X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ffld%2Ffld_handler.c;h=84d90cea0491cfe6187fe9d1eec1e93b7c74c8a4;hb=da430ce6173149b4bafcad71ea1d650d54ea39c0;hp=00c9d123001bb2af7afe0eb34a5da3d76d9b4246;hpb=1f8b7d9dd5ecf9eecaf52327f114201c3c171554;p=fs%2Flustre-release.git diff --git a/lustre/fld/fld_handler.c b/lustre/fld/fld_handler.c index 00c9d12..84d90ce 100644 --- a/lustre/fld/fld_handler.c +++ b/lustre/fld/fld_handler.c @@ -1,34 +1,47 @@ -/* -*- MODE: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: +/* + * GPL HEADER START * - * lustre/fld/fld_handler.c - * FLD (Fids Location Database) + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Copyright (C) 2006 Cluster File Systems, Inc. - * Author: Yury Umanets - * WangDi + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * GPL HEADER END */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif +/* + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. + * + * Copyright (c) 2011, 2013, Intel Corporation. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/fld/fld_handler.c + * + * FLD (Fids Location Database) + * + * Author: Yury Umanets + * Author: WangDi + * Author: Pravin Shelar + */ + #define DEBUG_SUBSYSTEM S_FLD #ifdef __KERNEL__ @@ -48,8 +61,10 @@ #include #include +#include #include #include "fld_internal.h" +#include #ifdef __KERNEL__ @@ -57,7 +72,7 @@ LU_KEY_INIT_FINI(fld, struct fld_thread_info); /* context key: fld_thread_key */ -LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD); +LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD); cfs_proc_dir_entry_t *fld_type_proc_dir = NULL; @@ -83,107 +98,117 @@ static void __exit fld_mod_exit(void) } } -/* Insert index entry and update cache. */ -int fld_server_create(struct lu_server_fld *fld, - const struct lu_env *env, - seqno_t seq, mdsno_t mds) +int fld_declare_server_create(const struct lu_env *env, + struct lu_server_fld *fld, + struct lu_seq_range *range, + struct thandle *th) { - int rc; - ENTRY; - - rc = fld_index_create(fld, env, seq, mds); - - if (rc == 0) { - /* - * Do not return result of calling fld_cache_insert() - * here. First of all because it may return -EEXISTS. Another - * reason is that, we do not want to stop proceeding even after - * cache errors. - */ - fld_cache_insert(fld->lsf_cache, seq, mds); - } + int rc; - RETURN(rc); + rc = fld_declare_index_create(env, fld, range, th); + RETURN(rc); } -EXPORT_SYMBOL(fld_server_create); +EXPORT_SYMBOL(fld_declare_server_create); -/* Delete index entry. */ -int fld_server_delete(struct lu_server_fld *fld, - const struct lu_env *env, - seqno_t seq) +/** + * Insert FLD index entry and update FLD cache. + * + * This function is called from the sequence allocator when a super-sequence + * is granted to a server. + */ +int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld, + struct lu_seq_range *range, struct thandle *th) { - int rc; - ENTRY; + int rc; - fld_cache_delete(fld->lsf_cache, seq); - rc = fld_index_delete(fld, env, seq); - - RETURN(rc); + mutex_lock(&fld->lsf_lock); + rc = fld_index_create(env, fld, range, th); + mutex_unlock(&fld->lsf_lock); + + RETURN(rc); } -EXPORT_SYMBOL(fld_server_delete); +EXPORT_SYMBOL(fld_server_create); -/* Lookup mds by seq. */ -int fld_server_lookup(struct lu_server_fld *fld, - const struct lu_env *env, - seqno_t seq, mdsno_t *mds) +/** + * 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) { - int rc; - ENTRY; - - /* Lookup it in the cache. */ - rc = fld_cache_lookup(fld->lsf_cache, seq, mds); - if (rc == 0) - RETURN(0); - - rc = fld_index_lookup(fld, env, seq, mds); - if (rc == 0) { - /* - * Do not return error here as well. See previous comment in - * same situation in function fld_server_create(). - */ - fld_cache_insert(fld->lsf_cache, seq, *mds); - } - RETURN(rc); + struct lu_seq_range *erange; + struct fld_thread_info *info; + int rc; + ENTRY; + + info = lu_context_key_get(&env->le_ctx, &fld_thread_key); + LASSERT(info != NULL); + erange = &info->fti_lrange; + + /* Lookup it in the cache. */ + rc = fld_cache_lookup(fld->lsf_cache, seq, erange); + if (rc == 0) { + if (unlikely(fld_range_type(erange) != fld_range_type(range) && + !fld_range_is_any(range))) { + CERROR("%s: FLD cache range "DRANGE" does not match" + "requested flag %x: rc = %d\n", fld->lsf_name, + PRANGE(erange), range->lsr_flags, -EIO); + RETURN(-EIO); + } + *range = *erange; + RETURN(0); + } + + if (fld->lsf_obj) { + /* 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); + } else { + LASSERT(fld->lsf_control_exp); + /* 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); + if (rc == 0) + fld_cache_insert(fld->lsf_cache, range); + } + RETURN(rc); } 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 md_fld *mf, + __u32 opc, struct lu_seq_range *range, struct fld_thread_info *info) { int rc; ENTRY; switch (opc) { - case FLD_CREATE: - rc = fld_server_create(fld, env, - mf->mf_seq, mf->mf_mds); - - /* Do not return -EEXIST error for resent case */ - if ((info->fti_flags & MSG_RESENT) && rc == -EEXIST) - rc = 0; - break; - case FLD_DELETE: - rc = fld_server_delete(fld, env, mf->mf_seq); - - /* Do not return -ENOENT error for resent case */ - if ((info->fti_flags & MSG_RESENT) && rc == -ENOENT) - rc = 0; - break; case FLD_LOOKUP: - rc = fld_server_lookup(fld, env, - mf->mf_seq, &mf->mf_mds); + rc = fld_server_lookup(env, fld, range->lsr_start, range); break; default: rc = -EINVAL; break; } - CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, seq: " - LPX64", mds: "LPU64")\n", fld->lsf_name, rc, opc, - mf->mf_seq, mf->mf_mds); - + CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: " + DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range)); + RETURN(rc); } @@ -191,59 +216,58 @@ static int fld_server_handle(struct lu_server_fld *fld, static int fld_req_handle(struct ptlrpc_request *req, struct fld_thread_info *info) { - struct lu_site *site; - struct md_fld *in; - struct md_fld *out; + struct obd_export *exp = req->rq_export; + 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; ENTRY; - site = req->rq_export->exp_obd->obd_lu_dev->ld_site; - - rc = req_capsule_pack(&info->fti_pill); + rc = req_capsule_server_pack(info->fti_pill); if (rc) RETURN(err_serious(rc)); - opc = req_capsule_client_get(&info->fti_pill, &RMF_FLD_OPC); + opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC); if (opc != NULL) { - in = req_capsule_client_get(&info->fti_pill, &RMF_FLD_MDFLD); + in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD); if (in == NULL) RETURN(err_serious(-EPROTO)); - out = req_capsule_server_get(&info->fti_pill, &RMF_FLD_MDFLD); + out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD); if (out == NULL) RETURN(err_serious(-EPROTO)); *out = *in; - rc = fld_server_handle(site->ls_server_fld, - req->rq_svc_thread->t_env, - *opc, out, info); - } else + /* 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, + req->rq_svc_thread->t_env, + *opc, out, info); + } else { rc = err_serious(-EPROTO); + } - RETURN(rc); + RETURN(rc); } static void fld_thread_info_init(struct ptlrpc_request *req, struct fld_thread_info *info) { - int i; - - info->fti_flags = lustre_msg_get_flags(req->rq_reqmsg); - - /* Mark rep buffer as req-layout stuff expects. */ - for (i = 0; i < ARRAY_SIZE(info->fti_rep_buf_size); i++) - info->fti_rep_buf_size[i] = -1; - + info->fti_pill = &req->rq_pill; /* Init request capsule. */ - req_capsule_init(&info->fti_pill, req, RCL_SERVER, - info->fti_rep_buf_size); - - req_capsule_set(&info->fti_pill, &RQF_FLD_QUERY); + req_capsule_init(info->fti_pill, req, RCL_SERVER); + req_capsule_set(info->fti_pill, &RQF_FLD_QUERY); } static void fld_thread_info_fini(struct fld_thread_info *info) { - req_capsule_fini(&info->fti_pill); + req_capsule_fini(info->fti_pill); } static int fld_handle(struct ptlrpc_request *req) @@ -270,7 +294,7 @@ static int fld_handle(struct ptlrpc_request *req) */ int fld_query(struct com_thread_info *info) { - return fld_handle(info->cti_pill.rc_req); + return fld_handle(info->cti_pill->rc_req); } EXPORT_SYMBOL(fld_query); @@ -282,21 +306,29 @@ EXPORT_SYMBOL(fld_query); * * fid_is_local() is supposed to be used in assertion checks only. */ -int fid_is_local(struct lu_site *site, const struct lu_fid *fid) +int fid_is_local(const struct lu_env *env, + struct lu_site *site, const struct lu_fid *fid) { - int result; - - result = 1; /* conservatively assume fid is local */ - if (site->ls_client_fld != NULL) { - mdsno_t mds; - int rc; - - rc = fld_cache_lookup(site->ls_client_fld->lcf_cache, - fid_seq(fid), &mds); - if (rc == 0) - result = (mds == site->ls_node_id); - } - return result; + int result; + struct seq_server_site *ss_site; + struct lu_seq_range *range; + struct fld_thread_info *info; + ENTRY; + + info = lu_context_key_get(&env->le_ctx, &fld_thread_key); + range = &info->fti_lrange; + + result = 1; /* conservatively assume fid is local */ + ss_site = lu_site2seq(site); + if (ss_site->ss_client_fld != NULL) { + int rc; + + rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache, + fid_seq(fid), range); + if (rc == 0) + result = (range->lsr_index == ss_site->ss_node_id); + } + return result; } EXPORT_SYMBOL(fid_is_local); @@ -316,7 +348,14 @@ static int fld_server_proc_init(struct lu_server_fld *fld) RETURN(rc); } - RETURN(rc); + rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444, + &fld_proc_seq_fops, fld); + if (rc) { + lprocfs_remove(&fld->lsf_proc_dir); + fld->lsf_proc_dir = NULL; + } + + RETURN(rc); } static void fld_server_proc_fini(struct lu_server_fld *fld) @@ -341,12 +380,13 @@ static void fld_server_proc_fini(struct lu_server_fld *fld) } #endif -int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt, - const char *prefix, const struct lu_env *env) +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) { - int cache_size, cache_threshold; - int rc; - ENTRY; + int cache_size, cache_threshold; + int rc; + ENTRY; snprintf(fld->lsf_name, sizeof(fld->lsf_name), "srv-%s", prefix); @@ -357,8 +397,8 @@ int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt, cache_threshold = cache_size * FLD_SERVER_CACHE_THRESHOLD / 100; + mutex_init(&fld->lsf_lock); fld->lsf_cache = fld_cache_init(fld->lsf_name, - FLD_SERVER_HTABLE_SIZE, cache_size, cache_threshold); if (IS_ERR(fld->lsf_cache)) { rc = PTR_ERR(fld->lsf_cache); @@ -366,41 +406,47 @@ int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt, GOTO(out, rc); } - rc = fld_index_init(fld, env, dt); - if (rc) - GOTO(out, rc); + if (!mds_node_id && type == LU_SEQ_RANGE_MDT) { + rc = fld_index_init(env, fld, dt); + if (rc) + GOTO(out, rc); + } else { + fld->lsf_obj = NULL; + } rc = fld_server_proc_init(fld); if (rc) GOTO(out, rc); - EXIT; + fld->lsf_control_exp = NULL; + + GOTO(out, rc); + out: - if (rc) - fld_server_fini(fld, env); - return rc; + if (rc) + fld_server_fini(env, fld); + return rc; } EXPORT_SYMBOL(fld_server_init); -void fld_server_fini(struct lu_server_fld *fld, - const struct lu_env *env) +void fld_server_fini(const struct lu_env *env, struct lu_server_fld *fld) { - ENTRY; + ENTRY; - fld_server_proc_fini(fld); - fld_index_fini(fld, env); + fld_server_proc_fini(fld); + fld_index_fini(env, fld); - if (fld->lsf_cache != NULL) { - if (!IS_ERR(fld->lsf_cache)) - fld_cache_fini(fld->lsf_cache); - fld->lsf_cache = NULL; - } + if (fld->lsf_cache != NULL) { + if (!IS_ERR(fld->lsf_cache)) + fld_cache_fini(fld->lsf_cache); + fld->lsf_cache = NULL; + } - EXIT; + EXIT; } EXPORT_SYMBOL(fld_server_fini); -MODULE_AUTHOR("Cluster File Systems, Inc. "); +MODULE_AUTHOR("Sun Microsystems, Inc. "); MODULE_DESCRIPTION("Lustre FLD"); MODULE_LICENSE("GPL");