X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Finclude%2Flu_target.h;h=5da2544d81de8d60e1d182eb01f27a8a8afb7576;hp=eb155f747b4eca757cc4b12784eee7403a90f664;hb=b32e55b600ca2c9bf8b62287d9f889791d157426;hpb=466b89e41c3917a8a836065f72bfa62e57c818bd diff --git a/lustre/include/lu_target.h b/lustre/include/lu_target.h index eb155f7..5da2544 100644 --- a/lustre/include/lu_target.h +++ b/lustre/include/lu_target.h @@ -27,7 +27,7 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,11 +41,87 @@ #include #include #include +#include + +/* Each one represents a distribute transaction replay + * operation, and updates on each MDTs are linked to + * dtr_sub_list */ +struct distribute_txn_replay_req { + /* update record */ + struct llog_update_record *dtrq_lur; + int dtrq_lur_size; + + /* linked to the distribute transaction replay + * list (tdtd_replay_list) */ + struct list_head dtrq_list; + __u64 dtrq_master_transno; + __u64 dtrq_batchid; + __u64 dtrq_xid; + + /* all of sub updates are linked here */ + struct list_head dtrq_sub_list; + spinlock_t dtrq_sub_list_lock; + + /* If the local update has been executed during replay */ + __u32 dtrq_local_update_executed:1; +}; + +/* Each one represents a sub replay item under a distribute + * transaction. A distribute transaction will be operated in + * two or more MDTs, and updates on each MDT will be represented + * by this structure */ +struct distribute_txn_replay_req_sub { + __u32 dtrqs_mdt_index; + + /* All of cookies for the update will be linked here */ + spinlock_t dtrqs_cookie_list_lock; + struct list_head dtrqs_cookie_list; + struct list_head dtrqs_list; +}; + +struct target_distribute_txn_data; +typedef int (*distribute_txn_replay_handler_t)(struct lu_env *env, + struct target_distribute_txn_data *tdtd, + struct distribute_txn_replay_req *dtrq); +struct target_distribute_txn_data { + /* Distribution ID is used to identify updates log on different + * MDTs for one operation */ + spinlock_t tdtd_batchid_lock; + __u64 tdtd_batchid; + struct lu_target *tdtd_lut; + struct dt_object *tdtd_batchid_obj; + struct dt_device *tdtd_dt; + + /* Committed batchid for distribute transaction */ + __u64 tdtd_committed_batchid; + + /* List for distribute transaction */ + struct list_head tdtd_list; + + /* Threads to manage distribute transaction */ + wait_queue_head_t tdtd_commit_thread_waitq; + atomic_t tdtd_refcount; + + /* recovery update */ + distribute_txn_replay_handler_t tdtd_replay_handler; + struct list_head tdtd_replay_list; + struct list_head tdtd_replay_finish_list; + spinlock_t tdtd_replay_list_lock; + /* last replay update transno */ + __u32 tdtd_replay_ready:1; + + /* Manage the llog recovery threads */ + atomic_t tdtd_recovery_threads_count; + wait_queue_head_t tdtd_recovery_threads_waitq; +}; struct lu_target { struct obd_device *lut_obd; struct dt_device *lut_bottom; + struct target_distribute_txn_data *lut_tdtd; + struct ptlrpc_thread lut_tdtd_commit_thread; + /* supported opcodes and handlers for this target */ struct tgt_opc_slice *lut_slice; __u32 lut_reply_fail_id; @@ -56,9 +132,7 @@ struct lu_target { struct sptlrpc_rule_set lut_sptlrpc_rset; spinlock_t lut_flags_lock; int lut_sec_level; - unsigned int lut_mds_capa:1, - lut_oss_capa:1, - lut_syncjournal:1, + unsigned int lut_syncjournal:1, lut_sync_lock_cancel:2, /* e.g. OST node */ lut_no_reconstruct:1; @@ -76,6 +150,35 @@ struct lu_target { spinlock_t lut_client_bitmap_lock; /** Bitmap of known clients */ unsigned long *lut_client_bitmap; + /* Number of clients supporting multiple modify RPCs + * recorded in the bitmap */ + atomic_t lut_num_clients; + /* Client generation to identify client slot reuse */ + atomic_t lut_client_generation; + /** reply_data file */ + struct dt_object *lut_reply_data; + /** Bitmap of used slots in the reply data file */ + unsigned long **lut_reply_bitmap; +}; + +/* number of slots in reply bitmap */ +#define LUT_REPLY_SLOTS_PER_CHUNK (1<<20) +#define LUT_REPLY_SLOTS_MAX_CHUNKS 16 + +/** + * Target reply data + */ +struct tg_reply_data { + /** chain of reply data anchored in tg_export_data */ + struct list_head trd_list; + /** copy of on-disk reply data */ + struct lsd_reply_data trd_reply; + /** versions for Version Based Recovery */ + __u64 trd_pre_versions[4]; + /** slot index in reply_data file */ + int trd_index; + /** tag the client used */ + __u16 trd_tag; }; extern struct lu_context_key tgt_session_key; @@ -117,8 +220,14 @@ struct tgt_session_info { * Additional fail id that can be set by handler. */ int tsi_reply_fail_id; + bool tsi_preprocessed; /* request JobID */ char *tsi_jobid; + + /* update replay */ + __u64 tsi_xid; + __u32 tsi_result; + __u32 tsi_client_gen; }; static inline struct tgt_session_info *tgt_ses_info(const struct lu_env *env) @@ -207,9 +316,9 @@ struct tgt_handler { /* Request version for this opcode */ int th_version; /* Handler function */ - int (*th_act)(struct tgt_session_info *tti); + int (*th_act)(struct tgt_session_info *tsi); /* Handler function for high priority requests */ - int (*th_hp)(struct tgt_session_info *tti); + void (*th_hp)(struct tgt_session_info *tsi); /* Request format for this request */ const struct req_format *th_fmt; }; @@ -237,6 +346,12 @@ static inline int req_is_replay(struct ptlrpc_request *req) return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY); } +static inline bool tgt_is_multimodrpcs_client(struct obd_export *exp) +{ + return exp_connect_flags(exp) & OBD_CONNECT_MULTIMODRPCS; +} + + /* target/tgt_handler.c */ int tgt_request_handle(struct ptlrpc_request *req); char *tgt_name(struct lu_target *tgt); @@ -261,9 +376,10 @@ int tgt_sec_ctx_init(struct tgt_session_info *tsi); int tgt_sec_ctx_init_cont(struct tgt_session_info *tsi); int tgt_sec_ctx_fini(struct tgt_session_info *tsi); int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob); +int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf); int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa); int tgt_sync(const struct lu_env *env, struct lu_target *tgt, - struct dt_object *obj); + struct dt_object *obj, __u64 start, __u64 end); int tgt_io_thread_init(struct ptlrpc_thread *thread); void tgt_io_thread_done(struct ptlrpc_thread *thread); @@ -271,17 +387,26 @@ void tgt_io_thread_done(struct ptlrpc_thread *thread); int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, __u64 start, __u64 end, struct lustre_handle *lh, int mode, __u64 *flags); -void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode); +void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode); int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, struct obd_ioobj *obj, struct niobuf_remote *nb, - struct lustre_handle *lh, int mode); + struct lustre_handle *lh, enum ldlm_mode mode); void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob, - struct lustre_handle *lh, int mode); + struct lustre_handle *lh, enum ldlm_mode mode); int tgt_brw_read(struct tgt_session_info *tsi); int tgt_brw_write(struct tgt_session_info *tsi); int tgt_hpreq_handler(struct ptlrpc_request *req); +void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *, + struct dt_device *, + struct lfsck_request *, + struct thandle *)); +void tgt_register_lfsck_query(int (*query)(const struct lu_env *, + struct dt_device *, + struct lfsck_request *)); +bool req_can_reconstruct(struct ptlrpc_request *req, struct tg_reply_data *trd); extern struct tgt_handler tgt_sec_ctx_handlers[]; +extern struct tgt_handler tgt_lfsck_handlers[]; extern struct tgt_handler tgt_obd_handlers[]; extern struct tgt_handler tgt_dlm_handlers[]; extern struct tgt_handler tgt_llog_handlers[]; @@ -296,6 +421,8 @@ struct tgt_commit_cb { void *tgt_cb_data; }; +int tgt_hpreq_handler(struct ptlrpc_request *req); + /* target/tgt_main.c */ void tgt_boot_epoch_update(struct lu_target *lut); int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *lut, @@ -322,7 +449,42 @@ int tgt_server_data_update(const struct lu_env *env, struct lu_target *tg, int sync); int tgt_truncate_last_rcvd(const struct lu_env *env, struct lu_target *tg, loff_t off); - +int tgt_reply_data_init(const struct lu_env *env, struct lu_target *tgt); +bool tgt_lookup_reply(struct ptlrpc_request *req, struct tg_reply_data *trd); +int tgt_add_reply_data(const struct lu_env *env, struct lu_target *tgt, + struct tg_export_data *ted, struct tg_reply_data *trd, + struct thandle *th, bool update_lrd_file); +struct tg_reply_data *tgt_lookup_reply_by_xid(struct tg_export_data *ted, + __u64 xid); + +/* target/update_trans.c */ +int distribute_txn_init(const struct lu_env *env, + struct lu_target *lut, + struct target_distribute_txn_data *tdtd, + __u32 index); +void distribute_txn_fini(const struct lu_env *env, + struct target_distribute_txn_data *tdtd); + +/* target/update_recovery.c */ +int insert_update_records_to_replay_list(struct target_distribute_txn_data *, + struct llog_update_record *, + struct llog_cookie *, __u32); +void dtrq_list_dump(struct target_distribute_txn_data *tdtd, + unsigned int mask); +void dtrq_list_destroy(struct target_distribute_txn_data *tdtd); +int distribute_txn_replay_handle(struct lu_env *env, + struct target_distribute_txn_data *tdtd, + struct distribute_txn_replay_req *dtrq); +__u64 distribute_txn_get_next_transno(struct target_distribute_txn_data *tdtd); +struct distribute_txn_replay_req * +distribute_txn_get_next_req(struct target_distribute_txn_data *tdtd); +void dtrq_destroy(struct distribute_txn_replay_req *dtrq); +struct distribute_txn_replay_req_sub * +dtrq_sub_lookup(struct distribute_txn_replay_req *dtrq, __u32 mdt_index); +struct distribute_txn_replay_req * +distribute_txn_lookup_finish_list(struct target_distribute_txn_data *tdtd, + __u64 transno); +bool is_req_replayed_by_update(struct ptlrpc_request *req); enum { ESERIOUS = 0x0001000 }; @@ -360,7 +522,7 @@ static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa) /* * Unified target generic handers macros and generic functions. */ -#define TGT_RPC_HANDLER(base, flags, opc, fn, fmt, version) \ +#define TGT_RPC_HANDLER_HP(base, flags, opc, fn, hp, fmt, version) \ [opc - base] = { \ .th_name = #opc, \ .th_fail_id = OBD_FAIL_ ## opc ## _NET, \ @@ -368,8 +530,11 @@ static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa) .th_flags = flags, \ .th_act = fn, \ .th_fmt = fmt, \ - .th_version = version \ + .th_version = version, \ + .th_hp = hp, \ } +#define TGT_RPC_HANDLER(base, flags, opc, fn, fmt, version) \ + TGT_RPC_HANDLER_HP(base, flags, opc, fn, NULL, fmt, version) /* MDT Request with a format known in advance */ #define TGT_MDT_HDL(flags, name, fn) \ @@ -380,10 +545,13 @@ static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa) TGT_RPC_HANDLER(MDS_FIRST_OPC, flags, name, fn, NULL, \ LUSTRE_MDS_VERSION) -/* MDT Request with a format known in advance */ +/* OST Request with a format known in advance */ #define TGT_OST_HDL(flags, name, fn) \ TGT_RPC_HANDLER(OST_FIRST_OPC, flags, name, fn, &RQF_ ## name, \ LUSTRE_OST_VERSION) +#define TGT_OST_HDL_HP(flags, name, fn, hp) \ + TGT_RPC_HANDLER_HP(OST_FIRST_OPC, flags, name, fn, hp, \ + &RQF_ ## name, LUSTRE_OST_VERSION) /* MGS request with a format known in advance */ #define TGT_MGS_HDL(flags, name, fn) \ @@ -440,13 +608,18 @@ static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa) LUSTRE_MDS_VERSION) /* FID Location Database handlers */ -#define TGT_FLD_HDL(flags, name, fn) \ - TGT_RPC_HANDLER(FLD_QUERY, flags, name, fn, &RQF_ ## name, \ +#define TGT_FLD_HDL_VAR(flags, name, fn) \ + TGT_RPC_HANDLER(FLD_QUERY, flags, name, fn, NULL, \ LUSTRE_MDS_VERSION) +/* LFSCK handlers */ +#define TGT_LFSCK_HDL(flags, name, fn) \ + TGT_RPC_HANDLER(LFSCK_FIRST_OPC, flags, name, fn, \ + &RQF_ ## name, LUSTRE_OBD_VERSION) + /* Request with a format known in advance */ #define TGT_UPDATE_HDL(flags, name, fn) \ - TGT_RPC_HANDLER(UPDATE_OBJ, flags, name, fn, &RQF_ ## name, \ + TGT_RPC_HANDLER(OUT_UPDATE, flags, name, fn, &RQF_ ## name, \ LUSTRE_MDS_VERSION) #endif /* __LUSTRE_LU_TARGET_H */