fio: factor out endian_check
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index bce13a72306fccc62d84020ee4d57a3a2d8d8722..99d6209adb5bc4c6c6cd6d555ec2a3ac3040c935 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -2,7 +2,7 @@
  * fio - the flexible io tester
  *
  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
- * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
+ * Copyright (C) 2006-2012 Jens Axboe <axboe@kernel.dk>
  *
  * The license below covers all files distributed with fio unless otherwise
  * noted in the file itself.
  *
  */
 #include <unistd.h>
-#include <fcntl.h>
-#include <string.h>
-#include <signal.h>
+#include <locale.h>
 #include <time.h>
-#include <assert.h>
-#include <sys/stat.h>
-#include <sys/wait.h>
-#include <sys/ipc.h>
-#include <sys/shm.h>
-#include <sys/ioctl.h>
-#include <sys/mman.h>
 
 #include "fio.h"
-#include "os.h"
-
-static unsigned long page_mask;
-#define ALIGN(buf)     \
-       (char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
-
-int groupid = 0;
-int thread_number = 0;
-int shm_id = 0;
-int temp_stall_ts;
-
-static volatile int startup_sem;
-static volatile int fio_abort;
-
-struct io_log *agg_io_log[2];
-
-#define TERMINATE_ALL          (-1)
-#define JOB_START_TIMEOUT      (5 * 1000)
-
-static inline void td_set_runstate(struct thread_data *td, int runstate)
-{
-       td->runstate = runstate;
-}
-
-static void terminate_threads(int group_id, int forced_kill)
-{
-       struct thread_data *td;
-       int i;
-
-       for_each_td(td, i) {
-               if (group_id == TERMINATE_ALL || groupid == td->groupid) {
-                       td->terminate = 1;
-                       td->start_delay = 0;
-                       if (forced_kill)
-                               td_set_runstate(td, TD_EXITED);
-               }
-       }
-}
-
-static void sig_handler(int sig)
-{
-       switch (sig) {
-               case SIGALRM:
-                       update_io_ticks();
-                       disk_util_timer_arm();
-                       print_thread_status();
-                       break;
-               case SIGSEGV:
-                       fprintf(stderr, "fio: got segfault, aborting\n");
-                       terminate_threads(TERMINATE_ALL, 1);
-                       fio_abort = 1;
-                       exit(0);
-               default:
-                       printf("\nfio: terminating on signal %d\n", sig);
-                       fflush(stdout);
-                       terminate_threads(TERMINATE_ALL, 0);
-                       break;
-       }
-}
-
-/*
- * Check if we are above the minimum rate given.
- */
-static int check_min_rate(struct thread_data *td, struct timeval *now)
-{
-       unsigned long spent;
-       unsigned long rate;
-       int ddir = td->ddir;
-
-       /*
-        * allow a 2 second settle period in the beginning
-        */
-       if (mtime_since(&td->start, now) < 2000)
-               return 0;
-
-       /*
-        * if rate blocks is set, sample is running
-        */
-       if (td->rate_bytes) {
-               spent = mtime_since(&td->lastrate, now);
-               if (spent < td->ratecycle)
-                       return 0;
-
-               rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
-               if (rate < td->ratemin) {
-                       fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate);
-                       return 1;
-               }
-       }
-
-       td->rate_bytes = td->this_io_bytes[ddir];
-       memcpy(&td->lastrate, now, sizeof(*now));
-       return 0;
-}
-
-static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
-{
-       if (!td->timeout)
-               return 0;
-       if (mtime_since(&td->epoch, t) >= td->timeout * 1000)
-               return 1;
-
-       return 0;
-}
-
-static struct fio_file *get_next_file(struct thread_data *td)
-{
-       unsigned int old_next_file = td->next_file;
-       struct fio_file *f;
-
-       do {
-               f = &td->files[td->next_file];
-
-               td->next_file++;
-               if (td->next_file >= td->nr_files)
-                       td->next_file = 0;
-
-               if (f->fd != -1)
-                       break;
-
-               f = NULL;
-       } while (td->next_file != old_next_file);
-
-       return f;
-}
-
-/*
- * When job exits, we can cancel the in-flight IO if we are using async
- * io. Attempt to do so.
