Whamcloud - gitweb
LU-9341 lod: Add special O_APPEND striping
[fs/lustre-release.git] / lustre / lod / lod_object.c
index 27c5e8d..13a7e31 100644 (file)
@@ -1076,12 +1076,16 @@ static int lod_attr_get(const struct lu_env *env,
 }
 
 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
-                                         struct lov_desc *desc)
+                                         struct lov_desc *desc,
+                                         int append_stripes)
 {
        if (comp->llc_pattern != LOV_PATTERN_MDT) {
-               if (!comp->llc_stripe_count)
+               if (append_stripes) {
+                       comp->llc_stripe_count = append_stripes;
+               } else if (!comp->llc_stripe_count) {
                        comp->llc_stripe_count =
                                desc->ld_default_stripe_count;
+               }
        }
        if (comp->llc_stripe_size <= 0)
                comp->llc_stripe_size = desc->ld_default_stripe_size;
@@ -1419,7 +1423,7 @@ static int lod_attr_set(const struct lu_env *env,
                buf->lb_len = info->lti_ea_store_size;
                lmm = info->lti_ea_store;
                magic = le32_to_cpu(lmm->lmm_magic);
-               if (magic == LOV_MAGIC_COMP_V1) {
+               if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
                        struct lov_comp_md_v1 *lcm = buf->lb_buf;
                        struct lov_comp_md_entry_v1 *lcme =
                                                &lcm->lcm_entries[0];
@@ -1453,7 +1457,8 @@ static int lod_attr_set(const struct lu_env *env,
                buf->lb_buf = info->lti_ea_store;
                buf->lb_len = info->lti_ea_store_size;
                lcm = buf->lb_buf;
-               if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
+               if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
+                   le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
                        RETURN(-EINVAL);
 
                le32_add_cpu(&lcm->lcm_layout_gen, 1);
@@ -1543,6 +1548,13 @@ static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
                RETURN(rc = rc1 != 0 ? rc1 : rc);
        }
 
+       if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
+               struct lov_comp_md_v1 *lcm = buf->lb_buf;
+
+               if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
+                       lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
+       }
+
        if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
                RETURN(rc);
 
@@ -2032,7 +2044,7 @@ static int lod_prep_md_striped_create(const struct lu_env *env,
                                continue;
 
                        tgt_dt = tgt->ltd_tgt;
-                       rc = dt_statfs(env, tgt_dt, &info->lti_osfs, NULL);
+                       rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
                        if (rc) {
                                /* this OSP doesn't feel well */
                                rc = 0;
@@ -2796,7 +2808,7 @@ static int lod_declare_layout_add(const struct lu_env *env,
 
                lod_comp->llc_stripe_count = v1->lmm_stripe_count;
                lod_comp->llc_stripe_size = v1->lmm_stripe_size;
-               lod_adjust_stripe_info(lod_comp, desc);
+               lod_adjust_stripe_info(lod_comp, desc, 0);
 
                if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
                        v3 = (struct lov_user_md_v3 *) v1;
@@ -3228,6 +3240,7 @@ static int lod_declare_layout_merge(const struct lu_env *env,
        struct lov_comp_md_v1   *cur_lcm;
        struct lov_comp_md_v1   *merge_lcm;
        struct lov_comp_md_entry_v1     *lcme;
+       struct lov_mds_md_v1 *lmm;
        size_t size = 0;
        size_t offset;
        __u16 cur_entry_count;
@@ -3236,6 +3249,8 @@ static int lod_declare_layout_merge(const struct lu_env *env,
        __u16 mirror_id = 0;
        __u32 mirror_count;
        int     rc, i;
+       bool merge_has_dom;
+
        ENTRY;
 
        merge_lcm = mbuf->lb_buf;
@@ -3264,6 +3279,7 @@ static int lod_declare_layout_merge(const struct lu_env *env,
                rc = lod_layout_convert(info);
                break;
        case LOV_MAGIC_COMP_V1:
+       case LOV_MAGIC_SEL:
                rc = 0;
                break;
        default:
@@ -3322,6 +3338,13 @@ static int lod_declare_layout_merge(const struct lu_env *env,
        }
 
        mirror_id = mirror_id_of(id) + 1;
+
+       /* check if first entry in new layout is DOM */
+       lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
+                                       merge_lcm->lcm_entries[0].lcme_offset);
+       merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
+                       LOV_PATTERN_MDT;
+
        for (i = 0; i < merge_entry_count; i++) {
                struct lov_comp_md_entry_v1 *merge_lcme;
 
