ltd->ltd_refcount--;
if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
struct lod_tgt_desc *tgt_desc, *tmp;
- struct list_head kill;
- unsigned int idx;
+ LIST_HEAD(kill);
CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
ltd->ltd_death_row);
- INIT_LIST_HEAD(&kill);
-
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
- tgt_desc = LTD_TGT(ltd, idx);
+ ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
LASSERT(tgt_desc);
-
if (!tgt_desc->ltd_reap)
continue;
list_add(&tgt_desc->ltd_kill, &kill);
- LTD_TGT(ltd, idx) = NULL;
- /*FIXME: only support ost pool for now */
- if (ltd == &lod->lod_ost_descs) {
- lod_ost_pool_remove(&lod->lod_pool_info, idx);
- if (tgt_desc->ltd_active)
- lod->lod_desc.ld_active_tgt_count--;
- }
- ltd->ltd_tgtnr--;
- cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
+ tgt_pool_remove(<d->ltd_tgt_pool,
+ tgt_desc->ltd_index);
+ ltd_del_tgt(ltd, tgt_desc);
ltd->ltd_death_row--;
}
mutex_unlock(<d->ltd_mutex);
list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
int rc;
+
list_del(&tgt_desc->ltd_kill);
- if (ltd == &lod->lod_ost_descs) {
- /* remove from QoS structures */
- rc = qos_del_tgt(lod, tgt_desc);
- if (rc)
- CERROR("%s: qos_del_tgt(%s) failed:"
- "rc = %d\n",
- lod2obd(lod)->obd_name,
- obd_uuid2str(&tgt_desc->ltd_uuid),
- rc);
- }
rc = obd_disconnect(tgt_desc->ltd_exp);
if (rc)
CERROR("%s: failed to disconnect %s: rc = %d\n",
}
/**
- * Expand size of target table.
- *
- * When the target table is full, we have to extend the table. To do so,
- * we allocate new memory with some reserve, move data from the old table
- * to the new one and release memory consumed by the old table.
- * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
- *
- * \param[in] ltd target table
- * \param[in] newsize new size of the table
- *
- * \retval 0 on success
- * \retval -ENOMEM if reallocation failed
- */
-static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
-{
- struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
- int rc = 0;
- ENTRY;
-
- /* grab write reference on the lod. Relocating the array requires
- * exclusive access */
-
- down_write(<d->ltd_rw_sem);
- if (newsize <= ltd->ltd_tgts_size)
- /* someone else has already resize the array */
- GOTO(out, rc = 0);
-
- /* allocate new bitmap */
- new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
- if (!new_bitmap)
- GOTO(out, rc = -ENOMEM);
-
- if (ltd->ltd_tgts_size > 0) {
- /* the bitmap already exists, we need
- * to copy data from old one */
- cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
- old_bitmap = ltd->ltd_tgt_bitmap;
- }
-
- ltd->ltd_tgts_size = newsize;
- ltd->ltd_tgt_bitmap = new_bitmap;
-
- if (old_bitmap)
- CFS_FREE_BITMAP(old_bitmap);
-
- CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
-
- EXIT;
-out:
- up_write(<d->ltd_rw_sem);
- return rc;
-}
-
-/**
* Connect LOD to a new OSP and add it to the target table.
