LU-9489 lod: keep minimum LOVEA size

For a PFL file, some of its component could be un-instantiated, and
their lov_ost_data_v1 array is not needed, we should keep its LOVEA
as small as possible.

An unstantiated component's stripe offset should be set.

Signed-off-by: Bobi Jam <bobijam.xu@intel.com>
Change-Id: I1cd70399446f063dca4de5b61d7e1dc2c8dde37c
Reviewed-on: https://review.whamcloud.com/27089
Tested-by: Jenkins
Reviewed-by: Niu Yawei <yawei.niu@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>

diff --git a/lustre/lod/lod_internal.h b/lustre/lod/lod_internal.h
index 03c19cc..b388347 100644 (file)
--- a/lustre/lod/lod_internal.h
+++ b/lustre/lod/lod_internal.h
@@ -560,6 +560,26 @@ lod_comp_inited(const struct lod_layout_component *entry)
        return entry->llc_flags & LCME_FL_INIT;
 }
 
        return entry->llc_flags & LCME_FL_INIT;
 }
 

+/**
+ * For a PFL file, some of its component could be un-instantiated, so
+ * that their lov_ost_data_v1 array is not needed, we'd use this function
+ * to reduce the LOVEA buffer size.
+ *
+ * Note: if llc_ostlist contains value, we'd need lov_ost_data_v1 array to
+ * save the specified OST index list.
+ */
+static inline void
+lod_comp_shrink_stripecount(struct lod_layout_component *lod_comp,
+                           __u16 *stripe_count)
+{
+       /**
+        * Need one lov_ost_data_v1 to store invalid ost_idx, please refer to
+        * lod_parse_striping()
+        */
+       if (!lod_comp_inited(lod_comp) && lod_comp->llc_ostlist.op_count == 0)
+               *stripe_count = 1;
+}
+

 void lod_fix_desc(struct lov_desc *desc);
 void lod_fix_desc_qos_maxage(__u32 *val);
 void lod_fix_desc_pattern(__u32 *val);
 void lod_fix_desc(struct lov_desc *desc);
 void lod_fix_desc_qos_maxage(__u32 *val);
 void lod_fix_desc_pattern(__u32 *val);

diff --git a/lustre/lod/lod_lov.c b/lustre/lod/lod_lov.c
index 6b01a87..671b768 100644 (file)
--- a/lustre/lod/lod_lov.c
+++ b/lustre/lod/lod_lov.c
@@ -789,6 +789,8 @@ static int lod_gen_component_ea(const struct lu_env *env,
                objs = &v3->lmm_objects[0];
        }
        stripecnt = lod_comp_entry_stripecnt(lo, lod_comp, is_dir);
                objs = &v3->lmm_objects[0];
        }
        stripecnt = lod_comp_entry_stripecnt(lo, lod_comp, is_dir);

+       if (!is_dir && lo->ldo_is_composite)
+               lod_comp_shrink_stripecount(lod_comp, &stripecnt);

 
        if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
                GOTO(done, rc = 0);
 
        if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
                GOTO(done, rc = 0);


@@ -1273,26 +1275,42 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 
                /**
                 * If uninstantiated template component has valid l_ost_idx,
 
                /**
                 * If uninstantiated template component has valid l_ost_idx,

-                * then use has specified ost list for this component.
+                * then user has specified ost list for this component.

                 */
                 */

-               if (!lod_comp_inited(lod_comp) &&
-                   objs[0].l_ost_idx != (__u32)-1UL) {
-                       /**
-                        * load the user specified ost list, when this
-                        * component is instantiated later, it will be used
-                        * in lod_alloc_ost_list().
-                        */
-                       lod_comp->llc_ostlist.op_count = lod_comp->llc_stripenr;
-                       lod_comp->llc_ostlist.op_size =
+               if (!lod_comp_inited(lod_comp)) {
+                       if (objs[0].l_ost_idx != (__u32)-1UL) {
+                               /**
+                                * load the user specified ost list, when this
+                                * component is instantiated later, it will be
+                                * used in lod_alloc_ost_list().
+                                */
+                               lod_comp->llc_ostlist.op_count =
+                                       lod_comp->llc_stripenr;
+                               lod_comp->llc_ostlist.op_size =

