b=17447

r=deen,adilger

- fixes memory leaks;
- cleanups and more accurate error handling.

diff --git a/lustre/lov/lov_log.c b/lustre/lov/lov_log.c
index 1c47d99..7a5923e 100644 (file)
--- a/lustre/lov/lov_log.c
+++ b/lustre/lov/lov_log.c
@@ -218,7 +218,7 @@ int lov_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
         rc = llog_setup(obd, olg, LLOG_SIZE_REPL_CTXT, tgt, 0, NULL,
                         &lov_size_repl_logops);
         if (rc)
-                RETURN(rc);
+                GOTO(err_cleanup, rc);
 
         lov_getref(obd);
         /* count may not match lov->desc.ld_tgt_count during dynamic ost add */
@@ -238,7 +238,18 @@ int lov_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
                 break;
         }
         lov_putref(obd);
-        RETURN(rc);
+        GOTO(err_cleanup, rc);
+err_cleanup:
+        if (rc) {
+                struct llog_ctxt *ctxt = 
+                        llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
+                if (ctxt)
+                        llog_cleanup(ctxt);
+                ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
+                if (ctxt)
+                        llog_cleanup(ctxt);
+        }
+        return rc;
 }
 
 int lov_llog_finish(struct obd_device *obd, int count)

diff --git a/lustre/mds/handler.c b/lustre/mds/handler.c
index 27e6bb0..5bf89a2 100644 (file)
--- a/lustre/mds/handler.c
+++ b/lustre/mds/handler.c
@@ -184,6 +184,7 @@ static int mds_lov_clean(struct obd_device *obd)
 static int mds_postsetup(struct obd_device *obd)
 {
         struct mds_obd *mds = &obd->u.mds;
+        struct llog_ctxt *ctxt;
         int rc = 0;
         ENTRY;
 
@@ -195,7 +196,7 @@ static int mds_postsetup(struct obd_device *obd)
         rc = llog_setup(obd, &obd->obd_olg, LLOG_LOVEA_ORIG_CTXT, obd, 0, NULL,
                         &llog_lvfs_ops);
         if (rc)
-                RETURN(rc);
+                GOTO(err_llog, rc);
 
         if (mds->mds_profile) {
                 struct lustre_profile *lprof;
@@ -218,9 +219,14 @@ static int mds_postsetup(struct obd_device *obd)
 
 err_cleanup:
         mds_lov_clean(obd);
-        llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
-        llog_cleanup(llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT));
-        RETURN(rc);
+        ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
+        if (ctxt)
+                llog_cleanup(ctxt);
+err_llog:
+        ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+        if (ctxt)
+                llog_cleanup(ctxt);
+        return rc;
 }
 
 int mds_postrecov(struct obd_device *obd)

diff --git a/lustre/mds/mds_log.c b/lustre/mds/mds_log.c
index 0e8447f..705d730 100644 (file)
--- a/lustre/mds/mds_log.c
+++ b/lustre/mds/mds_log.c
@@ -118,6 +118,7 @@ int mds_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
                   struct obd_uuid *uuid)
 {
         struct obd_device *lov_obd = obd->u.mds.mds_osc_obd;
+        struct llog_ctxt *ctxt;
         int rc;
         ENTRY;
 
@@ -130,13 +131,24 @@ int mds_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
         rc = llog_setup(obd, &obd->obd_olg, LLOG_SIZE_REPL_CTXT, tgt, 0, NULL,
                         &mds_size_repl_logops);
         if (rc)
-                RETURN(rc);
+                GOTO(err_llog, rc);
 
         rc = obd_llog_init(lov_obd, &lov_obd->obd_olg, tgt, count, logid, uuid);
-        if (rc)
+        if (rc) {
                 CERROR("lov_llog_init err %d\n", rc);
+                GOTO(err_cleanup, rc);
+        }
 
         RETURN(rc);
+err_cleanup:
+        ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
+        if (ctxt)
+                llog_cleanup(ctxt);
+err_llog:
+        ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
+        if (ctxt)
+                llog_cleanup(ctxt);
+        return rc;
 }
 
 int mds_llog_finish(struct obd_device *obd, int count)

diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c
index bd1f214..c99aab6 100644 (file)
--- a/lustre/mgc/mgc_request.c
+++ b/lustre/mgc/mgc_request.c
@@ -922,6 +922,10 @@ static int mgc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
                 llog_initiator_connect(ctxt);
                 llog_ctxt_put(ctxt);
+        } else {
+                ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+                if (ctxt)
+                        llog_cleanup(ctxt);
         }
 
         RETURN(rc);

diff --git a/lustre/obdclass/class_hash.c b/lustre/obdclass/class_hash.c
index 55b41df..0dd18f1 100644 (file)
--- a/lustre/obdclass/class_hash.c
+++ b/lustre/obdclass/class_hash.c
@@ -128,8 +128,7 @@ lustre_hash_exit(lustre_hash_t *lh)
         int                   i;
         ENTRY;
 
