+
+#define LOV_U64_MAX ((__u64)~0ULL)
+#define LOV_SUM_MAX(tot, add) \
+ do { \
+ if ((tot) + (add) < (tot)) \
+ (tot) = LOV_U64_MAX; \
+ else \
+ (tot) += (add); \
+ } while(0)
+
+int lov_fini_statfs(struct obd_device *obd, struct obd_statfs *osfs,int success)
+{
+ ENTRY;
+
+ if (success) {
+ __u32 expected_stripes = lov_get_stripecnt(&obd->u.lov,
+ LOV_MAGIC, 0);
+ if (osfs->os_files != LOV_U64_MAX)
+ do_div(osfs->os_files, expected_stripes);
+ if (osfs->os_ffree != LOV_U64_MAX)
+ do_div(osfs->os_ffree, expected_stripes);
+
+ cfs_spin_lock(&obd->obd_osfs_lock);
+ memcpy(&obd->obd_osfs, osfs, sizeof(*osfs));
+ obd->obd_osfs_age = cfs_time_current_64();
+ cfs_spin_unlock(&obd->obd_osfs_lock);
+ RETURN(0);
+ }
+
+ RETURN(-EIO);
+}
+
+int lov_fini_statfs_set(struct lov_request_set *set)
+{
+ int rc = 0;
+ ENTRY;
+
+ if (set == NULL)
+ RETURN(0);
+
+ if (cfs_atomic_read(&set->set_completes)) {
+ rc = lov_fini_statfs(set->set_obd, set->set_oi->oi_osfs,
+ cfs_atomic_read(&set->set_success));
+ }
+ lov_put_reqset(set);
+ RETURN(rc);
+}
+
+void lov_update_statfs(struct obd_statfs *osfs, struct obd_statfs *lov_sfs,
+ int success)
+{
+ int shift = 0, quit = 0;
+ __u64 tmp;
+
+ if (success == 0) {
+ memcpy(osfs, lov_sfs, sizeof(*lov_sfs));
+ } else {
+ if (osfs->os_bsize != lov_sfs->os_bsize) {
+ /* assume all block sizes are always powers of 2 */
+ /* get the bits difference */
+ tmp = osfs->os_bsize | lov_sfs->os_bsize;
+ for (shift = 0; shift <= 64; ++shift) {
+ if (tmp & 1) {
+ if (quit)
+ break;
+ else
+ quit = 1;
+ shift = 0;
+ }
+ tmp >>= 1;
+ }
+ }
+
+ if (osfs->os_bsize < lov_sfs->os_bsize) {
+ osfs->os_bsize = lov_sfs->os_bsize;
+
+ osfs->os_bfree >>= shift;
+ osfs->os_bavail >>= shift;
+ osfs->os_blocks >>= shift;
+ } else if (shift != 0) {
+ lov_sfs->os_bfree >>= shift;
+ lov_sfs->os_bavail >>= shift;
+ lov_sfs->os_blocks >>= shift;
+ }
+#ifdef MIN_DF
+ /* Sandia requested that df (and so, statfs) only
+ returned minimal available space on
+ a single OST, so people would be able to
+ write this much data guaranteed. */
+ if (osfs->os_bavail > lov_sfs->os_bavail) {
+ /* Presumably if new bavail is smaller,
+ new bfree is bigger as well */
+ osfs->os_bfree = lov_sfs->os_bfree;
+ osfs->os_bavail = lov_sfs->os_bavail;
+ }
+#else
+ osfs->os_bfree += lov_sfs->os_bfree;
+ osfs->os_bavail += lov_sfs->os_bavail;
+#endif
+ osfs->os_blocks += lov_sfs->os_blocks;
+ /* XXX not sure about this one - depends on policy.
+ * - could be minimum if we always stripe on all OBDs
+ * (but that would be wrong for any other policy,
+ * if one of the OBDs has no more objects left)
+ * - could be sum if we stripe whole objects
+ * - could be average, just to give a nice number
+ *
+ * To give a "reasonable" (if not wholly accurate)
+ * number, we divide the total number of free objects
+ * by expected stripe count (watch out for overflow).
+ */
+ LOV_SUM_MAX(osfs->os_files, lov_sfs->os_files);
+ LOV_SUM_MAX(osfs->os_ffree, lov_sfs->os_ffree);
+ }
+}
+
+/* The callback for osc_statfs_async that finilizes a request info when a
+ * response is received. */
+static int cb_statfs_update(void *cookie, int rc)
+{
+ struct obd_info *oinfo = cookie;
+ struct lov_request *lovreq;
+ struct lov_request_set *set;
+ struct obd_statfs *osfs, *lov_sfs;
+ struct lov_obd *lov;
+ struct lov_tgt_desc *tgt;
+ struct obd_device *lovobd, *tgtobd;
+ int success;
+ ENTRY;
+
+ lovreq = container_of(oinfo, struct lov_request, rq_oi);
+ set = lovreq->rq_rqset;
+ lovobd = set->set_obd;
+ lov = &lovobd->u.lov;
+ osfs = set->set_oi->oi_osfs;
+ lov_sfs = oinfo->oi_osfs;
+ success = cfs_atomic_read(&set->set_success);
+ /* XXX: the same is done in lov_update_common_set, however
+ lovset->set_exp is not initialized. */
+ lov_update_set(set, lovreq, rc);
+ if (rc)
+ GOTO(out, rc);
+
+ obd_getref(lovobd);
+ tgt = lov->lov_tgts[lovreq->rq_idx];
+ if (!tgt || !tgt->ltd_active)
+ GOTO(out_update, rc);
+
+ tgtobd = class_exp2obd(tgt->ltd_exp);
+ cfs_spin_lock(&tgtobd->obd_osfs_lock);
+ memcpy(&tgtobd->obd_osfs, lov_sfs, sizeof(*lov_sfs));
+ if ((oinfo->oi_flags & OBD_STATFS_FROM_CACHE) == 0)
+ tgtobd->obd_osfs_age = cfs_time_current_64();
+ cfs_spin_unlock(&tgtobd->obd_osfs_lock);
+
+out_update:
+ lov_update_statfs(osfs, lov_sfs, success);
+ qos_update(lov);
+ obd_putref(lovobd);
+
+out:
+ if (set->set_oi->oi_flags & OBD_STATFS_PTLRPCD &&
+ lov_finished_set(set)) {
+ lov_statfs_interpret(NULL, set, set->set_count !=
+ cfs_atomic_read(&set->set_success));
+ if (lov->lov_qos.lq_statfs_in_progress)
+ qos_statfs_done(lov);
+ }
+
+ RETURN(0);
+}
+
+int lov_prep_statfs_set(struct obd_device *obd, struct obd_info *oinfo,
+ struct lov_request_set **reqset)
+{
+ struct lov_request_set *set;
+ struct lov_obd *lov = &obd->u.lov;
+ int rc = 0, i;
+ ENTRY;
+
+ OBD_ALLOC(set, sizeof(*set));
+ if (set == NULL)
+ RETURN(-ENOMEM);
+ lov_init_set(set);
+
+ set->set_obd = obd;
+ set->set_oi = oinfo;
+
+ /* We only get block data from the OBD */
+ for (i = 0; i < lov->desc.ld_tgt_count; i++) {
+ struct lov_request *req;
+
+ if (!lov->lov_tgts[i] || (!lov->lov_tgts[i]->ltd_active
+ && (oinfo->oi_flags & OBD_STATFS_NODELAY))) {
+ CDEBUG(D_HA, "lov idx %d inactive\n", i);
+ continue;
+ }
+
+ /* skip targets that have been explicitely disabled by the
+ * administrator */
+ if (!lov->lov_tgts[i]->ltd_exp) {
+ CDEBUG(D_HA, "lov idx %d administratively disabled\n", i);
+ continue;
+ }
+
+ OBD_ALLOC(req, sizeof(*req));
+ if (req == NULL)
+ GOTO(out_set, rc = -ENOMEM);
+
+ OBD_ALLOC(req->rq_oi.oi_osfs, sizeof(*req->rq_oi.oi_osfs));
+ if (req->rq_oi.oi_osfs == NULL) {
+ OBD_FREE(req, sizeof(*req));
+ GOTO(out_set, rc = -ENOMEM);
+ }
+
+ req->rq_idx = i;
+ req->rq_oi.oi_cb_up = cb_statfs_update;
+ req->rq_oi.oi_flags = oinfo->oi_flags;
+
+ lov_set_add_req(req, set);
+ }
+ if (!set->set_count)
+ GOTO(out_set, rc = -EIO);
+ *reqset = set;
+ RETURN(rc);
+out_set:
+ lov_fini_statfs_set(set);
+ RETURN(rc);
+}