/*
* Active pool health status.
*
* To determine the status for a pool, we make several passes over the config,
* picking the most egregious error we find. In order of importance, we do the
* following:
*
* - Check for a complete and valid configuration
* - Look for any faulted or missing devices in a non-replicated config
* - Check for any data errors
* - Check for any faulted or missing devices in a replicated config
* - Look for any devices showing errors
* - Check for any resilvering devices
*
* There can obviously be multiple errors within a single pool, so this routine
* only picks the most damaging of all the current errors to report.
*/
static zpool_status_t
check_status(nvlist_t *config, boolean_t isimport, zpool_errata_t *erratap)
{
nvlist_t *nvroot;
vdev_stat_t *vs;
pool_scan_stat_t *ps = NULL;
uint_t vsc, psc;
uint64_t nerr;
uint64_t version;
uint64_t stateval;
uint64_t suspended;
uint64_t hostid = 0;
uint64_t errata = 0;
unsigned long system_hostid = gethostid() & 0xffffffff;
verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
&version) == 0);
verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
&nvroot) == 0);
verify(nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
(uint64_t **)&vs, &vsc) == 0);
verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
&stateval) == 0);
/*
* Currently resilvering a vdev
*/
(void) nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_SCAN_STATS,
(uint64_t **)&ps, &psc);
if (ps && ps->pss_func == POOL_SCAN_RESILVER &&
ps->pss_state == DSS_SCANNING)
return (ZPOOL_STATUS_RESILVERING);
/*
* Pool last accessed by another system.
*/
(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_HOSTID, &hostid);
if (hostid != 0 && (unsigned long)hostid != system_hostid &&
stateval == POOL_STATE_ACTIVE)
return (ZPOOL_STATUS_HOSTID_MISMATCH);
/*
* Newer on-disk version.
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
vs->vs_aux == VDEV_AUX_VERSION_NEWER)
return (ZPOOL_STATUS_VERSION_NEWER);
/*
* Unsupported feature(s).
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
vs->vs_aux == VDEV_AUX_UNSUP_FEAT) {
nvlist_t *nvinfo;
verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
&nvinfo) == 0);
if (nvlist_exists(nvinfo, ZPOOL_CONFIG_CAN_RDONLY))
return (ZPOOL_STATUS_UNSUP_FEAT_WRITE);
return (ZPOOL_STATUS_UNSUP_FEAT_READ);
}
/*
* Check that the config is complete.
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
vs->vs_aux == VDEV_AUX_BAD_GUID_SUM)
return (ZPOOL_STATUS_BAD_GUID_SUM);
/*
* Check whether the pool has suspended due to failed I/O.
*/
if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_SUSPENDED,
&suspended) == 0) {
if (suspended == ZIO_FAILURE_MODE_CONTINUE)
return (ZPOOL_STATUS_IO_FAILURE_CONTINUE);
return (ZPOOL_STATUS_IO_FAILURE_WAIT);
}
/*
* Could not read a log.
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
vs->vs_aux == VDEV_AUX_BAD_LOG) {
return (ZPOOL_STATUS_BAD_LOG);
}
/*
* Bad devices in non-replicated config.
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
find_vdev_problem(nvroot, vdev_faulted))
return (ZPOOL_STATUS_FAULTED_DEV_NR);
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
find_vdev_problem(nvroot, vdev_missing))
return (ZPOOL_STATUS_MISSING_DEV_NR);
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
find_vdev_problem(nvroot, vdev_broken))
return (ZPOOL_STATUS_CORRUPT_LABEL_NR);
/*
* Corrupted pool metadata
*/
if (vs->vs_state == VDEV_STATE_CANT_OPEN &&
vs->vs_aux == VDEV_AUX_CORRUPT_DATA)
return (ZPOOL_STATUS_CORRUPT_POOL);
/*
* Persistent data errors.
*/
if (!isimport) {
if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_ERRCOUNT,
&nerr) == 0 && nerr != 0)
return (ZPOOL_STATUS_CORRUPT_DATA);
}
/*
* Missing devices in a replicated config.
*/
if (find_vdev_problem(nvroot, vdev_faulted))
return (ZPOOL_STATUS_FAULTED_DEV_R);
if (find_vdev_problem(nvroot, vdev_missing))
return (ZPOOL_STATUS_MISSING_DEV_R);
if (find_vdev_problem(nvroot, vdev_broken))
return (ZPOOL_STATUS_CORRUPT_LABEL_R);
/*
* Devices with errors
*/
if (!isimport && find_vdev_problem(nvroot, vdev_errors))
return (ZPOOL_STATUS_FAILING_DEV);
/*
* Offlined devices
*/
if (find_vdev_problem(nvroot, vdev_offlined))
return (ZPOOL_STATUS_OFFLINE_DEV);
/*
* Removed device
*/
if (find_vdev_problem(nvroot, vdev_removed))
return (ZPOOL_STATUS_REMOVED_DEV);
/*
* Outdated, but usable, version
*/
if (SPA_VERSION_IS_SUPPORTED(version) && version != SPA_VERSION)
return (ZPOOL_STATUS_VERSION_OLDER);
/*
* Usable pool with disabled features
*/
if (version >= SPA_VERSION_FEATURES) {
int i;
nvlist_t *feat;
if (isimport) {
feat = fnvlist_lookup_nvlist(config,
ZPOOL_CONFIG_LOAD_INFO);
feat = fnvlist_lookup_nvlist(feat,
ZPOOL_CONFIG_ENABLED_FEAT);
} else {
feat = fnvlist_lookup_nvlist(config,
ZPOOL_CONFIG_FEATURE_STATS);
}
for (i = 0; i < SPA_FEATURES; i++) {
zfeature_info_t *fi = &spa_feature_table[i];
if (!nvlist_exists(feat, fi->fi_guid))
return (ZPOOL_STATUS_FEAT_DISABLED);
}
}
/*
* Informational errata available.
*/
(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_ERRATA, &errata);
if (errata) {
*erratap = errata;
return (ZPOOL_STATUS_ERRATA);
}
return (ZPOOL_STATUS_OK);
}
/*
* Function: it_config_commit()
*
* Informs the iscsit service that the configuration has changed and
* commits the new configuration to persistent store by calling
* stmfSetProviderData. This function can be called multiple times
* during a configuration sequence if necessary.
*
* Parameters:
* cfg A C representation of the current iSCSI configuration
*
* Return Values:
* 0 Success
* ENOMEM Could not allocate resources
* EINVAL Invalid it_config_t structure
* TBD ioctl() failed
* TBD could not save config to STMF
*/
int
it_config_commit(it_config_t *cfg)
{
int ret;
nvlist_t *cfgnv = NULL;
char *packednv = NULL;
int iscsit_fd = -1;
size_t pnv_size;
iscsit_ioc_set_config_t iop;
it_tgt_t *tgtp;
if (!cfg) {
return (EINVAL);
}
ret = it_config_to_nv(cfg, &cfgnv);
if (ret == 0) {
ret = nvlist_size(cfgnv, &pnv_size, NV_ENCODE_NATIVE);
}
/*
* If the iscsit service is enabled, send the changes to the
* kernel first. Kernel will be the final sanity check before
* the config is saved persistently.
*
* This somewhat leaves open the simultaneous-change hole
* that STMF was trying to solve, but is a better sanity
* check and allows for graceful handling of target renames.
*/
if ((ret == 0) && is_iscsit_enabled()) {
packednv = malloc(pnv_size);
if (!packednv) {
ret = ENOMEM;
} else {
ret = nvlist_pack(cfgnv, &packednv, &pnv_size,
NV_ENCODE_NATIVE, 0);
}
if (ret == 0) {
iscsit_fd = open(ISCSIT_NODE, O_RDWR|O_EXCL);
if (iscsit_fd != -1) {
iop.set_cfg_vers = ISCSIT_API_VERS0;
iop.set_cfg_pnvlist = packednv;
iop.set_cfg_pnvlist_len = pnv_size;
if ((ioctl(iscsit_fd, ISCSIT_IOC_SET_CONFIG,
&iop)) != 0) {
ret = errno;
}
(void) close(iscsit_fd);
} else {
ret = errno;
}
}
if (packednv != NULL) {
free(packednv);
}
}
/*
* Before saving the config persistently, remove any
* PROP_OLD_TARGET_NAME entries. This is only interesting to
* the active service.
*/
if (ret == 0) {
boolean_t changed = B_FALSE;
tgtp = cfg->config_tgt_list;
for (; tgtp != NULL; tgtp = tgtp->tgt_next) {
if (!tgtp->tgt_properties) {
continue;
}
if (nvlist_exists(tgtp->tgt_properties,
PROP_OLD_TARGET_NAME)) {
(void) nvlist_remove_all(tgtp->tgt_properties,
PROP_OLD_TARGET_NAME);
changed = B_TRUE;
}
}
if (changed) {
/* rebuild the config nvlist */
nvlist_free(cfgnv);
cfgnv = NULL;
ret = it_config_to_nv(cfg, &cfgnv);
}
}
/*
* stmfGetProviderDataProt() checks to ensure
* that the config data hasn't changed since we fetched it.
*
* The kernel now has a version we need to save persistently.
* CLI will 'do the right thing' and warn the user if it
* gets STMF_ERROR_PROV_DATA_STALE. We'll try once to revert
* the kernel to the persistently saved data, but ultimately,
* it's up to the administrator to validate things are as they
* want them to be.