-        if (!lh)
-                return;
+        LASSERT(lh != NULL);
 
         write_lock(&lh->lh_rwlock);
 

diff --git a/lustre/obdclass/llog_obd.c b/lustre/obdclass/llog_obd.c
index 77c28eb..6d24d60 100644 (file)
--- a/lustre/obdclass/llog_obd.c
+++ b/lustre/obdclass/llog_obd.c
@@ -94,7 +94,14 @@ int __llog_ctxt_put(struct llog_ctxt *ctxt)
         obd = ctxt->loc_obd;
         spin_lock(&obd->obd_dev_lock);
         spin_unlock(&obd->obd_dev_lock); /* sync with llog ctxt user thread */
-        LASSERT(obd->obd_stopping == 1 || obd->obd_set_up == 0);
+
+        /* obd->obd_starting is needed for the case of cleanup
+         * in error case while obd is starting up. */
+        LASSERTF(obd->obd_starting == 1 || 
+                 obd->obd_stopping == 1 || obd->obd_set_up == 0,
+                 "wrong obd state: %d/%d/%d\n", !!obd->obd_starting, 
+                 !!obd->obd_stopping, !!obd->obd_set_up);
+
         /* cleanup the llog ctxt here */
         if (CTXTP(ctxt, cleanup))
                 rc = CTXTP(ctxt, cleanup)(ctxt);

diff --git a/lustre/obdclass/obd_config.c b/lustre/obdclass/obd_config.c
index 9968043..8e3a854 100644 (file)
--- a/lustre/obdclass/obd_config.c
+++ b/lustre/obdclass/obd_config.c
@@ -292,23 +292,23 @@ int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
         obd->obd_uuid_hash = lustre_hash_init("UUID_HASH", 128, 128,
                                               &uuid_hash_ops, 0);
         if (!obd->obd_uuid_hash)
-                GOTO(err_hash, -ENOMEM);
+                GOTO(err_hash, err = -ENOMEM);
  
         /* create a nid-export lustre hash */
         obd->obd_nid_hash = lustre_hash_init("NID_HASH", 128, 128,
                                              &nid_hash_ops, 0);
         if (!obd->obd_nid_hash)
-                GOTO(err_hash, -ENOMEM);
+                GOTO(err_hash, err = -ENOMEM);
  
         /* create a nid-stats lustre hash */
         obd->obd_nid_stats_hash = lustre_hash_init("NID_STATS", 128, 128,
                                                    &nid_stat_hash_ops, 0);
         if (!obd->obd_nid_stats_hash)
-                GOTO(err_hash, -ENOMEM);
+                GOTO(err_hash, err = -ENOMEM);
 
         exp = class_new_export(obd, &obd->obd_uuid);
         if (IS_ERR(exp))
-                RETURN(PTR_ERR(exp));
+                GOTO(err_hash, err = PTR_ERR(exp));
 
         obd->obd_self_export = exp;
         list_del_init(&exp->exp_obd_chain_timed);
@@ -329,17 +329,25 @@ int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
                obd->obd_name, obd->obd_uuid.uuid);
 
         RETURN(0);
-
 err_exp:
         class_unlink_export(obd->obd_self_export);
         obd->obd_self_export = NULL;
 err_hash:
-        lustre_hash_exit(obd->obd_uuid_hash);
-        lustre_hash_exit(obd->obd_nid_hash);
-        lustre_hash_exit(obd->obd_nid_stats_hash);
+        if (obd->obd_uuid_hash) {
+                lustre_hash_exit(obd->obd_uuid_hash);
+                obd->obd_uuid_hash = NULL;
+        }
+        if (obd->obd_nid_hash) {
+                lustre_hash_exit(obd->obd_nid_hash);
+                obd->obd_nid_hash = NULL;
+        }
+        if (obd->obd_nid_stats_hash) {
+                lustre_hash_exit(obd->obd_nid_stats_hash);
+                obd->obd_nid_stats_hash = NULL;
+        }
         obd->obd_starting = 0;
         CERROR("setup %s failed (%d)\n", obd->obd_name, err);
-        RETURN(err);
+        return err;
 }
 
 int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)

diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c
index 9cef323..f55604b 100644 (file)
--- a/lustre/obdclass/obd_mount.c
+++ b/lustre/obdclass/obd_mount.c
@@ -167,7 +167,7 @@ struct lustre_mount_info *server_get_mount(const char *name)
         lsi = s2lsi(lmi->lmi_sb);
         mntget(lmi->lmi_mnt);
         atomic_inc(&lsi->lsi_mounts);
-
+        
         CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
                lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
                atomic_read(&lmi->lmi_mnt->mnt_count));
@@ -1132,7 +1132,6 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
         if (rc) {
                 CERROR("failed to start server %s: %d\n",
                        lsi->lsi_ldd->ldd_svname, rc);
-                server_deregister_mount(lsi->lsi_ldd->ldd_svname);
                 GOTO(out_mgc, rc);
         }
 
@@ -1193,10 +1192,8 @@ static int lustre_free_lsi(struct super_block *sb)
         struct lustre_sb_info *lsi = s2lsi(sb);
         ENTRY;
 
-        if (!lsi)
-                RETURN(0);
-
-        CDEBUG(D_MOUNT, "Freeing lsi\n");
+        LASSERT(lsi != NULL);
+        CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
 
         /* someone didn't call server_put_mount. */
         LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
@@ -1235,10 +1232,9 @@ static int lustre_put_lsi(struct super_block *sb)
         struct lustre_sb_info *lsi = s2lsi(sb);
         ENTRY;
 
-        LASSERT(lsi);
+        LASSERT(lsi != NULL);
 
         CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
-
         if (atomic_dec_and_test(&lsi->lsi_mounts)) {
                 lustre_free_lsi(sb);
                 RETURN(1);
@@ -1586,9 +1582,9 @@ static int server_fill_super(struct super_block *sb)
         if (IS_ERR(mnt)) {
                 rc = PTR_ERR(mnt);
                 CERROR("Unable to mount device %s: %d\n",
-                      lsi->lsi_lmd->lmd_dev, rc);
+                       lsi->lsi_lmd->lmd_dev, rc);
                 lustre_put_lsi(sb);
-                GOTO(out, rc);
+                RETURN(rc);
         }
         lsi->lsi_srv_mnt = mnt;
 
@@ -1602,12 +1598,12 @@ static int server_fill_super(struct super_block *sb)
                                    "running. Double-mount may have compromised"
                                    " the disk journal.\n",
                                    lsi->lsi_ldd->ldd_svname);
-                unlock_mntput(mnt);
                 lustre_put_lsi(sb);
-                GOTO(out, rc = -EALREADY);
+                unlock_mntput(mnt);
+                RETURN(-EALREADY);
         }
 
-        /* start MGS before MGC */
+        /* Start MGS before MGC */
         if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
                 rc = server_start_mgs(sb);
                 if (rc)
@@ -1645,11 +1641,12 @@ static int server_fill_super(struct super_block *sb)
                       lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
 
         RETURN(0);
-
 out_mnt:
+        /* We jump here in case of failure while starting targets or MGS.
+         * In this case we can't just put @mnt and have to do real cleanup
+         * with stoping targets, etc. */
         server_put_super(sb);
-out:
-        RETURN(rc);
+        return rc;
 }
 
 /* Get the index from the obd name.
@@ -1945,7 +1942,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
         /* Figure out the lmd from the mount options */
         if (lmd_parse((char *)data, lmd)) {
                 lustre_put_lsi(sb);
-                RETURN(-EINVAL);
+                GOTO(out, rc = -EINVAL);
         }
 
         if (lmd_is_client(lmd)) {
@@ -1954,12 +1951,13 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                         LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
                                            "client mount! Is the 'lustre' "
                                            "module loaded?\n");
+                        lustre_put_lsi(sb);
                         rc = -ENODEV;
                 } else {
                         rc = lustre_start_mgc(sb);
                         if (rc) {
-                                lustre_stop_mgc(sb);
-                                goto out;
+                                lustre_put_lsi(sb);
+                                GOTO(out, rc);
                         }
                         /* Connect and start */
                         /* (should always be ll_fill_super) */
@@ -1976,14 +1974,18 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                 /* s_f_s will call server_put_super on failure */
         }
 
+        /* If error happens in fill_super() call, @lsi will be killed there.
+         * This is why we do not put it here. */
+        GOTO(out, rc);
 out:
-        if (rc){
+        if (rc) {
                 CERROR("Unable to mount %s (%d)\n",
                        s2lsi(sb) ? lmd->lmd_dev : "", rc);
         } else {
-                CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
+                CDEBUG(D_SUPER, "Mount %s complete\n", 
+                       lmd->lmd_dev);
         }
-        RETURN(rc);
+        return rc;
 }
 
 

diff --git a/lustre/osc/osc_request.c b/lustre/osc/osc_request.c
index ec49db6..8ce49e2 100644 (file)
--- a/lustre/osc/osc_request.c
+++ b/lustre/osc/osc_request.c
@@ -3866,8 +3866,14 @@ static int osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
 
         rc = llog_setup(obd, &obd->obd_olg, LLOG_SIZE_REPL_CTXT, tgt, count,
                         NULL, &osc_size_repl_logops);
-        if (rc)
+        if (rc) {
+                struct llog_ctxt *ctxt = 
+                        llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
+                if (ctxt)
+                        llog_cleanup(ctxt);
                 CERROR("failed LLOG_SIZE_REPL_CTXT\n");
+        }
+        GOTO(out, rc);
 out:
         if (rc) {
                 CERROR("osc '%s' tgt '%s' cnt %d catid %p rc=%d\n",
@@ -3875,7 +3881,7 @@ out:
                 CERROR("logid "LPX64":0x%x\n",
                        catid->lci_logid.lgl_oid, catid->lci_logid.lgl_ogen);
         }
-        RETURN(rc);
+        return rc;
 }
 
 static int osc_llog_finish(struct obd_device *obd, int count)