                                        lod_comp->llc_stripenr * sizeof(__u32);
                                        lod_comp->llc_stripenr * sizeof(__u32);

-                       OBD_ALLOC(lod_comp->llc_ostlist.op_array,
-                                 lod_comp->llc_ostlist.op_size);
-                       if (!lod_comp->llc_ostlist.op_array)
-                               GOTO(out, rc = -ENOMEM);
+                               OBD_ALLOC(lod_comp->llc_ostlist.op_array,
+                                         lod_comp->llc_ostlist.op_size);
+                               if (!lod_comp->llc_ostlist.op_array)
+                                       GOTO(out, rc = -ENOMEM);

 
 

-                       for (j = 0; j < lod_comp->llc_stripenr; j++)
-                               lod_comp->llc_ostlist.op_array[j] =
+                               for (j = 0; j < lod_comp->llc_stripenr; j++)
+                                       lod_comp->llc_ostlist.op_array[j] =

                                                le32_to_cpu(objs[j].l_ost_idx);
                                                le32_to_cpu(objs[j].l_ost_idx);

+
+                               /**
+                                * this component OST objects starts from the
+                                * first ost_idx, lod_alloc_ost_list() will
+                                * check this.
+                                */
+                               lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
+                       } else {
+                               /**
+                                * for uninstantiated component,
+                                * lmm_layout_gen stores default stripe offset.
+                                */
+                               lod_comp->llc_stripe_offset =
+                                                       lmm->lmm_layout_gen;
+                       }

                }
 
                /* skip un-instantiated component object initialization */
                }
 
                /* skip un-instantiated component object initialization */

diff --git a/lustre/lod/lod_object.c b/lustre/lod/lod_object.c
index 2318bdb..f4a86a9 100644 (file)
--- a/lustre/lod/lod_object.c
+++ b/lustre/lod/lod_object.c
@@ -2190,6 +2190,10 @@ static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
                magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
                stripenr = lod_comp_entry_stripecnt(lo, &comp_entries[i],
                                                    is_dir);
                magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
                stripenr = lod_comp_entry_stripecnt(lo, &comp_entries[i],
                                                    is_dir);

+               if (!is_dir && is_composite)
+                       lod_comp_shrink_stripecount(&comp_entries[i],
+                                                   &stripenr);
+

                size += lov_user_md_size(stripenr, magic);
                LASSERT(size % sizeof(__u64) == 0);
        }
                size += lov_user_md_size(stripenr, magic);
                LASSERT(size % sizeof(__u64) == 0);
        }

diff --git a/lustre/lov/lov_internal.h b/lustre/lov/lov_internal.h
index b9743b6..6f74868 100644 (file)
--- a/lustre/lov/lov_internal.h
+++ b/lustre/lov/lov_internal.h
@@ -80,6 +80,21 @@ struct lov_stripe_md {
        struct lov_stripe_md_entry *lsm_entries[];
 };
 
        struct lov_stripe_md_entry *lsm_entries[];
 };
 

+static inline bool lsme_inited(const struct lov_stripe_md_entry *lsme)
+{
+       return lsme->lsme_flags & LCME_FL_INIT;
+}
+
+static inline bool lsm_entry_inited(const struct lov_stripe_md *lsm, int index)
+{
+       return lsme_inited(lsm->lsm_entries[index]);
+}
+
+static inline bool lsm_is_composite(__u32 magic)
+{
+       return !!(magic & LOV_MAGIC_COMP_V1);
+}
+

 static inline size_t lov_comp_md_size(const struct lov_stripe_md *lsm)
 {
        struct lov_stripe_md_entry *lsme;
 static inline size_t lov_comp_md_size(const struct lov_stripe_md *lsm)
 {
        struct lov_stripe_md_entry *lsme;


@@ -94,8 +109,15 @@ static inline size_t lov_comp_md_size(const struct lov_stripe_md *lsm)
 
        size = sizeof(struct lov_comp_md_v1);
        for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
 
        size = sizeof(struct lov_comp_md_v1);
        for (entry = 0; entry < lsm->lsm_entry_count; entry++) {

+               u16 stripe_count;
+

                lsme = lsm->lsm_entries[entry];
 
                lsme = lsm->lsm_entries[entry];
 

+               if (lsme_inited(lsme))
+                       stripe_count = lsme->lsme_stripe_count;
+               else
+                       stripe_count = 0;
+

                size += sizeof(*lsme);
                size += lov_mds_md_size(lsme->lsme_stripe_count,
                                        lsme->lsme_magic);
                size += sizeof(*lsme);
                size += lov_mds_md_size(lsme->lsme_stripe_count,
                                        lsme->lsme_magic);


@@ -338,20 +360,4 @@ static inline void lov_lsm2layout(struct lov_stripe_md *lsm,
                ol->ol_comp_id = 0;
        }
 }
                ol->ol_comp_id = 0;
        }
 }

-
-static inline bool lsme_inited(const struct lov_stripe_md_entry *lsme)
-{
-       return lsme->lsme_flags & LCME_FL_INIT;
-}
-
-static inline bool lsm_entry_inited(const struct lov_stripe_md *lsm, int index)
-{
-       return lsme_inited(lsm->lsm_entries[index]);
-}
-
-static inline bool lsm_is_composite(__u32 magic)
-{
-       return !!(magic & LOV_MAGIC_COMP_V1);
-}
-

 #endif
 #endif

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 525d436..1fdde25 100755 (executable)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -6158,7 +6158,8 @@ test_87() { #LU-6544
        #set xattr
        $SETSTRIPE -E 1M -c 1 -E 64M -c 1 -E -1 -c -1 $file ||
                error "Create file with 3 components failed"
        #set xattr
        $SETSTRIPE -E 1M -c 1 -E 64M -c 1 -E -1 -c -1 $file ||
                error "Create file with 3 components failed"

-       i=$($GETSTRIPE -I 3 -c $file)
+       $TRUNCATE $file $((1024*1024*64+1)) || error "truncate file failed"
+       i=$($GETSTRIPE -I3 -c $file) || error "get 3rd stripe count failed"

        if [ $i -ne $OSTCOUNT ]; then
                left_size=$(expr $left_size + $(expr $OSTCOUNT - $i) \* 24)
                echo -n "Since only $i out $OSTCOUNT OSTs are used, "
        if [ $i -ne $OSTCOUNT ]; then
                left_size=$(expr $left_size + $(expr $OSTCOUNT - $i) \* 24)
                echo -n "Since only $i out $OSTCOUNT OSTs are used, "

diff --git a/lustre/tests/sanity-lfsck.sh b/lustre/tests/sanity-lfsck.sh
index a4d68b2..79e130e 100644 (file)
--- a/lustre/tests/sanity-lfsck.sh
+++ b/lustre/tests/sanity-lfsck.sh
@@ -2508,8 +2508,8 @@ test_18e() {
        sleep 2
 
        echo "Write new data to f2/f4 to modify the new created OST-object."
        sleep 2
 
        echo "Write new data to f2/f4 to modify the new created OST-object."

-       echo "dummy" >> $DIR/$tdir/a1/f2
-       echo "dummy" >> $DIR/$tdir/a1/f4
+       echo "dummy" >> $DIR/$tdir/a1/f2 || error "write a1/f2 failed"
+       echo "dummy" >> $DIR/$tdir/a1/f4 || error "write a1/f4 failed"

 
        do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
 
 
        do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
 

diff --git a/lustre/tests/sanity-pfl.sh b/lustre/tests/sanity-pfl.sh
index 3c83c37..cc4d01e 100644 (file)
--- a/lustre/tests/sanity-pfl.sh
+++ b/lustre/tests/sanity-pfl.sh
@@ -361,7 +361,7 @@ test_9() {
        local f2=$($LFS getstripe -I2 $comp_file |
                        awk '/l_fid:/ {print $7}')
        echo "after MDS recovery, the ost fid of 2nd component is $f2"
        local f2=$($LFS getstripe -I2 $comp_file |
                        awk '/l_fid:/ {print $7}')
        echo "after MDS recovery, the ost fid of 2nd component is $f2"

-       [ $f1 == $f2 ] || error "$f1 != $f2"
+       [ "x$f1" == "x$f2" ] || error "$f1 != $f2"

