LU-8342 utils: Set dnodesize/recordsize at zfs dataset create

After the zfs dataset is created, attempt to set the
dnodesize and recordsize properties. Moved xattr=sa to be
consistent with the new method of setting dataset properties.

Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Change-Id: I12e5863e4602496b85f8512ea780be4589489d01
Reviewed-on: https://review.whamcloud.com/21055
Tested-by: Jenkins
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Olaf Faaland-LLNL <faaland1@llnl.gov>
Reviewed-by: Nathaniel Clark <nathaniel.l.clark@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/lustre/utils/libmount_utils_zfs.c b/lustre/utils/libmount_utils_zfs.c
index 920682c..b616225 100644 (file)
--- a/lustre/utils/libmount_utils_zfs.c
+++ b/lustre/utils/libmount_utils_zfs.c
@@ -717,11 +717,10 @@ int zfs_make_lustre(struct mkfs_opts *mop)
        /*
         * Create the ZFS filesystem with any required mkfs options:
         * - canmount=off is set to prevent zfs automounting
-        * - xattr=sa is set to use system attribute based xattrs
         */
        memset(mkfs_cmd, 0, PATH_MAX);
        snprintf(mkfs_cmd, PATH_MAX,
-                "zfs create -o canmount=off -o xattr=sa%s %s",
+                "zfs create -o canmount=off %s %s",
                 zfs_mkfs_opts(mop, mkfs_tmp, PATH_MAX), ds);
 
        vprint("mkfs_cmd = %s\n", mkfs_cmd);
@@ -733,6 +732,42 @@ int zfs_make_lustre(struct mkfs_opts *mop)
                goto out;
        }
 
+       /*
+        * Attempt to set dataset properties to reasonable defaults
+        * to optimize performance, unless the values were specified
+        * at the mkfs command line. Some ZFS pools or ZFS versions
+        * do not support these properties. We can safely ignore the
+        * errors and continue in those cases.
+        *
+        * zfs 0.6.1 - system attribute based xattrs
+        * zfs 0.6.5 - large block support
+        * zfs 0.7.0 - large dnode support
+        *
+        * Check if zhp is NULL as a defensive measure. Any dataset
+        * validation errors that would cause zfs_open() to fail
+        * should have been caught earlier.
+        */
+       zhp = zfs_open(g_zfs, ds, ZFS_TYPE_FILESYSTEM);
+       if (zhp) {
+               /* zfs 0.6.1 - system attribute based xattrs */
+               if (!strstr(mop->mo_mkfsopts, "xattr="))
+                       zfs_set_prop_str(zhp, "xattr", "sa");
+
+               /* zfs 0.7.0 - large dnode support */
+               if (!strstr(mop->mo_mkfsopts, "dnodesize=") &&
+                   !strstr(mop->mo_mkfsopts, "dnsize="))
+                       zfs_set_prop_str(zhp, "dnodesize", "auto");
+
+               if (IS_OST(&mop->mo_ldd)) {
+                       /* zfs 0.6.5 - large block support */
+                       if (!strstr(mop->mo_mkfsopts, "recordsize=") &&
+                           !strstr(mop->mo_mkfsopts, "recsize="))
+                               zfs_set_prop_str(zhp, "recordsize", "1M");
+               }
+
+               zfs_close(zhp);
+       }
+
 out:
        if (pool != NULL)
                free(pool);