@@ -3330,6 +3353,8 @@ static int lod_declare_layout_merge(const struct lu_env *env,
 
                *lcme = *merge_lcme;
                lcme->lcme_offset = cpu_to_le32(offset);
+               if (merge_has_dom && i == 0)
+                       lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
 
                id = pflr_id(mirror_id, i + 1);
                lcme->lcme_id = cpu_to_le32(id);
@@ -3630,8 +3655,23 @@ static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
                                             pool_name);
                break;
        case LOV_USER_MAGIC_COMP_V1:
+       {
+               struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
+               struct lov_comp_md_entry_v1 *lcme;
+               int i, comp_cnt;
+
+               comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
+               for (i = 0; i < comp_cnt; i++) {
+                       lcme = &lcm->lcm_entries[i];
+                       if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
+                               lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
+                               break;
+                       }
+               }
+
                is_del = false;
                break;
+       }
        default:
                CERROR("Invalid magic %x\n", lum->lmm_magic);
                RETURN(-EINVAL);
@@ -3793,7 +3833,7 @@ static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
                /* if it's source stripe of migrating directory, don't create */
                if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
                      i >= lo->ldo_dir_migrate_offset)) {
-                       dt_write_lock(env, dto, MOR_TGT_CHILD);
+                       dt_write_lock(env, dto, DT_TGT_CHILD);
                        rc = lod_sub_create(env, dto, attr, NULL, dof, th);
                        if (rc != 0) {
                                dt_write_unlock(env, dto);
@@ -4279,6 +4319,9 @@ static int lod_layout_del_prep_layout(const struct lu_env *env,
 
        ENTRY;
 
+       LASSERT(lo->ldo_is_composite);
+       LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
+
        rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
        if (rc)
                RETURN(rc);
@@ -4388,9 +4431,7 @@ static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
        struct lu_attr *attr = &lod_env_info(env)->lti_attr;
        int rc;
 
-       LASSERT(lo->ldo_is_composite);
        LASSERT(lo->ldo_mirror_count == 1);
-       LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
 
        rc = lod_layout_del_prep_layout(env, lo, th);
        if (rc < 0)
@@ -4426,7 +4467,8 @@ out:
 
 static int lod_get_default_lov_striping(const struct lu_env *env,
                                        struct lod_object *lo,
-                                       struct lod_default_striping *lds);
+                                       struct lod_default_striping *lds,
+                                       struct dt_allocation_hint *ah);
 /**
  * Implementation of dt_object_operations::do_xattr_set.
  *
@@ -4482,7 +4524,8 @@ static int lod_xattr_set(const struct lu_env *env,
                bool is_del;
 
                /* get existing striping config */
-               rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
+               rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
+                                                 NULL);
                if (rc)
                        RETURN(rc);
 
@@ -4613,7 +4656,13 @@ static int lod_declare_xattr_del(const struct lu_env *env,
        if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
                RETURN(0);
 
-       /* set xattr to each stripes, if needed */
+       /* NB: don't delete stripe LMV, because when we do this, normally we
+        * will remove stripes, besides, if directory LMV is corrupt, this will
+        * prevent deleting its LMV and fixing it (via LFSCK).
+        */
+       if (!strcmp(name, XATTR_NAME_LMV))
+               RETURN(0);
+
        rc = lod_striping_load(env, lo);
        if (rc != 0)
                RETURN(rc);
@@ -4659,6 +4708,9 @@ static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
        if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
                RETURN(rc);
 
