X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=0b31287d58d14c1779c16db957923a3a52ae66e1;hb=aa9938a26b2fd58afc53bd25020e8940464d1855;hp=58bd62442476668f7f8e648b08ce46c4bf99e269;hpb=5c50b8609e54cc6fa72253a4c0ceb9c77961da3c;p=fs%2Flustre-release.git

diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c
index 58bd624..0b31287 100644
--- a/lustre/mdt/mdt_handler.c
+++ b/lustre/mdt/mdt_handler.c
@@ -49,6 +49,7 @@
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/module.h>
+#include <linux/pagemap.h>
 
 #include <dt_object.h>
 #include <lustre_acl.h>
@@ -379,6 +380,7 @@ static int mdt_get_root(struct tgt_session_info *tsi)
 
 	nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 	if (nodemap_fileset && nodemap_fileset[0]) {
+		CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 		if (fileset) {
 			/* consider fileset from client as a sub-fileset
 			 * of the nodemap one */
@@ -396,6 +398,7 @@ static int mdt_get_root(struct tgt_session_info *tsi)
 	}
 
 	if (fileset) {
+		CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 		rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 		if (rc < 0)
 			GOTO(out, rc = err_serious(rc));
@@ -448,15 +451,15 @@ static int mdt_statfs(struct tgt_session_info *tsi)
 		rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 		if (rc)
 			GOTO(out, rc);
-		spin_lock(&info->mti_mdt->mdt_osfs_lock);
+		spin_lock(&info->mti_mdt->mdt_lock);
 		info->mti_mdt->mdt_osfs = *osfs;
 		info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
-		spin_unlock(&info->mti_mdt->mdt_osfs_lock);
+		spin_unlock(&info->mti_mdt->mdt_lock);
 	} else {
 		/** use cached statfs data */
-		spin_lock(&info->mti_mdt->mdt_osfs_lock);
+		spin_lock(&info->mti_mdt->mdt_lock);
 		*osfs = info->mti_mdt->mdt_osfs;
-		spin_unlock(&info->mti_mdt->mdt_osfs_lock);
+		spin_unlock(&info->mti_mdt->mdt_lock);
 	}
 
 	if (rc == 0)
@@ -582,9 +585,6 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 		       PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 	}
 
-	if (info != NULL)
-		mdt_body_reverse_idmap(info, b);
-
 	if (!(attr->la_valid & LA_TYPE))
 		return;
 
@@ -1017,7 +1017,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
 
 	rc = mdt_attr_get_complex(info, o, ma);
 	if (unlikely(rc)) {
-		CERROR("%s: getattr error for "DFID": rc = %d\n",
+		CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
+		       "%s: getattr error for "DFID": rc = %d\n",
 		       mdt_obd_name(info->mti_mdt),
 		       PFID(mdt_object_fid(o)), rc);
 		RETURN(rc);
@@ -1138,23 +1139,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
 		       repbody->mbo_max_mdsize);
 	}
 
-	if (exp_connect_rmtclient(info->mti_exp) &&
-	    reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
-		void *buf = req_capsule_server_get(pill, &RMF_ACL);
-
-		/* mdt_getattr_lock only */
-		rc = mdt_pack_remote_perm(info, o, buf);
-		if (rc) {
-			repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
-			repbody->mbo_aclsize = 0;
-			RETURN(rc);
-		} else {
-			repbody->mbo_valid |= OBD_MD_FLRMTPERM;
-			repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
-		}
-	}
 #ifdef CONFIG_FS_POSIX_ACL
-	else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
+	if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 		 (reqbody->mbo_valid & OBD_MD_FLACL)) {
 		struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
 		if (IS_ERR(nodemap))
@@ -1227,18 +1213,13 @@ static int mdt_getattr(struct tgt_session_info *tsi)
 	repbody->mbo_eadatasize = 0;
 	repbody->mbo_aclsize = 0;
 
-	if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
-		rc = mdt_init_ucred(info, reqbody);
-	else
-		rc = mdt_check_ucred(info);
+	rc = mdt_check_ucred(info);
 	if (unlikely(rc))
 		GOTO(out_shrink, rc);
 
 	info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
 
 	rc = mdt_getattr_internal(info, obj, 0);
-	if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
-                mdt_exit_ucred(info);
         EXIT;
 out_shrink:
         mdt_client_compatibility(info);
@@ -1260,17 +1241,13 @@ out:
  */
 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
 {
-	__u64	o1_flags;
+	unsigned int o1_lov_created = o1->mot_lov_created;
 
 	mutex_lock(&o1->mot_lov_mutex);
 	mutex_lock(&o2->mot_lov_mutex);
 
