1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_FLD
51 # include <libcfs/libcfs.h>
52 # include <linux/module.h>
53 # include <linux/jbd.h>
54 # include <asm/div64.h>
55 #else /* __KERNEL__ */
56 # include <liblustre.h>
57 # include <libcfs/list.h>
61 #include <obd_class.h>
62 #include <lustre_ver.h>
63 #include <obd_support.h>
64 #include <lprocfs_status.h>
66 #include <md_object.h>
67 #include <lustre_fid.h>
68 #include <lustre_req_layout.h>
69 #include "fld_internal.h"
73 /* context key constructor/destructor: fld_key_init, fld_key_fini */
74 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
76 /* context key: fld_thread_key */
77 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD);
79 cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
81 static struct lu_local_obj_desc llod_fld_index = {
82 .llod_name = fld_index_name,
83 .llod_oid = FLD_INDEX_OID,
85 .llod_feat = &fld_index_features,
88 static int __init fld_mod_init(void)
90 fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
93 if (IS_ERR(fld_type_proc_dir))
94 return PTR_ERR(fld_type_proc_dir);
96 llo_local_obj_register(&llod_fld_index);
98 LU_CONTEXT_KEY_INIT(&fld_thread_key);
99 lu_context_key_register(&fld_thread_key);
103 static void __exit fld_mod_exit(void)
105 llo_local_obj_unregister(&llod_fld_index);
106 lu_context_key_degister(&fld_thread_key);
107 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
108 lprocfs_remove(&fld_type_proc_dir);
109 fld_type_proc_dir = NULL;
114 * Insert FLD index entry and update FLD cache.
116 * First it try to merge given range with existing range then update
117 * FLD index and FLD cache accordingly. FLD index consistency is maintained
119 * This function is called from the sequence allocator when a super-sequence
120 * is granted to a server.
123 int fld_server_create(struct lu_server_fld *fld,
124 const struct lu_env *env,
125 struct lu_seq_range *add_range,
128 struct lu_seq_range *erange;
129 struct lu_seq_range *new;
130 struct fld_thread_info *info;
136 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
137 mutex_lock(&fld->lsf_lock);
139 erange = &info->fti_lrange;
140 new = &info->fti_irange;
143 /* STEP 1: try to merge with previous range */
144 rc = fld_index_lookup(fld, env, new->lsr_start, erange);
146 /* in case of range overlap, mdt ID must be same for both ranges */
147 if (new->lsr_mdt != erange->lsr_mdt) {
148 CERROR("mdt[%x] for given range is different from"
149 "existing overlapping range mdt[%x]\n",
150 new->lsr_mdt, erange->lsr_mdt);
155 if (new->lsr_end < erange->lsr_end)
159 } else if (rc == -ENOENT) {
160 /* check for merge case: optimizes for single mds lustre.
161 * As entry does not exist, returned entry must be left side
162 * entry compared to start of new range (ref dio_lookup()).
163 * So try to merge from left.
165 if (new->lsr_start == erange->lsr_end &&
166 new->lsr_mdt == erange->lsr_mdt)
169 /* no overlap allowed in fld, so failure in lookup is error */
174 /* new range can be combined with existing one.
175 * So delete existing range.
178 rc = fld_index_delete(fld, env, erange, th);
180 new->lsr_start = min(erange->lsr_start, new->lsr_start);
181 new->lsr_end = max(erange->lsr_end, new->lsr_end);
188 /* STEP 2: try to merge with next range */
189 rc = fld_index_lookup(fld, env, new->lsr_end, erange);
191 /* case range overlap: with right side entry. */
192 if (new->lsr_mdt == erange->lsr_mdt)
194 } else if (rc == -ENOENT) {
195 /* this range is left of new range end point */
196 LASSERT(erange->lsr_end <= new->lsr_end);
198 if (new->lsr_end == erange->lsr_end)
200 if (new->lsr_start <= erange->lsr_start)
206 if (new->lsr_mdt != erange->lsr_mdt) {
207 CERROR("mdt[%x] for given range is different from"
208 "existing overlapping range mdt[%x]\n",
209 new->lsr_mdt, erange->lsr_mdt);
214 /* merge with next range */
215 rc = fld_index_delete(fld, env, erange, th);
217 new->lsr_start = min(erange->lsr_start, new->lsr_start);
218 new->lsr_end = max(erange->lsr_end, new->lsr_end);
223 /* now update fld entry. */
224 rc = fld_index_create(fld, env, new, th);
226 LASSERT(rc != -EEXIST);
229 fld_cache_insert(fld->lsf_cache, new);
231 mutex_unlock(&fld->lsf_lock);
233 CDEBUG((rc != 0 ? D_ERROR : D_INFO),
234 "%s: FLD create: given range : "DRANGE
235 "after merge "DRANGE" rc = %d \n", fld->lsf_name,
236 PRANGE(add_range), PRANGE(new), rc);
241 EXPORT_SYMBOL(fld_server_create);
244 * Lookup mds by seq, returns a range for given seq.
246 * If that entry is not cached in fld cache, request is sent to super
247 * sequence controller node (MDT0). All other MDT[1...N] and client
248 * cache fld entries, but this cache is not persistent.
251 int fld_server_lookup(struct lu_server_fld *fld,
252 const struct lu_env *env,
253 seqno_t seq, struct lu_seq_range *range)
258 /* Lookup it in the cache. */
259 rc = fld_cache_lookup(fld->lsf_cache, seq, range);
264 rc = fld_index_lookup(fld, env, seq, range);
266 LASSERT(fld->lsf_control_exp);
267 /* send request to mdt0 i.e. super seq. controller.
268 * This is temporary solution, long term solution is fld
269 * replication on all mdt servers.
271 rc = fld_client_rpc(fld->lsf_control_exp,
276 fld_cache_insert(fld->lsf_cache, range);
280 EXPORT_SYMBOL(fld_server_lookup);
283 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
284 * if entry is not found in cache we need to forward lookup request to MDT0
287 static int fld_server_handle(struct lu_server_fld *fld,
288 const struct lu_env *env,
289 __u32 opc, struct lu_seq_range *range,
290 struct fld_thread_info *info)
297 rc = fld_server_lookup(fld, env,
298 range->lsr_start, range);
305 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
306 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
312 static int fld_req_handle(struct ptlrpc_request *req,
313 struct fld_thread_info *info)
315 struct lu_site *site;
316 struct lu_seq_range *in;
317 struct lu_seq_range *out;
322 site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
324 rc = req_capsule_server_pack(info->fti_pill);
326 RETURN(err_serious(rc));
328 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
330 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
332 RETURN(err_serious(-EPROTO));
333 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
335 RETURN(err_serious(-EPROTO));
338 rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
339 req->rq_svc_thread->t_env,
342 rc = err_serious(-EPROTO);
347 static void fld_thread_info_init(struct ptlrpc_request *req,
348 struct fld_thread_info *info)
350 info->fti_pill = &req->rq_pill;
351 /* Init request capsule. */
352 req_capsule_init(info->fti_pill, req, RCL_SERVER);
353 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
356 static void fld_thread_info_fini(struct fld_thread_info *info)
358 req_capsule_fini(info->fti_pill);
361 static int fld_handle(struct ptlrpc_request *req)
363 struct fld_thread_info *info;
364 const struct lu_env *env;
367 env = req->rq_svc_thread->t_env;
368 LASSERT(env != NULL);
370 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
371 LASSERT(info != NULL);
373 fld_thread_info_init(req, info);
374 rc = fld_req_handle(req, info);
375 fld_thread_info_fini(info);
381 * Entry point for handling FLD RPCs called from MDT.
383 int fld_query(struct com_thread_info *info)
385 return fld_handle(info->cti_pill->rc_req);
387 EXPORT_SYMBOL(fld_query);
390 * Returns true, if fid is local to this server node.
392 * WARNING: this function is *not* guaranteed to return false if fid is
393 * remote: it makes an educated conservative guess only.
395 * fid_is_local() is supposed to be used in assertion checks only.
397 int fid_is_local(const struct lu_env *env,
398 struct lu_site *site, const struct lu_fid *fid)
401 struct md_site *msite;
402 struct lu_seq_range *range;
403 struct fld_thread_info *info;
406 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
407 range = &info->fti_lrange;
409 result = 1; /* conservatively assume fid is local */
410 msite = lu_site2md(site);
411 if (msite->ms_client_fld != NULL) {
414 rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
415 fid_seq(fid), range);
417 result = (range->lsr_mdt == msite->ms_node_id);
421 EXPORT_SYMBOL(fid_is_local);
423 static void fld_server_proc_fini(struct lu_server_fld *fld);
426 static int fld_server_proc_init(struct lu_server_fld *fld)
431 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
433 fld_server_proc_list, fld);
434 if (IS_ERR(fld->lsf_proc_dir)) {
435 rc = PTR_ERR(fld->lsf_proc_dir);
442 static void fld_server_proc_fini(struct lu_server_fld *fld)
445 if (fld->lsf_proc_dir != NULL) {
446 if (!IS_ERR(fld->lsf_proc_dir))
447 lprocfs_remove(&fld->lsf_proc_dir);
448 fld->lsf_proc_dir = NULL;
453 static int fld_server_proc_init(struct lu_server_fld *fld)
458 static void fld_server_proc_fini(struct lu_server_fld *fld)
464 int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
465 const char *prefix, const struct lu_env *env,
468 int cache_size, cache_threshold;
472 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
475 cache_size = FLD_SERVER_CACHE_SIZE /
476 sizeof(struct fld_cache_entry);
478 cache_threshold = cache_size *
479 FLD_SERVER_CACHE_THRESHOLD / 100;
481 mutex_init(&fld->lsf_lock);
482 fld->lsf_cache = fld_cache_init(fld->lsf_name,
483 cache_size, cache_threshold);
484 if (IS_ERR(fld->lsf_cache)) {
485 rc = PTR_ERR(fld->lsf_cache);
486 fld->lsf_cache = NULL;
491 rc = fld_index_init(fld, env, dt);
497 rc = fld_server_proc_init(fld);
501 fld->lsf_control_exp = NULL;
505 fld_server_fini(fld, env);
508 EXPORT_SYMBOL(fld_server_init);
510 void fld_server_fini(struct lu_server_fld *fld,
511 const struct lu_env *env)
515 fld_server_proc_fini(fld);
516 fld_index_fini(fld, env);
518 if (fld->lsf_cache != NULL) {
519 if (!IS_ERR(fld->lsf_cache))
520 fld_cache_fini(fld->lsf_cache);
521 fld->lsf_cache = NULL;
526 EXPORT_SYMBOL(fld_server_fini);
528 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
529 MODULE_DESCRIPTION("Lustre FLD");
530 MODULE_LICENSE("GPL");
532 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);