+       if (!strcmp(name, XATTR_NAME_LMV))
+               RETURN(0);
+
        if (lo->ldo_dir_stripe_count == 0)
                RETURN(0);
 
@@ -4752,7 +4804,8 @@ skip:
  */
 static int lod_get_default_lov_striping(const struct lu_env *env,
                                        struct lod_object *lo,
-                                       struct lod_default_striping *lds)
+                                       struct lod_default_striping *lds,
+                                       struct dt_allocation_hint *ah)
 {
        struct lod_thread_info *info = lod_env_info(env);
        struct lov_user_md_v1 *v1 = NULL;
@@ -4783,17 +4836,21 @@ static int lod_get_default_lov_striping(const struct lu_env *env,
                lustre_swab_lov_user_md_v3(v3);
                lustre_swab_lov_user_md_objects(v3->lmm_objects,
                                                v3->lmm_stripe_count);
-       } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
+       } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
+                  v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
                comp_v1 = (struct lov_comp_md_v1 *)v1;
                lustre_swab_lov_comp_md_v1(comp_v1);
        }
 
        if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
            v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
+           v1->lmm_magic != LOV_MAGIC_SEL &&
            v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
                RETURN(-ENOTSUPP);
 
-       if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
+       if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
+           v1->lmm_magic == LOV_MAGIC_SEL) &&
+            !(ah && ah->dah_append_stripes)) {
                comp_v1 = (struct lov_comp_md_v1 *)v1;
                comp_cnt = comp_v1->lcm_entry_count;
                if (comp_cnt == 0)
@@ -4844,19 +4901,24 @@ static int lod_get_default_lov_striping(const struct lu_env *env,
                        RETURN(-EINVAL);
                }
 
-               CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
-                      "stripe_offset=%d\n",
+               CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
                       PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
                       (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
-                      (int)v1->lmm_stripe_offset);
+                      (int)v1->lmm_stripe_offset,
+                      ah ? ah->dah_append_stripes : 0);
 
-               lod_comp->llc_stripe_count = v1->lmm_stripe_count;
+               if (ah && ah->dah_append_stripes)
+                       lod_comp->llc_stripe_count = ah->dah_append_stripes;
+               else
+                       lod_comp->llc_stripe_count = v1->lmm_stripe_count;
                lod_comp->llc_stripe_size = v1->lmm_stripe_size;
                lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
                lod_comp->llc_pattern = v1->lmm_pattern;
 
                pool = NULL;
-               if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
+               if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
+                       pool = ah->dah_append_pool;
+               } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
                        /* XXX: sanity check here */
                        v3 = (struct lov_user_md_v3 *) v1;
                        if (v3->lmm_pool_name[0] != '\0')
@@ -4938,7 +5000,7 @@ static int lod_get_default_striping(const struct lu_env *env,
 {
        int rc, rc1;
 
-       rc = lod_get_default_lov_striping(env, lo, lds);
+       rc = lod_get_default_lov_striping(env, lo, lds, NULL);
        rc1 = lod_get_default_lmv_striping(env, lo, lds);
        if (rc == 0 && rc1 < 0)
                rc = rc1;
@@ -5021,7 +5083,7 @@ static void lod_striping_from_default(struct lod_object *lo,
                        if (!lo->ldo_is_composite)
                                continue;
 
-                       lod_adjust_stripe_info(obj_comp, desc);
+                       lod_adjust_stripe_info(obj_comp, desc, 0);
                }
        } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
                if (lo->ldo_dir_stripe_count == 0)
@@ -5041,7 +5103,8 @@ static void lod_striping_from_default(struct lod_object *lo,
        }
 }
 
-static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
+static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
+                                        char *append_pool)
 {
        struct lod_layout_component *lod_comp;
 
@@ -5061,6 +5124,9 @@ static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
                          lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
                return true;
 
