LU-10248 mdd: set PFID for swap and merge layout

It should update PFID for the OST objects if the layout has been
moved to another file. So far only swap and merge layout would
have this requirement.

Signed-off-by: Jinshan Xiong <jinshan.xiong@intel.com>
Change-Id: Ic2fd7d7a88cd0ef4501d2f02baf58dd177e07973
Reviewed-on: https://review.whamcloud.com/30292
Tested-by: Jenkins
Reviewed-by: Fan Yong <fan.yong@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/lustre/lod/lod_object.c b/lustre/lod/lod_object.c
index bdffb85..bf56b50 100644 (file)
--- a/lustre/lod/lod_object.c
+++ b/lustre/lod/lod_object.c
@@ -2133,6 +2133,7 @@ lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
                                     struct lod_obj_stripe_cb_data *data)
 {
        struct lod_thread_info *info = lod_env_info(env);
+       struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
        struct filter_fid *ff = &info->lti_ff;
        struct lu_buf *buf = &info->lti_buf;
        int rc;
@@ -2140,15 +2141,28 @@ lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
        buf->lb_buf = ff;
        buf->lb_len = sizeof(*ff);
        rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
-       if (rc == -ENODATA)
-               return 0;
-
-       if (rc < 0)
+       if (rc < 0) {
+               if (rc == -ENODATA)
+                       return 0;
                return rc;
+       }
+
+       filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
+       if (lu_fid_eq(lu_object_fid(&lo->ldo_obj.do_lu), &ff->ff_parent) &&
+           ff->ff_layout.ol_comp_id == comp->llc_id)
+               return 0;
 
+       /* rewrite filter_fid */
+       memset(ff, 0, sizeof(*ff));
        ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
        ff->ff_parent.f_ver = stripe_idx;
-       fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
+       ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
+       ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
+       ff->ff_layout.ol_comp_id = comp->llc_id;
+       ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
+       ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
+       filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
+
        if (data->locd_declare)
                rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
                                               LU_XATTR_REPLACE, th);

diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c
index e15141e..fe246d1 100644 (file)
--- a/lustre/mdd/mdd_object.c
+++ b/lustre/mdd/mdd_object.c
@@ -1084,6 +1084,40 @@ free:
        return rc;
 }
 
+static int mdd_object_pfid_replace(const struct lu_env *env,
+                                  struct mdd_object *o)
+{
+       struct mdd_device *mdd = mdo2mdd(&o->mod_obj);
+       struct thandle *handle;
+       int rc;
+
+       handle = mdd_trans_create(env, mdd);
+       if (IS_ERR(handle))
+               RETURN(PTR_ERR(handle));
+
+       handle->th_complex = 1;
+
+       /* it doesn't need to track the PFID update via llog, because LFSCK
+        * will repair it even it goes wrong */
+       rc = mdd_declare_xattr_set(env, mdd, o, NULL, XATTR_NAME_FID,
+                                  0, handle);
+       if (rc)
+               GOTO(out, rc);
+
+       rc = mdd_trans_start(env, mdd, handle);
+       if (rc != 0)
+               GOTO(out, rc);
+
+       rc = mdo_xattr_set(env, o, NULL, XATTR_NAME_FID, 0, handle);
+       if (rc)
+               GOTO(out, rc);
+
+out:
+       mdd_trans_stop(env, mdd, rc, handle);
+       return rc;
+}
+
+
 static int mdd_declare_xattr_del(const struct lu_env *env,
                                 struct mdd_device *mdd,
                                 struct mdd_object *obj,
@@ -1162,22 +1196,23 @@ static int mdd_xattr_merge(const struct lu_env *env, struct md_object *md_obj,
                GOTO(out, rc);
 
        rc = mdo_xattr_del(env, vic, XATTR_NAME_LOV, handle);
-       if (rc) { /* wtf? */
-               int rc2;
+       if (rc) /* wtf? */
+               GOTO(out_restore, rc);
 
-               rc2 = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV,
-                                   LU_XATTR_REPLACE, handle);
+       (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
+       (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
+       EXIT;
+
+out_restore:
+       if (rc) {
+               int rc2 = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV,
+                                       LU_XATTR_REPLACE, handle);
                if (rc2)
                        CERROR("%s: failed to rollback of layout of: "DFID
                               ": %d, file state unknown\n",
                               mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2);
-               GOTO(out, rc);
        }
 
-       (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
-       (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
-       EXIT;
-
 out:
        mdd_trans_stop(env, mdd, rc, handle);
        mdd_write_unlock(env, obj);
@@ -1185,6 +1220,9 @@ out:
        lu_buf_free(buf);
        lu_buf_free(buf_vic);
 
+       if (!rc)
+               (void) mdd_object_pfid_replace(env, obj);
+
        return rc;
 }
 
@@ -1797,7 +1835,20 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
                else
                        rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
        }
+       if (rc != 0)
+               GOTO(out_restore, rc);
+
+       /* Issue one changelog record per file */
+       rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
+       if (rc)
+               GOTO(stop, rc);
+
+       rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
+       if (rc)
+               GOTO(stop, rc);
+       EXIT;
 
+out_restore:
        if (rc != 0) {
                int steps = 0;
 
@@ -1836,19 +1887,8 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
                         */
                        LBUG();
                }
-               GOTO(stop, rc);
        }
 
-       /* Issue one changelog record per file */
-       rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
-       if (rc)
-               GOTO(stop, rc);
-
-       rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
-       if (rc)
-               GOTO(stop, rc);
-       EXIT;
-
 stop:
        rc = mdd_trans_stop(env, mdd, rc, handle);
 
@@ -1859,6 +1899,11 @@ stop:
        lu_buf_free(snd_buf);
        lu_buf_free(fst_hsm_buf);
        lu_buf_free(snd_hsm_buf);
+
+       if (!rc) {
+               (void) mdd_object_pfid_replace(env, fst_o);
+               (void) mdd_object_pfid_replace(env, snd_o);
+       }
        return rc;
 }
 

diff --git a/lustre/tests/sanity-flr.sh b/lustre/tests/sanity-flr.sh
index abf81d5..ba90384 100644 (file)
--- a/lustre/tests/sanity-flr.sh
+++ b/lustre/tests/sanity-flr.sh
@@ -381,6 +381,8 @@ test_0d() {
                verify_comp_extent $tf ${ids[$i]} 0 EOF
        done
 
+       lfsck_verify_pfid $tf || error "PFID is not set"
+
        # create a mirrored file and extend it
        $LFS mirror create -N $tf-1 || error "create mirrored file $tf-1 failed"
        $LFS mirror create -N $tf-2 || error "create mirrored file $tf-2 failed"

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 73211a3..d66fa79 100755 (executable)
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -12266,6 +12266,8 @@ test_184a() {
 
        cmp $ref1 $file2 || error "content compare failed ($ref1 != $file2)"
        cmp $ref2 $file1 || error "content compare failed ($ref2 != $file1)"
+
+       lfsck_verify_pfid $file1 $file2 || error "PFID are not transferred"
 }
 run_test 184a "Basic layout swap"
 

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index cb9d21a..113ea76 100755 (executable)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -8420,3 +8420,27 @@ get_layout_param()
        local param=$($LFS getstripe -d $1 | parse_layout_param)
        echo "$param"
 }
+
+lfsck_verify_pfid()
+{
+       local f
+       local rc=0
+
+       # Cancel locks before setting lfsck_verify_pfid so that errors are more
+        # controllable
+       cancel_lru_locks mdc
+       cancel_lru_locks osc
+        
+       # make sure PFID is set correctly for files
+       do_nodes $(comma_list $(osts_nodes)) \
+              "$LCTL set_param -n obdfilter.${FSNAME}-OST*.lfsck_verify_pfid=1"
+
+       for f in "$@"; do
+               cat $f &> /dev/nullA ||
+                       { rc=$?; echo "verify $f failed"; break; }
+       done
+
+       do_nodes $(comma_list $(osts_nodes)) \
+              "$LCTL set_param -n obdfilter.${FSNAME}-OST*.lfsck_verify_pfid=0"
+       return $rc
+}