X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flod%2Flod_internal.h;h=f47454139bb0fa75905057818a264a929e855047;hp=83fb5398e5a15ec9646802387f2a7e9b481328d7;hb=3b0094779b58c162e5dc553f6d7cdff1910d0c3b;hpb=c872afa36ff5de5910f0f524f7a487982fa0c776 diff --git a/lustre/lod/lod_internal.h b/lustre/lod/lod_internal.h index 83fb539..f474541 100644 --- a/lustre/lod/lod_internal.h +++ b/lustre/lod/lod_internal.h @@ -54,113 +54,30 @@ #define LMVEA_DELETE_VALUES(count, offset) \ ((count) == 0 && (offset) == (typeof(offset))(-1)) -struct lod_qos_rr { - spinlock_t lqr_alloc; /* protect allocation index */ - __u32 lqr_start_idx; /* start index of new inode */ - __u32 lqr_offset_idx;/* aliasing for start_idx */ - int lqr_start_count;/* reseed counter */ - struct ost_pool lqr_pool; /* round-robin optimized list */ - unsigned long lqr_dirty:1; /* recalc round-robin list */ -}; - 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 lod_qos_rr pool_rr; - struct hlist_node pool_hash; /* access by poolname */ + struct lu_qos_rr pool_rr; + 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) -struct lod_qos { - struct list_head lq_oss_list; - struct rw_semaphore lq_rw_sem; - __u32 lq_active_oss_count; - unsigned int lq_prio_free; /* priority for free space */ - unsigned int lq_threshold_rr;/* priority for rr */ - struct lod_qos_rr lq_rr; /* round robin qos data */ - bool lq_dirty:1, /* recalc qos data */ - lq_same_space:1,/* the ost's all have approx. - the same space avail */ - lq_reset:1; /* zero current penalties */ -}; - -struct lod_qos_oss { - struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */ - struct list_head lqo_oss_list; /* link to lov_qos */ - __u64 lqo_bavail; /* total bytes avail on OSS */ - __u64 lqo_penalty; /* current penalty */ - __u64 lqo_penalty_per_obj; /* penalty decrease - every obj*/ - time64_t lqo_used; /* last used time, seconds */ - __u32 lqo_ost_count; /* number of osts on this oss */ - __u32 lqo_id; /* unique oss id */ -}; - -struct ltd_qos { - struct lod_qos_oss *ltq_oss; /* oss info */ - __u64 ltq_penalty; /* current penalty */ - __u64 ltq_penalty_per_obj; /* penalty decrease - every obj*/ - __u64 ltq_weight; /* net weighting */ - time64_t ltq_used; /* last used time, seconds */ - bool ltq_usable:1; /* usable for striping */ -}; - -struct lod_tgt_desc { - struct dt_device *ltd_tgt; - struct list_head ltd_kill; - struct obd_export *ltd_exp; - struct obd_uuid ltd_uuid; - __u32 ltd_gen; - __u32 ltd_index; - struct ltd_qos ltd_qos; /* qos info per target */ - struct obd_statfs ltd_statfs; - struct ptlrpc_thread *ltd_recovery_thread; - unsigned long ltd_active:1,/* is this target up for requests */ - ltd_activate:1,/* should target be activated */ - ltd_reap:1, /* should this target be deleted */ - ltd_got_update_log:1, /* Already got update log */ - ltd_connecting:1; /* target is connecting */ -}; - -#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_desc lu_tgt_desc +#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 */ @@ -187,9 +104,6 @@ struct lod_device { 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; @@ -199,23 +113,16 @@ struct lod_device { 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 lod_qos lod_qos; /* qos info per lod */ - /* 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; @@ -228,19 +135,9 @@ struct lod_device { 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 lod_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 lod_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; @@ -255,7 +152,7 @@ struct lod_layout_component { __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; }; @@ -306,7 +203,8 @@ struct lod_object { struct lod_mirror_entry *ldo_mirrors; __u32 ldo_is_composite:1, ldo_flr_state:2, - ldo_comp_cached:1; + ldo_comp_cached:1, + ldo_is_foreign:1; }; /* directory stripe (LMV) */ struct { @@ -322,19 +220,36 @@ struct lod_object { __u32 