X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flod%2Flod_object.c;h=fc9efb5362e09a69acced049d29ada07c329a252;hp=27c5e8db108fe6be92dc2de2ad8cd53111b67d73;hb=e7d91d9b8cf291d638857b6bd904469b46d1273a;hpb=76ca884398cae59e455caf3ae2ab1609c5fb1eea diff --git a/lustre/lod/lod_object.c b/lustre/lod/lod_object.c index 27c5e8d..fc9efb5 100644 --- a/lustre/lod/lod_object.c +++ b/lustre/lod/lod_object.c @@ -367,6 +367,50 @@ static struct dt_index_operations lod_index_ops = { }; /** + * Implementation of dt_index_operations::dio_lookup + * + * Used with striped directories. + * + * \see dt_index_operations::dio_lookup() in the API description for details. + */ +static int lod_striped_lookup(const struct lu_env *env, struct dt_object *dt, + struct dt_rec *rec, const struct dt_key *key) +{ + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object *next; + const char *name = (const char *)key; + + LASSERT(lo->ldo_dir_stripe_count > 0); + + if (strcmp(name, dot) == 0) { + struct lu_fid *fid = (struct lu_fid *)rec; + + *fid = *lod_object_fid(lo); + return 1; + } + + if (strcmp(name, dotdot) == 0) { + next = dt_object_child(dt); + } else { + int index; + + index = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type, + lo->ldo_dir_stripe_count, + lo->ldo_dir_migrate_hash, + lo->ldo_dir_migrate_offset, + name, strlen(name), true); + if (index < 0) + return index; + + next = lo->ldo_stripe[index]; + if (!next || !dt_object_exists(next)) + return -ENODEV; + } + + return next->do_index_ops->dio_lookup(env, next, rec, key); +} + +/** * Implementation of dt_it_ops::init. * * Used with striped objects. Internally just initializes the iterator @@ -744,7 +788,7 @@ static int lod_striped_it_load(const struct lu_env *env, } static struct dt_index_operations lod_striped_index_ops = { - .dio_lookup = lod_lookup, + .dio_lookup = lod_striped_lookup, .dio_declare_insert = lod_declare_insert, .dio_insert = lod_insert, .dio_declare_delete = lod_declare_delete, @@ -871,8 +915,8 @@ int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo, goto next; } - len = snprintf(name, sizeof(name), - DFID":", PFID(&ent->lde_fid)); + len = scnprintf(name, sizeof(name), + DFID":", PFID(&ent->lde_fid)); /* The ent->lde_name is composed of ${FID}:${index} */ if (ent->lde_namelen < len + 1 || memcmp(ent->lde_name, name, len) != 0) { @@ -1075,16 +1119,51 @@ static int lod_attr_get(const struct lu_env *env, return dt_attr_get(env, dt_object_child(dt), attr); } +void lod_adjust_stripe_size(struct lod_layout_component *comp, + __u32 def_stripe_size) +{ + __u64 comp_end = comp->llc_extent.e_end; + + /* Choose stripe size if not set. Note that default stripe size can't + * be used as is, because it must be multiplier of given component end. + * - first check if default stripe size can be used + * - if not than select the lowest set bit from component end and use + * that value as stripe size + */ + if (!comp->llc_stripe_size) { + if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size)) + comp->llc_stripe_size = def_stripe_size; + else + comp->llc_stripe_size = comp_end & ~(comp_end - 1); + } else { + /* check stripe size is multiplier of comp_end */ + if (comp_end != LUSTRE_EOF && + comp_end % comp->llc_stripe_size) { + /* fix that even for defined stripe size but warn + * about the problem, that must not happen + */ + CWARN("Component end %llu is not aligned by the stripe size %u\n", + comp_end, comp->llc_stripe_size); + dump_stack(); + comp->llc_stripe_size = comp_end & ~(comp_end - 1); + } + } +} + 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; + + lod_adjust_stripe_size(comp, desc->ld_default_stripe_size); } int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo, @@ -1419,7 +1498,7 @@ 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 +1532,8 @@ 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); @@ -1500,7 +1580,7 @@ static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt, RETURN(rc = rc > 0 ? -EINVAL : rc); if (buf->lb_buf == NULL || buf->lb_len == 0) { - CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key)); + BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key)); /* lti_buf is large enough for *lmv1 or a short * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV @@ -1527,10 +1607,10 @@ static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt, if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1) rc = lmv_mds_md_size( le32_to_cpu(lmv1->lmv_stripe_count), - LMV_MAGIC_V1); + le32_to_cpu(lmv1->lmv_magic)); } else { - lfm = buf->lb_buf; - if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) + lmv1 = buf->lb_buf; + if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1) RETURN(rc); if (rc != sizeof(*lmv1)) @@ -1543,6 +1623,13 @@ 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); @@ -1559,7 +1646,7 @@ static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt, if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) { struct lov_user_md *lum = buf->lb_buf; - struct lov_desc *desc = &dev->lod_desc; + struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc; if (buf->lb_buf == NULL) { rc = sizeof(*lum); @@ -1669,7 +1756,8 @@ static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt, lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC); lmm1->lmv_stripe_count = cpu_to_le32(stripe_count); lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type); - if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) { + lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version); + if (lod_is_layout_changing(lo)) { lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash); lmm1->lmv_migrate_offset = cpu_to_le32(lo->ldo_dir_migrate_offset); @@ -1726,15 +1814,11 @@ int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo, RETURN(0); } - if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1) + if (!lmv_is_sane(lmv1)) RETURN(-EINVAL); - if (le32_to_cpu(lmv1->lmv_stripe_count) < 1) - RETURN(0); - LASSERT(lo->ldo_stripe == NULL); - OBD_ALLOC(stripe, sizeof(stripe[0]) * - (le32_to_cpu(lmv1->lmv_stripe_count))); + OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count)); if (stripe == NULL) RETURN(-ENOMEM); @@ -1777,6 +1861,10 @@ out: lo->ldo_stripe = stripe; lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count); lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count); + lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version); + lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset); + lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash); + lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type); if (rc != 0) lod_striping_free_nolock(env, lo); @@ -1847,31 +1935,74 @@ static int lod_dir_declare_create_stripes(const struct lu_env *env, if (!dto) continue; - rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th); - if (rc != 0) - GOTO(out, rc); + /* directory split skip create for existing stripes */ + if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) { + rc = lod_sub_declare_create(env, dto, attr, NULL, dof, + th); + if (rc != 0) + GOTO(out, rc); - if (!dt_try_as_dir(env, dto)) - GOTO(out, rc = -EINVAL); + if (!dt_try_as_dir(env, dto)) + GOTO(out, rc = -EINVAL); - rc = lod_sub_declare_ref_add(env, dto, th); - if (rc != 0) - GOTO(out, rc); + rc = lod_sub_declare_ref_add(env, dto, th); + if (rc != 0) + GOTO(out, rc); - rec->rec_fid = lu_object_fid(&dto->do_lu); - rc = lod_sub_declare_insert(env, dto, - (const struct dt_rec *)rec, - (const struct dt_key *)dot, th); - if (rc != 0) - GOTO(out, rc); + rec->rec_fid = lu_object_fid(&dto->do_lu); + rc = lod_sub_declare_insert(env, dto, + (const struct dt_rec *)rec, + (const struct dt_key *)dot, + th); + if (rc != 0) + GOTO(out, rc); - /* master stripe FID will be put to .. */ - rec->rec_fid = lu_object_fid(&dt->do_lu); - rc = lod_sub_declare_insert(env, dto, - (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th); - if (rc != 0) - GOTO(out, rc); + /* master stripe FID will be put to .. */ + rec->rec_fid = lu_object_fid(&dt->do_lu); + rc = lod_sub_declare_insert(env, dto, + (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, + th); + if (rc != 0) + GOTO(out, rc); + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) && + cfs_fail_val == i) + snprintf(stripe_name, sizeof(info->lti_key), + DFID":%u", + PFID(lu_object_fid(&dto->do_lu)), + i + 1); + else + snprintf(stripe_name, sizeof(info->lti_key), + DFID":%u", + PFID(lu_object_fid(&dto->do_lu)), i); + + sname = lod_name_get(env, stripe_name, + strlen(stripe_name)); + rc = linkea_links_new(&ldata, &info->lti_linkea_buf, + sname, lu_object_fid(&dt->do_lu)); + if (rc != 0) + GOTO(out, rc); + + linkea_buf.lb_buf = ldata.ld_buf->lb_buf; + linkea_buf.lb_len = ldata.ld_leh->leh_len; + rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf, + XATTR_NAME_LINK, 0, th); + if (rc != 0) + GOTO(out, rc); + + rec->rec_fid = lu_object_fid(&dto->do_lu); + rc = lod_sub_declare_insert(env, dt_object_child(dt), + (const struct dt_rec *)rec, + (const struct dt_key *)stripe_name, th); + if (rc != 0) + GOTO(out, rc); + + rc = lod_sub_declare_ref_add(env, dt_object_child(dt), + th); + if (rc != 0) + GOTO(out, rc); + } if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) || cfs_fail_val != i) { @@ -1887,39 +2018,6 @@ static int lod_dir_declare_create_stripes(const struct lu_env *env, if (rc != 0) GOTO(out, rc); } - - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) && - cfs_fail_val == i) - snprintf(stripe_name, sizeof(info->lti_key), DFID":%u", - PFID(lu_object_fid(&dto->do_lu)), i + 1); - else - snprintf(stripe_name, sizeof(info->lti_key), DFID":%u", - PFID(lu_object_fid(&dto->do_lu)), i); - - sname = lod_name_get(env, stripe_name, strlen(stripe_name)); - rc = linkea_links_new(&ldata, &info->lti_linkea_buf, - sname, lu_object_fid(&dt->do_lu)); - if (rc != 0) - GOTO(out, rc); - - linkea_buf.lb_buf = ldata.ld_buf->lb_buf; - linkea_buf.lb_len = ldata.ld_leh->leh_len; - rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf, - XATTR_NAME_LINK, 0, th); - if (rc != 0) - GOTO(out, rc); - - rec->rec_fid = lu_object_fid(&dto->do_lu); - rc = lod_sub_declare_insert(env, dt_object_child(dt), - (const struct dt_rec *)rec, - (const struct dt_key *)stripe_name, - th); - if (rc != 0) - GOTO(out, rc); - - rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th); - if (rc != 0) - GOTO(out, rc); } rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), @@ -1933,74 +2031,72 @@ out: RETURN(rc); } -static int lod_prep_md_striped_create(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - const struct lmv_user_md_v1 *lum, - struct dt_object_format *dof, - struct thandle *th) +/** + * Allocate a striping on a predefined set of MDTs. + * + * Allocates new striping using the MDT index range provided by the data from + * the lum_obejcts contained in the lmv_user_md passed to this method if + * \a is_specific is true; or allocates new layout starting from MDT index in + * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and + * varies depending on MDT status. The number of stripes needed and stripe + * offset are taken from the object. If that number cannot be met, then the + * function returns an error and then it's the caller's responsibility to + * release the stripes allocated. All the internal structures are protected, + * but no concurrent allocation is allowed on the same objects. + * + * \param[in] env execution environment for this thread + * \param[in] lo LOD object + * \param[out] stripes striping created + * \param[out] mdt_indices MDT indices of striping created + * \param[in] is_specific true if the MDTs are provided by lum; false if + * only the starting MDT index is provided + * + * \retval positive stripes allocated, including the first stripe allocated + * outside + * \retval negative errno on failure + */ +static int lod_mdt_alloc_specific(const struct lu_env *env, + struct lod_object *lo, + struct dt_object **stripes, + __u32 *mdt_indices, bool is_specific) { - struct lod_thread_info *info = lod_env_info(env); - struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev); - struct lod_tgt_descs *ltd = &lod->lod_mdt_descs; - struct lod_object *lo = lod_dt_obj(dt); - struct dt_object **stripe; - __u32 stripe_count; - int *idx_array; - __u32 master_index; - int rc = 0; - __u32 i; - __u32 j; - bool is_specific = false; - ENTRY; - - /* The lum has been verifed in lod_verify_md_striping */ - LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC || - le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC); - - stripe_count = lo->ldo_dir_stripe_count; - - OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count); - if (idx_array == NULL) - RETURN(-ENOMEM); - - OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count); - if (stripe == NULL) - GOTO(out_free, rc = -ENOMEM); + struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); + struct lu_tgt_descs *ltd = &lod->lod_mdt_descs; + struct lu_tgt_desc *tgt = NULL; + struct lu_object_conf conf = { .loc_flags = LOC_F_NEW }; + struct dt_device *tgt_dt = NULL; + struct lu_fid fid = { 0 }; + struct dt_object *dto; + u32 master_index; + u32 stripe_count = lo->ldo_dir_stripe_count; + int stripe_idx = 1; + int j; + int idx; + int rc; - /* Start index must be the master MDT */ master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id; - idx_array[0] = master_index; - if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) { - is_specific = true; - for (i = 1; i < stripe_count; i++) - idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds); - } - - for (i = 0; i < stripe_count; i++) { - struct lod_tgt_desc *tgt = NULL; - struct dt_object *dto; - struct lu_fid fid = { 0 }; - int idx; - struct lu_object_conf conf = { 0 }; - struct dt_device *tgt_dt = NULL; + if (!is_specific && stripe_count > 1) + /* Set the start index for the 2nd stripe allocation */ + mdt_indices[1] = (mdt_indices[0] + 1) % + (lod->lod_remote_mdt_count + 1); + for (; stripe_idx < stripe_count; stripe_idx++) { /* Try to find next avaible target */ - idx = idx_array[i]; + idx = mdt_indices[stripe_idx]; for (j = 0; j < lod->lod_remote_mdt_count; j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) { bool already_allocated = false; __u32 k; CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n", - idx, lod->lod_remote_mdt_count + 1, i); + idx, lod->lod_remote_mdt_count + 1, stripe_idx); if (likely(!is_specific && !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) { /* check whether the idx already exists * in current allocated array */ - for (k = 0; k < i; k++) { - if (idx_array[k] == idx) { + for (k = 0; k < stripe_idx; k++) { + if (mdt_indices[k] == idx) { already_allocated = true; break; } @@ -2012,38 +2108,33 @@ static int lod_prep_md_striped_create(const struct lu_env *env, /* Sigh, this index is not in the bitmap, let's check * next available target */ - if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) && + if (!test_bit(idx, ltd->ltd_tgt_bitmap) && idx != master_index) continue; if (idx == master_index) { /* Allocate the FID locally */ - rc = obd_fid_alloc(env, lod->lod_child_exp, - &fid, NULL); - if (rc < 0) - GOTO(out_put, rc); tgt_dt = lod->lod_child; + rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, + NULL); + if (rc < 0) + continue; break; } /* check the status of the OSP */ tgt = LTD_TGT(ltd, idx); - if (tgt == NULL) + if (!tgt) continue; tgt_dt = tgt->ltd_tgt; - rc = dt_statfs(env, tgt_dt, &info->lti_osfs, NULL); - if (rc) { + if (!tgt->ltd_active) /* this OSP doesn't feel well */ - rc = 0; continue; - } - rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL); - if (rc < 0) { - rc = 0; + rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL); + if (rc < 0) continue; - } break; } @@ -2051,15 +2142,16 @@ static int lod_prep_md_striped_create(const struct lu_env *env, /* Can not allocate more stripes */ if (j == lod->lod_remote_mdt_count) { CDEBUG(D_INFO, "%s: require stripes %u only get %d\n", - lod2obd(lod)->obd_name, stripe_count, i); + lod2obd(lod)->obd_name, stripe_count, + stripe_idx); break; } CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n", - idx, i, PFID(&fid)); - idx_array[i] = idx; + idx, stripe_idx, PFID(&fid)); + mdt_indices[stripe_idx] = idx; /* Set the start index for next stripe allocation */ - if (!is_specific && i < stripe_count - 1) { + if (!is_specific && stripe_idx < stripe_count - 1) { /* * for large dir test, put all other slaves on one * remote MDT, otherwise we may save too many local @@ -2067,7 +2159,7 @@ static int lod_prep_md_striped_create(const struct lu_env *env, */ if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) idx = master_index; - idx_array[i + 1] = (idx + 1) % + mdt_indices[stripe_idx + 1] = (idx + 1) % (lod->lod_remote_mdt_count + 1); } /* tgt_dt and fid must be ready after search avaible OSP @@ -2076,47 +2168,125 @@ static int lod_prep_md_striped_create(const struct lu_env *env, LASSERT(fid_is_sane(&fid)); /* fail a remote stripe FID allocation */ - if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID)) + if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID)) continue; - conf.loc_flags = LOC_F_NEW; dto = dt_locate_at(env, tgt_dt, &fid, - dt->do_lu.lo_dev->ld_site->ls_top_dev, - &conf); - if (IS_ERR(dto)) - GOTO(out_put, rc = PTR_ERR(dto)); - stripe[i] = dto; + lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev, + &conf); + if (IS_ERR(dto)) { + rc = PTR_ERR(dto); + goto error; + } + + stripes[stripe_idx] = dto; + } + + return stripe_idx; + +error: + for (j = 1; j < stripe_idx; j++) { + LASSERT(stripes[j] != NULL); + dt_object_put(env, stripes[j]); + stripes[j] = NULL; + } + return rc; +} + +static int lod_prep_md_striped_create(const struct lu_env *env, + struct dt_object *dt, + struct lu_attr *attr, + const struct lmv_user_md_v1 *lum, + struct dt_object_format *dof, + struct thandle *th) +{ + struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev); + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object **stripes; + struct lu_object_conf conf = { .loc_flags = LOC_F_NEW }; + struct lu_fid fid = { 0 }; + __u32 stripe_count; + int i; + int rc = 0; + + ENTRY; + + /* The lum has been verifed in lod_verify_md_striping */ + LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC || + le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC); + + stripe_count = lo->ldo_dir_stripe_count; + + OBD_ALLOC_PTR_ARRAY(stripes, stripe_count); + if (!stripes) + RETURN(-ENOMEM); + + /* Allocate the first stripe locally */ + rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL); + if (rc < 0) + GOTO(out, rc); + + stripes[0] = dt_locate_at(env, lod->lod_child, &fid, + dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf); + if (IS_ERR(stripes[0])) + GOTO(out, rc = PTR_ERR(stripes[0])); + + if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) { + lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs); + rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count); + if (rc == -EAGAIN) + rc = lod_mdt_alloc_rr(env, lo, stripes, 