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 <lustre_mdt.h> /* err_serious() */
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 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD);
65 int fld_server_mod_init(void)
67 LU_CONTEXT_KEY_INIT(&fld_thread_key);
68 return lu_context_key_register(&fld_thread_key);
71 void fld_server_mod_exit(void)
73 lu_context_key_degister(&fld_thread_key);
76 int fld_declare_server_create(const struct lu_env *env,
77 struct lu_server_fld *fld,
78 const struct lu_seq_range *range,
83 rc = fld_declare_index_create(env, fld, range, th);
86 EXPORT_SYMBOL(fld_declare_server_create);
89 * Insert FLD index entry and update FLD cache.
91 * This function is called from the sequence allocator when a super-sequence
92 * is granted to a server.
94 int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld,
95 const struct lu_seq_range *range, struct thandle *th)
99 mutex_lock(&fld->lsf_lock);
100 rc = fld_index_create(env, fld, range, th);
101 mutex_unlock(&fld->lsf_lock);
105 EXPORT_SYMBOL(fld_server_create);
108 * Lookup mds by seq, returns a range for given seq.
110 * If that entry is not cached in fld cache, request is sent to super
111 * sequence controller node (MDT0). All other MDT[1...N] and client
112 * cache fld entries, but this cache is not persistent.
114 int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld,
115 seqno_t seq, struct lu_seq_range *range)
117 struct lu_seq_range *erange;
118 struct fld_thread_info *info;
122 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
123 LASSERT(info != NULL);
124 erange = &info->fti_lrange;
126 /* Lookup it in the cache. */
127 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
129 if (unlikely(fld_range_type(erange) != fld_range_type(range) &&
130 !fld_range_is_any(range))) {
131 CERROR("%s: FLD cache range "DRANGE" does not match"
132 "requested flag %x: rc = %d\n", fld->lsf_name,
133 PRANGE(erange), range->lsr_flags, -EIO);
141 /* On server side, all entries should be in cache.
142 * If we can not find it in cache, just return error */
143 CERROR("%s: Cannot find sequence "LPX64": rc = %d\n",
144 fld->lsf_name, seq, -EIO);
147 LASSERT(fld->lsf_control_exp);
148 /* send request to mdt0 i.e. super seq. controller.
149 * This is temporary solution, long term solution is fld
150 * replication on all mdt servers.
152 range->lsr_start = seq;
153 rc = fld_client_rpc(fld->lsf_control_exp,
156 fld_cache_insert(fld->lsf_cache, range);
160 EXPORT_SYMBOL(fld_server_lookup);
163 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
164 * if entry is not found in cache we need to forward lookup request to MDT0
167 static int fld_server_handle(struct lu_server_fld *fld,
168 const struct lu_env *env,
169 __u32 opc, struct lu_seq_range *range,
170 struct fld_thread_info *info)
177 rc = fld_server_lookup(env, fld, range->lsr_start, range);
184 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
185 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
191 static int fld_req_handle(struct ptlrpc_request *req,
192 struct fld_thread_info *info)
194 struct obd_export *exp = req->rq_export;
195 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
196 struct lu_seq_range *in;
197 struct lu_seq_range *out;
202 rc = req_capsule_server_pack(info->fti_pill);
204 RETURN(err_serious(rc));
206 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
208 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
210 RETURN(err_serious(-EPROTO));
211 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
213 RETURN(err_serious(-EPROTO));
216 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
217 * Set it as 'LU_SEQ_RANGE_MDT' by default. */
218 if (!(exp_connect_flags(exp) & OBD_CONNECT_64BITHASH) &&
219 !(exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
220 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) &&
222 fld_range_set_mdt(out);
224 rc = fld_server_handle(lu_site2seq(site)->ss_server_fld,
225 req->rq_svc_thread->t_env,
228 rc = err_serious(-EPROTO);
234 static void fld_thread_info_init(struct ptlrpc_request *req,
235 struct fld_thread_info *info)
237 info->fti_pill = &req->rq_pill;
238 /* Init request capsule. */
239 req_capsule_init(info->fti_pill, req, RCL_SERVER);
240 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
243 static void fld_thread_info_fini(struct fld_thread_info *info)
245 req_capsule_fini(info->fti_pill);
248 static int fld_handle(struct ptlrpc_request *req)
250 struct fld_thread_info *info;
251 const struct lu_env *env;
254 env = req->rq_svc_thread->t_env;
255 LASSERT(env != NULL);
257 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
258 LASSERT(info != NULL);
260 fld_thread_info_init(req, info);
261 rc = fld_req_handle(req, info);
262 fld_thread_info_fini(info);
268 * Entry point for handling FLD RPCs called from MDT.
270 int fld_query(struct com_thread_info *info)
272 return fld_handle(info->cti_pill->rc_req);
274 EXPORT_SYMBOL(fld_query);
277 * Returns true, if fid is local to this server node.
279 * WARNING: this function is *not* guaranteed to return false if fid is
280 * remote: it makes an educated conservative guess only.
282 * fid_is_local() is supposed to be used in assertion checks only.
284 int fid_is_local(const struct lu_env *env,
285 struct lu_site *site, const struct lu_fid *fid)
288 struct seq_server_site *ss_site;
289 struct lu_seq_range *range;
290 struct fld_thread_info *info;
293 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
294 range = &info->fti_lrange;
296 result = 1; /* conservatively assume fid is local */
297 ss_site = lu_site2seq(site);
298 if (ss_site->ss_client_fld != NULL) {
301 rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache,
302 fid_seq(fid), range);
304 result = (range->lsr_index == ss_site->ss_node_id);
308 EXPORT_SYMBOL(fid_is_local);
310 static void fld_server_proc_fini(struct lu_server_fld *fld);
313 static int fld_server_proc_init(struct lu_server_fld *fld)
318 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
320 fld_server_proc_list, fld);
321 if (IS_ERR(fld->lsf_proc_dir)) {
322 rc = PTR_ERR(fld->lsf_proc_dir);
326 rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444,
327 &fld_proc_seq_fops, fld);
329 lprocfs_remove(&fld->lsf_proc_dir);
330 fld->lsf_proc_dir = NULL;
336 static void fld_server_proc_fini(struct lu_server_fld *fld)
339 if (fld->lsf_proc_dir != NULL) {
340 if (!IS_ERR(fld->lsf_proc_dir))
341 lprocfs_remove(&fld->lsf_proc_dir);
342 fld->lsf_proc_dir = NULL;
347 static int fld_server_proc_init(struct lu_server_fld *fld)
352 static void fld_server_proc_fini(struct lu_server_fld *fld)
358 int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld,
359 struct dt_device *dt, const char *prefix, int mds_node_id,
362 int cache_size, cache_threshold;
366 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
369 cache_size = FLD_SERVER_CACHE_SIZE /
370 sizeof(struct fld_cache_entry);
372 cache_threshold = cache_size *
373 FLD_SERVER_CACHE_THRESHOLD / 100;
375 mutex_init(&fld->lsf_lock);
376 fld->lsf_cache = fld_cache_init(fld->lsf_name,
377 cache_size, cache_threshold);
378 if (IS_ERR(fld->lsf_cache)) {
379 rc = PTR_ERR(fld->lsf_cache);
380 fld->lsf_cache = NULL;
384 if (!mds_node_id && type == LU_SEQ_RANGE_MDT) {
385 rc = fld_index_init(env, fld, dt);
392 rc = fld_server_proc_init(fld);
396 fld->lsf_control_exp = NULL;
402 fld_server_fini(env, fld);
405 EXPORT_SYMBOL(fld_server_init);
407 void fld_server_fini(const struct lu_env *env, struct lu_server_fld *fld)
411 fld_server_proc_fini(fld);
412 fld_index_fini(env, fld);
414 if (fld->lsf_cache != NULL) {
415 if (!IS_ERR(fld->lsf_cache))
416 fld_cache_fini(fld->lsf_cache);
417 fld->lsf_cache = NULL;
422 EXPORT_SYMBOL(fld_server_fini);