static int
lx_ptm_data_check(dev_t dev, int ignore_eof, int *rvalp)
{
ldi_handle_t lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
int err;
*rvalp = 0;
if (ignore_eof) {
int size, rval;
if ((err = ldi_ioctl(lh, FIONREAD, (intptr_t)&size,
FKIOCTL, kcred, &rval)) != 0)
return (err);
if (size != 0)
*rvalp = 1;
} else {
int msg_size, msg_count;
if ((err = ldi_ioctl(lh, I_NREAD, (intptr_t)&msg_size,
FKIOCTL, kcred, &msg_count)) != 0)
return (err);
if (msg_count != 0)
*rvalp = 1;
}
return (0);
}
/* Must be called with vd->vdev_tsd_lock taken */
static uint64_t
vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
{
ASSERT(vd->vdev_wholedisk);
ASSERT(rw_lock_held(&vd->vdev_tsd_lock));
vdev_disk_t *dvd = vd->vdev_tsd;
dk_efi_t dk_ioc;
efi_gpt_t *efi;
uint64_t avail_space = 0;
int rc = ENXIO, efisize = EFI_LABEL_SIZE * 2;
dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
dk_ioc.dki_lba = 1;
dk_ioc.dki_length = efisize;
dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
efi = dk_ioc.dki_data;
/*
* Here we are called with vdev_tsd_lock taken,
* so it's safe to use dvd and vd_lh if not NULL
*/
if (dvd != NULL && dvd->vd_lh != NULL) {
rc = ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
FKIOCTL, kcred, NULL);
}
if (rc == 0) {
uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
if (capacity > efi_altern_lba)
avail_space = (capacity - efi_altern_lba) * blksz;
}
kmem_free(dk_ioc.dki_data, efisize);
return (avail_space);
}
static void
vdev_disk_io_done(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
zfs_zone_zio_done(zio);
/*
* If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
* the device has been removed. If this is the case, then we trigger an
* asynchronous removal of the device. Otherwise, probe the device and
* make sure it's still accessible.
*/
if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
vdev_disk_t *dvd = vd->vdev_tsd;
int state = DKIO_NONE;
if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
/*
* We post the resource as soon as possible, instead of
* when the async removal actually happens, because the
* DE is using this information to discard previous I/O
* errors.
*/
zfs_post_remove(zio->io_spa, vd);
vd->vdev_remove_wanted = B_TRUE;
spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
} else if (!vd->vdev_delayed_close) {
vd->vdev_delayed_close = B_TRUE;
}
}
}
static uint64_t
vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
{
ASSERT(vd->vdev_wholedisk);
vdev_disk_t *dvd = vd->vdev_tsd;
dk_efi_t dk_ioc;
efi_gpt_t *efi;
uint64_t avail_space = 0;
int efisize = EFI_LABEL_SIZE * 2;
dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
dk_ioc.dki_lba = 1;
dk_ioc.dki_length = efisize;
dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
efi = dk_ioc.dki_data;
if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
FKIOCTL, kcred, NULL) == 0) {
uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
if (capacity > efi_altern_lba)
avail_space = (capacity - efi_altern_lba) * blksz;
}
kmem_free(dk_ioc.dki_data, efisize);
return (avail_space);
}
static int
lx_ptm_eof_drop_1(dev_t dev, int *rvalp)
{
ldi_handle_t lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
int err, msg_size, msg_count;
*rvalp = 0;
/*
* Check if there is an EOF message (represented by a zero length
* data message) at the head of the stream. Note that the
* I_NREAD ioctl is a streams framework ioctl so it will succeed
* even if there have been previous write errors on this stream.
*/
if ((err = ldi_ioctl(lh, I_NREAD, (intptr_t)&msg_size,
FKIOCTL, kcred, &msg_count)) != 0)
return (err);
if ((msg_count == 0) || (msg_size != 0)) {
/* No EOF message found */
return (0);
}
/* Record the fact that the slave device has been closed. */
lx_ptm_lh_eofed_set(DEVT_TO_INDEX(dev));
/* drop the EOF */
lx_ptm_eof_read(lh);
*rvalp = 1;
return (0);
}
/*
* Public interface for creating an IP stream with shared conn_t
* Handles multiple callers in parallel by using conn_lock.