*
* Connect to the OSP device passed, initialize all the internal
struct lustre_cfg *lcfg;
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;
tgt_desc->ltd_index = index;
tgt_desc->ltd_active = active;
- lod_getref(ltd);
- if (index >= ltd->ltd_tgts_size) {
- /* we have to increase the size of the lod_osts array */
- __u32 newsize;
-
- newsize = max(ltd->ltd_tgts_size, (__u32)2);
- while (newsize < index + 1)
- newsize = newsize << 1;
-
- /* lod_bitmap_resize() needs lod_rw_sem
- * which we hold with th reference */
- lod_putref(lod, ltd);
-
- rc = ltd_bitmap_resize(ltd, newsize);
- if (rc)
- GOTO(out_desc, rc);
-
- lod_getref(ltd);
- }
-
+ down_write(<d->ltd_rw_sem);
mutex_lock(<d->ltd_mutex);
- lock = true;
- if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
- CERROR("%s: device %d is registered already\n", obd->obd_name,
- index);
- GOTO(out_mutex, rc = -EEXIST);
- }
-
- if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
- OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
- if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
- CERROR("can't allocate index to add %s\n",
- obd->obd_name);
- GOTO(out_mutex, rc = -ENOMEM);
- }
- }
-
- if (for_ost) {
- /* pool and qos are not supported for MDS stack yet */
- rc = lod_ost_pool_add(&lod->lod_pool_info, index,
- lod->lod_osts_size);
- if (rc) {
- CERROR("%s: can't set up pool, failed with %d\n",
- obd->obd_name, rc);
- GOTO(out_mutex, rc);
- }
+ rc = ltd_add_tgt(ltd, tgt_desc);
+ if (rc)
+ GOTO(out_mutex, rc);
- rc = qos_add_tgt(lod, tgt_desc);
- if (rc) {
- CERROR("%s: qos_add_tgt failed with %d\n",
- obd->obd_name, rc);
- GOTO(out_pool, rc);
- }
+ rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
+ if (rc)
+ GOTO(out_del_tgt, rc);
- /* The new OST is now a full citizen */
- if (index >= lod->lod_desc.ld_tgt_count)
- lod->lod_desc.ld_tgt_count = index + 1;
- if (active)
- lod->lod_desc.ld_active_tgt_count++;
+ rc = tgt_pool_add(<d->ltd_tgt_pool, index,
+ ltd->ltd_lov_desc.ld_tgt_count);
+ if (rc) {
+ CERROR("%s: can't set up pool, failed with %d\n",
+ obd->obd_name, rc);
+ GOTO(out_del_tgt, rc);
}
- LTD_TGT(ltd, index) = tgt_desc;
- cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
- ltd->ltd_tgtnr++;
mutex_unlock(<d->ltd_mutex);
- lod_putref(lod, ltd);
- lock = false;
+ up_write(<d->ltd_rw_sem);
+
if (lod->lod_recovery_completed)
lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
RETURN(rc);
out_fini_llog:
lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
- tgt_desc->ltd_recovery_thread);
+ &tgt_desc->ltd_recovery_task);
out_ltd:
- lod_getref(ltd);
+ down_write(<d->ltd_rw_sem);
mutex_lock(<d->ltd_mutex);
- lock = true;
- if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
- struct ptlrpc_thread *thread;
-
- thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
- OBD_FREE_PTR(thread);
- }
- ltd->ltd_tgtnr--;
- cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
- LTD_TGT(ltd, index) = NULL;
-out_pool:
- lod_ost_pool_remove(&lod->lod_pool_info, index);
+ tgt_pool_remove(<d->ltd_tgt_pool, index);
+out_del_tgt:
+ ltd_del_tgt(ltd, tgt_desc);
out_mutex:
- if (lock) {
- mutex_unlock(<d->ltd_mutex);
- lod_putref(lod, ltd);
- }
-out_desc:
+ mutex_unlock(<d->ltd_mutex);
+ up_write(<d->ltd_rw_sem);
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;
}
* \param[in] env execution environment for this thread
* \param[in] lod LOD device the target table belongs to
* \param[in] ltd target table
- * \param[in] idx index of the target
- * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
+ * \param[in] tgt target
*/
static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
- struct lod_tgt_descs *ltd, unsigned idx,
- bool for_ost)
+ struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
{
- LASSERT(LTD_TGT(ltd, idx));
-
- lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
- idx, for_ost);
-
- if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
- struct ptlrpc_thread *thread;
+ lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
+ !ltd->ltd_is_mdt);
- thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
- OBD_FREE_PTR(thread);
- }
-
- if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
- LTD_TGT(ltd, idx)->ltd_reap = 1;
+ if (!tgt->ltd_reap) {
+ tgt->ltd_reap = 1;
ltd->ltd_death_row++;
}
}
* \param[in] env execution environment for this thread
* \param[in] lod LOD device the target table belongs to
* \param[in] ltd target table
- * \param[in] for_ost type of the target: MDT or OST
*
* \retval 0 always
*/
int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
- struct lod_tgt_descs *ltd, bool for_ost)
+ struct lod_tgt_descs *ltd)
{
- unsigned int idx;
+ struct lu_tgt_desc *tgt;
if (ltd->ltd_tgts_size <= 0)
return 0;
+
lod_getref(ltd);
mutex_lock(<d->ltd_mutex);
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
- __lod_del_device(env, lod, ltd, idx, for_ost);
+ ltd_foreach_tgt(ltd, tgt)
+ __lod_del_device(env, lod, ltd, tgt);
mutex_unlock(<d->ltd_mutex);
lod_putref(lod, ltd);
- CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
- for (idx = 0; idx < TGT_PTRS; idx++) {
- if (ltd->ltd_tgt_idx[idx])
- OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
- }
- ltd->ltd_tgts_size = 0;
+
+ lu_tgt_descs_fini(ltd);
+
return 0;
}
* \param[in] osp name of OSP device to be removed
* \param[in] idx index of the target
* \param[in] gen generation number, not used currently
- * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
*
* \retval 0 if the device was scheduled for removal
* \retval -EINVAL if no device was found
*/
int lod_del_device(const struct lu_env *env, struct lod_device *lod,
- struct lod_tgt_descs *ltd, char *osp, unsigned idx,
- unsigned gen, bool for_ost)
+ struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
+ unsigned int gen)
{
struct obd_device *obd;
- int rc = 0;
- struct obd_uuid uuid;
+ struct lu_tgt_desc *tgt;
+ struct obd_uuid uuid;
+ int rc = 0;
+
ENTRY;
CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
lod_getref(ltd);
mutex_lock(<d->ltd_mutex);
+ tgt = LTD_TGT(ltd, idx);
/* check that the index is allocated in the bitmap */
- if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
- !LTD_TGT(ltd, idx)) {
+ if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || !tgt) {
CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
GOTO(out, rc = -EINVAL);
}
/* check that the UUID matches */
- if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
+ if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
- obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
- idx, osp);
+ obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
GOTO(out, rc = -EINVAL);
}
- __lod_del_device(env, lod, ltd, idx, for_ost);
+ __lod_del_device(env, lod, ltd, tgt);
EXIT;
out:
mutex_unlock(<d->ltd_mutex);
{
__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);
void lod_free_comp_entries(struct lod_object *lo)
{
if (lo->ldo_mirrors) {
- OBD_FREE(lo->ldo_mirrors,
- sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
+ OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
lo->ldo_mirrors = NULL;
lo->ldo_mirror_count = 0;
}
LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
if (mirror_count > 0) {
- OBD_ALLOC(lo->ldo_mirrors,
- sizeof(*lo->ldo_mirrors) * mirror_count);
+ OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
if (!lo->ldo_mirrors)
return -ENOMEM;
OBD_ALLOC_LARGE(lo->ldo_comp_entries,
sizeof(*lo->ldo_comp_entries) * comp_count);
if (lo->ldo_comp_entries == NULL) {
- OBD_FREE(lo->ldo_mirrors,
- sizeof(*lo->ldo_mirrors) * mirror_count);
+ OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
lo->ldo_mirror_count = 0;
return -ENOMEM;
}
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);
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) {
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 =
return -EINVAL;
}
- if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
+ if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
CERROR("%s: invalid lod device, for idx: %d\n",
lod2obd(md)->obd_name , idx);
return -EINVAL;
LASSERT(lod_comp->llc_stripe_size > 0);
stripe_len = lod_comp->llc_stripe_count;
- OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
+ OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
if (stripe == NULL)
RETURN(-ENOMEM);
- OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
+ OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
if (!ost_indices)
GOTO(out, rc = -ENOMEM);
GOTO(out, rc);
}
- nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
+ nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
lod_putref(md, &md->lod_ost_descs);
/* In the function below, .hs_keycmp resolves to
if (stripe[i] != NULL)
dt_object_put(env, stripe[i]);
- OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
+ OBD_FREE_PTR_ARRAY(stripe, stripe_len);
lod_comp->llc_stripe_count = 0;
if (ost_indices)
- OBD_FREE(ost_indices,
- sizeof(*ost_indices) * stripe_len);
+ OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
} else {
lod_comp->llc_stripe = stripe;
lod_comp->llc_ost_indices = ost_indices;
int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
const struct lu_buf *buf)
{
- struct lov_mds_md_v1 *lmm;
- struct lov_comp_md_v1 *comp_v1 = NULL;
- struct lov_ost_data_v1 *objs;
- __u32 magic, pattern;
- int i, j, rc = 0;
- __u16 comp_cnt;
- __u16 mirror_cnt = 0;
+ struct lov_mds_md_v1 *lmm;
+ struct lov_comp_md_v1 *comp_v1 = NULL;
+ struct lov_foreign_md *foreign = NULL;
+ struct lov_ost_data_v1 *objs;
+ __u32 magic, pattern;
+ __u16 mirror_cnt = 0;
+ __u16 comp_cnt;
+ int i, rc;
ENTRY;
LASSERT(buf);
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);
GOTO(out, rc);
for (i = 0; i < comp_cnt; i++) {
- struct lod_layout_component *lod_comp;
- struct lu_extent *ext;
- __u32 offs;
+ struct lod_layout_component *lod_comp;
+ struct lu_extent *ext;
+ __u32 offs;
lod_comp = &lo->ldo_comp_entries[i];
if (lo->ldo_is_composite) {
offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
- magic = le32_to_cpu(lmm->lmm_magic);
ext = &comp_v1->lcm_entries[i].lcme_extent;
lod_comp->llc_extent.e_start =
le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
if (lod_comp->llc_id == LCME_ID_INVAL)
GOTO(out, rc = -EINVAL);
+
+ if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
+ comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
+ struct lod_device *d =
+ lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+
+ CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
+ lod2obd(d)->obd_name,
+ lod_comp->llc_flags, lod_comp->llc_id, i,
+ PFID(lod_object_fid(lo)),
+ le32_to_cpu(comp_v1->lcm_magic));
+ }
} else {
lod_comp_set_init(lod_comp);
}
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;
lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
- if (magic == LOV_MAGIC_V3) {
+ if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
+
lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
objs = &v3->lmm_objects[0];
} else {
__u16 stripe_count;
if (objs[0].l_ost_idx != (__u32)-1UL) {
+ int j;
+
stripe_count = lod_comp_entry_stripe_count(
lo, lod_comp, false);
+ if (stripe_count == 0 &&
+ !(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
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 (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
/* if offset is not within valid range [0, osts_size) */
- if (stripe_offset >= d->lod_osts_size) {
+ if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
- stripe_offset, d->lod_osts_size);
+ stripe_offset, d->lod_ost_descs.ltd_tgts_size);
GOTO(out, rc = -EINVAL);
}
le16_to_cpu(comp->lcm_entry_count) - 1); \
entry++)
-int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
+int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
+ struct lov_comp_md_entry_v1 *dom_ent)
{
- struct lov_comp_md_entry_v1 *ent, *dom_ent;
+ struct lov_comp_md_entry_v1 *ent;
__u16 entries;
- __u32 dom_off, dom_size, comp_size;
- void *blob_src, *blob_dst;
- unsigned int blob_size, blob_shift;
+ __u32 dom_off, dom_size, comp_size, off;
+ void *src, *dst;
+ unsigned int size, 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;
+ LASSERT(entries > 0);
+ comp_v1->lcm_entry_count = cpu_to_le16(entries);
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));
+ /* all entries offsets are shifted by entry size at least */
+ shift = sizeof(*dom_ent);
+ for_each_comp_entry_v1(comp_v1, ent) {
+ off = le32_to_cpu(ent->lcme_offset);
+ if (off == dom_off) {
+ /* Entry deletion creates two holes in layout data:
+ * - hole in entries array
+ * - hole in layout data at dom_off with dom_size
+ *
+ * First memmove is one entry shift from next entry
+ * start with size up to dom_off in blob
+ */
+ dst = (void *)ent;
+ src = (void *)(ent + 1);
+ size = (unsigned long)((void *)comp_v1 + dom_off - src);
+ memmove(dst, src, size);
+ /* take 'off' from just moved entry */
+ off = le32_to_cpu(ent->lcme_offset);
+ /* second memmove is blob tail after 'off' up to
+ * component end
+ */
+ dst = (void *)comp_v1 + dom_off - sizeof(*ent);
+ src = (void *)comp_v1 + off;
+ size = (unsigned long)(comp_size - off);
+ memmove(dst, src, size);
