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.
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 cfs_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, the location must be same */
148 if (range_compare_loc(new, erange) != 0) {
149 CERROR("the start of given range "DRANGE" conflict to"
150 "an existing range "DRANGE"\n",
151 PRANGE(new), PRANGE(erange));
152 GOTO(out, rc = -EIO);
155 if (new->lsr_end < erange->lsr_end)
158 } else if (rc == -ENOENT) {
159 /* check for merge case: optimizes for single mds lustre.
160 * As entry does not exist, returned entry must be left side
161 * entry compared to start of new range (ref dio_lookup()).
162 * So try to merge from left.
164 if (new->lsr_start == erange->lsr_end &&
165 range_compare_loc(new, erange) == 0)
168 /* no overlap allowed in fld, so failure in lookup is error */
173 /* new range will be merged with the existing one.
174 * delete this range at first. */
175 rc = fld_index_delete(fld, env, erange, th);
179 new->lsr_start = min(erange->lsr_start, new->lsr_start);
180 new->lsr_end = max(erange->lsr_end, new->lsr_end);
184 /* STEP 2: try to merge with next range */
185 rc = fld_index_lookup(fld, env, new->lsr_end, erange);
187 /* found a matched range, meaning we're either
188 * overlapping or ajacent, must merge with it. */
190 } else if (rc == -ENOENT) {
191 /* this range is left of new range end point */
192 LASSERT(erange->lsr_end <= new->lsr_end);
194 * the found left range must be either:
195 * 1. withing new range.
196 * 2. left of new range (no overlapping).
197 * because if they're partly overlapping, the STEP 1 must have
198 * been removed this range.
200 LASSERTF(erange->lsr_start > new->lsr_start ||
201 erange->lsr_end < new->lsr_start ||
202 (erange->lsr_end == new->lsr_start &&
203 range_compare_loc(new, erange) != 0),
204 "left "DRANGE", new "DRANGE"\n",
205 PRANGE(erange), PRANGE(new));
207 /* if it's within the new range, merge it */
208 if (erange->lsr_start > new->lsr_start)
215 if (range_compare_loc(new, erange) != 0) {
216 CERROR("the end of given range "DRANGE" overlaps "
217 "with an existing range "DRANGE"\n",
218 PRANGE(new), PRANGE(erange));
219 GOTO(out, rc = -EIO);
222 /* merge with next range */
223 rc = fld_index_delete(fld, env, erange, th);
227 new->lsr_start = min(erange->lsr_start, new->lsr_start);
228 new->lsr_end = max(erange->lsr_end, new->lsr_end);
231 /* now update fld entry. */
232 rc = fld_index_create(fld, env, new, th);
234 LASSERT(rc != -EEXIST);
237 fld_cache_insert(fld->lsf_cache, new);
239 cfs_mutex_unlock(&fld->lsf_lock);
241 CDEBUG((rc != 0 ? D_ERROR : D_INFO),
242 "%s: FLD create: given range : "DRANGE
243 "after merge "DRANGE" rc = %d \n", fld->lsf_name,
244 PRANGE(add_range), PRANGE(new), rc);
249 EXPORT_SYMBOL(fld_server_create);
252 * Lookup mds by seq, returns a range for given seq.
254 * If that entry is not cached in fld cache, request is sent to super
255 * sequence controller node (MDT0). All other MDT[1...N] and client
256 * cache fld entries, but this cache is not persistent.
259 int fld_server_lookup(struct lu_server_fld *fld,
260 const struct lu_env *env,
261 seqno_t seq, struct lu_seq_range *range)
263 struct lu_seq_range *erange;
264 struct fld_thread_info *info;
268 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
269 erange = &info->fti_lrange;
271 /* Lookup it in the cache. */
272 rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
274 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
275 CERROR("FLD cache found a range "DRANGE" doesn't "
276 "match the requested flag %x\n",
277 PRANGE(erange), range->lsr_flags);
285 rc = fld_index_lookup(fld, env, seq, erange);
287 if (unlikely(erange->lsr_flags != range->lsr_flags)) {
288 CERROR("FLD found a range "DRANGE" doesn't "
289 "match the requested flag %x\n",
290 PRANGE(erange), range->lsr_flags);
296 LASSERT(fld->lsf_control_exp);
297 /* send request to mdt0 i.e. super seq. controller.
298 * This is temporary solution, long term solution is fld
299 * replication on all mdt servers.
301 rc = fld_client_rpc(fld->lsf_control_exp,
306 fld_cache_insert(fld->lsf_cache, range);
310 EXPORT_SYMBOL(fld_server_lookup);
313 * All MDT server handle fld lookup operation. But only MDT0 has fld index.
314 * if entry is not found in cache we need to forward lookup request to MDT0
317 static int fld_server_handle(struct lu_server_fld *fld,
318 const struct lu_env *env,
319 __u32 opc, struct lu_seq_range *range,
320 struct fld_thread_info *info)
327 rc = fld_server_lookup(fld, env,
328 range->lsr_start, range);
335 CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
336 DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
342 static int fld_req_handle(struct ptlrpc_request *req,
343 struct fld_thread_info *info)
345 struct obd_export *exp = req->rq_export;
346 struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
347 struct lu_seq_range *in;
348 struct lu_seq_range *out;
353 rc = req_capsule_server_pack(info->fti_pill);
355 RETURN(err_serious(rc));
357 opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
359 in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
361 RETURN(err_serious(-EPROTO));
362 out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
364 RETURN(err_serious(-EPROTO));
367 /* For old 2.0 client, the 'lsr_flags' is uninitialized.
