int
vdev_disk_physio(vdev_t *vd, caddr_t data,
size_t size, uint64_t offset, int flags, boolean_t isdump)
{
int rc = EIO;
vdev_disk_t *dvd;
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)
goto out;
ASSERT(vd->vdev_ops == &vdev_disk_ops);
/*
* If in the context of an active crash dump, use the ldi_dump(9F)
* call instead of ldi_strategy(9F) as usual.
*/
if (isdump) {
ASSERT3P(dvd, !=, NULL);
rc = ldi_dump(dvd->vd_lh, data, lbtodb(offset), lbtodb(size));
goto out;
}
rc = vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags);
out:
rw_exit(&vd->vdev_tsd_lock);
return (rc);
}
int
vdev_disk_physio(vdev_t *vd, caddr_t data,
size_t size, uint64_t offset, int flags, boolean_t isdump)
{
vdev_disk_t *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_ldi_offline && dvd->vd_lh == NULL))
return (EIO);
ASSERT(vd->vdev_ops == &vdev_disk_ops);
/* XXX Apple - no equivalent crash dump mechanism on OS X */
#ifdef illumos
/*
* If in the context of an active crash dump, use the ldi_dump(9F)
* call instead of ldi_strategy(9F) as usual.
*/
if (isdump) {
ASSERT3P(dvd, !=, NULL);
return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
lbtodb(size)));
}
#endif
return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
}
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
zvol_size_changed(zvol_state_t *zv, dev_t dev)
{
dev = makedevice(getmajor(dev), zv->zv_minor);
VERIFY(ddi_prop_update_int64(dev, zfs_dip,
"Size", zv->zv_volsize) == DDI_SUCCESS);
VERIFY(ddi_prop_update_int64(dev, zfs_dip,
"Nblocks", lbtodb(zv->zv_volsize)) == DDI_SUCCESS);
}
static void
zvol_size_changed(uint64_t volsize, major_t maj, minor_t min)
{
dev_t dev = makedevice(maj, min);
VERIFY(ddi_prop_update_int64(dev, zfs_dip,
"Size", volsize) == DDI_SUCCESS);
VERIFY(ddi_prop_update_int64(dev, zfs_dip,
"Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
/* Notify specfs to invalidate the cached size */
spec_size_invalidate(dev, VBLK);
spec_size_invalidate(dev, VCHR);
}
/*
* Called from roll thread;
* buffer set for reading master
* Returns
* 0 - success, can continue with next buffer
* 1 - failure due to logmap deltas being in use
*/
int
top_read_roll(rollbuf_t *rbp, ml_unit_t *ul)
{
buf_t *bp = &rbp->rb_bh;
offset_t mof = ldbtob(bp->b_blkno);
/*
* get a list of deltas
*/
if (logmap_list_get_roll(ul->un_logmap, mof, rbp)) {
/* logmap deltas are in use */
return (1);
}
/*
* no deltas were found, nothing to roll
*/
if (rbp->rb_age == NULL) {
bp->b_flags |= B_INVAL;
return (0);
}
/*
* If there is one cached roll buffer that cover all the deltas then
* we can use that instead of copying to a separate roll buffer.
*/
if (rbp->rb_crb) {
rbp->rb_bh.b_blkno = lbtodb(rbp->rb_crb->c_mof);
return (0);
}
/*
* Set up the read.
* If no read is needed logmap_setup_read() returns 0.
*/
if (logmap_setup_read(rbp->rb_age, rbp)) {
/*
* async read the data from master
*/
logstats.ls_rreads.value.ui64++;
bp->b_bcount = MAPBLOCKSIZE;
(void) bdev_strategy(bp);
lwp_stat_update(LWP_STAT_INBLK, 1);
} else {
sema_v(&bp->b_io); /* mark read as complete */
}
return (0);
}
static int
vdev_file_io_start(zio_t *zio)
{
spa_t *spa = zio->io_spa;
vdev_t *vd = zio->io_vd;
vdev_file_t *vf = vd->vdev_tsd;
vdev_buf_t *vb;
buf_t *bp;
if (zio->io_type == ZIO_TYPE_IOCTL) {
/* XXPOLICY */
if (!vdev_readable(vd)) {
zio->io_error = ENXIO;
return (ZIO_PIPELINE_CONTINUE);
}
switch (zio->io_cmd) {
case DKIOCFLUSHWRITECACHE:
zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
kcred, NULL);
break;
default:
zio->io_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 = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
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_private = vf->vf_vnode;
bp->b_iodone = (int (*)())vdev_file_io_intr;
taskq_dispatch_ent(spa->spa_zio_taskq[ZIO_TYPE_FREE][ZIO_TASKQ_ISSUE],
vdev_file_io_strategy, bp, 0, &zio->io_tqent);
return (ZIO_PIPELINE_STOP);
}
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);
}
/*
* Find something to roll, then if we don't have cached roll buffers
* covering all the deltas in that MAPBLOCK then read the master
* and overlay the deltas.
* returns;
* 0 if sucessful
* 1 on finding nothing to roll
* 2 on error
*/
int
log_roll_read(ml_unit_t *ul, rollbuf_t *rbs, int nmblk, caddr_t roll_bufs,
int *retnbuf)
{
offset_t mof;
buf_t *bp;
rollbuf_t *rbp;
mt_map_t *logmap = ul->un_logmap;
daddr_t mblkno;
int i;
int error;
int nbuf;
/*
* Make sure there is really something to roll
*/
mof = 0;
if (!logmap_next_roll(logmap, &mof)) {
return (1);
}
/*
* build some master blocks + deltas to roll forward
*/
rw_enter(&logmap->mtm_rwlock, RW_READER);
nbuf = 0;
do {
mof = mof & (offset_t)MAPBLOCKMASK;
mblkno = lbtodb(mof);
/*
* Check for the case of a new delta to a set up buffer
*/
for (i = 0, rbp = rbs; i < nbuf; ++i, ++rbp) {
if (P2ALIGN(rbp->rb_bh.b_blkno,
MAPBLOCKSIZE / DEV_BSIZE) == mblkno) {
TNF_PROBE_0(trans_roll_new_delta, "lufs",
/* CSTYLED */);
trans_roll_new_delta++;
/* Flush out the current set of buffers */
goto flush_bufs;
}
}
/*
* Work out what to roll next. If it isn't cached then read
* it asynchronously from the master.
*/
bp = &rbp->rb_bh;
bp->b_blkno = mblkno;
bp->b_flags = B_READ;
bp->b_un.b_addr = roll_bufs + (nbuf << MAPBLOCKSHIFT);
bp->b_bufsize = MAPBLOCKSIZE;
if (top_read_roll(rbp, ul)) {
/* logmap deltas were in use */
if (nbuf == 0) {
/*
* On first buffer wait for the logmap user
* to finish by grabbing the logmap lock
* exclusively rather than spinning
*/
rw_exit(&logmap->mtm_rwlock);
lrr_wait++;
rw_enter(&logmap->mtm_rwlock, RW_WRITER);
rw_exit(&logmap->mtm_rwlock);
return (1);
}
/* we have at least one buffer - flush it */
goto flush_bufs;
}
if ((bp->b_flags & B_INVAL) == 0) {
nbuf++;
}
mof += MAPBLOCKSIZE;
} while ((nbuf < nmblk) && logmap_next_roll(logmap, &mof));
