X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=fio.c;h=0f72fd92c15c883e75603a19b2a4fc89e915ba6f;hb=a2f77c9f5f01186882762264100b6d30ee77a55d;hp=1c2748e06ee4483b0cdef5457757985140a133a6;hpb=e916b390684ec1ca6247f98138fa9c1682701d29;p=fio.git diff --git a/fio.c b/fio.c index 1c2748e0..0f72fd92 100644 --- a/fio.c +++ b/fio.c @@ -149,7 +149,9 @@ static void cleanup_pending_aio(struct thread_data *td) /* * get immediately available events, if any */ - io_u_queued_complete(td, 0, NULL); + r = io_u_queued_complete(td, 0, NULL); + if (r < 0) + return; /* * now cancel remaining active events @@ -158,14 +160,24 @@ static void cleanup_pending_aio(struct thread_data *td) list_for_each_safe(entry, n, &td->io_u_busylist) { io_u = list_entry(entry, struct io_u, list); - r = td->io_ops->cancel(td, io_u); - if (!r) + /* + * if the io_u isn't in flight, then that generally + * means someone leaked an io_u. complain but fix + * it up, so we don't stall here. + */ + if ((io_u->flags & IO_U_F_FLIGHT) == 0) { + log_err("fio: non-busy IO on busy list\n"); put_io_u(td, io_u); + } else { + r = td->io_ops->cancel(td, io_u); + if (!r) + put_io_u(td, io_u); + } } } if (td->cur_depth) - io_u_queued_complete(td, td->cur_depth, NULL); + r = io_u_queued_complete(td, td->cur_depth, NULL); } /* @@ -195,7 +207,7 @@ requeue: put_io_u(td, io_u); return 1; } else if (ret == FIO_Q_QUEUED) { - if (io_u_queued_complete(td, 1, NULL)) + if (io_u_queued_complete(td, 1, NULL) < 0) return 1; } else if (ret == FIO_Q_COMPLETED) { if (io_u->error) { @@ -203,7 +215,8 @@ requeue: return 1; } - io_u_sync_complete(td, io_u, NULL); + if (io_u_sync_complete(td, io_u, NULL) < 0) + return 1; } else if (ret == FIO_Q_BUSY) { if (td_io_commit(td)) return 1; @@ -228,10 +241,15 @@ static void do_verify(struct thread_data *td) * read from disk. */ for_each_file(td, f, i) { - fio_io_sync(td, f); - file_invalidate_cache(td, f); + if (fio_io_sync(td, f)) + break; + if (file_invalidate_cache(td, f)) + break; } + if (td->error) + return; + td_set_runstate(td, TD_VERIFYING); io_u = NULL; @@ -240,15 +258,20 @@ static void do_verify(struct thread_data *td) if (!io_u) break; - if (runtime_exceeded(td, &io_u->start_time)) + if (runtime_exceeded(td, &io_u->start_time)) { + put_io_u(td, io_u); break; + } - if (get_next_verify(td, io_u)) + if (get_next_verify(td, io_u)) { + put_io_u(td, io_u); break; + } - if (td_io_prep(td, io_u)) + if (td_io_prep(td, io_u)) { + put_io_u(td, io_u); break; - + } requeue: ret = td_io_queue(td, io_u); @@ -264,7 +287,7 @@ requeue: goto requeue; } ret = io_u_sync_complete(td, io_u, verify_io_u); - if (ret) + if (ret < 0) break; continue; case FIO_Q_QUEUED: @@ -298,13 +321,10 @@ requeue: * Reap required number of io units, if any, and do the * verification on them through the callback handler */ - if (io_u_queued_complete(td, min_events, verify_io_u)) + if (io_u_queued_complete(td, min_events, verify_io_u) < 0) break; } - if (io_u) - put_io_u(td, io_u); - if (td->cur_depth) cleanup_pending_aio(td); @@ -543,9 +563,12 @@ static int init_io_u(struct thread_data *td) fill_rand_buf(io_u, max_bs); io_u->index = i; + io_u->flags = IO_U_F_FREE; list_add(&io_u->list, &td->io_u_freelist); } + io_u_init_timeout(); + return 0; } @@ -694,6 +717,7 @@ static void *thread_main(void *data) } fio_gettime(&td->epoch, NULL); + memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); getrusage(RUSAGE_SELF, &td->ts.ru_start); runtime[0] = runtime[1] = 0; @@ -761,7 +785,7 @@ err: close_ioengine(td); cleanup_io_u(td); td_set_runstate(td, TD_EXITED); - return (void *) td->error; + return (void *) (unsigned long) td->error; } /* @@ -784,7 +808,7 @@ static int fork_main(int shmid, int offset) td = data + offset * sizeof(struct thread_data); ret = thread_main(td); shmdt(data); - return (int) ret; + return (int) (unsigned long) ret; } /* @@ -800,6 +824,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) */ pending = cputhreads = 0; for_each_td(td, i) { + int flags; + /* * ->io_ops is NULL for a thread that has closed its * io engine @@ -807,55 +833,57 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) if (td->io_ops && td->io_ops->flags & FIO_CPUIO) cputhreads++; - if (td->runstate < TD_EXITED) { - /* - * check if someone quit or got killed in an unusual way - */ - ret = waitpid(td->pid, &status, WNOHANG); - if (ret < 0) - perror("waitpid"); - else if ((ret == td->pid) && WIFSIGNALED(status)) { + if (!td->pid || td->runstate == TD_REAPED) + continue; + + flags = WNOHANG; + if (td->runstate == TD_EXITED) + flags = 0; + + /* + * check if someone quit or got killed in an unusual way + */ + ret = waitpid(td->pid, &status, flags); + if (ret < 0) { + if (errno == ECHILD) { + log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate); + td_set_runstate(td, TD_REAPED); + goto reaped; + } + perror("waitpid"); + } else if (ret == td->pid) { + if (WIFSIGNALED(status)) { int sig = WTERMSIG(status); log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); td_set_runstate(td, TD_REAPED); goto reaped; } - } - - if (td->runstate != TD_EXITED) { - if (td->runstate < TD_RUNNING) - pending++; - - continue; - } - - if (td->error) - exit_value++; - - td_set_runstate(td, TD_REAPED); - - if (td->use_thread) { - long ret; - - if (pthread_join(td->thread, (void *) &ret)) - perror("thread_join"); - } else { - int status; - - ret = waitpid(td->pid, &status, 0); - if (ret < 0) - perror("waitpid"); - else if (WIFEXITED(status) && WEXITSTATUS(status)) { - if (!exit_value) - exit_value++; + if (WIFEXITED(status)) { + if (WEXITSTATUS(status) && !td->error) + td->error = WEXITSTATUS(status); + if (td->use_thread) { + long ret; + + if (pthread_join(td->thread, (void *) &ret)) + perror("pthread_join"); + } + td_set_runstate(td, TD_REAPED); + goto reaped; } } + /* + * thread is not dead, continue + */ + continue; reaped: (*nr_running)--; (*m_rate) -= td->ratemin; (*t_rate) -= td->rate; + + if (td->error) + exit_value++; } if (*nr_running == cputhreads && !pending) @@ -911,6 +939,8 @@ static void run_threads(void) init_disk_util(td); } + set_genesis_time(); + while (todo) { struct thread_data *map[MAX_JOBS]; struct timeval this_start; @@ -1069,6 +1099,8 @@ int main(int argc, char *argv[]) setup_log(&agg_io_log[DDIR_WRITE]); } + set_genesis_time(); + disk_util_timer_arm(); run_threads();