X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=5f862abeead5ba6fc16c82a4112a33e0f32fc7d5;hp=3b42c4df4dd285678cb09433fa7b4d26c7fb2511;hb=4896473eb0c69d870c41a6a679765d2d5f03e199;hpb=d0b937ed8841484c2d651919943e8e4c70dfd460 diff --git a/backend.c b/backend.c index 3b42c4df..5f862abe 100644 --- a/backend.c +++ b/backend.c @@ -34,10 +34,12 @@ #include #include #include -#include #include #include "fio.h" +#ifndef FIO_NO_HAVE_SHM_H +#include +#endif #include "hash.h" #include "smalloc.h" #include "verify.h" @@ -48,6 +50,8 @@ #include "lib/rand.h" #include "memalign.h" #include "server.h" +#include "lib/getrusage.h" +#include "idletime.h" static pthread_t disk_util_thread; static struct fio_mutex *disk_thread_mutex; @@ -62,6 +66,7 @@ struct io_log *agg_io_log[DDIR_RWDIR_CNT]; int groupid = 0; unsigned int thread_number = 0; +unsigned int stat_number = 0; unsigned int nr_process = 0; unsigned int nr_thread = 0; int shm_id = 0; @@ -213,7 +218,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, } static int check_min_rate(struct thread_data *td, struct timeval *now, - unsigned long *bytes_done) + uint64_t *bytes_done) { int ret = 0; @@ -317,6 +322,21 @@ requeue: return 0; } +static int fio_file_fsync(struct thread_data *td, struct fio_file *f) +{ + int ret; + + if (fio_file_open(f)) + return fio_io_sync(td, f); + + if (td_io_open_file(td, f)) + return 1; + + ret = fio_io_sync(td, f); + td_io_close_file(td, f); + return ret; +} + static inline void __update_tv_cache(struct thread_data *td) { fio_gettime(&td->tv_cache, NULL); @@ -386,12 +406,22 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, return 0; } +static void check_update_rusage(struct thread_data *td) +{ + if (td->update_rusage) { + td->update_rusage = 0; + update_rusage_stat(td); + fio_mutex_up(td->rusage_sem); + } +} + /* * 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) +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; @@ -412,6 +442,8 @@ static void do_verify(struct thread_data *td) break; } + check_update_rusage(td); + if (td->error) return; @@ -419,9 +451,11 @@ static void do_verify(struct thread_data *td) io_u = NULL; while (!td->terminate) { + enum fio_ddir ddir; int ret2, full; update_tv_cache(td); + check_update_rusage(td); if (runtime_exceeded(td, &td->tv_cache)) { __update_tv_cache(td); @@ -434,18 +468,53 @@ static void do_verify(struct thread_data *td) if (flow_threshold_exceeded(td)) continue; - io_u = __get_io_u(td); - if (!io_u) - break; + if (!td->o.experimental_verify) { + io_u = __get_io_u(td); + if (!io_u) + break; - if (get_next_verify(td, io_u)) { - put_io_u(td, io_u); - break; - } + if (get_next_verify(td, io_u)) { + put_io_u(td, io_u); + break; + } - if (td_io_prep(td, io_u)) { - put_io_u(td, io_u); - break; + if (td_io_prep(td, io_u)) { + put_io_u(td, io_u); + break; + } + } else { + if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) + break; + + while ((io_u = get_io_u(td)) != NULL) { + /* + * We are only interested in the places where + * we wrote or trimmed IOs. Turn those into + * reads for verification purposes. + */ + if (io_u->ddir == DDIR_READ) { + /* + * Pretend we issued it for rwmix + * accounting + */ + td->io_issues[DDIR_READ]++; + put_io_u(td, io_u); + continue; + } else if (io_u->ddir == DDIR_TRIM) { + io_u->ddir = DDIR_READ; + io_u->flags |= IO_U_F_TRIMMED; + break; + } else if (io_u->ddir == DDIR_WRITE) { + io_u->ddir = DDIR_READ; + break; + } else { + put_io_u(td, io_u); + continue; + } + } + + if (!io_u) + break; } if (td->o.verify_async) @@ -453,6 +522,8 @@ static void do_verify(struct thread_data *td) else io_u->end_io = verify_io_u; + ddir = io_u->ddir; + ret = td_io_queue(td, io_u); switch (ret) { case FIO_Q_COMPLETED: @@ -485,7 +556,7 @@ static void do_verify(struct thread_data *td) requeue_io_u(td, &io_u); } else { sync_done: - ret = io_u_sync_complete(td, io_u, NULL); + ret = io_u_sync_complete(td, io_u, bytes_done); if (ret < 0) break; } @@ -504,7 +575,7 @@ sync_done: break; } - if (break_on_this_error(td, io_u->ddir, &ret)) + if (break_on_this_error(td, ddir, &ret)) break; /* @@ -528,7 +599,7 @@ sync_done: * and do the verification on them through * the callback handler */ - if (io_u_queued_complete(td, min_events, NULL) < 0) { + if (io_u_queued_complete(td, min_events, bytes_done) < 0) { ret = -1; break; } @@ -538,6 +609,8 @@ sync_done: break; } + check_update_rusage(td); + if (!td->error) { min_events = td->cur_depth; @@ -570,9 +643,12 @@ static int io_bytes_exceeded(struct thread_data *td) /* * Main IO worker function. It retrieves io_u's to process and queues * and reaps them, checking for rate and errors along the way. + * + * Returns number of bytes written and trimmed. */ -static void do_io(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; @@ -585,13 +661,14 @@ static void do_io(struct thread_data *td) (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || td->o.time_based) { struct timeval comp_time; - unsigned long bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; int min_evts = 0; struct io_u *io_u; int ret2, full; enum fio_ddir ddir; - if (td->terminate) + check_update_rusage(td); + + if (td->terminate || td->done) break; update_tv_cache(td); @@ -726,7 +803,7 @@ sync_done: if (ret < 0) break; - if (!ddir_rw_sum(bytes_done)) + if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO)) continue; if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { @@ -755,6 +832,8 @@ sync_done: } } + check_update_rusage(td); + if (td->trim_entries) log_err("fio: %d trim entries leaked?\n", td->trim_entries); @@ -767,7 +846,7 @@ sync_done: i = td->cur_depth; if (i) { - ret = io_u_queued_complete(td, i, NULL); + ret = io_u_queued_complete(td, i, bytes_done); if (td->o.fill_device && td->error == ENOSPC) td->error = 0; } @@ -776,9 +855,11 @@ sync_done: td_set_runstate(td, TD_FSYNCING); for_each_file(td, f, i) { - if (!fio_file_open(f)) + if (!fio_file_fsync(td, f)) continue; - fio_io_sync(td, f); + + log_err("fio: end_fsync failed for file %s\n", + f->file_name); } } } else @@ -789,6 +870,8 @@ sync_done: */ if (!ddir_rw_sum(td->this_io_bytes)) td->done = 1; + + return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; } static void cleanup_io_u(struct thread_data *td) @@ -800,6 +883,10 @@ static void cleanup_io_u(struct thread_data *td) io_u = flist_entry(entry, struct io_u, list); flist_del(&io_u->list); + + if (td->io_ops->io_u_free) + td->io_ops->io_u_free(td, io_u); + fio_memfree(io_u, sizeof(*io_u)); } @@ -815,8 +902,7 @@ static int init_io_u(struct thread_data *td) char *p; max_units = td->o.iodepth; - max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); - max_bs = max(td->o.max_bs[DDIR_TRIM], max_bs); + 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; @@ -882,6 +968,16 @@ static int init_io_u(struct thread_data *td) io_u->index = i; io_u->flags = IO_U_F_FREE; flist_add(&io_u->list, &td->io_u_freelist); + + 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; } @@ -955,8 +1051,19 @@ static int keep_running(struct thread_data *td) return 1; } - if (ddir_rw_sum(td->io_bytes) < td->o.size) + if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) { + uint64_t diff; + + /* + * If the difference is less than the minimum IO size, we + * are done. + */ + diff = td->o.size - ddir_rw_sum(td->io_bytes); + if (diff < td_max_bs(td)) + return 0; + return 1; + } return 0; } @@ -985,15 +1092,18 @@ static void *thread_main(void *data) { unsigned long long elapsed; struct thread_data *td = data; + struct thread_options *o = &td->o; pthread_condattr_t attr; int clear_state; - if (!td->o.use_thread) { + if (!o->use_thread) { setsid(); td->pid = getpid(); } else td->pid = gettid(); + fio_local_clock_init(o->use_thread); + dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); INIT_FLIST_HEAD(&td->io_u_freelist); @@ -1003,6 +1113,7 @@ static void *thread_main(void *data) INIT_FLIST_HEAD(&td->io_hist_list); 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; @@ -1027,11 +1138,11 @@ static void *thread_main(void *data) * A new gid requires privilege, so we need to do this before setting * the uid. */ - if (td->o.gid != -1U && setgid(td->o.gid)) { + if (o->gid != -1U && setgid(o->gid)) { td_verror(td, errno, "setgid"); goto err; } - if (td->o.uid != -1U && setuid(td->o.uid)) { + if (o->uid != -1U && setuid(o->uid)) { td_verror(td, errno, "setuid"); goto err; } @@ -1040,21 +1151,23 @@ static void *thread_main(void *data) * 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); + if (o->gtod_cpu) + fio_cpu_clear(&o->cpumask, 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; + if (o->cpumask_set) { + if (fio_setaffinity(td->pid, o->cpumask) == -1) { + td_verror(td, errno, "cpu_set_affinity"); + goto err; + } } -#ifdef FIO_HAVE_LIBNUMA +#ifdef CONFIG_LIBNUMA /* numa node setup */ - if (td->o.numa_cpumask_set || td->o.numa_memmask_set) { + if (o->numa_cpumask_set || o->numa_memmask_set) { int ret; if (numa_available() < 0) { @@ -1062,8 +1175,8 @@ static void *thread_main(void *data) goto err; } - if (td->o.numa_cpumask_set) { - ret = numa_run_on_node_mask(td->o.numa_cpunodesmask); + if (o->numa_cpumask_set) { + ret = numa_run_on_node_mask(o->numa_cpunodesmask); if (ret == -1) { td_verror(td, errno, \ "numa_run_on_node_mask failed\n"); @@ -1071,20 +1184,20 @@ static void *thread_main(void *data) } } - if (td->o.numa_memmask_set) { + if (o->numa_memmask_set) { - switch (td->o.numa_mem_mode) { + switch (o->numa_mem_mode) { case MPOL_INTERLEAVE: - numa_set_interleave_mask(td->o.numa_memnodesmask); + numa_set_interleave_mask(o->numa_memnodesmask); break; case MPOL_BIND: - numa_set_membind(td->o.numa_memnodesmask); + numa_set_membind(o->numa_memnodesmask); break; case MPOL_LOCAL: numa_set_localalloc(); break; case MPOL_PREFERRED: - numa_set_preferred(td->o.numa_mem_prefer_node); + numa_set_preferred(o->numa_mem_prefer_node); break; case MPOL_DEFAULT: default: @@ -1095,6 +1208,9 @@ static void *thread_main(void *data) } #endif + if (fio_pin_memory(td)) + goto err; + /* * May alter parameters that init_io_u() will use, so we need to * do this first. @@ -1105,7 +1221,7 @@ static void *thread_main(void *data) if (init_io_u(td)) goto err; - if (td->o.verify_async && verify_async_init(td)) + if (o->verify_async && verify_async_init(td)) goto err; if (td->ioprio_set) { @@ -1115,19 +1231,19 @@ static void *thread_main(void *data) } } - if (td->o.cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) + if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) goto err; errno = 0; - if (nice(td->o.nice) == -1 && errno != 0) { + if (nice(o->nice) == -1 && errno != 0) { td_verror(td, errno, "nice"); goto err; } - if (td->o.ioscheduler && switch_ioscheduler(td)) + if (o->ioscheduler && switch_ioscheduler(td)) goto err; - if (!td->o.create_serialize && setup_files(td)) + if (!o->create_serialize && setup_files(td)) goto err; if (td_io_init(td)) @@ -1136,28 +1252,29 @@ static void *thread_main(void *data) if (init_random_map(td)) goto err; - if (td->o.exec_prerun) { - if (exec_string(td->o.exec_prerun)) - goto err; - } + if (o->exec_prerun && exec_string(o->exec_prerun)) + goto err; - if (td->o.pre_read) { + if (o->pre_read) { if (pre_read_files(td) < 0) goto err; } - fio_gettime(&td->epoch, NULL); - getrusage(RUSAGE_SELF, &td->ru_start); + fio_verify_init(td); + fio_gettime(&td->epoch, NULL); + fio_getrusage(&td->ru_start); 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 (td->o.ratemin[DDIR_READ] || td->o.ratemin[DDIR_WRITE] || - td->o.ratemin[DDIR_TRIM]) { + 