LU-8068 osd-zfs: large dnode support

In ZFS 0.7.x support for large dnode has been added.  This allows
Lustre to optionally specify the size of a dnode which ZFS will
then use to store metadata such as xattrs.

The default dnode size specified by the 'dnodesize' property on the
dataset will be used unless a specific value is provided.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Change-Id: Iaa7211ca50e7089af4b1fabcccdaa6e16477e641
Reviewed-on: http://review.whamcloud.com/20367
Tested-by: Jenkins
Reviewed-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Reviewed-by: Nathaniel Clark <nathaniel.l.clark@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/config/lustre-build-zfs.m4 b/config/lustre-build-zfs.m4
index 611e09d..3aa9314 100644 (file)
--- a/config/lustre-build-zfs.m4
+++ b/config/lustre-build-zfs.m4
@@ -450,6 +450,35 @@ your distribution.
                        AC_DEFINE(HAVE_SPA_MAXBLOCKSIZE, 1,
                                [Have spa_maxblocksize in ZFS])
                ])
+               dnl #
+               dnl # ZFS 0.7.x adds support for large dnodes.  This
+               dnl # allows Lustre to optionally specify the size of a
+               dnl # dnode which ZFS will then use to store metadata such
+               dnl # as xattrs. The default dnode size specified by the
+               dnl # 'dnodesize' dataset property will be used unless a
+               dnl # specific value is provided.
+               dnl #
+               LB_CHECK_COMPILE([if zfs defines dmu_object_alloc_dnsize],
+               dmu_object_alloc_dnsize, [
+                       #include <sys/dmu.h>
+                       #include <sys/dnode.h>
+               ],[
+                       objset_t *os = NULL;
+                       dmu_object_type_t objtype = DMU_OT_NONE;
+                       int blocksize = 0;
+                       dmu_object_type_t bonustype = DMU_OT_SA;
+                       int dnodesize = DNODE_MIN_SIZE;
+                       dmu_tx_t *tx = NULL;
+                       uint64_t id;
+
+                       id = dmu_object_alloc_dnsize(os, objtype, blocksize,
+                                                    bonustype,
+                                                    DN_BONUS_SIZE(dnodesize),
+                                                    dnodesize, tx);
+               ],[
+                       AC_DEFINE(HAVE_DMU_OBJECT_ALLOC_DNSIZE, 1,
+                               [Have dmu_object_alloc_dnsize in ZFS])
+               ])
        ])
 
        AM_CONDITIONAL(ZFS_ENABLED, [test "x$enable_zfs" = xyes])

diff --git a/lustre/osd-zfs/osd_internal.h b/lustre/osd-zfs/osd_internal.h
index a59ff87..e3daa1f 100644 (file)
--- a/lustre/osd-zfs/osd_internal.h
+++ b/lustre/osd-zfs/osd_internal.h
@@ -595,4 +595,49 @@ osd_zio_buf_free(void *buf, size_t size)
 #define        osd_zio_buf_free(buf, size)     zio_buf_free(buf, size)
 #endif
 
+#ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
+static inline uint64_t
+osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
+                    int dnodesize, dmu_tx_t *tx)
+{
+       if (dnodesize == 0)
+               dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
+
+       return dmu_object_alloc_dnsize(os, objtype, blocksize, DMU_OT_SA,
+                                      DN_BONUS_SIZE(dnodesize), dnodesize, tx);
+}
+
+static inline uint64_t
+osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
+                    dmu_object_type_t ot, int leaf_blockshift,
+                    int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
+{
+       if (dnodesize == 0)
+               dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
+
+       return zap_create_flags_dnsize(os, normflags, flags, ot,
+                                      leaf_blockshift, indirect_blockshift,
+                                      DMU_OT_SA, DN_BONUS_SIZE(dnodesize),
+                                      dnodesize, tx);
+}
+#else
+static inline uint64_t
+osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
+                    int dnodesize, dmu_tx_t *tx)
+{
+       return dmu_object_alloc(os, objtype, blocksize, DMU_OT_SA,
+                               DN_MAX_BONUSLEN, tx);
+}
+
+static inline uint64_t
+osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
+                    dmu_object_type_t ot, int leaf_blockshift,
+                    int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
+{
+       return zap_create_flags(os, normflags, flags, ot, leaf_blockshift,
+                               indirect_blockshift, DMU_OT_SA,
+                               DN_MAX_BONUSLEN, tx);
+}
+#endif /* HAVE_DMU_OBJECT_ALLOC_DNSIZE */
+
 #endif /* _OSD_INTERNAL_H */

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index 2575ea4..b534906 100644 (file)
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -1293,9 +1293,8 @@ int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
                     fid_seq_is_local_file(fid_seq(fid))))
                type = DMU_OTN_UINT8_METADATA;
 
-       /* Create a new DMU object. */
-       oid = dmu_object_alloc(osd->od_os, type, 0,
-                              DMU_OT_SA, DN_MAX_BONUSLEN, tx);
+       /* Create a new DMU object using the default dnode size. */
+       oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
        rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
        LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
 
@@ -1335,11 +1334,11 @@ int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
           transaction group. */
        LASSERT(tx->tx_txg != 0);
 
-       oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
-                              DMU_OT_DIRECTORY_CONTENTS,
-                              14, /* == ZFS fzap_default_block_shift */
-                              DN_MAX_INDBLKSHIFT, /* indirect block shift */
-                              DMU_OT_SA, DN_MAX_BONUSLEN, tx);
+       oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
+                                  DMU_OT_DIRECTORY_CONTENTS,
+                                  14, /* == ZFS fzap_default_blockshift */
+                                  DN_MAX_INDBLKSHIFT, /* indirect blockshift */
+                                  0, tx);
 
        rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
        if (rc)