+ /* all entries offsets after DoM entry are shifted by
+ * dom_size additionally
+ */
+ shift += dom_size;
+ }
+ ent->lcme_offset = cpu_to_le32(off - shift);
+ }
+ comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
- /* 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;
+ /* notify a caller to re-check entry */
+ return -ERESTART;
+}
+
+void lod_dom_stripesize_recalc(struct lod_device *d)
+{
+ __u64 threshold_mb = d->lod_dom_threshold_free_mb;
+ __u32 max_size = d->lod_dom_stripesize_max_kb;
+ __u32 def_size = d->lod_dom_stripesize_cur_kb;
+
+ /* use maximum allowed value if free space is above threshold */
+ if (d->lod_lsfs_free_mb >= threshold_mb) {
+ def_size = max_size;
+ } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
+ def_size = 0;
+ } else {
+ /* recalc threshold like it would be with def_size as max */
+ threshold_mb = mult_frac(threshold_mb, def_size, max_size);
+ if (d->lod_lsfs_free_mb < threshold_mb)
+ def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
+ else if (d->lod_lsfs_free_mb > threshold_mb * 2)
+ def_size = max_t(unsigned int, def_size * 2,
+ LOD_DOM_MIN_SIZE_KB);
+ }
- memmove(blob_dst, blob_src, blob_size);
+ if (d->lod_dom_stripesize_cur_kb != def_size) {
+ CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
+ d->lod_dom_stripesize_cur_kb, def_size);
+ d->lod_dom_stripesize_cur_kb = def_size;
+ }
+}
- for_each_comp_entry_v1(comp_v1, ent) {
- __u32 off;
+static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
+ struct lod_device *d)
+{
+ int rc;
- off = le32_to_cpu(ent->lcme_offset);
- ent->lcme_offset = cpu_to_le32(off - blob_shift);
+ /* set bfree as fraction of total space */
+ if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
+ spin_lock(&d->lod_lsfs_lock);
+ d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
+ min_t(int, cfs_fail_val, 100), 100);
+ GOTO(recalc, rc = 0);
}
- comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
+ if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
+ struct obd_statfs sfs;
- /* notify a caller to re-check entry */
- return -ERESTART;
+ spin_lock(&d->lod_lsfs_lock);
+ if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
+ GOTO(unlock, rc = 0);
+
+ d->lod_lsfs_age = ktime_get_seconds();
+ spin_unlock(&d->lod_lsfs_lock);
+ rc = dt_statfs(env, d->lod_child, &sfs);
+ if (rc) {
+ CDEBUG(D_LAYOUT,
+ "%s: failed to get OSD statfs: rc = %d\n",
+ lod2obd(d)->obd_name, rc);
+ GOTO(out, rc);
+ }
+ /* udpate local OSD cached statfs data */
+ spin_lock(&d->lod_lsfs_lock);
+ d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
+ d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
+recalc:
+ lod_dom_stripesize_recalc(d);
+unlock:
+ spin_unlock(&d->lod_lsfs_lock);
+ }
+out:
+ return d->lod_dom_stripesize_cur_kb << 10;
}
-int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
+int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
+ struct lov_comp_md_v1 *comp_v1,
+ struct lov_comp_md_entry_v1 *dom_ent,
+ __u32 stripe_size)
{
- struct lov_comp_md_entry_v1 *ent, *dom_ent;
+ struct lov_comp_md_entry_v1 *ent;
struct lu_extent *dom_ext, *ext;
struct lov_user_md_v1 *lum;
- __u32 stripe_size;
+ __u32 max_stripe_size;
__u16 mid, dom_mid;
int rc = 0;
+ bool dom_next_entry = false;
- 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;
+ max_stripe_size = lod_dom_stripesize_limit(env, d);
+
+ /* Check stripe size againts current per-MDT limit */
+ if (stripe_size <= max_stripe_size)
+ return 0;
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);
+ CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
+ stripe_size, max_stripe_size);
+ lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
+ /* In common case the DoM stripe is first entry in a mirror and
+ * can be deleted only if it is not single entry in layout or
+ * mirror, otherwise error should be returned.
+ */
for_each_comp_entry_v1(comp_v1, ent) {
if (ent == dom_ent)
continue;
* 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);
+ ext->e_start = cpu_to_le64(max_stripe_size);
+ dom_next_entry = true;
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);
+ if (max_stripe_size == 0) {
+ /* DoM component size is zero due to server setting, remove
+ * it from the layout but only if next component exists in
+ * the same mirror. That must be checked prior calling the
+ * lod_erase_dom_stripe().
