list_del(&tgt_desc->ltd_kill);
if (ltd == &lod->lod_ost_descs) {
/* remove from QoS structures */
- rc = qos_del_tgt(lod, tgt_desc);
+ rc = lqos_del_tgt(&lod->lod_qos, tgt_desc);
if (rc)
CERROR("%s: qos_del_tgt(%s) failed:"
"rc = %d\n",
struct obd_uuid obd_uuid;
bool for_ost;
bool lock = false;
+ bool connected = false;
ENTRY;
CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
obd->obd_name, osp, rc);
GOTO(out_cleanup, rc);
}
+ connected = true;
/* Allocate ost descriptor and fill it */
OBD_ALLOC_PTR(tgt_desc);
if (!tgt_desc)
- GOTO(out_conn, rc = -ENOMEM);
+ GOTO(out_cleanup, rc = -ENOMEM);
tgt_desc->ltd_tgt = dt_dev;
tgt_desc->ltd_exp = exp;
GOTO(out_mutex, rc);
}
- rc = qos_add_tgt(lod, tgt_desc);
+ rc = lqos_add_tgt(&lod->lod_qos, tgt_desc);
if (rc) {
CERROR("%s: qos_add_tgt failed with %d\n",
obd->obd_name, rc);
}
out_desc:
OBD_FREE_PTR(tgt_desc);
-out_conn:
- obd_disconnect(exp);
out_cleanup:
/* XXX OSP needs us to send down LCFG_CLEANUP because it uses
* objects from the MDT stack. See LU-7184. */
lcfg->lcfg_command = LCFG_CLEANUP;
lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
+ if (connected)
+ obd_disconnect(exp);
+
return rc;
}
{
__u32 round = size_roundup_power2(size);
- LASSERT(round <=
- lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
if (info->lti_ea_store) {
LASSERT(info->lti_ea_store_size);
LASSERT(info->lti_ea_store_size < round);
objs = &v3->lmm_objects[0];
}
stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
+ if (stripe_count == 0 && !is_dir &&
+ !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
+ !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
+ RETURN(-E2BIG);
+
if (!is_dir && lo->ldo_is_composite)
lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
PFID(&info->lti_fid), rc);
RETURN(rc);
}
- } else if (lod_comp->llc_ostlist.op_array) {
+ } else if (lod_comp->llc_ostlist.op_array &&
+ lod_comp->llc_ostlist.op_count) {
/* user specified ost list */
ost_idx = lod_comp->llc_ostlist.op_array[i];
}
struct lov_comp_md_v1 *lcm;
struct lod_layout_component *comp_entries;
__u16 comp_cnt, mirror_cnt;
- bool is_composite;
+ bool is_composite, is_foreign = false;
int i, rc = 0, offset;
ENTRY;
mirror_cnt = lo->ldo_mirror_count;
comp_entries = lo->ldo_comp_entries;
is_composite = lo->ldo_is_composite;
+ is_foreign = lo->ldo_is_foreign;
}
LASSERT(lmm_size != NULL);
+
+ if (is_foreign) {
+ struct lov_foreign_md *lfm;
+
+ lfm = (struct lov_foreign_md *)lmm;
+ memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
+ /* need to store little-endian */
+ if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
+ __swab32s(&lfm->lfm_magic);
+ __swab32s(&lfm->lfm_length);
+ __swab32s(&lfm->lfm_type);
+ __swab32s(&lfm->lfm_flags);
+ }
+ *lmm_size = lo->ldo_foreign_lov_size;
+ RETURN(0);
+ }
+
LASSERT(comp_cnt != 0 && comp_entries != NULL);
if (!is_composite) {
/* component could be un-inistantiated */
lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
+ if (lod_comp->llc_flags & LCME_FL_NOSYNC)
+ lcme->lcme_timestamp =
+ cpu_to_le64(lod_comp->llc_timestamp);
+ if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
+ lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
+
lcme->lcme_extent.e_start =
cpu_to_le64(lod_comp->llc_extent.e_start);
lcme->lcme_extent.e_end =
int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
struct lov_ost_data_v1 *objs, int comp_idx)
{
- struct lod_layout_component *lod_comp;
- struct lod_thread_info *info = lod_env_info(env);
- struct lod_device *md;
- struct lu_object *o, *n;
- struct lu_device *nd;
- struct dt_object **stripe;
- int stripe_len;
- int i, rc = 0;
- __u32 idx;
+ struct lod_layout_component *lod_comp;
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_device *md;
+ struct lu_object *o, *n;
+ struct lu_device *nd;
+ struct dt_object **stripe = NULL;
+ __u32 *ost_indices = NULL;
+ int stripe_len;
+ int i, rc = 0;
+ __u32 idx;
ENTRY;
LASSERT(lo != NULL);
OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
if (stripe == NULL)
RETURN(-ENOMEM);
+ OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
+ if (!ost_indices)
+ GOTO(out, rc = -ENOMEM);
for (i = 0; i < lod_comp->llc_stripe_count; i++) {
if (unlikely(lovea_slot_is_dummy(&objs[i])))
LASSERT(n);
stripe[i] = container_of(n, struct dt_object, do_lu);
+ ost_indices[i] = idx;
}
out:
OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
lod_comp->llc_stripe_count = 0;
+ if (ost_indices)
+ OBD_FREE(ost_indices,
+ sizeof(*ost_indices) * stripe_len);
} else {
lod_comp->llc_stripe = stripe;
+ lod_comp->llc_ost_indices = ost_indices;
lod_comp->llc_stripes_allocated = stripe_len;
}
{
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;
LASSERT(buf);
LASSERT(buf->lb_buf);
LASSERT(buf->lb_len);
+ LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
magic = le32_to_cpu(lmm->lmm_magic);
if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
- magic != LOV_MAGIC_COMP_V1)
+ magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
+ magic != LOV_MAGIC_SEL)
GOTO(out, rc = -EINVAL);
- lod_free_comp_entries(lo);
+ if (lo->ldo_is_foreign)
+ lod_free_foreign_lov(lo);
+ else
+ lod_free_comp_entries(lo);
- if (magic == LOV_MAGIC_COMP_V1) {
+ if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
comp_v1 = (struct lov_comp_md_v1 *)lmm;
comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
if (comp_cnt == 0)
lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
LCM_FL_FLR_MASK;
mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
+ } else if (magic == LOV_MAGIC_FOREIGN) {
+ size_t length;
+
+ foreign = (struct lov_foreign_md *)buf->lb_buf;
+ length = offsetof(typeof(*foreign), lfm_value);
+ if (buf->lb_len < length ||
+ buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
+ CDEBUG(D_LAYOUT,
+ "buf len %zu too small for lov_foreign_md\n",
+ buf->lb_len);
+ GOTO(out, rc = -EINVAL);
+ }
+
+ /* just cache foreign LOV EA raw */
+ rc = lod_alloc_foreign_lov(lo, length);
+ if (rc)
+ GOTO(out, rc);
+ memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
+ GOTO(out, rc);
} else {
comp_cnt = 1;
lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
lod_comp->llc_flags =
le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
+ if (lod_comp->llc_flags & LCME_FL_NOSYNC)
+ lod_comp->llc_timestamp = le64_to_cpu(
+ comp_v1->lcm_entries[i].lcme_timestamp);
lod_comp->llc_id =
le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
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) {
+ 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);
+ }
} else {
lod_comp_set_init(lod_comp);
}
pattern = le32_to_cpu(lmm->lmm_pattern);
- if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
- lov_pattern(pattern) != LOV_PATTERN_MDT)
+ if (!lov_pattern_supported(lov_pattern(pattern)))
GOTO(out, rc = -EINVAL);
lod_comp->llc_pattern = pattern;
if (objs[0].l_ost_idx != (__u32)-1UL) {
stripe_count = lod_comp_entry_stripe_count(
lo, lod_comp, false);
+ if (stripe_count == 0 &&
+ !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
+ !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
+ GOTO(out, rc = -E2BIG);
/**
* load the user specified ost list, when this
* component is instantiated later, it will be
out:
if (rc)
- lod_object_free_striping(env, lo);
+ lod_striping_free_nolock(env, lo);
RETURN(rc);
}
return true;
if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
- if (lo->ldo_stripe != NULL || lo->ldo_dir_stripe_loaded)
+ if (lo->ldo_dir_stripe_loaded)
return true;
/* Never load LMV stripe for slaves of striped dir */
}
/**
- * Initialize the object representing the stripes.
