X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Ffld%2Ffld_handler.c;h=eae79eaea88d36ab4b206481c1d40f735ab1d3a3;hb=63851b5816bb30687fbf3750380d6b448e9400f1;hp=661381752d2efd4e8207b6dbcd6234405c7dd2b8;hpb=641d2da14c3c00ae2f63d9814f9b1f4ef4040f1c;p=fs%2Flustre-release.git diff --git a/lustre/fld/fld_handler.c b/lustre/fld/fld_handler.c index 6613817..eae79ea 100644 --- a/lustre/fld/fld_handler.c +++ b/lustre/fld/fld_handler.c @@ -51,7 +51,7 @@ #include #include #include -#include /* err_serious() */ +#include #include #include #include "fld_internal.h" @@ -106,14 +106,105 @@ int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld, EXPORT_SYMBOL(fld_server_create); /** - * Lookup mds by seq, returns a range for given seq. - * - * If that entry is not cached in fld cache, request is sent to super - * sequence controller node (MDT0). All other MDT[1...N] and client - * cache fld entries, but this cache is not persistent. - */ -int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld, - seqno_t seq, struct lu_seq_range *range) + * Extract index information from fld name like srv-fsname-MDT0000 + **/ +int fld_name_to_index(const char *name, __u32 *index) +{ + char *dash; + int rc; + ENTRY; + + CDEBUG(D_INFO, "get index from %s\n", name); + dash = strrchr(name, '-'); + if (dash == NULL) + RETURN(-EINVAL); + dash++; + rc = target_name2index(dash, index, NULL); + RETURN(rc); +} + +/** + * Retrieve fldb entry from MDT0 and add to local FLDB and cache. + **/ +int fld_update_from_controller(const struct lu_env *env, + struct lu_server_fld *fld) +{ + struct fld_thread_info *info; + struct lu_seq_range *range; + struct lu_seq_range_array *lsra; + __u32 index; + struct ptlrpc_request *req; + int rc; + int i; + ENTRY; + + /* Update only happens during initalization, i.e. local FLDB + * does not exist yet */ + if (!fld->lsf_new) + RETURN(0); + + rc = fld_name_to_index(fld->lsf_name, &index); + if (rc < 0) + RETURN(rc); + + /* No need update fldb for MDT0 */ + if (index == 0) + RETURN(0); + + info = lu_context_key_get(&env->le_ctx, &fld_thread_key); + LASSERT(info != NULL); + range = &info->fti_lrange; + memset(range, 0, sizeof(*range)); + range->lsr_index = index; + fld_range_set_mdt(range); + + do { + rc = fld_client_rpc(fld->lsf_control_exp, range, FLD_READ, + &req); + if (rc != 0 && rc != -EAGAIN) + GOTO(out, rc); + + LASSERT(req != NULL); + lsra = (struct lu_seq_range_array *)req_capsule_server_get( + &req->rq_pill, &RMF_GENERIC_DATA); + if (lsra == NULL) + GOTO(out, rc = -EPROTO); + + range_array_le_to_cpu(lsra, lsra); + for (i = 0; i < lsra->lsra_count; i++) { + int rc1; + + if (lsra->lsra_lsr[i].lsr_flags != LU_SEQ_RANGE_MDT) + GOTO(out, rc = -EINVAL); + + if (lsra->lsra_lsr[i].lsr_index != index) + GOTO(out, rc = -EINVAL); + + mutex_lock(&fld->lsf_lock); + rc1 = fld_insert_entry(env, fld, &lsra->lsra_lsr[i]); + mutex_unlock(&fld->lsf_lock); + + if (rc1 != 0) + GOTO(out, rc = rc1); + } + if (rc == -EAGAIN) + *range = lsra->lsra_lsr[lsra->lsra_count - 1]; + } while (rc == -EAGAIN); + + fld->lsf_new = 1; +out: + if (req != NULL) + ptlrpc_req_finished(req); + + RETURN(rc); +} +EXPORT_SYMBOL(fld_update_from_controller); + +/** + * Lookup sequece in local cache/fldb. + **/ +int fld_local_lookup(const struct lu_env *env, struct lu_server_fld *fld, + seqno_t seq, struct lu_seq_range *range) { struct lu_seq_range *erange; struct fld_thread_info *info; @@ -137,22 +228,51 @@ int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld, *range = *erange; RETURN(0); } + RETURN(rc); +} +EXPORT_SYMBOL(fld_local_lookup); + +/** + * Lookup MDT/OST by seq, returns a range for given seq. + * + * If that entry is not cached in fld cache, request is sent to super + * sequence controller node (MDT0). All other MDT[1...N] and client + * cache fld entries, but this cache is not persistent. + */ +int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld, + seqno_t seq, struct lu_seq_range *range) +{ + __u32 index; + int rc; + ENTRY; - if (fld->lsf_obj) { + rc = fld_local_lookup(env, fld, seq, range); + if (likely(rc == 0)) + RETURN(rc); + + rc = fld_name_to_index(fld->lsf_name, &index); + if (rc < 0) + RETURN(rc); + + if (index == 0 && rc == LDD_F_SV_TYPE_MDT) { /* On server side, all entries should be in cache. * If we can not find it in cache, just return error */ CERROR("%s: Cannot find sequence "LPX64": rc = %d\n", - fld->lsf_name, seq, -EIO); - RETURN(-EIO); + fld->lsf_name, seq, -ENOENT); + RETURN(-ENOENT); } else { - LASSERT(fld->lsf_control_exp); + if (fld->lsf_control_exp == NULL) { + CERROR("%s: lookup "LPX64", but not connects to MDT0" + "yet: rc = %d.\n", fld->lsf_name, seq, -EIO); + RETURN(-EIO); + } /* send request to mdt0 i.e. super seq. controller. * This is temporary solution, long term solution is fld * replication on all mdt servers. */ range->lsr_start = seq; rc = fld_client_rpc(fld->lsf_control_exp, - range, FLD_LOOKUP); + range, FLD_QUERY, NULL); if (rc == 0) fld_cache_insert(fld->lsf_cache, range); } @@ -164,63 +284,79 @@ EXPORT_SYMBOL(fld_server_lookup); * All MDT server handle fld lookup operation. But only MDT0 has fld index. * if entry is not found in cache we need to forward lookup request to MDT0 */ -static int fld_server_handle(struct lu_server_fld *fld, - const struct lu_env *env, - __u32 opc, struct lu_seq_range *range) +static int fld_handle_lookup(struct tgt_session_info *tsi) { - int rc; + struct obd_export *exp = tsi->tsi_exp; + struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site; + struct lu_server_fld *fld; + struct lu_seq_range *in; + struct lu_seq_range *out; + int rc; ENTRY; - switch (opc) { - case FLD_LOOKUP: - rc = fld_server_lookup(env, fld, range->lsr_start, range); - break; - default: - rc = -EINVAL; - break; - } + in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD); + if (in == NULL) + RETURN(err_serious(-EPROTO)); - CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: " - DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range)); + rc = req_capsule_server_pack(tsi->tsi_pill); + if (unlikely(rc != 0)) + RETURN(err_serious(rc)); + + out = req_capsule_server_get(tsi->tsi_pill, &RMF_FLD_MDFLD); + if (out == NULL) + RETURN(err_serious(-EPROTO)); + *out = *in; + + fld = lu_site2seq(site)->ss_server_fld; + + rc = fld_server_lookup(tsi->tsi_env, fld, in->lsr_start, out); + + CDEBUG(D_INFO, "%s: FLD req handle: error %d (range: "DRANGE")\n", + fld->lsf_name, rc, PRANGE(out)); RETURN(rc); } -static int fld_handler(struct tgt_session_info *tsi) +static int fld_handle_read(struct tgt_session_info *tsi) { struct obd_export *exp = tsi->tsi_exp; struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site; struct lu_seq_range *in; - struct lu_seq_range *out; - int rc; - __u32 *opc; + void *data; + int rc; ENTRY; - opc = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_OPC); - if (opc != NULL) { - in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD); - if (in == NULL) - RETURN(err_serious(-EPROTO)); - out = req_capsule_server_get(tsi->tsi_pill, &RMF_FLD_MDFLD); - if (out == NULL) - RETURN(err_serious(-EPROTO)); - *out = *in; - - /* For old 2.0 client, the 'lsr_flags' is uninitialized. - * Set it as 'LU_SEQ_RANGE_MDT' by default. */ - if (!(exp_connect_flags(exp) & OBD_CONNECT_64BITHASH) && - !(exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) && - !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) && - !exp->exp_libclient) - fld_range_set_mdt(out); - - rc = fld_server_handle(lu_site2seq(site)->ss_server_fld, - tsi->tsi_env, *opc, out); - } else { - rc = err_serious(-EPROTO); - } + req_capsule_set(tsi->tsi_pill, &RQF_FLD_READ); + + in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD); + if (in == NULL) + RETURN(err_serious(-EPROTO)); + + req_capsule_set_size(tsi->tsi_pill, &RMF_GENERIC_DATA, RCL_SERVER, + PAGE_CACHE_SIZE); + + rc = req_capsule_server_pack(tsi->tsi_pill); + if (unlikely(rc != 0)) + RETURN(err_serious(rc)); + + data = req_capsule_server_get(tsi->tsi_pill, &RMF_GENERIC_DATA); + + rc = fld_server_read(tsi->tsi_env, lu_site2seq(site)->ss_server_fld, + in, data, PAGE_CACHE_SIZE); + RETURN(rc); +} + +static int fld_handle_query(struct tgt_session_info *tsi) +{ + int rc; + + ENTRY; + + req_capsule_set(tsi->tsi_pill, &RQF_FLD_QUERY); + + rc = fld_handle_lookup(tsi); RETURN(rc); } @@ -267,13 +403,13 @@ static int fld_server_proc_init(struct lu_server_fld *fld) int rc = 0; ENTRY; - fld->lsf_proc_dir = lprocfs_register(fld->lsf_name, - fld_type_proc_dir, - fld_server_proc_list, fld); - if (IS_ERR(fld->lsf_proc_dir)) { - rc = PTR_ERR(fld->lsf_proc_dir); - RETURN(rc); - } + fld->lsf_proc_dir = lprocfs_seq_register(fld->lsf_name, + fld_type_proc_dir, + fld_server_proc_list, fld); + if (IS_ERR(fld->lsf_proc_dir)) { + rc = PTR_ERR(fld->lsf_proc_dir); + RETURN(rc); + } rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444, &fld_proc_seq_fops, fld); @@ -308,16 +444,14 @@ static void fld_server_proc_fini(struct lu_server_fld *fld) #endif int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld, - struct dt_device *dt, const char *prefix, int mds_node_id, - int type) + struct dt_device *dt, const char *prefix, int type) { int cache_size, cache_threshold; int rc; ENTRY; - snprintf(fld->lsf_name, sizeof(fld->lsf_name), - "srv-%s", prefix); + snprintf(fld->lsf_name, sizeof(fld->lsf_name), "srv-%s", prefix); cache_size = FLD_SERVER_CACHE_SIZE / sizeof(struct fld_cache_entry); @@ -332,13 +466,9 @@ int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld, RETURN(rc); } - if (!mds_node_id && type == LU_SEQ_RANGE_MDT) { - rc = fld_index_init(env, fld, dt); - if (rc) - GOTO(out_cache, rc); - } else { - fld->lsf_obj = NULL; - } + rc = fld_index_init(env, fld, dt, type); + if (rc) + GOTO(out_cache, rc); rc = fld_server_proc_init(fld); if (rc) @@ -373,6 +503,7 @@ void fld_server_fini(const struct lu_env *env, struct lu_server_fld *fld) EXPORT_SYMBOL(fld_server_fini); struct tgt_handler fld_handlers[] = { -TGT_FLD_HDL(HABEO_REFERO, FLD_QUERY, fld_handler), +TGT_FLD_HDL_VAR(0, FLD_QUERY, fld_handle_query), +TGT_FLD_HDL_VAR(0, FLD_READ, fld_handle_read), }; EXPORT_SYMBOL(fld_handlers);