+ */
+ if (!dom_next_entry)
+ return -EFBIG;
+
+ rc = lod_erase_dom_stripe(comp_v1, dom_ent);
} else {
/* Update DoM extent end finally */
- dom_ext->e_end = cpu_to_le64(stripe_size);
+ dom_ext->e_end = cpu_to_le64(max_stripe_size);
}
return rc;
* \retval 0 if the striping is valid
* \retval -EINVAL if striping is invalid
*/
-int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
- const struct lu_buf *buf, bool is_from_disk)
+int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
+ struct lod_object *lo, const struct lu_buf *buf,
+ bool is_from_disk)
{
- struct lov_desc *desc = &d->lod_desc;
struct lov_user_md_v1 *lum;
struct lov_comp_md_v1 *comp_v1;
struct lov_comp_md_entry_v1 *ent;
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));
lum = tmp.lb_buf;
if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
LOV_PATTERN_MDT) {
- /* DoM component can be only the first stripe */
+ /* DoM component must be the first in a mirror */
if (le64_to_cpu(ext->e_start) > 0) {
CDEBUG(D_LAYOUT, "invalid DoM component "
"with %llu extent start\n",
stripe_size, prev_end);
RETURN(-EINVAL);
}
- /* Check stripe size againts per-MDT limit */
- if (stripe_size > d->lod_dom_max_stripesize) {
- CDEBUG(D_LAYOUT, "DoM component size "
- "%u is bigger than MDT limit %u, check "
- "dom_max_stripesize parameter\n",
- stripe_size, d->lod_dom_max_stripesize);
- rc = lod_fix_dom_stripe(d, comp_v1);
- if (rc == -ERESTART) {
- /* DoM entry was removed, re-check
- * new layout from start */
- goto recheck;
- } else if (rc) {
- RETURN(rc);
+ /* Check and adjust stripe size by per-MDT limit */
+ rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
+ stripe_size);
+ /* DoM entry was removed, re-check layout from start */
+ if (rc == -ERESTART)
+ goto recheck;
+ else if (rc)
+ RETURN(rc);
+
+ /* Any stripe count is forbidden on DoM component */
+ if (lum->lmm_stripe_count) {
+ CDEBUG(D_LAYOUT,
+ "invalid DoM layout stripe count %u, must be 0\n",
+ le16_to_cpu(lum->lmm_stripe_count));
+ RETURN(-EINVAL);
+ }
+
+ /* Any pool is forbidden on DoM component */
+ if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
+ struct lov_user_md_v3 *v3 = (void *)lum;
+
+ if (v3->lmm_pool_name[0] != '\0') {
+ CDEBUG(D_LAYOUT,
+ "DoM component cannot have pool assigned\n");
+ RETURN(-EINVAL);
}
}
}
/* extent end must be aligned with the stripe_size */
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 (stripe_size && 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;
+ }
+}
+
+void lod_fix_lmv_desc_pattern(__u32 *val)
+{
+ if ((*val) && !lmv_is_known_hash_type(*val)) {
+ LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
*val = 0;
}
}
lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
}
+static void lod_fix_lmv_desc(struct lmv_desc *desc)
+{
+ desc->ld_active_tgt_count = 0;
+ lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
+ lod_fix_lmv_desc_pattern(&desc->ld_pattern);
+ lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
+}
+
/**
* Initialize the structures used to store pools and default striping.
*
lod_fix_desc(desc);
desc->ld_active_tgt_count = 0;
- lod->lod_desc = *desc;
+ lod->lod_ost_descs.ltd_lov_desc = *desc;
- lod->lod_sp_me = LUSTRE_SP_CLI;
+ /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
+ lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
- /* Set up allocation policy (QoS and RR) */
- INIT_LIST_HEAD(&lod->lod_qos.lq_oss_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;
+ lod->lod_sp_me = LUSTRE_SP_CLI;
/* Set up OST pool environment */
- lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
- HASH_POOLS_MAX_BITS,
- HASH_POOLS_BKT_BITS, 0,
- CFS_HASH_MIN_THETA,
- CFS_HASH_MAX_THETA,
- &pool_hash_operations,
- CFS_HASH_DEFAULT);
- if (lod->lod_pools_hash_body == NULL)
+ lod->lod_pool_count = 0;
+ rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
+ if (rc)
RETURN(-ENOMEM);
INIT_LIST_HEAD(&lod->lod_pool_list);
lod->lod_pool_count = 0;
- rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
+ rc = tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 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);
+
+ rc = tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
+ if (rc)
+ GOTO(out_mdt_pool, rc);
+
+ rc = tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
+ if (rc)
+ GOTO(out_mdt_rr_pool, rc);
+
+ rc = tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
if (rc)
- GOTO(out_pool_info, rc);
+ GOTO(out_ost_pool, rc);
RETURN(0);
-out_pool_info:
- lod_ost_pool_free(&lod->lod_pool_info);
+out_ost_pool:
+ tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
+out_mdt_rr_pool:
+ tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
+out_mdt_pool:
+ tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
out_hash:
- cfs_hash_putref(lod->lod_pools_hash_body);
+ lod_pool_hash_destroy(&lod->lod_pools_hash_body);
return rc;
}
lod_pool_del(obd, pool->pool_name);
}
- cfs_hash_putref(lod->lod_pools_hash_body);
- lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
- lod_ost_pool_free(&lod->lod_pool_info);
+ lod_pool_hash_destroy(&lod->lod_pools_hash_body);
+ tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
+ tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
+ tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
+ tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
RETURN(0);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff