+struct osd_metadnode_it {
+ struct osd_device *mit_dev;
+ __u64 mit_pos;
+ struct lu_fid mit_fid;
+ int mit_prefetched;
+ __u64 mit_prefetched_dnode;
+};
+
+static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
+ struct dt_object *dt, __u32 attr,
+ struct lustre_capa *capa)
+{
+ struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
+ struct osd_metadnode_it *it;
+ ENTRY;
+
+ OBD_ALLOC_PTR(it);
+ if (unlikely(it == NULL))
+ RETURN(ERR_PTR(-ENOMEM));
+
+ it->mit_dev = dev;
+
+ RETURN((struct dt_it *)it);
+}
+
+static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ OBD_FREE_PTR(it);
+}
+
+static int osd_zfs_otable_it_get(const struct lu_env *env,
+ struct dt_it *di, const struct dt_key *key)
+{
+ return 0;
+}
+
+static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
+{
+}
+
+#define OTABLE_PREFETCH 256
+
+static void osd_zfs_otable_prefetch(const struct lu_env *env,
+ struct osd_metadnode_it *it)
+{
+ struct osd_device *dev = it->mit_dev;
+ udmu_objset_t *uos = &dev->od_objset;
+ int rc;
+
+ /* can go negative on the very first access to the iterator
+ * or if some non-Lustre objects were found */
+ if (unlikely(it->mit_prefetched < 0))
+ it->mit_prefetched = 0;
+
+ if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
+ return;
+
+ if (it->mit_prefetched_dnode == 0)
+ it->mit_prefetched_dnode = it->mit_pos;
+
+ while (it->mit_prefetched < OTABLE_PREFETCH) {
+ rc = -dmu_object_next(uos->os, &it->mit_prefetched_dnode,
+ B_FALSE, 0);
+ if (unlikely(rc != 0))
+ break;
+
+ /* dmu_prefetch() was exported in 0.6.2, if you use with
+ * an older release, just comment it out - this is an
+ * optimization */
+ dmu_prefetch(uos->os, it->mit_prefetched_dnode, 0, 0);
+
+ it->mit_prefetched++;
+ }
+}
+
+static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+ struct lustre_mdt_attrs *lma;
+ struct osd_device *dev = it->mit_dev;
+ udmu_objset_t *uos = &dev->od_objset;
+ nvlist_t *nvbuf = NULL;
+ uchar_t *v;
+ __u64 dnode;
+ int rc, s;
+ ENTRY;
+
+ memset(&it->mit_fid, 0, sizeof(it->mit_fid));
+
+ dnode = it->mit_pos;
+ do {
+ rc = -dmu_object_next(uos->os, &it->mit_pos, B_FALSE, 0);
+ if (unlikely(rc != 0))
+ GOTO(out, rc = 1);
+ it->mit_prefetched--;
+
+ /* LMA is required for this to be a Lustre object.
+ * If there is no xattr skip it. */
+ rc = __osd_xattr_load(uos, it->mit_pos, &nvbuf);
+ if (unlikely(rc != 0))
+ continue;
+
+ LASSERT(nvbuf != NULL);
+ rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
+ if (likely(rc == 0)) {
+ /* Lustre object */
+ lma = (struct lustre_mdt_attrs *)v;
+ lustre_lma_swab(lma);
+ it->mit_fid = lma->lma_self_fid;
+ nvlist_free(nvbuf);
+ break;
+ } else {
+ /* not a Lustre object, try next one */
+ nvlist_free(nvbuf);
+ }
+
+ } while (1);
+
+
+ /* we aren't prefetching in the above loop because the number of
+ * non-Lustre objects is very small and we will be repeating very
+ * rare. in case we want to use this to iterate over non-Lustre
+ * objects (i.e. when we convert regular ZFS in Lustre) it makes
+ * sense to initiate prefetching in the loop */
+
+ /* 0 - there are more items, +1 - the end */
+ if (likely(rc == 0))
+ osd_zfs_otable_prefetch(env, it);
+
+ CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
+ it->mit_pos, PFID(&it->mit_fid), rc);
+
+out:
+ RETURN(rc);
+}
+
+static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ return NULL;
+}
+
+static int osd_zfs_otable_it_key_size(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ return sizeof(__u64);
+}
+
+static int osd_zfs_otable_it_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ struct dt_rec *rec, __u32 attr)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+ struct lu_fid *fid = (struct lu_fid *)rec;
+ ENTRY;
+
+ *fid = it->mit_fid;
+
+ RETURN(0);
+}
+
+
+static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ return it->mit_pos;
+}
+
+static int osd_zfs_otable_it_load(const struct lu_env *env,
+ const struct dt_it *di, __u64 hash)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ it->mit_pos = hash;
+ it->mit_prefetched = 0;
+ it->mit_prefetched_dnode = 0;
+
+ return osd_zfs_otable_it_next(env, (struct dt_it *)di);
+}
+
+static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
+ const struct dt_it *di, void *key_rec)
+{
+ return 0;
+}
+
+const struct dt_index_operations osd_zfs_otable_ops = {
+ .dio_it = {
+ .init = osd_zfs_otable_it_init,
+ .fini = osd_zfs_otable_it_fini,
+ .get = osd_zfs_otable_it_get,
+ .put = osd_zfs_otable_it_put,
+ .next = osd_zfs_otable_it_next,
+ .key = osd_zfs_otable_it_key,
+ .key_size = osd_zfs_otable_it_key_size,
+ .rec = osd_zfs_otable_it_rec,
+ .store = osd_zfs_otable_it_store,
+ .load = osd_zfs_otable_it_load,
+ .key_rec = osd_zfs_otable_it_key_rec,
+ }
+};
+
git://git.whamcloud.com / fs / lustre-release.git / commitdiff