+       if (append_pool && append_pool[0])
+               return true;
+
        return false;
 }
 
@@ -5095,6 +5161,9 @@ static void lod_ah_init(const struct lu_env *env,
 
        LASSERT(child);
 
+       if (ah->dah_append_stripes == -1)
+               ah->dah_append_stripes = d->lod_desc.ld_tgt_count;
+
        if (likely(parent)) {
                nextp = dt_object_child(parent);
                lp = lod_dt_obj(parent);
@@ -5196,7 +5265,7 @@ static void lod_ah_init(const struct lu_env *env,
         * Try from the parent first.
         */
        if (likely(lp != NULL)) {
-               rc = lod_get_default_lov_striping(env, lp, lds);
+               rc = lod_get_default_lov_striping(env, lp, lds, ah);
                if (rc == 0)
                        lod_striping_from_default(lc, lds, child_mode);
        }
@@ -5224,8 +5293,10 @@ static void lod_ah_init(const struct lu_env *env,
         *  - parent has plain(v1/v3) default layout, and some attributes
         *    are not specified in the default layout;
         */
-       if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
-               rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
+       if (d->lod_md_root != NULL &&
+           lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
+               rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
+                                                 ah);
                if (rc)
                        goto out;
                if (lc->ldo_comp_cnt == 0) {
@@ -5256,7 +5327,7 @@ out:
         * fs default striping may not be explicitly set, or historically set
         * in config log, use them.
         */
-       if (lod_need_inherit_more(lc, false)) {
+       if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
                if (lc->ldo_comp_cnt == 0) {
                        rc = lod_alloc_comp_entries(lc, 0, 1);
                        if (rc)
@@ -5270,7 +5341,9 @@ out:
                LASSERT(!lc->ldo_is_composite);
                lod_comp = &lc->ldo_comp_entries[0];
                desc = &d->lod_desc;
-               lod_adjust_stripe_info(lod_comp, desc);
+               lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
+               if (ah->dah_append_pool && ah->dah_append_pool[0])
+                       lod_obj_set_pool(lc, 0, ah->dah_append_pool);
        }
 
        EXIT;
@@ -5558,12 +5631,10 @@ static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
                                struct lod_tgt_descs *ltd;
                                struct lod_tgt_desc *tgt = NULL;
                                bool found_mdt = false;
-                               int i;
 
                                lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
                                ltd = &lod->lod_mdt_descs;
-                               cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
-                                       tgt = LTD_TGT(ltd, i);
+                               ltd_foreach_tgt(ltd, tgt) {
                                        if (tgt->ltd_index ==
                                                lo->ldo_dir_stripe_offset) {
                                                found_mdt = true;
@@ -5982,7 +6053,7 @@ static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
 
                        if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
                            i == cfs_fail_val) {
-                               dt_write_lock(env, stripe, MOR_TGT_CHILD);
+                               dt_write_lock(env, stripe, DT_TGT_CHILD);
                                rc = lod_sub_ref_del(env, stripe, th);
                                dt_write_unlock(env, stripe);
                                if (rc != 0)
@@ -6327,7 +6398,7 @@ static bool lod_sel_osts_allowed(const struct lu_env *env,
                        break;
                }
 
-               rc = dt_statfs(env, ost->ltd_ost, sfs, &info);
+               rc = dt_statfs_info(env, ost->ltd_ost, sfs, &info);
                if (rc) {
                        CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
                               index, rc);
@@ -7786,16 +7857,8 @@ static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
        ENTRY;
 
        rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
-       if (rc != 0) {
-               /* Note: Sometimes, it will Return EAGAIN here, see
-                * ptrlpc_import_delay_req(), which might confuse
-                * lu_object_find_at() and make it wait there incorrectly.
-                * so we convert it to EIO here.*/
-               if (rc == -EAGAIN)
-                       rc = -EIO;
-
+       if (rc != 0)
                RETURN(rc);
-       }
 
        if (type == LU_SEQ_RANGE_MDT &&
            idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {