X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Fosd-zfs%2Fosd_object.c;h=e4a2649433a70037cdfcdc0acb69b7eb597ef526;hb=d35997e6e100acca3effb75ca402c9df7f6252ef;hp=9513748bfa13b91dc2d6f943afc0a4fcaf86729e;hpb=20544bd96269dbda51a76d31c2d09935637e8b2d;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index 9513748..e4a2649 100644
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -28,7 +28,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
  * Use is subject to license terms.
  */
 /*
@@ -42,14 +42,10 @@
  * Author: Johann Lombardi <johann@whamcloud.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_OSD
 
 #include <lustre_ver.h>
 #include <libcfs/libcfs.h>
-#include <lustre_fsfilt.h>
 #include <obd_support.h>
 #include <lustre_net.h>
 #include <obd.h>
@@ -79,7 +75,7 @@ static struct lu_object_operations osd_lu_obj_ops;
 extern struct dt_body_operations osd_body_ops;
 static struct dt_object_operations osd_obj_otable_it_ops;
 
-extern cfs_mem_cache_t *osd_object_kmem;
+extern struct kmem_cache *osd_object_kmem;
 
 static void
 osd_object_sa_fini(struct osd_object *obj)
@@ -121,13 +117,13 @@ osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
 static void
 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 {
-	if (!cfs_list_empty(&obj->oo_sa_linkage))
+	if (!list_empty(&obj->oo_sa_linkage))
 		return;
 
 	down(&oh->ot_sa_lock);
 	write_lock(&obj->oo_attr_lock);
-	if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
-		cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
+	if (likely(list_empty(&obj->oo_sa_linkage)))
+		list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 	write_unlock(&obj->oo_attr_lock);
 	up(&oh->ot_sa_lock);
 }
@@ -140,12 +136,12 @@ void osd_object_sa_dirty_rele(struct osd_thandle *oh)
 	struct osd_object *obj;
 
 	down(&oh->ot_sa_lock);
-	while (!cfs_list_empty(&oh->ot_sa_list)) {
-		obj = cfs_list_entry(oh->ot_sa_list.next,
+	while (!list_empty(&oh->ot_sa_list)) {
+		obj = list_entry(oh->ot_sa_list.next,
 				     struct osd_object, oo_sa_linkage);
 		sa_spill_rele(obj->oo_sa_hdl);
 		write_lock(&obj->oo_attr_lock);
-		cfs_list_del_init(&obj->oo_sa_linkage);
+		list_del_init(&obj->oo_sa_linkage);
 		write_unlock(&obj->oo_attr_lock);
 	}
 	up(&oh->ot_sa_lock);
@@ -290,7 +286,7 @@ struct lu_object *osd_object_alloc(const struct lu_env *env,
 {
 	struct osd_object *mo;
 
-	OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, CFS_ALLOC_IO);
+	OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 	if (mo != NULL) {
 		struct lu_object *l;
 
@@ -298,7 +294,7 @@ struct lu_object *osd_object_alloc(const struct lu_env *env,
 		dt_object_init(&mo->oo_dt, NULL, d);
 		mo->oo_dt.do_ops = &osd_obj_ops;
 		l->lo_ops = &osd_lu_obj_ops;
-		CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
+		INIT_LIST_HEAD(&mo->oo_sa_linkage);
 		init_rwsem(&mo->oo_sem);
 		sema_init(&mo->oo_guard, 1);
 		rwlock_init(&mo->oo_attr_lock);
@@ -343,7 +339,7 @@ int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 	 */
 	obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 
-	cfs_mb();
+	smp_mb();
 	obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 
 	RETURN(0);
@@ -678,7 +674,7 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 			obj->oo_sa_xattr = NULL;
 		}
 		sa_buf_rele(obj->oo_db, osd_obj_tag);
-		cfs_list_del(&obj->oo_sa_linkage);
+		list_del(&obj->oo_sa_linkage);
 		obj->oo_db = NULL;
 	}
 }
@@ -878,7 +874,10 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	oh = container_of0(handle, struct osd_thandle, ot_super);
 
 	LASSERT(obj->oo_sa_hdl != NULL);
+	LASSERT(oh->ot_tx != NULL);
 	dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
+	if (oh->ot_tx->tx_err != 0)
+		RETURN(-oh->ot_tx->tx_err);
 
 	sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 	bspace = toqb(bspace * blksize);
@@ -934,6 +933,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 	struct osd_thandle	*oh;
 	struct osa_attr		*osa = &osd_oti_get(env)->oti_osa;
 	sa_bulk_attr_t		*bulk;
+	__u64			 valid = la->la_valid;
 	int			 cnt;
 	int			 rc = 0;
 
@@ -948,7 +948,11 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 	   transaction group. */
 	LASSERT(oh->ot_tx->tx_txg != 0);
 
-	if (la->la_valid == 0)
+	/* Only allow set size for regular file */
+	if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
+		valid &= ~(LA_SIZE | LA_BLOCKS);
+
+	if (valid == 0)
 		RETURN(0);
 
 	OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
@@ -956,7 +960,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 		RETURN(-ENOMEM);
 
