-
-/*
- * 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 elapsed;
- struct thread_data *td = data;
- pthread_condattr_t attr;
- int clear_state;
-
- if (!td->o.use_thread) {
- setsid();
- td->pid = getpid();
- } else
- td->pid = gettid();
-
- dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
-
- INIT_FLIST_HEAD(&td->io_u_freelist);
- INIT_FLIST_HEAD(&td->io_u_busylist);
- INIT_FLIST_HEAD(&td->io_u_requeues);
- INIT_FLIST_HEAD(&td->io_log_list);
- INIT_FLIST_HEAD(&td->io_hist_list);
- INIT_FLIST_HEAD(&td->verify_list);
- INIT_FLIST_HEAD(&td->trim_list);
- pthread_mutex_init(&td->io_u_lock, NULL);
- td->io_hist_tree = RB_ROOT;
-
- pthread_condattr_init(&attr);
- pthread_cond_init(&td->verify_cond, &attr);
- pthread_cond_init(&td->free_cond, &attr);
-
- td_set_runstate(td, TD_INITIALIZED);
- dprint(FD_MUTEX, "up startup_mutex\n");
- fio_mutex_up(startup_mutex);
- dprint(FD_MUTEX, "wait on td->mutex\n");
- fio_mutex_down(td->mutex);
- dprint(FD_MUTEX, "done waiting on td->mutex\n");
-
- /*
- * the ->mutex mutex is now no longer used, close it to avoid
- * eating a file descriptor
- */
- fio_mutex_remove(td->mutex);
-
- /*
- * A new gid requires privilege, so we need to do this before setting
- * the uid.
- */
- if (td->o.gid != -1U && setgid(td->o.gid)) {
- td_verror(td, errno, "setgid");
- goto err;
- }
- if (td->o.uid != -1U && setuid(td->o.uid)) {
- td_verror(td, errno, "setuid");
- goto err;
- }
-
- /*
- * 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);
-
- /*
- * 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;
- }
-
- /*
- * May alter parameters that init_io_u() will use, so we need to
- * do this first.
- */
- if (init_iolog(td))
- goto err;
-
- if (init_io_u(td))
- goto err;
-
- if (td->o.verify_async && verify_async_init(td))
- goto err;
-
- if (td->ioprio_set) {
- if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
- td_verror(td, errno, "ioprio_set");
- goto err;
- }
- }
-
- if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt))
- goto err;
-
- if (nice(td->o.nice) == -1) {
- td_verror(td, errno, "nice");
- goto err;
- }
-
- if (td->o.ioscheduler && switch_ioscheduler(td))
- goto err;
-
- if (!td->o.create_serialize && setup_files(td))
- goto err;
-
- if (td_io_init(td))
- goto err;
-
- if (init_random_map(td))
- goto err;
-
- if (td->o.exec_prerun) {
- if (exec_string(td->o.exec_prerun))
- goto err;
- }
-
- if (td->o.pre_read) {
- if (pre_read_files(td) < 0)
- goto err;
- }
-
- fio_gettime(&td->epoch, NULL);
- getrusage(RUSAGE_SELF, &td->ru_start);
-
- clear_state = 0;
- while (keep_running(td)) {
- fio_gettime(&td->start, NULL);
- memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
- memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
- memcpy(&td->tv_cache, &td->start, sizeof(td->start));
-
- if (td->o.ratemin[0] || td->o.ratemin[1])
- memcpy(&td->lastrate, &td->bw_sample_time,
- sizeof(td->lastrate));
-
- if (clear_state)
- clear_io_state(td);
-
- prune_io_piece_log(td);
-
- do_io(td);
-
- clear_state = 1;
-
- if (td_read(td) && td->io_bytes[DDIR_READ]) {
- elapsed = utime_since_now(&td->start);
- td->ts.runtime[DDIR_READ] += elapsed;
- }
- if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
- elapsed = utime_since_now(&td->start);
- td->ts.runtime[DDIR_WRITE] += elapsed;
- }
-
- if (td->error || td->terminate)
- break;
-
- if (!td->o.do_verify ||
- td->o.verify == VERIFY_NONE ||
- (td->io_ops->flags & FIO_UNIDIR))
- continue;
-
- clear_io_state(td);
-
- fio_gettime(&td->start, NULL);
-
- do_verify(td);
-
- td->ts.runtime[DDIR_READ] += utime_since_now(&td->start);
-
- if (td->error || td->terminate)
- break;
- }
-
- update_rusage_stat(td);
- td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000;
- td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 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];
-
- fio_mutex_down(writeout_mutex);
- if (td->bw_log) {
- if (td->o.bw_log_file) {
- finish_log_named(td, td->bw_log,
- td->o.bw_log_file, "bw");
- } else
- finish_log(td, td->bw_log, "bw");
- }
- if (td->lat_log) {
- if (td->o.lat_log_file) {
- finish_log_named(td, td->lat_log,
