X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_oi.c;h=794035ef0bff27a9a18f44f592e34cf98733098d;hp=e629821120101482d3ac1c775155a71fb731980b;hb=42bedf90159bcaf0c8415f19a821ba11b2957155;hpb=33c7936e9175f5cb151ad0cfc8c9694751eb4430

diff --git a/lustre/osd-zfs/osd_oi.c b/lustre/osd-zfs/osd_oi.c
index e629821..794035e 100644
--- a/lustre/osd-zfs/osd_oi.c
+++ b/lustre/osd-zfs/osd_oi.c
@@ -40,6 +40,7 @@
  *
  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  * Author: Mike Pershin <tappro@whamcloud.com>
+ * Author: Di Wang <di.wang@intel.com>
  */
 
 #ifndef EXPORT_SYMTAB
@@ -89,7 +90,7 @@ struct named_oid {
 };
 
 static const struct named_oid oids[] = {
-	{ OFD_LAST_RECV_OID,		LAST_RCVD },
+	{ LAST_RECV_OID,		LAST_RCVD },
 	{ OFD_LAST_GROUP_OID,		"LAST_GROUP" },
 	{ LLOG_CATALOGS_OID,		"CATALOGS" },
 	{ MGS_CONFIGS_OID,              NULL /*MOUNT_CONFIGS_DIR*/ },
@@ -98,7 +99,6 @@ static const struct named_oid oids[] = {
 	{ MDD_CAPA_KEYS_OID,            NULL /*CAPA_KEYS*/ },
 	{ FLD_INDEX_OID,                "fld" },
 	{ MDD_LOV_OBJ_OID,		LOV_OBJID },
-	{ MDT_LAST_RECV_OID,		LAST_RCVD },
 	{ OFD_HEALTH_CHECK_OID,		HEALTH_CHECK },
 	{ ACCT_USER_OID,		"acct_usr_inode" },
 	{ ACCT_GROUP_OID,		"acct_grp_inode" },
@@ -119,6 +119,262 @@ static char *oid2name(const unsigned long oid)
 	return NULL;
 }
 