- */
-static void cleanup_pending_aio(struct thread_data *td)
-{
-       struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
-       struct list_head *entry, *n;
-       struct io_completion_data icd;
-       struct io_u *io_u;
-       int r;
-
-       /*
-        * get immediately available events, if any
-        */
-       r = td_io_getevents(td, 0, td->cur_depth, &ts);
-       if (r > 0) {
-               icd.nr = r;
-               ios_completed(td, &icd);
-       }
-
-       /*
-        * now cancel remaining active events
-        */
-       if (td->io_ops->cancel) {
-               list_for_each_safe(entry, n, &td->io_u_busylist) {
-                       io_u = list_entry(entry, struct io_u, list);
-
-                       r = td->io_ops->cancel(td, io_u);
-                       if (!r)
-                               put_io_u(td, io_u);
-               }
-       }
-
-       if (td->cur_depth) {
-               r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL);
-               if (r > 0) {
-                       icd.nr = r;
-                       ios_completed(td, &icd);
-               }
-       }
-}
-
-/*
- * Helper to handle the final sync of a file. Works just like the normal
- * io path, just does everything sync.
- */
-static int fio_io_sync(struct thread_data *td, struct fio_file *f)
-{
-       struct io_u *io_u = __get_io_u(td);
-       struct io_completion_data icd;
-       int ret;
-
-       if (!io_u)
-               return 1;
-
-       io_u->ddir = DDIR_SYNC;
-       io_u->file = f;
-
-       if (td_io_prep(td, io_u)) {
-               put_io_u(td, io_u);
-               return 1;
-       }
-
-       ret = td_io_queue(td, io_u);
-       if (ret) {
-               td_verror(td, io_u->error);
-               put_io_u(td, io_u);
-               return 1;
-       }
-
-       ret = td_io_getevents(td, 1, td->cur_depth, NULL);
-       if (ret < 0) {
-               td_verror(td, ret);
-               return 1;
-       }
-
-       icd.nr = ret;
-       ios_completed(td, &icd);
-       if (icd.error) {
-               td_verror(td, icd.error);
-               return 1;
-       }
-
-       return 0;
-}
-
-/*
- * The main verify engine. Runs over the writes we previusly submitted,
- * reads the blocks back in, and checks the crc/md5 of the data.
- */
-static void do_verify(struct thread_data *td)
-{
-       struct io_u *io_u, *v_io_u = NULL;
-       struct io_completion_data icd;
-       struct fio_file *f;
-       int ret, i;
-
-       /*
-        * sync io first and invalidate cache, to make sure we really
-        * read from disk.
-        */
-       for_each_file(td, f, i) {
-               fio_io_sync(td, f);
-               file_invalidate_cache(td, f);
-       }
-
-       td_set_runstate(td, TD_VERIFYING);
-
-       do {
-               if (td->terminate)
-                       break;
-
-               io_u = __get_io_u(td);
-               if (!io_u)
-                       break;
-
-               if (runtime_exceeded(td, &io_u->start_time)) {
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               if (get_next_verify(td, io_u)) {
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               f = get_next_file(td);
-               if (!f)
-                       break;
-
-               io_u->file = f;
-
-               if (td_io_prep(td, io_u)) {
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               ret = td_io_queue(td, io_u);
-               if (ret) {
-                       td_verror(td, io_u->error);
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               /*
-                * we have one pending to verify, do that while
-                * we are doing io on the next one
-                */
-               if (do_io_u_verify(td, &v_io_u))
-                       break;
-
-               ret = td_io_getevents(td, 1, 1, NULL);
-               if (ret != 1) {
-                       if (ret < 0)
-                               td_verror(td, ret);
-                       break;
-               }
-
-               v_io_u = td->io_ops->event(td, 0);
-               icd.nr = 1;
-               icd.error = 0;
-               fio_gettime(&icd.time, NULL);
-               io_completed(td, v_io_u, &icd);
-
-               if (icd.error) {
-                       td_verror(td, icd.error);
-                       put_io_u(td, v_io_u);
-                       v_io_u = NULL;
-                       break;
-               }
-
-               /*
-                * if we can't submit more io, we need to verify now
-                */
-               if (queue_full(td) && do_io_u_verify(td, &v_io_u))
-                       break;
-
-       } while (1);
-
-       do_io_u_verify(td, &v_io_u);
-
-       if (td->cur_depth)
-               cleanup_pending_aio(td);
-
-       td_set_runstate(td, TD_RUNNING);
-}
-
-/*
- * Not really an io thread, all it does is burn CPU cycles in the specified
- * manner.
