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.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/fld/fld_handler.c
40 * FLD (Fids Location Database)
42 * Author: Yury Umanets <umka@clusterfs.com>
43 * Author: WangDi <wangdi@clusterfs.com>
44 * Author: Pravin Shelar <pravin.shelar@sun.com>
48 # define EXPORT_SYMTAB
50 #define DEBUG_SUBSYSTEM S_FLD
53 # include <libcfs/libcfs.h>
54 # include <linux/module.h>
55 # include <linux/jbd.h>
56 # include <asm/div64.h>
57 #else /* __KERNEL__ */
58 # include <liblustre.h>
59 # include <libcfs/list.h>
63 #include <obd_class.h>
64 #include <lustre_ver.h>
65 #include <obd_support.h>
66 #include <lprocfs_status.h>
68 #include <md_object.h>
69 #include <lustre_fid.h>
70 #include <lustre_req_layout.h>
71 #include "fld_internal.h"
72 #include <lustre_fid.h>
76 /* context key constructor/destructor: fld_key_init, fld_key_fini */
77 LU_KEY_INIT_FINI(fld, struct fld_thread_info);
79 /* context key: fld_thread_key */
80 LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD);
82 cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
84 static struct lu_local_obj_desc llod_fld_index = {
85 .llod_name = fld_index_name,
86 .llod_oid = FLD_INDEX_OID,
88 .llod_feat = &fld_index_features,
91 static int __init fld_mod_init(void)
93 fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
96 if (IS_ERR(fld_type_proc_dir))
97 return PTR_ERR(fld_type_proc_dir);
99 llo_local_obj_register(&llod_fld_index);
101 LU_CONTEXT_KEY_INIT(&fld_thread_key);
102 lu_context_key_register(&fld_thread_key);
106 static void __exit fld_mod_exit(void)
108 llo_local_obj_unregister(&llod_fld_index);
109 lu_context_key_degister(&fld_thread_key);
110 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
111 lprocfs_remove(&fld_type_proc_dir);
112 fld_type_proc_dir = NULL;
116 int fld_declare_server_create(struct lu_server_fld *fld,
117 const struct lu_env *env,
120 struct dt_object *dt_obj = fld->lsf_obj;
125 /* for ldiskfs OSD it's enough to declare operation with any ops
126 * with DMU we'll probably need to specify exact key/value */
127 rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
130 rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
133 rc = dt_obj->do_index_ops->dio_declare_insert(env, dt_obj,
138 EXPORT_SYMBOL(fld_declare_server_create);
141 * Insert FLD index entry and update FLD cache.
143 * First it try to merge given range with existing range then update
144 * FLD index and FLD cache accordingly. FLD index consistency is maintained
146 * This function is called from the sequence allocator when a super-sequence
147 * is granted to a server.
150 int fld_server_create(struct lu_server_fld *fld,
151 const struct lu_env *env,
152 struct lu_seq_range *add_range,
155 struct lu_seq_range *erange;
156 struct lu_seq_range *new;
157 struct fld_thread_info *info;
163 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
164 cfs_mutex_lock(&fld->lsf_lock);
166 erange = &info->fti_lrange;
167 new = &info->fti_irange;
170 /* STEP 1: try to merge with previous range */
171 rc = fld_index_lookup(fld, env, new->lsr_start, erange);
173 /* in case of range overlap, the location must be same */
174 if (range_compare_loc(new, erange) != 0) {
175 CERROR("the start of given range "DRANGE" conflict to"
176 "an existing range "DRANGE"\n",
177 PRANGE(new), PRANGE(erange));
178 GOTO(out, rc = -EIO);
181 if (new->lsr_end < erange->lsr_end)
184 } else if (rc == -ENOENT) {
185 /* check for merge case: optimizes for single mds lustre.
186 * As entry does not exist, returned entry must be left side
187 * entry compared to start of new range (ref dio_lookup()).
188 * So try to merge from left.
190 if (new->lsr_start == erange->lsr_end &&
191 range_compare_loc(new, erange) == 0)
194 /* no overlap allowed in fld, so failure in lookup is error */
199 /* new range will be merged with the existing one.
200 * delete this range at first. */
201 rc = fld_index_delete(fld, env, erange, th);
205 new->lsr_start = min(erange->lsr_start, new->lsr_start);
206 new->lsr_end = max(erange->lsr_end, new->lsr_end);
210 /* STEP 2: try to merge with next range */
211 rc = fld_index_lookup(fld, env, new->lsr_end, erange);
213 /* found a matched range, meaning we're either
214 * overlapping or ajacent, must merge with it. */
216 } else if (rc == -ENOENT) {
217 /* this range is left of new range end point */
218 LASSERT(erange->lsr_end <= new->lsr_end);
220 * the found left range must be either:
221 * 1. withing new range.
222 * 2. left of new range (no overlapping).
223 * because if they're partly overlapping, the STEP 1 must have
224 * been removed this range.
226 LASSERTF(erange->lsr_start > new->lsr_start ||
227 erange->lsr_end < new->lsr_start ||
228 (erange->lsr_end == new->lsr_start &&
229 range_compare_loc(new, erange) != 0),
230 "left "DRANGE", new "DRANGE"\n",
231 PRANGE(erange), PRANGE(new));
233 /* if it's within the new range, merge it */
234 if (erange->lsr_start > new->lsr_start)
241 if (range_compare_loc(new, erange) != 0) {
242 CERROR("the end of given range "DRANGE" overlaps "
243 "with an existing range "DRANGE"\n",
244 PRANGE(new), PRANGE(erange));
245 GOTO(out, rc = -EIO);
248 /* merge with next range */
249 rc = fld_index_delete(fld, env, erange, th);
253 new->lsr_start = min(erange->lsr_start, new->lsr_start);
254 new->lsr_end = max(erange->lsr_end, new->lsr_end);
257 /* now update fld entry. */
258 rc = fld_index_create(fld, env, new, th);
260 LASSERT(rc != -EEXIST);
263 fld_cache_insert(fld->lsf_cache, new);
265 cfs_mutex_unlock(&fld->lsf_lock);
267 CDEBUG((rc != 0 ? D_ERROR : D_INFO),
268 "%s: FLD create: given range : "DRANGE
269 "after merge "DRANGE" rc = %d \n", fld->lsf_name,
270 PRANGE(add_range), PRANGE(new), rc);
274 EXPORT_SYMBOL(fld_server_create);
277 * Lookup mds by seq, returns a range for given seq.
279 * If that entry is not cached in fld cache, request is sent to super
280 * sequence controller node (MDT0). All other MDT[1...N] and client
281 * cache fld entries, but this cache is not persistent.
284 int fld_server_lookup(struct lu_server_fld *fld,
285 const struct lu_env *env,
286 seqno_t seq, struct lu_seq_range *range)
288 struct lu_seq_range *erange;
289 struct fld_thread_info *info;
293 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
294 erange = &info->fti_lrange;
296 /* Lookup it in the cache. */
297 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
299 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
300 CERROR("FLD cache found a range "DRANGE" doesn't "
301 "match the requested flag %x\n",
302 PRANGE(erange), range->lsr_flags);
310 rc = fld_index_lookup(fld, env, seq, erange);
312 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
313 CERROR("FLD found a range "DRANGE" doesn't "
314 "match the requested flag %x\n",
315 PRANGE(erange), range->lsr_flags);
321 LASSERT(fld->lsf_control_exp);
322 /* send request to mdt0 i.e. super seq. controller.
323 * This is temporary solution, long term solution is fld
324 * replication on all mdt servers.
326 rc = fld_client_rpc(fld->lsf_control_exp,
331 fld_cache_insert(fld->lsf_cache, range);
335 EXPORT_SYMBOL(fld_server_lookup);
338 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
339 * if entry is not found in cache we need to forward lookup request to MDT0
342 static int fld_server_handle(struct lu_server_fld *fld,
343 const struct lu_env *env,
344 __u32 opc, struct lu_seq_range *range,
345 struct fld_thread_info *info)
352 rc = fld_server_lookup(fld, env,
353 range->lsr_start, range);
360 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
361 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
367 static int fld_req_handle(struct ptlrpc_request *req,
368 struct fld_thread_info *info)
370 struct obd_export *exp = req->rq_export;
371 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
372 struct lu_seq_range *in;
373 struct lu_seq_range *out;
378 rc = req_capsule_server_pack(info->fti_pill);
380 RETURN(err_serious(rc));
382 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
384 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
386 RETURN(err_serious(-EPROTO));
387 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
389 RETURN(err_serious(-EPROTO));
392 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
393 * Set it as 'LU_SEQ_RANGE_MDT' by default.
394 * Old 2.0 liblustre client cannot talk with new 2.1 server. */
395 if (!(exp->exp_connect_flags & OBD_CONNECT_64BITHASH) &&
397 out->lsr_flags = LU_SEQ_RANGE_MDT;
399 rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
400 req->rq_svc_thread->t_env,
403 rc = err_serious(-EPROTO);
408 static void fld_thread_info_init(struct ptlrpc_request *req,
409 struct fld_thread_info *info)
411 info->fti_pill = &req->rq_pill;
412 /* Init request capsule. */
413 req_capsule_init(info->fti_pill, req, RCL_SERVER);
414 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
417 static void fld_thread_info_fini(struct fld_thread_info *info)
419 req_capsule_fini(info->fti_pill);
422 static int fld_handle(struct ptlrpc_request *req)
424 struct fld_thread_info *info;
425 const struct lu_env *env;
428 env = req->rq_svc_thread->t_env;
429 LASSERT(env != NULL);
431 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
432 LASSERT(info != NULL);
434 fld_thread_info_init(req, info);
435 rc = fld_req_handle(req, info);
436 fld_thread_info_fini(info);
442 * Entry point for handling FLD RPCs called from MDT.
444 int fld_query(struct com_thread_info *info)
446 return fld_handle(info->cti_pill->rc_req);
448 EXPORT_SYMBOL(fld_query);
451 * Returns true, if fid is local to this server node.
453 * WARNING: this function is *not* guaranteed to return false if fid is
454 * remote: it makes an educated conservative guess only.
456 * fid_is_local() is supposed to be used in assertion checks only.
458 int fid_is_local(const struct lu_env *env,
459 struct lu_site *site, const struct lu_fid *fid)
462 struct md_site *msite;
463 struct lu_seq_range *range;
464 struct fld_thread_info *info;
467 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
468 range = &info->fti_lrange;
470 result = 1; /* conservatively assume fid is local */
471 msite = lu_site2md(site);
472 if (msite->ms_client_fld != NULL) {
475 rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
476 fid_seq(fid), range);
478 result = (range->lsr_index == msite->ms_node_id);
482 EXPORT_SYMBOL(fid_is_local);
484 static void fld_server_proc_fini(struct lu_server_fld *fld);
487 static int fld_server_proc_init(struct lu_server_fld *fld)
492 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
494 fld_server_proc_list, fld);
495 if (IS_ERR(fld->lsf_proc_dir)) {
496 rc = PTR_ERR(fld->lsf_proc_dir);
503 static void fld_server_proc_fini(struct lu_server_fld *fld)
506 if (fld->lsf_proc_dir != NULL) {
507 if (!IS_ERR(fld->lsf_proc_dir))
508 lprocfs_remove(&fld->lsf_proc_dir);
509 fld->lsf_proc_dir = NULL;
514 static int fld_server_proc_init(struct lu_server_fld *fld)
519 static void fld_server_proc_fini(struct lu_server_fld *fld)
525 int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
526 const char *prefix, const struct lu_env *env,
529 int cache_size, cache_threshold;
530 struct lu_seq_range range;
534 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
537 cache_size = FLD_SERVER_CACHE_SIZE /
538 sizeof(struct fld_cache_entry);
540 cache_threshold = cache_size *
541 FLD_SERVER_CACHE_THRESHOLD / 100;
543 cfs_mutex_init(&fld->lsf_lock);
544 fld->lsf_cache = fld_cache_init(fld->lsf_name,
545 cache_size, cache_threshold);
546 if (IS_ERR(fld->lsf_cache)) {
547 rc = PTR_ERR(fld->lsf_cache);
548 fld->lsf_cache = NULL;
553 rc = fld_index_init(fld, env, dt);
559 rc = fld_server_proc_init(fld);
563 fld->lsf_control_exp = NULL;
565 /* Insert reserved sequence number of ".lustre" into fld cache. */
566 range.lsr_start = FID_SEQ_DOT_LUSTRE;
567 range.lsr_end = FID_SEQ_DOT_LUSTRE + 1;
569 range.lsr_flags = LU_SEQ_RANGE_MDT;
570 fld_cache_insert(fld->lsf_cache, &range);
575 fld_server_fini(fld, env);
578 EXPORT_SYMBOL(fld_server_init);
580 void fld_server_fini(struct lu_server_fld *fld,
581 const struct lu_env *env)
585 fld_server_proc_fini(fld);
586 fld_index_fini(fld, env);
588 if (fld->lsf_cache != NULL) {
589 if (!IS_ERR(fld->lsf_cache))
590 fld_cache_fini(fld->lsf_cache);
591 fld->lsf_cache = NULL;
596 EXPORT_SYMBOL(fld_server_fini);
598 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
599 MODULE_DESCRIPTION("Lustre FLD");
600 MODULE_LICENSE("GPL");
602 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);