* Note that we allocate the helper stream without any locks, which means
* we might need to free it if we had two threads doing this concurrently
* for the conn_t.
*/
int
ip_create_helper_stream(conn_t *connp, ldi_ident_t li)
{
ip_helper_stream_info_t *helper;
int error;
int ret;
ASSERT(!servicing_interrupt());
if (connp->conn_helper_info != NULL) {
/* Already allocated */
return (0);
}
error = 0;
helper = kmem_alloc(sizeof (ip_helper_stream_info_t), KM_SLEEP);
/*
* open ip device via the layered interface.
* pass in kcred as some threads do not have the
* priviledge to open /dev/ip and the check in
* secpolicy_spec_open() will fail the open
*/
error = ldi_open_by_name((connp->conn_family == AF_INET6 ? DEV_IP6 :
DEV_IP), IP_HELPER_STR, kcred, &helper->iphs_handle, li);
if (error != 0) {
kmem_free(helper, sizeof (ip_helper_stream_info_t));
return (error);
}
/* Make sure we are the only one */
mutex_enter(&connp->conn_lock);
if (connp->conn_helper_info != NULL) {
/* Some other thread won - discard this stream */
mutex_exit(&connp->conn_lock);
(void) ldi_close(helper->iphs_handle, 0, kcred);
kmem_free(helper, sizeof (ip_helper_stream_info_t));
return (0);
}
connp->conn_helper_info = helper;
/*
* Share connp with the helper stream. We hold conn_lock across this
* operation.
*/
error = ldi_ioctl(helper->iphs_handle, SIOCSQPTR, (intptr_t)connp,
FKIOCTL, kcred, &ret);
if (error != 0) {
/*
* Passing in a zero flag indicates that an error
* occured and stream was not shared
*/
(void) ldi_close(helper->iphs_handle, 0, kcred);
kmem_free(helper, sizeof (ip_helper_stream_info_t));
connp->conn_helper_info = NULL;
}
mutex_exit(&connp->conn_lock);
return (error);
}
static void
vdev_disk_io_done(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
/*
* If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
* the device has been removed. If this is the case, then we trigger an
* asynchronous removal of the device.
*/
if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
/* Apple handle device removal in zfs_osx.cpp - read errors etc
* should be retried by zio
*/
#ifdef __APPLE__
return;
#else
state = DKIO_NONE;
if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
FKIOCTL, kcred, NULL) == 0 &&
state != DKIO_INSERTED)
{
/*
* We post the resource as soon as possible, instead of
* when the async removal actually happens, because the
* DE is using this information to discard previous I/O
* errors.
*/
zfs_post_remove(zio->io_spa, vd);
vd->vdev_remove_wanted = B_TRUE;
spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
spa_async_dispatch(zio->io_spa);
} else if (!vd->vdev_delayed_close) {
vd->vdev_delayed_close = B_TRUE;
}
#endif
}
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
int isssd;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_offline) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
devvp = dvd->vd_devvp;
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/* ### APPLE TODO ### */
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
context = vfs_context_create( spl_vfs_context_kernel() );
/* Obtain an opened/referenced vnode for the device. */
if ((error = vnode_open(vd->vdev_path, spa_mode(spa), 0, 0,
&devvp, context))) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/*
* ### APPLE TODO ###
* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
} else {
goto out;
}
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0,
context) != 0 ||
VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt, 0,
context) != 0) {
error = EINVAL;
goto out;
}
*psize = blkcnt * (uint64_t)blksize;
*max_psize = *psize;
dvd->vd_ashift = highbit(blksize) - 1;
dprintf("vdev_disk: Device %p ashift set to %d\n", devvp,
dvd->vd_ashift);
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* ### APPLE TODO ###
*/
#ifdef illumos
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
#endif
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (VNOP_IOCTL(devvp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0,
context) == 0) {
if (isssd)
vd->vdev_nonrot = B_TRUE;
}
dprintf("ZFS: vdev_disk(%s) isSSD %d\n", vd->vdev_path ? vd->vdev_path : "",
isssd);
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp) {
vnode_close(devvp, fmode, context);
dvd->vd_devvp = NULL;
}
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context)
(void) vfs_context_rele(context);
if (error) printf("ZFS: vdev_disk_open('%s') failed error %d\n",
vd->vdev_path ? vd->vdev_path : "", error);
return (error);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
ldi_ev_cookie_t ecookie;
vdev_disk_ldi_cb_t *lcb;
union {
struct dk_minfo_ext ude;
struct dk_minfo ud;
} dks;
struct dk_minfo_ext *dkmext = &dks.ude;
struct dk_minfo *dkm = &dks.ud;
int error;
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
dev_t dev;
int otyp;
boolean_t validate_devid = B_FALSE;
ddi_devid_t devid;
#endif
uint64_t capacity = 0, blksz = 0, pbsize;
#ifdef __APPLE__
int isssd;
#endif
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/*
* XXX We must not set or modify the devid as this check would prevent
* import on Solaris/illumos.
