struct pool_desc {
char pool_name[LOV_MAXPOOLNAME + 1];
- struct ost_pool pool_obds; /* pool members */
+ struct lu_tgt_pool pool_obds; /* pool members */
atomic_t pool_refcount;
struct lu_qos_rr pool_rr;
- struct hlist_node pool_hash; /* access by poolname */
+ struct rhash_head pool_hash; /* access by poolname */
struct list_head pool_list;
+ struct rcu_head pool_rcu;
struct proc_dir_entry *pool_proc_entry;
struct obd_device *pool_lobd; /* owner */
};
+int lod_pool_hash_init(struct rhashtable *tbl);
+void lod_pool_hash_destroy(struct rhashtable *tbl);
+
#define pool_tgt_count(p) ((p)->pool_obds.op_count)
#define pool_tgt_array(p) ((p)->pool_obds.op_array)
#define pool_tgt_rw_sem(p) ((p)->pool_obds.op_rw_sem)
#define lod_tgt_desc lu_tgt_desc
-
-#define TGT_PTRS 256 /* number of pointers at 1st level */
-#define TGT_PTRS_PER_BLOCK 256 /* number of pointers at 2nd level */
-
-struct lod_tgt_desc_idx {
- struct lod_tgt_desc *ldi_tgt[TGT_PTRS_PER_BLOCK];
-};
-
-#define LTD_TGT(ltd, index) \
- ((ltd)->ltd_tgt_idx[(index) / \
- TGT_PTRS_PER_BLOCK]->ldi_tgt[(index) % TGT_PTRS_PER_BLOCK])
+#define lod_tgt_descs lu_tgt_descs
#define OST_TGT(lod, index) LTD_TGT(&lod->lod_ost_descs, index)
#define MDT_TGT(lod, index) LTD_TGT(&lod->lod_mdt_descs, index)
-struct lod_tgt_descs {
- /* list of known TGTs */
- struct lod_tgt_desc_idx *ltd_tgt_idx[TGT_PTRS];
- /* Size of the lod_tgts array, granted to be a power of 2 */
- __u32 ltd_tgts_size;
- /* number of registered TGTs */
- __u32 ltd_tgtnr;
- /* bitmap of TGTs available */
- struct cfs_bitmap *ltd_tgt_bitmap;
- /* TGTs scheduled to be deleted */
- __u32 ltd_death_row;
- /* Table refcount used for delayed deletion */
- int ltd_refcount;
- /* mutex to serialize concurrent updates to the tgt table */
- struct mutex ltd_mutex;
- /* read/write semaphore used for array relocation */
- struct rw_semaphore ltd_rw_sem;
-};
struct lod_avoid_guide {
/* ids of OSSs avoid guidance */
__u32 lag_ost_avail;
};
+#define LOD_DOM_MIN_SIZE_KB (LOV_MIN_STRIPE_SIZE >> 10)
+#define LOD_DOM_SFS_MAX_AGE 10
+
struct lod_device {
struct dt_device lod_dt_dev;
struct obd_export *lod_child_exp;
lod_lmv_failout:1,
lod_child_got_update_log:1;
- /* lov settings descriptor storing static information */
- struct lov_desc lod_desc;
-
/* protect ld_active_tgt_count, ltd_active and lod_md_root */
spinlock_t lod_lock;
struct lod_tgt_descs lod_mdt_descs;
/* Recovery thread for lod_child */
- struct ptlrpc_thread lod_child_recovery_thread;
+ struct task_struct *lod_child_recovery_task;
/* maximum EA size underlied OSD may have */
unsigned int lod_osd_max_easize;
/* maximum size of MDT stripe for Data-on-MDT files. */
- unsigned int lod_dom_max_stripesize;
-
- /*FIXME: When QOS and pool is implemented for MDT, probably these
- * structure should be moved to lod_tgt_descs as well.
- */
- /* QoS info per LOD */
- struct lu_qos lod_qos; /* qos info per lod */
+ unsigned int lod_dom_stripesize_max_kb;
+ /* current DOM default stripe size adjusted by threshold */
+ unsigned int lod_dom_stripesize_cur_kb;
+ /* Threshold at which DOM default stripe will start decreasing */
+ __u64 lod_dom_threshold_free_mb;
+
+ /* Local OSD statfs cache */
+ spinlock_t lod_lsfs_lock;
+ time64_t lod_lsfs_age;
+ __u64 lod_lsfs_total_mb;
+ __u64 lod_lsfs_free_mb;
/* OST pool data */
- struct ost_pool lod_pool_info; /* all OSTs in a packed array */
int lod_pool_count;
- struct cfs_hash *lod_pools_hash_body; /* used for key access */
+ struct rhashtable lod_pools_hash_body; /* used for key access */
struct list_head lod_pool_list; /* used for sequential access */
struct proc_dir_entry *lod_pool_proc_entry;
struct lod_object *lod_md_root;
};
-#define lod_osts lod_ost_descs.ltd_tgts
-#define lod_ost_bitmap lod_ost_descs.ltd_tgt_bitmap
-#define lod_ostnr lod_ost_descs.ltd_tgtnr
-#define lod_osts_size lod_ost_descs.ltd_tgts_size
-#define ltd_ost ltd_tgt
-#define lod_ost_desc lu_tgt_desc
-
-#define lod_mdts lod_mdt_descs.ltd_tgts
-#define lod_mdt_bitmap lod_mdt_descs.ltd_tgt_bitmap
-#define lod_remote_mdt_count lod_mdt_descs.ltd_tgtnr
-#define lod_mdts_size lod_mdt_descs.ltd_tgts_size
-#define ltd_mdt ltd_tgt
-#define lod_mdt_desc lu_tgt_desc
+#define lod_ost_bitmap lod_ost_descs.ltd_tgt_bitmap
+#define lod_ost_count lod_ost_descs.ltd_lov_desc.ld_tgt_count
+#define lod_remote_mdt_count lod_mdt_descs.ltd_lmv_desc.ld_tgt_count
struct lod_layout_component {
struct lu_extent llc_extent;
__u64 llc_timestamp; /* snapshot time */
char *llc_pool;
/* ost list specified with LOV_USER_MAGIC_SPECIFIC lum */
- struct ost_pool llc_ostlist;
+ struct lu_tgt_pool llc_ostlist;
struct dt_object **llc_stripe;
__u32 *llc_ost_indices;
};
__u32 ldo_dir_hash_type;
__u32 ldo_dir_migrate_offset;
__u32 ldo_dir_migrate_hash;
+ __u32 ldo_dir_layout_version;
/* Is a slave stripe of striped directory? */
__u32 ldo_dir_slave_stripe:1,
ldo_dir_striped:1,
};
};
+#define ldo_dir_split_offset ldo_dir_migrate_offset
+#define ldo_dir_split_hash ldo_dir_migrate_hash
+
#define lod_foreach_mirror_comp(comp, lo, mirror_idx) \
for (comp = &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_start]; \
comp <= &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_end]; \
return (lo->ldo_flr_state & LCM_FL_FLR_MASK) != LCM_FL_NONE;
}
+static inline bool lod_is_splitting(const struct lod_object *lo)
+{
+ return lmv_hash_is_splitting(lo->ldo_dir_hash_type);
+}
+
+static inline bool lod_is_migrating(const struct lod_object *lo)
+{
+ return lmv_hash_is_migrating(lo->ldo_dir_hash_type);
+}
+
+static inline bool lod_is_layout_changing(const struct lod_object *lo)
+{
+ return lmv_hash_is_layout_changing(lo->ldo_dir_hash_type);
+}
+
static inline int lod_set_pool(char **pool, const char *new_pool)
{
int len;
struct lu_attr lti_layout_attr;
/* object allocation avoid guide info */
struct lod_avoid_guide lti_avoid;
+ union lmv_mds_md lti_lmv;
};
extern const struct lu_device_operations lod_lu_ops;
static inline struct lod_device* lu2lod_dev(struct lu_device *d)
{
LASSERT(lu_device_is_lod(d));
- return container_of0(d, struct lod_device, lod_dt_dev.dd_lu_dev);
+ return container_of_safe(d, struct lod_device, lod_dt_dev.dd_lu_dev);
}
static inline struct lu_device *lod2lu_dev(struct lod_device *d)
static inline struct lod_device *dt2lod_dev(struct dt_device *d)
{
LASSERT(lu_device_is_lod(&d->dd_lu_dev));
- return container_of0(d, struct lod_device, lod_dt_dev);
+ return container_of(d, struct lod_device, lod_dt_dev);
}
static inline struct lod_object *lu2lod_obj(struct lu_object *o)
{
LASSERT(ergo(o != NULL, lu_device_is_lod(o->lo_dev)));
- return container_of0(o, struct lod_object, ldo_obj.do_lu);
+ return container_of_safe(o, struct lod_object, ldo_obj.do_lu);
}
static inline struct lu_object *lod2lu_obj(struct lod_object *obj)
static inline struct lod_object *lod_obj(const struct lu_object *o)
{
LASSERT(lu_device_is_lod(o->lo_dev));
- return container_of0(o, struct lod_object, ldo_obj.do_lu);
+ return container_of(o, struct lod_object, ldo_obj.do_lu);
}
static inline struct lod_object *lod_dt_obj(const struct dt_object *d)
static inline struct dt_object* lod_object_child(struct lod_object *o)
{
- return container_of0(lu_object_next(lod2lu_obj(o)),