+/**
+ * Lookup an existing OI by the given name.
+ */
+static int
+osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
+	      uint64_t parent, const char *name, struct osd_oi *oi)
+{
+	struct zpl_direntry	*zde = &osd_oti_get(env)->oti_zde.lzd_reg;
+	int			 rc;
+
+	rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
+	if (rc)
+		return rc;
+
+	strncpy(oi->oi_name, name, OSD_OI_NAME_SIZE - 1);
+	oi->oi_zapid = zde->zde_dnode;
+
+	return rc;
+}
+
+/**
+ * Create a new OI with the given name.
+ */
+static int
+osd_oi_create(const struct lu_env *env, struct osd_device *o,
+	      uint64_t parent, const char *name, uint64_t *child)
+{
+	struct zpl_direntry	*zde = &osd_oti_get(env)->oti_zde.lzd_reg;
+	struct lu_attr		*la = &osd_oti_get(env)->oti_la;
+	dmu_buf_t		*db;
+	dmu_tx_t		*tx;
+	int			 rc;
+
+	/* verify it doesn't already exist */
+	rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
+	if (rc == 0)
+		return -EEXIST;
+
+	/* create fid-to-dnode index */
+	tx = dmu_tx_create(o->od_objset.os);
+	if (tx == NULL)
+		return -ENOMEM;
+
+	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
+	dmu_tx_hold_bonus(tx, parent);
+	dmu_tx_hold_zap(tx, parent, TRUE, name);
+	LASSERT(tx->tx_objset->os_sa);
+	dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
+
+	rc = -dmu_tx_assign(tx, TXG_WAIT);
+	if (rc) {
+		dmu_tx_abort(tx);
+		return rc;
+	}
+
+	la->la_valid = LA_MODE | LA_UID | LA_GID;
+	la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
+	la->la_uid = la->la_gid = 0;
+	__osd_zap_create(env, &o->od_objset, &db, tx, la, oi_tag, 0);
+
+	zde->zde_dnode = db->db_object;
+	zde->zde_pad = 0;
+	zde->zde_type = IFTODT(S_IFDIR);
+
+	rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
+
+	dmu_tx_commit(tx);
+
+	*child = db->db_object;
+	sa_buf_rele(db, oi_tag);
+
+	return rc;
+}
+
+static int
+osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
+		      uint64_t parent, const char *name, uint64_t *child)
+{
+	struct osd_oi	oi;
+	int		rc;
+
+	rc = osd_oi_lookup(env, o, parent, name, &oi);
+	if (rc == 0)
+		*child = oi.oi_zapid;
+	else if (rc == -ENOENT)
+		rc = osd_oi_create(env, o, parent, name, child);
+
+	return rc;
+}
+
+/**
+ * Lookup the target index/flags of the fid, so it will know where
+ * the object is located (tgt index) and it is MDT or OST object.
+ */
+int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
+		   const struct lu_fid *fid, struct lu_seq_range *range)
+{
+	struct seq_server_site	*ss = osd_seq_site(osd);
+	int			rc;
+
+	if (fid_is_idif(fid)) {
+		range->lsr_flags = LU_SEQ_RANGE_OST;
+		range->lsr_index = fid_idif_ost_idx(fid);
+		return 0;
+	}
+
+	if (!fid_is_norm(fid)) {
+		range->lsr_flags = LU_SEQ_RANGE_MDT;
+		if (ss != NULL)
+			range->lsr_index = ss->ss_node_id;
+		return 0;
+	}
+
+	LASSERT(ss != NULL);
+	range->lsr_flags = -1;
+	rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
+	if (rc != 0) {
+		CERROR("%s can not find "DFID": rc = %d\n",
+		       osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
+	}
+	return rc;
+}
+
+int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
+		  const struct lu_fid *fid)
+{
+	struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
+	int rc;
+	ENTRY;
+
+	if (fid_is_idif(fid))
+		RETURN(1);
+
+	rc = osd_fld_lookup(env, osd, fid, range);
+	if (rc != 0) {
+		CERROR("%s: Can not lookup fld for "DFID"\n",
+		       osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid));
+		RETURN(rc);
+	}
+
+	CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
+	       PRANGE(range));
+
+	if (range->lsr_flags == LU_SEQ_RANGE_OST)
+		RETURN(1);
+
+	RETURN(0);
+}
+
+static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
+					   obd_seq seq)
+{
+	struct osd_seq *osd_seq;
+
+	cfs_list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
+		if (osd_seq->os_seq == seq)
+			return osd_seq;
+	}
+	return NULL;
+}
+
+static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
+				    obd_seq seq)
+{
+	struct osd_seq *osd_seq;
+
+	read_lock(&seq_list->osl_seq_list_lock);
+	osd_seq = osd_seq_find_locked(seq_list, seq);
+	read_unlock(&seq_list->osl_seq_list_lock);
+
+	return osd_seq;
+}
+
+static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
+					   struct osd_device *osd, obd_seq seq)
+{
+	struct osd_seq_list	*seq_list = &osd->od_seq_list;
+	struct osd_seq		*osd_seq;
+	char			*key = osd_oti_get(env)->oti_buf;
+	char			*seq_name = osd_oti_get(env)->oti_str;
+	struct osd_oi		oi;
+	uint64_t		sdb, odb;
+	int			i;
+	int			rc = 0;
+	ENTRY;
+
+	osd_seq = osd_seq_find(seq_list, seq);
+	if (osd_seq != NULL)
+		RETURN(osd_seq);
+
+	down(&seq_list->osl_seq_init_sem);
+	/* Check again, in case some one else already add it
+	 * to the list */
+	osd_seq = osd_seq_find(seq_list, seq);
+	if (osd_seq != NULL)
+		GOTO(out, rc = 0);
+
+	OBD_ALLOC_PTR(osd_seq);
+	if (osd_seq == NULL)
+		GOTO(out, rc = -ENOMEM);
+
+	CFS_INIT_LIST_HEAD(&osd_seq->os_seq_list);
+	osd_seq->os_seq = seq;
+
+	/* Init subdir count to be 32, but each seq can have
+	 * different subdir count */
+	osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
+	OBD_ALLOC(osd_seq->os_compat_dirs,
+		  sizeof(uint64_t) * osd_seq->os_subdir_count);
+	if (osd_seq->os_compat_dirs == NULL)
+		GOTO(out, rc = -ENOMEM);
+
+	rc = osd_oi_lookup(env, osd, osd->od_root, "O", &oi);
+	if (rc != 0) {
+		CERROR("%s: Can not find O: rc = %d\n", osd_name(osd), rc);
+		GOTO(out, rc);
+	}
+
+	sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
+		fid_seq_is_mdt0(seq)) ?  LPU64 : LPX64i,
+		fid_seq_is_idif(seq) ? 0 : seq);
+
+	rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
+	if (rc != 0) {
+		CERROR("%s: Can not create %s : rc = %d\n",
+		       osd_name(osd), seq_name, rc);
+		GOTO(out, rc);
+	}
+
+	if (seq == 0)
+		osd->od_ost_compat_grp0 = odb;
+
+	for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
+		sprintf(key, "d%d", i);
+		rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
+		if (rc)
+			GOTO(out, osd_seq = ERR_PTR(rc));
+		osd_seq->os_compat_dirs[i] = sdb;
+	}
+
+	write_lock(&seq_list->osl_seq_list_lock);
+	cfs_list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
+	write_unlock(&seq_list->osl_seq_list_lock);
+out:
+	up(&seq_list->osl_seq_init_sem);
+	if (rc != 0) {
+		if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
+			OBD_FREE(osd_seq->os_compat_dirs,
+				 sizeof(uint64_t) * osd_seq->os_subdir_count);
+		if (osd_seq != NULL)
+			OBD_FREE_PTR(osd_seq);
+		osd_seq = ERR_PTR(rc);
+	}
+	RETURN(osd_seq);
+}
+
 /*
  * objects w/o a natural reference (unlike a file on a MDS)
  * are put under a special hierarchy /O/<seq>/d0..dXX
@@ -128,17 +384,24 @@ static uint64_t
 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
 			const struct lu_fid *fid, char *buf)
 {
+	struct osd_seq	*osd_seq;
 	unsigned long	b;
 	int		rc;
 