*/
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
vdev_dbgmsg(vd, "vdev_disk_open: invalid "
"vdev_devid '%s'", vd->vdev_devid);
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
/*
* XXX This assumes that if vdev_path refers to a device path /dev/dsk/cNtNdN,
* then the whole disk can be found by slice 0 at path /dev/dsk/cNtNdNs0.
*/
#ifdef illumos
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
error = ldi_open_by_name(buf, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
if (error == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
} else {
kmem_free(buf, len);
}
}
#endif
/*
* If we have not yet opened the device, try to open it by the
* specified path.
*/
if (error != 0) {
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = SET_ERROR(EINVAL);
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
#endif
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL) {
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
}
#endif
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
if (vd->vdev_devid != NULL)
validate_devid = B_TRUE;
#endif
/* XXX Apple to do - make ddi_ interface for this, using IORegistry path */
#ifdef illumos
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
#endif
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]",
error);
return (error);
}
/*
* XXX Apple - We must not set or modify the devid. Import on Solaris/illumos
* expects a valid devid and fails if it cannot be decoded.
*/
#ifdef illumos
/*
* Now that the device has been successfully opened, update the devid
* if necessary.
*/
if (validate_devid && spa_writeable(spa) &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
char *vd_devid;
vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
vdev_dbgmsg(vd, "vdev_disk_open: update devid from "
"'%s' to '%s'", vd->vdev_devid, vd_devid);
spa_strfree(vd->vdev_devid);
vd->vdev_devid = spa_strdup(vd_devid);
ddi_devid_str_free(vd_devid);
}
ddi_devid_free(devid);
}
#endif
/* XXX Apple to do, needs IORegistry physpath interface */
#ifdef illumos
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
#endif
/*
* Register callbacks for the LDI offline event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
}
/* XXX Apple to do - we could support the degrade event, or just no-op */
#ifdef illumos
/*
* Register callbacks for the LDI degrade event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
}
#endif
#if 0
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
#endif
skip_open:
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_dbgmsg(vd, "vdev_disk_open: failed to get size");
return (SET_ERROR(EINVAL));
}
*max_psize = *psize;
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
(intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
capacity = dkmext->dki_capacity - 1;
blksz = dkmext->dki_lbsize;
pbsize = dkmext->dki_pbsize;
} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
(intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
VDEV_DEBUG(
"vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
vd->vdev_path);
capacity = dkm->dki_capacity - 1;
blksz = dkm->dki_lbsize;
pbsize = blksz;
} else {
VDEV_DEBUG("vdev_disk_open(\"%s\"): "
"both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
vd->vdev_path, error);
pbsize = DEV_BSIZE;
}
*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
/* XXX Now that we opened the device, determine if it is a whole disk. */
#ifdef __APPLE__
/*
* XXX Apple to do - provide an ldi_ mechanism
* to report whether this is a whole disk or a
* partition.
* Return 0 (no), 1 (yes), or -1 (error).