-	o1_flags = o1->mot_flags;
-	o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
-			(o2->mot_flags & MOF_LOV_CREATED);
-
-	o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
-			(o1_flags & MOF_LOV_CREATED);
+	o1->mot_lov_created = o2->mot_lov_created;
+	o2->mot_lov_created = o1_lov_created;
 
 	mutex_unlock(&o2->mot_lov_mutex);
 	mutex_unlock(&o1->mot_lov_mutex);
@@ -1842,6 +1819,26 @@ free_rdpg:
 	return rc;
 }
 
+static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
+{
+	struct lu_ucred *uc = mdt_ucred_check(info);
+	struct lu_attr *attr = &info->mti_attr.ma_attr;
+
+	if (uc == NULL)
+		return -EINVAL;
+
+	if (op != REINT_SETATTR) {
+		if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
+			attr->la_uid = uc->uc_fsuid;
+		/* for S_ISGID, inherit gid from his parent, such work will be
+		 * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
+		if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
+			attr->la_gid = uc->uc_fsgid;
+	}
+
+	return 0;
+}
+
 static int mdt_reint_internal(struct mdt_thread_info *info,
                               struct mdt_lock_handle *lhc,
                               __u32 op)
@@ -2094,32 +2091,7 @@ static int mdt_quotactl(struct tgt_session_info *tsi)
 		GOTO(out_nodemap, rc = -EFAULT);
 	}
 
-	/* map uid/gid for remote client */
 	id = oqctl->qc_id;
-	if (exp_connect_rmtclient(exp)) {
-		struct lustre_idmap_table *idmap;
-
-		idmap = exp->exp_mdt_data.med_idmap;
-
-		if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
-			     oqctl->qc_cmd != Q_GETINFO))
-			GOTO(out_nodemap, rc = -EPERM);
-
-		if (oqctl->qc_type == USRQUOTA)
-			id = lustre_idmap_lookup_uid(NULL, idmap, 0,
-						     oqctl->qc_id);
-		else if (oqctl->qc_type == GRPQUOTA)
-			id = lustre_idmap_lookup_gid(NULL, idmap, 0,
-						     oqctl->qc_id);
-		else
-			GOTO(out_nodemap, rc = -EINVAL);
-
-		if (id == CFS_IDMAP_NOTFOUND) {
-			CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
-			GOTO(out_nodemap, rc = -EACCES);
-		}
-	}
-
 	if (oqctl->qc_type == USRQUOTA)
 		id = nodemap_map_id(nodemap, NODEMAP_UID,
 				    NODEMAP_CLIENT_TO_FS, id);
@@ -2218,21 +2190,9 @@ static int mdt_llog_ctxt_unclone(const struct lu_env *env,
  */
 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
 {
-	int rc;
-
-	rc = mdt_handle_idmap(tsi);
-	if (unlikely(rc)) {
-		struct ptlrpc_request	*req = tgt_ses_req(tsi);
-		__u32			 opc;
-
-		opc = lustre_msg_get_opc(req->rq_reqmsg);
-		if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
-			sptlrpc_svc_ctx_invalidate(req);
-	}
-
 	CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 
-	return rc;
+	return 0;
 }
 
 /*
@@ -2415,12 +2375,13 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 			    void *data, int flag)
 {
-	struct lustre_handle lockh;
-	int		  rc;
+	int rc = 0;
 	ENTRY;
 