+	osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
+	if (IS_ERR(osd_seq)) {
+		CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
+		       PFID(fid));
+		return PTR_ERR(osd_seq);
+	}
 	rc = fid_ostid_pack(fid, &osd_oti_get(env)->oti_ostid);
 	LASSERT(rc == 0); /* we should not get here with IGIF */
 	b = osd_oti_get(env)->oti_ostid.oi_id % OSD_OST_MAP_SIZE;
-	LASSERT(osd->od_ost_compat_dirs[b]);
+	LASSERT(osd_seq->os_compat_dirs[b]);
 
 	sprintf(buf, LPU64, osd_oti_get(env)->oti_ostid.oi_id);
 
-	return osd->od_ost_compat_dirs[b];
+	return osd_seq->os_compat_dirs[b];
 }
 
 /* XXX: f_ver is not counted, but may differ too */
@@ -174,7 +437,7 @@ uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
 	LASSERT(fid);
 	LASSERT(buf);
 
-	if (fid_is_idif(fid)) {
+	if (fid_is_on_ost(env, osd, fid) == 1) {
 		zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
 	} else if (fid_is_last_id(fid)) {
 		zapid = osd->od_ost_compat_grp0;
@@ -237,97 +500,6 @@ int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
 }
 
 /**
- * Lookup an existing OI by the given name.
- */
-static int
-osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
-	      uint64_t parent, const char *name, struct osd_oi *oi)
-{
-	struct zpl_direntry	*zde = &osd_oti_get(env)->oti_zde.lzd_reg;
-	int			 rc;
-
-	rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
-	if (rc)
-		return rc;
-
-	strncpy(oi->oi_name, name, OSD_OI_NAME_SIZE - 1);
-	oi->oi_zapid = zde->zde_dnode;
-
-	return rc;
-}
-
-/**
- * Create a new OI with the given name.
- */
-static int
-osd_oi_create(const struct lu_env *env, struct osd_device *o,
-	      uint64_t parent, const char *name, uint64_t *child)
-{
-	struct zpl_direntry	*zde = &osd_oti_get(env)->oti_zde.lzd_reg;
-	struct lu_attr		*la = &osd_oti_get(env)->oti_la;
-	dmu_buf_t		*db;
-	dmu_tx_t		*tx;
-	int			 rc;
-
-	/* verify it doesn't already exist */
-	rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
-	if (rc == 0)
-		return -EEXIST;
-
-	/* create fid-to-dnode index */
-	tx = dmu_tx_create(o->od_objset.os);
-	if (tx == NULL)
-		return -ENOMEM;
-
-	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
-	dmu_tx_hold_bonus(tx, parent);
-	dmu_tx_hold_zap(tx, parent, TRUE, name);
-	LASSERT(tx->tx_objset->os_sa);
-	dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
-
-	rc = -dmu_tx_assign(tx, TXG_WAIT);
-	if (rc) {
-		dmu_tx_abort(tx);
-		return rc;
-	}
-
-	la->la_valid = LA_MODE | LA_UID | LA_GID;
-	la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
-	la->la_uid = la->la_gid = 0;
-	__osd_zap_create(env, &o->od_objset, &db, tx, la, oi_tag, 0);
-
-	zde->zde_dnode = db->db_object;
-	zde->zde_pad = 0;
-	zde->zde_type = IFTODT(S_IFDIR);
-
-	rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
-
-	dmu_tx_commit(tx);
-
-	*child = db->db_object;
-	sa_buf_rele(db, oi_tag);
-
-	return rc;
-}
-
-static int
-osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
-		      uint64_t parent, const char *name, uint64_t *child)
-{
-	struct osd_oi	oi;
-	int		rc;
-
-	rc = osd_oi_lookup(env, o, parent, name, &oi);
-	if (rc == 0) {
-		*child = oi.oi_zapid;
-	} else if (rc == -ENOENT) {
-		rc = osd_oi_create(env, o, parent, name, child);
-	}
-
-	return rc;
-}
-
-/**
  * Close an entry in a specific slot.
  */
 static void
@@ -451,37 +623,48 @@ osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
 	RETURN(0);
 }
 
+static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
+{
+	struct osd_seq_list *osl = &osd->od_seq_list;
+
+	CFS_INIT_LIST_HEAD(&osl->osl_seq_list);
+	rwlock_init(&osl->osl_seq_list_lock);
+	sema_init(&osl->osl_seq_init_sem, 1);
+}
+
+static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
+{
+	struct osd_seq_list	*osl = &osd->od_seq_list;
+	struct osd_seq		*osd_seq, *tmp;
+
+	write_lock(&osl->osl_seq_list_lock);
+	cfs_list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
+				     os_seq_list) {
+		cfs_list_del(&osd_seq->os_seq_list);
+		OBD_FREE(osd_seq->os_compat_dirs,
+			 sizeof(uint64_t) * osd_seq->os_subdir_count);
+		OBD_FREE(osd_seq, sizeof(*osd_seq));
+	}
+	write_unlock(&osl->osl_seq_list_lock);
+
+	return;
+}
+
 /**
  * Create /O subdirectory to map legacy OST objects for compatibility.
  */
 static int
 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
 {
-	char		*key = osd_oti_get(env)->oti_buf;
 	uint64_t	 odb, sdb;
-	int		 i, rc;
+	int		 rc;
 	ENTRY;
 
 	rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
 	if (rc)
 		RETURN(rc);
 
-	/* create /O/0 subdirectory to map legacy OST objects */
-	rc = osd_oi_find_or_create(env, o, sdb, "0", &odb);
-	if (rc)
-		RETURN(rc);
-
-	o->od_ost_compat_grp0 = odb;
-
-	for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
-		sprintf(key, "d%d", i);
-		rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
-		if (rc)
-			RETURN(rc);
-
-		o->od_ost_compat_dirs[i] = sdb;
-	}
-
+	osd_ost_seq_init(env, o);
 	/* Create on-disk indexes to maintain per-UID/GID inode usage.
 	 * Those new indexes are created in the top-level ZAP outside the
 	 * namespace in order not to confuse ZPL which might interpret those
@@ -551,6 +734,8 @@ void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
 {
 	ENTRY;
 
+	osd_ost_seq_fini(env, o);
+
 	if (o->od_oi_table != NULL) {
 		(void) osd_oi_close_table(env, o);
 		OBD_FREE(o->od_oi_table,