X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_index.c;h=7fb832659db77ba90248aaa345b134857cbc3894;hb=17797975b27e1b5222fb75cac49d90066091486d;hp=6ebd0ab080ddc7a8a246106f0088db22907d1bb5;hpb=11db1a551172f596d1d284e8496530f9ce24ac81;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_index.c b/lustre/osd-zfs/osd_index.c
index 6ebd0ab..7fb8326 100644
--- a/lustre/osd-zfs/osd_index.c
+++ b/lustre/osd-zfs/osd_index.c
@@ -472,25 +472,40 @@ static inline void osd_object_put(const struct lu_env *env,
 	lu_object_put(env, &obj->oo_dt.do_lu);
 }
 
-static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
-			  struct lu_fid *fid)
+static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
+			  obd_seq seq)
 {
 	struct lu_seq_range	*range = &osd_oti_get(env)->oti_seq_range;
 	struct seq_server_site	*ss = osd_seq_site(osd);
 	int			rc;
 	ENTRY;
 
-	if (!fid_is_norm(fid) && !fid_is_root(fid))
-		RETURN(0);
+	if (ss == NULL)
+		RETURN(1);
 
-	rc = osd_fld_lookup(env, osd, fid, range);
+	rc = osd_fld_lookup(env, osd, seq, range);
 	if (rc != 0) {
-		CERROR("%s: Can not lookup fld for "DFID"\n",
-		       osd_name(osd), PFID(fid));
-		RETURN(rc);
+		CERROR("%s: Can not lookup fld for "LPX64"\n",
+		       osd_name(osd), seq);
+		RETURN(0);
 	}
 
-	RETURN(ss->ss_node_id != range->lsr_index);
+	RETURN(ss->ss_node_id == range->lsr_index);
+}
+
+static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
+			  struct lu_fid *fid)
+{
+	ENTRY;
+
+	/* FID seqs not in FLDB, must be local seq */
+	if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
+		RETURN(0);
+
+	if (osd_seq_exists(env, osd, fid_seq(fid)))
+		RETURN(0);
+
+	RETURN(1);
 }
 
 /**
@@ -918,11 +933,11 @@ static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
 
 	LASSERT(lde);
 
-	lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
-
-	if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
+	rc = -zap_cursor_retrieve(it->ozi_zc, za);
+	if (unlikely(rc != 0))
 		GOTO(out, rc);
 
+	lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
 	namelen = strlen(za->za_name);
 	if (namelen > NAME_MAX)
 		GOTO(out, rc = -EOVERFLOW);
@@ -954,6 +969,44 @@ out:
 	RETURN(rc);
 }
 
+static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
+			       __u32 attr)
+{
+	struct osd_zap_it   *it = (struct osd_zap_it *)di;
+	zap_attribute_t     *za = &osd_oti_get(env)->oti_za;
+	int		     rc, namelen = 0;
+	ENTRY;
+
+	if (it->ozi_pos <= 1)
+		namelen = 1;
+	else if (it->ozi_pos == 2)
+		namelen = 2;
+
+	if (namelen > 0) {
+		rc = lu_dirent_calc_size(namelen, attr);
+		RETURN(rc);
+	}
+
+	rc = -zap_cursor_retrieve(it->ozi_zc, za);
+	if (unlikely(rc != 0))
+		RETURN(rc);
+
+	if (za->za_integer_length != 8 || za->za_num_integers < 3) {
+		CERROR("%s: unsupported direntry format: %d %d\n",
+		       osd_obj2dev(it->ozi_obj)->od_svname,
+		       za->za_integer_length, (int)za->za_num_integers);
+		RETURN(-EIO);
+	}
+
+	namelen = strlen(za->za_name);
+	if (namelen > NAME_MAX)
+		RETURN(-EOVERFLOW);
+
+	rc = lu_dirent_calc_size(namelen, attr);
+
+	RETURN(rc);
+}
+
 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
 {
 	struct osd_zap_it *it = (struct osd_zap_it *)di;
@@ -1020,6 +1073,7 @@ static struct dt_index_operations osd_dir_ops = {
 		.key      = osd_dir_it_key,
 		.key_size = osd_dir_it_key_size,
 		.rec      = osd_dir_it_rec,
+		.rec_size = osd_dir_it_rec_size,
 		.store    = osd_dir_it_store,
 		.load     = osd_dir_it_load
 	}
@@ -1027,20 +1081,44 @@ static struct dt_index_operations osd_dir_ops = {
 
 /*
  * Primitives for index files using binary keys.
- * XXX: only 64-bit keys are supported for now.
  */
 
+/* key integer_size is 8 */
+static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
+				  const struct dt_key *src)
+{
+	int size;
+
+	LASSERT(dst);
+	LASSERT(src);
+
+	/* align keysize to 64bit */
+	size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
+	size *= sizeof(__u64);
+
+	LASSERT(size <= MAXNAMELEN);
+
+	if (unlikely(size > o->oo_keysize))
+		memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
+	memcpy(dst, (const char *)src, o->oo_keysize);
+
+	return (size/sizeof(__u64));
+}
+
 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
 			struct dt_rec *rec, const struct dt_key *key,
 			struct lustre_capa *capa)
 {
 	struct osd_object *obj = osd_dt_obj(dt);
 	struct osd_device *osd = osd_obj2dev(obj);
+	__u64		  *k = osd_oti_get(env)->oti_key64;
 	int                rc;
 	ENTRY;
 
+	rc = osd_prepare_key_uint64(obj, k, key);
+
 	rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
-				(const __u64 *)key, 1, 8, obj->oo_recsize,
+				k, rc, obj->oo_recusize, obj->oo_recsize,
 				(void *)rec);
 	RETURN(rc == 0 ? 1 : rc);
 }
@@ -1078,6 +1156,7 @@ static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
 	struct osd_object  *obj = osd_dt_obj(dt);
 	struct osd_device  *osd = osd_obj2dev(obj);
 	struct osd_thandle *oh;
+	__u64		   *k = osd_oti_get(env)->oti_key64;
 	int                 rc;
 	ENTRY;
 
@@ -1088,9 +1167,11 @@ static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
 
 	oh = container_of0(th, struct osd_thandle, ot_super);
 
+	rc = osd_prepare_key_uint64(obj, k, key);
+
 	/* Insert (key,oid) into ZAP */
 	rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
-			     (const __u64 *)key, 1, 8, obj->oo_recsize,
+			     k, rc, obj->oo_recusize, obj->oo_recsize,
 			     (void *)rec, oh->ot_tx);
 	RETURN(rc);
 }
@@ -1122,6 +1203,7 @@ static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
 	struct osd_object  *obj = osd_dt_obj(dt);
 	struct osd_device  *osd = osd_obj2dev(obj);
 	struct osd_thandle *oh;
+	__u64		   *k = osd_oti_get(env)->oti_key64;
 	int                 rc;
 	ENTRY;
 
@@ -1129,9 +1211,11 @@ static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
 	LASSERT(th != NULL);
 	oh = container_of0(th, struct osd_thandle, ot_super);
 
+	rc = osd_prepare_key_uint64(obj, k, key);
+
 	/* Remove binary key from the ZAP */
 	rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
-				(const __u64 *)key, 1, oh->ot_tx);
+				k, rc, oh->ot_tx);
 	RETURN(rc);
 }
 
@@ -1146,8 +1230,12 @@ static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
 	LASSERT(it);
 	LASSERT(it->ozi_zc);
 
-	/* XXX: API is broken at the moment */
-	LASSERT(*((const __u64 *)key) == 0);
+	/*
+	 * XXX: we need a binary version of zap_cursor_move_to_key()
+	 *	to implement this API */
+	if (*((const __u64 *)key) != 0)
+		CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
+		       *((__u64 *)key));
 
 	zap_cursor_fini(it->ozi_zc);
 	memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
@@ -1185,6 +1273,7 @@ static struct dt_key *osd_index_it_key(const struct lu_env *env,
 				       const struct dt_it *di)
 {
 	struct osd_zap_it *it = (struct osd_zap_it *)di;
+	struct osd_object *obj = it->ozi_obj;
 	zap_attribute_t   *za = &osd_oti_get(env)->oti_za;
 	int                rc = 0;
 	ENTRY;
@@ -1195,7 +1284,7 @@ static struct dt_key *osd_index_it_key(const struct lu_env *env,
 		RETURN(ERR_PTR(rc));
 