*/
if (ret == 0) {
ret = stmfSetProviderDataProt(ISCSIT_MODNAME, cfgnv,
STMF_PORT_PROVIDER_TYPE, &(cfg->stmf_token));
if (ret == STMF_STATUS_SUCCESS) {
ret = 0;
} else if (ret == STMF_ERROR_NOMEM) {
ret = ENOMEM;
} else if (ret == STMF_ERROR_PROV_DATA_STALE) {
int st;
it_config_t *rcfg = NULL;
st = it_config_load(&rcfg);
if (st == 0) {
(void) it_config_commit(rcfg);
it_config_free(rcfg);
}
}
}
if (cfgnv) {
nvlist_free(cfgnv);
}
return (ret);
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
nvlist_t *nvl = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
#ifdef _KERNEL
fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
#endif
if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
zfs_dbgmsg("command: %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
} else {
zfs_dbgmsg("txg %lld %s %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
}
} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
zfs_dbgmsg("ioctl %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
}
VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
KM_PUSHPAGE), ==, 0);
mutex_enter(&spa->spa_history_lock);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
/* The first command is the create, which we keep forever */
if (ret == 0 && shpp->sh_pool_create_len == 0 &&
nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
}
mutex_exit(&spa->spa_history_lock);
fnvlist_pack_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
fnvlist_free(nvl);
}
/*
* Goes through the config property list and validates
* each entry. If errs is non-NULL, will return explicit errors
* for each property that fails validation.
*/
static int
it_validate_configprops(nvlist_t *nvl, nvlist_t *errs)
{
int errcnt = 0;
nvpair_t *nvp = NULL;
data_type_t nvtype;
char *name;
char *val;
struct sockaddr_storage sa;
boolean_t update_rad_server = B_FALSE;
char *rad_server;
char *auth = NULL;
if (!nvl) {
return (0);
}
while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
name = nvpair_name(nvp);
nvtype = nvpair_type(nvp);
if (!name) {
continue;
}
val = NULL;
/* prefetch string value as we mostly need it */
if (nvtype == DATA_TYPE_STRING) {
(void) nvpair_value_string(nvp, &val);
}
if (strcmp(name, PROP_ALIAS) == 0) {
if (!val) {
PROPERR(errs, name,
gettext("must be a string value"));
errcnt++;
}
} else if (strcmp(name, PROP_AUTH) == 0) {
if (!val) {
PROPERR(errs, name,
gettext("must be a string value"));
errcnt++;
continue;
}
if ((strcmp(val, PA_AUTH_NONE) != 0) &&
(strcmp(val, PA_AUTH_CHAP) != 0) &&
(strcmp(val, PA_AUTH_RADIUS) != 0)) {
PROPERR(errs, PROP_AUTH,
gettext("must be none, chap or radius"));
errcnt++;
}
auth = val;
} else if (strcmp(name, PROP_ISNS_ENABLED) == 0) {
if (nvtype != DATA_TYPE_BOOLEAN_VALUE) {
PROPERR(errs, name,
gettext("must be a boolean value"));
errcnt++;
}
} else if (strcmp(name, PROP_ISNS_SERVER) == 0) {
char **arr = NULL;
uint32_t acount = 0;
(void) nvlist_lookup_string_array(nvl, name,
&arr, &acount);
while (acount > 0) {
if (strcasecmp(arr[acount - 1], "none") == 0) {
break;
}
if ((it_common_convert_sa(arr[acount - 1],
&sa, 0)) == NULL) {
PROPERR(errs, arr[acount - 1],
gettext("invalid address"));
errcnt++;
}
acount--;
}
} else if (strcmp(name, PROP_RADIUS_SECRET) == 0) {
if (!val) {
PROPERR(errs, name,
gettext("must be a string value"));
errcnt++;
continue;
}
} else if (strcmp(name, PROP_RADIUS_SERVER) == 0) {
struct sockaddr_storage sa;
if (!val) {
PROPERR(errs, name,
gettext("must be a string value"));
errcnt++;
continue;
}
if ((it_common_convert_sa(val, &sa,
DEFAULT_RADIUS_PORT)) == NULL) {
PROPERR(errs, name,
gettext("invalid address"));
errcnt++;
} else {
/*
* rewrite this property to ensure port
* number is added.
*/
if (sockaddr_to_str(&sa, &rad_server) == 0) {
update_rad_server = B_TRUE;
}
}
} else {
/* unrecognized property */
PROPERR(errs, name, gettext("unrecognized property"));
errcnt++;
}
}
/*
* If we successfully reformatted the radius server to add the port
* number then update the nvlist
*/
if (update_rad_server) {
(void) nvlist_add_string(nvl, PROP_RADIUS_SERVER, rad_server);
free(rad_server);
}
/*
* if auth = radius, ensure radius server & secret are set.
*/
if (auth) {
if (strcmp(auth, PA_AUTH_RADIUS) == 0) {
/* need server & secret for radius */
if (!nvlist_exists(nvl, PROP_RADIUS_SERVER)) {
PROPERR(errs, PROP_RADIUS_SERVER,
gettext("missing required property"));
errcnt++;
}
if (!nvlist_exists(nvl, PROP_RADIUS_SECRET)) {
PROPERR(errs, PROP_RADIUS_SECRET,
gettext("missing required property"));
errcnt++;
}
}
}
if (errcnt) {
return (EINVAL);
}
return (0);
}
