X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_internal.h;h=44a5d68c6be062c59eb267f38a15f58f2d8733b2;hp=90594593d73235f65170b96fddc98b49e6a44363;hb=bc962bde3b109b99c924137ed281d9400637e295;hpb=4fd92d576124fd7772c9f718b83eb67f500e5cec diff --git a/lustre/mdt/mdt_internal.h b/lustre/mdt/mdt_internal.h index 9059459..44a5d68 100644 --- a/lustre/mdt/mdt_internal.h +++ b/lustre/mdt/mdt_internal.h @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -29,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -73,6 +71,7 @@ #include #include #include +#include /* check if request's xid is equal to last one or not*/ static inline int req_xid_is_last(struct ptlrpc_request *req) @@ -86,10 +85,10 @@ struct mdt_object; /* file data for open files on MDS */ struct mdt_file_data { struct portals_handle mfd_handle; /* must be first */ + int mfd_mode; /* open mode provided by client */ cfs_list_t mfd_list; /* protected by med_open_lock */ __u64 mfd_xid; /* xid of the open request */ struct lustre_handle mfd_old_handle; /* old handle in replay case */ - int mfd_mode; /* open mode provided by client */ struct mdt_object *mfd_object; /* point to opened object */ }; @@ -100,9 +99,10 @@ struct mdt_file_data { struct mdt_device { /* super-class */ struct md_device mdt_md_dev; + struct seq_server_site mdt_seq_site; struct ptlrpc_service *mdt_regular_service; struct ptlrpc_service *mdt_readpage_service; - struct ptlrpc_service *mdt_xmds_service; + struct ptlrpc_service *mdt_out_service; struct ptlrpc_service *mdt_setattr_service; struct ptlrpc_service *mdt_mdsc_service; struct ptlrpc_service *mdt_mdss_service; @@ -113,8 +113,10 @@ struct mdt_device { /* ptlrpc handle for MDS->client connections (for lock ASTs). */ struct ptlrpc_client *mdt_ldlm_client; /* underlying device */ + struct obd_export *mdt_child_exp; struct md_device *mdt_child; struct dt_device *mdt_bottom; + struct obd_export *mdt_bottom_exp; /** target device */ struct lu_target mdt_lut; /* @@ -131,7 +133,7 @@ struct mdt_device { /* mdt state flags */ unsigned long mdt_state; /* lock to protect IOepoch */ - cfs_spinlock_t mdt_ioepoch_lock; + spinlock_t mdt_ioepoch_lock; __u64 mdt_ioepoch; /* transaction callbacks */ @@ -145,7 +147,7 @@ struct mdt_device { struct upcall_cache *mdt_identity_cache; /* sptlrpc rules */ - cfs_rwlock_t mdt_sptlrpc_lock; + rwlock_t mdt_sptlrpc_lock; struct sptlrpc_rule_set mdt_sptlrpc_rset; /* capability keys */ @@ -157,8 +159,14 @@ struct mdt_device { cfs_timer_t mdt_ck_timer; struct ptlrpc_thread mdt_ck_thread; struct lustre_capa_key mdt_capa_keys[2]; - unsigned int mdt_capa_conf:1, - mdt_som_conf:1; + unsigned int mdt_capa_conf:1, + mdt_som_conf:1, + mdt_enable_remote_dir:1; + + /* statfs optimization: we cache a bit */ + struct obd_statfs mdt_osfs; + __u64 mdt_osfs_age; + spinlock_t mdt_osfs_lock; /* root squash */ uid_t mdt_squash_uid; @@ -166,19 +174,23 @@ struct mdt_device { cfs_list_t mdt_nosquash_nids; char *mdt_nosquash_str; int mdt_nosquash_strlen; - cfs_rw_semaphore_t mdt_squash_sem; + struct rw_semaphore mdt_squash_sem; - cfs_proc_dir_entry_t *mdt_proc_entry; - struct lprocfs_stats *mdt_stats; int mdt_sec_level; struct rename_stats mdt_rename_stats; + struct lu_fid mdt_md_root_fid; + + /* connection to quota master */ + struct obd_export *mdt_qmt_exp; + /* quota master device associated with this MDT */ + struct lu_device *mdt_qmt_dev; }; -#define MDT_SERVICE_WATCHDOG_FACTOR (2) -#define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID) -#define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \ - OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \ - OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI) +#define MDT_SERVICE_WATCHDOG_FACTOR (2) +#define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID) +#define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \ + OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \ + OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI) #define MDT_COS_DEFAULT (0) struct mdt_object { @@ -189,9 +201,9 @@ struct mdt_object { int mot_ioepoch_count; int mot_writecount; /* Lock to protect object's IO epoch. */ - cfs_mutex_t mot_ioepoch_mutex; + struct mutex mot_ioepoch_mutex; /* Lock to protect create_data */ - cfs_mutex_t mot_lov_mutex; + struct mutex mot_lov_mutex; }; enum mdt_object_flags { @@ -232,6 +244,7 @@ enum { MDT_LH_PARENT, /* parent lockh */ MDT_LH_CHILD, /* child lockh */ MDT_LH_OLD, /* old lockh for rename */ + MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */ MDT_LH_NEW, /* new lockh for rename */ MDT_LH_RMT, /* used for return lh to caller */ MDT_LH_NR @@ -262,6 +275,57 @@ enum mdt_reint_flag { MRF_OPEN_TRUNC = 1 << 0, }; +struct mdt_thread_info; +struct tx_arg; +typedef int (*tx_exec_func_t)(struct mdt_thread_info *, struct thandle *, + struct tx_arg *); + +struct tx_arg { + tx_exec_func_t exec_fn; + tx_exec_func_t undo_fn; + struct dt_object *object; + char *file; + int line; + struct update_reply *reply; + int index; + union { + struct { + const struct dt_rec *rec; + const struct dt_key *key; + } insert; + struct { + } ref; + struct { + struct lu_attr attr; + } attr_set; + struct { + struct lu_buf buf; + const char *name; + int flags; + __u32 csum; + } xattr_set; + struct { + struct lu_attr attr; + struct dt_allocation_hint hint; + struct dt_object_format dof; + struct lu_fid fid; + } create; + struct { + struct lu_buf buf; + loff_t pos; + } write; + } u; +}; + +#define TX_MAX_OPS 10 +struct thandle_exec_args { + struct thandle *ta_handle; + struct dt_device *ta_dev; + int ta_err; + struct tx_arg ta_args[TX_MAX_OPS]; + int ta_argno; /* used args */ +}; + /* * Common data shared by mdt-level handlers. This is allocated per-thread to * reduce stack consumption. @@ -323,7 +387,6 @@ struct mdt_thread_info { const struct ldlm_request *mti_dlm_req; __u32 mti_has_trans:1, /* has txn already? */ - mti_no_need_trans:1, mti_cross_ref:1; /* opdata for mdt_reint_open(), has the same as @@ -371,7 +434,7 @@ struct mdt_thread_info { struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */ char ns_name[48]; /* for mdt_init0() */ struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */ - cfs_kstatfs_t ksfs; /* for mdt_statfs() */ + struct obd_statfs osfs; /* for mdt_statfs() */ struct { /* for mdt_readpage() */ struct lu_rdpg mti_rdpg; @@ -382,33 +445,90 @@ struct mdt_thread_info { struct md_attr attr; struct md_som_data data; } som; + struct { + struct dt_object_format mti_update_dof; + struct update_reply *mti_update_reply; + struct update *mti_update; + int mti_update_reply_index; + struct obdo mti_obdo; + struct dt_object *mti_dt_object; + } update; } mti_u; /* IO epoch related stuff. */ struct mdt_ioepoch *mti_ioepoch; __u64 mti_replayepoch; - /* server and client data buffers */ - struct lr_server_data mti_lsd; - struct lsd_client_data mti_lcd; loff_t mti_off; struct lu_buf mti_buf; struct lustre_capa_key mti_capa_key; /* Ops object filename */ struct lu_name mti_name; + /* per-thread values, can be re-used */ + void *mti_big_lmm; + int mti_big_lmmsize; + /* big_lmm buffer was used and must be used in reply */ + int mti_big_lmm_used; + /* should be enough to fit lustre_mdt_attrs */ + char mti_xattr_buf[128]; + struct thandle_exec_args mti_handle; + struct ldlm_enqueue_info mti_einfo; +}; + +/* ptlrpc request handler for MDT. All handlers are + * grouped into several slices - struct mdt_opc_slice, + * and stored in an array - mdt_handlers[]. + */ +struct mdt_handler { + /* The name of this handler. */ + const char *mh_name; + /* Fail id for this handler, checked at the beginning of this handler*/ + int mh_fail_id; + /* Operation code for this handler */ + __u32 mh_opc; + /* flags are listed in enum mdt_handler_flags below. */ + __u32 mh_flags; + /* The actual handler function to execute. */ + int (*mh_act)(struct mdt_thread_info *info); + /* Request format for this request. */ + const struct req_format *mh_fmt; }; -typedef void (*mdt_cb_t)(const struct mdt_device *mdt, __u64 transno, - void *data, int err); -struct mdt_commit_cb { - mdt_cb_t mdt_cb_func; - void *mdt_cb_data; +enum mdt_handler_flags { + /* + * struct mdt_body is passed in the incoming message, and object + * identified by this fid exists on disk. + * + * "habeo corpus" == "I have a body" + */ + HABEO_CORPUS = (1 << 0), + /* + * struct ldlm_request is passed in the incoming message. + * + * "habeo clavis" == "I have a key" + */ + HABEO_CLAVIS = (1 << 1), + /* + * this request has fixed reply format, so that reply message can be + * packed by generic code. + * + * "habeo refero" == "I have a reply" + */ + HABEO_REFERO = (1 << 2), + /* + * this request will modify something, so check whether the filesystem + * is readonly or not, then return -EROFS to client asap if necessary. + * + * "mutabor" == "I shall modify" + */ + MUTABOR = (1 << 3) }; -enum mdt_txn_op { - MDT_TXN_CAPA_KEYS_WRITE_OP, - MDT_TXN_LAST_RCVD_WRITE_OP, +struct mdt_opc_slice { + __u32 mos_opc_start; + int mos_opc_end; + struct mdt_handler *mos_hs; }; static inline const struct md_device_operations * @@ -477,9 +597,9 @@ static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt) return mdt->mdt_md_dev.md_lu_dev.ld_site; } -static inline struct md_site *mdt_md_site(const struct mdt_device *mdt) +static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt) { - return lu_site2md(mdt_lu_site(mdt)); + return &mdt->mdt_seq_site; } static inline void mdt_export_evict(struct obd_export *exp) @@ -510,11 +630,19 @@ int mdt_object_lock(struct mdt_thread_info *, struct mdt_lock_handle *, __u64, int); +int mdt_object_lock_try(struct mdt_thread_info *, + struct mdt_object *, + struct mdt_lock_handle *, + __u64, int); + void mdt_object_unlock(struct mdt_thread_info *, struct mdt_object *, struct mdt_lock_handle *, int decref); +struct mdt_object *mdt_object_new(const struct lu_env *, + struct mdt_device *, + const struct lu_fid *); struct mdt_object *mdt_object_find(const struct lu_env *, struct mdt_device *, const struct lu_fid *); @@ -553,14 +681,6 @@ int mdt_fs_setup(const struct lu_env *, struct mdt_device *, struct obd_device *, struct lustre_sb_info *lsi); void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *); -int mdt_client_del(const struct lu_env *env, - struct mdt_device *mdt); -int mdt_client_add(const struct lu_env *env, - struct mdt_device *mdt, - int cl_idx); -int mdt_client_new(const struct lu_env *env, - struct mdt_device *mdt); - int mdt_export_stats_init(struct obd_device *obd, struct obd_export *exp, void *client_nid); @@ -591,6 +711,8 @@ enum { MDT_SOM_ENABLE = 1, }; +int mdt_attr_get_complex(struct mdt_thread_info *info, + struct mdt_object *o, struct md_attr *ma); int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o, int created); int mdt_object_is_som_enabled(struct mdt_object *mo); @@ -603,24 +725,14 @@ void mdt_mfd_free(struct mdt_file_data *mfd); int mdt_close(struct mdt_thread_info *info); int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo, struct md_attr *ma, int flags); +int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo, + struct md_attr *ma); int mdt_done_writing(struct mdt_thread_info *info); int mdt_fix_reply(struct mdt_thread_info *info); int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *, const struct md_attr *); void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *); -struct thandle *mdt_trans_create(const struct lu_env *env, - struct mdt_device *mdt); -int mdt_trans_start(const struct lu_env *env, struct mdt_device *mdt, - struct thandle *th); -void mdt_trans_stop(const struct lu_env *env, - struct mdt_device *mdt, struct thandle *th); -int mdt_record_write(const struct lu_env *env, - struct dt_object *dt, const struct lu_buf *buf, - loff_t *pos, struct thandle *th); -int mdt_record_read(const struct lu_env *env, - struct dt_object *dt, struct lu_buf *buf, loff_t *pos); - struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len); const struct lu_buf *mdt_buf_const(const struct lu_env *env, const void *area, ssize_t len); @@ -635,7 +747,51 @@ int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int); void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int); int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *, int); - +int mdt_handle_common(struct ptlrpc_request *req, + struct mdt_opc_slice *supported); +int mdt_connect(struct mdt_thread_info *info); +int mdt_disconnect(struct mdt_thread_info *info); +int mdt_set_info(struct mdt_thread_info *info); +int mdt_get_info(struct mdt_thread_info *info); +int mdt_getstatus(struct mdt_thread_info *info); +int mdt_getattr(struct mdt_thread_info *info); +int mdt_getattr_name(struct mdt_thread_info *info); +int mdt_statfs(struct mdt_thread_info *info); +int mdt_reint(struct mdt_thread_info *info); +int mdt_sync(struct mdt_thread_info *info); +int mdt_is_subdir(struct mdt_thread_info *info); +int mdt_obd_ping(struct mdt_thread_info *info); +int mdt_obd_log_cancel(struct mdt_thread_info *info); +int mdt_obd_qc_callback(struct mdt_thread_info *info); +int mdt_enqueue(struct mdt_thread_info *info); +int mdt_convert(struct mdt_thread_info *info); +int mdt_bl_callback(struct mdt_thread_info *info); +int mdt_cp_callback(struct mdt_thread_info *info); +int mdt_llog_create(struct mdt_thread_info *info); +int mdt_llog_destroy(struct mdt_thread_info *info); +int mdt_llog_read_header(struct mdt_thread_info *info); +int mdt_llog_next_block(struct mdt_thread_info *info); +int mdt_llog_prev_block(struct mdt_thread_info *info); +int mdt_sec_ctx_handle(struct mdt_thread_info *info); +int mdt_readpage(struct mdt_thread_info *info); +int mdt_obd_idx_read(struct mdt_thread_info *info); + +extern struct mdt_opc_slice mdt_regular_handlers[]; +extern struct mdt_opc_slice mdt_seq_handlers[]; +extern struct mdt_opc_slice mdt_fld_handlers[]; + +int mdt_quotacheck(struct mdt_thread_info *info); +int mdt_quotactl(struct mdt_thread_info *info); +int mdt_quota_dqacq(struct mdt_thread_info *info); + +extern struct lprocfs_vars lprocfs_mds_module_vars[]; +extern struct lprocfs_vars lprocfs_mds_obd_vars[]; + +int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj, + struct md_hsm *mh); + +struct mdt_handler *mdt_handler_find(__u32 opc, + struct mdt_opc_slice *supported); /* mdt_idmap.c */ int mdt_init_sec_level(struct mdt_thread_info *); int mdt_init_idmap(struct mdt_thread_info *); @@ -680,6 +836,15 @@ __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t); int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *); +/* mdt/mdt_hsm.c */ +int mdt_hsm_state_get(struct mdt_thread_info *info); +int mdt_hsm_state_set(struct mdt_thread_info *info); +int mdt_hsm_action(struct mdt_thread_info *info); +int mdt_hsm_progress(struct mdt_thread_info *info); +int mdt_hsm_ct_register(struct mdt_thread_info *info); +int mdt_hsm_ct_unregister(struct mdt_thread_info *info); +int mdt_hsm_request(struct mdt_thread_info *info); + extern struct lu_context_key mdt_thread_key; /* debug issues helper starts here*/ static inline int mdt_fail_write(const struct lu_env *env, @@ -720,7 +885,8 @@ static inline int mdt_check_resent(struct mdt_thread_info *info, RETURN(0); } -struct md_ucred *mdt_ucred(const struct mdt_thread_info *info); +struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info); +struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info); static inline int is_identity_get_disabled(struct upcall_cache *cache) { @@ -735,7 +901,7 @@ static inline int mdt_fid_lock(struct ldlm_namespace *ns, ldlm_mode_t mode, ldlm_policy_data_t *policy, const struct ldlm_res_id *res_id, - int flags, const __u64 *client_cookie) + __u64 flags, const __u64 *client_cookie) { int rc; @@ -745,7 +911,7 @@ static inline int mdt_fid_lock(struct ldlm_namespace *ns, rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy, mode, &flags, mdt_blocking_ast, ldlm_completion_ast, NULL, NULL, 0, - client_cookie, lh); + LVB_T_NONE, client_cookie, lh); return rc == ELDLM_OK ? 0 : -EIO; } @@ -770,6 +936,9 @@ static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode) return mdt_dlm_lock_modes[mode]; } +/* mdt_lvb.c */ +extern struct ldlm_valblock_ops mdt_lvbo; + static inline struct lu_name *mdt_name(const struct lu_env *env, char *name, int namelen) { @@ -822,17 +991,16 @@ enum { LPROC_MDT_CROSSDIR_RENAME, LPROC_MDT_LAST, }; -void mdt_counter_incr(struct obd_export *exp, int opcode); +void mdt_counter_incr(struct ptlrpc_request *req, int opcode); void mdt_stats_counter_init(struct lprocfs_stats *stats); void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars); +void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars); int mdt_procfs_init(struct mdt_device *mdt, const char *name); int mdt_procfs_fini(struct mdt_device *mdt); void mdt_rename_counter_tally(struct mdt_thread_info *info, - struct mdt_device *mdt, struct obd_export *exp, - struct mdt_object *src, struct mdt_object *tgt); - -void mdt_time_start(const struct mdt_thread_info *info); -void mdt_time_end(const struct mdt_thread_info *info, int idx); + struct mdt_device *mdt, + struct ptlrpc_request *req, + struct mdt_object *src, struct mdt_object *tgt); /* Capability */ int mdt_ck_thread_start(struct mdt_device *mdt); @@ -846,7 +1014,7 @@ static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset, { struct md_capainfo *ci; - LASSERT(offset >= 0 && offset <= MD_CAPAINFO_MAX); + LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX); if (!info->mti_mdt->mdt_opts.mo_mds_capa || !(info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA)) return; @@ -883,5 +1051,58 @@ static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt) { return mdt->mdt_md_dev.md_lu_dev.ld_obd; } + +extern const struct lu_device_operations mdt_lu_ops; + +static inline int lu_device_is_mdt(struct lu_device *d) +{ + return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops); +} + +static inline struct mdt_device *lu2mdt_dev(struct lu_device *d) +{ + LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n", + d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops); + return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev); +} + +static inline char *mdt_obd_name(struct mdt_device *mdt) +{ + return mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name; +} + +int mds_mod_init(void); +void mds_mod_exit(void); + +/* Update handlers */ +int out_handle(struct mdt_thread_info *info); + +#define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \ + __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \ + __FILE__, __LINE__) + +#define out_tx_attr_set(info, obj, attr, th, reply, idx) \ + __out_tx_attr_set(info, obj, attr, th, reply, idx, \ + __FILE__, __LINE__) + +#define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \ + __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \ + __FILE__, __LINE__) + +#define out_tx_ref_add(info, obj, th, reply, idx) \ + __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__) + +#define out_tx_ref_del(info, obj, th, reply, idx) \ + __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__) + +#define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \ + __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \ + __FILE__, __LINE__) + +#define out_tx_index_delete(info, obj, th, name, reply, idx) \ + __out_tx_index_delete(info, obj, th, name, reply, idx, \ + __FILE__, __LINE__) + + #endif /* __KERNEL__ */ #endif /* _MDT_H */