X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=c8554bc1a102ec721b9f1d33b7e508680c0c094f;hp=3424a0982b534bc60b8ea037814427e9156b929f;hb=2a2804c95571c5bc9f3fc9e6e5b422fa90aa5670;hpb=66b6c5efc84f1a5fb25648f38409716682bad84c diff --git a/backend.c b/backend.c index 3424a098..c8554bc1 100644 --- a/backend.c +++ b/backend.c @@ -35,6 +35,7 @@ #include #include #include +#include #include "fio.h" #ifndef FIO_NO_HAVE_SHM_H @@ -48,12 +49,14 @@ #include "cgroup.h" #include "profile.h" #include "lib/rand.h" -#include "memalign.h" +#include "lib/memalign.h" #include "server.h" #include "lib/getrusage.h" #include "idletime.h" #include "err.h" -#include "lib/tp.h" +#include "workqueue.h" +#include "lib/mountcheck.h" +#include "rate-submit.h" static pthread_t helper_thread; static pthread_mutex_t helper_lock; @@ -98,7 +101,7 @@ static void sig_int(int sig) } } -static void sig_show_status(int sig) +void sig_show_status(int sig) { show_running_run_stats(); } @@ -141,8 +144,8 @@ static void set_sig_handlers(void) /* * Check if we are above the minimum rate given. */ -static int __check_min_rate(struct thread_data *td, struct timeval *now, - enum fio_ddir ddir) +static bool __check_min_rate(struct thread_data *td, struct timeval *now, + enum fio_ddir ddir) { unsigned long long bytes = 0; unsigned long iops = 0; @@ -155,13 +158,13 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, assert(ddir_rw(ddir)); if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) - return 0; + return false; /* * allow a 2 second settle period in the beginning */ if (mtime_since(&td->start, now) < 2000) - return 0; + return false; iops += td->this_io_blocks[ddir]; bytes += td->this_io_bytes[ddir]; @@ -175,16 +178,16 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { spent = mtime_since(&td->lastrate[ddir], now); if (spent < td->o.ratecycle) - return 0; + return false; - if (td->o.rate[ddir]) { + if (td->o.rate[ddir] || td->o.ratemin[ddir]) { /* * check bandwidth specified rate */ if (bytes < td->rate_bytes[ddir]) { log_err("%s: min rate %u not met\n", td->o.name, ratemin); - return 1; + return true; } else { if (spent) rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; @@ -196,7 +199,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, log_err("%s: min rate %u not met, got" " %luKB/sec\n", td->o.name, ratemin, rate); - return 1; + return true; } } } else { @@ -206,7 +209,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, if (iops < rate_iops) { log_err("%s: min iops rate %u not met\n", td->o.name, rate_iops); - return 1; + return true; } else { if (spent) rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; @@ -218,6 +221,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, log_err("%s: min iops rate %u not met," " got %lu\n", td->o.name, rate_iops_min, rate); + return true; } } } @@ -226,19 +230,18 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, td->rate_bytes[ddir] = bytes; td->rate_blocks[ddir] = iops; memcpy(&td->lastrate[ddir], now, sizeof(*now)); - return 0; + return false; } -static int check_min_rate(struct thread_data *td, struct timeval *now, - uint64_t *bytes_done) +static bool check_min_rate(struct thread_data *td, struct timeval *now) { - int ret = 0; + bool ret = false; - if (bytes_done[DDIR_READ]) + if (td->bytes_done[DDIR_READ]) ret |= __check_min_rate(td, now, DDIR_READ); - if (bytes_done[DDIR_WRITE]) + if (td->bytes_done[DDIR_WRITE]) ret |= __check_min_rate(td, now, DDIR_WRITE); - if (bytes_done[DDIR_TRIM]) + if (td->bytes_done[DDIR_TRIM]) ret |= __check_min_rate(td, now, DDIR_TRIM); return ret; @@ -255,7 +258,7 @@ static void cleanup_pending_aio(struct thread_data *td) /* * get immediately available events, if any */ - r = io_u_queued_complete(td, 0, NULL); + r = io_u_queued_complete(td, 0); if (r < 0) return; @@ -276,27 +279,27 @@ static void cleanup_pending_aio(struct thread_data *td) } if (td->cur_depth) - r = io_u_queued_complete(td, td->cur_depth, NULL); + r = io_u_queued_complete(td, td->cur_depth); } /* * 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) +static bool fio_io_sync(struct thread_data *td, struct fio_file *f) { struct io_u *io_u = __get_io_u(td); int ret; if (!io_u) - return 1; + return true; io_u->ddir = DDIR_SYNC; io_u->file = f; if (td_io_prep(td, io_u)) { put_io_u(td, io_u); - return 1; + return true; } requeue: @@ -304,25 +307,25 @@ requeue: if (ret < 0) { td_verror(td, io_u->error, "td_io_queue"); put_io_u(td, io_u); - return 1; + return true; } else if (ret == FIO_Q_QUEUED) { - if (io_u_queued_complete(td, 1, NULL) < 0) - return 1; + if (io_u_queued_complete(td, 1) < 0) + return true; } else if (ret == FIO_Q_COMPLETED) { if (io_u->error) { td_verror(td, io_u->error, "td_io_queue"); - return 1; + return true; } - if (io_u_sync_complete(td, io_u, NULL) < 0) - return 1; + if (io_u_sync_complete(td, io_u) < 0) + return true; } else if (ret == FIO_Q_BUSY) { if (td_io_commit(td)) - return 1; + return true; goto requeue; } - return 0; + return false; } static int fio_file_fsync(struct thread_data *td, struct fio_file *f) @@ -351,20 +354,37 @@ static inline void update_tv_cache(struct thread_data *td) __update_tv_cache(td); } -static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) +static inline bool runtime_exceeded(struct thread_data *td, struct timeval *t) { if (in_ramp_time(td)) - return 0; + return false; if (!td->o.timeout) - return 0; + return false; if (utime_since(&td->epoch, t) >= td->o.timeout) - return 1; + return true; - return 0; + return false; +} + +/* + * We need to update the runtime consistently in ms, but keep a running + * tally of the current elapsed time in microseconds for sub millisecond + * updates. + */ +static inline void update_runtime(struct thread_data *td, + unsigned long long *elapsed_us, + const enum fio_ddir ddir) +{ + if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) + return; + + td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; + elapsed_us[ddir] += utime_since_now(&td->start); + td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; } -static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, - int *retptr) +static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, + int *retptr) { int ret = *retptr; @@ -377,7 +397,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, eb = td_error_type(ddir, err); if (!(td->o.continue_on_error & (1 << eb))) - return 1; + return true; if (td_non_fatal_error(td, eb, err)) { /* @@ -387,7 +407,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, update_error_count(td, err); td_clear_error(td); *retptr = 0; - return 0; + return false; } else if (td->o.fill_device && err == ENOSPC) { /* * We expect to hit this error if @@ -395,18 +415,18 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, */ td_clear_error(td); fio_mark_td_terminate(td); - return 1; + return true; } else { /* * Stop the I/O in case of a fatal * error. */ update_error_count(td, err); - return 1; + return true; } } - return 0; + return false; } static void check_update_rusage(struct thread_data *td) @@ -418,8 +438,7 @@ static void check_update_rusage(struct thread_data *td) } } -static int wait_for_completions(struct thread_data *td, struct timeval *time, - uint64_t *bytes_done) +static int wait_for_completions(struct thread_data *td, struct timeval *time) { const int full = queue_full(td); int min_evts = 0; @@ -428,8 +447,8 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time, /* * if the queue is full, we MUST reap at least 1 event */ - min_evts = min(td->o.iodepth_batch_complete, td->cur_depth); - if (full && !min_evts) + min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); + if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) min_evts = 1; if (time && (__should_check_rate(td, DDIR_READ) || @@ -438,7 +457,7 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time, fio_gettime(time, NULL); do { - ret = io_u_queued_complete(td, min_evts, bytes_done); + ret = io_u_queued_complete(td, min_evts); if (ret < 0) break; } while (full && (td->cur_depth > td->o.iodepth_low)); @@ -446,13 +465,105 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time, return ret; } +int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, + enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, + struct timeval *comp_time) +{ + int ret2; + + switch (*ret) { + case FIO_Q_COMPLETED: + if (io_u->error) { + *ret = -io_u->error; + clear_io_u(td, io_u); + } else if (io_u->resid) { + int 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); + + /* + * zero read, fail + */ + if (!bytes) { + if (!from_verify) + unlog_io_piece(td, io_u); + td_verror(td, EIO, "full resid"); + put_io_u(td, io_u); + break; + } + + io_u->xfer_buflen = io_u->resid; + io_u->xfer_buf += bytes; + io_u->offset += bytes; + + if (ddir_rw(io_u->ddir)) + td->ts.short_io_u[io_u->ddir]++; + + f = io_u->file; + if (io_u->offset == f->real_file_size) + goto sync_done; + + 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))) + fio_gettime(comp_time, NULL); + + *ret = io_u_sync_complete(td, io_u); + if (*ret < 0) + break; + } + return 0; + case FIO_Q_QUEUED: + /* + * if the engine doesn't have a commit hook, + * the io_u is really queued. if it does have such + * a hook, it has to call io_u_queued() itself. + */ + if (td->io_ops->commit == NULL) + io_u_queued(td, io_u); + if (bytes_issued) + *bytes_issued += io_u->xfer_buflen; + break; + case FIO_Q_BUSY: + if (!from_verify) + unlog_io_piece(td, io_u); + requeue_io_u(td, &io_u); + ret2 = td_io_commit(td); + if (ret2 < 0) + *ret = ret2; + break; + default: + assert(*ret < 0); + td_verror(td, -(*ret), "td_io_queue"); + break; + } + + if (break_on_this_error(td, ddir, ret)) + return 1; + + return 0; +} + +static inline bool io_in_polling(struct thread_data *td) +{ + return !td->o.iodepth_batch_complete_min && + !td->o.iodepth_batch_complete_max; +} + /* * The main verify engine. Runs over the writes we previously submitted, * reads the blocks back in, and checks the crc/md5 of the data. */ static void do_verify(struct thread_data *td, uint64_t verify_bytes) { - uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; struct fio_file *f; struct io_u *io_u; int ret, min_events; @@ -483,7 +594,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) io_u = NULL; while (!td->terminate) { enum fio_ddir ddir; - int ret2, full; + int full; update_tv_cache(td); check_update_rusage(td); @@ -514,7 +625,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; } } else { - if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) + if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) break; while ((io_u = get_io_u(td)) != NULL) { @@ -539,7 +650,7 @@ 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->flags |= IO_U_F_TRIMMED; + io_u_set(io_u, IO_U_F_TRIMMED); break; } else if (io_u->ddir == DDIR_WRITE) { io_u->ddir = DDIR_READ; @@ -565,59 +676,12 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) io_u->end_io = verify_io_u; ddir = io_u->ddir; + if (!td->o.disable_slat) + fio_gettime(&io_u->start_time, NULL); ret = td_io_queue(td, io_u); - switch (ret) { - case FIO_Q_COMPLETED: - if (io_u->error) { - ret = -io_u->error; - clear_io_u(td, io_u); - } else if (io_u->resid) { - int bytes = io_u->xfer_buflen - io_u->resid; - /* - * zero read, fail - */ - if (!bytes) { - td_verror(td, EIO, "full resid"); - put_io_u(td, io_u); - break; - } - - io_u->xfer_buflen = io_u->resid; - io_u->xfer_buf += bytes; - io_u->offset += bytes; - - if (ddir_rw(io_u->ddir)) - td->ts.short_io_u[io_u->ddir]++; - - f = io_u->file; - if (io_u->offset == f->real_file_size) - goto sync_done; - - requeue_io_u(td, &io_u); - } else { -sync_done: - ret = io_u_sync_complete(td, io_u, bytes_done); - if (ret < 0) - break; - } - continue; - case FIO_Q_QUEUED: - break; - case FIO_Q_BUSY: - requeue_io_u(td, &io_u); - ret2 = td_io_commit(td); - if (ret2 < 0) - ret = ret2; - break; - default: - assert(ret < 0); - td_verror(td, -ret, "td_io_queue"); - break; - } - - if (break_on_this_error(td, ddir, &ret)) + if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) break; /* @@ -627,8 +691,8 @@ sync_done: */ reap: full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); - if (full || !td->o.iodepth_batch_complete) - ret = wait_for_completions(td, NULL, bytes_done); + if (full || io_in_polling(td)) + ret = wait_for_completions(td, NULL); if (ret < 0) break; @@ -640,7 +704,7 @@ reap: min_events = td->cur_depth; if (min_events) - ret = io_u_queued_complete(td, min_events, NULL); + ret = io_u_queued_complete(td, min_events); } else cleanup_pending_aio(td); @@ -649,20 +713,42 @@ reap: dprint(FD_VERIFY, "exiting loop\n"); } -static unsigned int exceeds_number_ios(struct thread_data *td) +static bool exceeds_number_ios(struct thread_data *td) { unsigned long long number_ios; if (!td->o.number_ios) - return 0; + return false; - number_ios = ddir_rw_sum(td->this_io_blocks); + number_ios = ddir_rw_sum(td->io_blocks); number_ios += td->io_u_queued + td->io_u_in_flight; - return number_ios >= td->o.number_ios; + return number_ios >= (td->o.number_ios * td->loops); } -static int io_bytes_exceeded(struct thread_data *td) +static bool io_issue_bytes_exceeded(struct thread_data *td) +{ + unsigned long long bytes, limit; + + if (td_rw(td)) + bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE]; + else if (td_write(td)) + bytes = td->io_issue_bytes[DDIR_WRITE]; + else if (td_read(td)) + bytes = td->io_issue_bytes[DDIR_READ]; + else + bytes = td->io_issue_bytes[DDIR_TRIM]; + + if (td->o.io_limit) + limit = td->o.io_limit; + else + limit = td->o.size; + + limit *= td->loops; + return bytes >= limit || exceeds_number_ios(td); +} + +static bool io_complete_bytes_exceeded(struct thread_data *td) { unsigned long long bytes, limit; @@ -680,9 +766,42 @@ static int io_bytes_exceeded(struct thread_data *td) else limit = td->o.size; + limit *= td->loops; return bytes >= limit || exceeds_number_ios(td); } +/* + * used to calculate the next io time for rate control + * + */ +static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) +{ + uint64_t secs, remainder, bps, bytes, iops; + + assert(!(td->flags & TD_F_CHILD)); + bytes = td->rate_io_issue_bytes[ddir]; + bps = td->rate_bps[ddir]; + + if (td->o.rate_process == RATE_PROCESS_POISSON) { + uint64_t val; + iops = bps / td->o.bs[ddir]; + val = (int64_t) (1000000 / iops) * + -logf(__rand_0_1(&td->poisson_state)); + if (val) { + dprint(FD_RATE, "poisson rate iops=%llu\n", + (unsigned long long) 1000000 / val); + } + td->last_usec += val; + return td->last_usec; + } else if (bps) { + secs = bytes / bps; + remainder = bytes % bps; + return remainder * 1000000 / bps + secs * 1000000; + } + + return 0; +} + /* * Main IO worker function. It retrieves io_u's to process and queues * and reaps them, checking for rate and errors along the way. @@ -691,7 +810,6 @@ static int io_bytes_exceeded(struct thread_data *td) */ static uint64_t do_io(struct thread_data *td) { - uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; unsigned int i; int ret = 0; uint64_t total_bytes, bytes_issued = 0; @@ -703,22 +821,33 @@ static uint64_t do_io(struct thread_data *td) lat_target_init(td); + total_bytes = td->o.size; + /* + * Allow random overwrite workloads to write up to io_limit + * before starting verification phase as 'size' doesn't apply. + */ + if (td_write(td) && td_random(td) && td->o.norandommap) + total_bytes = max(total_bytes, (uint64_t) td->o.io_limit); /* * If verify_backlog is enabled, we'll run