- cleanup_pending_aio(td);
-}
-
-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;
- io_u->flags = IO_U_F_FREE;
- list_add(&io_u->list, &td->io_u_freelist);
- }
-
- io_u_init_timeout();
-
- return 0;
-}
-
-static int switch_ioscheduler(struct thread_data *td)
-{
- 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);
-
- f = fopen(tmp, "r+");
- if (!f) {
- td_verror(td, errno, "fopen");
- return 1;
- }
-
- /*
- * Set io scheduler.
- */
- ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f);
- if (ferror(f) || ret != 1) {
- td_verror(td, errno, "fwrite");
- fclose(f);
- return 1;
- }
-
- rewind(f);
-
- /*
- * Read back and check that the selected scheduler is now the default.
- */
- ret = fread(tmp, 1, sizeof(tmp), f);
- if (ferror(f) || ret < 0) {
- td_verror(td, errno, "fread");
- 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, "iosched_switch");
- fclose(f);
- return 1;
- }
-
- fclose(f);
- return 0;
-}
-
-static void clear_io_state(struct thread_data *td)
-{
- struct fio_file *f;
- int i;
-
- td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0;
- td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
- td->zone_bytes = 0;
-
- td->last_was_sync = 0;
-
- for_each_file(td, f, i) {
- f->last_completed_pos = 0;
-
- 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_u_requeues);
- 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, "cpu_set_affinity");
- 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, "ioprio_set");
- goto err;
- }
- }
-
- if (nice(td->nice) == -1) {
- td_verror(td, errno, "nice");
- 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) {
- if (system(td->exec_prerun) < 0)
- goto err;
- }
-
- fio_gettime(&td->epoch, NULL);
- memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch));
- getrusage(RUSAGE_SELF, &td->ts.ru_start);
-
- runtime[0] = runtime[1] = 0;
- while (td->loops--) {
- fio_gettime(&td->start, NULL);
- memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
-
- if (td->ratemin)
- memcpy(&td->lastrate, &td->ts.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);
-
- if (td_read(td) && td->io_bytes[DDIR_READ])
- runtime[DDIR_READ] += utime_since_now(&td->start);
- if (td_write(td) && td->io_bytes[DDIR_WRITE])
- runtime[DDIR_WRITE] += utime_since_now(&td->start);
-
- 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);
- td->ts.runtime[0] = runtime[0] / 1000;
- td->ts.runtime[1] = runtime[1] / 1000;
- td->ts.total_run_time = mtime_since_now(&td->epoch);
- td->ts.io_bytes[0] = td->io_bytes[0];
- td->ts.io_bytes[1] = td->io_bytes[1];
-
- if (td->ts.bw_log)
- finish_log(td, td->ts.bw_log, "bw");
- if (td->ts.slat_log)
- finish_log(td, td->ts.slat_log, "slat");
- if (td->ts.clat_log)
- finish_log(td, td->ts.clat_log, "clat");
- if (td->write_iolog_file)
- write_iolog_close(td);
- if (td->exec_postrun) {
- if (system(td->exec_postrun) < 0)
- log_err("fio: postrun %s failed\n", td->exec_postrun);
- }
-
- if (exitall_on_terminate)
- terminate_threads(td->groupid);
-
-err:
- if (td->error)
- printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
- close_files(td);
- close_ioengine(td);
- cleanup_io_u(td);
- td_set_runstate(td, TD_EXITED);
- return (void *) (unsigned long) td->error;
-}
-
-/*
- * We cannot pass the td data into a forked process, so attach the td and
- * pass it to the thread worker.
- */
-static int fork_main(int shmid, int offset)
-{
- struct thread_data *td;
- void *data, *ret;
-
- data = shmat(shmid, NULL, 0);
- if (data == (void *) -1) {
- int __err = errno;
-
- perror("shmat");
- return __err;
- }
-
- td = data + offset * sizeof(struct thread_data);
- ret = thread_main(td);
- shmdt(data);
- return (int) (unsigned long) ret;
-}
-
-/*
- * 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, status, ret;
-
- /*
- * reap exited threads (TD_EXITED -> TD_REAPED)
- */
- pending = cputhreads = 0;
- for_each_td(td, i) {
- int flags = 0;
-
- /*
- * ->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->pid || td->runstate == TD_REAPED)
- continue;
- if (td->use_thread) {
- if (td->runstate == TD_EXITED) {
- td_set_runstate(td, TD_REAPED);
- goto reaped;
- }
- continue;
- }
-
- flags = WNOHANG;
- if (td->runstate == TD_EXITED)
- flags = 0;
-
- /*
- * check if someone quit or got killed in an unusual way
- */
- ret = waitpid(td->pid, &status, flags);
- if (ret < 0) {
- if (errno == ECHILD) {
- log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate);
- td_set_runstate(td, TD_REAPED);
- goto reaped;
- }
- perror("waitpid");
- } else if (ret == td->pid) {
- if (WIFSIGNALED(status)) {
- int sig = WTERMSIG(status);
-
- if (sig != SIGQUIT)
- log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
- td_set_runstate(td, TD_REAPED);
- goto reaped;
- }
- if (WIFEXITED(status)) {
- if (WEXITSTATUS(status) && !td->error)
- td->error = WEXITSTATUS(status);
-
- td_set_runstate(td, TD_REAPED);
- goto reaped;
- }
- }
-
- /*
- * thread is not dead, continue
- */
- continue;
-reaped:
- if (td->use_thread) {
- long ret;
-
- if (pthread_join(td->thread, (void *) &ret))
- perror("pthread_join");
- }
-
- (*nr_running)--;
- (*m_rate) -= td->ratemin;
- (*t_rate) -= td->rate;
-
- if (td->error)
- exit_value++;
- }
-
- if (*nr_running == cputhreads && !pending)
- terminate_threads(TERMINATE_ALL);
-}
-
-/*
- * 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)) {
- exit_value++;
- if (td->error)
- log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
- td_set_runstate(td, TD_REAPED);
- todo--;
- }
-
- init_disk_util(td);
- }
-
- set_genesis_time();
-
- 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 {
- int ret = fork_main(shm_id, i);
-
- exit(ret);
- }
- }
- }
-
- /*
- * 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;
-
- /*
- * We need locale for number printing, if it isn't set then just
- * go with the US format.
- */
- if (!getenv("LC_NUMERIC"))
- setlocale(LC_NUMERIC, "en_US");
-
- if (parse_options(argc, argv))
- return 1;
-
- if (!thread_number) {
- log_err("Nothing to do\n");
- return 1;
- }
-
- ps = sysconf(_SC_PAGESIZE);
- if (ps < 0) {
- log_err("Failed to get page size\n");
- return 1;
- }
-
- page_mask = ps - 1;
-
- if (write_bw_log) {
- setup_log(&agg_io_log[DDIR_READ]);
- setup_log(&agg_io_log[DDIR_WRITE]);
- }
-
- set_genesis_time();
-
- disk_util_timer_arm();
-
- run_threads();
-
- 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");
- }
- }
-
- return exit_value;