 	switch (flag) {
-	case LDLM_CB_BLOCKING:
+	case LDLM_CB_BLOCKING: {
+		struct lustre_handle lockh;
+
 		ldlm_lock2handle(lock, &lockh);
 		rc = ldlm_cli_cancel(&lockh,
 			ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
@@ -2429,17 +2390,46 @@ int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 			RETURN(rc);
 		}
 		break;
-	case LDLM_CB_CANCELING:
-		LDLM_DEBUG(lock, "Revoke remote lock");
+	}
+	case LDLM_CB_CANCELING: {
+		LDLM_DEBUG(lock, "Revoke remote lock\n");
+
 		/* discard slc lock here so that it can be cleaned anytime,
 		 * especially for cleanup_resource() */
 		tgt_discard_slc_lock(lock);
+
+		/* once we cache lock, l_ast_data is set to mdt_object */
+		if (lock->l_ast_data != NULL) {
+			struct mdt_object *mo = lock->l_ast_data;
+			struct lu_env env;
+
+			rc = lu_env_init(&env, LCT_MD_THREAD);
+			if (unlikely(rc != 0)) {
+				struct obd_device *obd;
+
+				obd = ldlm_lock_to_ns(lock)->ns_obd;
+				CWARN("%s: lu_env initialization failed, object"
+				      "%p "DFID" is leaked!\n",
+				      obd->obd_name, mo,
+				      PFID(mdt_object_fid(mo)));
+				RETURN(rc);
+			}
+
+			if (lock->l_policy_data.l_inodebits.bits &
+			    (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
+				rc = mo_invalidate(&env, mdt_object_child(mo));
+				mo->mot_cache_attr = 0;
+			}
+			mdt_object_put(&env, mo);
+			lu_env_fini(&env);
+		}
 		break;
+	}
 	default:
 		LBUG();
 	}
 
-	RETURN(0);
+	RETURN(rc);
 }
 
 int mdt_check_resent_lock(struct mdt_thread_info *info,
@@ -2479,7 +2469,8 @@ int mdt_check_resent_lock(struct mdt_thread_info *info,
 
 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
 			   const struct lu_fid *fid, struct lustre_handle *lh,
-			   enum ldlm_mode mode, __u64 ibits, bool nonblock)
+			   enum ldlm_mode mode, __u64 ibits, bool nonblock,
+			   bool cache)
 {
 	struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
 	union ldlm_policy_data *policy = &mti->mti_policy;
@@ -2500,12 +2491,24 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
 	einfo->ei_res_id = res_id;
 	if (nonblock)
 		einfo->ei_nonblock = 1;
+	if (cache) {
+		/*
+		 * if we cache lock, couple lock with mdt_object, so that object
+		 * can be easily found in lock ASTs.
+		 */
+		mdt_object_get(mti->mti_env, o);
+		einfo->ei_cbdata = o;
+	}
 
 	memset(policy, 0, sizeof(*policy));
 	policy->l_inodebits.bits = ibits;
 
 	rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
 			    policy);
+	if (rc < 0 && cache) {
+		mdt_object_put(mti->mti_env, o);
+		einfo->ei_cbdata = NULL;
+	}
 	RETURN(rc);
 }
 
@@ -2654,7 +2657,8 @@ mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
 		rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
 					    &lh->mlh_rreg_lh,
 					    lh->mlh_rreg_mode,
-					    MDS_INODELOCK_UPDATE, nonblock);
+					    MDS_INODELOCK_UPDATE, nonblock,
+					    false);
 		if (rc != ELDLM_OK) {
 			if (local_lh != NULL)
 				mdt_object_unlock(info, o, local_lh, rc);
@@ -2782,17 +2786,24 @@ static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
  * \param decref force immediate lock releasing
  */
 static void mdt_save_remote_lock(struct mdt_thread_info *info,
-				 struct lustre_handle *h, enum ldlm_mode mode,
-				 int decref)
+				 struct mdt_object *o, struct lustre_handle *h,
+				 enum ldlm_mode mode, int decref)
 {
 	ENTRY;
 
 	if (lustre_handle_is_used(h)) {
+		struct ldlm_lock *lock = ldlm_handle2lock(h);
+
+		if (o != NULL &&
+		    (lock->l_policy_data.l_inodebits.bits &
+		     (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
+			mo_invalidate(info->mti_env, mdt_object_child(o));
+
 		if (decref || !info->mti_has_trans ||
 		    !(mode & (LCK_PW | LCK_EX))) {
 			ldlm_lock_decref_and_cancel(h, mode);
+			LDLM_LOCK_PUT(lock);
 		} else {
-			struct ldlm_lock *lock = ldlm_handle2lock(h);
 			struct ptlrpc_request *req = mdt_info_req(info);
 
 			LASSERT(req != NULL);
@@ -2824,7 +2835,8 @@ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
 
 	mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
 	mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
-	mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
+	mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
+			     decref);
 
 	EXIT;
 }
@@ -3034,11 +3046,6 @@ struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
 
 static int mdt_tgt_connect(struct tgt_session_info *tsi)
 {
-	struct ptlrpc_request	*req = tgt_ses_req(tsi);
-	int			 rc;
-
-	ENTRY;
-
 	if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
 	    cfs_fail_val ==
 	    tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
@@ -3046,17 +3053,7 @@ static int mdt_tgt_connect(struct tgt_session_info *tsi)
 		schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
 	}
 
-	rc = tgt_connect(tsi);
-	if (rc != 0)
-		RETURN(rc);
-
-	rc = mdt_init_idmap(tsi);
-	if (rc != 0)
-		GOTO(err, rc);
-	RETURN(0);
-err:
-	obd_disconnect(class_export_get(req->rq_export));
-	return rc;
+	return tgt_connect(tsi);
 }
 
 enum mdt_it_code {
@@ -4501,6 +4498,11 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
 	struct lfsck_stop	 stop;
 	ENTRY;
 
+	if (m->mdt_md_root != NULL) {
+		mdt_object_put(env, m->mdt_md_root);
+		m->mdt_md_root = NULL;
+	}
+
 	stop.ls_status = LS_PAUSED;
 	stop.ls_flags = 0;
 	next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
@@ -4621,7 +4623,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 	m->mdt_squash.rsi_gid = 0;
 	INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
 	init_rwsem(&m->mdt_squash.rsi_sem);
-	spin_lock_init(&m->mdt_osfs_lock);
+	spin_lock_init(&m->mdt_lock);
 	m->mdt_osfs_age = cfs_time_shift_64(-1000);
 	m->mdt_enable_remote_dir = 0;
 	m->mdt_enable_remote_dir_gid = 0;
@@ -4957,8 +4959,10 @@ static int mdt_object_print(const struct lu_env *env, void *cookie,
 	struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
 
 	return (*p)(env, cookie,
-		    LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
-		    mdto, mdto->mot_flags, mdto->mot_write_count);
+		    LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
+		    mdto, mdto->mot_lov_created ? "lov_created" : "",
+		    mdto->mot_cache_attr ? "cache_attr" : "",
+		    mdto->mot_write_count);
 }
 
 static int mdt_prepare(const struct lu_env *env,
@@ -5071,6 +5075,7 @@ static int mdt_connect_internal(struct obd_export *exp,
 	LASSERT(data != NULL);
 
 	data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
+	data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
 	data->ocd_ibits_known &= MDS_INODELOCK_FULL;
 
 	if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
@@ -5312,6 +5317,8 @@ static int mdt_obd_connect(const struct lu_env *env,
 	ENTRY;
 
 	LASSERT(env != NULL);
+	LASSERT(data != NULL);
+
 	if (!exp || !obd || !cluuid)
 		RETURN(-EINVAL);
 
@@ -5328,7 +5335,7 @@ static int mdt_obd_connect(const struct lu_env *env,
 	 * XXX: probably not very appropriate method is used now
 	 *      at some point we should find a better one
 	 */
-	if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
+	if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
 	    !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
 	    !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
 		rc = obd_get_info(env, mdt->mdt_child_exp,
@@ -5414,8 +5421,6 @@ static int mdt_init_export(struct obd_export *exp)
 
 	INIT_LIST_HEAD(&med->med_open_head);
 	spin_lock_init(&med->med_open_lock);
-	mutex_init(&med->med_idmap_mutex);
-	med->med_idmap = NULL;
 	spin_lock(&exp->exp_lock);
 	exp->exp_connecting = 1;
 	spin_unlock(&exp->exp_lock);
@@ -5447,9 +5452,6 @@ static int mdt_destroy_export(struct obd_export *exp)
 {
         ENTRY;
 
-        if (exp_connect_rmtclient(exp))
-                mdt_cleanup_idmap(&exp->exp_mdt_data);
-
         target_destroy_export(exp);
         /* destroy can be called from failed obd_setup, so
          * checking uuid is safer than obd_self_export */
@@ -5910,13 +5912,15 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
         if (rc)
                 RETURN(rc);
 
-        switch (cmd) {
-        case OBD_IOC_SYNC:
-                rc = mdt_device_sync(&env, mdt);
-                break;
-        case OBD_IOC_SET_READONLY:
-                rc = dt->dd_ops->dt_ro(&env, dt);
-                break;
+	switch (cmd) {
+	case OBD_IOC_SYNC:
+		rc = mdt_device_sync(&env, mdt);
+		break;
+	case OBD_IOC_SET_READONLY:
+		rc = dt_sync(&env, dt);
+		if (rc == 0)
+			rc = dt_ro(&env, dt);
+		break;
 	case OBD_IOC_ABORT_RECOVERY:
 		CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
 		obd->obd_abort_recovery = 1;