#include <string.h>
#include <signal.h>
#include <time.h>
-#include <assert.h>
+#include <locale.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <sys/ipc.h>
#include <sys/shm.h>
-#include <sys/ioctl.h>
#include <sys/mman.h>
#include "fio.h"
#include "os.h"
-#define MASK (4095)
-
-#define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK))
+static unsigned long page_mask;
+#define ALIGN(buf) \
+ (char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
int groupid = 0;
int thread_number = 0;
static volatile int startup_sem;
static volatile int fio_abort;
+static int exit_value;
struct io_log *agg_io_log[2];
disk_util_timer_arm();
print_thread_status();
break;
- case SIGSEGV:
- fprintf(stderr, "fio: got segfault, aborting\n");
- terminate_threads(TERMINATE_ALL, 1);
- fio_abort = 1;
- exit(0);
default:
printf("\nfio: terminating on signal %d\n", sig);
fflush(stdout);
td_set_runstate(td, TD_RUNNING);
- while (td->this_io_bytes[td->ddir] < td->io_size) {
+ while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) {
struct timespec *timeout;
int min_evts = 0;
struct io_u *io_u;
break;
memcpy(&s, &io_u->start_time, sizeof(s));
-
+requeue:
ret = td_io_queue(td, io_u);
if (ret) {
- td_verror(td, io_u->error);
- put_io_u(td, io_u);
- break;
+ if (ret > 0 && (io_u->xfer_buflen != io_u->resid) &&
+ io_u->resid) {
+ /*
+ * short read/write. requeue.
+ */
+ io_u->xfer_buflen = io_u->resid;
+ io_u->xfer_buf += ret;
+ goto requeue;
+ } else {
+ put_io_u(td, io_u);
+ break;
+ }
}
add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE)
td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1);
else
- td->orig_buffer_size += MASK;
+ td->orig_buffer_size += page_mask;
if (allocate_io_mem(td))
return 1;
if (td_io_init(td))
goto err;
- if (td->exec_prerun)
- system(td->exec_prerun);
+ if (td->exec_prerun) {
+ if (system(td->exec_prerun) < 0)
+ goto err;
+ }
fio_gettime(&td->epoch, NULL);
getrusage(RUSAGE_SELF, &td->ru_start);
finish_log(td, td->clat_log, "clat");
if (td->write_iolog_file)
write_iolog_close(td);
- if (td->exec_postrun)
- system(td->exec_postrun);
+ if (td->exec_postrun) {
+ if (system(td->exec_postrun) < 0)
+ log_err("fio: postrun %s failed\n", td->exec_postrun);
+ }
if (exitall_on_terminate)
terminate_threads(td->groupid, 0);
err:
+ if (td->error)
+ printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
close_files(td);
close_ioengine(td);
cleanup_io_u(td);
td_set_runstate(td, TD_EXITED);
- return NULL;
-
+ return (void *) td->error;
}
/*
* We cannot pass the td data into a forked process, so attach the td and
* pass it to the thread worker.
*/
-static void *fork_main(int shmid, int offset)
+static int fork_main(int shmid, int offset)
{
struct thread_data *td;
- void *data;
+ void *data, *ret;
data = shmat(shmid, NULL, 0);
if (data == (void *) -1) {
+ int __err = errno;
+
perror("shmat");
- return NULL;
+ return __err;
}
td = data + offset * sizeof(struct thread_data);
- thread_main(td);
+ ret = thread_main(td);
shmdt(data);
- return NULL;
+ return (int) ret;
}
/*
static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
{
struct thread_data *td;
- int i, cputhreads, pending;
+ int i, cputhreads, pending, status, ret;
/*
* reap exited threads (TD_EXITED -> TD_REAPED)
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)) {
+ 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) {
if (pthread_join(td->thread, (void *) &ret))
perror("thread_join");
- } else
- waitpid(td->pid, NULL, 0);
+ } else {
+ int status;
+ ret = waitpid(td->pid, &status, 0);
+ if (ret < 0)
+ perror("waitpid");
+ if (WIFEXITED(status) && WEXITSTATUS(status)) {
+ if (!exit_value)
+ exit_value++;
+ }
+ }
+
+reaped:
(*nr_running)--;
(*m_rate) -= td->ratemin;
(*t_rate) -= td->rate;
signal(SIGINT, sig_handler);
signal(SIGALRM, sig_handler);
- signal(SIGSEGV, sig_handler);
todo = thread_number;
nr_running = 0;
* client data interspersed on disk
*/
if (setup_files(td)) {
+ exit_value++;
+ if (td->error)
+ log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
td_set_runstate(td, TD_REAPED);
todo--;
}
if (fork())
fio_sem_down(&startup_sem);
else {
- fork_main(shm_id, i);
- exit(0);
+ int ret = fork_main(shm_id, i);
+
+ exit(ret);
}
}
}
int main(int argc, char *argv[])
{
+ long ps;
+
+ /*
+ * We need locale for number printing, if it isn't set then just
+ * go with the US format.
+ */
+ if (!getenv("LC_NUMERIC"))
+ setlocale(LC_NUMERIC, "en_US");
+
if (parse_options(argc, argv))
return 1;
return 1;
}
+ ps = sysconf(_SC_PAGESIZE);
+ if (ps < 0) {
+ log_err("Failed to get page size\n");
+ return 1;
+ }
+
+ page_mask = ps - 1;
+
if (write_bw_log) {
setup_log(&agg_io_log[DDIR_READ]);
setup_log(&agg_io_log[DDIR_WRITE]);
}
}
- return 0;
+ return exit_value;
}
projects / fio.git / blobdiff