LU-2803 osd: osd-zfs to handle echo sequence (2) properly

[fs/lustre-release.git] / lustre / osd-zfs / osd_oi.c

diff --git a/lustre/osd-zfs/osd_oi.c b/lustre/osd-zfs/osd_oi.c
index f5e941b..01e6067 100644 (file)
--- a/lustre/osd-zfs/osd_oi.c
+++ b/lustre/osd-zfs/osd_oi.c
@@ -177,7 +177,7 @@ osd_oi_create(const struct lu_env *env, struct osd_device *o,
        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);
+       __osd_zap_create(env, &o->od_objset, &db, tx, la, parent, oi_tag, 0);
 
        zde->zde_dnode = db->db_object;
        zde->zde_pad = 0;
@@ -341,16 +341,13 @@ static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
                fid_seq_is_mdt0(seq)) ?  LPU64 : LPX64i,
                fid_seq_is_idif(seq) ? 0 : seq);
 
-       rc = osd_oi_create(env, osd, oi.oi_zapid, seq_name, &odb);
+       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);
@@ -437,10 +434,8 @@ uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
        LASSERT(fid);
        LASSERT(buf);
 
-       if (fid_is_on_ost(env, osd, fid) == 1) {
+       if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
                zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
-       } else if (fid_is_last_id(fid)) {
-               zapid = osd->od_ost_compat_grp0;
        } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
                /* special objects with fixed known fids get their name */
                char *name = oid2name(fid_oid(fid));
@@ -684,6 +679,107 @@ osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
        RETURN(rc);
 }
 
+static char *root2convert = "ROOT";
+/*
+ * due to DNE requirements we have to change sequence of /ROOT object
+ * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
+ * but a normal sequence living on MDS#0
+ * this is the sole purpose of this function.
+ *
+ * This is only needed for pre-production 2.4 ZFS filesystems, and
+ * can be removed in the future.
+ */
+int osd_convert_root_to_new_seq(const struct lu_env *env,
+                                       struct osd_device *o)
+{
+       struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
+       char                *buf = osd_oti_get(env)->oti_str;
+       struct lu_fid        newfid;
+       uint64_t             zapid;
+       dmu_tx_t            *tx = NULL;
+       int                  rc;
+       ENTRY;
+
+       /* ignore OSTs */
+       if (strstr(o->od_svname, "MDT") == NULL)
+               RETURN(0);
+
+       /* lookup /ROOT */
+       rc = -zap_lookup(o->od_objset.os, o->od_root, root2convert, 8,
+                        sizeof(*lze) / 8, (void *)lze);
+       /* doesn't exist or let actual user to handle the error */
+       if (rc)
+               RETURN(0);
+
+       CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
+              PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
+
+       /* already right one? */
+       if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
+               return 0;
+
+       tx = dmu_tx_create(o->od_objset.os);
+       if (tx == NULL)
+               return -ENOMEM;
+
+       dmu_tx_hold_bonus(tx, o->od_root);
+
+       /* declare delete/insert of the name */
+       dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
+       dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
+
+       /* declare that we'll remove object from fid-dnode mapping */
+       zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
+       dmu_tx_hold_bonus(tx, zapid);
+       dmu_tx_hold_zap(tx, zapid, FALSE, buf);
+
+       /* declare that we'll add object to fid-dnode mapping */
+       newfid.f_seq = FID_SEQ_ROOT;
+       newfid.f_oid = 1;
+       newfid.f_ver = 0;
+       zapid = osd_get_name_n_idx(env, o, &newfid, buf);
+       dmu_tx_hold_bonus(tx, zapid);
+       dmu_tx_hold_zap(tx, zapid, TRUE, buf);
+
+       rc = -dmu_tx_assign(tx, TXG_WAIT);
+       if (rc)
+               GOTO(err, rc);
+
+       rc = -zap_remove(o->od_objset.os, o->od_root, root2convert, tx);
+       if (rc)
+               GOTO(err, rc);
+
+       /* remove from OI */
+       zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
+       rc = -zap_remove(o->od_objset.os, zapid, buf, tx);
+       if (rc)
+               GOTO(err, rc);
+
+       lze->lzd_fid = newfid;
+       rc = -zap_add(o->od_objset.os, o->od_root, root2convert,
+                     8, sizeof(*lze) / 8, (void *)lze, tx);
+       if (rc)
+               GOTO(err, rc);
+
+       /* add to OI with the new fid */
+       zapid = osd_get_name_n_idx(env, o, &newfid, buf);
+       rc = -zap_add(o->od_objset.os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
+       if (rc)
+               GOTO(err, rc);
+
+
+       /* LMA will be updated in mdd_compat_fixes */
+       dmu_tx_commit(tx);
+
+       RETURN(rc);
+
+err:
+       if (tx)
+               dmu_tx_abort(tx);
+       CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
+       RETURN(rc);
+}
+
 /**
  * Initialize the OIs by either opening or creating them as needed.
  */