 	/* the binary key is stored in the name */
-	it->ozi_key = *((__u64 *)za->za_name);
+	memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
 
 	RETURN((struct dt_key *)&it->ozi_key);
 }
@@ -1203,8 +1292,9 @@ static struct dt_key *osd_index_it_key(const struct lu_env *env,
 static int osd_index_it_key_size(const struct lu_env *env,
 				const struct dt_it *di)
 {
-	/* we only support 64-bit binary keys for the time being */
-	RETURN(sizeof(__u64));
+	struct osd_zap_it *it = (struct osd_zap_it *)di;
+	struct osd_object *obj = it->ozi_obj;
+	RETURN(obj->oo_keysize);
 }
 
 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
@@ -1214,6 +1304,7 @@ static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
 	struct osd_zap_it *it = (struct osd_zap_it *)di;
 	struct osd_object *obj = it->ozi_obj;
 	struct osd_device *osd = osd_obj2dev(obj);
+	__u64		  *k = osd_oti_get(env)->oti_key64;
 	int                rc;
 	ENTRY;
 
@@ -1222,9 +1313,11 @@ static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
 	if (rc)
 		RETURN(rc);
 
+	rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
+
 	rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
-				(const __u64 *)za->za_name, 1, 8,
-				obj->oo_recsize, (void *)rec);
+				k, rc, obj->oo_recusize, obj->oo_recsize,
+				(void *)rec);
 	RETURN(rc);
 }
 
@@ -1285,6 +1378,216 @@ static struct dt_index_operations osd_index_ops = {
 	}
 };
 
+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;
+
+	/* XXX: dmu_object_next() does NOT find dnodes allocated
+	 *	in the current non-committed txg, so we force txg
+	 *	commit to find all existing dnodes ... */
+	txg_wait_synced(dmu_objset_pool(dev->od_objset.os), 0ULL);
+
+	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;
+
+	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,
+	}
+};
+
 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
 		const struct dt_index_features *feat)
 {
@@ -1300,9 +1603,10 @@ int osd_index_try(const struct lu_env *env, struct dt_object *dt,
 	if (feat->dif_flags & DT_IND_RANGE)
 		RETURN(-ERANGE);
 
-	if (unlikely(feat == &dt_otable_features))
-		/* do not support oi scrub yet. */
-		RETURN(-ENOTSUPP);
+	if (unlikely(feat == &dt_otable_features)) {
+		dt->do_index_ops = &osd_zfs_otable_ops;
+		RETURN(0);
+	}
 
 	LASSERT(obj->oo_db != NULL);
 	if (likely(feat == &dt_directory_features)) {
@@ -1320,10 +1624,9 @@ int osd_index_try(const struct lu_env *env, struct dt_object *dt,
 		if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
 			RETURN(-EINVAL);
 
-		/* Although the zap_*_uint64() primitives support large keys, we
-		 * limit ourselves to 64-bit keys for now */
-		if (feat->dif_keysize_max != sizeof(__u64) ||
-		    feat->dif_keysize_min != sizeof(__u64))
+		if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
+			RETURN(-E2BIG);
+		if (feat->dif_keysize_max != feat->dif_keysize_min)
 			RETURN(-EINVAL);
 
 		/* As for the record size, it should be a multiple of 8 bytes
@@ -1331,14 +1634,20 @@ int osd_index_try(const struct lu_env *env, struct dt_object *dt,
 		 */
 		if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
 			RETURN(-E2BIG);
-		if (feat->dif_recsize_max != feat->dif_recsize_min ||
-		    (feat->dif_recsize_max & (sizeof(__u64) - 1)))
+		if (feat->dif_recsize_max != feat->dif_recsize_min)
 			RETURN(-EINVAL);
 
-		obj->oo_recsize = feat->dif_recsize_max / sizeof(__u64);
+		obj->oo_keysize = feat->dif_keysize_max;
+		obj->oo_recsize = feat->dif_recsize_max;
+		obj->oo_recusize = 1;
+
+		/* ZFS prefers to work with array of 64bits */
+		if ((obj->oo_recsize & 7) == 0) {
+			obj->oo_recsize >>= 3;
+			obj->oo_recusize = 8;
+		}
 		dt->do_index_ops = &osd_index_ops;
 	}
 
 	RETURN(0);
 }
-