368 * Set it as 'LU_SEQ_RANGE_MDT' by default.
369 * Old 2.0 liblustre client cannot talk with new 2.1 server. */
370 if (!(exp->exp_connect_flags & OBD_CONNECT_64BITHASH) &&
372 out->lsr_flags = LU_SEQ_RANGE_MDT;
374 rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
375 req->rq_svc_thread->t_env,
378 rc = err_serious(-EPROTO);
383 static void fld_thread_info_init(struct ptlrpc_request *req,
384 struct fld_thread_info *info)
386 info->fti_pill = &req->rq_pill;
387 /* Init request capsule. */
388 req_capsule_init(info->fti_pill, req, RCL_SERVER);
389 req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
392 static void fld_thread_info_fini(struct fld_thread_info *info)
394 req_capsule_fini(info->fti_pill);
397 static int fld_handle(struct ptlrpc_request *req)
399 struct fld_thread_info *info;
400 const struct lu_env *env;
403 env = req->rq_svc_thread->t_env;
404 LASSERT(env != NULL);
406 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
407 LASSERT(info != NULL);
409 fld_thread_info_init(req, info);
410 rc = fld_req_handle(req, info);
411 fld_thread_info_fini(info);
417 * Entry point for handling FLD RPCs called from MDT.
419 int fld_query(struct com_thread_info *info)
421 return fld_handle(info->cti_pill->rc_req);
423 EXPORT_SYMBOL(fld_query);
426 * Returns true, if fid is local to this server node.
428 * WARNING: this function is *not* guaranteed to return false if fid is
429 * remote: it makes an educated conservative guess only.
431 * fid_is_local() is supposed to be used in assertion checks only.
433 int fid_is_local(const struct lu_env *env,
434 struct lu_site *site, const struct lu_fid *fid)
437 struct md_site *msite;
438 struct lu_seq_range *range;
439 struct fld_thread_info *info;
442 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
443 range = &info->fti_lrange;
445 result = 1; /* conservatively assume fid is local */
446 msite = lu_site2md(site);
447 if (msite->ms_client_fld != NULL) {
450 rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
451 fid_seq(fid), range);
453 result = (range->lsr_index == msite->ms_node_id);
457 EXPORT_SYMBOL(fid_is_local);
459 static void fld_server_proc_fini(struct lu_server_fld *fld);
462 static int fld_server_proc_init(struct lu_server_fld *fld)
467 fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
469 fld_server_proc_list, fld);
470 if (IS_ERR(fld->lsf_proc_dir)) {
471 rc = PTR_ERR(fld->lsf_proc_dir);
478 static void fld_server_proc_fini(struct lu_server_fld *fld)
481 if (fld->lsf_proc_dir != NULL) {
482 if (!IS_ERR(fld->lsf_proc_dir))
483 lprocfs_remove(&fld->lsf_proc_dir);
484 fld->lsf_proc_dir = NULL;
489 static int fld_server_proc_init(struct lu_server_fld *fld)
494 static void fld_server_proc_fini(struct lu_server_fld *fld)
500 int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
501 const char *prefix, const struct lu_env *env,
504 int cache_size, cache_threshold;
505 struct lu_seq_range range;
509 snprintf(fld->lsf_name, sizeof(fld->lsf_name),
512 cache_size = FLD_SERVER_CACHE_SIZE /
513 sizeof(struct fld_cache_entry);
515 cache_threshold = cache_size *
516 FLD_SERVER_CACHE_THRESHOLD / 100;
518 cfs_mutex_init(&fld->lsf_lock);
519 fld->lsf_cache = fld_cache_init(fld->lsf_name,
520 cache_size, cache_threshold);
521 if (IS_ERR(fld->lsf_cache)) {
522 rc = PTR_ERR(fld->lsf_cache);
523 fld->lsf_cache = NULL;
528 rc = fld_index_init(fld, env, dt);
534 rc = fld_server_proc_init(fld);
538 fld->lsf_control_exp = NULL;
540 /* Insert reserved sequence number of ".lustre" into fld cache. */
541 range.lsr_start = FID_SEQ_DOT_LUSTRE;
542 range.lsr_end = FID_SEQ_DOT_LUSTRE + 1;
544 range.lsr_flags = LU_SEQ_RANGE_MDT;
545 fld_cache_insert(fld->lsf_cache, &range);
550 fld_server_fini(fld, env);
553 EXPORT_SYMBOL(fld_server_init);
555 void fld_server_fini(struct lu_server_fld *fld,
556 const struct lu_env *env)
560 fld_server_proc_fini(fld);
561 fld_index_fini(fld, env);
563 if (fld->lsf_cache != NULL) {
564 if (!IS_ERR(fld->lsf_cache))
565 fld_cache_fini(fld->lsf_cache);
566 fld->lsf_cache = NULL;
571 EXPORT_SYMBOL(fld_server_fini);
573 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
574 MODULE_DESCRIPTION("Lustre FLD");
575 MODULE_LICENSE("GPL");
577 cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);