- td->o.lat_log_file, "lat");
- } else
- finish_log(td, td->lat_log, "lat");
- }
- if (td->slat_log) {
- if (td->o.lat_log_file) {
- finish_log_named(td, td->slat_log,
- td->o.lat_log_file, "slat");
- } else
- finish_log(td, td->slat_log, "slat");
- }
- if (td->clat_log) {
- if (td->o.lat_log_file) {
- finish_log_named(td, td->clat_log,
- td->o.lat_log_file, "clat");
- } else
- finish_log(td, td->clat_log, "clat");
- }
- if (td->iops_log) {
- if (td->o.iops_log_file) {
- finish_log_named(td, td->iops_log,
- td->o.iops_log_file, "iops");
- } else
- finish_log(td, td->iops_log, "iops");
- }
-
- fio_mutex_up(writeout_mutex);
- if (td->o.exec_postrun)
- exec_string(td->o.exec_postrun);
-
- if (exitall_on_terminate)
- fio_terminate_threads(td->groupid);
-
-err:
- if (td->error)
- log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
- td->verror);
-
- if (td->o.verify_async)
- verify_async_exit(td);
-
- close_and_free_files(td);
- close_ioengine(td);
- cleanup_io_u(td);
- cgroup_shutdown(td, &cgroup_mnt);
-
- if (td->o.cpumask_set) {
- int ret = fio_cpuset_exit(&td->o.cpumask);
-
- td_verror(td, ret, "fio_cpuset_exit");
- }
-
- /*
- * do this very late, it will log file closing as well
- */
- if (td->o.write_iolog_file)
- write_iolog_close(td);
-
- options_mem_free(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;
-
-#ifndef __hpux
- data = shmat(shmid, NULL, 0);
- if (data == (void *) -1) {
- int __err = errno;
-
- perror("shmat");
- return __err;
- }
-#else
- /*
- * HP-UX inherits shm mappings?
- */
- data = threads;
-#endif
-
- 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, realthreads, pending, status, ret;
-
- /*
- * reap exited threads (TD_EXITED -> TD_REAPED)
- */
- realthreads = 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 && !strcmp(td->io_ops->name, "cpuio"))
- cputhreads++;
- else
- realthreads++;
-
- if (!td->pid) {
- pending++;
- continue;
- }
- if (td->runstate == TD_REAPED)
- continue;
- if (td->o.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",
- (int) 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 != SIGTERM)
- log_err("fio: pid=%d, got signal=%d\n",
- (int) 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
- */
- pending++;
- continue;
-reaped:
- (*nr_running)--;
- (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]);
- (*t_rate) -= (td->o.rate[0] + td->o.rate[1]);
- if (!td->pid)
- pending--;
-
- if (td->error)
- exit_value++;
-
- done_secs += mtime_since_now(&td->epoch) / 1000;
- }
-
- if (*nr_running == cputhreads && !pending && realthreads)
- fio_terminate_threads(TERMINATE_ALL);
-}
-
-static void *gtod_thread_main(void *data)
-{
- fio_mutex_up(startup_mutex);
-
- /*
- * As long as we have jobs around, update the clock. It would be nice
- * to have some way of NOT hammering that CPU with gettimeofday(),
- * but I'm not sure what to use outside of a simple CPU nop to relax
- * it - we don't want to lose precision.
- */
- while (threads) {
- fio_gtod_update();
- nop;
- }
-
- return NULL;
-}
-
-static int fio_start_gtod_thread(void)
-{
- pthread_attr_t attr;
- int ret;
-
- pthread_attr_init(&attr);
- pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN);
- ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL);
- pthread_attr_destroy(&attr);
- if (ret) {
- log_err("Can't create gtod thread: %s\n", strerror(ret));
- return 1;
- }
-
- ret = pthread_detach(gtod_thread);
- if (ret) {
- log_err("Can't detatch gtod thread: %s\n", strerror(ret));
- return 1;
- }
-
- dprint(FD_MUTEX, "wait on startup_mutex\n");
- fio_mutex_down(startup_mutex);
- dprint(FD_MUTEX, "done waiting on startup_mutex\n");
- return 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 (fio_gtod_offload && fio_start_gtod_thread())
- return;
-
- set_sig_handlers();
-
- if (!terse_output) {
- log_info("Starting ");
- if (nr_thread)
- log_info("%d thread%s", nr_thread,
- nr_thread > 1 ? "s" : "");
- if (nr_process) {
- if (nr_thread)
- printf(" and ");
- log_info("%d process%s", nr_process,
- nr_process > 1 ? "es" : "");
- }
- log_info("\n");
- fflush(stdout);
- }
-
- 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->o.create_serialize)
- 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",
- (int) td->pid, td->error, td->verror);
- td_set_runstate(td, TD_REAPED);
- todo--;
- } else {
- struct fio_file *f;
- unsigned int j;
-
- /*
- * 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, j) {
- if (fio_file_open(f))
- td_io_close_file(td, f);
- }
- }
- }
-
- set_genesis_time();
-
- while (todo) {
- struct thread_data *map[REAL_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->o.start_delay) {
- spent = mtime_since_genesis();
-
- if (td->o.start_delay * 1000 > spent)
- continue;
- }
-
- if (td->o.stonewall && (nr_started || nr_running)) {
- dprint(FD_PROCESS, "%s: stonewall wait\n",
- td->o.name);
- break;
- }
-
- init_disk_util(td);
-
- /*
- * 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;
- nr_started++;
-
- if (td->o.use_thread) {
- int ret;
-
- dprint(FD_PROCESS, "will pthread_create\n");
- ret = pthread_create(&td->thread, NULL,
- thread_main, td);
- if (ret) {
- log_err("pthread_create: %s\n",
- strerror(ret));
- nr_started--;
- break;
- }
- ret = pthread_detach(td->thread);
- if (ret)
- log_err("pthread_detach: %s",
- strerror(ret));
- } else {
- pid_t pid;
- dprint(FD_PROCESS, "will fork\n");
- pid = fork();
- if (!pid) {
- int ret = fork_main(shm_id, i);
-
- _exit(ret);
- } else if (i == fio_debug_jobno)
- *fio_debug_jobp = pid;
- }
- dprint(FD_MUTEX, "wait on startup_mutex\n");
- if (fio_mutex_down_timeout(startup_mutex, 10)) {
- log_err("fio: job startup hung? exiting.\n");
- fio_terminate_threads(TERMINATE_ALL);
- fio_abort = 1;
- nr_started--;
- break;
- }
- dprint(FD_MUTEX, "done waiting on startup_mutex\n");
- }
-
- /*
- * 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;
-
- if (in_ramp_time(td))
- td_set_runstate(td, TD_RAMP);
- else
- td_set_runstate(td, TD_RUNNING);
- nr_running++;
- nr_started--;
- m_rate += td->o.ratemin[0] + td->o.ratemin[1];
- t_rate += td->o.rate[0] + td->o.rate[1];
- todo--;
- fio_mutex_up(td->mutex);
- }
-
- reap_threads(&nr_running, &t_rate, &m_rate);
-
- if (todo) {
- if (is_backend)
- fio_server_idle_loop();
- else
- usleep(100000);
- }
- }
-
- while (nr_running) {
- reap_threads(&nr_running, &t_rate, &m_rate);
-
- if (is_backend)
- fio_server_idle_loop();
- else
- usleep(10000);
- }
-
- update_io_ticks();
- fio_unpin_memory();
-}
-
-int exec_run(void)
-{
- if (nr_clients)
- return fio_handle_clients();
- if (exec_profile && load_profile(exec_profile))
- return 1;
-
- if (!thread_number)
- return 0;
-
- if (write_bw_log) {
- setup_log(&agg_io_log[DDIR_READ]);
- setup_log(&agg_io_log[DDIR_WRITE]);
- }
-
- startup_mutex = fio_mutex_init(0);
- if (startup_mutex == NULL)
- return 1;
- writeout_mutex = fio_mutex_init(1);
- if (writeout_mutex == NULL)
- return 1;
-
- set_genesis_time();
- create_disk_util_thread();
-
- cgroup_list = smalloc(sizeof(*cgroup_list));
- INIT_FLIST_HEAD(cgroup_list);
-
- 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");
- }
- }
-
- cgroup_kill(cgroup_list);
- sfree(cgroup_list);
- sfree(cgroup_mnt);
-
- fio_mutex_remove(startup_mutex);
- fio_mutex_remove(writeout_mutex);
- return exit_value;
-}
-
-void reset_fio_state(void)
-{
- groupid = 0;
- thread_number = 0;
- nr_process = 0;
- nr_thread = 0;
- done_secs = 0;
-}
-
-static int endian_check(void)
-{
- union {
- uint8_t c[8];
- uint64_t v;
- } u;
- int le = 0, be = 0;
-
- u.v = 0x12;
- if (u.c[7] == 0x12)
- be = 1;
- else if (u.c[0] == 0x12)
- le = 1;
-
-#if defined(FIO_LITTLE_ENDIAN)
- if (be)
- return 1;
-#elif defined(FIO_BIG_ENDIAN)
- if (le)
- return 1;
-#else
- return 1;
-#endif
-
- if (!le && !be)
- return 1;
-
- return 0;
-}
-
-int main(int argc, char *argv[], char *envp[])
-{
- long ps;
-
- 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;
- }
-
- arch_init(envp);
-
- sinit();
-
- /*
- * 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");
-
- ps = sysconf(_SC_PAGESIZE);
- if (ps < 0) {
- log_err("Failed to get page size\n");
- return 1;
- }
-
- page_size = ps;
- page_mask = ps - 1;
-
- fio_keywords_init();
-
- if (parse_options(argc, argv))
- return 1;
-
- return exec_run();
-}