Whamcloud - gitweb
LU-12624 lod: alloc dir stripes by QoS
[fs/lustre-release.git] / lustre / lod / lod_lov.c
index c044bd1..e449db6 100644 (file)
@@ -77,30 +77,21 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
        if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
                struct lod_tgt_desc *tgt_desc, *tmp;
                struct list_head kill;
-               unsigned int idx;
 
                CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
                       ltd->ltd_death_row);
 
                INIT_LIST_HEAD(&kill);
 
-               cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
-                       tgt_desc = LTD_TGT(ltd, idx);
+               ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
                        LASSERT(tgt_desc);
-
                        if (!tgt_desc->ltd_reap)
                                continue;
 
                        list_add(&tgt_desc->ltd_kill, &kill);
-                       LTD_TGT(ltd, idx) = NULL;
-                       /*FIXME: only support ost pool for now */
-                       if (ltd == &lod->lod_ost_descs) {
-                               lod_ost_pool_remove(&lod->lod_pool_info, idx);
-                               if (tgt_desc->ltd_active)
-                                       lod->lod_desc.ld_active_tgt_count--;
-                       }
-                       ltd->ltd_tgtnr--;
-                       cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
+                       lod_tgt_pool_remove(&ltd->ltd_tgt_pool,
+                                           tgt_desc->ltd_index);
+                       ltd_del_tgt(ltd, tgt_desc);
                        ltd->ltd_death_row--;
                }
                mutex_unlock(&ltd->ltd_mutex);
@@ -108,17 +99,8 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
 
                list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
                        int rc;
+
                        list_del(&tgt_desc->ltd_kill);
-                       if (ltd == &lod->lod_ost_descs) {
-                               /* remove from QoS structures */
-                               rc = lqos_del_tgt(&lod->lod_qos, tgt_desc);
-                               if (rc)
-                                       CERROR("%s: qos_del_tgt(%s) failed:"
-                                              "rc = %d\n",
-                                              lod2obd(lod)->obd_name,
-                                             obd_uuid2str(&tgt_desc->ltd_uuid),
-                                              rc);
-                       }
                        rc = obd_disconnect(tgt_desc->ltd_exp);
                        if (rc)
                                CERROR("%s: failed to disconnect %s: rc = %d\n",
@@ -133,60 +115,6 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
 }
 
 /**
- * Expand size of target table.
- *
- * When the target table is full, we have to extend the table. To do so,
- * we allocate new memory with some reserve, move data from the old table
- * to the new one and release memory consumed by the old table.
- * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
- *
- * \param[in] ltd              target table
- * \param[in] newsize          new size of the table
- *
- * \retval                     0 on success
- * \retval                     -ENOMEM if reallocation failed
- */
-static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
-{
-       struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
-       int           rc = 0;
-       ENTRY;
-
-       /* grab write reference on the lod. Relocating the array requires
-        * exclusive access */
-
-       down_write(&ltd->ltd_rw_sem);
-       if (newsize <= ltd->ltd_tgts_size)
-               /* someone else has already resize the array */
-               GOTO(out, rc = 0);
-
-       /* allocate new bitmap */
-       new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
-       if (!new_bitmap)
-               GOTO(out, rc = -ENOMEM);
-
-       if (ltd->ltd_tgts_size > 0) {
-               /* the bitmap already exists, we need
-                * to copy data from old one */
-               cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
-               old_bitmap = ltd->ltd_tgt_bitmap;
-       }
-
-       ltd->ltd_tgts_size  = newsize;
-       ltd->ltd_tgt_bitmap = new_bitmap;
-
-       if (old_bitmap)
-               CFS_FREE_BITMAP(old_bitmap);
-
-       CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
-
-       EXIT;
-out:
-       up_write(&ltd->ltd_rw_sem);
-       return rc;
-}
-
-/**
  * Connect LOD to a new OSP and add it to the target table.
  *
  * Connect to the OSP device passed, initialize all the internal
@@ -219,7 +147,6 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
        struct lustre_cfg       *lcfg;
        struct obd_uuid         obd_uuid;
        bool                    for_ost;
-       bool lock = false;
        bool connected = false;
        ENTRY;
 
@@ -317,73 +244,27 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
        tgt_desc->ltd_index  = index;
        tgt_desc->ltd_active = active;
 