 }
 run_test 9 "Replay layout extend object instantiation"
 
 }
 run_test 9 "Replay layout extend object instantiation"
 


@@ -479,8 +479,14 @@ test_12() {
 
        # specify ost list for component
        $LFS setstripe -E1m -c2 -o0,1 -E2m -c2 -o1,2 -E3m -c2 -o2,1 \
 
        # specify ost list for component
        $LFS setstripe -E1m -c2 -o0,1 -E2m -c2 -o1,2 -E3m -c2 -o2,1 \

-               -E4m -c2 -o2,0 -E-1 $file ||
+               -E4m -c1 -i2 -E-1 $file ||

                error "Create $file failed"
                error "Create $file failed"

+
+       # clear lod component cache
+       stop $SINGLEMDS || error "stop MDS"
+       local MDT_DEV=$(mdsdevname ${SINGLEMDS//mds/})
+       start $SINGLEMDS $MDT_DEV $MDS_MOUNT_OPTS || error "start MDS"
+

        # instantiate all components
        $TRUNCATE $file $((1024*1024*4+1))
 
        # instantiate all components
        $TRUNCATE $file $((1024*1024*4+1))
 


@@ -499,7 +505,7 @@ test_12() {
 
        local o4=$($LFS getstripe -I4 $file |
                        awk '/l_ost_idx:/ {printf("%d",$5)}')
 
        local o4=$($LFS getstripe -I4 $file |
                        awk '/l_ost_idx:/ {printf("%d",$5)}')

-       [[ $o4 != "20" ]] && error "$o4 is not 20"
+       [[ $o4 != "2" ]] && error "$o4 is not 2"

 
        return 0
 }
 
        return 0
 }


@@ -630,6 +636,42 @@ test_15() {
 }
 run_test 15 "Verify component options for lfs find"
 
 }
 run_test 15 "Verify component options for lfs find"
 

+test_17() {
+       [ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
+       local file=$DIR/$tdir/$tfile
+       test_mkdir -p $DIR/$tdir
+       rm -f $file
+
+       $LFS setstripe -E1m -E2m -c2 -E-1 -c-1 $file ||
+               error "Create $file failed"
+
+       local s1=$($LFS getstripe -I1 -v $file | awk '/lcme_size:/{print $2}')
+       local s2=$($LFS getstripe -I2 -v $file | awk '/lcme_size:/{print $2}')
+       local s3=$($LFS getstripe -I3 -v $file | awk '/lcme_size:/{print $2}')
+       echo "1st init: comp size 1:$s1 2:$s2 3:$s3"
+
+       # init 2nd component
+       $TRUNCATE $file $((1024*1024+1))
+       local s1n=$($LFS getstripe -I1 -v $file | awk '/lcme_size:/{print $2}')
+       local s2n=$($LFS getstripe -I2 -v $file | awk '/lcme_size:/{print $2}')
+       echo "2nd init: comp size 1:$s1n 2:$s2n 3:$s3"
+
+       [ $s1 -eq $s1n ] || error "1st comp size $s1 should == $s1n"
+       [ $s2 -lt $s2n ] || error "2nd comp size $s2 should < $s2n"
+
+       # init 3rd component
+       $TRUNCATE $file $((1024*1024*2+1))
+       s1n=$($LFS getstripe -I1 -v $file | awk '/lcme_size:/{print $2}')
+       s2n=$($LFS getstripe -I2 -v $file | awk '/lcme_size:/{print $2}')
+       local s3n=$($LFS getstripe -I3 -v $file | awk '/lcme_size:/{print $2}')
+       echo "3rd init: comp size 1:$s1n 2:$s2n 3:$s3n"
+
+       [ $s1 -eq $s1n ] || error "1st comp size $s1 should == $s1n"
+       [ $s2 -lt $s2n ] || error "2nd comp size $s2 should < $s2n"
+       [ $s3 -lt $s3n ] || error "3rd comp size $s3 should < $s3n"
+}
+run_test 17 "Verify LOVEA grows with more component inited"
+

 complete $SECONDS
 check_and_cleanup_lustre
 exit_status
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status