struct io_u **io_u_index;
+ int *fds;
+
struct io_sq_ring sq_ring;
struct io_uring_sqe *sqes;
struct iovec *iovecs;
void *pad;
unsigned int hipri;
unsigned int fixedbufs;
+ unsigned int registerfiles;
unsigned int sqpoll_thread;
unsigned int sqpoll_set;
unsigned int sqpoll_cpu;
+ unsigned int nonvectored;
+ unsigned int uncached;
+};
+
+static const int ddir_to_op[2][2] = {
+ { IORING_OP_READV, IORING_OP_READ },
+ { IORING_OP_WRITEV, IORING_OP_WRITE }
};
static int fio_ioring_sqpoll_cb(void *data, unsigned long long *val)
.off1 = offsetof(struct ioring_options, hipri),
.help = "Use polled IO completions",
.category = FIO_OPT_C_ENGINE,
- .group = FIO_OPT_G_LIBAIO,
+ .group = FIO_OPT_G_IOURING,
},
{
.name = "fixedbufs",
.off1 = offsetof(struct ioring_options, fixedbufs),
.help = "Pre map IO buffers",
.category = FIO_OPT_C_ENGINE,
- .group = FIO_OPT_G_LIBAIO,
+ .group = FIO_OPT_G_IOURING,
+ },
+ {
+ .name = "registerfiles",
+ .lname = "Register file set",
+ .type = FIO_OPT_STR_SET,
+ .off1 = offsetof(struct ioring_options, registerfiles),
+ .help = "Pre-open/register files",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
},
{
.name = "sqthread_poll",
.off1 = offsetof(struct ioring_options, sqpoll_thread),
.help = "Offload submission/completion to kernel thread",
.category = FIO_OPT_C_ENGINE,
- .group = FIO_OPT_G_LIBAIO,
+ .group = FIO_OPT_G_IOURING,
},
{
.name = "sqthread_poll_cpu",
.cb = fio_ioring_sqpoll_cb,
.help = "What CPU to run SQ thread polling on",
.category = FIO_OPT_C_ENGINE,
- .group = FIO_OPT_G_LIBAIO,
+ .group = FIO_OPT_G_IOURING,
+ },
+ {
+ .name = "nonvectored",
+ .lname = "Non-vectored",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_options, nonvectored),
+ .help = "Use non-vectored read/write commands",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
+ },
+ {
+ .name = "uncached",
+ .lname = "Uncached",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_options, uncached),
+ .help = "Use RWF_UNCACHED for buffered read/writes",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
},
{
.name = NULL,
struct io_uring_sqe *sqe;
sqe = &ld->sqes[io_u->index];
- sqe->fd = f->fd;
- sqe->flags = 0;
- sqe->ioprio = 0;
- sqe->buf_index = 0;
+
+ /* zero out fields not used in this submission */
+ memset(sqe, 0, sizeof(*sqe));
+
+ if (o->registerfiles) {
+ sqe->fd = f->engine_pos;
+ sqe->flags = IOSQE_FIXED_FILE;
+ } else {
+ sqe->fd = f->fd;
+ }
if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) {
+ sqe->opcode = ddir_to_op[io_u->ddir][!!o->nonvectored];
if (o->fixedbufs) {
- if (io_u->ddir == DDIR_READ)
- sqe->opcode = IORING_OP_READ_FIXED;
- else
- sqe->opcode = IORING_OP_WRITE_FIXED;
sqe->addr = (unsigned long) io_u->xfer_buf;
sqe->len = io_u->xfer_buflen;
sqe->buf_index = io_u->index;
} else {
- if (io_u->ddir == DDIR_READ)
- sqe->opcode = IORING_OP_READV;
- else
- sqe->opcode = IORING_OP_WRITEV;
- sqe->addr = (unsigned long) &ld->iovecs[io_u->index];
- sqe->len = 1;
+ if (o->nonvectored) {
+ sqe->addr = (unsigned long)
+ ld->iovecs[io_u->index].iov_base;
+ sqe->len = ld->iovecs[io_u->index].iov_len;
+ } else {
+ sqe->addr = (unsigned long) &ld->iovecs[io_u->index];
+ sqe->len = 1;
+ }
}
+ if (!td->o.odirect && o->uncached)
+ sqe->rw_flags = RWF_UNCACHED;
+ if (fio_option_is_set(&td->o, ioprio_class))
+ sqe->ioprio = td->o.ioprio_class << 13;
+ if (fio_option_is_set(&td->o, ioprio))
+ sqe->ioprio |= td->o.ioprio;
sqe->off = io_u->offset;
} else if (ddir_sync(io_u->ddir)) {
- sqe->fsync_flags = 0;
- if (io_u->ddir == DDIR_DATASYNC)
- sqe->fsync_flags |= IORING_FSYNC_DATASYNC;
- sqe->opcode = IORING_OP_FSYNC;
+ if (io_u->ddir == DDIR_SYNC_FILE_RANGE) {
+ sqe->off = f->first_write;
+ sqe->len = f->last_write - f->first_write;
+ sqe->sync_range_flags = td->o.sync_file_range;
+ sqe->opcode = IORING_OP_SYNC_FILE_RANGE;
+ } else {
+ if (io_u->ddir == DDIR_DATASYNC)
+ sqe->fsync_flags |= IORING_FSYNC_DATASYNC;
+ sqe->opcode = IORING_OP_FSYNC;
+ }
}
sqe->user_data = (unsigned long) io_u;
r = fio_ioring_cqring_reap(td, events, max);
if (r) {
events += r;
+ if (actual_min != 0)
+ actual_min -= r;
continue;
}
r = io_uring_enter(ld, 0, actual_min,
IORING_ENTER_GETEVENTS);
if (r < 0) {
- if (errno == EAGAIN)
+ if (errno == EAGAIN || errno == EINTR)
continue;
td_verror(td, errno, "io_uring_enter");
break;
io_u_mark_submit(td, ret);
continue;
} else {
- if (errno == EAGAIN) {
+ if (errno == EAGAIN || errno == EINTR) {
ret = fio_ioring_cqring_reap(td, 0, ld->queued);
if (ret)
continue;
free(ld->io_u_index);
free(ld->iovecs);
+ free(ld->fds);
free(ld);
}
}
return fio_ioring_mmap(ld, &p);
}
+static int fio_ioring_register_files(struct thread_data *td)
+{
+ struct ioring_data *ld = td->io_ops_data;
+ struct fio_file *f;
+ unsigned int i;
+ int ret;
+
+ ld->fds = calloc(td->o.nr_files, sizeof(int));
+
+ for_each_file(td, f, i) {
+ ret = generic_open_file(td, f);
+ if (ret)
+ goto err;
+ ld->fds[i] = f->fd;
+ f->engine_pos = i;
+ }
+
+ ret = syscall(__NR_sys_io_uring_register, ld->ring_fd,
+ IORING_REGISTER_FILES, ld->fds, td->o.nr_files);
+ if (ret) {
+err:
+ free(ld->fds);
+ ld->fds = NULL;
+ }
+
+ /*
+ * Pretend the file is closed again, and really close it if we hit
+ * an error.
+ */
+ for_each_file(td, f, i) {
+ if (ret) {
+ int fio_unused ret2;
+ ret2 = generic_close_file(td, f);
+ } else
+ f->fd = -1;
+ }
+
+ return ret;
+}
+
static int fio_ioring_post_init(struct thread_data *td)
{
struct ioring_data *ld = td->io_ops_data;
+ struct ioring_options *o = td->eo;
struct io_u *io_u;
int err, i;
return 1;
}
+ if (o->registerfiles) {
+ err = fio_ioring_register_files(td);
+ if (err) {
+ td_verror(td, errno, "ioring_register_files");
+ return 1;
+ }
+ }
+
return 0;
}
static int fio_ioring_init(struct thread_data *td)
{
+ struct ioring_options *o = td->eo;
struct ioring_data *ld;
+ /* sqthread submission requires registered files */
+ if (o->sqpoll_thread)
+ o->registerfiles = 1;
+
+ if (o->registerfiles && td->o.nr_files != td->o.open_files) {
+ log_err("fio: io_uring registered files require nr_files to "
+ "be identical to open_files\n");
+ return 1;
+ }
+
ld = calloc(1, sizeof(*ld));
/* ring depth must be a power-of-2 */
return 0;
}
+static int fio_ioring_open_file(struct thread_data *td, struct fio_file *f)
+{
+ struct ioring_data *ld = td->io_ops_data;
+ struct ioring_options *o = td->eo;
+
+ if (!ld || !o->registerfiles)
+ return generic_open_file(td, f);
+
+ f->fd = ld->fds[f->engine_pos];
+ return 0;
+}
+
+static int fio_ioring_close_file(struct thread_data *td, struct fio_file *f)
+{
+ struct ioring_data *ld = td->io_ops_data;
+ struct ioring_options *o = td->eo;
+
+ if (!ld || !o->registerfiles)
+ return generic_close_file(td, f);
+
+ f->fd = -1;
+ return 0;
+}
+
static struct ioengine_ops ioengine = {
.name = "io_uring",
.version = FIO_IOOPS_VERSION,
+ .flags = FIO_ASYNCIO_SYNC_TRIM,
.init = fio_ioring_init,
.post_init = fio_ioring_post_init,
.io_u_init = fio_ioring_io_u_init,
.getevents = fio_ioring_getevents,
.event = fio_ioring_event,
.cleanup = fio_ioring_cleanup,
- .open_file = generic_open_file,
- .close_file = generic_close_file,
+ .open_file = fio_ioring_open_file,
+ .close_file = fio_ioring_close_file,
.get_file_size = generic_get_file_size,
.options = options,
.option_struct_size = sizeof(struct ioring_options),