X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_handler.c;h=0dc8c2e3bf9dcc0fc968c5c96376a6e47259c451;hb=9cda61a26bb1f2e70ac7ee9b583edf43201228cd;hp=9f0c53956d504a44ef8bf06bd2a759a4bd359c55;hpb=93e2795aa9c08f4ec4a73af0f7d051ddb5a4ab42;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_handler.c b/lustre/osd-zfs/osd_handler.c
index 9f0c539..0dc8c2e 100644
--- a/lustre/osd-zfs/osd_handler.c
+++ b/lustre/osd-zfs/osd_handler.c
@@ -28,7 +28,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2011, 2012 Whamcloud, Inc.
+ * Copyright (c) 2012, 2013, Intel Corporation.
  * Use is subject to license terms.
  *
  */
@@ -44,9 +44,6 @@
  * Author: Johann Lombardi <johann@whamcloud.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_OSD
 
 #include <lustre_ver.h>
@@ -57,6 +54,8 @@
 #include <obd_class.h>
 #include <lustre_disk.h>
 #include <lustre_fid.h>
+#include <lustre_param.h>
+#include <md_object.h>
 
 #include "osd_internal.h"
 
@@ -78,7 +77,7 @@ struct lu_context_key	osd_key;
 static char *root_tag = "osd_mount, rootdb";
 
 /* Slab for OSD object allocation */
-cfs_mem_cache_t *osd_object_kmem;
+struct kmem_cache *osd_object_kmem;
 
 static struct lu_kmem_descr osd_caches[] = {
 	{
@@ -164,7 +163,7 @@ static void osd_trans_commit_cb(void *cb_data, int error)
 	th->th_dev = NULL;
 	lu_context_exit(&th->th_ctx);
 	lu_context_fini(&th->th_ctx);
-	OBD_FREE_PTR(oh);
+	thandle_put(&oh->ot_super);
 
 	EXIT;
 }
@@ -228,7 +227,8 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 /*
  * Concurrency: shouldn't matter.
  */
-static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
+static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
+			  struct thandle *th)
 {
 	struct osd_device	*osd = osd_dt_dev(th->th_dev);
 	struct osd_thandle	*oh;
@@ -245,7 +245,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 		/* there won't be any commit, release reserved quota space now,
 		 * if any */
 		qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans);
-		OBD_FREE_PTR(oh);
+		thandle_put(&oh->ot_super);
 		RETURN(0);
 	}
 
@@ -299,12 +299,14 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 	oh->ot_tx = tx;
 	CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 	CFS_INIT_LIST_HEAD(&oh->ot_sa_list);
-	cfs_sema_init(&oh->ot_sa_lock, 1);
+	sema_init(&oh->ot_sa_lock, 1);
 	memset(&oh->ot_quota_trans, 0, sizeof(oh->ot_quota_trans));
 	th = &oh->ot_super;
 	th->th_dev = dt;
 	th->th_result = 0;
 	th->th_tags = LCT_TX_HANDLE;
+	atomic_set(&th->th_refc, 1);
+	th->th_alloc_size = sizeof(*oh);
 	RETURN(th);
 }
 
@@ -319,7 +321,7 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
 	ENTRY;
 
 	rc = udmu_objset_statfs(&osd->od_objset, osfs);
-	if (rc)
+	if (unlikely(rc))
 		RETURN(rc);
 	osfs->os_bavail -= min_t(obd_size,
 				 OSD_GRANT_FOR_LOCAL_OIDS / osfs->os_bsize,
@@ -362,8 +364,6 @@ static void osd_conf_get(const struct lu_env *env,
 	param->ddp_inodespace = OSD_DNODE_EST_COUNT;
 	/* per-fragment overhead to be used by the client code */
 	param->ddp_grant_frag = udmu_blk_insert_cost();
-
-	param->ddp_mnt = NULL;
 }
 
 /*
@@ -383,6 +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;
@@ -514,10 +515,11 @@ static void osd_xattr_changed_cb(void *arg, uint64_t newval)
 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;
+	struct dsl_dataset	*ds;
+	char			*dev  = lustre_cfg_string(cfg, 1);
+	dmu_buf_t		*rootdb;
+	dsl_pool_t		*dp;
+	int			 rc;
 	ENTRY;
 
 	if (o->od_objset.os != NULL)
@@ -530,6 +532,13 @@ static int osd_mount(const struct lu_env *env,
 	strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
 		sizeof(o->od_svname) - 1);
 
+	rc = osd_zfs_acct_init(env, o);
+	if (rc)
+		RETURN(rc);
+
+	if (server_name_is_ost(o->od_svname))
+		o->od_is_ost = 1;
+
 	rc = -udmu_objset_open(o->od_mntdev, &o->od_objset);
 	if (rc) {
 		CERROR("can't open objset %s: %d\n", o->od_mntdev, rc);
@@ -537,8 +546,12 @@ static int osd_mount(const struct lu_env *env,
 	}
 
 	ds = dmu_objset_ds(o->od_objset.os);
+	dp = dmu_objset_pool(o->od_objset.os);
 	LASSERT(ds);
+	LASSERT(dp);
+	dsl_pool_config_enter(dp, FTAG);
 	rc = dsl_prop_register(ds, "xattr", osd_xattr_changed_cb, o);
+	dsl_pool_config_exit(dp, FTAG);
 	if (rc)
 		CERROR("%s: cat not register xattr callback, ignore: %d\n",
 		       o->od_svname, rc);
@@ -568,6 +581,10 @@ static int osd_mount(const struct lu_env *env,
 	if (rc)
 		GOTO(err, rc);
 
+	rc = osd_convert_root_to_new_seq(env, o);
+	if (rc)
+		GOTO(err, rc);
+
 	/* Use our own ZAP for inode accounting by default, this can be changed
 	 * via procfs to estimate the inode usage from the block usage */
 	o->od_quota_iused_est = 0;
@@ -594,15 +611,17 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o)
 {
 	ENTRY;
 
-	if (cfs_atomic_read(&o->od_zerocopy_alloc))
+	if (atomic_read(&o->od_zerocopy_alloc))
 		CERROR("%s: lost %d allocated page(s)\n", o->od_svname,
-		       cfs_atomic_read(&o->od_zerocopy_alloc));
-	if (cfs_atomic_read(&o->od_zerocopy_loan))
+		       atomic_read(&o->od_zerocopy_alloc));
+	if (atomic_read(&o->od_zerocopy_loan))
 		CERROR("%s: lost %d loaned abuf(s)\n", o->od_svname,
-		       cfs_atomic_read(&o->od_zerocopy_loan));
-	if (cfs_atomic_read(&o->od_zerocopy_pin))
+		       atomic_read(&o->od_zerocopy_loan));
+	if (atomic_read(&o->od_zerocopy_pin))
 		CERROR("%s: lost %d pinned dbuf(s)\n", o->od_svname,
-		       cfs_atomic_read(&o->od_zerocopy_pin));
+		       atomic_read(&o->od_zerocopy_pin));
+
+	osd_zfs_acct_fini(env, o);
 
 	if (o->od_objset.os != NULL)
 		udmu_objset_close(&o->od_objset);
@@ -695,6 +714,7 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
 	ENTRY;
 
 
+	osd_shutdown(env, o);
 	osd_oi_fini(env, o);
 
 	if (o->od_objset.os) {
@@ -735,27 +755,46 @@ static int osd_process_config(const struct lu_env *env,
 			      struct lu_device *d, struct lustre_cfg *cfg)
 {
 	struct osd_device	*o = osd_dev(d);
-	int			 err;
+	int			rc;
 	ENTRY;
 
 	switch(cfg->lcfg_command) {
 	case LCFG_SETUP:
-		err = osd_mount(env, o, cfg);
+		rc = osd_mount(env, o, cfg);
 		break;
 	case LCFG_CLEANUP:
-		err = osd_shutdown(env, o);
+		rc = osd_shutdown(env, o);
 		break;
+	case LCFG_PARAM: {
+		LASSERT(&o->od_dt_dev);
+		rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
+					      cfg, &o->od_dt_dev);
+		if (rc > 0 || rc == -ENOSYS)
+			rc = class_process_proc_param(PARAM_OST,
+						      lprocfs_osd_obd_vars,
+						      cfg, &o->od_dt_dev);
+		break;
+	}
 	default:
-		err = -ENOTTY;
+		rc = -ENOTTY;
 	}
 
-	RETURN(err);
+	RETURN(rc);
 }
 
 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);
 }
 
 /*
@@ -778,9 +817,9 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
 
 	*exp = class_conn2export(&conn);
 
-	cfs_spin_lock(&osd->od_objset.lock);
+	spin_lock(&osd->od_objset.lock);
 	osd->od_connects++;
-	cfs_spin_unlock(&osd->od_objset.lock);
+	spin_unlock(&osd->od_objset.lock);
 
 	RETURN(0);
 }
@@ -797,11 +836,11 @@ static int osd_obd_disconnect(struct obd_export *exp)
 	ENTRY;
 
 	/* Only disconnect the underlying layers on the final disconnect. */
-	cfs_spin_lock(&osd->od_objset.lock);
+	spin_lock(&osd->od_objset.lock);
 	osd->od_connects--;
 	if (osd->od_connects == 0)
 		release = 1;
-	cfs_spin_unlock(&osd->od_objset.lock);
+	spin_unlock(&osd->od_objset.lock);
 
 	rc = class_disconnect(exp); /* bz 9811 */
 
@@ -817,15 +856,6 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
 	int			 rc = 0;
 	ENTRY;
 
-	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);
@@ -888,8 +918,10 @@ int __init osd_init(void)
 	if (rc)
 		return rc;
 
-	rc = class_register_type(&osd_obd_device_ops, NULL,
+	rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
+#ifndef HAVE_ONLY_PROCFS_SEQ
 				 lprocfs_osd_module_vars,
+#endif
 				 LUSTRE_OSD_ZFS_NAME, &osd_device_type);
 	if (rc)
 		lu_kmem_fini(osd_caches);