the verify in this * handler as well. For that case, we may need up to twice the * amount of bytes. */ - total_bytes = td->o.size; if (td->o.verify != VERIFY_NONE && (td_write(td) && td->o.verify_backlog)) total_bytes += td->o.size; + /* In trimwrite mode, each byte is trimmed and then written, so + * allow total_bytes to be twice as big */ + if (td_trimwrite(td)) + total_bytes += td->total_io_size; + while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || - (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || + (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || td->o.time_based) { struct timeval comp_time; struct io_u *io_u; - int ret2, full; + int full; enum fio_ddir ddir; check_update_rusage(td); @@ -739,7 +868,7 @@ static uint64_t do_io(struct thread_data *td) if (flow_threshold_exceeded(td)) continue; - if (bytes_issued >= total_bytes) + if (!td->o.time_based && bytes_issued >= total_bytes) break; io_u = get_io_u(td); @@ -798,98 +927,54 @@ static uint64_t do_io(struct thread_data *td) !td->o.experimental_verify) log_io_piece(td, io_u); - ret = td_io_queue(td, io_u); - switch (ret) { - case FIO_Q_COMPLETED: - if (io_u->error) { - ret = -io_u->error; - unlog_io_piece(td, io_u); - clear_io_u(td, io_u); - } else if (io_u->resid) { - int bytes = io_u->xfer_buflen - io_u->resid; - struct fio_file *f = io_u->file; - - bytes_issued += bytes; + 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); - trim_io_piece(td, io_u); + if (td->error) + break; - /* - * zero read, fail - */ - if (!bytes) { - unlog_io_piece(td, io_u); - td_verror(td, EIO, "full resid"); - put_io_u(td, io_u); - break; - } + workqueue_enqueue(&td->io_wq, &io_u->work); + ret = FIO_Q_QUEUED; - io_u->xfer_buflen = io_u->resid; - io_u->xfer_buf += bytes; - io_u->offset += bytes; + if (ddir_rw(ddir)) { + td->io_issues[ddir]++; + td->io_issue_bytes[ddir] += blen; + td->rate_io_issue_bytes[ddir] += blen; + } - if (ddir_rw(io_u->ddir)) - td->ts.short_io_u[io_u->ddir]++; + if (should_check_rate(td)) + td->rate_next_io_time[ddir] = usec_for_io(td, ddir); - if (io_u->offset == f->real_file_size) - goto sync_done; + } else { + ret = td_io_queue(td, io_u); - requeue_io_u(td, &io_u); - } else { -sync_done: - if (__should_check_rate(td, DDIR_READ) || - __should_check_rate(td, DDIR_WRITE) || - __should_check_rate(td, DDIR_TRIM)) - fio_gettime(&comp_time, NULL); + if (should_check_rate(td)) + td->rate_next_io_time[ddir] = usec_for_io(td, ddir); + + if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) + break; - ret = io_u_sync_complete(td, io_u, bytes_done); - if (ret < 0) - break; - bytes_issued += io_u->xfer_buflen; - } - break; - case FIO_Q_QUEUED: /* - * if the engine doesn't have a commit hook, - * the io_u is really queued. if it does have such - * a hook, it has to call io_u_queued() itself. + * See if we need to complete some commands. Note that + * we can get BUSY even without IO queued, if the + * system is resource starved. */ - if (td->io_ops->commit == NULL) - io_u_queued(td, io_u); - bytes_issued += io_u->xfer_buflen; - break; - case FIO_Q_BUSY: - unlog_io_piece(td, io_u); - requeue_io_u(td, &io_u); - ret2 = td_io_commit(td); - if (ret2 < 0) - ret = ret2; - break; - default: - assert(ret < 0); - put_io_u(td, io_u); - break; - } - - if (break_on_this_error(td, ddir, &ret)) - break; - - /* - * See if we need to complete some commands. Note that we - * can get BUSY even without IO queued, if the system is - * resource starved. - */ reap: - full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); - if (full || !td->o.iodepth_batch_complete) - ret = wait_for_completions(td, &comp_time, bytes_done); + full = queue_full(td) || + (ret == FIO_Q_BUSY && td->cur_depth); + if (full || io_in_polling(td)) + ret = wait_for_completions(td, &comp_time); + } if (ret < 0) break; - if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO)) + if (!ddir_rw_sum(td->bytes_done) && + !(td->io_ops->flags & FIO_NOIO)) continue; - if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { - if (check_min_rate(td, &comp_time, bytes_done)) { - if (exitall_on_terminate) + 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); td_verror(td, EIO, "check_min_rate"); break; @@ -929,9 +1014,14 @@ reap: if (!td->error) { struct fio_file *f; - i = td->cur_depth; + if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { + workqueue_flush(&td->io_wq); + i = 0; + } else + i = td->cur_depth; + if (i) { - ret = io_u_queued_complete(td, i, bytes_done); + ret = io_u_queued_complete(td, i); if (td->o.fill_device && td->error == ENOSPC) td->error = 0; } @@ -956,7 +1046,7 @@ reap: if (!ddir_rw_sum(td->this_io_bytes)) td->done = 1; - return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; + return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; } static void cleanup_io_u(struct thread_data *td) @@ -1143,13 +1233,17 @@ static int switch_ioscheduler(struct thread_data *td) /* * Read back and check that the selected scheduler is now the default. */ + memset(tmp, 0, sizeof(tmp)); ret = fread(tmp, sizeof(tmp), 1, f); if (ferror(f) || ret < 0) { td_verror(td, errno, "fread"); fclose(f); return 1; } - tmp[sizeof(tmp) - 1] = '\0'; + /* + * either a list of io schedulers or "none\n" is expected. + */ + tmp[strlen(tmp) - 1] = '\0'; sprintf(tmp2, "[%s]", td->o.ioscheduler); @@ -1164,20 +1258,20 @@ static int switch_ioscheduler(struct thread_data *td) return 0; } -static int keep_running(struct thread_data *td) +static bool keep_running(struct thread_data *td) { unsigned long long limit; if (td->done) - return 0; + return false; if (td->o.time_based) - return 1; + return true; if (td->o.loops) { td->o.loops--; - return 1; + return true; } if (exceeds_number_ios(td)) - return 0; + return false; if (td->o.io_limit) limit = td->o.io_limit; @@ -1193,20 +1287,21 @@ static int keep_running(struct thread_data *td) */ diff = limit - ddir_rw_sum(td->io_bytes); if (diff < td_max_bs(td)) - return 0; + return false; if (fio_files_done(td)) - return 0; + return false; - return 1; + return true; } - return 0; + return false; } static int exec_string(struct thread_options *o, const char *string, const char *mode) { - int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; + size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; + int ret; char *str; str = malloc(newlen); @@ -1226,12 +1321,10 @@ static int exec_string(struct thread_options *o, const char *string, const char */ static uint64_t do_dry_run(struct thread_data *td) { - uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; - td_set_runstate(td, TD_RUNNING); while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || - (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) { + (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { struct io_u *io_u; int ret; @@ -1242,7 +1335,7 @@ static uint64_t do_dry_run(struct thread_data *td) if (!io_u) break; - io_u->flags |= IO_U_F_FLIGHT; + io_u_set(io_u, IO_U_F_FLIGHT); io_u->error = 0; io_u->resid = 0; if (ddir_rw(acct_ddir(io_u))) @@ -1258,11 +1351,11 @@ static uint64_t do_dry_run(struct thread_data *td) !td->o.experimental_verify) log_io_piece(td, io_u); - ret = io_u_sync_complete(td, io_u, bytes_done); + ret = io_u_sync_complete(td, io_u); (void) ret; } - return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; + return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; } /* @@ -1271,7 +1364,7 @@ static uint64_t do_dry_run(struct thread_data *td) */ static void *thread_main(void *data) { - unsigned long long elapsed; + unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; struct thread_data *td = data; struct thread_options *o = &td->o; pthread_condattr_t attr; @@ -1457,57 +1550,67 @@ static void *thread_main(void *data) goto err; } - if (td->flags & TD_F_COMPRESS_LOG) - tp_init(&td->tp_data); + if (iolog_compress_init(td)) + goto err; fio_verify_init(td); + if (rate_submit_init(td)) + goto err; + fio_gettime(&td->epoch, NULL); 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)); + + if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || + o->ratemin[DDIR_TRIM]) { + memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, + sizeof(td->bw_sample_time)); + memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, + sizeof(td->bw_sample_time)); + memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, + sizeof(td->bw_sample_time)); + } + clear_state = 0; while (keep_running(td)) { uint64_t verify_bytes; 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 (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || - o->ratemin[DDIR_TRIM]) { - memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, - sizeof(td->bw_sample_time)); - memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, - sizeof(td->bw_sample_time)); - memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, - sizeof(td->bw_sample_time)); - } - if (clear_state) - clear_io_state(td); + clear_io_state(td, 0); prune_io_piece_log(td); if (td->o.verify_only && (td_write(td) || td_rw(td))) verify_bytes = do_dry_run(td); - else + else { verify_bytes = do_io(td); + if (!verify_bytes) + fio_mark_td_terminate(td); + } clear_state = 1; + /* + * Make sure we've successfully updated the rusage stats + * before waiting on the stat mutex. Otherwise we could have + * the stat thread holding stat mutex and waiting for + * the rusage_sem, which would never get upped because + * this thread is waiting for the stat mutex. + */ + check_update_rusage(td); + fio_mutex_down(stat_mutex); - if (td_read(td) && td->io_bytes[DDIR_READ]) { - elapsed = mtime_since_now(&td->start); - td->ts.runtime[DDIR_READ] += elapsed; - } - if (td_write(td) && td->io_bytes[DDIR_WRITE]) { - elapsed = mtime_since_now(&td->start); - td->ts.runtime[DDIR_WRITE] += elapsed; - } - if (td_trim(td) && td->io_bytes[DDIR_TRIM]) { - elapsed = mtime_since_now(&td->start); - td->ts.runtime[DDIR_TRIM] += elapsed; - } + if (td_read(td) && td->io_bytes[DDIR_READ]) + update_runtime(td, elapsed_us, DDIR_READ); + if (td_write(td) && td->io_bytes[DDIR_WRITE]) + update_runtime(td, elapsed_us, DDIR_WRITE); + if (td_trim(td) && td->io_bytes[DDIR_TRIM]) + update_runtime(td, elapsed_us, DDIR_TRIM); fio_gettime(&td->start, NULL); fio_mutex_up(stat_mutex); @@ -1519,14 +1622,19 @@ static void *thread_main(void *data) (td->io_ops->flags & FIO_UNIDIR)) continue; - clear_io_state(td); + clear_io_state(td, 0); fio_gettime(&td->start, NULL); do_verify(td, verify_bytes); + /* + * See comment further up for why this is done here. + */ + check_update_rusage(td); + fio_mutex_down(stat_mutex); - td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start); + update_runtime(td, elapsed_us, DDIR_READ); fio_gettime(&td->start, NULL); fio_mutex_up(stat_mutex); @@ -1541,28 +1649,20 @@ static void *thread_main(void *data) td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && - (td->o.verify != VERIFY_NONE && td_write(td))) { - struct all_io_list *state; - size_t sz; - - state = get_all_io_list(td->thread_number, &sz); - if (state) { - __verify_save_state(state, "local"); - free(state); - } - } + (td->o.verify != VERIFY_NONE && td_write(td))) + verify_save_state(td->thread_number); fio_unpin_memory(td); fio_writeout_logs(td); - if (td->flags & TD_F_COMPRESS_LOG) - tp_exit(&td->tp_data); + iolog_compress_exit(td); + rate_submit_exit(td); if (o->exec_postrun) exec_string(o, o->exec_postrun, (const char *)"postrun"); - if (exitall_on_terminate) + if (exitall_on_terminate || (o->exitall_error && td->error)) fio_terminate_threads(td->groupid); err: @@ -1754,29 +1854,29 @@ reaped: fio_terminate_threads(TERMINATE_ALL); } -static int __check_trigger_file(void) +static bool __check_trigger_file(void) { struct stat sb; if (!trigger_file) - return 0; + return false; if (stat(trigger_file, &sb)) - return 0; + return false; if (unlink(trigger_file) < 0) log_err("fio: failed to unlink %s: %s\n", trigger_file, strerror(errno)); - return 1; + return true; } -static int trigger_timedout(void) +static bool trigger_timedout(void) { if (trigger_timeout) return time_since_genesis() >= trigger_timeout; - return 0; + return false; } void exec_trigger(const char *cmd) @@ -1797,7 +1897,7 @@ void check_trigger_file(void) if (nr_clients) fio_clients_send_trigger(trigger_remote_cmd); else { - verify_save_state(); + verify_save_state(IO_LIST_ALL); fio_terminate_threads(TERMINATE_ALL); exec_trigger(trigger_cmd); } @@ -1813,11 +1913,12 @@ 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); + td->thread_number - 1, &data, &ver); if (!ret) - verify_convert_assign_state(td, data); + verify_convert_assign_state(td, data, ver); } else ret = verify_load_state(td, "local"); @@ -1831,6 +1932,27 @@ static void do_usleep(unsigned int usecs) usleep(usecs); } +static bool check_mount_writes(struct thread_data *td) +{ + struct fio_file *f; + unsigned int i; + + if (!td_write(td) || td->o.allow_mounted_write) + return false; + + for_each_file(td, f, i) { + if (f->filetype != FIO_TYPE_BD) + continue; + if (device_is_mounted(f->file_name)) + goto mounted; + } + + return false; +mounted: + log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name); + return true; +} + /* * Main function for kicking off and reaping jobs, as needed. */ @@ -1849,13 +1971,15 @@ static void run_threads(void) nr_thread = nr_process = 0; for_each_td(td, i) { + if (check_mount_writes(td)) + return; if (td->o.use_thread) nr_thread++; else nr_process++; } - if (output_format == FIO_OUTPUT_NORMAL) { + if (output_format & FIO_OUTPUT_NORMAL) { log_info("Starting "); if (nr_thread) log_info("%d thread%s", nr_thread, @@ -1993,7 +2117,7 @@ reap: *fio_debug_jobp = pid; } dprint(FD_MUTEX, "wait on startup_mutex\n"); - if (fio_mutex_down_timeout(startup_mutex, 10)) { + if (fio_mutex_down_timeout(startup_mutex, 10000)) { log_err("fio: job startup hung? exiting.\n"); fio_terminate_threads(TERMINATE_ALL); fio_abort = 1; @@ -2096,8 +2220,12 @@ static void free_disk_util(void) static void *helper_thread_main(void *data) { + struct backend_data *d = data; int ret = 0; + if (d) + pthread_setspecific(d->key, d->ptr); + fio_mutex_up(startup_mutex); while (!ret) { @@ -2109,7 +2237,8 @@ static void *helper_thread_main(void *data) gettimeofday(&tv, NULL); ts.tv_sec = tv.tv_sec + sec; ts.tv_nsec = (tv.tv_usec * 1000) + nsec; - if (ts.tv_nsec > 1000000000ULL) { + + if (ts.tv_nsec >= 1000000000ULL) { ts.tv_nsec -= 1000000000ULL; ts.tv_sec++; } @@ -2130,7 +2259,7 @@ static void *helper_thread_main(void *data) return NULL; } -static int create_helper_thread(void) +static int create_helper_thread(struct backend_data *data) { int ret; @@ -2139,7 +2268,7 @@ static int create_helper_thread(void) pthread_cond_init(&helper_cond, NULL); pthread_mutex_init(&helper_lock, NULL); - ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL); + ret = pthread_create(&helper_thread, NULL, helper_thread_main, data); if (ret) { log_err("Can't create helper thread: %s\n", strerror(ret)); return 1; @@ -2151,7 +2280,7 @@ static int create_helper_thread(void) return 0; } -int fio_backend(void) +int fio_backend(struct backend_data *data) { struct thread_data *td; int i; @@ -2181,7 +2310,7 @@ int fio_backend(void) set_genesis_time(); stat_init(); - create_helper_thread(); + create_helper_thread(data); cgroup_list = smalloc(sizeof(*cgroup_list)); INIT_FLIST_HEAD(cgroup_list); @@ -2196,7 +2325,7 @@ int fio_backend(void) for (i = 0; i < DDIR_RWDIR_CNT; i++) { struct io_log *log = agg_io_log[i]; - flush_log(log); + flush_log(log, 0); free_log(log); } }