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"
70 #include <lustre_fid.h>
74 /* context key constructor/destructor: fld_key_init, fld_key_fini */
75 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
77 /* context key: fld_thread_key */
78 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD);
80 cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
82 static struct lu_local_obj_desc llod_fld_index = {
83 .llod_name = fld_index_name,
84 .llod_oid = FLD_INDEX_OID,
86 .llod_feat = &fld_index_features,
89 static int __init fld_mod_init(void)
91 fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
94 if (IS_ERR(fld_type_proc_dir))
95 return PTR_ERR(fld_type_proc_dir);
97 llo_local_obj_register(&llod_fld_index);
99 LU_CONTEXT_KEY_INIT(&fld_thread_key);
100 lu_context_key_register(&fld_thread_key);
104 static void __exit fld_mod_exit(void)
106 llo_local_obj_unregister(&llod_fld_index);
107 lu_context_key_degister(&fld_thread_key);
108 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
109 lprocfs_remove(&fld_type_proc_dir);
110 fld_type_proc_dir = NULL;
115 * Insert FLD index entry and update FLD cache.
117 * First it try to merge given range with existing range then update
118 * FLD index and FLD cache accordingly. FLD index consistency is maintained
120 * This function is called from the sequence allocator when a super-sequence
121 * is granted to a server.
124 int fld_server_create(struct lu_server_fld *fld,
125 const struct lu_env *env,
126 struct lu_seq_range *add_range,
129 struct lu_seq_range *erange;
130 struct lu_seq_range *new;
131 struct fld_thread_info *info;
137 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
138 mutex_lock(&fld->lsf_lock);
140 erange = &info->fti_lrange;
141 new = &info->fti_irange;
144 /* STEP 1: try to merge with previous range */
145 rc = fld_index_lookup(fld, env, new->lsr_start, erange);
147 /* in case of range overlap, mdt ID must be same for both ranges */
148 if (new->lsr_mdt != erange->lsr_mdt) {
149 CERROR("mdt[%x] for given range is different from"
150 "existing overlapping range mdt[%x]\n",
151 new->lsr_mdt, erange->lsr_mdt);
156 if (new->lsr_end < erange->lsr_end)
160 } else if (rc == -ENOENT) {
161 /* check for merge case: optimizes for single mds lustre.
162 * As entry does not exist, returned entry must be left side
163 * entry compared to start of new range (ref dio_lookup()).
164 * So try to merge from left.
166 if (new->lsr_start == erange->lsr_end &&
167 new->lsr_mdt == erange->lsr_mdt)
170 /* no overlap allowed in fld, so failure in lookup is error */
175 /* new range can be combined with existing one.
176 * So delete existing range.
179 rc = fld_index_delete(fld, env, erange, th);
181 new->lsr_start = min(erange->lsr_start, new->lsr_start);
182 new->lsr_end = max(erange->lsr_end, new->lsr_end);
189 /* STEP 2: try to merge with next range */
190 rc = fld_index_lookup(fld, env, new->lsr_end, erange);
192 /* case range overlap: with right side entry. */
193 if (new->lsr_mdt == erange->lsr_mdt)
195 } else if (rc == -ENOENT) {
196 /* this range is left of new range end point */
197 LASSERT(erange->lsr_end <= new->lsr_end);
199 if (new->lsr_end == erange->lsr_end)
201 if (new->lsr_start <= erange->lsr_start)
207 if (new->lsr_mdt != erange->lsr_mdt) {
208 CERROR("mdt[%x] for given range is different from"
209 "existing overlapping range mdt[%x]\n",
210 new->lsr_mdt, erange->lsr_mdt);
215 /* merge with next range */
216 rc = fld_index_delete(fld, env, erange, th);
218 new->lsr_start = min(erange->lsr_start, new->lsr_start);
219 new->lsr_end = max(erange->lsr_end, new->lsr_end);
224 /* now update fld entry. */
225 rc = fld_index_create(fld, env, new, th);
227 LASSERT(rc != -EEXIST);
230 fld_cache_insert(fld->lsf_cache, new);
232 mutex_unlock(&fld->lsf_lock);
234 CDEBUG((rc != 0 ? D_ERROR : D_INFO),
235 "%s: FLD create: given range : "DRANGE
236 "after merge "DRANGE" rc = %d \n", fld->lsf_name,
237 PRANGE(add_range), PRANGE(new), rc);
242 EXPORT_SYMBOL(fld_server_create);
245 * Lookup mds by seq, returns a range for given seq.
247 * If that entry is not cached in fld cache, request is sent to super
248 * sequence controller node (MDT0). All other MDT[1...N] and client
249 * cache fld entries, but this cache is not persistent.
252 int fld_server_lookup(struct lu_server_fld *fld,
253 const struct lu_env *env,
254 seqno_t seq, struct lu_seq_range *range)
259 /* Lookup it in the cache. */
260 rc = fld_cache_lookup(fld->lsf_cache, seq, range);
265 rc = fld_index_lookup(fld, env, seq, range);
267 LASSERT(fld->lsf_control_exp);
268 /* send request to mdt0 i.e. super seq. controller.
269 * This is temporary solution, long term solution is fld
270 * replication on all mdt servers.
272 rc = fld_client_rpc(fld->lsf_control_exp,
277 fld_cache_insert(fld->lsf_cache, range);
281 EXPORT_SYMBOL(fld_server_lookup);
284 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
285 * if entry is not found in cache we need to forward lookup request to MDT0
288 static int fld_server_handle(struct lu_server_fld *fld,
289 const struct lu_env *env,
290 __u32 opc, struct lu_seq_range *range,
291 struct fld_thread_info *info)
298 rc = fld_server_lookup(fld, env,
299 range->lsr_start, range);
306 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
307 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
313 static int fld_req_handle(struct ptlrpc_request *req,
314 struct fld_thread_info *info)
316 struct lu_site *site;
317 struct lu_seq_range *in;
318 struct lu_seq_range *out;
323 site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
325 rc = req_capsule_server_pack(info->fti_pill);
327 RETURN(err_serious(rc));
329 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
331 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
333 RETURN(err_serious(-EPROTO));
334 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
336 RETURN(err_serious(-EPROTO));
339 rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
340 req->rq_svc_thread->t_env,
343 rc = err_serious(-EPROTO);
348 static void fld_thread_info_init(struct ptlrpc_request *req,
349 struct fld_thread_info *info)
351 info->fti_pill = &req->rq_pill;
352 /* Init request capsule. */
353 req_capsule_init(info->fti_pill, req, RCL_SERVER);
354 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
357 static void fld_thread_info_fini(struct fld_thread_info *info)
359 req_capsule_fini(info->fti_pill);
362 static int fld_handle(struct ptlrpc_request *req)
364 struct fld_thread_info *info;
365 const struct lu_env *env;
368 env = req->rq_svc_thread->t_env;
369 LASSERT(env != NULL);
371 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
372 LASSERT(info != NULL);
374 fld_thread_info_init(req, info);
375 rc = fld_req_handle(req, info);
376 fld_thread_info_fini(info);
382 * Entry point for handling FLD RPCs called from MDT.
384 int fld_query(struct com_thread_info *info)
386 return fld_handle(info->cti_pill->rc_req);
388 EXPORT_SYMBOL(fld_query);
391 * Returns true, if fid is local to this server node.
393 * WARNING: this function is *not* guaranteed to return false if fid is
394 * remote: it makes an educated conservative guess only.
396 * fid_is_local() is supposed to be used in assertion checks only.
398 int fid_is_local(const struct lu_env *env,
399 struct lu_site *site, const struct lu_fid *fid)
402 struct md_site *msite;
403 struct lu_seq_range *range;
404 struct fld_thread_info *info;
407 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
408 range = &info->fti_lrange;
410 result = 1; /* conservatively assume fid is local */
411 msite = lu_site2md(site);
412 if (msite->ms_client_fld != NULL) {
415 rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
416 fid_seq(fid), range);
418 result = (range->lsr_mdt == msite->ms_node_id);
422 EXPORT_SYMBOL(fid_is_local);
424 static void fld_server_proc_fini(struct lu_server_fld *fld);
427 static int fld_server_proc_init(struct lu_server_fld *fld)
432 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
434 fld_server_proc_list, fld);
435 if (IS_ERR(fld->lsf_proc_dir)) {
436 rc = PTR_ERR(fld->lsf_proc_dir);
443 static void fld_server_proc_fini(struct lu_server_fld *fld)
446 if (fld->lsf_proc_dir != NULL) {
447 if (!IS_ERR(fld->lsf_proc_dir))
448 lprocfs_remove(&fld->lsf_proc_dir);
449 fld->lsf_proc_dir = NULL;
454 static int fld_server_proc_init(struct lu_server_fld *fld)
459 static void fld_server_proc_fini(struct lu_server_fld *fld)
465 int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
466 const char *prefix, const struct lu_env *env,
469 int cache_size, cache_threshold;
470 struct lu_seq_range range;
474 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
477 cache_size = FLD_SERVER_CACHE_SIZE /
478 sizeof(struct fld_cache_entry);
480 cache_threshold = cache_size *
481 FLD_SERVER_CACHE_THRESHOLD / 100;
483 mutex_init(&fld->lsf_lock);
484 fld->lsf_cache = fld_cache_init(fld->lsf_name,
485 cache_size, cache_threshold);
486 if (IS_ERR(fld->lsf_cache)) {
487 rc = PTR_ERR(fld->lsf_cache);
488 fld->lsf_cache = NULL;
493 rc = fld_index_init(fld, env, dt);
499 rc = fld_server_proc_init(fld);
503 fld->lsf_control_exp = NULL;
505 /* Insert reserved sequence number of ".lustre" into fld cache. */
506 range.lsr_start = LU_DOT_LUSTRE_SEQ;
507 range.lsr_end = LU_DOT_LUSTRE_SEQ + 1;
509 fld_cache_insert(fld->lsf_cache, &range);
514 fld_server_fini(fld, env);
517 EXPORT_SYMBOL(fld_server_init);
519 void fld_server_fini(struct lu_server_fld *fld,
520 const struct lu_env *env)
524 fld_server_proc_fini(fld);
525 fld_index_fini(fld, env);
527 if (fld->lsf_cache != NULL) {
528 if (!IS_ERR(fld->lsf_cache))
529 fld_cache_fini(fld->lsf_cache);
530 fld->lsf_cache = NULL;
535 EXPORT_SYMBOL(fld_server_fini);
537 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
538 MODULE_DESCRIPTION("Lustre FLD");
539 MODULE_LICENSE("GPL");
541 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);