4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/fld/fld_handler.c
38 * FLD (Fids Location Database)
40 * Author: Yury Umanets <umka@clusterfs.com>
41 * Author: WangDi <wangdi@clusterfs.com>
42 * Author: Pravin Shelar <pravin.shelar@sun.com>
45 #define DEBUG_SUBSYSTEM S_FLD
47 # include <linux/libcfs/libcfs.h>
48 # include <linux/module.h>
49 # include <linux/jbd.h>
50 # include <asm/div64.h>
53 #include <obd_class.h>
54 #include <lustre_ver.h>
55 #include <obd_support.h>
56 #include <lprocfs_status.h>
58 #include <md_object.h>
59 #include <lustre_fid.h>
60 #include <lustre_req_layout.h>
61 #include "fld_internal.h"
62 #include <lustre_fid.h>
65 /* context key constructor/destructor: fld_key_init, fld_key_fini */
66 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
68 /* context key: fld_thread_key */
69 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD);
71 proc_dir_entry_t *fld_type_proc_dir = NULL;
73 static int __init fld_mod_init(void)
75 fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
78 if (IS_ERR(fld_type_proc_dir))
79 return PTR_ERR(fld_type_proc_dir);
81 LU_CONTEXT_KEY_INIT(&fld_thread_key);
82 lu_context_key_register(&fld_thread_key);
86 static void __exit fld_mod_exit(void)
88 lu_context_key_degister(&fld_thread_key);
89 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
90 lprocfs_remove(&fld_type_proc_dir);
91 fld_type_proc_dir = NULL;
95 int fld_declare_server_create(const struct lu_env *env,
96 struct lu_server_fld *fld,
97 struct lu_seq_range *range,
102 rc = fld_declare_index_create(env, fld, range, th);
105 EXPORT_SYMBOL(fld_declare_server_create);
108 * Insert FLD index entry and update FLD cache.
110 * This function is called from the sequence allocator when a super-sequence
111 * is granted to a server.
113 int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld,
114 struct lu_seq_range *range, struct thandle *th)
118 mutex_lock(&fld->lsf_lock);
119 rc = fld_index_create(env, fld, range, th);
120 mutex_unlock(&fld->lsf_lock);
124 EXPORT_SYMBOL(fld_server_create);
127 * Lookup mds by seq, returns a range for given seq.
129 * If that entry is not cached in fld cache, request is sent to super
130 * sequence controller node (MDT0). All other MDT[1...N] and client
131 * cache fld entries, but this cache is not persistent.
133 int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld,
134 seqno_t seq, struct lu_seq_range *range)
136 struct lu_seq_range *erange;
137 struct fld_thread_info *info;
141 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
142 LASSERT(info != NULL);
143 erange = &info->fti_lrange;
145 /* Lookup it in the cache. */
146 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
148 if (unlikely(fld_range_type(erange) != fld_range_type(range) &&
149 !fld_range_is_any(range))) {
150 CERROR("%s: FLD cache range "DRANGE" does not match"
151 "requested flag %x: rc = %d\n", fld->lsf_name,
152 PRANGE(erange), range->lsr_flags, -EIO);
160 /* On server side, all entries should be in cache.
161 * If we can not find it in cache, just return error */
162 CERROR("%s: Cannot find sequence "LPX64": rc = %d\n",
163 fld->lsf_name, seq, -EIO);
166 LASSERT(fld->lsf_control_exp);
167 /* send request to mdt0 i.e. super seq. controller.
168 * This is temporary solution, long term solution is fld
169 * replication on all mdt servers.
171 range->lsr_start = seq;
172 rc = fld_client_rpc(fld->lsf_control_exp,
175 fld_cache_insert(fld->lsf_cache, range);
179 EXPORT_SYMBOL(fld_server_lookup);
182 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
183 * if entry is not found in cache we need to forward lookup request to MDT0
186 static int fld_server_handle(struct lu_server_fld *fld,
187 const struct lu_env *env,
188 __u32 opc, struct lu_seq_range *range,
189 struct fld_thread_info *info)
196 rc = fld_server_lookup(env, fld, range->lsr_start, range);
203 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
204 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
210 static int fld_req_handle(struct ptlrpc_request *req,
211 struct fld_thread_info *info)
213 struct obd_export *exp = req->rq_export;
214 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
215 struct lu_seq_range *in;
216 struct lu_seq_range *out;
221 rc = req_capsule_server_pack(info->fti_pill);
223 RETURN(err_serious(rc));
225 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
227 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
229 RETURN(err_serious(-EPROTO));
230 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
232 RETURN(err_serious(-EPROTO));
235 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
236 * Set it as 'LU_SEQ_RANGE_MDT' by default. */
237 if (!(exp_connect_flags(exp) & OBD_CONNECT_64BITHASH) &&
238 !(exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
239 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) &&
241 fld_range_set_mdt(out);
243 rc = fld_server_handle(lu_site2seq(site)->ss_server_fld,
244 req->rq_svc_thread->t_env,
247 rc = err_serious(-EPROTO);
253 static void fld_thread_info_init(struct ptlrpc_request *req,
254 struct fld_thread_info *info)
256 info->fti_pill = &req->rq_pill;
257 /* Init request capsule. */
258 req_capsule_init(info->fti_pill, req, RCL_SERVER);
259 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
262 static void fld_thread_info_fini(struct fld_thread_info *info)
264 req_capsule_fini(info->fti_pill);
267 static int fld_handle(struct ptlrpc_request *req)
269 struct fld_thread_info *info;
270 const struct lu_env *env;
273 env = req->rq_svc_thread->t_env;
274 LASSERT(env != NULL);
276 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
277 LASSERT(info != NULL);
279 fld_thread_info_init(req, info);
280 rc = fld_req_handle(req, info);
281 fld_thread_info_fini(info);
287 * Entry point for handling FLD RPCs called from MDT.