ldo_dir_slave_stripe:1, ldo_dir_striped:1, /* the stripe has been loaded */ - ldo_dir_stripe_loaded:1; + ldo_dir_stripe_loaded:1, + /* foreign directory */ + ldo_dir_is_foreign; /* - * default striping is not cached, so this field is - * invalid after create, make sure it's used by + * This default LMV is parent default LMV, which will be + * used in child creation, and it's not cached, so this + * field is invalid after create, make sure it's used by * lod_dir_striping_create_internal() only. */ struct lod_default_striping *ldo_def_striping; }; }; - /* file stripe (LOV) */ - struct lod_layout_component *ldo_comp_entries; - /* slave stripes of striped directory (LMV) */ - struct dt_object **ldo_stripe; + union { + struct { + /* foreign/raw format LOV */ + char *ldo_foreign_lov; + size_t ldo_foreign_lov_size; + }; + struct { + /* foreign/raw format LMV */ + char *ldo_foreign_lmv; + size_t ldo_foreign_lmv_size; + }; + struct { + /* file stripe (LOV) */ + struct lod_layout_component *ldo_comp_entries; + /* slave stripes of striped directory (LMV) */ + struct dt_object **ldo_stripe; + }; + }; }; #define lod_foreach_mirror_comp(comp, lo, mirror_idx) \ @@ -519,6 +434,9 @@ static inline bool lod_obj_is_striped(struct dt_object *dt) if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) return lo->ldo_dir_stripe_count != 0; + if (lo->ldo_is_foreign) + return false; + for (i = 0; i < lo->ldo_comp_cnt; i++) { if (lo->ldo_comp_entries[i].llc_stripe == NULL) continue; @@ -572,12 +490,8 @@ static inline void lod_layout_get_pool(struct lod_layout_component *entries, } } -#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; @@ -586,7 +500,7 @@ int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod, 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); @@ -597,10 +511,10 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, 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); @@ -668,6 +582,7 @@ void lod_fix_desc_qos_maxage(__u32 *val); 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, @@ -688,15 +603,14 @@ int lod_alloc_comp_entries(struct lod_object *lo, int mirror_cnt, int comp_cnt); 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); +int lod_tgt_pool_init(struct lu_tgt_pool *op, unsigned int count); +int lod_tgt_pool_free(struct lu_tgt_pool *op); +int lod_tgt_pool_add(struct lu_tgt_pool *op, __u32 idx, unsigned int min_count); +int lod_tgt_pool_remove(struct lu_tgt_pool *op, __u32 idx); +int lod_tgt_pool_extend(struct lu_tgt_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); @@ -727,12 +641,13 @@ struct lod_obj_stripe_cb_data { }; /* lod_qos.c */ +int lod_mdt_alloc_qos(const struct lu_env *env, struct lod_object *lo, + struct dt_object **stripes); +int lod_mdt_alloc_rr(const struct lu_env *env, struct lod_object *lo, + struct dt_object **stripe); 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); -int qos_add_tgt(struct lod_device*, struct lod_tgt_desc *); -int qos_del_tgt(struct lod_device *, struct lod_tgt_desc *); -void lod_qos_rr_init(struct lod_qos_rr *lqr); int lod_use_defined_striping(const struct lu_env *, struct lod_object *, const struct lu_buf *); int lod_qos_parse_config(const struct lu_env *env, struct lod_object *lo, @@ -744,8 +659,9 @@ __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); -void lod_qos_statfs_update(const struct lu_env *env, struct lod_device *lod); + __u16 stripe_count, bool overstriping); +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); @@ -763,6 +679,8 @@ int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt, int lod_striped_create(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr, struct dt_object_format *dof, struct thandle *th); +int lod_alloc_foreign_lov(struct lod_object *lo, size_t size); +void lod_free_foreign_lov(struct lod_object *lo); void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo); void lod_striping_free(const struct lu_env *env, struct lod_object *lo);