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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
31 * Copyright (c) 2011 Whamcloud, Inc.
34 * This file is part of Lustre, http://www.lustre.org/
35 * Lustre is a trademark of Sun Microsystems, Inc.
37 * lustre/fld/fld_handler.c
39 * FLD (Fids Location Database)
41 * Author: Yury Umanets <umka@clusterfs.com>
42 * Author: WangDi <wangdi@clusterfs.com>
43 * Author: Pravin Shelar <pravin.shelar@sun.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_FLD
52 # include <libcfs/libcfs.h>
53 # include <linux/module.h>
54 # include <linux/jbd.h>
55 # include <asm/div64.h>
56 #else /* __KERNEL__ */
57 # include <liblustre.h>
58 # include <libcfs/list.h>
62 #include <obd_class.h>
63 #include <lustre_ver.h>
64 #include <obd_support.h>
65 #include <lprocfs_status.h>
67 #include <md_object.h>
68 #include <lustre_fid.h>
69 #include <lustre_req_layout.h>
70 #include "fld_internal.h"
71 #include <lustre_fid.h>
75 /* context key constructor/destructor: fld_key_init, fld_key_fini */
76 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
78 /* context key: fld_thread_key */
79 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD);
81 cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
83 static struct lu_local_obj_desc llod_fld_index = {
84 .llod_name = fld_index_name,
85 .llod_oid = FLD_INDEX_OID,
87 .llod_feat = &fld_index_features,
90 static int __init fld_mod_init(void)
92 fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
95 if (IS_ERR(fld_type_proc_dir))
96 return PTR_ERR(fld_type_proc_dir);
98 llo_local_obj_register(&llod_fld_index);
100 LU_CONTEXT_KEY_INIT(&fld_thread_key);
101 lu_context_key_register(&fld_thread_key);
105 static void __exit fld_mod_exit(void)
107 llo_local_obj_unregister(&llod_fld_index);
108 lu_context_key_degister(&fld_thread_key);
109 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
110 lprocfs_remove(&fld_type_proc_dir);
111 fld_type_proc_dir = NULL;
115 int fld_declare_server_create(struct lu_server_fld *fld,
116 const struct lu_env *env,
119 struct dt_object *dt_obj = fld->lsf_obj;
124 /* for ldiskfs OSD it's enough to declare operation with any ops
125 * with DMU we'll probably need to specify exact key/value */
126 rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
129 rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
132 rc = dt_obj->do_index_ops->dio_declare_insert(env, dt_obj,
137 EXPORT_SYMBOL(fld_declare_server_create);
140 * Insert FLD index entry and update FLD cache.
142 * First it try to merge given range with existing range then update
143 * FLD index and FLD cache accordingly. FLD index consistency is maintained
145 * This function is called from the sequence allocator when a super-sequence
146 * is granted to a server.
149 int fld_server_create(struct lu_server_fld *fld,
150 const struct lu_env *env,
151 struct lu_seq_range *add_range,
154 struct lu_seq_range *erange;
155 struct lu_seq_range *new;
156 struct fld_thread_info *info;
162 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
163 cfs_mutex_lock(&fld->lsf_lock);
165 erange = &info->fti_lrange;
166 new = &info->fti_irange;
169 /* STEP 1: try to merge with previous range */
170 rc = fld_index_lookup(fld, env, new->lsr_start, erange);
172 /* in case of range overlap, the location must be same */
173 if (range_compare_loc(new, erange) != 0) {
174 CERROR("the start of given range "DRANGE" conflict to"
175 "an existing range "DRANGE"\n",
176 PRANGE(new), PRANGE(erange));
177 GOTO(out, rc = -EIO);
180 if (new->lsr_end < erange->lsr_end)
183 } else if (rc == -ENOENT) {
184 /* check for merge case: optimizes for single mds lustre.
185 * As entry does not exist, returned entry must be left side
186 * entry compared to start of new range (ref dio_lookup()).
187 * So try to merge from left.
189 if (new->lsr_start == erange->lsr_end &&
190 range_compare_loc(new, erange) == 0)
193 /* no overlap allowed in fld, so failure in lookup is error */
198 /* new range will be merged with the existing one.
199 * delete this range at first. */
200 rc = fld_index_delete(fld, env, erange, th);
204 new->lsr_start = min(erange->lsr_start, new->lsr_start);
205 new->lsr_end = max(erange->lsr_end, new->lsr_end);
209 /* STEP 2: try to merge with next range */
210 rc = fld_index_lookup(fld, env, new->lsr_end, erange);
212 /* found a matched range, meaning we're either
213 * overlapping or ajacent, must merge with it. */
215 } else if (rc == -ENOENT) {
216 /* this range is left of new range end point */
217 LASSERT(erange->lsr_end <= new->lsr_end);
219 * the found left range must be either:
220 * 1. withing new range.
