Update 'overwrite' option documentation
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index 04a06bac1bee010b82866de1ade9ffd5178f3d7f..b79aa93e644296be9bdf0439863760d8dd4a6cfd 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -36,6 +36,7 @@
 
 #include "fio.h"
 #include "hash.h"
+#include "smalloc.h"
 
 unsigned long page_mask;
 unsigned long page_size;
@@ -48,8 +49,9 @@ int nr_process = 0;
 int nr_thread = 0;
 int shm_id = 0;
 int temp_stall_ts;
+unsigned long done_secs = 0;
 
-static struct fio_sem *startup_sem;
+static struct fio_mutex *startup_mutex;
 static volatile int fio_abort;
 static int exit_value;
 
@@ -60,6 +62,9 @@ struct io_log *agg_io_log[2];
 
 static inline void td_set_runstate(struct thread_data *td, int runstate)
 {
+       if (td->runstate == runstate)
+               return;
+
        dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", td->pid, td->runstate,
                                                                runstate);
        td->runstate = runstate;
@@ -70,16 +75,26 @@ static void terminate_threads(int group_id)
        struct thread_data *td;
        int i;
 
+       dprint(FD_PROCESS, "terminate group_id=%d\n", group_id);
+
        for_each_td(td, i) {
                if (group_id == TERMINATE_ALL || groupid == td->groupid) {
-                       dprint(FD_PROCESS, "setting terminate on %d\n",td->pid);
+                       dprint(FD_PROCESS, "setting terminate on %s/%d\n",
+                                                       td->o.name, td->pid);
+                       td->terminate = 1;
+                       td->o.start_delay = 0;
+
                        /*
                         * if the thread is running, just let it exit
                         */
                        if (td->runstate < TD_RUNNING)
                                kill(td->pid, SIGQUIT);
-                       td->terminate = 1;
-                       td->o.start_delay = 0;
+                       else {
+                               struct ioengine_ops *ops = td->io_ops;
+
+                               if (ops && (ops->flags & FIO_SIGQUIT))
+                                       kill(td->pid, SIGQUIT);
+                       }
                }
        }
 }
@@ -87,16 +102,16 @@ static void terminate_threads(int group_id)
 static void sig_handler(int sig)
 {
        switch (sig) {
-               case SIGALRM:
-                       update_io_ticks();
-                       disk_util_timer_arm();
-                       print_thread_status();
-                       break;
-               default:
-                       printf("\nfio: terminating on signal %d\n", sig);
-                       fflush(stdout);
-                       terminate_threads(TERMINATE_ALL);
-                       break;
+       case SIGALRM:
+               update_io_ticks();
+               disk_util_timer_arm();
+               print_thread_status();
+               break;
+       default:
+               printf("\nfio: terminating on signal %d\n", sig);
+               fflush(stdout);
+               terminate_threads(TERMINATE_ALL);
+               break;
        }
 }
 
@@ -144,12 +159,16 @@ static int check_min_rate(struct thread_data *td, struct timeval *now)
                         * check bandwidth specified rate
                         */
                        if (bytes < td->rate_bytes) {
-                               log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin);
+                               log_err("%s: min rate %u not met\n", td->o.name,
+                                                               td->o.ratemin);
                                return 1;
                        } else {
                                rate = (bytes - td->rate_bytes) / spent;
-                               if (rate < td->o.ratemin || bytes < td->rate_bytes) {
-                                       log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate);
+                               if (rate < td->o.ratemin ||
+                                   bytes < td->rate_bytes) {
+                                       log_err("%s: min rate %u not met, got"
+                                               " %luKiB/sec\n", td->o.name,
+                                                       td->o.ratemin, rate);
                                        return 1;
                                }
                        }
@@ -158,12 +177,17 @@ static int check_min_rate(struct thread_data *td, struct timeval *now)
                         * checks iops specified rate
                         */
                        if (iops < td->o.rate_iops) {
-                               log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops);
+                               log_err("%s: min iops rate %u not met\n",
+                                               td->o.name, td->o.rate_iops);
                                return 1;
                        } else {
                                rate = (iops - td->rate_blocks) / spent;
-                               if (rate < td->o.rate_iops_min || iops < td->rate_blocks) {
-                                       log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate);
+                               if (rate < td->o.rate_iops_min ||
+                                   iops < td->rate_blocks) {
+                                       log_err("%s: min iops rate %u not met,"
+                                               " got %lu\n", td->o.name,
+                                                       td->o.rate_iops_min,
+                                                       rate);
                                }
                        }
                }
@@ -351,7 +375,6 @@ static void do_verify(struct thread_data *td)
                                io_u->xfer_buflen = io_u->resid;
                                io_u->xfer_buf += bytes;
                                io_u->offset += bytes;
-                               f->last_completed_pos = io_u->offset;
 
                                td->ts.short_io_u[io_u->ddir]++;
 
@@ -453,8 +476,11 @@ static void do_io(struct thread_data *td)
                 * Add verification end_io handler, if asked to verify
                 * a previously written file.
                 */
-               if (td->o.verify != VERIFY_NONE)
+               if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) {
                        io_u->end_io = verify_io_u;
+                       td_set_runstate(td, TD_VERIFYING);
+               } else
+                       td_set_runstate(td, TD_RUNNING);
 
                ret = td_io_queue(td, io_u);
                switch (ret) {
@@ -477,7 +503,6 @@ static void do_io(struct thread_data *td)
                                io_u->xfer_buflen = io_u->resid;
                                io_u->xfer_buf += bytes;
                                io_u->offset += bytes;
-                               f->last_completed_pos = io_u->offset;
 
                                td->ts.short_io_u[io_u->ddir]++;
 
@@ -526,7 +551,8 @@ sync_done:
                                min_evts = 1;
 
                                if (td->cur_depth > td->o.iodepth_low)
-                                       min_evts = td->cur_depth - td->o.iodepth_low;
+                                       min_evts = td->cur_depth
+                                                       - td->o.iodepth_low;
                        }
 
                        fio_gettime(&comp_time, NULL);
@@ -639,16 +665,17 @@ static int init_io_u(struct thread_data *td)
        int i, max_units;
        char *p;
 
-       if (td->io_ops->flags & FIO_SYNCIO)
-               max_units = 1;
-       else
-               max_units = td->o.iodepth;
-
+       max_units = td->o.iodepth;
        max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
-       td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units;
+       td->orig_buffer_size = (unsigned long long) max_bs
+                                       * (unsigned long long) max_units;
 
-       if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE)
-               td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1);
+       if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
+               unsigned long bs;
+
+               bs = td->orig_buffer_size + td->o.hugepage_size - 1;
+               td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1);
+       }
 
        if (td->orig_buffer_size != (size_t) td->orig_buffer_size) {
                log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
@@ -701,7 +728,8 @@ static int switch_ioscheduler(struct thread_data *td)
        f = fopen(tmp, "r+");
        if (!f) {
                if (errno == ENOENT) {
-                       log_err("fio: os or kernel doesn't support IO scheduler switching\n");
+                       log_err("fio: os or kernel doesn't support IO scheduler"
+                               " switching\n");
                        return 0;
                }
                td_verror(td, errno, "fopen iosched");
@@ -755,7 +783,8 @@ static int keep_running(struct thread_data *td)
                return 1;
        }
 
-       io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes;
+       io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]
+                       + td->io_skip_bytes;
        if (io_done < td->o.size)
                return 1;
 
@@ -783,8 +812,7 @@ static int clear_io_state(struct thread_data *td)
        if (td->o.time_based || td->o.loops)
                td->nr_done_files = 0;
 
-       for_each_file(td, f, i)
-               td_io_close_file(td, f);
+       close_files(td);
 
        ret = 0;
        for_each_file(td, f, i) {
@@ -822,14 +850,14 @@ static void *thread_main(void *data)
        td->io_hist_tree = RB_ROOT;
 
        td_set_runstate(td, TD_INITIALIZED);
-       fio_sem_up(startup_sem);
-       fio_sem_down(td->mutex);
+       fio_mutex_up(startup_mutex);
+       fio_mutex_down(td->mutex);
 
        /*
-        * the ->mutex semaphore is now no longer used, close it to avoid
+        * the ->mutex mutex is now no longer used, close it to avoid
         * eating a file descriptor
         */
-       fio_sem_remove(td->mutex);
+       fio_mutex_remove(td->mutex);
 
        /*
         * May alter parameters that init_io_u() will use, so we need to
@@ -889,7 +917,8 @@ static void *thread_main(void *data)
                memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
 
                if (td->o.ratemin)
-                       memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate));
+                       memcpy(&td->lastrate, &td->ts.stat_sample_time,
+                                                       sizeof(td->lastrate));
 
                if (clear_state && clear_io_state(td))
                        break;
@@ -902,7 +931,8 @@ static void *thread_main(void *data)
 
                if (td_read(td) && td->io_bytes[DDIR_READ]) {
                        if (td->rw_end_set[DDIR_READ])
-                               elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]);
+                               elapsed = utime_since(&td->start,
+                                                     &td->rw_end[DDIR_READ]);
                        else
                                elapsed = utime_since_now(&td->start);
 
@@ -910,13 +940,14 @@ static void *thread_main(void *data)
                }
                if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
                        if (td->rw_end_set[DDIR_WRITE])
-                               elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]);
+                               elapsed = utime_since(&td->start,
+                                                     &td->rw_end[DDIR_WRITE]);
                        else
                                elapsed = utime_since_now(&td->start);
 
                        runtime[DDIR_WRITE] += elapsed;
                }
-               
+
                if (td->error || td->terminate)
                        break;
 
@@ -961,8 +992,9 @@ static void *thread_main(void *data)
 
 err:
        if (td->error)
-               printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
-       close_files(td);
+               printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error,
+                                                       td->verror);
+       close_and_free_files(td);
        close_ioengine(td);
        cleanup_io_u(td);
 
@@ -1024,7 +1056,11 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                else
                        realthreads++;
 
-               if (!td->pid || td->runstate == TD_REAPED)
+               if (!td->pid) {
+                       pending++;
+                       continue;
+               }
+               if (td->runstate == TD_REAPED)
                        continue;
                if (td->o.use_thread) {
                        if (td->runstate == TD_EXITED) {
@@ -1044,7 +1080,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                ret = waitpid(td->pid, &status, flags);
                if (ret < 0) {
                        if (errno == ECHILD) {
-                               log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate);
+                               log_err("fio: pid=%d disappeared %d\n", td->pid,
+                                                               td->runstate);
                                td_set_runstate(td, TD_REAPED);
                                goto reaped;
                        }
@@ -1054,7 +1091,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                                int sig = WTERMSIG(status);
 
                                if (sig != SIGQUIT)
-                                       log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
+                                       log_err("fio: pid=%d, got signal=%d\n",
+                                                               td->pid, sig);
                                td_set_runstate(td, TD_REAPED);
                                goto reaped;
                        }
@@ -1073,23 +1111,16 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                pending++;
                continue;
 reaped:
-               if (td->o.use_thread) {
-                       long ret;
-
-                       dprint(FD_PROCESS, "joining tread %d\n", td->pid);
-                       if (pthread_join(td->thread, (void *) &ret)) {
-                               dprint(FD_PROCESS, "join failed %ld\n", ret);
-                               perror("pthread_join");
-                       }
-               }
-
                (*nr_running)--;
                (*m_rate) -= td->o.ratemin;
                (*t_rate) -= td->o.rate;
-               pending--;
+               if (!td->pid)
+                       pending--;
 
                if (td->error)
                        exit_value++;
+
+               done_secs += mtime_since_now(&td->epoch) / 1000;
        }
 
        if (*nr_running == cputhreads && !pending && realthreads)
@@ -1111,11 +1142,13 @@ static void run_threads(void)
        if (!terse_output) {
                printf("Starting ");
                if (nr_thread)
-                       printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : "");
+                       printf("%d thread%s", nr_thread,
+                                               nr_thread > 1 ? "s" : "");
                if (nr_process) {
                        if (nr_thread)
                                printf(" and ");
-                       printf("%d process%s", nr_process, nr_process > 1 ? "es" : "");
+                       printf("%d process%s", nr_process,
+                                               nr_process > 1 ? "es" : "");
                }
                printf("\n");
                fflush(stdout);
@@ -1145,9 +1178,21 @@ static void run_threads(void)
                if (setup_files(td)) {
                        exit_value++;
                        if (td->error)
-                               log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
+                               log_err("fio: pid=%d, err=%d/%s\n", td->pid,
+                                                       td->error, td->verror);
                        td_set_runstate(td, TD_REAPED);
                        todo--;
+               } else {
+                       struct fio_file *f;
+                       unsigned int i;
+
+                       /*
+                        * for sharing to work, each job must always open
+                        * its own files. so close them, if we opened them
+                        * for creation
+                        */
+                       for_each_file(td, f, i)
+                               td_io_close_file(td, f);
                }
 
                init_disk_util(td);
@@ -1183,8 +1228,11 @@ static void run_threads(void)
                                        continue;
                        }
 
-                       if (td->o.stonewall && (nr_started || nr_running))
+                       if (td->o.stonewall && (nr_started || nr_running)) {
+                               dprint(FD_PROCESS, "%s: stonewall wait\n",
+                                                       td->o.name);
                                break;
+                       }
 
                        /*
                         * Set state to created. Thread will transition
@@ -1196,11 +1244,14 @@ static void run_threads(void)
 
                        if (td->o.use_thread) {
                                dprint(FD_PROCESS, "will pthread_create\n");
-                               if (pthread_create(&td->thread, NULL, thread_main, td)) {
-                                       perror("thread_create");
+                               if (pthread_create(&td->thread, NULL,
+                                                  thread_main, td)) {
+                                       perror("pthread_create");
                                        nr_started--;
                                        break;
                                }
+                               if (pthread_detach(td->thread) < 0)
+                                       perror("pthread_detach");
                        } else {
                                dprint(FD_PROCESS, "will fork\n");
                                if (!fork()) {
@@ -1209,7 +1260,7 @@ static void run_threads(void)
                                        exit(ret);
                                }
                        }
-                       fio_sem_down(startup_sem);
+                       fio_mutex_down(startup_mutex);
                }
 
                /*
@@ -1264,7 +1315,7 @@ static void run_threads(void)
                        m_rate += td->o.ratemin;
                        t_rate += td->o.rate;
                        todo--;
-                       fio_sem_up(td->mutex);
+                       fio_mutex_up(td->mutex);
                }
 
                reap_threads(&nr_running, &t_rate, &m_rate);
@@ -1286,6 +1337,8 @@ int main(int argc, char *argv[])
 {
        long ps;
 
+       sinit();
+
        /*
         * We need locale for number printing, if it isn't set then just
         * go with the US format.
@@ -1313,7 +1366,7 @@ int main(int argc, char *argv[])
                setup_log(&agg_io_log[DDIR_WRITE]);
        }
 
-       startup_sem = fio_sem_init(0);
+       startup_mutex = fio_mutex_init(0);
 
        set_genesis_time();
 
@@ -1324,11 +1377,12 @@ int main(int argc, char *argv[])
        if (!fio_abort) {
                show_run_stats();
                if (write_bw_log) {
-                       __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log");
-                       __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log");
+                       __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log");
+                       __finish_log(agg_io_log[DDIR_WRITE],
+                                       "agg-write_bw.log");
                }
        }
 
-       fio_sem_remove(startup_sem);
+       fio_mutex_remove(startup_mutex);
        return exit_value;
 }