int
vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
size_t size, uint64_t offset, int flags)
{
#ifdef illumos
buf_t *bp;
#else
ldi_buf_t *bp;
#endif
int error = 0;
if (vd_lh == NULL)
return (SET_ERROR(EINVAL));
ASSERT(flags & B_READ || flags & B_WRITE);
bp = getrbuf(KM_SLEEP);
bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
bp->b_bcount = size;
bp->b_un.b_addr = (void *)data;
bp->b_lblkno = lbtodb(offset);
bp->b_bufsize = size;
error = ldi_strategy(vd_lh, bp);
ASSERT(error == 0);
if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
error = SET_ERROR(EIO);
freerbuf(bp);
return (error);
}
static void
vdev_disk_io_start(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
vdev_disk_t *dvd;
vdev_buf_t *vb;
struct dk_callback *dkc;
buf_t *bp;
int error;
rw_enter(&vd->vdev_tsd_lock, RW_READER);
dvd = vd->vdev_tsd;
/*
* If the vdev is closed, it's likely in the REMOVED or FAULTED state.
* Nothing to be done here but return failure.
*/
if (dvd == NULL || dvd->vd_lh == NULL) {
zio->io_error = ENXIO;
rw_exit(&vd->vdev_tsd_lock);
zio_interrupt(zio);
return;
}
if (zio->io_type == ZIO_TYPE_IOCTL) {
/* XXPOLICY */
if (!vdev_readable(vd)) {
zio->io_error = SET_ERROR(ENXIO);
rw_exit(&vd->vdev_tsd_lock);
zio_interrupt(zio);
return;
}
switch (zio->io_cmd) {
case DKIOCFLUSHWRITECACHE:
if (zfs_nocacheflush)
break;
if (vd->vdev_nowritecache) {
zio->io_error = SET_ERROR(ENOTSUP);
break;
}
zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
zio->io_vsd_ops = &vdev_disk_vsd_ops;
dkc->dkc_callback = vdev_disk_ioctl_done;
dkc->dkc_flag = FLUSH_VOLATILE;
dkc->dkc_cookie = zio;
error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
(uintptr_t)dkc, FKIOCTL, kcred, NULL);
if (error == 0) {
/*
* The ioctl will be done asychronously,
* and will call vdev_disk_ioctl_done()
* upon completion.
*/
rw_exit(&vd->vdev_tsd_lock);
return;
}
if (error == ENOTSUP || error == ENOTTY) {
/*
* If we get ENOTSUP or ENOTTY, we know that
* no future attempts will ever succeed.
* In this case we set a persistent bit so
* that we don't bother with the ioctl in the
* future.
*/
vd->vdev_nowritecache = B_TRUE;
}
zio->io_error = error;
break;
case DKIOCFREE:
/*
* We perform device support checks here instead of
* in zio_trim(), as zio_trim() might be invoked on
* top of a top-level vdev, whereas vdev_disk_io_start
* is guaranteed to be operating a leaf vdev.
*/
if (vd->vdev_notrim &&
spa_get_force_trim(vd->vdev_spa) !=
SPA_FORCE_TRIM_ON) {
zio->io_error = SET_ERROR(ENOTSUP);
break;
}
/*
* zio->io_private contains a dkioc_free_list_t
* specifying which offsets are to be freed
*/
ASSERT(zio->io_private != NULL);
error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
(uintptr_t)zio->io_private, FKIOCTL, kcred, NULL);
if (error == ENOTSUP || error == ENOTTY)
vd->vdev_notrim = B_TRUE;
zio->io_error = error;
break;
default:
zio->io_error = SET_ERROR(ENOTSUP);
}
rw_exit(&vd->vdev_tsd_lock);
zio_execute(zio);
return;
}
vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
vb->vb_io = zio;
bp = &vb->vb_buf;
bioinit(bp);
bp->b_flags = B_BUSY | B_NOCACHE |
(zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
bp->b_flags |= B_FAILFAST;
bp->b_bcount = zio->io_size;
bp->b_un.b_addr = zio->io_data;
bp->b_lblkno = lbtodb(zio->io_offset);
bp->b_bufsize = zio->io_size;
bp->b_iodone = (int (*)())vdev_disk_io_intr;
/* ldi_strategy() will return non-zero only on programming errors */
VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
rw_exit(&vd->vdev_tsd_lock);
}
static int
vdev_disk_io_start(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
vdev_disk_t *dvd = vd->vdev_tsd;
vdev_buf_t *vb;
struct dk_callback *dkc;
buf_t *bp;
int error;
/*
* If the vdev is closed, it's likely in the REMOVED or FAULTED state.
* Nothing to be done here but return failure.
*/
if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
zio->io_error = ENXIO;
return (ZIO_PIPELINE_CONTINUE);
}
if (zio->io_type == ZIO_TYPE_IOCTL) {
/* XXPOLICY */
if (!vdev_readable(vd)) {
zio->io_error = SET_ERROR(ENXIO);
return (ZIO_PIPELINE_CONTINUE);
}
switch (zio->io_cmd) {
case DKIOCFLUSHWRITECACHE:
if (zfs_nocacheflush)
break;
if (vd->vdev_nowritecache) {
zio->io_error = SET_ERROR(ENOTSUP);
break;
}
zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
zio->io_vsd_ops = &vdev_disk_vsd_ops;
dkc->dkc_callback = vdev_disk_ioctl_done;
dkc->dkc_flag = FLUSH_VOLATILE;
dkc->dkc_cookie = zio;
error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
(uintptr_t)dkc, FKIOCTL, kcred, NULL);
if (error == 0) {
/*
* The ioctl will be done asychronously,
* and will call vdev_disk_ioctl_done()
* upon completion.
*/
return (ZIO_PIPELINE_STOP);
}
if (error == ENOTSUP || error == ENOTTY) {
/*
* If we get ENOTSUP or ENOTTY, we know that
* no future attempts will ever succeed.
* In this case we set a persistent bit so
* that we don't bother with the ioctl in the
* future.
*/
vd->vdev_nowritecache = B_TRUE;
}
zio->io_error = error;
break;
default:
zio->io_error = SET_ERROR(ENOTSUP);
}
return (ZIO_PIPELINE_CONTINUE);
}
vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
vb->vb_io = zio;
bp = &vb->vb_buf;
bioinit(bp);
bp->b_flags = B_BUSY | B_NOCACHE |
(zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
bp->b_flags |= B_FAILFAST;
bp->b_bcount = zio->io_size;
bp->b_un.b_addr = zio->io_data;
bp->b_lblkno = lbtodb(zio->io_offset);
bp->b_bufsize = zio->io_size;
bp->b_iodone = (int (*)())vdev_disk_io_intr;
zfs_zone_zio_start(zio);
/* ldi_strategy() will return non-zero only on programming errors */
VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
return (ZIO_PIPELINE_STOP);
}