- struct dt_object, do_lu);
+ return container_of(lu_object_next(lod2lu_obj(o)),
+ struct dt_object, do_lu);
}
static inline bool lod_obj_is_striped(struct dt_object *dt)
}
}
-#define lod_foreach_ost(__dev, index) \
- if ((__dev)->lod_osts_size > 0) \
- cfs_foreach_bit((__dev)->lod_ost_bitmap, (index))
-
-#define lod_foreach_mdt(mdt_dev, index) \
- cfs_foreach_bit((mdt_dev)->lod_mdt_bitmap, (index))
+#define lod_foreach_mdt(lod, mdt) ltd_foreach_tgt(&(lod)->lod_mdt_descs, mdt)
+#define lod_foreach_ost(lod, ost) ltd_foreach_tgt(&(lod)->lod_ost_descs, ost)
/* lod_dev.c */
extern struct kmem_cache *lod_object_kmem;
int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
struct dt_device *dt);
void lod_sub_fini_llog(const struct lu_env *env,
- struct dt_device *dt, struct ptlrpc_thread *thread);
+ struct dt_device *dt, struct task_struct **taskp);
int lodname2mdt_index(char *lodname, __u32 *mdt_index);
extern void target_recovery_fini(struct obd_device *obd);
char *osp, unsigned index, unsigned gen, int mdt_index,
char *type, int active);
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);
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);
int lod_striping_load(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);
+void lod_dom_stripesize_recalc(struct lod_device *d);
int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
const char *name);
void lod_fix_desc_pattern(__u32 *val);
void lod_fix_desc_stripe_count(__u32 *val);
void lod_fix_desc_stripe_size(__u64 *val);
+void lod_fix_lmv_desc_pattern(__u32 *val);
int lod_pools_init(struct lod_device *m, struct lustre_cfg *cfg);
int lod_pools_fini(struct lod_device *m);
int lod_parse_striping(const struct lu_env *env, struct lod_object *mo,
const struct lu_buf *buf);
int lod_initialize_objects(const struct lu_env *env, struct lod_object *mo,
struct lov_ost_data_v1 *objs, int index);
-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);
int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
struct lov_mds_md *lmm, int *lmm_size, bool is_dir);
int lod_ea_store_resize(struct lod_thread_info *info, size_t size);
int lod_fill_mirrors(struct lod_object *lo);
/* lod_pool.c */
-int lod_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count);
-int lod_ost_pool_remove(struct ost_pool *op, __u32 idx);
-int lod_ost_pool_extend(struct ost_pool *op, unsigned int min_count);
struct pool_desc *lod_find_pool(struct lod_device *lod, char *poolname);
void lod_pool_putref(struct pool_desc *pool);
-int lod_ost_pool_free(struct ost_pool *op);
int lod_pool_del(struct obd_device *obd, char *poolname);
-int lod_ost_pool_init(struct ost_pool *op, unsigned int count);
-extern struct cfs_hash_ops pool_hash_operations;
int lod_check_index_in_pool(__u32 idx, struct pool_desc *pool);
int lod_pool_new(struct obd_device *obd, char *poolname);
int lod_pool_add(struct obd_device *obd, char *poolname, char *ostname);
};
/* lod_qos.c */
+int lod_mdt_alloc_qos(const struct lu_env *env, struct lod_object *lo,
+ struct dt_object **stripes, u32 stripe_idx,
+ u32 stripe_count);
+int lod_mdt_alloc_rr(const struct lu_env *env, struct lod_object *lo,
+ struct dt_object **stripes, u32 stripe_idx,
+ u32 stripe_count);
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);
const struct lu_buf *buf);
int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo,
struct lu_attr *attr, struct thandle *th,
- int comp_idx);
+ int comp_idx, __u64 reserve);
__u16 lod_comp_entry_stripe_count(struct lod_object *lo,
struct lod_layout_component *entry,
bool is_dir);
__u16 lod_get_stripe_count(struct lod_device *lod, struct lod_object *lo,
__u16 stripe_count, bool overstriping);
-void lod_qos_statfs_update(const struct lu_env *env, struct lod_device *lod);
-int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d,
- int index, struct obd_statfs *sfs,
- struct obd_statfs_info *info);
+void lod_qos_statfs_update(const struct lu_env *env, struct lod_device *lod,
+ struct lu_tgt_descs *ltd);
/* lproc_lod.c */
int lod_procfs_init(struct lod_device *lod);
struct lod_obj_stripe_cb_data *data);
int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
struct lov_user_md_v3 *v3);
+void lod_adjust_stripe_size(struct lod_layout_component *comp,
+ __u32 def_stripe_size);
/* lod_sub_object.c */
struct thandle *lod_sub_get_thandle(const struct lu_env *env,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff