return 0;
}
-/**
- * Check is OST used in a composite layout
- *
- * \param[in] inuse all inuse ost indexs
- * \param[in] ost OST target index to check
- *
- * \retval 0 not used
- * \retval 1 used
- */
-static inline int lod_comp_is_ost_used(struct ost_pool *inuse, int ost)
+static inline bool
+lod_obj_is_ost_use_skip_cb(const struct lu_env *env, struct lod_object *lo,
+ int comp_idx, struct lod_obj_stripe_cb_data *data)
{
- __u32 j;
- LASSERT(inuse != NULL);
+ struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
- if (inuse->op_size == 0)
- return 0;
+ return comp->llc_ost_indices == NULL;
+}
- LASSERT(inuse->op_count * sizeof(inuse->op_array[0]) <= inuse->op_size);
- for (j = 0; j < inuse->op_count; j++) {
- if (inuse->op_array[j] == ost)
- return 1;
+static inline int
+lod_obj_is_ost_use_cb(const struct lu_env *env, struct lod_object *lo,
+ int comp_idx, struct lod_obj_stripe_cb_data *data)
+{
+ struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
+ int i;
+
+ for (i = 0; i < comp->llc_stripe_count; i++) {
+ if (comp->llc_ost_indices[i] == data->locd_ost_index) {
+ data->locd_ost_index = -1;
+ return -EEXIST;
+ }
}
+
return 0;
}
/**
- * Mark the given target as used for a composite layout
+ * Check is OST used in a composite layout
*
- * \param[in] inuse inuse ost index array
- * \param[in] idx index in the array
+ * \param[in] lo lod object
+ * \param[in] ost OST target index to check
+ *
+ * \retval false not used
+ * \retval true used
*/
-static inline void lod_comp_ost_in_use(struct ost_pool *inuse, int ost)
+static inline bool lod_comp_is_ost_used(const struct lu_env *env,
+ struct lod_object *lo, int ost)
{
- LASSERT(inuse != NULL);
- if (inuse->op_size && !lod_comp_is_ost_used(inuse, ost)) {
- LASSERTF(inuse->op_count * sizeof(inuse->op_array[0]) <
- inuse->op_size,
- "count %d size %u", inuse->op_count, inuse->op_size);
- inuse->op_array[inuse->op_count] = ost;
- inuse->op_count++;
- }
+ struct lod_obj_stripe_cb_data data = { { 0 } };
+
+ data.locd_ost_index = ost;
+ data.locd_comp_skip_cb = lod_obj_is_ost_use_skip_cb;
+ data.locd_comp_cb = lod_obj_is_ost_use_cb;
+
+ (void)lod_obj_for_each_stripe(env, lo, NULL, &data);
+
+ return data.locd_ost_index == -1;
}
static inline void lod_avoid_update(struct lod_object *lo,
bool used = false;
int i;
- if (!cfs_bitmap_check(lod->lod_ost_bitmap, index))
+ if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
+ QOS_DEBUG("OST%d: been used in conflicting mirror component\n",
+ index);
return true;
+ }
/**
* we've tried our best, all available OSTs have been used in
if (!cfs_bitmap_check(lag->lag_ost_avoid_bitmap, index))
used = false;
else
- QOS_DEBUG("OST%d: has been used in overlapped component "
- "in other mirror\n", index);
+ QOS_DEBUG("OST%d: been used in conflicting mirror component\n",
+ index);
return used;
}
__u32 speed, __u32 *s_idx,
struct dt_object **stripe,
__u32 *ost_indices,
- struct thandle *th,
- struct ost_pool *inuse)
+ struct thandle *th)
{
struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
struct lod_avoid_guide *lag = &lod_env_info(env)->lti_avoid;
struct dt_object *o;
__u32 stripe_idx = *s_idx;
int rc;
+ ENTRY;
rc = lod_statfs_and_check(env, lod, ost_idx, sfs);
- if (rc) {
- /* this OSP doesn't feel well */
- goto out_return;
- }
+ if (rc)
+ RETURN(rc);
/*
* We expect number of precreated objects in f_ffree at
*/
if (sfs->os_fprecreated == 0 && speed == 0) {
QOS_DEBUG("#%d: precreation is empty\n", ost_idx);
- goto out_return;
+ RETURN(rc);
}
/*
*/
if (sfs->os_state & OS_STATE_DEGRADED && speed < 2) {
QOS_DEBUG("#%d: degraded\n", ost_idx);
- goto out_return;
+ RETURN(rc);
}
/*
* try not allocate on OST which has been used by other
* component
*/
- if (speed == 0 && lod_comp_is_ost_used(inuse, ost_idx)) {
- QOS_DEBUG("#%d: used by other component\n", ost_idx);
- goto out_return;
+ if (speed == 0 && lod_comp_is_ost_used(env, lo, ost_idx)) {
+ QOS_DEBUG("iter %d: OST%d used by other component\n",
+ speed, ost_idx);
+ RETURN(rc);
}
/**
* for the first and second time.
*/
if (speed < 2 && lod_should_avoid_ost(lo, lag, ost_idx)) {
- QOS_DEBUG("#%d: used by overlapped component of other mirror\n",
- ost_idx);
- goto out_return;
+ QOS_DEBUG("iter %d: OST%d used by conflicting mirror "
+ "component\n", speed, ost_idx);
+ RETURN(rc);
}
/*
* do not put >1 objects on a single OST
*/
if (lod_qos_is_ost_used(env, ost_idx, stripe_idx))
- goto out_return;
+ RETURN(rc);
o = lod_qos_declare_object_on(env, lod, ost_idx, th);
if (IS_ERR(o)) {
CDEBUG(D_OTHER, "can't declare new object on #%u: %d\n",
ost_idx, (int) PTR_ERR(o));
rc = PTR_ERR(o);
- goto out_return;
+ RETURN(rc);
}
/*
*/
lod_avoid_update(lo, lag);
lod_qos_ost_in_use(env, stripe_idx, ost_idx);
- lod_comp_ost_in_use(inuse, ost_idx);
stripe[stripe_idx] = o;
ost_indices[stripe_idx] = ost_idx;
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LOV_CREATE_RACE, 2);
stripe_idx++;
*s_idx = stripe_idx;
-out_return:
- return rc;
+ RETURN(rc);
}
/**
* \param[in] flags allocation flags (0 or LOV_USES_DEFAULT_STRIPE)
* \param[in] th transaction handle
* \param[in] comp_idx index of ldo_comp_entries
- * \param[in|out] inuse array of inuse ost index
*
* \retval 0 on success
* \retval -ENOSPC if not enough OSTs are found
*/
static int lod_alloc_rr(const struct lu_env *env, struct lod_object *lo,
struct dt_object **stripe, __u32 *ost_indices,
- int flags, struct thandle *th, int comp_idx,
- struct ost_pool *inuse)
+ int flags, struct thandle *th, int comp_idx)
{
struct lod_layout_component *lod_comp;
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
spin_unlock(&lqr->lqr_alloc);
rc = lod_check_and_reserve_ost(env, lo, sfs, ost_idx, speed,
&stripe_idx, stripe, ost_indices,
- th, inuse);
+ th);
spin_lock(&lqr->lqr_alloc);
if (rc != 0 && OST_TGT(m, ost_idx)->ltd_connecting)
* \param[out] ost_indices ost indices of striping created
* \param[in] th transaction handle
* \param[in] comp_idx index of ldo_comp_entries
- * \param[in|out] inuse array of inuse ost index
*
* \retval 0 on success
* \retval -ENODEV OST index does not exist on file system
*/
static int lod_alloc_ost_list(const struct lu_env *env, struct lod_object *lo,
struct dt_object **stripe, __u32 *ost_indices,
- struct thandle *th, int comp_idx,
- struct ost_pool *inuse)
+ struct thandle *th, int comp_idx)
{
struct lod_layout_component *lod_comp;
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
LASSERT(lo->ldo_comp_cnt > comp_idx && lo->ldo_comp_entries != NULL);
lod_comp = &lo->ldo_comp_entries[comp_idx];
LASSERT(lod_comp->llc_ostlist.op_array);
+ LASSERT(lod_comp->llc_ostlist.op_count);
rc = lod_qos_ost_in_use_clear(env, lod_comp->llc_stripe_count);
if (rc < 0)
RETURN(rc);
+ if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
+ lod_comp->llc_stripe_offset =
+ lod_comp->llc_ostlist.op_array[0];
+
for (i = 0; i < lod_comp->llc_stripe_count; i++) {
if (lod_comp->llc_ostlist.op_array[i] ==
lod_comp->llc_stripe_offset) {
* We've successfully declared (reserved) an object
*/
lod_qos_ost_in_use(env, stripe_count, ost_idx);
- lod_comp_ost_in_use(inuse, ost_idx);
stripe[stripe_count] = o;
ost_indices[stripe_count] = ost_idx;
stripe_count++;
* \param[in] flags not used
* \param[in] th transaction handle
* \param[in] comp_idx index of ldo_comp_entries
- * \param[in|out]inuse array of inuse ost index
*
* \retval 0 on success
* \retval -ENOSPC if no OST objects are available at all
*/
static int lod_alloc_specific(const struct lu_env *env, struct lod_object *lo,
struct dt_object **stripe, __u32 *ost_indices,
- int flags, struct thandle *th, int comp_idx,
- struct ost_pool *inuse)
+ int flags, struct thandle *th, int comp_idx)
{
struct lod_layout_component *lod_comp;
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
* try not allocate on the OST used by other component
*/
if (speed == 0 && i != 0 &&
- lod_comp_is_ost_used(inuse, ost_idx))
+ lod_comp_is_ost_used(env, lo, ost_idx))
continue;
/* Drop slow OSCs if we can, but not for requested start idx.
* We've successfully declared (reserved) an object
*/
lod_qos_ost_in_use(env, stripe_num, ost_idx);
- lod_comp_ost_in_use(inuse, ost_idx);
stripe[stripe_num] = o;
ost_indices[stripe_num] = ost_idx;
stripe_num++;
* \param[in] flags 0 or LOV_USES_DEFAULT_STRIPE
* \param[in] th transaction handle
* \param[in] comp_idx index of ldo_comp_entries
- * \param[in|out]inuse array of inuse ost index
*
* \retval 0 on success
* \retval -EAGAIN not enough OSTs are found for specified stripe count
*/
static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo,
struct dt_object **stripe, __u32 *ost_indices,
- int flags, struct thandle *th, int comp_idx,
- struct ost_pool *inuse)
+ int flags, struct thandle *th, int comp_idx)
{
struct lod_layout_component *lod_comp;
struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
struct ost_pool *osts;
unsigned int i;
__u32 nfound, good_osts, stripe_count, stripe_count_min;
- __u32 inuse_old_count = inuse->op_count;
int rc = 0;
ENTRY;
* do not put >1 objects on a single OST
*/
if (lod_qos_is_ost_used(env, idx, nfound) ||
- lod_comp_is_ost_used(inuse, idx))
+ lod_comp_is_ost_used(env, lo, idx))
continue;
o = lod_qos_declare_object_on(env, lod, idx, th);
lod_avoid_update(lo, lag);
lod_qos_ost_in_use(env, nfound, idx);
- lod_comp_ost_in_use(inuse, idx);
stripe[nfound] = o;
ost_indices[nfound] = idx;
lod_qos_used(lod, osts, idx, &total_weight);
dt_object_put(env, stripe[i]);
stripe[i] = NULL;
}
- LASSERTF(nfound <= inuse->op_count,
- "nfound:%d, op_count:%u\n", nfound, inuse->op_count);
- inuse->op_count = inuse_old_count;
/* makes sense to rebalance next time */
lod->lod_qos.lq_dirty = 1;
if (mo->ldo_is_composite) {
offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
v1 = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
+ v3 = (struct lov_mds_md_v3 *)v1;
magic = le32_to_cpu(v1->lmm_magic);
ext = &comp_v1->lcm_entries[i].lcme_extent;
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)
pool_name = NULL;
if (v1->lmm_magic == LOV_USER_MAGIC_V3 ||
v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
- int j;
-
v3 = (struct lov_user_md_v3 *)v1;
if (v3->lmm_pool_name[0] != '\0')
pool_name = v3->lmm_pool_name;
if (v3->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
- if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
- v3->lmm_stripe_offset =
- v3->lmm_objects[0].l_ost_idx;
-
- /* copy ost list from lmm */
- lod_comp->llc_ostlist.op_count =
- v3->lmm_stripe_count;
- lod_comp->llc_ostlist.op_size =
- v3->lmm_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(free_comp, rc = -ENOMEM);
-
- for (j = 0; j < v3->lmm_stripe_count; j++)
- lod_comp->llc_ostlist.op_array[j] =
- v3->lmm_objects[j].l_ost_idx;
+ rc = lod_comp_copy_ost_lists(lod_comp, v3);
+ if (rc)
+ GOTO(free_comp, rc);
}
}
* OSTs to read the data.
*/
comp = &lo->ldo_comp_entries[lo->ldo_mirrors[i].lme_start];
- if (comp->llc_id == LCME_ID_INVAL ||
+ if (comp->llc_id != LCME_ID_INVAL &&
mirror_id_of(comp->llc_id) ==
mirror_id_of(lod_comp->llc_id))
continue;
/* iterate components of a mirror */
lod_foreach_mirror_comp(comp, lo, i) {
- /* skip non-overlapped or un-instantiated components */
+ /**
+ * skip non-overlapped or un-instantiated components,
+ * NOTE: don't use lod_comp_inited(comp) to judge
+ * whether @comp has been inited, since during
+ * declare phase, comp->llc_stripe has been allocated
+ * while it's init flag not been set until the exec
+ * phase.
+ */
if (!lu_extent_is_overlapped(&comp->llc_extent,
&lod_comp->llc_extent) ||
- !lod_comp_inited(comp) || !comp->llc_stripe)
+ !comp->llc_stripe)
continue;
/**
if (cfs_bitmap_check(bitmap, ost->ltd_index))
continue;
+ QOS_DEBUG("OST%d used in conflicting mirror "
+ "component\n", ost->ltd_index);
cfs_bitmap_set(bitmap, ost->ltd_index);
lag->lag_ost_avail--;
* \param[in] attr attributes OST objects will be declared with
* \param[in] th transaction handle
* \param[in] comp_idx index of ldo_comp_entries
- * \param[in|out] inuse array of inuse ost index
*
* \retval 0 on success
* \retval negative negated errno on error
*/
int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo,
struct lu_attr *attr, struct thandle *th,
- int comp_idx, struct ost_pool *inuse)
+ int comp_idx)
{
struct lod_layout_component *lod_comp;
struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
CDEBUG(D_OTHER, "tgt_count %d stripe_count %d\n",
d->lod_desc.ld_tgt_count, stripe_len);
- if (lod_comp->llc_ostlist.op_array) {
+ if (lod_comp->llc_ostlist.op_array &&
+ lod_comp->llc_ostlist.op_count) {
rc = lod_alloc_ost_list(env, lo, stripe, ost_indices,
- th, comp_idx, inuse);
+ th, comp_idx);
} else if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT) {
/**
* collect OSTs and OSSs used in other mirrors whose
if (rc)
GOTO(put_ldts, rc);
+ QOS_DEBUG("collecting conflict osts for comp[%d]\n",
+ comp_idx);
lod_collect_avoidance(lo, lag, comp_idx);
rc = lod_alloc_qos(env, lo, stripe, ost_indices, flag,
- th, comp_idx, inuse);
+ th, comp_idx);
if (rc == -EAGAIN)
rc = lod_alloc_rr(env, lo, stripe, ost_indices,
- flag, th, comp_idx, inuse);
+ flag, th, comp_idx);
} else {
rc = lod_alloc_specific(env, lo, stripe, ost_indices,
- flag, th, comp_idx, inuse);
+ flag, th, comp_idx);
}
put_ldts:
lod_putref(d, &d->lod_ost_descs);
RETURN(rc);
}
-int lod_obj_stripe_set_inuse_cb(const struct lu_env *env,
- struct lod_object *lo,
- struct dt_object *dt, struct thandle *th,
- int comp_idx, int stripe_idx,
- struct lod_obj_stripe_cb_data *data)
-{
- struct lod_thread_info *info = lod_env_info(env);
- struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
- struct lu_fid *fid = &info->lti_fid;
- __u32 index;
- int rc, type = LU_SEQ_RANGE_OST;
-
- *fid = *lu_object_fid(&dt->do_lu);
- rc = lod_fld_lookup(env, d, fid, &index, &type);
- if (rc < 0) {
- CERROR("%s: fail to locate "DFID": rc = %d\n",
- lod2obd(d)->obd_name, PFID(fid), rc);
- return rc;
- }
- lod_comp_ost_in_use(data->locd_inuse, index);
- return 0;
-}
-
-/**
- * Resize per-thread ost list to hold OST target index list already used.
