static void
zfs_purge_cases(fmd_hdl_t *hdl)
{
zfs_case_t *zcp;
uu_list_walk_t *walk;
libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
/*
* There is no way to open a pool by GUID, or lookup a vdev by GUID. No
* matter what we do, we're going to have to stomach an O(vdevs * cases)
* algorithm. In reality, both quantities are likely so small that
* neither will matter. Given that iterating over pools is more
* expensive than iterating over the in-memory case list, we opt for a
* 'present' flag in each case that starts off cleared. We then iterate
* over all pools, marking those that are still present, and removing
* those that aren't found.
*
* Note that we could also construct an FMRI and rely on
* fmd_nvl_fmri_present(), but this would end up doing the same search.
*/
/*
* Mark the cases as not present.
*/
for (zcp = uu_list_first(zfs_cases); zcp != NULL;
zcp = uu_list_next(zfs_cases, zcp))
zcp->zc_present = B_FALSE;
/*
* Iterate over all pools and mark the pools and vdevs found. If this
* fails (most probably because we're out of memory), then don't close
* any of the cases and we cannot be sure they are accurate.
*/
if (zpool_iter(zhdl, zfs_mark_pool, NULL) != 0)
return;
/*
* Remove those cases which were not found.
*/
walk = uu_list_walk_start(zfs_cases, UU_WALK_ROBUST);
while ((zcp = uu_list_walk_next(walk)) != NULL) {
if (!zcp->zc_present)
fmd_case_close(hdl, zcp->zc_case);
}
uu_list_walk_end(walk);
}
/*
* Solve a given ZFS case. This first checks to make sure the diagnosis is
* still valid, as well as cleaning up any pending timer associated with the
* case.
*/
static void
zfs_case_solve(fmd_hdl_t *hdl, zfs_case_t *zcp, const char *faultname,
boolean_t checkunusable)
{
libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
nvlist_t *detector, *fault;
boolean_t serialize;
nvlist_t *fmri, *fru;
topo_hdl_t *thp;
int err;
/*
* Construct the detector from the case data. The detector is in the
* ZFS scheme, and is either the pool or the vdev, depending on whether
* this is a vdev or pool fault.
*/
detector = fmd_nvl_alloc(hdl, FMD_SLEEP);
(void) nvlist_add_uint8(detector, FM_VERSION, ZFS_SCHEME_VERSION0);
(void) nvlist_add_string(detector, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS);
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_POOL,
zcp->zc_data.zc_pool_guid);
if (zcp->zc_data.zc_vdev_guid != 0) {
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_VDEV,
zcp->zc_data.zc_vdev_guid);
}
/*
* We also want to make sure that the detector (pool or vdev) properly
* reflects the diagnosed state, when the fault corresponds to internal
* ZFS state (i.e. not checksum or I/O error-induced). Otherwise, a
* device which was unavailable early in boot (because the driver/file
* wasn't available) and is now healthy will be mis-diagnosed.
*/
if (!fmd_nvl_fmri_present(hdl, detector) ||
(checkunusable && !fmd_nvl_fmri_unusable(hdl, detector))) {
fmd_case_close(hdl, zcp->zc_case);
nvlist_free(detector);
return;
}
fru = NULL;
if (zcp->zc_fru != NULL &&
(thp = fmd_hdl_topo_hold(hdl, TOPO_VERSION)) != NULL) {
/*
* If the vdev had an associated FRU, then get the FRU nvlist
* from the topo handle and use that in the suspect list. We
* explicitly lookup the FRU because the fmri reported from the
* kernel may not have up to date details about the disk itself
* (serial, part, etc).
*/
if (topo_fmri_str2nvl(thp, zcp->zc_fru, &fmri, &err) == 0) {
/*
* If the disk is part of the system chassis, but the
* FRU indicates a different chassis ID than our
* current system, then ignore the error. This
* indicates that the device was part of another
* cluster head, and for obvious reasons cannot be
* imported on this system.
*/
if (libzfs_fru_notself(zhdl, zcp->zc_fru)) {
fmd_case_close(hdl, zcp->zc_case);
nvlist_free(fmri);
fmd_hdl_topo_rele(hdl, thp);
nvlist_free(detector);
return;
}
/*
* If the device is no longer present on the system, or
* topo_fmri_fru() fails for other reasons, then fall
* back to the fmri specified in the vdev.
*/
if (topo_fmri_fru(thp, fmri, &fru, &err) != 0)
fru = fmd_nvl_dup(hdl, fmri, FMD_SLEEP);
nvlist_free(fmri);
}
fmd_hdl_topo_rele(hdl, thp);
}
fault = fmd_nvl_create_fault(hdl, faultname, 100, detector,
fru, detector);
fmd_case_add_suspect(hdl, zcp->zc_case, fault);
nvlist_free(fru);
fmd_case_solve(hdl, zcp->zc_case);
serialize = B_FALSE;
if (zcp->zc_data.zc_has_remove_timer) {
fmd_timer_remove(hdl, zcp->zc_remove_timer);
zcp->zc_data.zc_has_remove_timer = 0;
serialize = B_TRUE;
}
if (serialize)
zfs_case_serialize(hdl, zcp);
nvlist_free(detector);
}