221 * 2. left of new range (no overlapping).
222 * because if they're partly overlapping, the STEP 1 must have
223 * been removed this range.
225 LASSERTF(erange->lsr_start > new->lsr_start ||
226 erange->lsr_end < new->lsr_start ||
227 (erange->lsr_end == new->lsr_start &&
228 range_compare_loc(new, erange) != 0),
229 "left "DRANGE", new "DRANGE"\n",
230 PRANGE(erange), PRANGE(new));
232 /* if it's within the new range, merge it */
233 if (erange->lsr_start > new->lsr_start)
240 if (range_compare_loc(new, erange) != 0) {
241 CERROR("the end of given range "DRANGE" overlaps "
242 "with an existing range "DRANGE"\n",
243 PRANGE(new), PRANGE(erange));
244 GOTO(out, rc = -EIO);
247 /* merge with next range */
248 rc = fld_index_delete(fld, env, erange, th);
252 new->lsr_start = min(erange->lsr_start, new->lsr_start);
253 new->lsr_end = max(erange->lsr_end, new->lsr_end);
256 /* now update fld entry. */
257 rc = fld_index_create(fld, env, new, th);
259 LASSERT(rc != -EEXIST);
262 fld_cache_insert(fld->lsf_cache, new);
264 cfs_mutex_unlock(&fld->lsf_lock);
266 CDEBUG((rc != 0 ? D_ERROR : D_INFO),
267 "%s: FLD create: given range : "DRANGE
268 "after merge "DRANGE" rc = %d \n", fld->lsf_name,
269 PRANGE(add_range), PRANGE(new), rc);
273 EXPORT_SYMBOL(fld_server_create);
276 * Lookup mds by seq, returns a range for given seq.
278 * If that entry is not cached in fld cache, request is sent to super
279 * sequence controller node (MDT0). All other MDT[1...N] and client
280 * cache fld entries, but this cache is not persistent.
283 int fld_server_lookup(struct lu_server_fld *fld,
284 const struct lu_env *env,
285 seqno_t seq, struct lu_seq_range *range)
287 struct lu_seq_range *erange;
288 struct fld_thread_info *info;
292 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
293 erange = &info->fti_lrange;
295 /* Lookup it in the cache. */
296 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
298 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
299 CERROR("FLD cache found a range "DRANGE" doesn't "
300 "match the requested flag %x\n",
301 PRANGE(erange), range->lsr_flags);
309 rc = fld_index_lookup(fld, env, seq, erange);
311 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
312 CERROR("FLD found a range "DRANGE" doesn't "
313 "match the requested flag %x\n",
314 PRANGE(erange), range->lsr_flags);
320 LASSERT(fld->lsf_control_exp);
321 /* send request to mdt0 i.e. super seq. controller.
322 * This is temporary solution, long term solution is fld
323 * replication on all mdt servers.
325 rc = fld_client_rpc(fld->lsf_control_exp,
330 fld_cache_insert(fld->lsf_cache, range);
334 EXPORT_SYMBOL(fld_server_lookup);
337 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
338 * if entry is not found in cache we need to forward lookup request to MDT0
341 static int fld_server_handle(struct lu_server_fld *fld,
342 const struct lu_env *env,
343 __u32 opc, struct lu_seq_range *range,
344 struct fld_thread_info *info)
351 rc = fld_server_lookup(fld, env,
352 range->lsr_start, range);
359 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
360 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
366 static int fld_req_handle(struct ptlrpc_request *req,
367 struct fld_thread_info *info)
369 struct obd_export *exp = req->rq_export;
370 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
371 struct lu_seq_range *in;
372 struct lu_seq_range *out;
377 rc = req_capsule_server_pack(info->fti_pill);
379 RETURN(err_serious(rc));
381 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
383 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
385 RETURN(err_serious(-EPROTO));
386 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
388 RETURN(err_serious(-EPROTO));
391 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
392 * Set it as 'LU_SEQ_RANGE_MDT' by default.
393 * Old 2.0 liblustre client cannot talk with new 2.1 server. */
394 if (!(exp->exp_connect_flags & OBD_CONNECT_64BITHASH) &&
396 out->lsr_flags = LU_SEQ_RANGE_MDT;
398 rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
399 req->rq_svc_thread->t_env,
402 rc = err_serious(-EPROTO);
407 static void fld_thread_info_init(struct ptlrpc_request *req,
408 struct fld_thread_info *info)
410 info->fti_pill = &req->rq_pill;
411 /* Init request capsule. */
412 req_capsule_init(info->fti_pill, req, RCL_SERVER);
413 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
416 static void fld_thread_info_fini(struct fld_thread_info *info)
418 req_capsule_fini(info->fti_pill);
421 static int fld_handle(struct ptlrpc_request *req)
423 struct fld_thread_info *info;
424 const struct lu_env *env;
427 env = req->rq_svc_thread->t_env;
428 LASSERT(env != NULL);
430 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
431 LASSERT(info != NULL);
433 fld_thread_info_init(req, info);
434 rc = fld_req_handle(req, info);
435 fld_thread_info_fini(info);
441 * Entry point for handling FLD RPCs called from MDT.
443 int fld_query(struct com_thread_info *info)
445 return fld_handle(info->cti_pill->rc_req);
447 EXPORT_SYMBOL(fld_query);
450 * Returns true, if fid is local to this server node.
452 * WARNING: this function is *not* guaranteed to return false if fid is
453 * remote: it makes an educated conservative guess only.
455 * fid_is_local() is supposed to be used in assertion checks only.
457 int fid_is_local(const struct lu_env *env,
458 struct lu_site *site, const struct lu_fid *fid)
461 struct md_site *msite;
462 struct lu_seq_range *range;
463 struct fld_thread_info *info;
466 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
467 range = &info->fti_lrange;
469 result = 1; /* conservatively assume fid is local */
470 msite = lu_site2md(site);
471 if (msite->ms_client_fld != NULL) {
474 rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
475 fid_seq(fid), range);
477 result = (range->lsr_index == msite->ms_node_id);
481 EXPORT_SYMBOL(fid_is_local);
483 static void fld_server_proc_fini(struct lu_server_fld *fld);
486 static int fld_server_proc_init(struct lu_server_fld *fld)
491 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
493 fld_server_proc_list, fld);
494 if (IS_ERR(fld->lsf_proc_dir)) {
495 rc = PTR_ERR(fld->lsf_proc_dir);
502 static void fld_server_proc_fini(struct lu_server_fld *fld)
505 if (fld->lsf_proc_dir != NULL) {
506 if (!IS_ERR(fld->lsf_proc_dir))
507 lprocfs_remove(&fld->lsf_proc_dir);
508 fld->lsf_proc_dir = NULL;
513 static int fld_server_proc_init(struct lu_server_fld *fld)
518 static void fld_server_proc_fini(struct lu_server_fld *fld)
524 int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
525 const char *prefix, const struct lu_env *env,
528 int cache_size, cache_threshold;
529 struct lu_seq_range range;
533 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
536 cache_size = FLD_SERVER_CACHE_SIZE /
537 sizeof(struct fld_cache_entry);
539 cache_threshold = cache_size *
540 FLD_SERVER_CACHE_THRESHOLD / 100;
542 cfs_mutex_init(&fld->lsf_lock);
543 fld->lsf_cache = fld_cache_init(fld->lsf_name,
544 cache_size, cache_threshold);
545 if (IS_ERR(fld->lsf_cache)) {
546 rc = PTR_ERR(fld->lsf_cache);
547 fld->lsf_cache = NULL;
552 rc = fld_index_init(fld, env, dt);
558 rc = fld_server_proc_init(fld);
562 fld->lsf_control_exp = NULL;
564 /* Insert reserved sequence number of ".lustre" into fld cache. */
565 range.lsr_start = FID_SEQ_DOT_LUSTRE;
566 range.lsr_end = FID_SEQ_DOT_LUSTRE + 1;
568 range.lsr_flags = LU_SEQ_RANGE_MDT;
569 fld_cache_insert(fld->lsf_cache, &range);
574 fld_server_fini(fld, env);
577 EXPORT_SYMBOL(fld_server_init);
579 void fld_server_fini(struct lu_server_fld *fld,
580 const struct lu_env *env)
584 fld_server_proc_fini(fld);
585 fld_index_fini(fld, env);
587 if (fld->lsf_cache != NULL) {
588 if (!IS_ERR(fld->lsf_cache))
589 fld_cache_fini(fld->lsf_cache);
590 fld->lsf_cache = NULL;
595 EXPORT_SYMBOL(fld_server_fini);
597 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
598 MODULE_DESCRIPTION("Lustre FLD");
599 MODULE_LICENSE("GPL");
601 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);