- *
- * \param[in,out] inuse structure contains ost list array
- * \param[in] cnt total stripe count of all components
- * \param[in] max array's max size if @max > 0
- *
- * \retval 0 on success
- * \retval -ENOMEM reallocation failed
- */
-static int lod_inuse_resize(struct ost_pool *inuse, __u16 cnt, __u16 max)
-{
- __u32 *array;
- __u32 new = cnt * sizeof(inuse->op_array[0]);
-
- inuse->op_count = 0;
-
- if (new <= inuse->op_size)
- return 0;
-
- if (max)
- new = min_t(__u32, new, max);
-
- OBD_ALLOC(array, new);
- if (!array)
- return -ENOMEM;
-
- if (inuse->op_array)
- OBD_FREE(inuse->op_array, inuse->op_size);
-
- inuse->op_array = array;
- inuse->op_size = new;
-
- return 0;
-}
-
-int lod_prepare_inuse(const struct lu_env *env, struct lod_object *lo)
-{
- struct lod_thread_info *info = lod_env_info(env);
- struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
- struct ost_pool *inuse = &info->lti_inuse_osts;
- struct lod_obj_stripe_cb_data data = { { 0 } };
- __u32 stripe_count = 0;
- int i;
- int rc;
-
- for (i = 0; i < lo->ldo_comp_cnt; i++)
- stripe_count += lod_comp_entry_stripe_count(lo,
- &lo->ldo_comp_entries[i], false);
- rc = lod_inuse_resize(inuse, stripe_count, d->lod_osd_max_easize);
- if (rc)
- return rc;
-
- data.locd_inuse = inuse;
- data.locd_stripe_cb = lod_obj_stripe_set_inuse_cb;
- return lod_obj_for_each_stripe(env, lo, NULL, &data);
-}
-
int lod_prepare_create(const struct lu_env *env, struct lod_object *lo,
struct lu_attr *attr, const struct lu_buf *buf,
struct thandle *th)
{
- struct lod_thread_info *info = lod_env_info(env);
struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
- struct ost_pool *inuse = &info->lti_inuse_osts;
uint64_t size = 0;
int i;
int rc;
if (attr->la_valid & LA_SIZE)
size = attr->la_size;
- /* prepare inuse */
- rc = lod_prepare_inuse(env, lo);
- if (rc)
- RETURN(rc);
-
/**
* prepare OST object creation for the component covering file's
* size, the 1st component (including plain layout file) is always
lod_comp = &lo->ldo_comp_entries[i];
extent = &lod_comp->llc_extent;
- CDEBUG(D_QOS, "%lld [%lld, %lld)\n",
- size, extent->e_start, extent->e_end);
+ QOS_DEBUG("comp[%d] %lld "DEXT"\n", i, size, PEXT(extent));
if (!lo->ldo_is_composite || size >= extent->e_start) {
- rc = lod_qos_prep_create(env, lo, attr, th, i, inuse);
+ rc = lod_qos_prep_create(env, lo, attr, th, i);
if (rc)
break;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff