X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=c167f908625f39fd8e344c9f75c705f7fd71adf4;hp=033d5a707f2a896f161eab84299e2b1bf6a23de0;hb=52a0b9ed71c3e929461e64b39059281948107071;hpb=b153f94a22d86ab342b7cc7a0f1fa5051311f2b4 diff --git a/backend.c b/backend.c index 033d5a70..c167f908 100644 --- a/backend.c +++ b/backend.c @@ -29,6 +29,7 @@ #include #include #include +#include #include "fio.h" #include "smalloc.h" @@ -47,12 +48,13 @@ #include "rate-submit.h" #include "helper_thread.h" #include "pshared.h" +#include "zone-dist.h" static struct fio_sem *startup_sem; static struct flist_head *cgroup_list; -static char *cgroup_mnt; +static struct cgroup_mnt *cgroup_mnt; static int exit_value; -static volatile int fio_abort; +static volatile bool fio_abort; static unsigned int nr_process = 0; static unsigned int nr_thread = 0; @@ -60,16 +62,22 @@ struct io_log *agg_io_log[DDIR_RWDIR_CNT]; int groupid = 0; unsigned int thread_number = 0; +unsigned int nr_segments = 0; +unsigned int cur_segment = 0; unsigned int stat_number = 0; -int shm_id = 0; int temp_stall_ts; unsigned long done_secs = 0; +#ifdef PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP +pthread_mutex_t overlap_check = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP; +#else +pthread_mutex_t overlap_check = PTHREAD_MUTEX_INITIALIZER; +#endif #define JOB_START_TIMEOUT (5 * 1000) static void sig_int(int sig) { - if (threads) { + if (nr_segments) { if (is_backend) fio_server_got_signal(sig); else { @@ -78,7 +86,7 @@ static void sig_int(int sig) exit_value = 128; } - fio_terminate_threads(TERMINATE_ALL); + fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); } } @@ -131,8 +139,8 @@ static bool __check_min_rate(struct thread_data *td, struct timespec *now, unsigned long long bytes = 0; unsigned long iops = 0; unsigned long spent; - unsigned long rate; - unsigned int ratemin = 0; + unsigned long long rate; + unsigned long long ratemin = 0; unsigned int rate_iops = 0; unsigned int rate_iops_min = 0; @@ -166,7 +174,7 @@ static bool __check_min_rate(struct thread_data *td, struct timespec *now, * check bandwidth specified rate */ if (bytes < td->rate_bytes[ddir]) { - log_err("%s: rate_min=%uB/s not met, only transferred %lluB\n", + log_err("%s: rate_min=%lluB/s not met, only transferred %lluB\n", td->o.name, ratemin, bytes); return true; } else { @@ -177,7 +185,7 @@ static bool __check_min_rate(struct thread_data *td, struct timespec *now, if (rate < ratemin || bytes < td->rate_bytes[ddir]) { - log_err("%s: rate_min=%uB/s not met, got %luB/s\n", + log_err("%s: rate_min=%lluB/s not met, got %lluB/s\n", td->o.name, ratemin, rate); return true; } @@ -198,7 +206,7 @@ static bool __check_min_rate(struct thread_data *td, struct timespec *now, if (rate < rate_iops_min || iops < td->rate_blocks[ddir]) { - log_err("%s: rate_iops_min=%u not met, got %lu IOPS\n", + log_err("%s: rate_iops_min=%u not met, got %llu IOPS\n", td->o.name, rate_iops_min, rate); return true; } @@ -216,12 +224,10 @@ static bool check_min_rate(struct thread_data *td, struct timespec *now) { bool ret = false; - if (td->bytes_done[DDIR_READ]) - ret |= __check_min_rate(td, now, DDIR_READ); - if (td->bytes_done[DDIR_WRITE]) - ret |= __check_min_rate(td, now, DDIR_WRITE); - if (td->bytes_done[DDIR_TRIM]) - ret |= __check_min_rate(td, now, DDIR_TRIM); + for_each_rw_ddir(ddir) { + if (td->bytes_done[ddir]) + ret |= __check_min_rate(td, now, ddir); + } return ret; } @@ -238,8 +244,6 @@ static void cleanup_pending_aio(struct thread_data *td) * get immediately available events, if any */ r = io_u_queued_complete(td, 0); - if (r < 0) - return; /* * now cancel remaining active events @@ -275,6 +279,7 @@ static bool fio_io_sync(struct thread_data *td, struct fio_file *f) io_u->ddir = DDIR_SYNC; io_u->file = f; + io_u_set(td, io_u, IO_U_F_NO_FILE_PUT); if (td_io_prep(td, io_u)) { put_io_u(td, io_u); @@ -308,7 +313,7 @@ requeue: static int fio_file_fsync(struct thread_data *td, struct fio_file *f) { - int ret; + int ret, ret2; if (fio_file_open(f)) return fio_io_sync(td, f); @@ -317,8 +322,10 @@ static int fio_file_fsync(struct thread_data *td, struct fio_file *f) return 1; ret = fio_io_sync(td, f); - td_io_close_file(td, f); - return ret; + ret2 = 0; + if (fio_file_open(f)) + ret2 = td_io_close_file(td, f); + return (ret || ret2); } static inline void __update_ts_cache(struct thread_data *td) @@ -386,7 +393,7 @@ static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, td_clear_error(td); *retptr = 0; return false; - } else if (td->o.fill_device && err == ENOSPC) { + } else if (td->o.fill_device && (err == ENOSPC || err == EDQUOT)) { /* * We expect to hit this error if * fill_device option is set. @@ -432,9 +439,7 @@ static int wait_for_completions(struct thread_data *td, struct timespec *time) if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) min_evts = 1; - if (time && (__should_check_rate(td, DDIR_READ) || - __should_check_rate(td, DDIR_WRITE) || - __should_check_rate(td, DDIR_TRIM))) + if (time && should_check_rate(td)) fio_gettime(time, NULL); do { @@ -456,14 +461,14 @@ int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, *ret = -io_u->error; clear_io_u(td, io_u); } else if (io_u->resid) { - int bytes = io_u->xfer_buflen - io_u->resid; + long long bytes = io_u->xfer_buflen - io_u->resid; struct fio_file *f = io_u->file; if (bytes_issued) *bytes_issued += bytes; if (!from_verify) - trim_io_piece(td, io_u); + trim_io_piece(io_u); /* * zero read, fail @@ -489,9 +494,7 @@ int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, requeue_io_u(td, &io_u); } else { sync_done: - if (comp_time && (__should_check_rate(td, DDIR_READ) || - __should_check_rate(td, DDIR_WRITE) || - __should_check_rate(td, DDIR_TRIM))) + if (comp_time && should_check_rate(td)) fio_gettime(comp_time, NULL); *ret = io_u_sync_complete(td, io_u); @@ -570,7 +573,7 @@ static int unlink_all_files(struct thread_data *td) /* * Check if io_u will overlap an in-flight IO in the queue */ -static bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) +bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) { bool overlap; struct io_u *check_io_u; @@ -587,7 +590,7 @@ static bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) if (x1 < y2 && y1 < x2) { overlap = true; - dprint(FD_IO, "in-flight overlap: %llu/%lu, %llu/%lu\n", + dprint(FD_IO, "in-flight overlap: %llu/%llu, %llu/%llu\n", x1, io_u->buflen, y1, check_io_u->buflen); break; @@ -834,7 +837,7 @@ static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) if (td->o.rate_process == RATE_PROCESS_POISSON) { uint64_t val, iops; - iops = bps / td->o.bs[ddir]; + iops = bps / td->o.min_bs[ddir]; val = (int64_t) (1000000 / iops) * -logf(__rand_0_1(&td->poisson_state[ddir])); if (val) { @@ -855,14 +858,47 @@ static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) return 0; } -static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) +static void init_thinktime(struct thread_data *td) +{ + if (td->o.thinktime_blocks_type == THINKTIME_BLOCKS_TYPE_COMPLETE) + td->thinktime_blocks_counter = td->io_blocks; + else + td->thinktime_blocks_counter = td->io_issues; + td->last_thinktime = td->epoch; + td->last_thinktime_blocks = 0; +} + +static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir, + struct timespec *time) { unsigned long long b; uint64_t total; int left; + struct timespec now; + bool stall = false; + + if (td->o.thinktime_iotime) { + fio_gettime(&now, NULL); + if (utime_since(&td->last_thinktime, &now) + >= td->o.thinktime_iotime + td->o.thinktime) { + stall = true; + } else if (!fio_option_is_set(&td->o, thinktime_blocks)) { + /* + * When thinktime_iotime is set and thinktime_blocks is + * not set, skip the thinktime_blocks check, since + * thinktime_blocks default value 1 does not work + * together with thinktime_iotime. + */ + return; + } + + } + + b = ddir_rw_sum(td->thinktime_blocks_counter); + if (b >= td->last_thinktime_blocks + td->o.thinktime_blocks) + stall = true; - b = ddir_rw_sum(td->io_blocks); - if (b % td->o.thinktime_blocks) + if (!stall) return; io_u_quiesce(td); @@ -892,7 +928,16 @@ static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) over = (usperop - total) / usperop * -bs; td->rate_io_issue_bytes[ddir] += (missed - over); + /* adjust for rate_process=poisson */ + td->last_usec[ddir] += total; } + + if (time && should_check_rate(td)) + fio_gettime(time, NULL); + + td->last_thinktime_blocks = b; + if (td->o.thinktime_iotime) + td->last_thinktime = now; } /* @@ -968,8 +1013,10 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) * 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. + * Jobs read from iolog do not use this stop condition. */ if (bytes_issued >= total_bytes && + !td->o.read_iolog_file && (!td->o.time_based || (td->o.time_based && td->o.verify != VERIFY_NONE))) break; @@ -1002,12 +1049,6 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { - if (!td->o.verify_pattern_bytes) { - io_u->rand_seed = __rand(&td->verify_state); - if (sizeof(int) != sizeof(long *)) - io_u->rand_seed *= __rand(&td->verify_state); - } - if (verify_state_should_stop(td, io_u)) { put_io_u(td, io_u); break; @@ -1035,8 +1076,8 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) log_io_piece(td, io_u); if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { - const unsigned long blen = io_u->xfer_buflen; - const enum fio_ddir ddir = acct_ddir(io_u); + const unsigned long long blen = io_u->xfer_buflen; + const enum fio_ddir __ddir = acct_ddir(io_u); if (td->error) break; @@ -1044,14 +1085,14 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) workqueue_enqueue(&td->io_wq, &io_u->work); ret = FIO_Q_QUEUED; - if (ddir_rw(ddir)) { - td->io_issues[ddir]++; - td->io_issue_bytes[ddir] += blen; - td->rate_io_issue_bytes[ddir] += blen; + if (ddir_rw(__ddir)) { + td->io_issues[__ddir]++; + td->io_issue_bytes[__ddir] += blen; + td->rate_io_issue_bytes[__ddir] += blen; } if (should_check_rate(td)) - td->rate_next_io_time[ddir] = usec_for_io(td, ddir); + td->rate_next_io_time[__ddir] = usec_for_io(td, __ddir); } else { ret = io_u_submit(td, io_u); @@ -1075,6 +1116,10 @@ reap: } if (ret < 0) break; + + if (ddir_rw(ddir) && td->o.thinktime) + handle_thinktime(td, ddir, &comp_time); + if (!ddir_rw_sum(td->bytes_done) && !td_ioengine_flagged(td, FIO_NOIO)) continue; @@ -1082,16 +1127,13 @@ reap: if (!in_ramp_time(td) && should_check_rate(td)) { if (check_min_rate(td, &comp_time)) { if (exitall_on_terminate || td->o.exitall_error) - fio_terminate_threads(td->groupid); + fio_terminate_threads(td->groupid, td->o.exit_what); td_verror(td, EIO, "check_min_rate"); break; } } if (!in_ramp_time(td) && td->o.latency_target) lat_target_check(td); - - if (ddir_rw(ddir) && td->o.thinktime) - handle_thinktime(td, ddir); } check_update_rusage(td); @@ -1099,7 +1141,7 @@ reap: if (td->trim_entries) log_err("fio: %lu trim entries leaked?\n", td->trim_entries); - if (td->o.fill_device && td->error == ENOSPC) { + if (td->o.fill_device && (td->error == ENOSPC || td->error == EDQUOT)) { td->error = 0; fio_mark_td_terminate(td); } @@ -1114,11 +1156,12 @@ reap: if (i) { ret = io_u_queued_complete(td, i); - if (td->o.fill_device && td->error == ENOSPC) + if (td->o.fill_device && + (td->error == ENOSPC || td->error == EDQUOT)) td->error = 0; } - if (should_fsync(td) && td->o.end_fsync) { + if (should_fsync(td) && (td->o.end_fsync || td->o.fsync_on_close)) { td_set_runstate(td, TD_FSYNCING); for_each_file(td, f, i) { @@ -1186,14 +1229,14 @@ static void cleanup_io_u(struct thread_data *td) if (td->io_ops->io_u_free) td->io_ops->io_u_free(td, io_u); - fio_memfree(io_u, sizeof(*io_u)); + fio_memfree(io_u, sizeof(*io_u), td_offload_overlap(td)); } free_io_mem(td); io_u_rexit(&td->io_u_requeues); - io_u_qexit(&td->io_u_freelist); - io_u_qexit(&td->io_u_all); + io_u_qexit(&td->io_u_freelist, false); + io_u_qexit(&td->io_u_all, td_offload_overlap(td)); free_file_completion_logging(td); } @@ -1201,30 +1244,85 @@ static void cleanup_io_u(struct thread_data *td) static int init_io_u(struct thread_data *td) { struct io_u *io_u; - unsigned int max_bs, min_write; int cl_align, i, max_units; - int data_xfer = 1, err; - char *p; + int err; max_units = td->o.iodepth; - max_bs = td_max_bs(td); - min_write = td->o.min_bs[DDIR_WRITE]; - td->orig_buffer_size = (unsigned long long) max_bs - * (unsigned long long) max_units; - - if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) - data_xfer = 0; err = 0; err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); - err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth); - err += !io_u_qinit(&td->io_u_all, td->o.iodepth); + err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth, false); + err += !io_u_qinit(&td->io_u_all, td->o.iodepth, td_offload_overlap(td)); if (err) { log_err("fio: failed setting up IO queues\n"); return 1; } + cl_align = os_cache_line_size(); + + for (i = 0; i < max_units; i++) { + void *ptr; + + if (td->terminate) + return 1; + + ptr = fio_memalign(cl_align, sizeof(*io_u), td_offload_overlap(td)); + if (!ptr) { + log_err("fio: unable to allocate aligned memory\n"); + return 1; + } + + io_u = ptr; + memset(io_u, 0, sizeof(*io_u)); + INIT_FLIST_HEAD(&io_u->verify_list); + dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); + + io_u->index = i; + io_u->flags = IO_U_F_FREE; + io_u_qpush(&td->io_u_freelist, io_u); + + /* + * io_u never leaves this stack, used for iteration of all + * io_u buffers. + */ + io_u_qpush(&td->io_u_all, io_u); + + if (td->io_ops->io_u_init) { + int ret = td->io_ops->io_u_init(td, io_u); + + if (ret) { + log_err("fio: failed to init engine data: %d\n", ret); + return 1; + } + } + } + + init_io_u_buffers(td); + + if (init_file_completion_logging(td, max_units)) + return 1; + + return 0; +} + +int init_io_u_buffers(struct thread_data *td) +{ + struct io_u *io_u; + unsigned long long max_bs, min_write; + int i, max_units; + int data_xfer = 1; + char *p; + + max_units = td->o.iodepth; + max_bs = td_max_bs(td); + min_write = td->o.min_bs[DDIR_WRITE]; + td->orig_buffer_size = (unsigned long long) max_bs + * (unsigned long long) max_units; + + if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) + data_xfer = 0; + /* * if we may later need to do address alignment, then add any * possible adjustment here so that we don't cause a buffer @@ -1236,7 +1334,7 @@ static int init_io_u(struct thread_data *td) td->orig_buffer_size += page_mask + td->o.mem_align; if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { - unsigned long bs; + unsigned long long bs; bs = td->orig_buffer_size + td->o.hugepage_size - 1; td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); @@ -1256,23 +1354,8 @@ static int init_io_u(struct thread_data *td) else p = td->orig_buffer; - cl_align = os_cache_line_size(); - for (i = 0; i < max_units; i++) { - void *ptr; - - if (td->terminate) - return 1; - - ptr = fio_memalign(cl_align, sizeof(*io_u)); - if (!ptr) { - log_err("fio: unable to allocate aligned memory\n"); - break; - } - - io_u = ptr; - memset(io_u, 0, sizeof(*io_u)); - INIT_FLIST_HEAD(&io_u->verify_list); + io_u = td->io_u_all.io_us[i]; dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); if (data_xfer) { @@ -1289,51 +1372,25 @@ static int init_io_u(struct thread_data *td) fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); } } - - io_u->index = i; - io_u->flags = IO_U_F_FREE; - io_u_qpush(&td->io_u_freelist, io_u); - - /* - * io_u never leaves this stack, used for iteration of all - * io_u buffers. - */ - io_u_qpush(&td->io_u_all, io_u); - - if (td->io_ops->io_u_init) { - int ret = td->io_ops->io_u_init(td, io_u); - - if (ret) { - log_err("fio: failed to init engine data: %d\n", ret); - return 1; - } - } - p += max_bs; } - if (init_file_completion_logging(td, max_units)) - return 1; - return 0; } +#ifdef FIO_HAVE_IOSCHED_SWITCH /* - * This function is Linux specific. + * These functions are Linux specific. * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux. */ -static int switch_ioscheduler(struct thread_data *td) +static int set_ioscheduler(struct thread_data *td, struct fio_file *file) { -#ifdef FIO_HAVE_IOSCHED_SWITCH char tmp[256], tmp2[128], *p; FILE *f; int ret; - if (td_ioengine_flagged(td, FIO_DISKLESSIO)) - return 0; - - assert(td->files && td->files[0]); - sprintf(tmp, "%s/queue/scheduler", td->files[0]->du->sysfs_root); + assert(file->du && file->du->sysfs_root); + sprintf(tmp, "%s/queue/scheduler", file->du->sysfs_root); f = fopen(tmp, "r+"); if (!f) { @@ -1386,7 +1443,7 @@ static int switch_ioscheduler(struct thread_data *td) sprintf(tmp2, "[%s]", td->o.ioscheduler); if (!strstr(tmp, tmp2)) { - log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); + log_err("fio: unable to set io scheduler to %s\n", td->o.ioscheduler); td_verror(td, EINVAL, "iosched_switch"); fclose(f); return 1; @@ -1394,11 +1451,55 @@ static int switch_ioscheduler(struct thread_data *td) fclose(f); return 0; +} + +static int switch_ioscheduler(struct thread_data *td) +{ + struct fio_file *f; + unsigned int i; + int ret = 0; + + if (td_ioengine_flagged(td, FIO_DISKLESSIO)) + return 0; + + assert(td->files && td->files[0]); + + for_each_file(td, f, i) { + + /* Only consider regular files and block device files */ + switch (f->filetype) { + case FIO_TYPE_FILE: + case FIO_TYPE_BLOCK: + /* + * Make sure that the device hosting the file could + * be determined. + */ + if (!f->du) + continue; + break; + case FIO_TYPE_CHAR: + case FIO_TYPE_PIPE: + default: + continue; + } + + ret = set_ioscheduler(td, f); + if (ret) + return ret; + } + + return 0; +} + #else + +static int switch_ioscheduler(struct thread_data *td) +{ return 0; -#endif } +#endif /* FIO_HAVE_IOSCHED_SWITCH */ + static bool keep_running(struct thread_data *td) { unsigned long long limit; @@ -1441,16 +1542,17 @@ static bool keep_running(struct thread_data *td) return false; } -static int exec_string(struct thread_options *o, const char *string, const char *mode) +static int exec_string(struct thread_options *o, const char *string, + const char *mode) { - size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; int ret; char *str; - str = malloc(newlen); - sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); + if (asprintf(&str, "%s > %s.%s.txt 2>&1", string, o->name, mode) < 0) + return -1; - log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); + log_info("%s : Saving output of %s in %s.%s.txt\n", o->name, mode, + o->name, mode); ret = system(str); if (ret == -1) log_err("fio: exec of cmd <%s> failed\n", str); @@ -1519,8 +1621,8 @@ static void *thread_main(void *data) struct sk_out *sk_out = fd->sk_out; uint64_t bytes_done[DDIR_RWDIR_CNT]; int deadlock_loop_cnt; - bool clear_state, did_some_io; - int ret; + bool clear_state; + int res, ret; sk_out_assign(sk_out); free(fd); @@ -1531,7 +1633,7 @@ static void *thread_main(void *data) } else td->pid = gettid(); - fio_local_clock_init(o->use_thread); + fio_local_clock_init(); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); @@ -1575,6 +1677,8 @@ static void *thread_main(void *data) goto err; } + td_zone_gen_index(td); + /* * 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 @@ -1673,9 +1777,15 @@ static void *thread_main(void *data) if (!init_iolog(td)) goto err; + if (td_io_init(td)) + goto err; + if (init_io_u(td)) goto err; + if (td->io_ops->post_init && td->io_ops->post_init(td)) + goto err; + if (o->verify_async && verify_async_init(td)) goto err; @@ -1686,6 +1796,7 @@ static void *thread_main(void *data) td_verror(td, errno, "ioprio_set"); goto err; } + td->ioprio = ioprio_value(o->ioprio_class, o->ioprio); } if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) @@ -1703,13 +1814,10 @@ static void *thread_main(void *data) if (!o->create_serialize && setup_files(td)) goto err; - if (td_io_init(td)) - goto err; - if (!init_random_map(td)) goto err; - if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) + if (o->exec_prerun && exec_string(o, o->exec_prerun, "prerun")) goto err; if (o->pre_read && !pre_read_files(td)) @@ -1726,6 +1834,8 @@ static void *thread_main(void *data) memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); + init_thinktime(td); + if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || o->ratemin[DDIR_TRIM]) { memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, @@ -1738,7 +1848,6 @@ static void *thread_main(void *data) memset(bytes_done, 0, sizeof(bytes_done)); clear_state = false; - did_some_io = false; while (keep_running(td)) { uint64_t verify_bytes; @@ -1816,9 +1925,6 @@ static void *thread_main(void *data) td_ioengine_flagged(td, FIO_UNIDIR)) continue; - if (ddir_rw_sum(bytes_done)) - did_some_io = true; - clear_io_state(td, 0); fio_gettime(&td->start, NULL); @@ -1840,25 +1946,25 @@ static void *thread_main(void *data) } /* - * If td ended up with no I/O when it should have had, - * then something went wrong unless FIO_NOIO or FIO_DISKLESSIO. - * (Are we not missing other flags that can be ignored ?) + * Acquire this lock if we were doing overlap checking in + * offload mode so that we don't clean up this job while + * another thread is checking its io_u's for overlap */ - if ((td->o.size || td->o.io_size) && !ddir_rw_sum(bytes_done) && - !did_some_io && !td->o.create_only && - !(td_ioengine_flagged(td, FIO_NOIO) || - td_ioengine_flagged(td, FIO_DISKLESSIO))) - log_err("%s: No I/O performed by %s, " - "perhaps try --debug=io option for details?\n", - td->o.name, td->io_ops->name); - + if (td_offload_overlap(td)) { + int res = pthread_mutex_lock(&overlap_check); + assert(res == 0); + } td_set_runstate(td, TD_FINISHING); + if (td_offload_overlap(td)) { + res = pthread_mutex_unlock(&overlap_check); + assert(res == 0); + } 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]; - td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; - td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; + for_each_rw_ddir(ddir) { + td->ts.io_bytes[ddir] = td->io_bytes[ddir]; + } if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && (td->o.verify != VERIFY_NONE && td_write(td))) @@ -1872,10 +1978,10 @@ static void *thread_main(void *data) rate_submit_exit(td); if (o->exec_postrun) - exec_string(o, o->exec_postrun, (const char *)"postrun"); + exec_string(o, o->exec_postrun, "postrun"); if (exitall_on_terminate || (o->exitall_error && td->error)) - fio_terminate_threads(td->groupid); + fio_terminate_threads(td->groupid, td->o.exit_what); err: if (td->error) @@ -1888,17 +1994,9 @@ err: close_and_free_files(td); cleanup_io_u(td); close_ioengine(td); - cgroup_shutdown(td, &cgroup_mnt); + 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; - } + td_zone_free_index(td); if (fio_option_is_set(o, cpumask)) { ret = fio_cpuset_exit(&o->cpumask); @@ -1911,6 +2009,8 @@ err: */ if (o->write_iolog_file) write_iolog_close(td); + if (td->io_log_rfile) + fclose(td->io_log_rfile); td_set_runstate(td, TD_EXITED); @@ -2030,10 +2130,11 @@ reaped: done_secs += mtime_since_now(&td->epoch) / 1000; profile_td_exit(td); + flow_exit_job(td); } if (*nr_running == cputhreads && !pending && realthreads) - fio_terminate_threads(TERMINATE_ALL); + fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); } static bool __check_trigger_file(void) @@ -2083,7 +2184,7 @@ void check_trigger_file(void) fio_clients_send_trigger(trigger_remote_cmd); else { verify_save_state(IO_LIST_ALL); - fio_terminate_threads(TERMINATE_ALL); + fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); exec_trigger(trigger_cmd); } } @@ -2103,8 +2204,16 @@ static int fio_verify_load_state(struct thread_data *td) td->thread_number - 1, &data); if (!ret) verify_assign_state(td, data); - } else - ret = verify_load_state(td, "local"); + } else { + char prefix[PATH_MAX]; + + if (aux_path) + sprintf(prefix, "%s%clocal", aux_path, + FIO_OS_PATH_SEPARATOR); + else + strcpy(prefix, "local"); + ret = verify_load_state(td, prefix); + } return ret; } @@ -2199,18 +2308,22 @@ static void run_threads(struct sk_out *sk_out) } if (output_format & FIO_OUTPUT_NORMAL) { - log_info("Starting "); + struct buf_output out; + + buf_output_init(&out); + __log_buf(&out, "Starting "); if (nr_thread) - log_info("%d thread%s", nr_thread, + __log_buf(&out, "%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); if (nr_process) { if (nr_thread) - log_info(" and "); - log_info("%d process%s", nr_process, + __log_buf(&out, " and "); + __log_buf(&out, "%d process%s", nr_process, nr_process > 1 ? "es" : ""); } - log_info("\n"); - log_info_flush(); + __log_buf(&out, "\n"); + log_info_buf(out.buf, out.buflen); + buf_output_free(&out); } todo = thread_number; @@ -2352,8 +2465,8 @@ reap: dprint(FD_MUTEX, "wait on startup_sem\n"); if (fio_sem_down_timeout(startup_sem, 10000)) { log_err("fio: job startup hung? exiting.\n"); - fio_terminate_threads(TERMINATE_ALL); - fio_abort = 1; + fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); + fio_abort = true; nr_started--; free(fd); break; @@ -2467,12 +2580,15 @@ int fio_backend(struct sk_out *sk_out) } startup_sem = fio_sem_init(FIO_SEM_LOCKED); + if (!sk_out) + is_local_backend = true; if (startup_sem == NULL) return 1; set_genesis_time(); stat_init(); - helper_thread_create(startup_sem, sk_out); + if (helper_thread_create(startup_sem, sk_out)) + log_err("fio: failed to create helper thread\n"); cgroup_list = smalloc(sizeof(*cgroup_list)); if (cgroup_list) @@ -2497,6 +2613,7 @@ int fio_backend(struct sk_out *sk_out) for_each_td(td, i) { steadystate_free(td); fio_options_free(td); + fio_dump_options_free(td); if (td->rusage_sem) { fio_sem_remove(td->rusage_sem); td->rusage_sem = NULL; @@ -2510,7 +2627,6 @@ int fio_backend(struct sk_out *sk_out) cgroup_kill(cgroup_list); sfree(cgroup_list); } - sfree(cgroup_mnt); fio_sem_remove(startup_sem); stat_exit();