_add_value_object(clients_array, tsobj);
@@ -866,7 +1017,41 @@ static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 
-       show_group_stats(gs);
+       if (output_format & FIO_OUTPUT_NORMAL)
+               show_group_stats(gs, NULL);
+}
+
+static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
+       struct print_option *p;
+
+       if (!job_opt_object)
+               return;
+
+       pdu->global = le16_to_cpu(pdu->global);
+       pdu->truncated = le16_to_cpu(pdu->truncated);
+       pdu->groupid = le32_to_cpu(pdu->groupid);
+
+       p = malloc(sizeof(*p));
+       p->name = strdup((char *) pdu->name);
+       if (pdu->value[0] != '\0')
+               p->value = strdup((char *) pdu->value);
+       else
+               p->value = NULL;
+
+       if (pdu->global) {
+               const char *pos = "";
+
+               if (p->value)
+                       pos = p->value;
+
+               json_object_add_value_string(job_opt_object, p->name, pos);
+       } else if (client->opt_lists) {
+               struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
+
+               flist_add_tail(&p->list, opt_list);
+       }
 }
 
 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
@@ -890,16 +1075,16 @@ static void convert_agg(struct disk_util_agg *agg)
        int i;
 
        for (i = 0; i < 2; i++) {
-               agg->ios[i]     = le32_to_cpu(agg->ios[i]);
-               agg->merges[i]  = le32_to_cpu(agg->merges[i]);
+               agg->ios[i]     = le64_to_cpu(agg->ios[i]);
+               agg->merges[i]  = le64_to_cpu(agg->merges[i]);
                agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
-               agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
+               agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
        }
 
-       agg->io_ticks           = le32_to_cpu(agg->io_ticks);
-       agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
+       agg->io_ticks           = le64_to_cpu(agg->io_ticks);
+       agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
        agg->slavecount         = le32_to_cpu(agg->slavecount);
-       agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
+       agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
 }
 
 static void convert_dus(struct disk_util_stat *dus)
@@ -907,14 +1092,14 @@ static void convert_dus(struct disk_util_stat *dus)
        int i;
 
        for (i = 0; i < 2; i++) {
-               dus->s.ios[i]           = le32_to_cpu(dus->s.ios[i]);
-               dus->s.merges[i]        = le32_to_cpu(dus->s.merges[i]);
+               dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
+               dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
                dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
-               dus->s.ticks[i]         = le32_to_cpu(dus->s.ticks[i]);
+               dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
        }
 
-       dus->s.io_ticks         = le32_to_cpu(dus->s.io_ticks);
-       dus->s.time_in_queue    = le32_to_cpu(dus->s.time_in_queue);
+       dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
+       dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
        dus->s.msec             = le64_to_cpu(dus->s.msec);
 }
 
@@ -927,13 +1112,16 @@ static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
                log_info("\nDisk stats (read/write):\n");
        }
 
-       if (output_format == FIO_OUTPUT_JSON) {
+       if (output_format & FIO_OUTPUT_JSON) {
                struct json_object *duobj;
                json_array_add_disk_util(&du->dus, &du->agg, du_array);
                duobj = json_array_last_value_object(du_array);
                json_object_add_client_info(duobj, client);
-       } else
-               print_disk_util(&du->dus, &du->agg, output_format == FIO_OUTPUT_TERSE);
+       }
+       if (output_format & FIO_OUTPUT_TERSE)
+               print_disk_util(&du->dus, &du->agg, 1, NULL);
+       if (output_format & FIO_OUTPUT_NORMAL)
+               print_disk_util(&du->dus, &du->agg, 0, NULL);
 }
 
 static void convert_jobs_eta(struct jobs_eta *je)
@@ -987,15 +1175,12 @@ void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
                dst->eta_sec = je->eta_sec;
 
        dst->nr_threads         += je->nr_threads;
-       /* we need to handle je->run_str too ... */
-}
 
-void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
-{
-       if (!--eta->pending) {
-               eta_fn(&eta->eta);
-               free(eta);
-       }
+       /*
+        * This wont be correct for multiple strings, but at least it
+        * works for the basic cases.
+        */
+       strcpy((char *) dst->run_str, (char *) je->run_str);
 }
 
 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
@@ -1057,6 +1242,7 @@ static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 
        client->eta_in_flight = NULL;
        flist_del_init(&client->eta_list);
+       client->eta_timeouts = 0;
 
        if (client->ops->jobs_eta)
                client->ops->jobs_eta(client, je);
@@ -1065,6 +1251,56 @@ static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
        fio_client_dec_jobs_eta(eta, client->ops->eta);
 }
 
+static int fio_client_handle_iolog(struct fio_client *client,
+                                  struct fio_net_cmd *cmd)
+{
+       struct cmd_iolog_pdu *pdu;
+       bool store_direct;
+
+       pdu = convert_iolog(cmd, &store_direct);
+       if (!pdu) {
+               log_err("fio: failed converting IO log\n");
+               return 1;
+       }
+
+       if (store_direct) {
+               ssize_t ret;
+               size_t sz;
+               int fd;
+
+               fd = open((const char *) pdu->name,
+                               O_WRONLY | O_CREAT | O_TRUNC, 0644);
+               if (fd < 0) {
+                       log_err("fio: open log: %s\n", strerror(errno));
+                       return 1;
+               }
+
+               sz = cmd->pdu_len - sizeof(*pdu);
+               ret = write(fd, pdu->samples, sz);
+               close(fd);
+
+               if (ret != sz) {
+                       log_err("fio: short write on compressed log\n");
+                       return 1;
+               }
+
+               return 0;
+       } else {
+               FILE *f;
+
+               f = fopen((const char *) pdu->name, "w");
+               if (!f) {
+                       log_err("fio: fopen log: %s\n", strerror(errno));
+                       return 1;
+               }
+
+               flush_samples(f, pdu->samples,
+                               pdu->nr_samples * sizeof(struct io_sample));
+               fclose(f);
+               return 0;
+       }
+}
+
 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
@@ -1098,8 +1334,16 @@ static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
        client->jobs = le32_to_cpu(pdu->jobs);
        client->nr_stat = le32_to_cpu(pdu->stat_outputs);
 
-       if (sum_stat_clients > 1)
-               do_output_all_clients = 1;
+       if (client->jobs) {
+               int i;
+
+               if (client->opt_lists)
+                       free(client->opt_lists);
+
+               client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
+               for (i = 0; i < client->jobs; i++)
+                       INIT_FLIST_HEAD(&client->opt_lists[i]);
+       }
 
        sum_stat_clients += client->nr_stat;
 }
@@ -1149,9 +1393,9 @@ static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
        /*
         * Get header first, it's not compressed
         */
-       nr_samples = le32_to_cpu(pdu->nr_samples);
+       nr_samples = le64_to_cpu(pdu->nr_samples);
 
-       total = nr_samples * sizeof(struct io_sample);
+       total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
        ret = malloc(total + sizeof(*pdu));
        ret->nr_samples = nr_samples;
 
@@ -1197,17 +1441,23 @@ err:
  * This has been compressed on the server side, since it can be big.
  * Uncompress here.
  */
-static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
+static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
+                                          bool *store_direct)
 {
        struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
        struct cmd_iolog_pdu *ret;
-       int i;
+       uint64_t i;
+       int compressed;
+       void *samples;
+
+       *store_direct = false;
 
        /*
         * Convert if compressed and we support it. If it's not
         * compressed, we need not do anything.
         */
-       if (le32_to_cpu(pdu->compressed)) {
+       compressed = le32_to_cpu(pdu->compressed);
+       if (compressed == XMIT_COMPRESSED) {
 #ifndef CONFIG_ZLIB
                log_err("fio: server sent compressed data by mistake\n");
                return NULL;
@@ -1217,26 +1467,82 @@ static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
                        log_err("fio: failed decompressing log\n");
                        return NULL;
                }
+       } else if (compressed == STORE_COMPRESSED) {
+               *store_direct = true;
+               ret = pdu;
        } else
                ret = pdu;
 
+       ret->nr_samples         = le64_to_cpu(ret->nr_samples);
        ret->thread_number      = le32_to_cpu(ret->thread_number);
-       ret->nr_samples         = le32_to_cpu(ret->nr_samples);
        ret->log_type           = le32_to_cpu(ret->log_type);
        ret->compressed         = le32_to_cpu(ret->compressed);
+       ret->log_offset         = le32_to_cpu(ret->log_offset);
 
+       if (*store_direct)
+               return ret;
+
+       samples = &ret->samples[0];
        for (i = 0; i < ret->nr_samples; i++) {
-               struct io_sample *s = &ret->samples[i];
+               struct io_sample *s;
+
+               s = __get_sample(samples, ret->log_offset, i);
+               s->time         = le64_to_cpu(s->time);
+               s->val          = le64_to_cpu(s->val);
+               s->__ddir       = le32_to_cpu(s->__ddir);
+               s->bs           = le32_to_cpu(s->bs);
+
+               if (ret->log_offset) {
+                       struct io_sample_offset *so = (void *) s;
 
-               s->time = le64_to_cpu(s->time);
-               s->val  = le64_to_cpu(s->val);
-               s->ddir = le32_to_cpu(s->ddir);
-               s->bs   = le32_to_cpu(s->bs);
+                       so->offset = le64_to_cpu(so->offset);
+               }
        }
 
        return ret;
 }
 
+static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
+                          size_t size, uint64_t tag)
+{
+       rep->error = cpu_to_le32(rep->error);
+       fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
+}
+
+static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
+                        uint64_t tag)
+{
+       struct cmd_sendfile_reply *rep;
+       struct stat sb;
+       size_t size;
+       int fd;
+
+       size = sizeof(*rep);
+       rep = malloc(size);
+
+       if (stat((char *)pdu->path, &sb) < 0) {
+fail:
+               rep->error = errno;
+               sendfile_reply(client->fd, rep, size, tag);
+               free(rep);
+               return 1;
+       }
+
+       size += sb.st_size;
+       rep = realloc(rep, size);
+       rep->size = cpu_to_le32((uint32_t) sb.st_size);
+
+       fd = open((char *)pdu->path, O_RDONLY);
+       if (fd == -1 )
+               goto fail;
+
+       rep->error = read_data(fd, &rep->data, sb.st_size);
+       sendfile_reply(client->fd, rep, size, tag);
+       free(rep);
+       close(fd);
+       return 0;
+}
+
 int fio_handle_client(struct fio_client *client)
 {
        struct client_ops *ops = client->ops;
@@ -1244,7 +1550,7 @@ int fio_handle_client(struct fio_client *client)
 
        dprint(FD_NET, "client: handle %s\n", client->hostname);
 
-       cmd = fio_net_recv_cmd(client->fd);
+       cmd = fio_net_recv_cmd(client->fd, true);
        if (!cmd)
                return 0;
 
@@ -1256,12 +1562,10 @@ int fio_handle_client(struct fio_client *client)
                if (ops->quit)
                        ops->quit(client, cmd);
                remove_client(client);
-               free(cmd);
                break;
        case FIO_NET_CMD_TEXT:
                convert_text(cmd);
                ops->text(client, cmd);
-               free(cmd);
                break;
        case FIO_NET_CMD_DU: {
                struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
@@ -1270,7 +1574,6 @@ int fio_handle_client(struct fio_client *client)
                convert_agg(&du->agg);
 
                ops->disk_util(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_TS: {
@@ -1280,7 +1583,6 @@ int fio_handle_client(struct fio_client *client)
                convert_gs(&p->rs, &p->rs);
 
                ops->thread_status(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_GS: {
@@ -1289,7 +1591,6 @@ int fio_handle_client(struct fio_client *client)
                convert_gs(gs, gs);
 
                ops->group_stats(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_ETA: {
@@ -1298,26 +1599,22 @@ int fio_handle_client(struct fio_client *client)
                remove_reply_cmd(client, cmd);
                convert_jobs_eta(je);
                handle_eta(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_PROBE:
                remove_reply_cmd(client, cmd);
                ops->probe(client, cmd);
-               free(cmd);
                break;
        case FIO_NET_CMD_SERVER_START:
                client->state = Client_running;
                if (ops->job_start)
                        ops->job_start(client, cmd);
-               free(cmd);
                break;
        case FIO_NET_CMD_START: {
                struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 
                pdu->jobs = le32_to_cpu(pdu->jobs);
                ops->start(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_STOP: {
@@ -1328,7 +1625,6 @@ int fio_handle_client(struct fio_client *client)
                client->error = le32_to_cpu(pdu->error);
                client->signal = le32_to_cpu(pdu->signal);
                ops->stop(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_ADD_JOB: {
@@ -1339,32 +1635,77 @@ int fio_handle_client(struct fio_client *client)
 
                if (ops->add_job)
                        ops->add_job(client, cmd);
-               free(cmd);
                break;
                }
        case FIO_NET_CMD_IOLOG:
-               if (ops->iolog) {
-                       struct cmd_iolog_pdu *pdu;
-
-                       pdu = convert_iolog(cmd);
-                       ops->iolog(client, pdu);
-               }
-               free(cmd);
+               fio_client_handle_iolog(client, cmd);
                break;
        case FIO_NET_CMD_UPDATE_JOB:
                ops->update_job(client, cmd);
                remove_reply_cmd(client, cmd);
-               free(cmd);
                break;
+       case FIO_NET_CMD_VTRIGGER: {
+               struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
+               char buf[128];
+               int off = 0;
+
+               if (aux_path) {
+                       strcpy(buf, aux_path);
+                       off = strlen(buf);
+               }
+
+               __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
+               exec_trigger(trigger_cmd);
+               break;
+               }
+       case FIO_NET_CMD_SENDFILE: {
+               struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
+               fio_send_file(client, pdu, cmd->tag);
+               break;
+               }
+       case FIO_NET_CMD_JOB_OPT: {
+               handle_job_opt(client, cmd);
+               break;
+       }
        default:
                log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
-               free(cmd);
                break;
        }
 
+       free(cmd);
        return 1;
 }
 
+int fio_clients_send_trigger(const char *cmd)
+{
+       struct flist_head *entry;
+       struct fio_client *client;
+       size_t slen;
+
+       dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
+
+       if (!cmd)
+               slen = 0;
+       else
+               slen = strlen(cmd);
+
+       flist_for_each(entry, &client_list) {
+               struct cmd_vtrigger_pdu *pdu;
+
+               client = flist_entry(entry, struct fio_client, list);
+
+               pdu = malloc(sizeof(*pdu) + slen);
+               pdu->len = cpu_to_le16((uint16_t) slen);
+               if (slen)
+                       memcpy(pdu->cmd, cmd, slen);
+               fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
+                                       sizeof(*pdu) + slen, NULL, NULL);
+               free(pdu);
+       }
+
+       return 0;
+}
+
 static void request_client_etas(struct client_ops *ops)
 {
        struct fio_client *client;
@@ -1374,8 +1715,7 @@ static void request_client_etas(struct client_ops *ops)
 
        dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 
-       eta = malloc(sizeof(*eta));
-       memset(&eta->eta, 0, sizeof(eta->eta));
+       eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
        eta->pending = nr_clients;
 
        flist_for_each(entry, &client_list) {
@@ -1395,12 +1735,43 @@ static void request_client_etas(struct client_ops *ops)
                                        (uintptr_t) eta, &client->cmd_list);
        }
 
-       while (skipped--)
-               fio_client_dec_jobs_eta(eta, ops->eta);
+       while (skipped--) {
+               if (!fio_client_dec_jobs_eta(eta, ops->eta))
+                       break;
+       }
 
        dprint(FD_NET, "client: requested eta tag %p\n", eta);
 }
 
+/*
+ * A single SEND_ETA timeout isn't fatal. Attempt to recover.
+ */
+static int handle_cmd_timeout(struct fio_client *client,
+                             struct fio_net_cmd_reply *reply)
+{
+       flist_del(&reply->list);
+       free(reply);
+
+       if (reply->opcode != FIO_NET_CMD_SEND_ETA)
+               return 1;
+
+       log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
+
+       flist_del_init(&client->eta_list);
+       if (client->eta_in_flight) {
+               fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
+               client->eta_in_flight = NULL;
+       }
+
+       /*
+        * If we fail 5 in a row, give up...
+        */
+       if (client->eta_timeouts++ > 5)
+               return 1;
+
+       return 0;
+}
+
 static int client_check_cmd_timeout(struct fio_client *client,
                                    struct timeval *now)
 {
@@ -1414,10 +1785,11 @@ static int client_check_cmd_timeout(struct fio_client *client,
                if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
                        continue;
 
+               if (!handle_cmd_timeout(client, reply))
+                       continue;
+
                log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
                                                fio_server_op(reply->opcode));
-               flist_del(&reply->list);
-               free(reply);
                ret = 1;
        }
 
@@ -1447,6 +1819,7 @@ static int fio_check_clients_timed_out(void)
                else
                        log_err("fio: client %s timed out\n", client->hostname);
 
+               client->error = ETIMEDOUT;
                remove_client(client);
                ret = 1;
        }
@@ -1492,6 +1865,7 @@ int fio_handle_clients(struct client_ops *ops)
 
                do {
                        struct timeval tv;
+                       int timeout;
 
                        fio_gettime(&tv, NULL);
                        if (mtime_since(&eta_tv, &tv) >= 900) {
@@ -1502,7 +1876,11 @@ int fio_handle_clients(struct client_ops *ops)
                                        break;
                        }
 
-                       ret = poll(pfds, nr_clients, ops->eta_msec);
+                       check_trigger_file();
+
+                       timeout = min(100u, ops->eta_msec);
+
+                       ret = poll(pfds, nr_clients, timeout);
                        if (ret < 0) {
                                if (errno == EINTR)
                                        continue;
@@ -1535,5 +1913,5 @@ int fio_handle_clients(struct client_ops *ops)
        fio_client_json_fini();
 
        free(pfds);
-       return retval;
+       return retval || error_clients;
 }