- */
-static void do_cpuio(struct thread_data *td)
-{
-       struct timeval e;
-       int split = 100 / td->cpuload;
-       int i = 0;
-
-       while (!td->terminate) {
-               fio_gettime(&e, NULL);
-
-               if (runtime_exceeded(td, &e))
-                       break;
-
-               if (!(i % split))
-                       __usec_sleep(10000);
-               else
-                       usec_sleep(td, 10000);
-
-               i++;
-       }
-}
-
-/*
- * Main IO worker function. It retrieves io_u's to process and queues
- * and reaps them, checking for rate and errors along the way.
- */
-static void do_io(struct thread_data *td)
-{
-       struct io_completion_data icd;
-       struct timeval s;
-       unsigned long usec;
-       struct fio_file *f;
-       int i, ret = 0;
-
-       td_set_runstate(td, TD_RUNNING);
-
-       while (td->this_io_bytes[td->ddir] < td->io_size) {
-               struct timespec *timeout;
-               int min_evts = 0;
-               struct io_u *io_u;
-
-               if (td->terminate)
-                       break;
-
-               f = get_next_file(td);
-               if (!f)
-                       break;
-
-               io_u = get_io_u(td, f);
-               if (!io_u)
-                       break;
-
-               memcpy(&s, &io_u->start_time, sizeof(s));
-
-               ret = td_io_queue(td, io_u);
-               if (ret) {
-                       td_verror(td, io_u->error);
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
-
-               if (td->cur_depth < td->iodepth) {
-                       struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
-
-                       timeout = &ts;
-                       min_evts = 0;
-               } else {
-                       timeout = NULL;
-                       min_evts = 1;
-               }
-
-               ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
-               if (ret < 0) {
-                       td_verror(td, ret);
-                       break;
-               } else if (!ret)
-                       continue;
-
-               icd.nr = ret;
-               ios_completed(td, &icd);
-               if (icd.error) {
-                       td_verror(td, icd.error);
-                       break;
-               }
-
-               /*
-                * the rate is batched for now, it should work for batches
-                * of completions except the very first one which may look
-                * a little bursty
-                */
-               usec = utime_since(&s, &icd.time);
-
-               rate_throttle(td, usec, icd.bytes_done[td->ddir], td->ddir);
-
-               if (check_min_rate(td, &icd.time)) {
-                       if (exitall_on_terminate)
-                               terminate_threads(td->groupid, 0);
-                       td_verror(td, ENODATA);
-                       break;
-               }
-
-               if (runtime_exceeded(td, &icd.time))
-                       break;
-
-               if (td->thinktime) {
-                       unsigned long long b;
-
-                       b = td->io_blocks[0] + td->io_blocks[1];
-                       if (!(b % td->thinktime_blocks))
-                               usec_sleep(td, td->thinktime);
-               }
-       }
-
-       if (!td->error) {
-               if (td->cur_depth)
-                       cleanup_pending_aio(td);
-
-               if (should_fsync(td) && td->end_fsync) {
-                       td_set_runstate(td, TD_FSYNCING);
-                       for_each_file(td, f, i)
-                               fio_io_sync(td, f);
-               }
-       }
-}
-
-static void cleanup_io_u(struct thread_data *td)
-{
-       struct list_head *entry, *n;
-       struct io_u *io_u;
-
-       list_for_each_safe(entry, n, &td->io_u_freelist) {
-               io_u = list_entry(entry, struct io_u, list);
-
-               list_del(&io_u->list);
-               free(io_u);
-       }
-
-       free_io_mem(td);
-}
-
-/*
- * "randomly" fill the buffer contents
- */
-static void fill_rand_buf(struct io_u *io_u, int max_bs)
-{
-       int *ptr = io_u->buf;
-
-       while ((void *) ptr - io_u->buf < max_bs) {
-               *ptr = rand() * 0x9e370001;
-               ptr++;
-       }
-}
-
-static int init_io_u(struct thread_data *td)
-{
-       struct io_u *io_u;
-       unsigned int max_bs;
-       int i, max_units;
-       char *p;
-
-       if (td->io_ops->flags & FIO_CPUIO)
-               return 0;
-
-       if (td->io_ops->flags & FIO_SYNCIO)
-               max_units = 1;
-       else
-               max_units = td->iodepth;
-
-       max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]);
-       td->orig_buffer_size = max_bs * max_units;
-
-       if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE)
-               td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1);
-       else
-               td->orig_buffer_size += page_mask;
-
-       if (allocate_io_mem(td))
-               return 1;
-
-       p = ALIGN(td->orig_buffer);
-       for (i = 0; i < max_units; i++) {
-               io_u = malloc(sizeof(*io_u));
-               memset(io_u, 0, sizeof(*io_u));
-               INIT_LIST_HEAD(&io_u->list);
-
-               io_u->buf = p + max_bs * i;
-               if (td_write(td) || td_rw(td))
-                       fill_rand_buf(io_u, max_bs);
-
-               io_u->index = i;
-               list_add(&io_u->list, &td->io_u_freelist);
-       }
-
-       return 0;
-}
-
-static int switch_ioscheduler(struct thread_data *td)
+#include "hash.h"
+#include "smalloc.h"
+#include "verify.h"
+#include "trim.h"
+#include "diskutil.h"
+#include "profile.h"
+#include "lib/rand.h"
+#include "memalign.h"
+#include "client.h"
+#include "server.h"
+#include "endian_check.h"
+
+unsigned long page_mask;
+unsigned long page_size;
+
+int main(int argc, char *argv[], char *envp[])
 {
-       char tmp[256], tmp2[128];
-       FILE *f;
-       int ret;
-
-       if (td->io_ops->flags & FIO_CPUIO)
-               return 0;
-
-       sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
+       long ps;
 
-       f = fopen(tmp, "r+");
-       if (!f) {
-               td_verror(td, errno);
+       if (endian_check()) {
+               log_err("fio: endianness settings appear wrong.\n");
+               log_err("fio: please report this to fio@vger.kernel.org\n");
                return 1;
        }
 
-       /*
-        * Set io scheduler.
-        */
-       ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f);
-       if (ferror(f) || ret != 1) {
-               td_verror(td, errno);
-               fclose(f);
-               return 1;
-       }
+       arch_init(envp);
 
-       rewind(f);
+       sinit();
 
        /*
-        * Read back and check that the selected scheduler is now the default.
+        * We need locale for number printing, if it isn't set then just
+        * go with the US format.
         */
-       ret = fread(tmp, 1, sizeof(tmp), f);
-       if (ferror(f) || ret < 0) {
-               td_verror(td, errno);
-               fclose(f);
-               return 1;
-       }
-
-       sprintf(tmp2, "[%s]", td->ioscheduler);
-       if (!strstr(tmp, tmp2)) {
-               log_err("fio: io scheduler %s not found\n", td->ioscheduler);
-               td_verror(td, EINVAL);
-               fclose(f);
-               return 1;
-       }
-
-       fclose(f);
-       return 0;
-}
-
-static void clear_io_state(struct thread_data *td)
-{
-       struct fio_file *f;
-       int i;
-
-       td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0;
-       td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
-       td->zone_bytes = 0;
-
-       for_each_file(td, f, i) {
-               f->last_pos = 0;
-               if (td->io_ops->flags & FIO_SYNCIO)
-                       lseek(f->fd, SEEK_SET, 0);
-
-               if (f->file_map)
-                       memset(f->file_map, 0, f->num_maps * sizeof(long));
-       }
-}
-
-/*
- * Entry point for the thread based jobs. The process based jobs end up
- * here as well, after a little setup.
- */
-static void *thread_main(void *data)
-{
-       unsigned long long runtime[2];
-       struct thread_data *td = data;
-
-       if (!td->use_thread)
-               setsid();
-
-       td->pid = getpid();
-
-       INIT_LIST_HEAD(&td->io_u_freelist);
-       INIT_LIST_HEAD(&td->io_u_busylist);
-       INIT_LIST_HEAD(&td->io_hist_list);
-       INIT_LIST_HEAD(&td->io_log_list);
-
-       if (init_io_u(td))
-               goto err;
-
-       if (fio_setaffinity(td) == -1) {
-               td_verror(td, errno);
-               goto err;
-       }
-
-       if (init_iolog(td))
-               goto err;
-
-       if (td->ioprio) {
-               if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
-                       td_verror(td, errno);
-                       goto err;
-               }
-       }
-
-       if (nice(td->nice) == -1) {
-               td_verror(td, errno);
-               goto err;
-       }
-
-       if (init_random_state(td))
-               goto err;
-
-       if (td->ioscheduler && switch_ioscheduler(td))
-               goto err;
-
-       td_set_runstate(td, TD_INITIALIZED);
-       fio_sem_up(&startup_sem);
-       fio_sem_down(&td->mutex);
-
-       if (!td->create_serialize && setup_files(td))
-               goto err;
-       if (open_files(td))
-               goto err;
-
-       /*
-        * Do this late, as some IO engines would like to have the
-        * files setup prior to initializing structures.
-        */
-       if (td_io_init(td))
-               goto err;
-
-       if (td->exec_prerun)
-               system(td->exec_prerun);
-
-       fio_gettime(&td->epoch, NULL);
-       getrusage(RUSAGE_SELF, &td->ru_start);
-
-       runtime[0] = runtime[1] = 0;
-       while (td->loops--) {
-               fio_gettime(&td->start, NULL);
-               memcpy(&td->stat_sample_time, &td->start, sizeof(td->start));
-
-               if (td->ratemin)
-                       memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate));
-
-               clear_io_state(td);
-               prune_io_piece_log(td);
-
-               if (td->io_ops->flags & FIO_CPUIO)
-                       do_cpuio(td);
-               else
-                       do_io(td);
-
-               runtime[td->ddir] += utime_since_now(&td->start);
-               if (td_rw(td) && td->io_bytes[td->ddir ^ 1])
-                       runtime[td->ddir ^ 1] = runtime[td->ddir];
-
-               if (td->error || td->terminate)
-                       break;
-
-               if (td->verify == VERIFY_NONE)
-                       continue;
-
-               clear_io_state(td);
-               fio_gettime(&td->start, NULL);
-
-               do_verify(td);
-
-               runtime[DDIR_READ] += utime_since_now(&td->start);
-
-               if (td->error || td->terminate)
-                       break;
-       }
-
-       update_rusage_stat(td);
-       fio_gettime(&td->end_time, NULL);
-       td->runtime[0] = runtime[0] / 1000;
-       td->runtime[1] = runtime[1] / 1000;
-
-       if (td->bw_log)
-               finish_log(td, td->bw_log, "bw");
-       if (td->slat_log)
-               finish_log(td, td->slat_log, "slat");
-       if (td->clat_log)
-               finish_log(td, td->clat_log, "clat");
-       if (td->write_iolog_file)
-               write_iolog_close(td);
-       if (td->exec_postrun)
-               system(td->exec_postrun);
-
-       if (exitall_on_terminate)
-               terminate_threads(td->groupid, 0);
-
-err:
-       close_files(td);
-       close_ioengine(td);
-       cleanup_io_u(td);
-       td_set_runstate(td, TD_EXITED);
-       return NULL;
-
-}
-
-/*
- * We cannot pass the td data into a forked process, so attach the td and
- * pass it to the thread worker.
- */
-static void *fork_main(int shmid, int offset)
-{
-       struct thread_data *td;
-       void *data;
-
-       data = shmat(shmid, NULL, 0);
-       if (data == (void *) -1) {
-               perror("shmat");
-               return NULL;
-       }
-
-       td = data + offset * sizeof(struct thread_data);
-       thread_main(td);
-       shmdt(data);
-       return NULL;
-}
-
-/*
- * Run over the job map and reap the threads that have exited, if any.
- */
-static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
-{
-       struct thread_data *td;
-       int i, cputhreads, pending;
-
-       /*
-        * reap exited threads (TD_EXITED -> TD_REAPED)
-        */
-       pending = cputhreads = 0;
-       for_each_td(td, i) {
-               /*
-                * ->io_ops is NULL for a thread that has closed its
-                * io engine
-                */
-               if (td->io_ops && td->io_ops->flags & FIO_CPUIO)
-                       cputhreads++;
-
-               if (td->runstate != TD_EXITED) {
-                       if (td->runstate < TD_RUNNING)
-                               pending++;
-
-                       continue;
-               }
-
-               td_set_runstate(td, TD_REAPED);
-
-               if (td->use_thread) {
-                       long ret;
-
-                       if (pthread_join(td->thread, (void *) &ret))
-                               perror("thread_join");
-               } else
-                       waitpid(td->pid, NULL, 0);
-
-               (*nr_running)--;
-               (*m_rate) -= td->ratemin;
-               (*t_rate) -= td->rate;
-       }
-
-       if (*nr_running == cputhreads && !pending)
-               terminate_threads(TERMINATE_ALL, 0);
-}
-
-/*
- * Main function for kicking off and reaping jobs, as needed.
- */
-static void run_threads(void)
-{
-       struct thread_data *td;
-       unsigned long spent;
-       int i, todo, nr_running, m_rate, t_rate, nr_started;
-
-       if (fio_pin_memory())
-               return;
-
-       if (!terse_output) {
-               printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
-               fflush(stdout);
-       }
-
-       signal(SIGINT, sig_handler);
-       signal(SIGALRM, sig_handler);
-
-       todo = thread_number;
-       nr_running = 0;
-       nr_started = 0;
-       m_rate = t_rate = 0;
-
-       for_each_td(td, i) {
-               print_status_init(td->thread_number - 1);
-
-               if (!td->create_serialize) {
-                       init_disk_util(td);
-                       continue;
-               }
-
-               /*
-                * do file setup here so it happens sequentially,
-                * we don't want X number of threads getting their
-                * client data interspersed on disk
-                */
-               if (setup_files(td)) {
-                       td_set_runstate(td, TD_REAPED);
-                       todo--;
-               }
-
-               init_disk_util(td);
-       }
-
-       while (todo) {
-               struct thread_data *map[MAX_JOBS];
-               struct timeval this_start;
-               int this_jobs = 0, left;
-
-               /*
-                * create threads (TD_NOT_CREATED -> TD_CREATED)
-                */
-               for_each_td(td, i) {
-                       if (td->runstate != TD_NOT_CREATED)
-                               continue;
-
-                       /*
-                        * never got a chance to start, killed by other
-                        * thread for some reason
-                        */
-                       if (td->terminate) {
-                               todo--;
-                               continue;
-                       }
-
-                       if (td->start_delay) {
-                               spent = mtime_since_genesis();
-
-                               if (td->start_delay * 1000 > spent)
-                                       continue;
-                       }
-
-                       if (td->stonewall && (nr_started || nr_running))
-                               break;
-
-                       /*
-                        * Set state to created. Thread will transition
-                        * to TD_INITIALIZED when it's done setting up.
-                        */
-                       td_set_runstate(td, TD_CREATED);
-                       map[this_jobs++] = td;
-                       fio_sem_init(&startup_sem, 1);
-                       nr_started++;
-
-                       if (td->use_thread) {
-                               if (pthread_create(&td->thread, NULL, thread_main, td)) {
-                                       perror("thread_create");
-                                       nr_started--;
-                               }
-                       } else {
-                               if (fork())
-                                       fio_sem_down(&startup_sem);
-                               else {
-                                       fork_main(shm_id, i);
-                                       exit(0);
-                               }
-                       }
-               }
-
-               /*
-                * Wait for the started threads to transition to
-                * TD_INITIALIZED.
-                */
-               fio_gettime(&this_start, NULL);
-               left = this_jobs;
-               while (left && !fio_abort) {
-                       if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
-                               break;
-
-                       usleep(100000);
-
-                       for (i = 0; i < this_jobs; i++) {
-                               td = map[i];
-                               if (!td)
-                                       continue;
-                               if (td->runstate == TD_INITIALIZED) {
-                                       map[i] = NULL;
-                                       left--;
-                               } else if (td->runstate >= TD_EXITED) {
-                                       map[i] = NULL;
-                                       left--;
-                                       todo--;
-                                       nr_running++; /* work-around... */
-                               }
-                       }
-               }
-
-               if (left) {
-                       log_err("fio: %d jobs failed to start\n", left);
-                       for (i = 0; i < this_jobs; i++) {
-                               td = map[i];
-                               if (!td)
-                                       continue;
-                               kill(td->pid, SIGTERM);
-                       }
-                       break;
-               }
-
-               /*
-                * start created threads (TD_INITIALIZED -> TD_RUNNING).
-                */
-               for_each_td(td, i) {
-                       if (td->runstate != TD_INITIALIZED)
-                               continue;
-
-                       td_set_runstate(td, TD_RUNNING);
-                       nr_running++;
-                       nr_started--;
-                       m_rate += td->ratemin;
-                       t_rate += td->rate;
-                       todo--;
-                       fio_sem_up(&td->mutex);
-               }
-
-               reap_threads(&nr_running, &t_rate, &m_rate);
-
-               if (todo)
-                       usleep(100000);
-       }
-
-       while (nr_running) {
-               reap_threads(&nr_running, &t_rate, &m_rate);
-               usleep(10000);
-       }
-
-       update_io_ticks();
-       fio_unpin_memory();
-}
-
-int main(int argc, char *argv[])
-{
-       long ps;
-
-       if (parse_options(argc, argv))
-               return 1;
-
-       if (!thread_number) {
-               log_err("Nothing to do\n");
-               return 1;
-       }
+       if (!getenv("LC_NUMERIC"))
+               setlocale(LC_NUMERIC, "en_US");
 
        ps = sysconf(_SC_PAGESIZE);
        if (ps < 0) {
@@ -1011,24 +68,16 @@ int main(int argc, char *argv[])
                return 1;
        }
 
+       page_size = ps;
        page_mask = ps - 1;
 
-       if (write_bw_log) {
-               setup_log(&agg_io_log[DDIR_READ]);
-               setup_log(&agg_io_log[DDIR_WRITE]);
-       }
-
-       disk_util_timer_arm();
-
-       run_threads();
+       fio_keywords_init();
 
-       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");
-               }
-       }
+       if (parse_options(argc, argv))
+               return 1;
 
-       return 0;
+       if (nr_clients)
+               return fio_handle_clients(&fio_client_ops);
+       else
+               return fio_backend();
 }