+ * A generic function to initialize the stripe objects.
*
- * Unless the stripes are initialized already, fetch LOV (for regular
- * objects) or LMV (for directory objects) EA and call lod_parse_striping()
- * to instantiate the objects representing the stripes. Caller should
- * hold the dt_write_lock(next).
+ * A protected version of lod_striping_load_locked() - load the striping
+ * information from storage, parse that and instantiate LU objects to
+ * represent the stripes. The LOD object \a lo supplies a pointer to the
+ * next sub-object in the LU stack so we can lock it. Also use \a lo to
+ * return an array of references to the newly instantiated objects.
*
* \param[in] env execution environment for this thread
- * \param[in,out] lo LOD object
+ * \param[in,out] lo LOD object, where striping is stored and
+ * which gets an array of references
*
* \retval 0 if parsing and object creation succeed
* \retval negative error number on failure
- */
-int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
+ **/
+int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
{
- struct lod_thread_info *info = lod_env_info(env);
- struct lu_buf *buf = &info->lti_buf;
- struct dt_object *next = dt_object_child(&lo->ldo_obj);
- int rc = 0;
+ struct lod_thread_info *info = lod_env_info(env);
+ struct dt_object *next = dt_object_child(&lo->ldo_obj);
+ struct lu_buf *buf = &info->lti_buf;
+ int rc = 0;
+
ENTRY;
if (!dt_object_exists(next))
- GOTO(out, rc = 0);
+ RETURN(0);
if (lod_striping_loaded(lo))
- GOTO(out, rc = 0);
+ RETURN(0);
+
+ mutex_lock(&lo->ldo_layout_mutex);
+ if (lod_striping_loaded(lo))
+ GOTO(unlock, rc = 0);
if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
rc = lod_get_lov_ea(env, lo);
if (rc <= 0)
- GOTO(out, rc);
+ GOTO(unlock, rc);
+
/*
* there is LOV EA (striping information) in this object
* let's parse it and create in-core objects for the stripes
lo->ldo_comp_cached = 1;
} else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
rc = lod_get_lmv_ea(env, lo);
- if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
+ if (rc > sizeof(struct lmv_foreign_md)) {
+ struct lmv_foreign_md *lfm = info->lti_ea_store;
+
+ if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
+ lo->ldo_foreign_lmv = info->lti_ea_store;
+ lo->ldo_foreign_lmv_size =
+ info->lti_ea_store_size;
+ info->lti_ea_store = NULL;
+ info->lti_ea_store_size = 0;
+
+ lo->ldo_dir_stripe_loaded = 1;
+ lo->ldo_dir_is_foreign = 1;
+ GOTO(unlock, rc = 0);
+ }
+ }
+
+ if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
/* Let's set stripe_loaded to avoid further
* stripe loading especially for non-stripe directory,
* which can hurt performance. (See LU-9840)
*/
if (rc == 0)
lo->ldo_dir_stripe_loaded = 1;
- GOTO(out, rc = rc > 0 ? -EINVAL : rc);
+ GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
}
buf->lb_buf = info->lti_ea_store;
buf->lb_len = info->lti_ea_store_size;
}
if (rc < 0)
- GOTO(out, rc);
+ GOTO(unlock, rc);
}
/*
if (rc == 0)
lo->ldo_dir_stripe_loaded = 1;
}
-out:
- RETURN(rc);
+ EXIT;
+unlock:
+ mutex_unlock(&lo->ldo_layout_mutex);
+
+ return rc;
}
-/**
- * A generic function to initialize the stripe objects.
- *
- * A protected version of lod_load_striping_locked() - load the striping
- * information from storage, parse that and instantiate LU objects to
- * represent the stripes. The LOD object \a lo supplies a pointer to the
- * next sub-object in the LU stack so we can lock it. Also use \a lo to
- * return an array of references to the newly instantiated objects.
- *
- * \param[in] env execution environment for this thread
- * \param[in,out] lo LOD object, where striping is stored and
- * which gets an array of references
- *
- * \retval 0 if parsing and object creation succeed
- * \retval negative error number on failure
- **/
-int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
+int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
+ const struct lu_buf *buf)
{
- struct dt_object *next = dt_object_child(&lo->ldo_obj);
- int rc;
+ int rc;
- if (!dt_object_exists(next))
- return 0;
+ ENTRY;
- /* Check without locking first */
- if (lod_striping_loaded(lo))
- return 0;
+ mutex_lock(&lo->ldo_layout_mutex);
+ lod_striping_free_nolock(env, lo);
+ rc = lod_parse_striping(env, lo, buf);
+ mutex_unlock(&lo->ldo_layout_mutex);
- /* currently this code is supposed to be called from declaration
- * phase only, thus the object is not expected to be locked by caller */
- dt_write_lock(env, next, 0);
- rc = lod_load_striping_locked(env, lo);
- dt_write_unlock(env, next);
- return rc;
+ RETURN(rc);
}
/**
int rc = 0;
ENTRY;
+ if (buf->lb_len < sizeof(lum->lmm_magic)) {
+ CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
+ RETURN(-EINVAL);
+ }
+
lum = buf->lb_buf;
+ magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
+ /* treat foreign LOV EA/object case first
+ * XXX is it expected to try setting again a foreign?
+ * XXX should we care about different current vs new layouts ?
+ */
+ if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
+ struct lov_foreign_md *lfm = buf->lb_buf;
+
+ if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
+ CDEBUG(D_LAYOUT,
+ "buf len %zu < min lov_foreign_md size (%zu)\n",
+ buf->lb_len, offsetof(typeof(*lfm),
+ lfm_value));
+ RETURN(-EINVAL);
+ }
+
+ if (foreign_size_le(lfm) > buf->lb_len) {
+ CDEBUG(D_LAYOUT,
+ "buf len %zu < this lov_foreign_md size (%zu)\n",
+ buf->lb_len, foreign_size_le(lfm));
+ RETURN(-EINVAL);
+ }
+ /* Don't do anything with foreign layouts */
+ RETURN(0);
+ }
+
+ /* normal LOV/layout cases */
+
if (buf->lb_len < sizeof(*lum)) {
CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
buf->lb_len);
RETURN(-EINVAL);
}
- magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
if (magic != LOV_USER_MAGIC_V1 &&
magic != LOV_USER_MAGIC_V3 &&
magic != LOV_USER_MAGIC_SPECIFIC &&
for_each_comp_entry_v1(comp_v1, ent) {
ext = &ent->lcme_extent;
- if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
+ if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end)) {
CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
le64_to_cpu(ext->e_start),
le64_to_cpu(ext->e_end));
stripe_size = le32_to_cpu(lum->lmm_stripe_size);
if (stripe_size == 0)
stripe_size = desc->ld_default_stripe_size;
- if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
+ if (prev_end % stripe_size) {
CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
"stripe_sz: %u, [%llu, %llu)\n",
stripe_size, ext->e_start, prev_end);
void lod_fix_desc_pattern(__u32 *val)
{
/* from lov_setstripe */
- if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
- (*val != LOV_PATTERN_MDT)) {
- LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
+ if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
+ LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
*val = 0;
}
}
lod->lod_sp_me = LUSTRE_SP_CLI;
/* Set up allocation policy (QoS and RR) */
- INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
+ 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_rr.lqr_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;
+ lu_qos_rr_init(&lod->lod_qos.lq_rr);
+
/* Set up OST pool environment */
lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
HASH_POOLS_MAX_BITS,
rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
if (rc)
GOTO(out_hash, rc);
- lod_qos_rr_init(&lod->lod_qos.lq_rr);
rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
if (rc)
GOTO(out_pool_info, rc);