X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=9c1bed379df586ba8f3efea187aaaf319ce2d387;hp=ba9e384b6609950a8c500dab96de39d0c8bce904;hb=d3cc4ebf58ec05037c3b0dd7c83c075a56f1dc62;hpb=86375e3446fdeb6c4dd1406cc0e4c456af954a2c diff --git a/fio.c b/fio.c index ba9e384b..9c1bed37 100644 --- a/fio.c +++ b/fio.c @@ -24,10 +24,12 @@ #include #include #include +#include #include #include #include #include +#include #include #include #include @@ -38,7 +40,12 @@ #include "hash.h" #include "smalloc.h" #include "verify.h" +#include "trim.h" #include "diskutil.h" +#include "cgroup.h" +#include "profile.h" +#include "lib/rand.h" +#include "memalign.h" unsigned long page_mask; unsigned long page_size; @@ -58,8 +65,12 @@ static struct fio_mutex *startup_mutex; static struct fio_mutex *writeout_mutex; static volatile int fio_abort; static int exit_value; -static struct itimerval itimer; static pthread_t gtod_thread; +static pthread_t disk_util_thread; +static struct flist_head *cgroup_list; +static char *cgroup_mnt; + +unsigned long arch_flags = 0; struct io_log *agg_io_log[2]; @@ -93,91 +104,87 @@ static void terminate_threads(int group_id) /* * if the thread is running, just let it exit */ - if (td->runstate < TD_RUNNING) - kill(td->pid, SIGQUIT); + if (!td->pid) + continue; + else if (td->runstate < TD_RAMP) + kill(td->pid, SIGTERM); else { struct ioengine_ops *ops = td->io_ops; - if (ops && (ops->flags & FIO_SIGQUIT)) - kill(td->pid, SIGQUIT); + if (ops && (ops->flags & FIO_SIGTERM)) + kill(td->pid, SIGTERM); } } } } -static void status_timer_arm(void) +static void sig_int(int sig) { - itimer.it_value.tv_sec = 0; - itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; - setitimer(ITIMER_REAL, &itimer, NULL); + if (threads) { + log_info("\nfio: terminating on signal %d\n", sig); + fflush(stdout); + exit_value = 128; + terminate_threads(TERMINATE_ALL); + } } -static void sig_alrm(int fio_unused sig) +static void *disk_thread_main(void *data) { - if (threads) { + fio_mutex_up(startup_mutex); + + while (threads) { + usleep(DISK_UTIL_MSEC * 1000); + if (!threads) + break; update_io_ticks(); print_thread_status(); - status_timer_arm(); } -} -/* - * Happens on thread runs with ctrl-c, ignore our own SIGQUIT - */ -static void sig_quit(int sig) -{ + return NULL; } -static void sig_int(int sig) +static int create_disk_util_thread(void) { - if (threads) { - printf("\nfio: terminating on signal %d\n", sig); - fflush(stdout); - terminate_threads(TERMINATE_ALL); + int ret; + + ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); + if (ret) { + log_err("Can't create disk util thread: %s\n", strerror(ret)); + return 1; } -} -static void sig_ill(int fio_unused sig) -{ - if (!threads) - return; + ret = pthread_detach(disk_util_thread); + if (ret) { + log_err("Can't detatch disk util thread: %s\n", strerror(ret)); + return 1; + } - log_err("fio: illegal instruction. your cpu does not support " - "the sse4.2 instruction for crc32c\n"); - terminate_threads(TERMINATE_ALL); - exit(4); + dprint(FD_MUTEX, "wait on startup_mutex\n"); + fio_mutex_down(startup_mutex); + dprint(FD_MUTEX, "done waiting on startup_mutex\n"); + return 0; } static void set_sig_handlers(void) { struct sigaction act; - memset(&act, 0, sizeof(act)); - act.sa_handler = sig_alrm; - act.sa_flags = SA_RESTART; - sigaction(SIGALRM, &act, NULL); - memset(&act, 0, sizeof(act)); act.sa_handler = sig_int; act.sa_flags = SA_RESTART; sigaction(SIGINT, &act, NULL); memset(&act, 0, sizeof(act)); - act.sa_handler = sig_ill; - act.sa_flags = SA_RESTART; - sigaction(SIGILL, &act, NULL); - - memset(&act, 0, sizeof(act)); - act.sa_handler = sig_quit; + act.sa_handler = sig_int; act.sa_flags = SA_RESTART; - sigaction(SIGQUIT, &act, NULL); + sigaction(SIGTERM, &act, NULL); } /* * Check if we are above the minimum rate given. */ static int __check_min_rate(struct thread_data *td, struct timeval *now, - enum td_ddir ddir) + enum fio_ddir ddir) { unsigned long long bytes = 0; unsigned long iops = 0; @@ -187,6 +194,8 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, unsigned int rate_iops = 0; unsigned int rate_iops_min = 0; + assert(ddir_rw(ddir)); + if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) return 0; @@ -367,10 +376,15 @@ requeue: return 0; } +static inline void __update_tv_cache(struct thread_data *td) +{ + fio_gettime(&td->tv_cache, NULL); +} + static inline void update_tv_cache(struct thread_data *td) { if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) - fio_gettime(&td->tv_cache, NULL); + __update_tv_cache(td); } static int break_on_this_error(struct thread_data *td, int *retptr) @@ -429,6 +443,8 @@ static void do_verify(struct thread_data *td) int ret, min_events; unsigned int i; + dprint(FD_VERIFY, "starting loop\n"); + /* * sync io first and invalidate cache, to make sure we really * read from disk. @@ -454,8 +470,11 @@ static void do_verify(struct thread_data *td) update_tv_cache(td); if (runtime_exceeded(td, &td->tv_cache)) { - td->terminate = 1; - break; + __update_tv_cache(td); + if (runtime_exceeded(td, &td->tv_cache)) { + td->terminate = 1; + break; + } } io_u = __get_io_u(td); @@ -485,7 +504,6 @@ static void do_verify(struct thread_data *td) 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; /* * zero read, fail @@ -500,8 +518,10 @@ static void do_verify(struct thread_data *td) io_u->xfer_buf += bytes; io_u->offset += bytes; - td->ts.short_io_u[io_u->ddir]++; + 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; @@ -532,12 +552,14 @@ sync_done: /* * if we can queue more, do so. but check if there are - * completed io_u's first. + * completed io_u's first. Note that we can get BUSY even + * without IO queued, if the system is resource starved. */ - full = queue_full(td) || ret == FIO_Q_BUSY; + full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); if (full || !td->o.iodepth_batch_complete) { - min_events = td->o.iodepth_batch_complete; - if (full && !min_events) + min_events = min(td->o.iodepth_batch_complete, + td->cur_depth); + if (full && !min_events && td->o.iodepth_batch_complete != 0) min_events = 1; do { @@ -565,6 +587,8 @@ sync_done: cleanup_pending_aio(td); td_set_runstate(td, TD_RUNNING); + + dprint(FD_VERIFY, "exiting loop\n"); } /* @@ -581,7 +605,9 @@ static void do_io(struct thread_data *td) else td_set_runstate(td, TD_RUNNING); - while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { + while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || + (!flist_empty(&td->trim_list)) || + ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { struct timeval comp_time; unsigned long bytes_done[2] = { 0, 0 }; int min_evts = 0; @@ -594,8 +620,11 @@ static void do_io(struct thread_data *td) update_tv_cache(td); if (runtime_exceeded(td, &td->tv_cache)) { - td->terminate = 1; - break; + __update_tv_cache(td); + if (runtime_exceeded(td, &td->tv_cache)) { + td->terminate = 1; + break; + } } io_u = get_io_u(td); @@ -606,7 +635,8 @@ static void do_io(struct thread_data *td) * Add verification end_io handler, if asked to verify * a previously written file. */ - if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { + if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && + !td_rw(td)) { if (td->o.verify_async) io_u->end_io = verify_io_u_async; else @@ -640,7 +670,8 @@ static void do_io(struct thread_data *td) io_u->xfer_buf += bytes; io_u->offset += bytes; - td->ts.short_io_u[io_u->ddir]++; + if (ddir_rw(io_u->ddir)) + td->ts.short_io_u[io_u->ddir]++; if (io_u->offset == f->real_file_size) goto sync_done; @@ -682,12 +713,15 @@ sync_done: break; /* - * See if we need to complete some commands + * See if we need to complete some commands. Note that we + * can get BUSY even without IO queued, if the system is + * resource starved. */ - full = queue_full(td) || ret == FIO_Q_BUSY; + full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); if (full || !td->o.iodepth_batch_complete) { - min_evts = td->o.iodepth_batch_complete; - if (full && !min_evts) + min_evts = min(td->o.iodepth_batch_complete, + td->cur_depth); + if (full && !min_evts && td->o.iodepth_batch_complete != 0) min_evts = 1; if (__should_check_rate(td, 0) || @@ -733,6 +767,9 @@ sync_done: } } + if (td->trim_entries) + printf("trim entries %ld\n", td->trim_entries); + if (td->o.fill_device && td->error == ENOSPC) { td->error = 0; td->terminate = 1; @@ -741,8 +778,11 @@ sync_done: struct fio_file *f; i = td->cur_depth; - if (i) + if (i) { ret = io_u_queued_complete(td, i, NULL); + if (td->o.fill_device && td->error == ENOSPC) + td->error = 0; + } if (should_fsync(td) && td->o.end_fsync) { td_set_runstate(td, TD_FSYNCING); @@ -772,7 +812,7 @@ static void cleanup_io_u(struct thread_data *td) io_u = flist_entry(entry, struct io_u, list); flist_del(&io_u->list); - free(io_u); + fio_memfree(io_u, sizeof(*io_u)); } free_io_mem(td); @@ -805,7 +845,8 @@ static int init_io_u(struct thread_data *td) if (allocate_io_mem(td)) return 1; - if (td->o.odirect || td->o.mem_align) + if (td->o.odirect || td->o.mem_align || + (td->io_ops->flags & FIO_RAWIO)) p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; else p = td->orig_buffer; @@ -818,8 +859,9 @@ static int init_io_u(struct thread_data *td) if (td->terminate) return 1; - if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { - log_err("fio: posix_memalign=%s\n", strerror(errno)); + ptr = fio_memalign(cl_align, sizeof(*io_u)); + if (!ptr) { + log_err("fio: unable to allocate aligned memory\n"); break; } @@ -834,6 +876,13 @@ static int init_io_u(struct thread_data *td) if (td_write(td) && !td->o.refill_buffers) io_u_fill_buffer(td, io_u, max_bs); + else if (td_write(td) && td->o.verify_pattern_bytes) { + /* + * Fill the buffer with the pattern if we are + * going to be doing writes. + */ + fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); + } } io_u->index = i; @@ -936,11 +985,6 @@ static void reset_io_counters(struct thread_data *td) */ if (td->o.time_based || td->o.loops) td->nr_done_files = 0; - - /* - * Set the same seed to get repeatable runs - */ - td_fill_rand_seeds(td); } void reset_all_stats(struct thread_data *td) @@ -956,8 +1000,10 @@ void reset_all_stats(struct thread_data *td) td->io_issues[i] = 0; td->ts.total_io_u[i] = 0; } - + fio_gettime(&tv, NULL); + td->ts.runtime[0] = 0; + td->ts.runtime[1] = 0; memcpy(&td->epoch, &tv, sizeof(tv)); memcpy(&td->start, &tv, sizeof(tv)); } @@ -972,6 +1018,11 @@ static void clear_io_state(struct thread_data *td) close_files(td); for_each_file(td, f, i) fio_file_clear_done(f); + + /* + * Set the same seed to get repeatable runs + */ + td_fill_rand_seeds(td); } static int exec_string(const char *string) @@ -996,15 +1047,16 @@ static int exec_string(const char *string) */ static void *thread_main(void *data) { - unsigned long long runtime[2], elapsed; + unsigned long long elapsed; struct thread_data *td = data; pthread_condattr_t attr; int clear_state; - if (!td->o.use_thread) + if (!td->o.use_thread) { setsid(); - - td->pid = getpid(); + td->pid = getpid(); + } else + td->pid = gettid(); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); @@ -1014,6 +1066,7 @@ static void *thread_main(void *data) INIT_FLIST_HEAD(&td->io_log_list); INIT_FLIST_HEAD(&td->io_hist_list); INIT_FLIST_HEAD(&td->verify_list); + INIT_FLIST_HEAD(&td->trim_list); pthread_mutex_init(&td->io_u_lock, NULL); td->io_hist_tree = RB_ROOT; @@ -1035,34 +1088,46 @@ static void *thread_main(void *data) fio_mutex_remove(td->mutex); /* - * May alter parameters that init_io_u() will use, so we need to - * do this first. + * A new gid requires privilege, so we need to do this before setting + * the uid. */ - if (init_iolog(td)) + if (td->o.gid != -1U && setgid(td->o.gid)) { + td_verror(td, errno, "setgid"); goto err; - - if (init_io_u(td)) + } + if (td->o.uid != -1U && setuid(td->o.uid)) { + td_verror(td, errno, "setuid"); goto err; + } - if (td->o.verify_async && verify_async_init(td)) - goto err; + /* + * If we have a gettimeofday() thread, make sure we exclude that + * thread from this job + */ + if (td->o.gtod_cpu) + fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); + /* + * Set affinity first, in case it has an impact on the memory + * allocations. + */ if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { td_verror(td, errno, "cpu_set_affinity"); goto err; } /* - * If we have a gettimeofday() thread, make sure we exclude that - * thread from this job + * May alter parameters that init_io_u() will use, so we need to + * do this first. */ - if (td->o.gtod_cpu) { - fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); - if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { - td_verror(td, errno, "cpu_set_affinity"); - goto err; - } - } + if (init_iolog(td)) + goto err; + + if (init_io_u(td)) + goto err; + + if (td->o.verify_async && verify_async_init(td)) + goto err; if (td->ioprio_set) { if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { @@ -1071,6 +1136,9 @@ static void *thread_main(void *data) } } + if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) + goto err; + if (nice(td->o.nice) == -1) { td_verror(td, errno, "nice"); goto err; @@ -1101,11 +1169,13 @@ static void *thread_main(void *data) fio_gettime(&td->epoch, NULL); getrusage(RUSAGE_SELF, &td->ts.ru_start); - runtime[0] = runtime[1] = 0; clear_state = 0; while (keep_running(td)) { fio_gettime(&td->start, NULL); - memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); + memcpy(&td->ts.stat_sample_time[0], &td->start, + sizeof(td->start)); + memcpy(&td->ts.stat_sample_time[1], &td->start, + sizeof(td->start)); memcpy(&td->tv_cache, &td->start, sizeof(td->start)); if (td->o.ratemin[0] || td->o.ratemin[1]) @@ -1123,11 +1193,11 @@ static void *thread_main(void *data) if (td_read(td) && td->io_bytes[DDIR_READ]) { elapsed = utime_since_now(&td->start); - runtime[DDIR_READ] += elapsed; + td->ts.runtime[DDIR_READ] += elapsed; } if (td_write(td) && td->io_bytes[DDIR_WRITE]) { elapsed = utime_since_now(&td->start); - runtime[DDIR_WRITE] += elapsed; + td->ts.runtime[DDIR_WRITE] += elapsed; } if (td->error || td->terminate) @@ -1144,15 +1214,15 @@ static void *thread_main(void *data) do_verify(td); - runtime[DDIR_READ] += utime_since_now(&td->start); + td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); if (td->error || td->terminate) break; } update_rusage_stat(td); - td->ts.runtime[0] = (runtime[0] + 999) / 1000; - td->ts.runtime[1] = (runtime[1] + 999) / 1000; + td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; + td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; td->ts.total_run_time = mtime_since_now(&td->epoch); td->ts.io_bytes[0] = td->io_bytes[0]; td->ts.io_bytes[1] = td->io_bytes[1]; @@ -1165,6 +1235,13 @@ static void *thread_main(void *data) } else finish_log(td, td->ts.bw_log, "bw"); } + if (td->ts.lat_log) { + if (td->o.lat_log_file) { + finish_log_named(td, td->ts.lat_log, + td->o.lat_log_file, "lat"); + } else + finish_log(td, td->ts.lat_log, "lat"); + } if (td->ts.slat_log) { if (td->o.lat_log_file) { finish_log_named(td, td->ts.slat_log, @@ -1188,11 +1265,16 @@ static void *thread_main(void *data) err: if (td->error) - printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, + log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, td->verror); + + if (td->o.verify_async) + verify_async_exit(td); + close_and_free_files(td); close_ioengine(td); cleanup_io_u(td); + cgroup_shutdown(td, &cgroup_mnt); if (td->o.cpumask_set) { int ret = fio_cpuset_exit(&td->o.cpumask); @@ -1200,9 +1282,6 @@ err: td_verror(td, ret, "fio_cpuset_exit"); } - if (td->o.verify_async) - verify_async_exit(td); - /* * do this very late, it will log file closing as well */ @@ -1223,6 +1302,7 @@ static int fork_main(int shmid, int offset) struct thread_data *td; void *data, *ret; +#ifndef __hpux data = shmat(shmid, NULL, 0); if (data == (void *) -1) { int __err = errno; @@ -1230,6 +1310,12 @@ static int fork_main(int shmid, int offset) perror("shmat"); return __err; } +#else + /* + * HP-UX inherits shm mappings? + */ + data = threads; +#endif td = data + offset * sizeof(struct thread_data); ret = thread_main(td); @@ -1295,7 +1381,7 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) if (WIFSIGNALED(status)) { int sig = WTERMSIG(status); - if (sig != SIGQUIT) + if (sig != SIGTERM) log_err("fio: pid=%d, got signal=%d\n", (int) td->pid, sig); td_set_runstate(td, TD_REAPED); @@ -1352,9 +1438,13 @@ static void *gtod_thread_main(void *data) static int fio_start_gtod_thread(void) { + pthread_attr_t attr; int ret; - ret = pthread_create(>od_thread, NULL, gtod_thread_main, NULL); + pthread_attr_init(&attr); + pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); + ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); + pthread_attr_destroy(&attr); if (ret) { log_err("Can't create gtod thread: %s\n", strerror(ret)); return 1; @@ -1388,17 +1478,17 @@ static void run_threads(void) return; if (!terse_output) { - printf("Starting "); + log_info("Starting "); if (nr_thread) - printf("%d thread%s", nr_thread, + log_info("%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); if (nr_process) { if (nr_thread) printf(" and "); - printf("%d process%s", nr_process, + log_info("%d process%s", nr_process, nr_process > 1 ? "es" : ""); } - printf("\n"); + log_info("\n"); fflush(stdout); } @@ -1431,14 +1521,14 @@ static void run_threads(void) todo--; } else { struct fio_file *f; - unsigned int i; + unsigned int j; /* * for sharing to work, each job must always open * its own files. so close them, if we opened them * for creation */ - for_each_file(td, f, i) { + for_each_file(td, f, j) { if (fio_file_open(f)) td_io_close_file(td, f); } @@ -1450,7 +1540,7 @@ static void run_threads(void) set_genesis_time(); while (todo) { - struct thread_data *map[MAX_JOBS]; + struct thread_data *map[REAL_MAX_JOBS]; struct timeval this_start; int this_jobs = 0, left; @@ -1602,11 +1692,14 @@ static void run_threads(void) fio_unpin_memory(); } -int main(int argc, char *argv[]) +int main(int argc, char *argv[], char *envp[]) { long ps; + arch_init(envp); + sinit(); + init_rand(&__fio_rand_state); /* * We need locale for number printing, if it isn't set then just @@ -1615,12 +1708,6 @@ int main(int argc, char *argv[]) if (!getenv("LC_NUMERIC")) setlocale(LC_NUMERIC, "en_US"); - if (parse_options(argc, argv)) - return 1; - - if (!thread_number) - return 0; - ps = sysconf(_SC_PAGESIZE); if (ps < 0) { log_err("Failed to get page size\n"); @@ -1630,17 +1717,34 @@ int main(int argc, char *argv[]) page_size = ps; page_mask = ps - 1; + fio_keywords_init(); + + if (parse_options(argc, argv)) + return 1; + + if (exec_profile && load_profile(exec_profile)) + return 1; + + if (!thread_number) + return 0; + if (write_bw_log) { setup_log(&agg_io_log[DDIR_READ]); setup_log(&agg_io_log[DDIR_WRITE]); } startup_mutex = fio_mutex_init(0); + if (startup_mutex == NULL) + return 1; writeout_mutex = fio_mutex_init(1); + if (writeout_mutex == NULL) + return 1; set_genesis_time(); + create_disk_util_thread(); - status_timer_arm(); + cgroup_list = smalloc(sizeof(*cgroup_list)); + INIT_FLIST_HEAD(cgroup_list); run_threads(); @@ -1653,6 +1757,10 @@ int main(int argc, char *argv[]) } } + cgroup_kill(cgroup_list); + sfree(cgroup_list); + sfree(cgroup_mnt); + fio_mutex_remove(startup_mutex); fio_mutex_remove(writeout_mutex); return exit_value;