289 int fld_query(struct com_thread_info *info)
291 return fld_handle(info->cti_pill->rc_req);
293 EXPORT_SYMBOL(fld_query);
296 * Returns true, if fid is local to this server node.
298 * WARNING: this function is *not* guaranteed to return false if fid is
299 * remote: it makes an educated conservative guess only.
301 * fid_is_local() is supposed to be used in assertion checks only.
303 int fid_is_local(const struct lu_env *env,
304 struct lu_site *site, const struct lu_fid *fid)
307 struct seq_server_site *ss_site;
308 struct lu_seq_range *range;
309 struct fld_thread_info *info;
312 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
313 range = &info->fti_lrange;
315 result = 1; /* conservatively assume fid is local */
316 ss_site = lu_site2seq(site);
317 if (ss_site->ss_client_fld != NULL) {
320 rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache,
321 fid_seq(fid), range);
323 result = (range->lsr_index == ss_site->ss_node_id);
327 EXPORT_SYMBOL(fid_is_local);
329 static void fld_server_proc_fini(struct lu_server_fld *fld);
332 static int fld_server_proc_init(struct lu_server_fld *fld)
337 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
339 fld_server_proc_list, fld);
340 if (IS_ERR(fld->lsf_proc_dir)) {
341 rc = PTR_ERR(fld->lsf_proc_dir);
345 rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444,
346 &fld_proc_seq_fops, fld);
348 lprocfs_remove(&fld->lsf_proc_dir);
349 fld->lsf_proc_dir = NULL;
355 static void fld_server_proc_fini(struct lu_server_fld *fld)
358 if (fld->lsf_proc_dir != NULL) {
359 if (!IS_ERR(fld->lsf_proc_dir))
360 lprocfs_remove(&fld->lsf_proc_dir);
361 fld->lsf_proc_dir = NULL;
366 static int fld_server_proc_init(struct lu_server_fld *fld)
371 static void fld_server_proc_fini(struct lu_server_fld *fld)
377 int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld,
378 struct dt_device *dt, const char *prefix, int mds_node_id,
381 int cache_size, cache_threshold;
385 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
388 cache_size = FLD_SERVER_CACHE_SIZE /
389 sizeof(struct fld_cache_entry);
391 cache_threshold = cache_size *
392 FLD_SERVER_CACHE_THRESHOLD / 100;
394 mutex_init(&fld->lsf_lock);
395 fld->lsf_cache = fld_cache_init(fld->lsf_name,
396 cache_size, cache_threshold);
397 if (IS_ERR(fld->lsf_cache)) {
398 rc = PTR_ERR(fld->lsf_cache);
399 fld->lsf_cache = NULL;
403 if (!mds_node_id && type == LU_SEQ_RANGE_MDT) {
404 rc = fld_index_init(env, fld, dt);
411 rc = fld_server_proc_init(fld);
415 fld->lsf_control_exp = NULL;
421 fld_server_fini(env, fld);
424 EXPORT_SYMBOL(fld_server_init);
426 void fld_server_fini(const struct lu_env *env, struct lu_server_fld *fld)
430 fld_server_proc_fini(fld);
431 fld_index_fini(env, fld);
433 if (fld->lsf_cache != NULL) {
434 if (!IS_ERR(fld->lsf_cache))
435 fld_cache_fini(fld->lsf_cache);
436 fld->lsf_cache = NULL;
441 EXPORT_SYMBOL(fld_server_fini);
443 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
444 MODULE_DESCRIPTION("Lustre FLD");
445 MODULE_LICENSE("GPL");
447 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);