*/
// vd->vdev_wholedisk = ldi_is_wholedisk(vd->vd_lh);
#endif
if (vd->vdev_wholedisk == 1) {
int wce = 1;
/* Gets information about the disk if it has GPT partitions */
#ifdef illumos
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
#endif
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
#ifdef __APPLE__
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (ldi_ioctl(dvd->vd_lh, DKIOCISSOLIDSTATE, (intptr_t)&isssd,
FKIOCTL, kcred, NULL) == 0) {
vd->vdev_nonrot = (isssd ? B_TRUE : B_FALSE);
}
#endif //__APPLE__
return (0);
}
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_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd;
ldi_ev_cookie_t ecookie;
vdev_disk_ldi_cb_t *lcb;
union {
struct dk_minfo_ext ude;
struct dk_minfo ud;
} dks;
struct dk_minfo_ext *dkmext = &dks.ude;
struct dk_minfo *dkm = &dks.ud;
int error;
dev_t dev;
int otyp;
boolean_t validate_devid = B_FALSE;
ddi_devid_t devid;
uint64_t capacity = 0, blksz = 0, pbsize;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
rw_enter(&vd->vdev_tsd_lock, RW_WRITER);
dvd = vd->vdev_tsd;
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
ASSERT(vd->vdev_reopening);
/*
* Here vd_lh is protected by vdev_tsd_lock
*/
ASSERT(dvd->vd_lh != NULL);
/* This should not happen, but let's be safe */
if (dvd->vd_lh == NULL) {
/* What are we going to do here??? */
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(ENXIO));
}
goto skip_open;
}
/*
* Create dvd to be used as vd->vdev_tsd.
*/
vd->vdev_tsd = dvd = vdev_disk_alloc();
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(EINVAL));
}
}
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
error = ldi_open_by_name(buf, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
if (error == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
} else {
kmem_free(buf, len);
}
}
/*
* If we have not yet opened the device, try to open it by the
* specified path.
*/
if (error != 0) {
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = SET_ERROR(EINVAL);
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL) {
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
}
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
if (vd->vdev_devid != NULL)
validate_devid = B_TRUE;
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (error);
}
/*
* Now that the device has been successfully opened, update the devid
* if necessary.
*/
if (validate_devid && spa_writeable(spa) &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
char *vd_devid;
vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
zfs_dbgmsg("vdev %s: update devid from %s, "
"to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
spa_strfree(vd->vdev_devid);
vd->vdev_devid = spa_strdup(vd_devid);
ddi_devid_str_free(vd_devid);
}
ddi_devid_free(devid);
}
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
/*
* Register callbacks for the LDI offline event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
}
/*
* Register callbacks for the LDI degrade event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
}
/* Reset TRIM flag, as underlying device support may have changed */
vd->vdev_notrim = B_FALSE;
skip_open:
ASSERT(dvd != NULL);
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(EINVAL));
}
*max_psize = *psize;
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
(intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
capacity = dkmext->dki_capacity - 1;
blksz = dkmext->dki_lbsize;
pbsize = dkmext->dki_pbsize;
} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
(intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
VDEV_DEBUG(
"vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
vd->vdev_path);
capacity = dkm->dki_capacity - 1;
blksz = dkm->dki_lbsize;
pbsize = blksz;
} else {
VDEV_DEBUG("vdev_disk_open(\"%s\"): "
"both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
vd->vdev_path, error);
pbsize = DEV_BSIZE;
}
*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize += vdev_disk_get_space(vd, capacity, blksz);
zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
"max_psize %llu", vd->vdev_path, *psize,
*max_psize);
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* We are done with vd_lh and vdev_tsd, release the vdev_tsd_lock
*/
rw_exit(&vd->vdev_tsd_lock);
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
return (0);
}
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);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd;
struct dk_minfo_ext dkmext;
int error;
dev_t dev;
int otyp;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (vd->vdev_tsd != NULL) {
ASSERT(vd->vdev_reopening);
dvd = vd->vdev_tsd;
goto skip_open;
}
dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the save device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
}
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
ddi_devid_t devid;
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
ldi_handle_t lh;
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
if (ldi_open_by_name(buf, spa_mode(spa), kcred,
&lh, zfs_li) == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
(void) ldi_close(lh, spa_mode(spa), kcred);
} else {
kmem_free(buf, len);
}
}
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = EINVAL;
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL)
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (EINVAL);
}
/*
* If we own the whole disk, try to enable disk write caching.