-       lod_getref(ltd);
-       if (index >= ltd->ltd_tgts_size) {
-               /* we have to increase the size of the lod_osts array */
-               __u32  newsize;
-
-               newsize = max(ltd->ltd_tgts_size, (__u32)2);
-               while (newsize < index + 1)
-                       newsize = newsize << 1;
-
-               /* lod_bitmap_resize() needs lod_rw_sem
-                * which we hold with th reference */
-               lod_putref(lod, ltd);
-
-               rc = ltd_bitmap_resize(ltd, newsize);
-               if (rc)
-                       GOTO(out_desc, rc);
-
-               lod_getref(ltd);
-       }
-
+       down_write(&ltd->ltd_rw_sem);
        mutex_lock(&ltd->ltd_mutex);
-       lock = true;
-       if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
-               CERROR("%s: device %d is registered already\n", obd->obd_name,
-                      index);
-               GOTO(out_mutex, rc = -EEXIST);
-       }
-
-       if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
-               OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
-               if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
-                       CERROR("can't allocate index to add %s\n",
-                              obd->obd_name);
-                       GOTO(out_mutex, rc = -ENOMEM);
-               }
-       }
-
-       if (for_ost) {
-               /* pool and qos are not supported for MDS stack yet */
-               rc = lod_ost_pool_add(&lod->lod_pool_info, index,
-                                     lod->lod_osts_size);
-               if (rc) {
-                       CERROR("%s: can't set up pool, failed with %d\n",
-                              obd->obd_name, rc);
-                       GOTO(out_mutex, rc);
-               }
+       rc = ltd_add_tgt(ltd, tgt_desc);
+       if (rc)
+               GOTO(out_mutex, rc);
 
-               rc = lqos_add_tgt(&lod->lod_qos, tgt_desc);
-               if (rc) {
-                       CERROR("%s: qos_add_tgt failed with %d\n",
-                               obd->obd_name, rc);
-                       GOTO(out_pool, rc);
-               }
+       rc = lu_qos_add_tgt(&ltd->ltd_qos, tgt_desc);
+       if (rc)
+               GOTO(out_del_tgt, rc);
 
-               /* The new OST is now a full citizen */
-               if (index >= lod->lod_desc.ld_tgt_count)
-                       lod->lod_desc.ld_tgt_count = index + 1;
-               if (active)
-                       lod->lod_desc.ld_active_tgt_count++;
+       rc = lod_tgt_pool_add(&ltd->ltd_tgt_pool, index,
+                             ltd->ltd_lov_desc.ld_tgt_count);
+       if (rc) {
+               CERROR("%s: can't set up pool, failed with %d\n",
+                      obd->obd_name, rc);
+               GOTO(out_del_tgt, rc);
        }
 
-       LTD_TGT(ltd, index) = tgt_desc;
-       cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
-       ltd->ltd_tgtnr++;
        mutex_unlock(&ltd->ltd_mutex);
-       lod_putref(lod, ltd);
-       lock = false;
+       up_write(&ltd->ltd_rw_sem);
+
        if (lod->lod_recovery_completed)
                lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
 
@@ -407,26 +288,20 @@ out_fini_llog:
        lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
                          tgt_desc->ltd_recovery_thread);
 out_ltd:
-       lod_getref(ltd);
+       down_write(&ltd->ltd_rw_sem);
        mutex_lock(&ltd->ltd_mutex);
-       lock = true;
        if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
                struct ptlrpc_thread *thread;
 
                thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
                OBD_FREE_PTR(thread);
        }
-       ltd->ltd_tgtnr--;
-       cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
-       LTD_TGT(ltd, index) = NULL;
-out_pool:
-       lod_ost_pool_remove(&lod->lod_pool_info, index);
+       lod_tgt_pool_remove(&ltd->ltd_tgt_pool, index);
+out_del_tgt:
+       ltd_del_tgt(ltd, tgt_desc);
 out_mutex:
-       if (lock) {
-               mutex_unlock(&ltd->ltd_mutex);
-               lod_putref(lod, ltd);
-       }
-out_desc:
+       mutex_unlock(&ltd->ltd_mutex);
+       up_write(&ltd->ltd_rw_sem);
        OBD_FREE_PTR(tgt_desc);
 out_cleanup:
        /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
@@ -453,27 +328,19 @@ out_cleanup:
  * \param[in] env              execution environment for this thread
  * \param[in] lod              LOD device the target table belongs to
  * \param[in] ltd              target table
- * \param[in] idx              index of the target
- * \param[in] for_ost          type of the target: 0 - MDT, 1 - OST
+ * \param[in] tgt              target
  */
 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
-                            struct lod_tgt_descs *ltd, unsigned idx,
-                            bool for_ost)
+                            struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
 {
-       LASSERT(LTD_TGT(ltd, idx));
+       lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
+                        !ltd->ltd_is_mdt);
 
-       lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
-                        idx, for_ost);
+       if (ltd->ltd_is_mdt && tgt->ltd_recovery_thread)
+               OBD_FREE_PTR(tgt->ltd_recovery_thread);
 
-       if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
-               struct ptlrpc_thread *thread;
-
-               thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
-               OBD_FREE_PTR(thread);
-       }
-
-       if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
-               LTD_TGT(ltd, idx)->ltd_reap = 1;
+       if (!tgt->ltd_reap) {
+               tgt->ltd_reap = 1;
                ltd->ltd_death_row++;
        }
 }
@@ -486,29 +353,26 @@ static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
  * \param[in] env              execution environment for this thread
  * \param[in] lod              LOD device the target table belongs to
  * \param[in] ltd              target table
- * \param[in] for_ost          type of the target: MDT or OST
  *
  * \retval                     0 always
  */
 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
-                struct lod_tgt_descs *ltd, bool for_ost)
+                struct lod_tgt_descs *ltd)
 {
-       unsigned int idx;
+       struct lu_tgt_desc *tgt;
 
        if (ltd->ltd_tgts_size <= 0)
                return 0;
+
        lod_getref(ltd);
        mutex_lock(&ltd->ltd_mutex);
-       cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
-               __lod_del_device(env, lod, ltd, idx, for_ost);
+       ltd_foreach_tgt(ltd, tgt)
+               __lod_del_device(env, lod, ltd, tgt);
        mutex_unlock(&ltd->ltd_mutex);
        lod_putref(lod, ltd);
-       CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
-       for (idx = 0; idx < TGT_PTRS; idx++) {
-               if (ltd->ltd_tgt_idx[idx])
-                       OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
-       }
-       ltd->ltd_tgts_size = 0;
+
+       lu_tgt_descs_fini(ltd);
+
        return 0;
 }
 
@@ -524,18 +388,19 @@ int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
  * \param[in] osp              name of OSP device to be removed
  * \param[in] idx              index of the target
  * \param[in] gen              generation number, not used currently
- * \param[in] for_ost          type of the target: 0 - MDT, 1 - OST
  *
  * \retval                     0 if the device was scheduled for removal
  * \retval                     -EINVAL if no device was found
  */
 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
-                  struct lod_tgt_descs *ltd, char *osp, unsigned idx,
-                  unsigned gen, bool for_ost)
+                  struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
+                  unsigned int gen)
 {
        struct obd_device *obd;
-       int                rc = 0;
-       struct obd_uuid    uuid;
+       struct lu_tgt_desc *tgt;
+       struct obd_uuid uuid;
+       int rc = 0;
+
        ENTRY;
 
        CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
@@ -559,22 +424,21 @@ int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 
        lod_getref(ltd);
        mutex_lock(&ltd->ltd_mutex);
+       tgt = LTD_TGT(ltd, idx);
        /* check that the index is allocated in the bitmap */
-       if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
-           !LTD_TGT(ltd, idx)) {
+       if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || !tgt) {
                CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
                GOTO(out, rc = -EINVAL);
        }
 
        /* check that the UUID matches */
-       if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
+       if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
                CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
-                      obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
-                      idx, osp);
+                      obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
                GOTO(out, rc = -EINVAL);
        }
 
-       __lod_del_device(env, lod, ltd, idx, for_ost);
+       __lod_del_device(env, lod, ltd, tgt);
        EXIT;
 out:
        mutex_unlock(&ltd->ltd_mutex);
@@ -1123,7 +987,7 @@ static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
                return -EINVAL;
        }
 
-       if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
+       if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
                CERROR("%s: invalid lod device, for idx: %d\n",
                       lod2obd(md)->obd_name , idx);
                return -EINVAL;
@@ -1201,7 +1065,7 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
                        GOTO(out, rc);
                }
 
-               nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
+               nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
                lod_putref(md, &md->lod_ost_descs);
 
                /* In the function below, .hs_keycmp resolves to
@@ -1254,14 +1118,14 @@ out:
 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                       const struct lu_buf *buf)
 {
-       struct lov_mds_md_v1    *lmm;
-       struct lov_comp_md_v1   *comp_v1 = NULL;
-       struct lov_foreign_md   *foreign = NULL;
-       struct lov_ost_data_v1  *objs;
-       __u32   magic, pattern;
-       int     i, j, rc = 0;
-       __u16   comp_cnt;
-       __u16   mirror_cnt = 0;
+       struct lov_mds_md_v1 *lmm;
+       struct lov_comp_md_v1 *comp_v1 = NULL;
+       struct lov_foreign_md *foreign = NULL;
+       struct lov_ost_data_v1 *objs;
+       __u32 magic, pattern;
+       __u16 mirror_cnt = 0;
+       __u16 comp_cnt;
+       int i, rc;
        ENTRY;
 
