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;
}
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;
}
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];
}
/* 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);
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;
}
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);
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)
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
*/
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);
}
/**
RETURN(rc);
}
+static inline
+struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
+{
+ LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
+ LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
+ le32_to_cpu(comp->lcm_magic));
+ LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
+ "bad index %d, max = %d\n",
+ i, le16_to_cpu(comp->lcm_entry_count));
+
+ return &comp->lcm_entries[i];
+}
+
+#define for_each_comp_entry_v1(comp, entry) \
+ for (entry = comp_entry_v1(comp, 0); \
+ entry <= comp_entry_v1(comp, \
+ le16_to_cpu(comp->lcm_entry_count) - 1); \
+ entry++)
+
+int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
+{
+ struct lov_comp_md_entry_v1 *ent, *dom_ent;
+ __u16 entries;
+ __u32 dom_off, dom_size, comp_size;
+ void *blob_src, *blob_dst;
+ unsigned int blob_size, blob_shift;
+
+ entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
+ /* if file has only DoM stripe return just error */
+ if (entries == 0)
+ return -EFBIG;
+
+ comp_size = le32_to_cpu(comp_v1->lcm_size);
+ dom_ent = &comp_v1->lcm_entries[0];
+ dom_off = le32_to_cpu(dom_ent->lcme_offset);
+ dom_size = le32_to_cpu(dom_ent->lcme_size);
+
+ /* shift entries array first */
+ comp_v1->lcm_entry_count = cpu_to_le16(entries);
+ memmove(dom_ent, dom_ent + 1,
+ entries * sizeof(struct lov_comp_md_entry_v1));
+
+ /* now move blob of layouts */
+ blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
+ blob_src = (void *)comp_v1 + dom_off + dom_size;
+ blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
+ blob_shift = sizeof(*dom_ent) + dom_size;
+
+ memmove(blob_dst, blob_src, blob_size);
+
+ for_each_comp_entry_v1(comp_v1, ent) {
+ __u32 off;
+
+ off = le32_to_cpu(ent->lcme_offset);
+ ent->lcme_offset = cpu_to_le32(off - blob_shift);
+ }
+
+ comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
+
+ /* notify a caller to re-check entry */
+ return -ERESTART;
+}
+
+int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
+{
+ struct lov_comp_md_entry_v1 *ent, *dom_ent;
+ struct lu_extent *dom_ext, *ext;
+ struct lov_user_md_v1 *lum;
+ __u32 stripe_size;
+ __u16 mid, dom_mid;
+ int rc = 0;
+
+ dom_ent = &comp_v1->lcm_entries[0];
+ dom_ext = &dom_ent->lcme_extent;
+ dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
+ stripe_size = d->lod_dom_max_stripesize;
+
+ lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
+ CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
+ "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
+ d->lod_dom_max_stripesize, stripe_size);
+ lum->lmm_stripe_size = cpu_to_le32(stripe_size);
+
+ for_each_comp_entry_v1(comp_v1, ent) {
+ if (ent == dom_ent)
+ continue;
+
+ mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
+ if (mid != dom_mid)
+ continue;
+
+ ext = &ent->lcme_extent;
+ if (ext->e_start != dom_ext->e_end)
+ continue;
+
+ /* Found next component after the DoM one with the same
+ * mirror_id and adjust its start with DoM component end.
+ *
+ * NOTE: we are considering here that there can be only one
+ * DoM component in a file, all replicas are located on OSTs
+ * always and don't need adjustment since use own layouts.
+ */
+ ext->e_start = cpu_to_le64(stripe_size);
+ break;
+ }
+
+ if (stripe_size == 0) {
+ /* DoM component size is zero due to server setting,
+ * remove it from the layout */
+ rc = lod_erase_dom_stripe(comp_v1);
+ } else {
+ /* Update DoM extent end finally */
+ dom_ext->e_end = cpu_to_le64(stripe_size);
+ }
+
+ return rc;
+}
+
/**
* Verify LOV striping.
*
__u64 prev_end = 0;
__u32 stripe_size = 0;
__u16 prev_mid = -1, mirror_id = -1;
- __u32 mirror_count = 0;
+ __u32 mirror_count;
__u32 magic;
- int rc = 0, i;
+ int rc = 0;
ENTRY;
lum = buf->lb_buf;
RETURN(-EINVAL);
}
+recheck:
+ mirror_count = 0;
if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
CDEBUG(D_LAYOUT, "entry count is zero\n");
RETURN(-EINVAL);
++mirror_count;
}
- for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
- ent = &comp_v1->lcm_entries[i];
+ 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)) {
RETURN(-EINVAL);
}
- prev_end = le64_to_cpu(ext->e_end);
-
tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
tmp.lb_len = le32_to_cpu(ent->lcme_size);
lum = tmp.lb_buf;
if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
LOV_PATTERN_MDT) {
- /* DoM component can be only the first entry */
- if (i > 0) {
- CDEBUG(D_LAYOUT, "invalid DoM layout "
- "entry found at %i index\n", i);
+ /* DoM component can be only the first stripe */
+ if (le64_to_cpu(ext->e_start) > 0) {
+ CDEBUG(D_LAYOUT, "invalid DoM component "
+ "with %llu extent start\n",
+ le64_to_cpu(ext->e_start));
RETURN(-EINVAL);
}
stripe_size = le32_to_cpu(lum->lmm_stripe_size);
/* There is just one stripe on MDT and it must
* cover whole component size. */
- if (stripe_size != prev_end) {
+ if (stripe_size != le64_to_cpu(ext->e_end)) {
CDEBUG(D_LAYOUT, "invalid DoM layout "
"stripe size %u != %llu "
"(component size)\n",
"%u is bigger than MDT limit %u, check "
"dom_max_stripesize parameter\n",
stripe_size, d->lod_dom_max_stripesize);
- RETURN(-EINVAL);
+ rc = lod_fix_dom_stripe(d, comp_v1);
+ if (rc == -ERESTART) {
+ /* DoM entry was removed, re-check
+ * new layout from start */
+ goto recheck;
+ } else if (rc) {
+ RETURN(rc);
+ }
}
}
+ prev_end = le64_to_cpu(ext->e_end);
+
rc = lod_verify_v1v3(d, &tmp, is_from_disk);
if (rc)
RETURN(rc);