LU-13415 dom: use DoM optimization for DOM+OST files

[fs/lustre-release.git] / lustre / mdt / mdt_handler.c

diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c
index 0f3ff88..0891ba5 100644 (file)
--- a/lustre/mdt/mdt_handler.c
+++ b/lustre/mdt/mdt_handler.c
@@ -509,6 +509,58 @@ out:
        RETURN(rc);
 }
 
+__u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
+{
+       struct lov_comp_md_v1 *comp_v1;
+       struct lov_mds_md *v1;
+       __u32 off;
+       __u32 dom_stripesize = 0;
+       int i;
+       bool has_ost_stripes = false;
+
+       ENTRY;
+
+       if (is_dom_only)
+               *is_dom_only = 0;
+
+       if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
+               RETURN(0);
+
+       comp_v1 = (struct lov_comp_md_v1 *)lmm;
+       off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
+       v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
+
+       /* Fast check for DoM entry with no mirroring, should be the first */
+       if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
+           lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
+               RETURN(0);
+
+       /* check all entries otherwise */
+       for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
+               struct lov_comp_md_entry_v1 *lcme;
+
+               lcme = &comp_v1->lcm_entries[i];
+               if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
+                       continue;
+
+               off = le32_to_cpu(lcme->lcme_offset);
+               v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
+
+               if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) ==
+                   LOV_PATTERN_MDT)
+                       dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
+               else
+                       has_ost_stripes = true;
+
+               if (dom_stripesize && has_ost_stripes)
+                       RETURN(dom_stripesize);
+       }
+       /* DoM-only case exits here */
+       if (is_dom_only && dom_stripesize)
+               *is_dom_only = 1;
+       RETURN(dom_stripesize);
+}
+
 /**
  * Pack size attributes into the reply.
  */
@@ -517,7 +569,7 @@ int mdt_pack_size2body(struct mdt_thread_info *info,
 {
        struct mdt_body *b;
        struct md_attr *ma = &info->mti_attr;
-       int dom_stripe;
+       __u32 dom_stripe;
        bool dom_lock = false;
 
        ENTRY;
@@ -528,9 +580,9 @@ int mdt_pack_size2body(struct mdt_thread_info *info,
            !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
                RETURN(-ENODATA);
 
-       dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
+       dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
        /* no DoM stripe, no size in reply */
-       if (dom_stripe == LMM_NO_DOM)
+       if (!dom_stripe)
                RETURN(-ENOENT);
 
        if (lustre_handle_is_used(lh)) {
@@ -2497,14 +2549,11 @@ out_shrink:
 
        /*
         * Data-on-MDT optimization - read data along with OPEN and return it
-        * in reply. Do that only if we have both DOM and LAYOUT locks.
+        * in reply when possible.
         */
-       if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
-           info->mti_attr.ma_lmm != NULL &&
-           mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
+       if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
                rc = mdt_dom_read_on_open(info, info->mti_mdt,
                                          &lhc->mlh_reg_lh);
-       }
 
        return rc;
 }
@@ -5448,7 +5497,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                        m->mdt_skip_lfsck = 1;
        }
 
-       /* DoM files get IO lock at open by default */
+       /* DoM files get IO lock at open optionally by default */
        m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
        /* DoM files are read at open and data is packed in the reply */
        m->mdt_opts.mo_dom_read_open = 1;