/*
* Free a vdev_t that has been removed from service.
*/
static void
vdev_free_common(vdev_t *vd)
{
spa_t *spa = vd->vdev_spa;
if (vd->vdev_path)
spa_strfree(vd->vdev_path);
if (vd->vdev_devid)
spa_strfree(vd->vdev_devid);
if (vd->vdev_isspare)
spa_spare_remove(vd);
txg_list_destroy(&vd->vdev_ms_list);
txg_list_destroy(&vd->vdev_dtl_list);
mutex_enter(&vd->vdev_dtl_lock);
space_map_unload(&vd->vdev_dtl_map);
space_map_destroy(&vd->vdev_dtl_map);
space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL);
space_map_destroy(&vd->vdev_dtl_scrub);
mutex_exit(&vd->vdev_dtl_lock);
mutex_destroy(&vd->vdev_dtl_lock);
mutex_destroy(&vd->vdev_stat_lock);
if (vd == spa->spa_root_vdev)
spa->spa_root_vdev = NULL;
kmem_free(vd, sizeof (vdev_t));
}
/*
* Determine if the underlying device is accessible by reading and writing
* to a known location. We must be able to do this during syncing context
* and thus we cannot set the vdev state directly.
*/
static int
vdev_file_probe(vdev_t *vd)
{
vdev_t *nvd;
char *vl_boot;
uint64_t offset;
int l, error = 0, retries = 0;
if (vd == NULL)
return (EINVAL);
/* Hijack the current vdev */
nvd = vd;
/*
* Pick a random label to rewrite.
*/
l = spa_get_random(VDEV_LABELS);
ASSERT(l < VDEV_LABELS);
offset = vdev_label_offset(vd->vdev_psize, l,
offsetof(vdev_label_t, vl_boot_header));
vl_boot = kmem_alloc(VDEV_BOOT_HEADER_SIZE, KM_SLEEP);
while ((error = vdev_file_probe_io(nvd, vl_boot, VDEV_BOOT_HEADER_SIZE,
offset, UIO_READ)) != 0 && retries == 0) {
/*
* If we failed with the vdev that was passed in then
* try allocating a new one and try again.
*/
nvd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
if (vd->vdev_path)
nvd->vdev_path = spa_strdup(vd->vdev_path);
nvd->vdev_guid = vd->vdev_guid;
retries++;
if (vdev_file_open_common(nvd) != 0)
break;
}
if ((spa_mode & FWRITE) && !error) {
error = vdev_file_probe_io(nvd, vl_boot, VDEV_BOOT_HEADER_SIZE,
offset, UIO_WRITE);
}
if (retries) {
vdev_file_close(nvd);
if (nvd->vdev_path)
spa_strfree(nvd->vdev_path);
kmem_free(nvd, sizeof (vdev_t));
}
kmem_free(vl_boot, VDEV_BOOT_HEADER_SIZE);
if (!error)
vd->vdev_is_failing = B_FALSE;
return (error);
}
void
dsl_prop_fini(dsl_dir_t *dd)
{
dsl_prop_record_t *pr;
while ((pr = list_remove_head(&dd->dd_props)) != NULL) {
list_destroy(&pr->pr_cbs);
spa_strfree((char *)pr->pr_propname);
kmem_free(pr, sizeof (dsl_prop_record_t));
}
list_destroy(&dd->dd_props);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held. Synchronizing the pool cache is typically done after
* the configuration has been synced to the MOS. This exposes a window where
* the MOS config will have been updated but the cache file has not. If
* the system were to crash at that instant then the cached config may not
* contain the correct information to open the pool and an explicity import
* would be required.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
boolean_t ccw_failure;
int error;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
if (rootdir == NULL || !(spa_mode_global & FWRITE))
return;
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
ccw_failure = B_FALSE;
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa_t *spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
nvlist_t *nvroot = NULL;
/*
* Skip over our own pool if we're about to remove
* ourselves from the spa namespace or any pool that
* is readonly. Since we cannot guarantee that a
* readonly pool would successfully import upon reboot,
* we don't allow them to be written to the cache file.