1, + stripe_count); + } else { + int *idx_array; + bool is_specific = false; + + OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count); + if (!idx_array) + GOTO(out, rc = -ENOMEM); + + if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) { + is_specific = true; + for (i = 0; i < stripe_count; i++) + idx_array[i] = + le32_to_cpu(lum->lum_objects[i].lum_mds); + } + + /* stripe 0 is local */ + idx_array[0] = + lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id; + rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array, + is_specific); + OBD_FREE_PTR_ARRAY(idx_array, stripe_count); } + if (rc < 0) + GOTO(out, rc); + + LASSERT(rc > 0); + lo->ldo_dir_striped = 1; - lo->ldo_stripe = stripe; - lo->ldo_dir_stripe_count = i; + lo->ldo_stripe = stripes; + lo->ldo_dir_stripe_count = rc; lo->ldo_dir_stripes_allocated = stripe_count; smp_mb(); lo->ldo_dir_stripe_loaded = 1; - if (lo->ldo_dir_stripe_count == 0) - GOTO(out_put, rc = -ENOSPC); - rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th); - if (rc != 0) - GOTO(out_put, rc); + if (rc < 0) + lod_striping_free(env, lo); -out_put: - if (rc < 0) { - for (i = 0; i < stripe_count; i++) - if (stripe[i] != NULL) - dt_object_put(env, stripe[i]); - OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count); - lo->ldo_dir_stripe_count = 0; - lo->ldo_dir_stripes_allocated = 0; - lo->ldo_stripe = NULL; - } + RETURN(rc); -out_free: - OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count); +out: + LASSERT(rc < 0); + if (!IS_ERR_OR_NULL(stripes[0])) + dt_object_put(env, stripes[0]); + for (i = 1; i < stripe_count; i++) + LASSERT(!stripes[i]); + OBD_FREE_PTR_ARRAY(stripes, stripe_count); - RETURN(rc); + return rc; } /** @@ -2201,281 +2371,80 @@ out: } /** - * Append source stripes after target stripes for migrating directory. NB, we - * only need to declare this, the append is done inside lod_xattr_set_lmv(). + * Set or replace striped directory layout, and LFSCK may set layout on a plain + * directory, so don't check stripe count. * * \param[in] env execution environment * \param[in] dt target object * \param[in] buf LMV buf which contains source stripe fids + * \param[in] fl set or replace * \param[in] th transaction handle * * \retval 0 on success * \retval negative if failed */ -static int lod_dir_declare_layout_add(const struct lu_env *env, - struct dt_object *dt, - const struct lu_buf *buf, - struct thandle *th) +static int lod_dir_layout_set(const struct lu_env *env, + struct dt_object *dt, + const struct lu_buf *buf, + int fl, + struct thandle *th) { - struct lod_thread_info *info = lod_env_info(env); - struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev); - struct lod_tgt_descs *ltd = &lod->lod_mdt_descs; - struct lod_object *lo = lod_dt_obj(dt); struct dt_object *next = dt_object_child(dt); - struct dt_object_format *dof = &info->lti_format; + struct lod_object *lo = lod_dt_obj(dt); + struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev); struct lmv_mds_md_v1 *lmv = buf->lb_buf; - struct dt_object **stripe; - __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count); - struct lu_fid *fid = &info->lti_fid; - struct lod_tgt_desc *tgt; - struct dt_object *dto; - struct dt_device *tgt_dt; - int type = LU_SEQ_RANGE_ANY; - struct dt_insert_rec *rec = &info->lti_dt_rec; - char *stripe_name = info->lti_key; - struct lu_name *sname; - struct linkea_data ldata = { NULL }; - struct lu_buf linkea_buf; - __u32 idx; + struct lmv_mds_md_v1 *slave_lmv; + struct lu_buf slave_buf; int i; int rc; ENTRY; - if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1) + if (!lmv_is_sane2(lmv)) RETURN(-EINVAL); - if (stripe_count == 0) - RETURN(-EINVAL); + /* adjust hash for dir merge, which may not be set in user command */ + if (lmv_is_merging(lmv) && !lmv->lmv_migrate_hash) + lmv->lmv_merge_hash = + lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern; - dof->dof_type = DFT_DIR; + LMV_DEBUG(D_INFO, lmv, "set"); - OBD_ALLOC(stripe, - sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count)); - if (stripe == NULL) - RETURN(-ENOMEM); + rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th); + if (rc) + RETURN(rc); - for (i = 0; i < lo->ldo_dir_stripe_count; i++) - stripe[i] = lo->ldo_stripe[i]; + /* directory restripe may update stripe LMV directly */ + if (!lo->ldo_dir_stripe_count) + RETURN(0); - for (i = 0; i < stripe_count; i++) { - fid_le_to_cpu(fid, - &lmv->lmv_stripe_fids[i]); - if (!fid_is_sane(fid)) - continue; - - rc = lod_fld_lookup(env, lod, fid, &idx, &type); - if (rc) - GOTO(out, rc); - - if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) { - tgt_dt = lod->lod_child; - } else { - tgt = LTD_TGT(ltd, idx); - if (tgt == NULL) - GOTO(out, rc = -ESTALE); - tgt_dt = tgt->ltd_tgt; - } - - dto = dt_locate_at(env, tgt_dt, fid, - lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev, - NULL); - if (IS_ERR(dto)) - GOTO(out, rc = PTR_ERR(dto)); - - stripe[i + lo->ldo_dir_stripe_count] = dto; - - if (!dt_try_as_dir(env, dto)) - GOTO(out, rc = -ENOTDIR); - - rc = lod_sub_declare_ref_add(env, dto, th); - if (rc) - GOTO(out, rc); - - rc = lod_sub_declare_insert(env, dto, - (const struct dt_rec *)rec, - (const struct dt_key *)dot, th); - if (rc) - GOTO(out, rc); - - rc = lod_sub_declare_insert(env, dto, - (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th); - if (rc) - GOTO(out, rc); - - rc = lod_sub_declare_xattr_set(env, dto, buf, - XATTR_NAME_LMV, 0, th); - if (rc) - GOTO(out, rc); - - snprintf(stripe_name, sizeof(info->lti_key), DFID":%u", - PFID(lu_object_fid(&dto->do_lu)), - i + lo->ldo_dir_stripe_count); - - sname = lod_name_get(env, stripe_name, strlen(stripe_name)); - rc = linkea_links_new(&ldata, &info->lti_linkea_buf, - sname, lu_object_fid(&dt->do_lu)); - if (rc) - GOTO(out, rc); - - linkea_buf.lb_buf = ldata.ld_buf->lb_buf; - linkea_buf.lb_len = ldata.ld_leh->leh_len; - rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf, - XATTR_NAME_LINK, 0, th); - if (rc) - GOTO(out, rc); - - rc = lod_sub_declare_insert(env, next, - (const struct dt_rec *)rec, - (const struct dt_key *)stripe_name, - th); - if (rc) - GOTO(out, rc); - - rc = lod_sub_declare_ref_add(env, next, th); - if (rc) - GOTO(out, rc); - } - - if (lo->ldo_stripe) - OBD_FREE(lo->ldo_stripe, - sizeof(*stripe) * lo->ldo_dir_stripes_allocated); - lo->ldo_stripe = stripe; - lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count; - lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type); - lo->ldo_dir_stripe_count += stripe_count; - lo->ldo_dir_stripes_allocated += stripe_count; - lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION; - - RETURN(0); -out: - i = lo->ldo_dir_stripe_count; - while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i]) - dt_object_put(env, stripe[i++]); - - OBD_FREE(stripe, - sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count)); - RETURN(rc); -} - -static int lod_dir_declare_layout_delete(const struct lu_env *env, - struct dt_object *dt, - const struct lu_buf *buf, - struct thandle *th) -{ - struct lod_thread_info *info = lod_env_info(env); - struct lod_object *lo = lod_dt_obj(dt); - struct dt_object *next = dt_object_child(dt); - struct lmv_user_md *lmu = buf->lb_buf; - __u32 final_stripe_count; - char *stripe_name = info->lti_key; - struct dt_object *dto; - int i; - int rc = 0; + lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type); + lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset); + lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash); + lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version); - if (!lmu) - return -EINVAL; + OBD_ALLOC_PTR(slave_lmv); + if (!slave_lmv) + RETURN(-ENOMEM); - final_stripe_count = le32_to_cpu(lmu->lum_stripe_count); - if (final_stripe_count >= lo->ldo_dir_stripe_count) - return -EINVAL; + lod_prep_slave_lmv_md(slave_lmv, lmv); + slave_buf.lb_buf = slave_lmv; + slave_buf.lb_len = sizeof(*slave_lmv); - for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) { - dto = lo->ldo_stripe[i]; - if (!dto) + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + if (!lo->ldo_stripe[i]) continue; - if (!dt_try_as_dir(env, dto)) - return -ENOTDIR; - - rc = lod_sub_declare_delete(env, dto, - (const struct dt_key *)dot, th); - if (rc) - return rc; - - rc = lod_sub_declare_ref_del(env, dto, th); - if (rc) - return rc; - - rc = lod_sub_declare_delete(env, dto, - (const struct dt_key *)dotdot, th); - if (rc) - return rc; - - snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", - PFID(lu_object_fid(&dto->do_lu)), i); - - rc = lod_sub_declare_delete(env, next, - (const struct dt_key *)stripe_name, th); - if (rc) - return rc; - - rc = lod_sub_declare_ref_del(env, next, th); - if (rc) - return rc; - } - - return 0; -} - -/* - * delete stripes from dir master object, the lum_stripe_count in argument is - * the final stripe count, the stripes after that will be deleted, NB, they - * are not destroyed, but deleted from it's parent namespace, this function - * will be called in two places: - * 1. mdd_migrate_create() delete stripes from source, and append them to - * target. - * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them. - */ -static int lod_dir_layout_delete(const struct lu_env *env, - struct dt_object *dt, - const struct lu_buf *buf, - struct thandle *th) -{ - struct lod_thread_info *info = lod_env_info(env); - struct lod_object *lo = lod_dt_obj(dt); - struct dt_object *next = dt_object_child(dt); - struct lmv_user_md *lmu = buf->lb_buf; - __u32 final_stripe_count; - char *stripe_name = info->lti_key; - struct dt_object *dto; - int i; - int rc = 0; - - ENTRY; - - if (!lmu) - RETURN(-EINVAL); - - final_stripe_count = le32_to_cpu(lmu->lum_stripe_count); - if (final_stripe_count >= lo->ldo_dir_stripe_count) - RETURN(-EINVAL); - - for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) { - dto = lo->ldo_stripe[i]; - if (!dto) + if (!dt_object_exists(lo->ldo_stripe[i])) continue; - rc = lod_sub_delete(env, dto, - (const struct dt_key *)dotdot, th); - if (rc) - break; - - snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", - PFID(lu_object_fid(&dto->do_lu)), i); - - rc = lod_sub_delete(env, next, - (const struct dt_key *)stripe_name, th); - if (rc) - break; - - rc = lod_sub_ref_del(env, next, th); + rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf, + XATTR_NAME_LMV, fl, th); if (rc) break; } - lod_striping_free(env, lod_dt_obj(dt)); + OBD_FREE_PTR(slave_lmv); RETURN(rc); } @@ -2512,7 +2481,7 @@ static int lod_dir_declare_xattr_set(const struct lu_env *env, if (rc != 0) RETURN(rc); } else if (strcmp(name, XATTR_NAME_LOV) == 0) { - rc = lod_verify_striping(d, lo, buf, false); + rc = lod_verify_striping(env, d, lo, buf, false); if (rc != 0) RETURN(rc); } @@ -2673,7 +2642,7 @@ inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo, else if (lod_comp_inited(entry)) return entry->llc_stripe_count; else if ((__u16)-1 == entry->llc_stripe_count) - return lod->lod_desc.ld_tgt_count; + return lod->lod_ost_count; else return lod_get_stripe_count(lod, lo, entry->llc_stripe_count, false); @@ -2744,14 +2713,14 @@ static int lod_declare_layout_add(const struct lu_env *env, { struct lod_thread_info *info = lod_env_info(env); struct lod_layout_component *comp_array, *lod_comp, *old_array; - struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev); + struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev); struct dt_object *next = dt_object_child(dt); - struct lov_desc *desc = &d->lod_desc; - struct lod_object *lo = lod_dt_obj(dt); - struct lov_user_md_v3 *v3; - struct lov_comp_md_v1 *comp_v1 = buf->lb_buf; - __u32 magic; - int i, rc, array_cnt, old_array_cnt; + struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc; + struct lod_object *lo = lod_dt_obj(dt); + struct lov_user_md_v3 *v3; + struct lov_comp_md_v1 *comp_v1 = buf->lb_buf; + __u32 magic; + int i, rc, array_cnt, old_array_cnt; ENTRY; LASSERT(lo->ldo_is_composite); @@ -2759,7 +2728,7 @@ static int lod_declare_layout_add(const struct lu_env *env, if (lo->ldo_flr_state != LCM_FL_NONE) RETURN(-EBUSY); - rc = lod_verify_striping(d, lo, buf, false); + rc = lod_verify_striping(env, d, lo, buf, false); if (rc != 0) RETURN(rc); @@ -2773,7 +2742,7 @@ static int lod_declare_layout_add(const struct lu_env *env, RETURN(-EINVAL); array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count; - OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt); + OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt); if (comp_array == NULL) RETURN(-ENOMEM); @@ -2796,7 +2765,7 @@ 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; @@ -2827,7 +2796,7 @@ static int lod_declare_layout_add(const struct lu_env *env, GOTO(error, rc); } - OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt); + OBD_FREE_PTR_ARRAY(old_array, old_array_cnt); LASSERT(lo->ldo_mirror_count == 1); lo->ldo_mirrors[0].lme_end = array_cnt - 1; @@ -2843,15 +2812,53 @@ error: lod_comp->llc_pool = NULL; } } - OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt); + OBD_FREE_PTR_ARRAY(comp_array, array_cnt); RETURN(rc); } /** + * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror. + * @mirror_id: Mirror id to be checked. + * @lo: LOD object. + * + * This function checks if a mirror with specified @mirror_id is the last + * non-stale mirror of a LOD object @lo. + * + * Return: true or false. + */ +static inline +bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo) +{ + struct lod_layout_component *lod_comp; + bool has_stale_flag; + int i; + + for (i = 0; i < lo->ldo_mirror_count; i++) { + if (lo->ldo_mirrors[i].lme_id == mirror_id || + lo->ldo_mirrors[i].lme_stale) + continue; + + has_stale_flag = false; + lod_foreach_mirror_comp(lod_comp, lo, i) { + if (lod_comp->llc_flags & LCME_FL_STALE) { + has_stale_flag = true; + break; + } + } + if (!has_stale_flag) + return false; + } + + return true; +} + +/** * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field, * the '$field' can only be 'flags' now. The xattr value is binary * lov_comp_md_v1 which contains the component ID(s) and the value of * the field to be modified. + * Please update allowed_lustre_lov macro if $field groks more values + * in the future. * * \param[in] env execution environment * \param[in] dt dt_object to be modified @@ -2877,6 +2884,9 @@ static int lod_declare_layout_set(const struct lu_env *env, bool changed = false; ENTRY; + /* Please update allowed_lustre_lov macro if op + * groks more values in the future + */ if (strcmp(op, "set.flags") != 0) { CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n", lod2obd(d)->obd_name, op); @@ -2902,6 +2912,7 @@ static int lod_declare_layout_set(const struct lu_env *env, __u32 id = comp_v1->lcm_entries[i].lcme_id; __u32 flags = comp_v1->lcm_entries[i].lcme_flags; __u32 mirror_flag = flags & LCME_MIRROR_FLAGS; + __u16 mirror_id = mirror_id_of(id); bool neg = flags & LCME_FL_NEG; if (flags & LCME_FL_INIT) { @@ -2916,7 +2927,7 @@ static int lod_declare_layout_set(const struct lu_env *env, /* lfs only put one flag in each entry */ if ((flags && id != lod_comp->llc_id) || - (mirror_flag && mirror_id_of(id) != + (mirror_flag && mirror_id != mirror_id_of(lod_comp->llc_id))) continue; @@ -2926,8 +2937,13 @@ static int lod_declare_layout_set(const struct lu_env *env, if (mirror_flag) lod_comp->llc_flags &= ~mirror_flag; } else { - if (flags) + if (flags) { + if ((flags & LCME_FL_STALE) && + lod_last_non_stale_mirror(mirror_id, + lo)) + RETURN(-EUCLEAN); lod_comp->llc_flags |= flags; + } if (mirror_flag) { lod_comp->llc_flags |= mirror_flag; if (mirror_flag & LCME_FL_NOSYNC) @@ -3190,13 +3206,13 @@ static int lod_layout_convert(struct lod_thread_info *info) } lcm = info->lti_ea_store; + memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme)); lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1); lcm->lcm_size = cpu_to_le32(size); lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu( lmm_save->lmm_layout_gen)); lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE); lcm->lcm_entry_count = cpu_to_le16(1); - lcm->lcm_mirror_count = 0; lcme = &lcm->lcm_entries[0]; lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT); @@ -3228,6 +3244,7 @@ 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 +3253,8 @@ 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 +3283,7 @@ 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: @@ -3318,10 +3338,17 @@ static int lod_declare_layout_merge(const struct lu_env *env, lcme->lcme_id = cpu_to_le32(id); } - id = MAX(le32_to_cpu(lcme->lcme_id), id); + id = max(le32_to_cpu(lcme->lcme_id), id); } 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 +3357,8 @@ 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); @@ -3441,9 +3470,8 @@ static int lod_declare_xattr_set(const struct lu_env *env, strcmp(name, XATTR_LUSTRE_LOV) == 0); rc = lod_declare_layout_split(env, dt, buf, th); } else if (S_ISREG(mode) && - strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 && - strncmp(name, XATTR_LUSTRE_LOV, - strlen(XATTR_LUSTRE_LOV)) == 0) { + strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 && + allowed_lustre_lov(name)) { /* * this is a request to modify object's striping. * add/set/del component(s). @@ -3452,20 +3480,6 @@ static int lod_declare_xattr_set(const struct lu_env *env, RETURN(-ENOENT); rc = lod_declare_modify_layout(env, dt, name, buf, th); - } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 && - strlen(name) > strlen(XATTR_NAME_LMV) + 1) { - const char *op = name + strlen(XATTR_NAME_LMV) + 1; - - rc = -ENOTSUPP; - if (strcmp(op, "add") == 0) - rc = lod_dir_declare_layout_add(env, dt, buf, th); - else if (strcmp(op, "del") == 0) - rc = lod_dir_declare_layout_delete(env, dt, buf, th); - else if (strcmp(op, "set") == 0) - rc = lod_sub_declare_xattr_set(env, next, buf, - XATTR_NAME_LMV, fl, th); - - RETURN(rc); } else if (S_ISDIR(mode)) { rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th); } else if (strcmp(name, XATTR_NAME_FID) == 0) { @@ -3630,8 +3644,23 @@ 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); @@ -3688,8 +3717,7 @@ static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env, if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)), le32_to_cpu(lum->lum_stripe_offset)) && - le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC && - !(le32_to_cpu(lum->lum_hash_type) & LMV_HASH_FLAG_SPACE)) { + le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) { rc = lod_xattr_del_internal(env, dt, name, th); if (rc == -ENODATA) rc = 0; @@ -3757,7 +3785,7 @@ static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt, if (rc != 0) RETURN(rc); - attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | + attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS | LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID; dof->dof_type = DFT_DIR; @@ -3790,10 +3818,16 @@ static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt, if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE)) continue; - /* 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); + /* don't create stripe if: + * 1. it's source stripe of migrating directory + * 2. it's existed stripe of splitting directory + */ + if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) || + (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) { + if (!dt_object_exists(dto)) + GOTO(out, rc = -EINVAL); + } else { + 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); @@ -3813,12 +3847,6 @@ static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt, GOTO(out, rc); } - rec->rec_fid = lu_object_fid(&dt->do_lu); - rc = lod_sub_insert(env, dto, (struct dt_rec *)rec, - (const struct dt_key *)dotdot, th); - if (rc != 0) - GOTO(out, rc); - if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) || cfs_fail_val != i) { if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) && @@ -3830,11 +3858,26 @@ static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt, cpu_to_le32(i); rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf, - XATTR_NAME_LMV, fl, th); + XATTR_NAME_LMV, 0, th); if (rc != 0) GOTO(out, rc); } + /* don't insert stripe if it's existed stripe of splitting + * directory (this directory is striped). + * NB, plain directory will insert itself as the first + * stripe in target. + */ + if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 && + lo->ldo_dir_split_offset > i) + continue; + + rec->rec_fid = lu_object_fid(&dt->do_lu); + rc = lod_sub_insert(env, dto, (struct dt_rec *)rec, + (const struct dt_key *)dotdot, th); + if (rc != 0) + GOTO(out, rc); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) && cfs_fail_val == i) snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", @@ -4158,7 +4201,7 @@ static int lod_layout_repeat_comp(const struct lu_env *env, CDEBUG(D_LAYOUT, "repeating component %d\n", index); - OBD_ALLOC(comp_array, sizeof(*comp_array) * new_cnt); + OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt); if (comp_array == NULL) GOTO(out, rc = -ENOMEM); @@ -4180,6 +4223,7 @@ static int lod_layout_repeat_comp(const struct lu_env *env, new_comp->llc_flags &= ~LCME_FL_INIT; new_comp->llc_stripe = NULL; new_comp->llc_stripes_allocated = 0; + new_comp->llc_ost_indices = NULL; new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT; /* for uninstantiated components, layout gen stores default stripe * offset */ @@ -4205,8 +4249,7 @@ static int lod_layout_repeat_comp(const struct lu_env *env, new_comp->llc_ostlist.op_array = op_array; } - OBD_FREE(lo->ldo_comp_entries, - sizeof(*comp_array) * lo->ldo_comp_cnt); + OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt); lo->ldo_comp_entries = comp_array; lo->ldo_comp_cnt = new_cnt; @@ -4220,7 +4263,7 @@ static int lod_layout_repeat_comp(const struct lu_env *env, EXIT; out: if (rc) - OBD_FREE(comp_array, sizeof(*comp_array) * new_cnt); + OBD_FREE_PTR_ARRAY(comp_array, new_cnt); return rc; } @@ -4237,12 +4280,11 @@ static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt) RETURN(0); if (info->lti_comp_size > 0) { - OBD_FREE(info->lti_comp_idx, - info->lti_comp_size * sizeof(__u32)); + OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size); info->lti_comp_size = 0; } - OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32)); + OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt); if (!info->lti_comp_idx) RETURN(-ENOMEM); @@ -4279,6 +4321,9 @@ 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); @@ -4326,11 +4371,11 @@ static int lod_layout_del_prep_layout(const struct lu_env *env, lu_object_put(env, &obj->do_lu); lod_comp->llc_stripe[j] = NULL; } - OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) * - lod_comp->llc_stripes_allocated); + OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe, + lod_comp->llc_stripes_allocated); lod_comp->llc_stripe = NULL; - OBD_FREE(lod_comp->llc_ost_indices, - sizeof(__u32) * lod_comp->llc_stripes_allocated); + OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices, + lod_comp->llc_stripes_allocated); lod_comp->llc_ost_indices = NULL; lod_comp->llc_stripes_allocated = 0; } @@ -4343,7 +4388,7 @@ static int lod_layout_del_prep_layout(const struct lu_env *env, if (info->lti_count > 0) { struct lod_layout_component *comp_array; - OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count); + OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count); if (comp_array == NULL) GOTO(out, rc = -ENOMEM); @@ -4353,8 +4398,7 @@ static int lod_layout_del_prep_layout(const struct lu_env *env, sizeof(*comp_array)); } - OBD_FREE(lo->ldo_comp_entries, - sizeof(*comp_array) * lo->ldo_comp_cnt); + OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt); lo->ldo_comp_entries = comp_array; lo->ldo_comp_cnt = info->lti_count; } else { @@ -4388,9 +4432,7 @@ 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 +4468,8 @@ 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. * @@ -4449,40 +4492,36 @@ static int lod_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, struct thandle *th) { - struct dt_object *next = dt_object_child(dt); - int rc; - ENTRY; - - if (S_ISDIR(dt->do_lu.lo_header->loh_attr) && - strcmp(name, XATTR_NAME_LMV) == 0) { - rc = lod_dir_striping_create(env, dt, NULL, NULL, th); - RETURN(rc); - } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) && - strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 && - strlen(name) > strlen(XATTR_NAME_LMV) + 1) { - const char *op = name + strlen(XATTR_NAME_LMV) + 1; + struct dt_object *next = dt_object_child(dt); + int rc; - rc = -ENOTSUPP; - /* - * XATTR_NAME_LMV".add" is never called, but only declared, - * because lod_xattr_set_lmv() will do the addition. - */ - if (strcmp(op, "del") == 0) - rc = lod_dir_layout_delete(env, dt, buf, th); - else if (strcmp(op, "set") == 0) - rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, - fl, th); + ENTRY; + + if (S_ISDIR(dt->do_lu.lo_header->loh_attr) && + !strcmp(name, XATTR_NAME_LMV)) { + switch (fl) { + case LU_XATTR_CREATE: + rc = lod_dir_striping_create(env, dt, NULL, NULL, th); + break; + case 0: + case LU_XATTR_REPLACE: + rc = lod_dir_layout_set(env, dt, buf, fl, th); + break; + default: + LBUG(); + } RETURN(rc); } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) && - strcmp(name, XATTR_NAME_LOV) == 0) { + strcmp(name, XATTR_NAME_LOV) == 0) { struct lod_default_striping *lds = lod_lds_buf_get(env); struct lov_user_md_v1 *v1 = buf->lb_buf; char pool[LOV_MAXPOOLNAME + 1]; 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); @@ -4536,9 +4575,9 @@ static int lod_xattr_set(const struct lu_env *env, th); RETURN(rc); } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) && - (!strcmp(name, XATTR_NAME_LOV) || - !strncmp(name, XATTR_LUSTRE_LOV, - strlen(XATTR_LUSTRE_LOV)))) { + (strcmp(name, XATTR_NAME_LOV) == 0 || + strcmp(name, XATTR_LUSTRE_LOV) == 0 || + allowed_lustre_lov(name))) { /* in case of lov EA swap, just set it * if not, it is a replay so check striping match what we * already have during req replay, declare_xattr_set() @@ -4613,7 +4652,13 @@ 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 +4704,9 @@ 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 +4800,8 @@ 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 +4832,21 @@ 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 +4897,24 @@ 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 +4996,7 @@ 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; @@ -4961,10 +5019,11 @@ static void lod_striping_from_default(struct lod_object *lo, umode_t mode) { struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); - struct lov_desc *desc = &d->lod_desc; int i, rc; if (lds->lds_def_striping_set && S_ISREG(mode)) { + struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc; + rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt, lds->lds_def_comp_cnt); if (rc != 0) @@ -5021,7 +5080,7 @@ 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) @@ -5031,8 +5090,7 @@ static void lod_striping_from_default(struct lod_object *lo, lo->ldo_dir_stripe_offset = lds->lds_dir_def_stripe_offset; if (lo->ldo_dir_hash_type == 0) - lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type & - ~LMV_HASH_FLAG_SPACE; + lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type; CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, " "offset:%u, hash_type:%u\n", @@ -5041,7 +5099,8 @@ 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 +5120,9 @@ 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 +5157,10 @@ static void lod_ah_init(const struct lu_env *env, LASSERT(child); + if (ah->dah_append_stripes == -1) + ah->dah_append_stripes = + d->lod_ost_descs.ltd_lov_desc.ld_tgt_count; + if (likely(parent)) { nextp = dt_object_child(parent); lp = lod_dt_obj(parent); @@ -5133,12 +5199,15 @@ static void lod_ah_init(const struct lu_env *env, * If parent object is not root directory, * then get default striping from parent object. */ - if (likely(lp != NULL) && !fid_is_root(lod_object_fid(lp))) + if (likely(lp != NULL)) { lod_get_default_striping(env, lp, lds); - /* set child default striping info, default value is NULL */ - if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set) - lc->ldo_def_striping = lds; + /* inherit default striping except ROOT */ + if ((lds->lds_def_striping_set || + lds->lds_dir_def_striping_set) && + !fid_is_root(lod_object_fid(lp))) + lc->ldo_def_striping = lds; + } /* It should always honour the specified stripes */ /* Note: old client (< 2.7)might also do lfs mkdir, whose EA @@ -5176,6 +5245,10 @@ static void lod_ah_init(const struct lu_env *env, lc->ldo_dir_stripe_count = 0; } + if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK)) + lc->ldo_dir_hash_type |= + d->lod_mdt_descs.ltd_lmv_desc.ld_pattern; + CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n", lc->ldo_dir_stripe_count, (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type); @@ -5196,7 +5269,7 @@ 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 +5297,10 @@ 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 +5331,7 @@ 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) @@ -5269,14 +5344,15 @@ out: } LASSERT(!lc->ldo_is_composite); lod_comp = &lc->ldo_comp_entries[0]; - desc = &d->lod_desc; - lod_adjust_stripe_info(lod_comp, desc); + desc = &d->lod_ost_descs.ltd_lov_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; } -#define ll_do_div64(aaa,bbb) do_div((aaa), (bbb)) /** * Size initialization on late striping. * @@ -5340,14 +5416,13 @@ static int lod_declare_init_size(const struct lu_env *env, continue; LASSERT(objects != NULL && stripe_size != 0); - /* ll_do_div64(a, b) returns a % b, and a = a / b */ - ll_do_div64(size, (__u64)stripe_size); - stripe = ll_do_div64(size, (__u64)stripe_count); + do_div(size, stripe_size); + stripe = do_div(size, stripe_count); LASSERT(objects[stripe] != NULL); size = size * stripe_size; offs = attr->la_size; - size += ll_do_div64(offs, stripe_size); + size += do_div(offs, stripe_size); attr->la_valid = LA_SIZE; attr->la_size = size; @@ -5473,7 +5548,7 @@ static inline int dt_object_qos_mkdir(const struct lu_env *env, return -EINVAL; lmu = info->lti_ea_store; - return !!(le32_to_cpu(lmu->lum_hash_type) & LMV_HASH_FLAG_SPACE); + return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT; } /** @@ -5539,32 +5614,31 @@ static int lod_declare_create(const struct lu_env *env, struct dt_object *dt, if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT)) GOTO(out, rc = -EREMOTE); - if (lo->ldo_dir_stripe_offset == -1) { + if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) { + struct lod_default_striping *lds; + + lds = lo->ldo_def_striping; /* - * child and parent should be in the same MDT, - * but if parent has plain layout, it's allowed. + * child and parent should be on the same MDT, + * but if parent has default LMV, and the start + * MDT offset is -1, it's allowed. This check + * is not necessary after 2.12.22 because client + * follows this already, but old client may not. */ if (hint->dah_parent && - dt_object_remote(hint->dah_parent)) { - rc = dt_object_qos_mkdir(env, - lo->ldo_obj.do_lu.lo_dev, - hint->dah_parent); - if (rc <= 0) - GOTO(out, rc ? rc : -EREMOTE); - } + dt_object_remote(hint->dah_parent) && lds && + lds->lds_dir_def_stripe_offset != + LMV_OFFSET_DEFAULT) + GOTO(out, rc = -EREMOTE); } else if (lo->ldo_dir_stripe_offset != ss->ss_node_id) { struct lod_device *lod; - struct lod_tgt_descs *ltd; - struct lod_tgt_desc *tgt = NULL; + struct lu_tgt_desc *mdt = 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); - if (tgt->ltd_index == + lod_foreach_mdt(lod, mdt) { + if (mdt->ltd_index == lo->ldo_dir_stripe_offset) { found_mdt = true; break; @@ -5982,7 +6056,7 @@ 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) @@ -6217,7 +6291,9 @@ static int lod_invalidate(const struct lu_env *env, struct dt_object *dt) } static int lod_declare_instantiate_components(const struct lu_env *env, - struct lod_object *lo, struct thandle *th) + struct lod_object *lo, + struct thandle *th, + __u64 reserve) { struct lod_thread_info *info = lod_env_info(env); int i; @@ -6228,7 +6304,7 @@ static int lod_declare_instantiate_components(const struct lu_env *env, for (i = 0; i < info->lti_count; i++) { rc = lod_qos_prep_create(env, lo, NULL, th, - info->lti_comp_idx[i]); + info->lti_comp_idx[i], reserve); if (rc) break; } @@ -6264,15 +6340,15 @@ static int lod_declare_instantiate_components(const struct lu_env *env, */ static bool lod_sel_osts_allowed(const struct lu_env *env, struct lod_object *lo, - int index, __u64 extension_size, + int index, __u64 reserve, struct lu_extent *extent, struct lu_extent *comp_extent, int write) { struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index]; struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); - struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs; + struct lod_thread_info *tinfo = lod_env_info(env); + struct obd_statfs *sfs = &tinfo->lti_osfs; __u64 available = 0; - __u64 size; bool ret = true; int i, rc; @@ -6280,21 +6356,6 @@ static bool lod_sel_osts_allowed(const struct lu_env *env, LASSERT(lod_comp->llc_stripe_count != 0); - if (write == 0 || - (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) { - /* truncate or append */ - size = extension_size; - } else { - /* In case of write op, check the real write extent, - * it may be larger than the extension_size */ - size = roundup(min(extent->e_end, comp_extent->e_end) - - max(extent->e_start, comp_extent->e_start), - extension_size); - } - /* extension_size is file level, so we must divide by stripe count to - * compare it to available space on a single OST */ - size /= lod_comp->llc_stripe_count; - lod_getref(&lod->lod_ost_descs); for (i = 0; i < lod_comp->llc_stripe_count; i++) { int index = lod_comp->llc_ost_indices[i]; @@ -6321,13 +6382,13 @@ static bool lod_sel_osts_allowed(const struct lu_env *env, if (j < lod_comp->llc_stripe_count) continue; - if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) { + if (!test_bit(index, lod->lod_ost_bitmap)) { CDEBUG(D_LAYOUT, "ost %d no longer present\n", index); ret = false; break; } - rc = dt_statfs(env, ost->ltd_ost, sfs, &info); + rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info); if (rc) { CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n", index, rc); @@ -6335,9 +6396,9 @@ static bool lod_sel_osts_allowed(const struct lu_env *env, break; } - if (sfs->os_state & OS_STATE_ENOSPC || - sfs->os_state & OS_STATE_READONLY || - sfs->os_state & OS_STATE_DEGRADED) { + if (sfs->os_state & OS_STATFS_ENOSPC || + sfs->os_state & OS_STATFS_READONLY || + sfs->os_state & OS_STATFS_DEGRADED) { CDEBUG(D_LAYOUT, "ost %d is not availble for SEL " "extension, state %u\n", index, sfs->os_state); ret = false; @@ -6355,11 +6416,11 @@ static bool lod_sel_osts_allowed(const struct lu_env *env, (100ull * sfs->os_bavail) / sfs->os_blocks, (100ull * sfs->os_bfree) / sfs->os_blocks); - if (size * repeated > available) { + if (reserve * repeated > available) { ret = false; CDEBUG(D_LAYOUT, "low space on ost %d, available %llu " - "< extension size %llu\n", index, available, - extension_size); + "< extension size %llu repeated %d\n", index, + available, reserve, repeated); break; } } @@ -6465,6 +6526,26 @@ static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end, return new_end; } +/** + * Calculate the exact reservation (per-OST extension_size) on the OSTs being + * instantiated. It needs to be calculated in advance and taken into account at + * the instantiation time, because otherwise lod_statfs_and_check() may consider + * an OST as OK, but SEL needs its extension_size to fit the free space and the + * OST may turn out to be low-on-space, thus inappropriate OST may be used and + * ENOSPC occurs. + * + * \param[in] lod_comp lod component we are checking + * + * \retval size to reserved on each OST of lod_comp's stripe. + */ +static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp) +{ + /* extension_size is file level, so we must divide by stripe count to + * compare it to available space on a single OST */ + return lod_comp->llc_stripe_size * SEL_UNIT_SIZE / + lod_comp->llc_stripe_count; +} + /* As lod_sel_handler() could be re-entered for the same component several * times, this is the data for the next call. Fields could be changed to * component indexes when needed, (e.g. if there is no need to instantiate @@ -6546,7 +6627,7 @@ static int lod_sel_handler(const struct lu_env *env, struct lod_layout_component *lod_comp; struct lod_layout_component *prev; struct lod_layout_component *next = NULL; - __u64 extension_size; + __u64 extension_size, reserve; __u64 new_end = 0; bool repeated; int change = 0; @@ -6583,11 +6664,13 @@ static int lod_sel_handler(const struct lu_env *env, RETURN(-EINVAL); } + reserve = lod_sel_stripe_reserved(lod_comp); + if (!prev->llc_stripe) { CDEBUG(D_LAYOUT, "Previous component not inited\n"); info->lti_count = 1; info->lti_comp_idx[0] = index - 1; - rc = lod_declare_instantiate_components(env, lo, th); + rc = lod_declare_instantiate_components(env, lo, th, reserve); /* ENOSPC tells us we can't use this component. If there is * a next or we are repeating, we either spill over (next) or * extend the original comp (repeat). Otherwise, return the @@ -6599,8 +6682,7 @@ static int lod_sel_handler(const struct lu_env *env, } if (sd->sd_force == 0 && rc == 0) - rc = !lod_sel_osts_allowed(env, lo, index - 1, - extension_size, extent, + rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent, &lod_comp->llc_extent, write); repeated = !!(sd->sd_repeat); @@ -6914,7 +6996,7 @@ static int lod_declare_update_plain(const struct lu_env *env, RETURN(-EALREADY); lod_obj_inc_layout_gen(lo); - rc = lod_declare_instantiate_components(env, lo, th); + rc = lod_declare_instantiate_components(env, lo, th, 0); EXIT; out: if (rc) @@ -7024,8 +7106,8 @@ static inline int lod_check_ost_avail(const struct lu_env *env, ost = OST_TGT(lod, idx); if (ost->ltd_statfs.os_state & - (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO | - OS_STATE_NOPRECREATE) || + (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO | + OS_STATFS_NOPRECREATE) || ost->ltd_active == 0) { CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n", PFID(lod_object_fid(lo)), index, idx, rc); @@ -7062,7 +7144,7 @@ static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo, * This algo can be revised later after knowing the topology of * cluster. */ - lod_qos_statfs_update(env, lod); + lod_qos_statfs_update(env, lod, &lod->lod_ost_descs); for (i = 0; i < lo->ldo_mirror_count; i++) { bool ost_avail = true; int index = (i + seq) % lo->ldo_mirror_count; @@ -7331,7 +7413,7 @@ static int lod_declare_update_rdonly(const struct lu_env *env, lo->ldo_layout_gen = layout_version & 0xffff; } - rc = lod_declare_instantiate_components(env, lo, th); + rc = lod_declare_instantiate_components(env, lo, th, 0); if (rc) GOTO(out, rc); @@ -7371,7 +7453,7 @@ static int lod_declare_update_write_pending(const struct lu_env *env, if (lo->ldo_mirrors[i].lme_stale) continue; - LASSERTF(primary < 0, DFID " has multiple primary: %u / %u", + LASSERTF(primary < 0, DFID " has multiple primary: %u / %u\n", PFID(lod_object_fid(lo)), lo->ldo_mirrors[i].lme_id, lo->ldo_mirrors[primary].lme_id); @@ -7475,7 +7557,7 @@ static int lod_declare_update_write_pending(const struct lu_env *env, lo->ldo_flr_state = LCM_FL_SYNC_PENDING; } - rc = lod_declare_instantiate_components(env, lo, th); + rc = lod_declare_instantiate_components(env, lo, th, 0); if (rc) GOTO(out, rc); @@ -7587,6 +7669,617 @@ out: RETURN(rc); } +typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th); + +/** + * Attach stripes after target's for migrating directory. NB, we + * only need to declare this, the actual work is done inside + * lod_xattr_set_lmv(). + * + * \param[in] env execution environment + * \param[in] dt target object + * \param[in] mlc layout change data + * \param[in] th transaction handle + * + * \retval 0 on success + * \retval negative if failed + */ +static int lod_dir_declare_layout_attach(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev); + struct lod_tgt_descs *ltd = &lod->lod_mdt_descs; + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object *next = dt_object_child(dt); + struct dt_object_format *dof = &info->lti_format; + struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf; + struct dt_object **stripes; + __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count); + struct lu_fid *fid = &info->lti_fid; + struct lod_tgt_desc *tgt; + struct dt_object *dto; + struct dt_device *tgt_dt; + int type = LU_SEQ_RANGE_ANY; + struct dt_insert_rec *rec = &info->lti_dt_rec; + char *stripe_name = info->lti_key; + struct lu_name *sname; + struct linkea_data ldata = { NULL }; + struct lu_buf linkea_buf; + __u32 idx; + int i; + int rc; + + ENTRY; + + if (!lmv_is_sane(lmv)) + RETURN(-EINVAL); + + if (!dt_try_as_dir(env, dt)) + return -ENOTDIR; + + dof->dof_type = DFT_DIR; + + OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count)); + if (!stripes) + RETURN(-ENOMEM); + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) + stripes[i] = lo->ldo_stripe[i]; + + rec->rec_type = S_IFDIR; + + for (i = 0; i < stripe_count; i++) { + fid_le_to_cpu(fid, + &lmv->lmv_stripe_fids[i]); + if (!fid_is_sane(fid)) + continue; + + rc = lod_fld_lookup(env, lod, fid, &idx, &type); + if (rc) + GOTO(out, rc); + + if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) { + tgt_dt = lod->lod_child; + } else { + tgt = LTD_TGT(ltd, idx); + if (tgt == NULL) + GOTO(out, rc = -ESTALE); + tgt_dt = tgt->ltd_tgt; + } + + dto = dt_locate_at(env, tgt_dt, fid, + lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev, + NULL); + if (IS_ERR(dto)) + GOTO(out, rc = PTR_ERR(dto)); + + stripes[i + lo->ldo_dir_stripe_count] = dto; + + if (!dt_try_as_dir(env, dto)) + GOTO(out, rc = -ENOTDIR); + + rc = lod_sub_declare_ref_add(env, dto, th); + if (rc) + GOTO(out, rc); + + rec->rec_fid = lu_object_fid(&dto->do_lu); + rc = lod_sub_declare_insert(env, dto, + (const struct dt_rec *)rec, + (const struct dt_key *)dot, th); + if (rc) + GOTO(out, rc); + + rc = lod_sub_declare_insert(env, dto, + (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, th); + if (rc) + GOTO(out, rc); + + rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf, + XATTR_NAME_LMV, 0, th); + if (rc) + GOTO(out, rc); + + snprintf(stripe_name, sizeof(info->lti_key), DFID":%u", + PFID(lu_object_fid(&dto->do_lu)), + i + lo->ldo_dir_stripe_count); + + sname = lod_name_get(env, stripe_name, strlen(stripe_name)); + rc = linkea_links_new(&ldata, &info->lti_linkea_buf, + sname, lu_object_fid(&dt->do_lu)); + if (rc) + GOTO(out, rc); + + linkea_buf.lb_buf = ldata.ld_buf->lb_buf; + linkea_buf.lb_len = ldata.ld_leh->leh_len; + rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf, + XATTR_NAME_LINK, 0, th); + if (rc) + GOTO(out, rc); + + rc = lod_sub_declare_insert(env, next, + (const struct dt_rec *)rec, + (const struct dt_key *)stripe_name, + th); + if (rc) + GOTO(out, rc); + + rc = lod_sub_declare_ref_add(env, next, th); + if (rc) + GOTO(out, rc); + } + + if (lo->ldo_stripe) + OBD_FREE_PTR_ARRAY(lo->ldo_stripe, + lo->ldo_dir_stripes_allocated); + lo->ldo_stripe = stripes; + lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count; + lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type); + lo->ldo_dir_stripe_count += stripe_count; + lo->ldo_dir_stripes_allocated += stripe_count; + + /* plain directory split creates target as a plain directory, while + * after source attached as the first stripe, it becomes a striped + * directory, set correct do_index_ops, otherwise it can't be unlinked. + */ + dt->do_index_ops = &lod_striped_index_ops; + + RETURN(0); +out: + i = lo->ldo_dir_stripe_count; + while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i]) + dt_object_put(env, stripes[i++]); + + OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count); + return rc; +} + +static int lod_dir_declare_layout_detach(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *unused, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object *next = dt_object_child(dt); + char *stripe_name = info->lti_key; + struct dt_object *dto; + int i; + int rc = 0; + + if (!dt_try_as_dir(env, dt)) + return -ENOTDIR; + + if (!lo->ldo_dir_stripe_count) + return lod_sub_declare_delete(env, next, + (const struct dt_key *)dotdot, th); + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (!dto) + continue; + + if (!dt_try_as_dir(env, dto)) + return -ENOTDIR; + + rc = lod_sub_declare_delete(env, dto, + (const struct dt_key *)dotdot, th); + if (rc) + return rc; + + snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", + PFID(lu_object_fid(&dto->do_lu)), i); + + rc = lod_sub_declare_delete(env, next, + (const struct dt_key *)stripe_name, th); + if (rc) + return rc; + + rc = lod_sub_declare_ref_del(env, next, th); + if (rc) + return rc; + } + + return 0; +} + +static int dt_dir_is_empty(const struct lu_env *env, + struct dt_object *obj) +{ + struct dt_it *it; + const struct dt_it_ops *iops; + int rc; + + ENTRY; + + if (!dt_try_as_dir(env, obj)) + RETURN(-ENOTDIR); + + iops = &obj->do_index_ops->dio_it; + it = iops->init(env, obj, LUDA_64BITHASH); + if (IS_ERR(it)) + RETURN(PTR_ERR(it)); + + rc = iops->get(env, it, (const struct dt_key *)""); + if (rc > 0) { + int i; + + for (rc = 0, i = 0; rc == 0 && i < 3; ++i) + rc = iops->next(env, it); + if (!rc) + rc = -ENOTEMPTY; + else if (rc == 1) + rc = 0; + } else if (!rc) { + /* Huh? Index contains no zero key? */ + rc = -EIO; + } + + iops->put(env, it); + iops->fini(env, it); + + RETURN(rc); +} + +static int lod_dir_declare_layout_shrink(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object *next = dt_object_child(dt); + struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf; + __u32 final_stripe_count; + char *stripe_name = info->lti_key; + struct lu_buf *lmv_buf = &info->lti_buf; + struct dt_object *dto; + int i; + int rc; + + LASSERT(lmu); + + if (!dt_try_as_dir(env, dt)) + return -ENOTDIR; + + /* shouldn't be called on plain directory */ + LASSERT(lo->ldo_dir_stripe_count); + + lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1; + lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1); + + final_stripe_count = le32_to_cpu(lmu->lum_stripe_count); + LASSERT(final_stripe_count && + final_stripe_count < lo->ldo_dir_stripe_count); + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (!dto) + continue; + + if (i < final_stripe_count) { + if (final_stripe_count == 1) + continue; + + rc = lod_sub_declare_xattr_set(env, dto, lmv_buf, + XATTR_NAME_LMV, + LU_XATTR_REPLACE, th); + if (rc) + return rc; + + continue; + } + + rc = dt_dir_is_empty(env, dto); + if (rc < 0) + return rc; + + rc = lod_sub_declare_ref_del(env, dto, th); + if (rc) + return rc; + + rc = lod_sub_declare_destroy(env, dto, th); + if (rc) + return rc; + + snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", + PFID(lu_object_fid(&dto->do_lu)), i); + + rc = lod_sub_declare_delete(env, next, + (const struct dt_key *)stripe_name, th); + if (rc) + return rc; + + rc = lod_sub_declare_ref_del(env, next, th); + if (rc) + return rc; + } + + rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, + LU_XATTR_REPLACE, th); + return rc; +} + +/** + * Allocate stripes for split directory. + * + * \param[in] env execution environment + * \param[in] dt target object + * \param[in] mlc layout change data + * \param[in] th transaction handle + * + * \retval 0 on success + * \retval negative if failed + */ +static int lod_dir_declare_layout_split(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev); + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object_format *dof = &info->lti_format; + struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata; + struct dt_object **stripes; + u32 stripe_count; + u32 saved_count; + int i; + int rc; + + ENTRY; + + LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC); + LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT); + + saved_count = lo->ldo_dir_stripes_allocated; + stripe_count = le32_to_cpu(lum->lum_stripe_count); + if (stripe_count <= saved_count) + RETURN(-EINVAL); + + dof->dof_type = DFT_DIR; + + OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count); + if (!stripes) + RETURN(-ENOMEM); + + for (i = 0; i < lo->ldo_dir_stripes_allocated; i++) + stripes[i] = lo->ldo_stripe[i]; + + lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs); + rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count); + if (rc == -EAGAIN) + rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count, + stripe_count); + if (rc < 0) { + OBD_FREE(stripes, sizeof(*stripes) * stripe_count); + RETURN(rc); + } + + LASSERT(rc > saved_count); + OBD_FREE(lo->ldo_stripe, + sizeof(*stripes) * lo->ldo_dir_stripes_allocated); + lo->ldo_stripe = stripes; + lo->ldo_dir_striped = 1; + lo->ldo_dir_stripe_count = rc; + lo->ldo_dir_stripes_allocated = stripe_count; + lo->ldo_dir_split_hash = lo->ldo_dir_hash_type; + lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type); + if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type)) + lo->ldo_dir_hash_type = + lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern; + lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION; + lo->ldo_dir_split_offset = saved_count; + lo->ldo_dir_layout_version++; + lo->ldo_dir_stripe_loaded = 1; + + rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th); + if (rc) + lod_striping_free(env, lo); + + RETURN(rc); +} + +/* + * detach all stripes from dir master object, NB, stripes are not destroyed, but + * deleted from it's parent namespace, this function is called in two places: + * 1. mdd_migrate_mdt() detach stripes from source, and attach them to + * target. + * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to + * a plain directory. + * + * \param[in] env execution environment + * \param[in] dt target object + * \param[in] mlc layout change data + * \param[in] th transaction handle + * + * \retval 0 on success + * \retval negative if failed + */ +static int lod_dir_layout_detach(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_object *lo = lod_dt_obj(dt); + struct dt_object *next = dt_object_child(dt); + char *stripe_name = info->lti_key; + struct dt_object *dto; + int i; + int rc = 0; + + ENTRY; + + if (!lo->ldo_dir_stripe_count) { + /* plain directory delete .. */ + rc = lod_sub_delete(env, next, + (const struct dt_key *)dotdot, th); + RETURN(rc); + } + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (!dto) + continue; + + rc = lod_sub_delete(env, dto, + (const struct dt_key *)dotdot, th); + if (rc) + break; + + snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", + PFID(lu_object_fid(&dto->do_lu)), i); + + rc = lod_sub_delete(env, next, + (const struct dt_key *)stripe_name, th); + if (rc) + break; + + rc = lod_sub_ref_del(env, next, th); + if (rc) + break; + } + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (dto) + dt_object_put(env, dto); + } + OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated); + lo->ldo_stripe = NULL; + lo->ldo_dir_stripes_allocated = 0; + lo->ldo_dir_stripe_count = 0; + dt->do_index_ops = &lod_index_ops; + + RETURN(rc); +} + +static int lod_dir_layout_shrink(const struct lu_env *env, + struct dt_object *dt, + const struct md_layout_change *mlc, + struct thandle *th) +{ + struct lod_thread_info *info = lod_env_info(env); + struct lod_object *lo = lod_dt_obj(dt); + struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); + struct dt_object *next = dt_object_child(dt); + struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf; + __u32 final_stripe_count; + char *stripe_name = info->lti_key; + struct dt_object *dto; + struct lu_buf *lmv_buf = &info->lti_buf; + struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1; + u32 mdtidx; + int type = LU_SEQ_RANGE_ANY; + int i; + int rc; + + ENTRY; + + final_stripe_count = le32_to_cpu(lmu->lum_stripe_count); + + lmv_buf->lb_buf = lmv; + lmv_buf->lb_len = sizeof(*lmv); + lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE); + lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count); + lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) & + cpu_to_le32(LMV_HASH_TYPE_MASK); + lmv->lmv_layout_version = + cpu_to_le32(lo->ldo_dir_layout_version + 1); + + for (i = 0; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (!dto) + continue; + + if (i < final_stripe_count) { + /* if only one stripe left, no need to update + * LMV because this stripe will replace master + * object and act as a plain directory. + */ + if (final_stripe_count == 1) + continue; + + + rc = lod_fld_lookup(env, lod, + lu_object_fid(&dto->do_lu), + &mdtidx, &type); + if (rc) + RETURN(rc); + + lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx); + rc = lod_sub_xattr_set(env, dto, lmv_buf, + XATTR_NAME_LMV, + LU_XATTR_REPLACE, th); + if (rc) + RETURN(rc); + + continue; + } + + dt_write_lock(env, dto, DT_TGT_CHILD); + rc = lod_sub_ref_del(env, dto, th); + dt_write_unlock(env, dto); + if (rc) + RETURN(rc); + + rc = lod_sub_destroy(env, dto, th); + if (rc) + RETURN(rc); + + snprintf(stripe_name, sizeof(info->lti_key), DFID":%d", + PFID(lu_object_fid(&dto->do_lu)), i); + + rc = lod_sub_delete(env, next, + (const struct dt_key *)stripe_name, th); + if (rc) + RETURN(rc); + + rc = lod_sub_ref_del(env, next, th); + if (rc) + RETURN(rc); + } + + rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx, + &type); + if (rc) + RETURN(rc); + + lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1); + lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx); + rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, + LU_XATTR_REPLACE, th); + if (rc) + RETURN(rc); + + for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) { + dto = lo->ldo_stripe[i]; + if (dto) + dt_object_put(env, dto); + } + lo->ldo_dir_stripe_count = final_stripe_count; + + RETURN(rc); +} + +static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = { + [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach, + [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach, + [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink, + [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split, +}; + +static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = { + [MD_LAYOUT_DETACH] = lod_dir_layout_detach, + [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink, +}; + static int lod_declare_layout_change(const struct lu_env *env, struct dt_object *dt, struct md_layout_change *mlc, struct thandle *th) @@ -7594,8 +8287,15 @@ static int lod_declare_layout_change(const struct lu_env *env, struct lod_thread_info *info = lod_env_info(env); struct lod_object *lo = lod_dt_obj(dt); int rc; + ENTRY; + if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) { + LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]); + rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th); + RETURN(rc); + } + if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) || dt_object_remote(dt_object_child(dt))) RETURN(-EINVAL); @@ -7643,13 +8343,21 @@ static int lod_layout_change(const struct lu_env *env, struct dt_object *dt, struct lod_object *lo = lod_dt_obj(dt); int rc; + ENTRY; + + if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) { + LASSERT(dir_mlc_ops[mlc->mlc_opc]); + rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th); + RETURN(rc); + } + rc = lod_striped_create(env, dt, attr, NULL, th); if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) { layout_attr->la_layout_version |= lo->ldo_layout_gen; rc = lod_attr_set(env, dt, layout_attr, th); } - return rc; + RETURN(rc); } struct dt_object_operations lod_obj_ops = { @@ -7786,16 +8494,8 @@ 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) { @@ -7812,7 +8512,7 @@ static int lod_object_init(const struct lu_env *env, struct lu_object *lo, if (ltd != NULL) { if (ltd->ltd_tgts_size > idx && - cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) { + test_bit(idx, ltd->ltd_tgt_bitmap)) { tgt = LTD_TGT(ltd, idx); LASSERT(tgt != NULL); @@ -7936,13 +8636,11 @@ void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo) lu_object_put(env, &lod_comp->llc_stripe[j]->do_lu); } - OBD_FREE(lod_comp->llc_stripe, - sizeof(struct dt_object *) * - lod_comp->llc_stripes_allocated); + OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe, + lod_comp->llc_stripes_allocated); lod_comp->llc_stripe = NULL; - OBD_FREE(lod_comp->llc_ost_indices, - sizeof(__u32) * - lod_comp->llc_stripes_allocated); + OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices, + lod_comp->llc_stripes_allocated); lod_comp->llc_ost_indices = NULL; lod_comp->llc_stripes_allocated = 0; } @@ -7971,6 +8669,7 @@ static void lod_object_free(const struct lu_env *env, struct lu_object *o) /* release all underlying object pinned */ lod_striping_free(env, lo); lu_object_fini(o); + /* lo doesn't contain a lu_object_header, so we don't need call_rcu */ OBD_SLAB_FREE_PTR(lo, lod_object_kmem); }