* We ignore errors because it's OK if we can't do it.
*/
if (vd->vdev_wholedisk == 1) {
int wce = 1;
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
FKIOCTL, kcred, NULL) != 0)
dkmext.dki_pbsize = DEV_BSIZE;
*ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
return (0);
}
static void
sdp_gen_ioctl(queue_t *q, mblk_t *mp)
{
struct iocblk *iocp;
int32_t enable = 0;
int ret;
boolean_t priv = B_TRUE;
/* LINTED */
iocp = (struct iocblk *)mp->b_rptr;
switch (iocp->ioc_cmd) {
int32_t send_enable;
case SIOCSENABLESDP:
bcopy(mp->b_cont->b_rptr, &enable, sizeof (int));
send_enable = enable;
/*
* Check for root privs.
* if not net config privs - return state of system SDP
*/
if (secpolicy_net_config(CRED(), B_FALSE) != 0) {
priv = B_FALSE;
}
/*
* The sdpib driver is loaded if root enables sdp the
* first time (sdp_transport_handle is NULL). It is
* unloaded during the following first disable. At all
* other times for root as well as non-root users, the
* action of enabling/disabling sdp is simply acked.
*/
rw_enter(&sdp_transport_lock, RW_READER);
if ((send_enable == 1) &&
(sdp_transport_handle == NULL) &&
(priv == B_TRUE)) {
/* Initialize sdpib transport driver */
rw_exit(&sdp_transport_lock);
ret = sdp_open_sdpib_driver();
rw_enter(&sdp_transport_lock,
RW_READER);
if (ret != 0) {
/* Transport failed to load */
rw_exit(&sdp_transport_lock);
enable = 0;
goto done;
}
(void) ldi_ioctl(sdp_transport_handle,
iocp->ioc_cmd, (intptr_t)&send_enable,
FKIOCTL, CRED(), (int *)&enable);
} else if (sdp_transport_handle != NULL) {
(void) ldi_ioctl(sdp_transport_handle,
iocp->ioc_cmd, (intptr_t)&send_enable,
FKIOCTL, CRED(), (int *)&enable);
if (send_enable == 0 && priv == B_TRUE) {
(void) ldi_close(sdp_transport_handle,
FNDELAY, kcred);
sdp_transport_handle = NULL;
}
} else {
enable = 0;
}
rw_exit(&sdp_transport_lock);
done:
bcopy(&enable, mp->b_cont->b_rptr, sizeof (int));
/* ACK the ioctl */
mp->b_datap->db_type = M_IOCACK;
iocp->ioc_count = sizeof (int);
qreply(q, mp);
break;
default:
miocnak(q, mp, 0, ENOTSUP);
}
}
static int
lx_ptm_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
int *rvalp)
{
ldi_handle_t lh = lx_ptm_lh_lookup(DEVT_TO_INDEX(dev));
int err;
/*
* here we need to make sure that we never allow the
* I_SETSIG and I_ESETSIG ioctls to pass through. we
* do this because we can't support them.
*
* the native Solaris ptm device supports these ioctls because
* they are streams framework ioctls and all streams devices
* support them by default. these ioctls cause the current
* process to be registered with a stream and receive signals
* when certain stream events occur.
*
* a problem arises with cleanup of these registrations
* for layered drivers.
*
* normally the streams framework is notified whenever a
* process closes any reference to a stream and it goes ahead
* and cleans up these registrations. but actual device drivers
* are not notified when a process performs a close operation
* unless the process is closing the last opened reference to
* the device on the entire system.
*
* so while we could pass these ioctls on and allow processes
* to register for signal delivery, we would never receive
* any notification when those processes exit (or close a
* stream) and we wouldn't be able to unregister them.
*
* luckily these operations are streams specific and Linux
* doesn't support streams devices. so it doesn't actually
* seem like we need to support these ioctls. if it turns
* out that we do need to support them for some reason in
* the future, the current driver model will have to be
* enhanced to better support streams device layering.
*/
if ((cmd == I_SETSIG) || (cmd == I_ESETSIG))
return (EINVAL);
/*
* here we fake up support for TIOCPKT. Linux applications expect
* /etc/ptmx to support this ioctl, but on Solaris it doesn't.
