X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=b55a5274f4ad56e698cdea9d154a1ef7935b36d6;hp=edc0644a113b259cf2d65cf91f96175a5f50201b;hb=16e56d250f72e4fec4591d562a5f4d7a16ba1bb8;hpb=9cc8cb91d355d7e47d5b930b352087ec71f85f85 diff --git a/backend.c b/backend.c index edc0644a..b55a5274 100644 --- a/backend.c +++ b/backend.c @@ -57,11 +57,7 @@ #include "workqueue.h" #include "lib/mountcheck.h" #include "rate-submit.h" - -static pthread_t helper_thread; -static pthread_mutex_t helper_lock; -pthread_cond_t helper_cond; -int helper_do_stat = 0; +#include "helper_thread.h" static struct fio_mutex *startup_mutex; static struct flist_head *cgroup_list; @@ -79,7 +75,6 @@ unsigned int stat_number = 0; int shm_id = 0; int temp_stall_ts; unsigned long done_secs = 0; -volatile int helper_exit = 0; #define PAGE_ALIGN(buf) \ (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask) @@ -309,6 +304,8 @@ requeue: put_io_u(td, io_u); return true; } else if (ret == FIO_Q_QUEUED) { + if (td_io_commit(td)) + return true; if (io_u_queued_complete(td, 1) < 0) return true; } else if (ret == FIO_Q_COMPLETED) { @@ -444,6 +441,12 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time) int min_evts = 0; int ret; + if (td->flags & TD_F_REGROW_LOGS) { + ret = io_u_quiesce(td); + regrow_logs(td); + return ret; + } + /* * if the queue is full, we MUST reap at least 1 event */ @@ -520,6 +523,17 @@ sync_done: if (*ret < 0) break; } + + if (td->flags & TD_F_REGROW_LOGS) + regrow_logs(td); + + /* + * when doing I/O (not when verifying), + * check for any errors that are to be ignored + */ + if (!from_verify) + break; + return 0; case FIO_Q_QUEUED: /* @@ -557,6 +571,28 @@ static inline bool io_in_polling(struct thread_data *td) return !td->o.iodepth_batch_complete_min && !td->o.iodepth_batch_complete_max; } +/* + * Unlinks files from thread data fio_file structure + */ +static int unlink_all_files(struct thread_data *td) +{ + struct fio_file *f; + unsigned int i; + int ret = 0; + + for_each_file(td, f, i) { + if (f->filetype != FIO_TYPE_FILE) + continue; + ret = td_io_unlink_file(td, f); + if (ret) + break; + } + + if (ret) + td_verror(td, ret, "unlink_all_files"); + + return ret; +} /* * The main verify engine. Runs over the writes we previously submitted, @@ -589,6 +625,15 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) if (td->error) return; + /* + * verify_state needs to be reset before verification + * proceeds so that expected random seeds match actual + * random seeds in headers. The main loop will reset + * all random number generators if randrepeat is set. + */ + if (!td->o.rand_repeatable) + td_fill_verify_state_seed(td); + td_set_runstate(td, TD_VERIFYING); io_u = NULL; @@ -629,7 +674,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; while ((io_u = get_io_u(td)) != NULL) { - if (IS_ERR(io_u)) { + if (IS_ERR_OR_NULL(io_u)) { io_u = NULL; ret = FIO_Q_BUSY; goto reap; @@ -650,7 +695,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) continue; } else if (io_u->ddir == DDIR_TRIM) { io_u->ddir = DDIR_READ; - io_u_set(io_u, IO_U_F_TRIMMED); + io_u_set(td, io_u, IO_U_F_TRIMMED); break; } else if (io_u->ddir == DDIR_WRITE) { io_u->ddir = DDIR_READ; @@ -871,7 +916,14 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) if (flow_threshold_exceeded(td)) continue; - if (!td->o.time_based && bytes_issued >= total_bytes) + /* + * Break if we exceeded the bytes. The exception is time + * based runs, but we still need to break out of the loop + * for those to run verification, if enabled. + */ + if (bytes_issued >= total_bytes && + (!td->o.time_based || + (td->o.time_based && td->o.verify != VERIFY_NONE))) break; io_u = get_io_u(td); @@ -972,7 +1024,7 @@ reap: if (ret < 0) break; if (!ddir_rw_sum(td->bytes_done) && - !(td->io_ops->flags & FIO_NOIO)) + !td_ioengine_flagged(td, FIO_NOIO)) continue; if (!in_ramp_time(td) && should_check_rate(td)) { @@ -1053,6 +1105,41 @@ reap: bytes_done[i] = td->bytes_done[i] - bytes_done[i]; } +static void free_file_completion_logging(struct thread_data *td) +{ + struct fio_file *f; + unsigned int i; + + for_each_file(td, f, i) { + if (!f->last_write_comp) + break; + sfree(f->last_write_comp); + } +} + +static int init_file_completion_logging(struct thread_data *td, + unsigned int depth) +{ + struct fio_file *f; + unsigned int i; + + if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save) + return 0; + + for_each_file(td, f, i) { + f->last_write_comp = scalloc(depth, sizeof(uint64_t)); + if (!f->last_write_comp) + goto cleanup; + } + + return 0; + +cleanup: + free_file_completion_logging(td); + log_err("fio: failed to alloc write comp data\n"); + return 1; +} + static void cleanup_io_u(struct thread_data *td) { struct io_u *io_u; @@ -1071,8 +1158,7 @@ static void cleanup_io_u(struct thread_data *td) io_u_qexit(&td->io_u_freelist); io_u_qexit(&td->io_u_all); - if (td->last_write_comp) - sfree(td->last_write_comp); + free_file_completion_logging(td); } static int init_io_u(struct thread_data *td) @@ -1089,7 +1175,7 @@ static int init_io_u(struct thread_data *td) td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units; - if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td))) + if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) data_xfer = 0; err = 0; @@ -1109,7 +1195,7 @@ static int init_io_u(struct thread_data *td) * lucky and the allocator gives us an aligned address. */ if (td->o.odirect || td->o.mem_align || td->o.oatomic || - (td->io_ops->flags & FIO_RAWIO)) + td_ioengine_flagged(td, FIO_RAWIO)) td->orig_buffer_size += page_mask + td->o.mem_align; if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { @@ -1128,7 +1214,7 @@ static int init_io_u(struct thread_data *td) return 1; if (td->o.odirect || td->o.mem_align || td->o.oatomic || - (td->io_ops->flags & FIO_RAWIO)) + td_ioengine_flagged(td, FIO_RAWIO)) p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; else p = td->orig_buffer; @@ -1189,24 +1275,20 @@ static int init_io_u(struct thread_data *td) p += max_bs; } - if (td->o.verify != VERIFY_NONE) { - td->last_write_comp = scalloc(max_units, sizeof(uint64_t)); - if (!td->last_write_comp) { - log_err("fio: failed to alloc write comp data\n"); - return 1; - } - } + if (init_file_completion_logging(td, max_units)) + return 1; return 0; } static int switch_ioscheduler(struct thread_data *td) { +#ifdef FIO_HAVE_IOSCHED_SWITCH char tmp[256], tmp2[128]; FILE *f; int ret; - if (td->io_ops->flags & FIO_DISKLESSIO) + if (td_ioengine_flagged(td, FIO_DISKLESSIO)) return 0; sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); @@ -1249,6 +1331,14 @@ static int switch_ioscheduler(struct thread_data *td) */ tmp[strlen(tmp) - 1] = '\0'; + /* + * Write to "none" entry doesn't fail, so check the result here. + */ + if (!strcmp(tmp, "none")) { + log_err("fio: io scheduler is not tunable\n"); + fclose(f); + return 0; + } sprintf(tmp2, "[%s]", td->o.ioscheduler); if (!strstr(tmp, tmp2)) { @@ -1260,6 +1350,9 @@ static int switch_ioscheduler(struct thread_data *td) fclose(f); return 0; +#else + return 0; +#endif } static bool keep_running(struct thread_data *td) @@ -1293,7 +1386,7 @@ static bool keep_running(struct thread_data *td) if (diff < td_max_bs(td)) return false; - if (fio_files_done(td)) + if (fio_files_done(td) && !td->o.io_limit) return false; return true; @@ -1336,10 +1429,10 @@ static uint64_t do_dry_run(struct thread_data *td) break; io_u = get_io_u(td); - if (!io_u) + if (IS_ERR_OR_NULL(io_u)) break; - io_u_set(io_u, IO_U_F_FLIGHT); + io_u_set(td, io_u, IO_U_F_FLIGHT); io_u->error = 0; io_u->resid = 0; if (ddir_rw(acct_ddir(io_u))) @@ -1378,7 +1471,6 @@ static void *thread_main(void *data) struct thread_data *td = fd->td; struct thread_options *o = &td->o; struct sk_out *sk_out = fd->sk_out; - pthread_condattr_t attr; int clear_state; int ret; @@ -1403,12 +1495,18 @@ static void *thread_main(void *data) INIT_FLIST_HEAD(&td->verify_list); INIT_FLIST_HEAD(&td->trim_list); INIT_FLIST_HEAD(&td->next_rand_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); + ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); + if (ret) { + td_verror(td, ret, "mutex_cond_init_pshared"); + goto err; + } + ret = cond_init_pshared(&td->verify_cond); + if (ret) { + td_verror(td, ret, "mutex_cond_pshared"); + goto err; + } td_set_runstate(td, TD_INITIALIZED); dprint(FD_MUTEX, "up startup_mutex\n"); @@ -1430,6 +1528,14 @@ static void *thread_main(void *data) goto err; } + /* + * Do this early, we don't want the compress threads to be limited + * to the same CPUs as the IO workers. So do this before we set + * any potential CPU affinity + */ + if (iolog_compress_init(td, sk_out)) + goto err; + /* * If we have a gettimeofday() thread, make sure we exclude that * thread from this job @@ -1564,9 +1670,6 @@ static void *thread_main(void *data) goto err; } - if (iolog_compress_init(td, sk_out)) - goto err; - fio_verify_init(td); if (rate_submit_init(td, sk_out)) @@ -1576,6 +1679,7 @@ static void *thread_main(void *data) fio_getrusage(&td->ru_start); memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); + memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || o->ratemin[DDIR_TRIM]) { @@ -1594,9 +1698,13 @@ static void *thread_main(void *data) fio_gettime(&td->start, NULL); memcpy(&td->tv_cache, &td->start, sizeof(td->start)); - if (clear_state) + if (clear_state) { clear_io_state(td, 0); + if (o->unlink_each_loop && unlink_all_files(td)) + break; + } + prune_io_piece_log(td); if (td->o.verify_only && (td_write(td) || td_rw(td))) @@ -1641,7 +1749,7 @@ static void *thread_main(void *data) if (!o->do_verify || o->verify == VERIFY_NONE || - (td->io_ops->flags & FIO_UNIDIR)) + td_ioengine_flagged(td, FIO_UNIDIR)) continue; clear_io_state(td, 0); @@ -1664,6 +1772,8 @@ static void *thread_main(void *data) break; } + td_set_runstate(td, TD_FINISHING); + update_rusage_stat(td); td->ts.total_run_time = mtime_since_now(&td->epoch); td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; @@ -1676,7 +1786,7 @@ static void *thread_main(void *data) fio_unpin_memory(td); - fio_writeout_logs(td); + td_writeout_logs(td, true); iolog_compress_exit(td); rate_submit_exit(td); @@ -1701,6 +1811,15 @@ err: cgroup_shutdown(td, &cgroup_mnt); verify_free_state(td); + if (td->zone_state_index) { + int i; + + for (i = 0; i < DDIR_RWDIR_CNT; i++) + free(td->zone_state_index[i]); + free(td->zone_state_index); + td->zone_state_index = NULL; + } + if (fio_option_is_set(o, cpumask)) { ret = fio_cpuset_exit(&o->cpumask); if (ret) @@ -1728,43 +1847,11 @@ err: return (void *) (uintptr_t) 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(struct sk_out *sk_out, int shmid, int offset) -{ - struct fork_data *fd; - void *data, *ret; - -#if !defined(__hpux) && !defined(CONFIG_NO_SHM) - 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 - - fd = calloc(1, sizeof(*fd)); - fd->td = data + offset * sizeof(struct thread_data); - fd->sk_out = sk_out; - ret = thread_main(fd); - shmdt(data); - return (int) (uintptr_t) ret; -} - static void dump_td_info(struct thread_data *td) { - log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to " - "be stuck. Doing forceful exit of this job.\n", td->o.name, + log_err("fio: job '%s' (state=%d) hasn't exited in %lu seconds, it " + "appears to be stuck. Doing forceful exit of this job.\n", + td->o.name, td->runstate, (unsigned long) time_since_now(&td->terminate_time)); } @@ -1850,6 +1937,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, * move on. */ if (td->terminate && + td->runstate < TD_FSYNCING && time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { dump_td_info(td); td_set_runstate(td, TD_REAPED); @@ -1938,12 +2026,11 @@ static int fio_verify_load_state(struct thread_data *td) if (is_backend) { void *data; - int ver; ret = fio_server_get_verify_state(td->o.name, - td->thread_number - 1, &data, &ver); + td->thread_number - 1, &data); if (!ret) - verify_convert_assign_state(td, data, ver); + verify_assign_state(td, data); } else ret = verify_load_state(td, "local"); @@ -2097,6 +2184,7 @@ reap: struct thread_data *map[REAL_MAX_JOBS]; struct timeval this_start; int this_jobs = 0, left; + struct fork_data *fd; /* * create threads (TD_NOT_CREATED -> TD_CREATED) @@ -2130,7 +2218,7 @@ reap: if (waitee_running(td)) { dprint(FD_PROCESS, "%s: waiting for %s\n", td->o.name, td->o.wait_for); - break; + continue; } init_disk_util(td); @@ -2146,14 +2234,13 @@ reap: map[this_jobs++] = td; nr_started++; + fd = calloc(1, sizeof(*fd)); + fd->td = td; + fd->sk_out = sk_out; + if (td->o.use_thread) { - struct fork_data *fd; int ret; - fd = calloc(1, sizeof(*fd)); - fd->td = td; - fd->sk_out = sk_out; - dprint(FD_PROCESS, "will pthread_create\n"); ret = pthread_create(&td->thread, NULL, thread_main, fd); @@ -2173,8 +2260,9 @@ reap: dprint(FD_PROCESS, "will fork\n"); pid = fork(); if (!pid) { - int ret = fork_main(sk_out, shm_id, i); + int ret; + ret = (int)(uintptr_t)thread_main(fd); _exit(ret); } else if (i == fio_debug_jobno) *fio_debug_jobp = pid; @@ -2265,82 +2353,10 @@ reap: update_io_ticks(); } -static void wait_for_helper_thread_exit(void) -{ - void *ret; - - helper_exit = 1; - pthread_cond_signal(&helper_cond); - pthread_join(helper_thread, &ret); -} - static void free_disk_util(void) { disk_util_prune_entries(); - - pthread_cond_destroy(&helper_cond); -} - -static void *helper_thread_main(void *data) -{ - struct sk_out *sk_out = data; - int ret = 0; - - sk_out_assign(sk_out); - - fio_mutex_up(startup_mutex); - - while (!ret) { - uint64_t sec = DISK_UTIL_MSEC / 1000; - uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000; - struct timespec ts; - struct timeval tv; - - gettimeofday(&tv, NULL); - ts.tv_sec = tv.tv_sec + sec; - ts.tv_nsec = (tv.tv_usec * 1000) + nsec; - - if (ts.tv_nsec >= 1000000000ULL) { - ts.tv_nsec -= 1000000000ULL; - ts.tv_sec++; - } - - pthread_cond_timedwait(&helper_cond, &helper_lock, &ts); - - ret = update_io_ticks(); - - if (helper_do_stat) { - helper_do_stat = 0; - __show_running_run_stats(); - } - - if (!is_backend) - print_thread_status(); - } - - sk_out_drop(); - return NULL; -} - -static int create_helper_thread(struct sk_out *sk_out) -{ - int ret; - - setup_disk_util(); - - pthread_cond_init(&helper_cond, NULL); - pthread_mutex_init(&helper_lock, NULL); - - ret = pthread_create(&helper_thread, NULL, helper_thread_main, sk_out); - if (ret) { - log_err("Can't create helper 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; + helper_thread_destroy(); } int fio_backend(struct sk_out *sk_out) @@ -2373,14 +2389,14 @@ int fio_backend(struct sk_out *sk_out) set_genesis_time(); stat_init(); - create_helper_thread(sk_out); + helper_thread_create(startup_mutex, sk_out); cgroup_list = smalloc(sizeof(*cgroup_list)); INIT_FLIST_HEAD(cgroup_list); run_threads(sk_out); - wait_for_helper_thread_exit(); + helper_thread_exit(); if (!fio_abort) { __show_run_stats(); @@ -2388,13 +2404,15 @@ int fio_backend(struct sk_out *sk_out) for (i = 0; i < DDIR_RWDIR_CNT; i++) { struct io_log *log = agg_io_log[i]; - flush_log(log, 0); + flush_log(log, false); free_log(log); } } } for_each_td(td, i) { + if (td->ss.dur) + free(td->ss.cache); fio_options_free(td); if (td->rusage_sem) { fio_mutex_remove(td->rusage_sem);