 	/* do both accounting updates outside oo_attr_lock below */
-	if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
+	if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 		/* Update user accounting. Failure isn't fatal, but we still
 		 * log an error message */
 		rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
@@ -971,7 +975,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 				"%d (%d)\n", osd->od_svname,
 				obj->oo_attr.la_uid, rc);
 	}
-	if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
+	if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 		/* Update group accounting. Failure isn't fatal, but we still
 		 * log an error message */
 		rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
@@ -989,22 +993,22 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 
 	write_lock(&obj->oo_attr_lock);
 	cnt = 0;
-	if (la->la_valid & LA_ATIME) {
+	if (valid & LA_ATIME) {
 		osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
 				 osa->atime, 16);
 	}
-	if (la->la_valid & LA_MTIME) {
+	if (valid & LA_MTIME) {
 		osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
 				 osa->mtime, 16);
 	}
-	if (la->la_valid & LA_CTIME) {
+	if (valid & LA_CTIME) {
 		osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
 				 osa->ctime, 16);
 	}
-	if (la->la_valid & LA_MODE) {
+	if (valid & LA_MODE) {
 		/* mode is stored along with type, so read it first */
 		obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 			(la->la_mode & ~S_IFMT);
@@ -1012,22 +1016,22 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
 				 &osa->mode, 8);
 	}
-	if (la->la_valid & LA_SIZE) {
+	if (valid & LA_SIZE) {
 		osa->size = obj->oo_attr.la_size = la->la_size;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
 				 &osa->size, 8);
 	}
-	if (la->la_valid & LA_NLINK) {
+	if (valid & LA_NLINK) {
 		osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
 				 &osa->nlink, 8);
 	}
-	if (la->la_valid & LA_RDEV) {
+	if (valid & LA_RDEV) {
 		osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
 				 &osa->rdev, 8);
 	}
-	if (la->la_valid & LA_FLAGS) {
+	if (valid & LA_FLAGS) {
 		osa->flags = attrs_fs2zfs(la->la_flags);
 		/* many flags are not supported by zfs, so ensure a good cached
 		 * copy */
@@ -1035,17 +1039,17 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
 				 &osa->flags, 8);
 	}
-	if (la->la_valid & LA_UID) {
+	if (valid & LA_UID) {
 		osa->uid = obj->oo_attr.la_uid = la->la_uid;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
 				 &osa->uid, 8);
 	}
-	if (la->la_valid & LA_GID) {
+	if (valid & LA_GID) {
 		osa->gid = obj->oo_attr.la_gid = la->la_gid;
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
 				 &osa->gid, 8);
 	}
-	obj->oo_attr.la_valid |= la->la_valid;
+	obj->oo_attr.la_valid |= valid;
 	write_unlock(&obj->oo_attr_lock);
 
 	rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
@@ -1062,7 +1066,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 
 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
 			struct dt_object *parent, struct dt_object *child,
-			cfs_umode_t child_mode)
+			umode_t child_mode)
 {
 	LASSERT(ah);
 
@@ -1227,10 +1231,6 @@ int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
 	int	 rc;
 
 	LASSERT(tag);
-	spin_lock(&uos->lock);
-	uos->objects++;
-	spin_unlock(&uos->lock);
-
 	/* Assert that the transaction has been assigned to a
 	   transaction group. */
 	LASSERT(tx->tx_txg != 0);
@@ -1239,14 +1239,24 @@ int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
 	oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
 			       DMU_OT_SA, DN_MAX_BONUSLEN, tx);
 	rc = -sa_buf_hold(uos->os, oid, tag, dbp);
-	if (rc)
-		return rc;
+	LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
 
 	LASSERT(la->la_valid & LA_MODE);
 	la->la_size = 0;
 	la->la_nlink = 1;
 
-	return __osd_attr_init(env, uos, oid, tx, la, parent);
+	rc = __osd_attr_init(env, uos, oid, tx, la, parent);
+	if (rc != 0) {
+		sa_buf_rele(*dbp, tag);
+		*dbp = NULL;
+		dmu_object_free(uos->os, oid, tx);
+		return rc;
+	}
+
+	spin_lock(&uos->lock);
+	uos->objects++;
+	spin_unlock(&uos->lock);
+	return 0;
 }
 
 /*
@@ -1278,7 +1288,9 @@ int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
 	LASSERT(tx->tx_txg != 0);
 
 	oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
-			       DMU_OT_DIRECTORY_CONTENTS, 12, 12,
+			       DMU_OT_DIRECTORY_CONTENTS,
+			       14, /* == ZFS fzap_default_block_shift */
+			       DN_MAX_INDBLKSHIFT, /* indirect block shift */
 			       DMU_OT_SA, DN_MAX_BONUSLEN, tx);
 
 	rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
@@ -1434,7 +1446,7 @@ static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
 	struct lu_buf		 buf;
 	int rc;
 
-	lustre_lma_init(lma, fid, 0);
+	lustre_lma_init(lma, fid, 0, 0);
 	lustre_lma_swab(lma);
 	buf.lb_buf = lma;
 	buf.lb_len = sizeof(*lma);
@@ -1492,7 +1504,7 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
 
 	db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
 	if (IS_ERR(db))
-		GOTO(out, rc = PTR_ERR(th));
+		GOTO(out, rc = PTR_ERR(db));
 
 	zde->zde_pad = 0;
 	zde->zde_dnode = db->db_object;
@@ -1752,13 +1764,16 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env,
 	RETURN(oc);
 }
 
-static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
+static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
+			   __u64 start, __u64 end)
 {
 	struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
 	ENTRY;
 
 	/* XXX: no other option than syncing the whole filesystem until we
-	 * support ZIL */
+	 * support ZIL.  If the object tracked the txg that it was last
+	 * modified in, it could pass that txg here instead of "0".  Maybe
+	 * the changes are already committed, so no wait is needed at all? */
 	txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
 
 	RETURN(0);