* (it is supported on older bsd style ptys.) so we'll fake
* up support for it here.
*
* the reason that this ioctl is emulated here instead of in
* userland is that this ioctl affects the results returned
* from read() operations. if this ioctl was emulated in
* userland the brand library would need to intercept all
* read operations and check to see if pktio was enabled
* for the fd being read from. since this ioctl only needs
* to be supported on the ptmx device it makes more sense
* to support it here where we can easily update the results
* returned for read() operations performed on ourselves.
*/
if (cmd == TIOCPKT) {
int pktio;
if (ddi_copyin((void *)arg, &pktio, sizeof (pktio),
mode) != DDI_SUCCESS)
return (EFAULT);
if (pktio == 0)
lx_ptm_lh_pktio_set(DEVT_TO_INDEX(dev), 0);
else
lx_ptm_lh_pktio_set(DEVT_TO_INDEX(dev), 1);
return (0);
}
err = ldi_ioctl(lh, cmd, arg, mode, credp, rvalp);
return (err);
}
/*ARGSUSED*/
static int
lx_ptm_open(dev_t *devp, int flag, int otyp, cred_t *credp)
{
struct strioctl iocb;
ptmptsopencb_t ppocb = { NULL, NULL };
ldi_handle_t lh;
major_t maj, our_major = getmajor(*devp);
minor_t min, lastmin;
uint_t index, anchor = 1;
dev_t ptm_dev;
int err, rval = 0;
/*
* Don't support the FNDELAY flag and FNONBLOCK until we either
* find a Linux app that opens /dev/ptmx with the O_NDELAY
* or O_NONBLOCK flags explicitly, or until we create test cases
* to determine how reads of master terminal devices opened with
* these flags behave in different situations on Linux. Supporting
* these flags will involve enhancing our read implementation
* and changing the way it deals with EOF notifications.
*/
if (flag & (FNDELAY | FNONBLOCK))
return (ENOTSUP);
/*
* we're layered on top of the ptm driver so open that driver
* first. (note that we're opening /dev/ptmx in the global
* zone, not ourselves in the Linux zone.)
*/
err = ldi_open_by_name(LP_PTM_PATH, flag, credp, &lh, lps.lps_li);
if (err != 0)
return (err);
/* get the devt returned by the ptmx open */
err = ldi_get_dev(lh, &ptm_dev);
if (err != 0) {
(void) ldi_close(lh, flag, credp);
return (err);
}
/*
* we're a cloning driver so here's well change the devt that we
* return. the ptmx is also a cloning driver so we'll just use
* it's minor number as our minor number (it already manages it's
* minor name space so no reason to duplicate the effort.)
*/
index = getminor(ptm_dev);
*devp = makedevice(our_major, INDEX_TO_MINOR(index));
/* Get a callback function to query if the pts device is open. */
iocb.ic_cmd = PTMPTSOPENCB;
iocb.ic_timout = 0;
iocb.ic_len = sizeof (ppocb);
iocb.ic_dp = (char *)&ppocb;
err = ldi_ioctl(lh, I_STR, (intptr_t)&iocb, FKIOCTL, kcred, &rval);
if ((err != 0) || (rval != 0)) {
(void) ldi_close(lh, flag, credp);
return (EIO); /* XXX return something else here? */
}
ASSERT(ppocb.ppocb_func != NULL);
/*
* now setup autopush for the terminal slave device. this is
* necessary so that when a Linux program opens the device we
* can push required strmod modules onto the stream. in Solaris
* this is normally done by the application that actually
* allocates the terminal.
*/
maj = lps.lps_pts_major;
min = index;
lastmin = 0;
err = kstr_autopush(SET_AUTOPUSH, &maj, &min, &lastmin,
&anchor, lx_pts_mods);
if (err != 0) {
(void) ldi_close(lh, flag, credp);
return (EIO); /* XXX return something else here? */
}
/* save off this layered handle for future accesses */
lx_ptm_lh_insert(index, lh);
lx_ptm_lh_set_ppocb(index, &ppocb);
return (0);
}
