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 <libcfs/libcfs.h>
48 #include <linux/module.h>
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_fld.h>
54 #include <md_object.h>
55 #include <lustre_req_layout.h>
56 #include <lprocfs_status.h>
57 #include "fld_internal.h"
59 /* context key constructor/destructor: fld_key_init, fld_key_fini */
60 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
62 /* context key: fld_thread_key */
63 /* MGS thread may create llog file causing FLD lookup */
64 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD);
66 int fld_server_mod_init(void)
68 LU_CONTEXT_KEY_INIT(&fld_thread_key);
69 return lu_context_key_register(&fld_thread_key);
72 void fld_server_mod_exit(void)
74 lu_context_key_degister(&fld_thread_key);
77 int fld_declare_server_create(const struct lu_env *env,
78 struct lu_server_fld *fld,
79 const struct lu_seq_range *range,
84 rc = fld_declare_index_create(env, fld, range, th);
87 EXPORT_SYMBOL(fld_declare_server_create);
90 * Insert FLD index entry and update FLD cache.
92 * This function is called from the sequence allocator when a super-sequence
93 * is granted to a server.
95 int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld,
96 const struct lu_seq_range *range, struct thandle *th)
100 mutex_lock(&fld->lsf_lock);
101 rc = fld_index_create(env, fld, range, th);
102 mutex_unlock(&fld->lsf_lock);
106 EXPORT_SYMBOL(fld_server_create);
109 * Lookup mds by seq, returns a range for given seq.
111 * If that entry is not cached in fld cache, request is sent to super
112 * sequence controller node (MDT0). All other MDT[1...N] and client
113 * cache fld entries, but this cache is not persistent.
115 int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld,
116 seqno_t seq, struct lu_seq_range *range)
118 struct lu_seq_range *erange;
119 struct fld_thread_info *info;
123 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
124 LASSERT(info != NULL);
125 erange = &info->fti_lrange;
127 /* Lookup it in the cache. */
128 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
130 if (unlikely(fld_range_type(erange) != fld_range_type(range) &&
131 !fld_range_is_any(range))) {
132 CERROR("%s: FLD cache range "DRANGE" does not match"
133 "requested flag %x: rc = %d\n", fld->lsf_name,
134 PRANGE(erange), range->lsr_flags, -EIO);
142 /* On server side, all entries should be in cache.
143 * If we can not find it in cache, just return error */
144 CERROR("%s: Cannot find sequence "LPX64": rc = %d\n",
145 fld->lsf_name, seq, -EIO);
148 if (fld->lsf_control_exp == NULL) {
149 CERROR("%s: lookup "LPX64", but not connects to MDT0"
150 "yet: rc = %d.\n", fld->lsf_name, seq, -EIO);
153 /* send request to mdt0 i.e. super seq. controller.
154 * This is temporary solution, long term solution is fld
155 * replication on all mdt servers.
157 range->lsr_start = seq;
158 rc = fld_client_rpc(fld->lsf_control_exp,
161 fld_cache_insert(fld->lsf_cache, range);
165 EXPORT_SYMBOL(fld_server_lookup);
168 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
169 * if entry is not found in cache we need to forward lookup request to MDT0
171 static int fld_server_handle(struct lu_server_fld *fld,
172 const struct lu_env *env,
173 __u32 opc, struct lu_seq_range *range)
181 rc = fld_server_lookup(env, fld, range->lsr_start, range);
188 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
189 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
194 static int fld_handler(struct tgt_session_info *tsi)
196 struct obd_export *exp = tsi->tsi_exp;
197 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
198 struct lu_seq_range *in;
199 struct lu_seq_range *out;
205 opc = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_OPC);
207 in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
209 RETURN(err_serious(-EPROTO));
210 out = req_capsule_server_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
212 RETURN(err_serious(-EPROTO));
215 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
216 * Set it as 'LU_SEQ_RANGE_MDT' by default. */
217 if (!(exp_connect_flags(exp) & OBD_CONNECT_64BITHASH) &&
218 !(exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
219 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) &&
221 fld_range_set_mdt(out);
223 rc = fld_server_handle(lu_site2seq(site)->ss_server_fld,
224 tsi->tsi_env, *opc, out);
226 rc = err_serious(-EPROTO);
233 * Returns true, if fid is local to this server node.
235 * WARNING: this function is *not* guaranteed to return false if fid is
236 * remote: it makes an educated conservative guess only.
238 * fid_is_local() is supposed to be used in assertion checks only.
240 int fid_is_local(const struct lu_env *env,
241 struct lu_site *site, const struct lu_fid *fid)
244 struct seq_server_site *ss_site;
245 struct lu_seq_range *range;
246 struct fld_thread_info *info;
249 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
250 range = &info->fti_lrange;
252 result = 1; /* conservatively assume fid is local */
253 ss_site = lu_site2seq(site);
254 if (ss_site->ss_client_fld != NULL) {
257 rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache,
258 fid_seq(fid), range);
260 result = (range->lsr_index == ss_site->ss_node_id);
264 EXPORT_SYMBOL(fid_is_local);
266 static void fld_server_proc_fini(struct lu_server_fld *fld);
269 static int fld_server_proc_init(struct lu_server_fld *fld)
274 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
276 fld_server_proc_list, fld);
277 if (IS_ERR(fld->lsf_proc_dir)) {
278 rc = PTR_ERR(fld->lsf_proc_dir);
282 rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444,
283 &fld_proc_seq_fops, fld);
285 lprocfs_remove(&fld->lsf_proc_dir);
286 fld->lsf_proc_dir = NULL;
292 static void fld_server_proc_fini(struct lu_server_fld *fld)
295 if (fld->lsf_proc_dir != NULL) {
296 if (!IS_ERR(fld->lsf_proc_dir))
297 lprocfs_remove(&fld->lsf_proc_dir);
298 fld->lsf_proc_dir = NULL;
303 static int fld_server_proc_init(struct lu_server_fld *fld)
308 static void fld_server_proc_fini(struct lu_server_fld *fld)
314 int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld,
315 struct dt_device *dt, const char *prefix, int mds_node_id,
318 int cache_size, cache_threshold;
323 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
326 cache_size = FLD_SERVER_CACHE_SIZE / sizeof(struct fld_cache_entry);
328 cache_threshold = cache_size * FLD_SERVER_CACHE_THRESHOLD / 100;
330 mutex_init(&fld->lsf_lock);
331 fld->lsf_cache = fld_cache_init(fld->lsf_name, cache_size,
333 if (IS_ERR(fld->lsf_cache)) {
334 rc = PTR_ERR(fld->lsf_cache);
335 fld->lsf_cache = NULL;
339 if (!mds_node_id && type == LU_SEQ_RANGE_MDT) {
340 rc = fld_index_init(env, fld, dt);
347 rc = fld_server_proc_init(fld);
351 fld->lsf_control_exp = NULL;
355 fld_index_fini(env, fld);
357 fld_cache_fini(fld->lsf_cache);
360 EXPORT_SYMBOL(fld_server_init);
362 void fld_server_fini(const struct lu_env *env, struct lu_server_fld *fld)
366 fld_server_proc_fini(fld);
367 fld_index_fini(env, fld);
369 if (fld->lsf_cache != NULL) {
370 if (!IS_ERR(fld->lsf_cache))
371 fld_cache_fini(fld->lsf_cache);
372 fld->lsf_cache = NULL;
377 EXPORT_SYMBOL(fld_server_fini);
379 struct tgt_handler fld_handlers[] = {
380 TGT_FLD_HDL(HABEO_REFERO, FLD_QUERY, fld_handler),
382 EXPORT_SYMBOL(fld_handlers);