* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* lustre/lod/lod_lov.c
if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
struct lod_tgt_desc *tgt_desc, *tmp;
struct list_head kill;
- unsigned int idx;
CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
ltd->ltd_death_row);
INIT_LIST_HEAD(&kill);
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
- tgt_desc = LTD_TGT(ltd, idx);
+ ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
LASSERT(tgt_desc);
-
if (!tgt_desc->ltd_reap)
continue;
list_add(&tgt_desc->ltd_kill, &kill);
- LTD_TGT(ltd, idx) = NULL;
/*FIXME: only support ost pool for now */
- if (ltd == &lod->lod_ost_descs) {
- lod_ost_pool_remove(&lod->lod_pool_info, idx);
- if (tgt_desc->ltd_active)
- lod->lod_desc.ld_active_tgt_count--;
- }
- ltd->ltd_tgtnr--;
- cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
+ if (ltd == &lod->lod_ost_descs)
+ lod_ost_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);
- }
+ rc = ltd_add_tgt(ltd, tgt_desc);
+ if (rc)
+ GOTO(out_mutex, rc);
- 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);
- }
- }
+ rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
+ if (rc)
+ GOTO(out_del_tgt, rc);
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);
+ /* pool is not supported for MDS stack yet */
+ rc = lod_ost_pool_add(<d->ltd_tgt_pool, index,
+ ltd->ltd_tgts_size);
if (rc) {
CERROR("%s: can't set up pool, failed with %d\n",
obd->obd_name, rc);
- GOTO(out_mutex, rc);
+ GOTO(out_del_tgt, 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);
- }
-
- /* 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++;
}
- 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);
lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
tgt_desc->ltd_recovery_thread);
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);
+ lod_ost_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->ltd_is_mdt && tgt->ltd_recovery_thread)
+ OBD_FREE_PTR(tgt->ltd_recovery_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);
+ lo->ldo_mirrors = NULL;
+ lo->ldo_mirror_count = 0;
+ }
lod_free_comp_buffer(lo->ldo_comp_entries,
lo->ldo_comp_cnt,
sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
lo->ldo_is_composite = 0;
}
-int lod_alloc_comp_entries(struct lod_object *lo, int cnt)
+int lod_alloc_comp_entries(struct lod_object *lo,
+ int mirror_count, int comp_count)
{
- LASSERT(cnt != 0);
+ LASSERT(comp_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);
+ if (!lo->ldo_mirrors)
+ return -ENOMEM;
+
+ lo->ldo_mirror_count = mirror_count;
+ }
+
OBD_ALLOC_LARGE(lo->ldo_comp_entries,
- sizeof(*lo->ldo_comp_entries) * cnt);
- if (lo->ldo_comp_entries == NULL)
+ sizeof(*lo->ldo_comp_entries) * comp_count);
+ if (lo->ldo_comp_entries == NULL) {
+ OBD_FREE(lo->ldo_mirrors,
+ sizeof(*lo->ldo_mirrors) * mirror_count);
+ lo->ldo_mirror_count = 0;
return -ENOMEM;
- lo->ldo_comp_cnt = cnt;
+ }
+
+ lo->ldo_comp_cnt = comp_count;
return 0;
}
+int lod_fill_mirrors(struct lod_object *lo)
+{
+ struct lod_layout_component *lod_comp;
+ int mirror_idx = -1;
+ __u16 mirror_id = 0xffff;
+ int i;
+ ENTRY;
+
+ LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
+
+ if (!lo->ldo_is_composite)
+ RETURN(0);
+
+ lod_comp = &lo->ldo_comp_entries[0];
+ for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
+ int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
+ int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
+
+ if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
+ lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
+ lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
+ lo->ldo_mirrors[mirror_idx].lme_end = i;
+ continue;
+ }
+
+ /* new mirror */
+ ++mirror_idx;
+ if (mirror_idx >= lo->ldo_mirror_count)
+ RETURN(-EINVAL);
+
+ mirror_id = mirror_id_of(lod_comp->llc_id);
+
+ lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
+ lo->ldo_mirrors[mirror_idx].lme_stale = stale;
+ lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
+ lo->ldo_mirrors[mirror_idx].lme_start = i;
+ lo->ldo_mirrors[mirror_idx].lme_end = i;
+ }
+ if (mirror_idx != lo->ldo_mirror_count - 1)
+ RETURN(-EINVAL);
+
+ RETURN(0);
+}
+
/**
* Generate on-disk lov_mds_md structure for each layout component based on
* the information in lod_object->ldo_comp_entries[i].