        LASSERT(buf);
@@ -1322,15 +1186,14 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                GOTO(out, rc);
 
        for (i = 0; i < comp_cnt; i++) {
-               struct lod_layout_component     *lod_comp;
-               struct lu_extent        *ext;
-               __u32   offs;
+               struct lod_layout_component *lod_comp;
+               struct lu_extent *ext;
+               __u32 offs;
 
                lod_comp = &lo->ldo_comp_entries[i];
                if (lo->ldo_is_composite) {
                        offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
                        lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
-                       magic = le32_to_cpu(lmm->lmm_magic);
 
                        ext = &comp_v1->lcm_entries[i].lcme_extent;
                        lod_comp->llc_extent.e_start =
@@ -1346,16 +1209,16 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                        if (lod_comp->llc_id == LCME_ID_INVAL)
                                GOTO(out, rc = -EINVAL);
 
-                       if (comp_v1->lcm_entries[i].lcme_flags &
-                           cpu_to_le32(LCME_FL_EXTENSION) &&
-                           magic != LOV_MAGIC_SEL) {
+                       if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
+                           comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
                                struct lod_device *d =
                                        lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 
-                               CDEBUG(D_WARNING, "%s: not SEL magic on SEL "
-                                      "file "DFID": %x\n",
-                                      lod2obd(d)->obd_name,
-                                      PFID(lod_object_fid(lo)), magic);
+                               CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
+                                     lod2obd(d)->obd_name,
+                                     lod_comp->llc_flags, lod_comp->llc_id, i,
+                                     PFID(lod_object_fid(lo)),
+                                     le32_to_cpu(comp_v1->lcm_magic));
                        }
                } else {
                        lod_comp_set_init(lod_comp);
@@ -1370,8 +1233,9 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
                lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 
-               if (magic == LOV_MAGIC_V3) {
+               if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
                        struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
+
                        lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
                        objs = &v3->lmm_objects[0];
                } else {
@@ -1387,6 +1251,8 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                        __u16 stripe_count;
 
                        if (objs[0].l_ost_idx != (__u32)-1UL) {
+                               int j;
+
                                stripe_count = lod_comp_entry_stripe_count(
                                                        lo, lod_comp, false);
                                if (stripe_count == 0 &&
@@ -1672,9 +1538,9 @@ static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
        if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
            lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
                /* if offset is not within valid range [0, osts_size) */
-               if (stripe_offset >= d->lod_osts_size) {
+               if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
                        CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
-                              stripe_offset, d->lod_osts_size);
+                              stripe_offset, d->lod_ost_descs.ltd_tgts_size);
                        GOTO(out, rc = -EINVAL);
                }
 
@@ -1752,9 +1618,10 @@ struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
                                   le16_to_cpu(comp->lcm_entry_count) - 1); \
             entry++)
 
-int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
+int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
+                        struct lov_comp_md_entry_v1 *dom_ent)
 {
-       struct lov_comp_md_entry_v1 *ent, *dom_ent;
+       struct lov_comp_md_entry_v1 *ent;
        __u16 entries;
        __u32 dom_off, dom_size, comp_size;
        void *blob_src, *blob_dst;
@@ -1766,7 +1633,6 @@ int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
                return -EFBIG;
 
        comp_size = le32_to_cpu(comp_v1->lcm_size);
-       dom_ent = &comp_v1->lcm_entries[0];
        dom_off = le32_to_cpu(dom_ent->lcme_offset);
        dom_size = le32_to_cpu(dom_ent->lcme_size);
 
@@ -1796,16 +1662,16 @@ int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
        return -ERESTART;
 }
 
-int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
+int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1,
+                      struct lov_comp_md_entry_v1 *dom_ent)
 {
-       struct lov_comp_md_entry_v1 *ent, *dom_ent;
+       struct lov_comp_md_entry_v1 *ent;
        struct lu_extent *dom_ext, *ext;
        struct lov_user_md_v1 *lum;
        __u32 stripe_size;
        __u16 mid, dom_mid;
        int rc = 0;
 
-       dom_ent = &comp_v1->lcm_entries[0];
        dom_ext = &dom_ent->lcme_extent;
        dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
        stripe_size = d->lod_dom_max_stripesize;
@@ -1842,7 +1708,7 @@ int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
        if (stripe_size == 0) {
                /* DoM component size is zero due to server setting,
                 * remove it from the layout */
-               rc = lod_erase_dom_stripe(comp_v1);
+               rc = lod_erase_dom_stripe(comp_v1, dom_ent);
        } else {
                /* Update DoM extent end finally */
                dom_ext->e_end = cpu_to_le64(stripe_size);
@@ -1866,7 +1732,7 @@ int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
                        const struct lu_buf *buf, bool is_from_disk)
 {
-       struct lov_desc *desc = &d->lod_desc;
+       struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
        struct lov_user_md_v1   *lum;
        struct lov_comp_md_v1   *comp_v1;
        struct lov_comp_md_entry_v1     *ent;
@@ -2015,7 +1881,7 @@ recheck:
                lum = tmp.lb_buf;
                if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
                    LOV_PATTERN_MDT) {
-                       /* DoM component can be only the first stripe */
+                       /* DoM component must be the first in a mirror */
                        if (le64_to_cpu(ext->e_start) > 0) {
                                CDEBUG(D_LAYOUT, "invalid DoM component "
                                       "with %llu extent start\n",
@@ -2038,7 +1904,7 @@ recheck:
                                       "%u is bigger than MDT limit %u, check "
                                       "dom_max_stripesize parameter\n",
                                       stripe_size, d->lod_dom_max_stripesize);
-                               rc = lod_fix_dom_stripe(d, comp_v1);
+                               rc = lod_fix_dom_stripe(d, comp_v1, ent);
                                if (rc == -ERESTART) {
                                        /* DoM entry was removed, re-check
                                         * new layout from start */
@@ -2135,6 +2001,14 @@ void lod_fix_desc_pattern(__u32 *val)
        }
 }
 
+void lod_fix_lmv_desc_pattern(__u32 *val)
+{
+       if ((*val) && !lmv_is_known_hash_type(*val)) {
+               LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
+               *val = 0;
+       }
+}
+
 void lod_fix_desc_qos_maxage(__u32 *val)
 {
        /* fix qos_maxage */
@@ -2155,6 +2029,14 @@ void lod_fix_desc(struct lov_desc *desc)
        lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 }
 
+static void lod_fix_lmv_desc(struct lmv_desc *desc)
+{
+       desc->ld_active_tgt_count = 0;
+       lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
+       lod_fix_lmv_desc_pattern(&desc->ld_pattern);
+       lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
+}
+
 /**
  * Initialize the structures used to store pools and default striping.
  *
@@ -2203,21 +2085,12 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
        lod_fix_desc(desc);
 
        desc->ld_active_tgt_count = 0;
-       lod->lod_desc = *desc;
+       lod->lod_ost_descs.ltd_lov_desc = *desc;
 
-       lod->lod_sp_me = LUSTRE_SP_CLI;
-
-       /* Set up allocation policy (QoS and RR) */
-       INIT_LIST_HEAD(&lod->lod_qos.lq_svr_list);
-       init_rwsem(&lod->lod_qos.lq_rw_sem);
-       lod->lod_qos.lq_dirty = 1;
-       lod->lod_qos.lq_reset = 1;
-       /* Default priority is toward free space balance */
-       lod->lod_qos.lq_prio_free = 232;
-       /* Default threshold for rr (roughly 17%) */
-       lod->lod_qos.lq_threshold_rr = 43;
+       /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
+       lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
 
-       lu_qos_rr_init(&lod->lod_qos.lq_rr);
+       lod->lod_sp_me = LUSTRE_SP_CLI;
 
        /* Set up OST pool environment */
        lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
@@ -2232,17 +2105,30 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 
        INIT_LIST_HEAD(&lod->lod_pool_list);
        lod->lod_pool_count = 0;
-       rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
+       rc = lod_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
        if (rc)
                GOTO(out_hash, rc);
-       rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
+
+       rc = lod_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
+       if (rc)
+               GOTO(out_mdt_pool, rc);
+
+       rc = lod_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
+       if (rc)
+               GOTO(out_mdt_rr_pool, rc);
+
+       rc = lod_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
        if (rc)
-               GOTO(out_pool_info, rc);
+               GOTO(out_ost_pool, rc);
 
        RETURN(0);
 
-out_pool_info:
-       lod_ost_pool_free(&lod->lod_pool_info);
+out_ost_pool:
+       lod_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
+out_mdt_rr_pool:
+       lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
+out_mdt_pool:
+       lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
 out_hash:
        cfs_hash_putref(lod->lod_pools_hash_body);
 
@@ -2272,8 +2158,10 @@ int lod_pools_fini(struct lod_device *lod)
        }
 
        cfs_hash_putref(lod->lod_pools_hash_body);
-       lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
-       lod_ost_pool_free(&lod->lod_pool_info);
+       lod_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
+       lod_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
+       lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
+       lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
 
        RETURN(0);
 }