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, @@ -1171,7 +1288,7 @@ static void *thread_main(void *data) prune_io_piece_log(td); - do_io(td); + verify_bytes = do_io(td); clear_state = 1; @@ -1191,8 +1308,8 @@ static void *thread_main(void *data) if (td->error || td->terminate) break; - if (!td->o.do_verify || - td->o.verify == VERIFY_NONE || + if (!o->do_verify || + o->verify == VERIFY_NONE || (td->io_ops->flags & FIO_UNIDIR)) continue; @@ -1200,7 +1317,7 @@ static void *thread_main(void *data) fio_gettime(&td->start, NULL); - do_verify(td); + do_verify(td, verify_bytes); td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); @@ -1217,46 +1334,48 @@ static void *thread_main(void *data) td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; + fio_unpin_memory(td); + fio_mutex_down(writeout_mutex); if (td->bw_log) { - if (td->o.bw_log_file) { + if (o->bw_log_file) { finish_log_named(td, td->bw_log, - td->o.bw_log_file, "bw"); + o->bw_log_file, "bw"); } else finish_log(td, td->bw_log, "bw"); } if (td->lat_log) { - if (td->o.lat_log_file) { + if (o->lat_log_file) { finish_log_named(td, td->lat_log, - td->o.lat_log_file, "lat"); + o->lat_log_file, "lat"); } else finish_log(td, td->lat_log, "lat"); } if (td->slat_log) { - if (td->o.lat_log_file) { + if (o->lat_log_file) { finish_log_named(td, td->slat_log, - td->o.lat_log_file, "slat"); + o->lat_log_file, "slat"); } else finish_log(td, td->slat_log, "slat"); } if (td->clat_log) { - if (td->o.lat_log_file) { + if (o->lat_log_file) { finish_log_named(td, td->clat_log, - td->o.lat_log_file, "clat"); + o->lat_log_file, "clat"); } else finish_log(td, td->clat_log, "clat"); } if (td->iops_log) { - if (td->o.iops_log_file) { + if (o->iops_log_file) { finish_log_named(td, td->iops_log, - td->o.iops_log_file, "iops"); + 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 (o->exec_postrun) + exec_string(o->exec_postrun); if (exitall_on_terminate) fio_terminate_threads(td->groupid); @@ -1266,16 +1385,16 @@ err: log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, td->verror); - if (td->o.verify_async) + if (o->verify_async) verify_async_exit(td); close_and_free_files(td); - close_ioengine(td); cleanup_io_u(td); + close_ioengine(td); cgroup_shutdown(td, &cgroup_mnt); - if (td->o.cpumask_set) { - int ret = fio_cpuset_exit(&td->o.cpumask); + if (o->cpumask_set) { + int ret = fio_cpuset_exit(&o->cpumask); td_verror(td, ret, "fio_cpuset_exit"); } @@ -1283,9 +1402,12 @@ err: /* * do this very late, it will log file closing as well */ - if (td->o.write_iolog_file) + if (o->write_iolog_file) write_iolog_close(td); + fio_mutex_remove(td->rusage_sem); + td->rusage_sem = NULL; + td_set_runstate(td, TD_EXITED); return (void *) (uintptr_t) td->error; } @@ -1382,7 +1504,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, if (WIFSIGNALED(status)) { int sig = WTERMSIG(status); - if (sig != SIGTERM) + if (sig != SIGTERM && sig != SIGUSR2) log_err("fio: pid=%d, got signal=%d\n", (int) td->pid, sig); td->sig = sig; @@ -1429,11 +1551,10 @@ static void run_threads(void) unsigned long spent; unsigned 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; + + fio_idle_prof_init(); set_sig_handlers(); @@ -1491,6 +1612,9 @@ static void run_threads(void) } } + /* start idle threads before io threads start to run */ + fio_idle_prof_start(); + set_genesis_time(); while (todo) { @@ -1529,6 +1653,9 @@ static void run_threads(void) init_disk_util(td); + td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); + td->update_rusage = 0; + /* * Set state to created. Thread will transition * to TD_INITIALIZED when it's done setting up. @@ -1653,8 +1780,9 @@ static void run_threads(void) usleep(10000); } + fio_idle_prof_stop(); + update_io_ticks(); - fio_unpin_memory(); } void wait_for_disk_thread_exit(void)