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];
}
}
/**
- * Generate component ID for new created component.
- *
- * \param[in] lo LOD object
- * \param[in] comp_idx index of ldo_comp_entries
- *
- * \retval component ID on success
- * \retval LCME_ID_INVAL on failure
- */
-static __u32 lod_gen_component_id(struct lod_object *lo, int comp_idx)
-{
- struct lod_layout_component *lod_comp;
- __u32 id, start, end;
- int i;
-
- LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
-
- lod_obj_inc_layout_gen(lo);
- id = lo->ldo_layout_gen;
- if (likely(id <= LCME_ID_MAX))
- return id;
-
- /* Layout generation wraps, need to check collisions. */
- start = id & LCME_ID_MASK;
- end = LCME_ID_MAX;
-again:
- for (id = start; id <= end; id++) {
- for (i = 0; i < lo->ldo_comp_cnt; i++) {
- lod_comp = &lo->ldo_comp_entries[i];
- if (id == lod_comp->llc_id)
- break;
- }
- /* Found the ununsed ID */
- if (i == lo->ldo_comp_cnt)
- return id;
- }
- if (end == LCME_ID_MAX) {
- start = 1;
- end = min(lo->ldo_layout_gen & LCME_ID_MASK,
- (__u32)(LCME_ID_MAX - 1));
- goto again;
- }
-
- return LCME_ID_INVAL;
-}
-
-/**
* Generate on-disk lov_mds_md structure based on the information in
* the lod_object->ldo_comp_entries.
*
struct lov_comp_md_entry_v1 *lcme;
struct lov_comp_md_v1 *lcm;
struct lod_layout_component *comp_entries;
- __u16 comp_cnt;
- bool is_composite;
+ __u16 comp_cnt, mirror_cnt;
+ bool is_composite, is_foreign = false;
int i, rc = 0, offset;
ENTRY;
if (is_dir) {
comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
+ mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
is_composite =
lo->ldo_def_striping->lds_def_striping_is_composite;
} else {
comp_cnt = lo->ldo_comp_cnt;
+ 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) {
}
lcm = (struct lov_comp_md_v1 *)lmm;
+ memset(lcm, 0, sizeof(*lcm));
+
lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
+ lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
+ lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
LASSERT(offset % sizeof(__u64) == 0);
lod_comp = &comp_entries[i];
lcme = &lcm->lcm_entries[i];
- if (lod_comp->llc_id == LCME_ID_INVAL && !is_dir) {
- lod_comp->llc_id = lod_gen_component_id(lo, i);
- if (lod_comp->llc_id == LCME_ID_INVAL)
- GOTO(out, rc = -ERANGE);
- }
+ LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
/* 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 =
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;
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])))
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
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;
}
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;
+ 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);
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)
GOTO(out, rc = -EINVAL);
lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
lo->ldo_is_composite = 1;
+ 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);
lo->ldo_is_composite = 0;
}
- rc = lod_alloc_comp_entries(lo, comp_cnt);
+ rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
if (rc)
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 =
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 ((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];
- /* no need to set pool, which is used in create only */
} else {
+ lod_set_pool(&lod_comp->llc_pool, NULL);
objs = &lmm->lmm_objects[0];
}
* then user has specified ost list for this component.
*/
if (!lod_comp_inited(lod_comp)) {
+ __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
* used in lod_alloc_ost_list().
*/
- lod_comp->llc_ostlist.op_count =
- lod_comp->llc_stripe_count;
+ lod_comp->llc_ostlist.op_count = stripe_count;
lod_comp->llc_ostlist.op_size =
- lod_comp->llc_stripe_count *
- sizeof(__u32);
+ stripe_count * sizeof(__u32);
OBD_ALLOC(lod_comp->llc_ostlist.op_array,
lod_comp->llc_ostlist.op_size);
if (!lod_comp->llc_ostlist.op_array)
GOTO(out, rc = -ENOMEM);
- for (j = 0; j < lod_comp->llc_stripe_count; j++)
+ for (j = 0; j < stripe_count; j++)
lod_comp->llc_ostlist.op_array[j] =
le32_to_cpu(objs[j].l_ost_idx);
GOTO(out, rc);
}
}
+
+ rc = lod_fill_mirrors(lo);
+ if (rc)
+ GOTO(out, rc);
+
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);
}
/**
GOTO(out, rc = -EINVAL);
}
- magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
+ 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) {
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);
}
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 *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;
+
+ 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_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 *dom_ent)
+{
+ struct lov_comp_md_entry_v1 *ent;
+ struct lu_extent *dom_ext, *ext;
+ struct lov_user_md_v1 *lum;
+ __u32 stripe_size;
+ __u16 mid, dom_mid;
+ int rc = 0;
+
+ dom_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, dom_ent);
+ } else {
+ /* Update DoM extent end finally */
+ dom_ext->e_end = cpu_to_le64(stripe_size);
+ }
+
+ return rc;
+}
+
/**
* Verify LOV striping.
*
* \retval 0 if the striping is valid
* \retval -EINVAL if striping is invalid
*/
-int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
- bool is_from_disk, __u64 start)
+int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
+ const struct lu_buf *buf, bool is_from_disk)
{
- struct lov_user_md_v1 *lum;
- struct lov_comp_md_v1 *comp_v1;
- __u32 magic;
- int rc = 0, i;
+ struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
+ struct lov_user_md_v1 *lum;
+ struct lov_comp_md_v1 *comp_v1;
+ struct lov_comp_md_entry_v1 *ent;
+ struct lu_extent *ext;
+ struct lu_buf tmp;
+ __u64 prev_end = 0;
+ __u32 stripe_size = 0;
+ __u16 prev_mid = -1, mirror_id = -1;
+ __u32 mirror_count;
+ __u32 magic;
+ 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_DEF;
if (magic != LOV_USER_MAGIC_V1 &&
magic != LOV_USER_MAGIC_V3 &&
magic != LOV_USER_MAGIC_SPECIFIC &&
RETURN(-EINVAL);
}
- if (magic == LOV_USER_MAGIC_COMP_V1) {
- struct lov_comp_md_entry_v1 *ent;
- struct lu_extent *ext;
- struct lov_desc *desc = &d->lod_desc;
- struct lu_buf tmp;
- __u32 stripe_size = 0;
- __u64 prev_end = start;
-
- comp_v1 = buf->lb_buf;
- if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
- CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
- buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
- RETURN(-EINVAL);
- }
+ if (magic != LOV_USER_MAGIC_COMP_V1)
+ RETURN(lod_verify_v1v3(d, buf, is_from_disk));
+
+ /* magic == LOV_USER_MAGIC_COMP_V1 */
+ comp_v1 = buf->lb_buf;
+ if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
+ CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
+ buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
+ RETURN(-EINVAL);
+ }
- if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
- CDEBUG(D_LAYOUT, "entry count is zero\n");
+recheck:
+ mirror_count = 0;
+ if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
+ CDEBUG(D_LAYOUT, "entry count is zero\n");
+ RETURN(-EINVAL);
+ }
+
+ if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
+ lo->ldo_comp_cnt > 0) {
+ /* could be called from lustre.lov.add */
+ __u32 cnt = lo->ldo_comp_cnt;
+
+ ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
+ prev_end = ext->e_end;
+
+ ++mirror_count;
+ }
+
+ 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)) {
+ CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
+ le64_to_cpu(ext->e_start),
+ le64_to_cpu(ext->e_end));
RETURN(-EINVAL);
}
- for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
- ent = &comp_v1->lcm_entries[i];
- ext = &ent->lcme_extent;
-
- if (is_from_disk &&
- (le32_to_cpu(ent->lcme_id) == 0 ||
- le32_to_cpu(ent->lcme_id) > LCME_ID_MAX)) {
+ if (is_from_disk) {
+ /* lcme_id contains valid value */
+ if (le32_to_cpu(ent->lcme_id) == 0 ||
+ le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
CDEBUG(D_LAYOUT, "invalid id %u\n",
le32_to_cpu(ent->lcme_id));
RETURN(-EINVAL);
}
- if (le64_to_cpu(ext->e_start) >=
- le64_to_cpu(ext->e_end)) {
- CDEBUG(D_LAYOUT, "invalid extent "
- "[%llu, %llu)\n",
- le64_to_cpu(ext->e_start),
- le64_to_cpu(ext->e_end));
- RETURN(-EINVAL);
- }
+ if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
+ mirror_id = mirror_id_of(
+ le32_to_cpu(ent->lcme_id));
- /* first component must start with 0, and the next
- * must be adjacent with the previous one */
- if (le64_to_cpu(ext->e_start) != prev_end) {
- CDEBUG(D_LAYOUT, "invalid start "
- "actual:%llu, expect:%llu\n",
- le64_to_cpu(ext->e_start), prev_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);
-
- /* Check DoM entry is always the first one */
- lum = tmp.lb_buf;
- if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
- LOV_PATTERN_MDT && i > 0) {
- CDEBUG(D_LAYOUT, "invalid DoM layout entry "
- "found at %i index\n", i);
- RETURN(-EINVAL);
+ /* first component must start with 0 */
+ if (mirror_id != prev_mid &&
+ le64_to_cpu(ext->e_start) != 0) {
+ CDEBUG(D_LAYOUT,
+ "invalid start:%llu, expect:0\n",
+ le64_to_cpu(ext->e_start));
+ RETURN(-EINVAL);
+ }
+ prev_mid = mirror_id;
}
- rc = lod_verify_v1v3(d, &tmp, is_from_disk);
- if (rc)
- break;
+ }
- lum = tmp.lb_buf;
+ if (le64_to_cpu(ext->e_start) == 0) {
+ ++mirror_count;
+ prev_end = 0;
+ }
- /* extent end must be aligned with the stripe_size */
+ /* the next must be adjacent with the previous one */
+ if (le64_to_cpu(ext->e_start) != prev_end) {
+ CDEBUG(D_LAYOUT,
+ "invalid start actual:%llu, expect:%llu\n",
+ le64_to_cpu(ext->e_start), prev_end);
+ RETURN(-EINVAL);
+ }
+
+ tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
+ tmp.lb_len = le32_to_cpu(ent->lcme_size);
+
+ /* Check DoM entry is always the first one */
+ lum = tmp.lb_buf;
+ if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
+ LOV_PATTERN_MDT) {
+ /* 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",
+ le64_to_cpu(ext->e_start));
+ RETURN(-EINVAL);
+ }
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 != LUSTRE_EOF &&
- (prev_end & (stripe_size - 1)))) {
- CDEBUG(D_LAYOUT, "stripe size isn't aligned. "
- " stripe_sz: %u, [%llu, %llu)\n",
- stripe_size, ext->e_start, prev_end);
+ /* There is just one stripe on MDT and it must
+ * cover whole component size. */
+ if (stripe_size != le64_to_cpu(ext->e_end)) {
+ CDEBUG(D_LAYOUT, "invalid DoM layout "
+ "stripe size %u != %llu "
+ "(component size)\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, ent);
+ 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);
+
+ if (prev_end == LUSTRE_EOF)
+ continue;
+
+ /* 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 (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);
+ RETURN(-EINVAL);
}
- } else {
- rc = lod_verify_v1v3(d, buf, is_from_disk);
}
- RETURN(rc);
+ /* make sure that the mirror_count is telling the truth */
+ if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
+ RETURN(-EINVAL);
+
+ RETURN(0);
}
/**
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_fix_desc(desc);
desc->ld_active_tgt_count = 0;
- lod->lod_desc = *desc;
+ lod->lod_ost_descs.ltd_lov_desc = *desc;
lod->lod_sp_me = LUSTRE_SP_CLI;
- /* Set up allocation policy (QoS and RR) */
- INIT_LIST_HEAD(&lod->lod_qos.lq_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;
-
/* Set up OST pool environment */
lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
HASH_POOLS_MAX_BITS,
INIT_LIST_HEAD(&lod->lod_pool_list);
lod->lod_pool_count = 0;
- rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
+ rc = lod_ost_pool_init(&lod->lod_ost_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 = lod_ost_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
if (rc)
GOTO(out_pool_info, rc);
RETURN(0);
out_pool_info:
- lod_ost_pool_free(&lod->lod_pool_info);
+ lod_ost_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
out_hash:
cfs_hash_putref(lod->lod_pools_hash_body);
}
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_ost_pool_free(&(lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool));
+ lod_ost_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
RETURN(0);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff