Merge branch 'master' into gfio
[fio.git] / backend.c
index f3f103038474e3c7e19bd2ece97203080bdab857..38efc2307809266d234cfeb183e33bda94b5ea68 100644 (file)
--- a/backend.c
+++ b/backend.c
@@ -56,13 +56,13 @@ static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
 static int exit_value;
 static volatile int fio_abort;
+static unsigned int nr_process = 0;
+static unsigned int nr_thread = 0;
 
 struct io_log *agg_io_log[2];
 
 int groupid = 0;
 unsigned int thread_number = 0;
-unsigned int nr_process = 0;
-unsigned int nr_thread = 0;
 int shm_id = 0;
 int temp_stall_ts;
 unsigned long done_secs = 0;
@@ -87,6 +87,11 @@ static void sig_int(int sig)
        }
 }
 
+static void sig_show_status(int sig)
+{
+       show_running_run_stats();
+}
+
 static void set_sig_handlers(void)
 {
        struct sigaction act;
@@ -101,6 +106,11 @@ static void set_sig_handlers(void)
        act.sa_flags = SA_RESTART;
        sigaction(SIGTERM, &act, NULL);
 
+       memset(&act, 0, sizeof(act));
+       act.sa_handler = sig_show_status;
+       act.sa_flags = SA_RESTART;
+       sigaction(SIGUSR1, &act, NULL);
+
        if (is_backend) {
                memset(&act, 0, sizeof(act));
                act.sa_handler = sig_int;
@@ -335,8 +345,8 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
                        return 1;
 
                if (td_non_fatal_error(err)) {
-                       /*
-                        * Continue with the I/Os in case of
+                       /*
+                        * Continue with the I/Os in case of
                         * a non fatal error.
                         */
                        update_error_count(td, err);
@@ -494,7 +504,10 @@ sync_done:
                if (full || !td->o.iodepth_batch_complete) {
                        min_events = min(td->o.iodepth_batch_complete,
                                         td->cur_depth);
-                       if (full && !min_events && td->o.iodepth_batch_complete != 0)
+                       /*
+                        * if the queue is full, we MUST reap at least 1 event
+                        */
+                       if (full && !min_events)
                                min_events = 1;
 
                        do {
@@ -526,6 +539,20 @@ sync_done:
        dprint(FD_VERIFY, "exiting loop\n");
 }
 
+static int io_bytes_exceeded(struct thread_data *td)
+{
+       unsigned long long bytes;
+
+       if (td_rw(td))
+               bytes = td->this_io_bytes[0] + td->this_io_bytes[1];
+       else if (td_write(td))
+               bytes = td->this_io_bytes[1];
+       else
+               bytes = td->this_io_bytes[0];
+
+       return bytes >= td->o.size;
+}
+
 /*
  * Main IO worker function. It retrieves io_u's to process and queues
  * and reaps them, checking for rate and errors along the way.
@@ -540,9 +567,9 @@ static void do_io(struct thread_data *td)
        else
                td_set_runstate(td, TD_RUNNING);
 
-       while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
-               (!flist_empty(&td->trim_list)) ||
-               ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) {
+       while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
+               (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
+               td->o.time_based) {
                struct timeval comp_time;
                unsigned long bytes_done[2] = { 0, 0 };
                int min_evts = 0;
@@ -573,11 +600,12 @@ static void do_io(struct thread_data *td)
                ddir = io_u->ddir;
 
                /*
-                * Add verification end_io handler, if asked to verify
-                * a previously written file.
+                * Add verification end_io handler if:
+                *      - Asked to verify (!td_rw(td))
+                *      - Or the io_u is from our verify list (mixed write/ver)
                 */
                if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
-                   !td_rw(td)) {
+                   ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
                        if (td->o.verify_async)
                                io_u->end_io = verify_io_u_async;
                        else
@@ -662,7 +690,10 @@ sync_done:
                if (full || !td->o.iodepth_batch_complete) {
                        min_evts = min(td->o.iodepth_batch_complete,
                                        td->cur_depth);
-                       if (full && !min_evts && td->o.iodepth_batch_complete != 0)
+                       /*
+                        * if the queue is full, we MUST reap at least 1 event
+                        */
+                       if (full && !min_evts)
                                min_evts = 1;
 
                        if (__should_check_rate(td, 0) ||
@@ -937,10 +968,12 @@ static void *thread_main(void *data)
 {
        unsigned long long elapsed;
        struct thread_data *td = data;
+       struct thread_options *o = &td->o;
        pthread_condattr_t attr;
        int clear_state;
+       int ret;
 
-       if (!td->o.use_thread) {
+       if (!o->use_thread) {
                setsid();
                td->pid = getpid();
        } else
@@ -948,6 +981,9 @@ static void *thread_main(void *data)
 
        dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
 
+       if (is_backend)
+               fio_server_send_start(td);
+
        INIT_FLIST_HEAD(&td->io_u_freelist);
        INIT_FLIST_HEAD(&td->io_u_busylist);
        INIT_FLIST_HEAD(&td->io_u_requeues);
@@ -974,16 +1010,17 @@ static void *thread_main(void *data)
         * eating a file descriptor
         */
        fio_mutex_remove(td->mutex);
+       td->mutex = NULL;
 
        /*
         * A new gid requires privilege, so we need to do this before setting
         * the uid.
         */
-       if (td->o.gid != -1U && setgid(td->o.gid)) {
+       if (o->gid != -1U && setgid(o->gid)) {
                td_verror(td, errno, "setgid");
                goto err;
        }
-       if (td->o.uid != -1U && setuid(td->o.uid)) {
+       if (o->uid != -1U && setuid(o->uid)) {
                td_verror(td, errno, "setuid");
                goto err;
        }
@@ -992,18 +1029,24 @@ static void *thread_main(void *data)
         * If we have a gettimeofday() thread, make sure we exclude that
         * thread from this job
         */
-       if (td->o.gtod_cpu)
-               fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu);
+       if (o->gtod_cpu)
+               fio_cpu_clear(&o->cpumask, o->gtod_cpu);
 
        /*
         * Set affinity first, in case it has an impact on the memory
         * allocations.
         */
-       if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) {
-               td_verror(td, errno, "cpu_set_affinity");
-               goto err;
+       if (o->cpumask_set) {
+               ret = fio_setaffinity(td->pid, o->cpumask);
+               if (ret == -1) {
+                       td_verror(td, errno, "cpu_set_affinity");
+                       goto err;
+               }
        }
 
+       if (fio_pin_memory(td))
+               goto err;
+
        /*
         * May alter parameters that init_io_u() will use, so we need to
         * do this first.
@@ -1014,29 +1057,30 @@ static void *thread_main(void *data)
        if (init_io_u(td))
                goto err;
 
-       if (td->o.verify_async && verify_async_init(td))
+       if (o->verify_async && verify_async_init(td))
                goto err;
 
-       if (td->ioprio_set) {
-               if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
+       if (o->ioprio) {
+               ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
+               if (ret == -1) {
                        td_verror(td, errno, "ioprio_set");
                        goto err;
                }
        }
 
-       if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt))
+       if (td->o.cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt))
                goto err;
 
        errno = 0;
-       if (nice(td->o.nice) == -1 && errno != 0) {
+       if (nice(o->nice) == -1 && errno != 0) {
                td_verror(td, errno, "nice");
                goto err;
        }
 
-       if (td->o.ioscheduler && switch_ioscheduler(td))
+       if (o->ioscheduler && switch_ioscheduler(td))
                goto err;
 
-       if (!td->o.create_serialize && setup_files(td))
+       if (!o->create_serialize && setup_files(td))
                goto err;
 
        if (td_io_init(td))
@@ -1045,16 +1089,16 @@ static void *thread_main(void *data)
        if (init_random_map(td))
                goto err;
 
-       if (td->o.exec_prerun) {
-               if (exec_string(td->o.exec_prerun))
-                       goto err;
-       }
+       if (o->exec_prerun && exec_string(o->exec_prerun))
+               goto err;
 
-       if (td->o.pre_read) {
+       if (o->pre_read) {
                if (pre_read_files(td) < 0)
                        goto err;
        }
 