*/
if ((spa == target && removing) ||
(spa_state(spa) == POOL_STATE_ACTIVE &&
!spa_writeable(spa)))
continue;
mutex_enter(&spa->spa_props_lock);
tdp = list_head(&spa->spa_config_list);
if (spa->spa_config == NULL ||
tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0) {
mutex_exit(&spa->spa_props_lock);
continue;
}
if (nvl == NULL)
VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
KM_SLEEP) == 0);
VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
spa->spa_config) == 0);
mutex_exit(&spa->spa_props_lock);
if (nvlist_lookup_nvlist(nvl, spa->spa_name, &nvroot) == 0)
spa_config_clean(nvroot);
}
error = spa_config_write(dp, nvl);
if (error != 0)
ccw_failure = B_TRUE;
nvlist_free(nvl);
}
if (ccw_failure) {
/*
* Keep trying so that configuration data is
* written if/when any temporary filesystem
* resource issues are resolved.
*/
if (target->spa_ccw_fail_time == 0) {
zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
target, NULL, NULL, 0, 0);
}
target->spa_ccw_fail_time = gethrtime();
spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
} else {
/*
* Do not rate limit future attempts to update
* the config cache.
*/
target->spa_ccw_fail_time = 0;
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held. Synchronizing the pool cache is typically done after
* the configuration has been synced to the MOS. This exposes a window where
* the MOS config will have been updated but the cache file has not. If
* the system were to crash at that instant then the cached config may not
* contain the correct information to open the pool and an explicity import
* would be required.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
char *pool_name;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
if (rootdir == NULL || !(spa_mode_global & FWRITE))
return;
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa_t *spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
/*
* Skip over our own pool if we're about to remove
* ourselves from the spa namespace or any pool that
* is readonly. Since we cannot guarantee that a
* readonly pool would successfully import upon reboot,
* we don't allow them to be written to the cache file.
*/
if ((spa == target && removing) ||
!spa_writeable(spa))
continue;
mutex_enter(&spa->spa_props_lock);
tdp = list_head(&spa->spa_config_list);
if (spa->spa_config == NULL ||
tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0) {
mutex_exit(&spa->spa_props_lock);
continue;
}
if (nvl == NULL)
VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
KM_SLEEP) == 0);
if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
VERIFY0(nvlist_lookup_string(spa->spa_config,
ZPOOL_CONFIG_POOL_NAME, &pool_name));
} else
pool_name = spa_name(spa);
VERIFY(nvlist_add_nvlist(nvl, pool_name,
spa->spa_config) == 0);
mutex_exit(&spa->spa_props_lock);
}
spa_config_write(dp, nvl);
nvlist_free(nvl);
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
if (rootdir == NULL || !(spa_mode_global & FWRITE))
return;
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa_t *spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
if (spa == target && removing)
continue;
mutex_enter(&spa->spa_props_lock);
tdp = list_head(&spa->spa_config_list);
if (spa->spa_config == NULL ||
tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0) {
mutex_exit(&spa->spa_props_lock);
continue;
}
if (nvl == NULL)
VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
KM_SLEEP) == 0);
VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
spa->spa_config) == 0);
mutex_exit(&spa->spa_props_lock);
}
spa_config_write(dp, nvl);
nvlist_free(nvl);
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
}
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 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_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);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_t *spa = NULL;
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
if (spa->spa_config == NULL || spa->spa_name == NULL)
continue;
if (spa == target && removing)
continue;
#ifdef __APPLE__
/* OS X - Omit disk based pools */
if (vdev_contains_disks(spa->spa_root_vdev))
continue;
#endif
tdp = list_head(&spa->spa_config_list);
ASSERT(tdp != NULL);
if (tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0)
continue;
if (nvl == NULL)
VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
KM_SLEEP) == 0);
VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
spa->spa_config) == 0);
}
spa_config_write(dp, nvl);
nvlist_free(nvl);
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
}
