LU-2224 osd-zfs: Fix osd_commit_async() locking

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

diff --git a/lustre/osd-zfs/osd_handler.c b/lustre/osd-zfs/osd_handler.c
index c7cb402..5a345d6 100644 (file)
--- a/lustre/osd-zfs/osd_handler.c
+++ b/lustre/osd-zfs/osd_handler.c
@@ -209,11 +209,10 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
                if (!lu_device_is_md(&d->dd_lu_dev) && rc == -ENOSPC)
                        CERROR("%s: failed to start transaction due to ENOSPC. "
                               "Metadata overhead is underestimated or "
-                              "grant_ratio is too low.\n",
-                              osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name);
+                              "grant_ratio is too low.\n", osd->od_svname);
                else
                        CERROR("%s: can't assign tx: rc = %d\n",
-                              osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name, rc);
+                              osd->od_svname, rc);
        } else {
                /* add commit callback */
                dmu_tx_callback_register(oh->ot_tx, osd_trans_commit_cb, oh);
@@ -384,6 +383,7 @@ static int osd_commit_async(const struct lu_env *env, struct dt_device *dev)
        tx_state_t        *tx = &dmu_objset_pool(osd->od_objset.os)->dp_tx;
        uint64_t           txg;
 
+       mutex_enter(&tx->tx_sync_lock);
        txg = tx->tx_open_txg + 1;
        if (tx->tx_quiesce_txg_waiting < txg) {
                tx->tx_quiesce_txg_waiting = txg;
@@ -505,9 +505,17 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
        RETURN(0);
 }
 
+static void osd_xattr_changed_cb(void *arg, uint64_t newval)
+{
+       struct osd_device *osd = arg;
+
+       osd->od_xattr_in_sa = (newval == ZFS_XATTR_SA);
+}
+
 static int osd_mount(const struct lu_env *env,
                     struct osd_device *o, struct lustre_cfg *cfg)
 {
+       struct dsl_dataset *ds;
        char      *dev  = lustre_cfg_string(cfg, 1);
        dmu_buf_t *rootdb;
        int        rc;
@@ -529,6 +537,13 @@ static int osd_mount(const struct lu_env *env,
                RETURN(rc);
        }
 
+       ds = dmu_objset_ds(o->od_objset.os);
+       LASSERT(ds);
+       rc = dsl_prop_register(ds, "xattr", osd_xattr_changed_cb, o);
+       if (rc)
+               CERROR("%s: cat not register xattr callback, ignore: %d\n",
+                      o->od_svname, rc);
+
        rc = __osd_obj2dbuf(env, o->od_objset.os, o->od_objset.root,
                                &rootdb, root_tag);
        if (rc) {
@@ -564,6 +579,14 @@ static int osd_mount(const struct lu_env *env,
 
        o->arc_prune_cb = arc_add_prune_callback(arc_prune_func, o);
 
+       /* initialize quota slave instance */
+       o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
+                                    o->od_proc_entry);
+       if (IS_ERR(o->od_quota_slave)) {
+               rc = PTR_ERR(o->od_quota_slave);
+               o->od_quota_slave = NULL;
+               GOTO(err, rc);
+       }
 err:
        RETURN(rc);
 }
@@ -668,13 +691,20 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
                                         struct lu_device *d)
 {
        struct osd_device *o = osd_dev(d);
+       struct dsl_dataset *ds;
        int                rc;
        ENTRY;
 
 
+       osd_shutdown(env, o);
        osd_oi_fini(env, o);
 
        if (o->od_objset.os) {
+               ds = dmu_objset_ds(o->od_objset.os);
+               rc = dsl_prop_unregister(ds, "xattr", osd_xattr_changed_cb, o);
+               if (rc)
+                       CERROR("%s: dsl_prop_unregister xattr error %d\n",
+                               o->od_svname, rc);
                arc_remove_prune_callback(o->arc_prune_cb);
                o->arc_prune_cb = NULL;
                osd_sync(env, lu2dt_dev(d));
@@ -726,8 +756,17 @@ static int osd_process_config(const struct lu_env *env,
 
 static int osd_recovery_complete(const struct lu_env *env, struct lu_device *d)
 {
+       struct osd_device       *osd = osd_dev(d);
+       int                      rc = 0;
        ENTRY;
-       RETURN(0);
+
+       if (osd->od_quota_slave == NULL)
+               RETURN(0);
+
+       /* start qsd instance on recovery completion, this notifies the quota
+        * slave code that we are about to process new requests now */
+       rc = qsd_start(env, osd->od_quota_slave);
+       RETURN(rc);
 }
 
 /*
@@ -789,14 +828,19 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
        int                      rc = 0;
        ENTRY;
 
-       /* initialize quota slave instance */
-       osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
-                                      osd->od_proc_entry);
-       if (IS_ERR(osd->od_quota_slave)) {
-               rc = PTR_ERR(osd->od_quota_slave);
-               osd->od_quota_slave = NULL;
+       if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
+               /* MDT/MDD still use old infrastructure to create
+                * special files */
+               rc = llo_local_objects_setup(env, lu2md_dev(pdev),
+                                            lu2dt_dev(dev));
+               if (rc)
+                       RETURN(rc);
        }
 
+       if (osd->od_quota_slave != NULL)
+               /* set up quota slave objects */
+               rc = qsd_prepare(env, osd->od_quota_slave);
+
        RETURN(rc);
 }