+       fio_verify_init(td);
+
        fio_gettime(&td->epoch, NULL);
        getrusage(RUSAGE_SELF, &td->ru_start);
 
@@ -1066,9 +1110,9 @@ static void *thread_main(void *data)
                memcpy(&td->tv_cache, &td->start, sizeof(td->start));
 
                if (td->o.ratemin[0] || td->o.ratemin[1]) {
-                       memcpy(&td->lastrate[0], &td->bw_sample_time,
+                       memcpy(&td->lastrate[0], &td->bw_sample_time,
                                                sizeof(td->bw_sample_time));
-                       memcpy(&td->lastrate[1], &td->bw_sample_time,
+                       memcpy(&td->lastrate[1], &td->bw_sample_time,
                                                sizeof(td->bw_sample_time));
                }
 
@@ -1117,6 +1161,8 @@ static void *thread_main(void *data)
        td->ts.io_bytes[0] = td->io_bytes[0];
        td->ts.io_bytes[1] = td->io_bytes[1];
 
+       fio_unpin_memory(td);
+
        fio_mutex_down(writeout_mutex);
        if (td->bw_log) {
                if (td->o.bw_log_file) {
@@ -1174,8 +1220,8 @@ err:
        cleanup_io_u(td);
        cgroup_shutdown(td, &cgroup_mnt);
 
-       if (td->o.cpumask_set) {
-               int ret = fio_cpuset_exit(&td->o.cpumask);
+       if (o->cpumask_set) {
+               int ret = fio_cpuset_exit(&o->cpumask);
 
                td_verror(td, ret, "fio_cpuset_exit");
        }
@@ -1273,6 +1319,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
                        if (errno == ECHILD) {
                                log_err("fio: pid=%d disappeared %d\n",
                                                (int) td->pid, td->runstate);
+                               td->sig = ECHILD;
                                td_set_runstate(td, TD_REAPED);
                                goto reaped;
                        }
@@ -1284,6 +1331,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
                                if (sig != SIGTERM)
                                        log_err("fio: pid=%d, got signal=%d\n",
                                                        (int) td->pid, sig);
+                               td->sig = sig;
                                td_set_runstate(td, TD_REAPED);
                                goto reaped;
                        }
@@ -1327,14 +1375,19 @@ static void run_threads(void)
        unsigned long spent;
        unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
 
-       if (fio_pin_memory())
-               return;
-
        if (fio_gtod_offload && fio_start_gtod_thread())
                return;
 
        set_sig_handlers();
 
+       nr_thread = nr_process = 0;
+       for_each_td(td, i) {
+               if (td->o.use_thread)
+                       nr_thread++;
+               else
+                       nr_process++;
+       }
+
        if (!terse_output) {
                log_info("Starting ");
                if (nr_thread)
@@ -1534,25 +1587,16 @@ static void run_threads(void)
 
                reap_threads(&nr_running, &t_rate, &m_rate);
 
-               if (todo) {
-                       if (is_backend)
-                               fio_server_idle_loop();
-                       else
-                               usleep(100000);
-               }
+               if (todo)
+                       usleep(100000);
        }
 
        while (nr_running) {
                reap_threads(&nr_running, &t_rate, &m_rate);
-
-               if (is_backend)
-                       fio_server_idle_loop();
-               else
-                       usleep(10000);
+               usleep(10000);
        }
 
        update_io_ticks();
-       fio_unpin_memory();
 }
 
 static void *disk_thread_main(void *data)
@@ -1609,8 +1653,8 @@ int fio_backend(void)
                return 0;
 
        if (write_bw_log) {
-               setup_log(&agg_io_log[DDIR_READ], 0);
-               setup_log(&agg_io_log[DDIR_WRITE], 0);
+               setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW);
+               setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW);
        }
 
        startup_mutex = fio_mutex_init(0);
@@ -1640,6 +1684,7 @@ int fio_backend(void)
        for_each_td(td, i)
                fio_options_free(td);
 
+       free_disk_util();
        cgroup_kill(cgroup_list);
        sfree(cgroup_list);
        sfree(cgroup_mnt);