4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/fld/fld_request.c
38 * FLD (Fids Location Database)
40 * Author: Yury Umanets <umka@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_FLD
46 # include <libcfs/libcfs.h>
47 # include <linux/module.h>
48 # include <linux/math64.h>
49 #else /* __KERNEL__ */
50 # include <liblustre.h>
51 # include <libcfs/list.h>
55 #include <obd_class.h>
56 #include <obd_support.h>
57 #include <lprocfs_status.h>
58 #include <lustre_req_layout.h>
59 #include <lustre_fld.h>
60 #include <lustre_mdc.h>
61 #include "fld_internal.h"
63 static int fld_rrb_hash(struct lu_client_fld *fld, u64 seq)
65 LASSERT(fld->lcf_count > 0);
66 return do_div(seq, fld->lcf_count);
69 static struct lu_fld_target *
70 fld_rrb_scan(struct lu_client_fld *fld, u64 seq)
72 struct lu_fld_target *target;
76 /* Because almost all of special sequence located in MDT0,
77 * it should go to index 0 directly, instead of calculating
78 * hash again, and also if other MDTs is not being connected,
79 * the fld lookup requests(for seq on MDT0) should not be
80 * blocked because of other MDTs */
81 if (fid_seq_is_norm(seq))
82 hash = fld_rrb_hash(fld, seq);
87 list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
88 if (target->ft_idx == hash)
93 /* It is possible the remote target(MDT) are not connected to
94 * with client yet, so we will refer this to MDT0, which should
95 * be connected during mount */
100 CERROR("%s: Can't find target by hash %d (seq "LPX64"). "
101 "Targets (%d):\n", fld->lcf_name, hash, seq,
104 list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
105 const char *srv_name = target->ft_srv != NULL ?
106 target->ft_srv->lsf_name : "<null>";
107 const char *exp_name = target->ft_exp != NULL ?
108 (char *)target->ft_exp->exp_obd->obd_uuid.uuid :
111 CERROR(" exp: 0x%p (%s), srv: 0x%p (%s), idx: "LPU64"\n",
112 target->ft_exp, exp_name, target->ft_srv,
113 srv_name, target->ft_idx);
117 * If target is not found, there is logical error anyway, so here is
118 * LBUG() to catch this situation.
124 struct lu_fld_hash fld_hash[] = {
127 .fh_hash_func = fld_rrb_hash,
128 .fh_scan_func = fld_rrb_scan
135 static struct lu_fld_target *
136 fld_client_get_target(struct lu_client_fld *fld, u64 seq)
138 struct lu_fld_target *target;
141 LASSERT(fld->lcf_hash != NULL);
143 spin_lock(&fld->lcf_lock);
144 target = fld->lcf_hash->fh_scan_func(fld, seq);
145 spin_unlock(&fld->lcf_lock);
147 if (target != NULL) {
148 CDEBUG(D_INFO, "%s: Found target (idx "LPU64
149 ") by seq "LPX64"\n", fld->lcf_name,
150 target->ft_idx, seq);
157 * Add export to FLD. This is usually done by CMM and LMV as they are main users
160 int fld_client_add_target(struct lu_client_fld *fld,
161 struct lu_fld_target *tar)
164 struct lu_fld_target *target, *tmp;
167 LASSERT(tar != NULL);
168 name = fld_target_name(tar);
169 LASSERT(name != NULL);
170 LASSERT(tar->ft_srv != NULL || tar->ft_exp != NULL);
172 if (fld->lcf_flags != LUSTRE_FLD_INIT) {
173 CERROR("%s: Attempt to add target %s (idx "LPU64") "
174 "on fly - skip it\n", fld->lcf_name, name,
178 CDEBUG(D_INFO, "%s: Adding target %s (idx "
179 LPU64")\n", fld->lcf_name, name, tar->ft_idx);
182 OBD_ALLOC_PTR(target);
186 spin_lock(&fld->lcf_lock);
187 list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
188 if (tmp->ft_idx == tar->ft_idx) {
189 spin_unlock(&fld->lcf_lock);
190 OBD_FREE_PTR(target);
191 CERROR("Target %s exists in FLD and known as %s:#"LPU64"\n",
192 name, fld_target_name(tmp), tmp->ft_idx);
197 target->ft_exp = tar->ft_exp;
198 if (target->ft_exp != NULL)
199 class_export_get(target->ft_exp);
200 target->ft_srv = tar->ft_srv;
201 target->ft_idx = tar->ft_idx;
203 list_add_tail(&target->ft_chain, &fld->lcf_targets);
206 spin_unlock(&fld->lcf_lock);
210 EXPORT_SYMBOL(fld_client_add_target);
212 /* Remove export from FLD */
213 int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
215 struct lu_fld_target *target, *tmp;
218 spin_lock(&fld->lcf_lock);
219 list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
220 if (target->ft_idx == idx) {
222 list_del(&target->ft_chain);
223 spin_unlock(&fld->lcf_lock);
225 if (target->ft_exp != NULL)
226 class_export_put(target->ft_exp);
228 OBD_FREE_PTR(target);
232 spin_unlock(&fld->lcf_lock);
235 EXPORT_SYMBOL(fld_client_del_target);
238 static int fld_client_proc_init(struct lu_client_fld *fld)
243 fld->lcf_proc_dir = lprocfs_seq_register(fld->lcf_name,
246 if (IS_ERR(fld->lcf_proc_dir)) {
247 CERROR("%s: LProcFS failed in fld-init\n",
249 rc = PTR_ERR(fld->lcf_proc_dir);
253 rc = lprocfs_seq_add_vars(fld->lcf_proc_dir,
254 fld_client_proc_list, fld);
256 CERROR("%s: Can't init FLD proc, rc %d\n",
258 GOTO(out_cleanup, rc);
264 fld_client_proc_fini(fld);
268 void fld_client_proc_fini(struct lu_client_fld *fld)
271 if (fld->lcf_proc_dir) {
272 if (!IS_ERR(fld->lcf_proc_dir))
273 lprocfs_remove(&fld->lcf_proc_dir);
274 fld->lcf_proc_dir = NULL;
279 static int fld_client_proc_init(struct lu_client_fld *fld)
284 void fld_client_proc_fini(struct lu_client_fld *fld)
290 EXPORT_SYMBOL(fld_client_proc_fini);
292 static inline int hash_is_sane(int hash)
294 return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
297 int fld_client_init(struct lu_client_fld *fld,
298 const char *prefix, int hash)
300 int cache_size, cache_threshold;
304 LASSERT(fld != NULL);
306 snprintf(fld->lcf_name, sizeof(fld->lcf_name),
309 if (!hash_is_sane(hash)) {
310 CERROR("%s: Wrong hash function %#x\n",
311 fld->lcf_name, hash);
316 spin_lock_init(&fld->lcf_lock);
317 fld->lcf_hash = &fld_hash[hash];
318 fld->lcf_flags = LUSTRE_FLD_INIT;
319 INIT_LIST_HEAD(&fld->lcf_targets);
321 cache_size = FLD_CLIENT_CACHE_SIZE /
322 sizeof(struct fld_cache_entry);
324 cache_threshold = cache_size *
325 FLD_CLIENT_CACHE_THRESHOLD / 100;
327 fld->lcf_cache = fld_cache_init(fld->lcf_name,
328 cache_size, cache_threshold);
329 if (IS_ERR(fld->lcf_cache)) {
330 rc = PTR_ERR(fld->lcf_cache);
331 fld->lcf_cache = NULL;
335 rc = fld_client_proc_init(fld);
341 fld_client_fini(fld);
343 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
344 fld->lcf_name, fld->lcf_hash->fh_name);
347 EXPORT_SYMBOL(fld_client_init);
349 void fld_client_fini(struct lu_client_fld *fld)
351 struct lu_fld_target *target, *tmp;
354 spin_lock(&fld->lcf_lock);
355 list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
357 list_del(&target->ft_chain);
358 if (target->ft_exp != NULL)
359 class_export_put(target->ft_exp);
360 OBD_FREE_PTR(target);
362 spin_unlock(&fld->lcf_lock);
364 if (fld->lcf_cache != NULL) {
365 if (!IS_ERR(fld->lcf_cache))
366 fld_cache_fini(fld->lcf_cache);
367 fld->lcf_cache = NULL;
372 EXPORT_SYMBOL(fld_client_fini);
374 int fld_client_rpc(struct obd_export *exp,
375 struct lu_seq_range *range, __u32 fld_op,
376 struct ptlrpc_request **reqp)
378 struct ptlrpc_request *req = NULL;
379 struct lu_seq_range *prange;
382 struct obd_import *imp;
385 LASSERT(exp != NULL);
388 imp = class_exp2cliimp(exp);
391 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
392 LUSTRE_MDS_VERSION, FLD_QUERY);
396 /* XXX: only needed when talking to old server(< 2.6), it should
397 * be removed when < 2.6 server is not supported */
398 op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
401 /* For MDS_MDS seq lookup, it will always use LWP connection,
402 * but LWP will be evicted after restart, so cause the error.
403 * so we will set no_delay for seq lookup request, once the
404 * request fails because of the eviction. always retry here */
405 if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS) {
406 req->rq_allow_replay = 1;
407 req->rq_no_delay = 1;
411 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
412 LUSTRE_MDS_VERSION, FLD_READ);
416 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
417 RCL_SERVER, PAGE_CACHE_SIZE);
427 prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
429 ptlrpc_request_set_replen(req);
430 req->rq_request_portal = FLD_REQUEST_PORTAL;
431 req->rq_reply_portal = MDC_REPLY_PORTAL;
432 ptlrpc_at_set_req_timeout(req);
434 obd_get_request_slot(&exp->exp_obd->u.cli);
435 rc = ptlrpc_queue_wait(req);
436 obd_put_request_slot(&exp->exp_obd->u.cli);
438 if (imp->imp_state != LUSTRE_IMP_CLOSED) {
439 /* Since LWP is not replayable, so it will keep
440 * trying unless umount happens, otherwise it would
441 * cause unecessary failure of the application. */
442 ptlrpc_req_finished(req);
449 if (fld_op == FLD_QUERY) {
450 prange = req_capsule_server_get(&req->rq_pill,
453 GOTO(out_req, rc = -EFAULT);
459 if (rc != 0 || reqp == NULL) {
460 ptlrpc_req_finished(req);
470 int fld_client_lookup(struct lu_client_fld *fld, u64 seq, u32 *mds,
471 __u32 flags, const struct lu_env *env)
473 struct lu_seq_range res = { 0 };
474 struct lu_fld_target *target;
478 fld->lcf_flags |= LUSTRE_FLD_RUN;
480 rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
482 *mds = res.lsr_index;
486 /* Can not find it in the cache */
487 target = fld_client_get_target(fld, seq);
488 LASSERT(target != NULL);
490 CDEBUG(D_INFO, "%s: Lookup fld entry (seq: "LPX64") on "
491 "target %s (idx "LPU64")\n", fld->lcf_name, seq,
492 fld_target_name(target), target->ft_idx);
495 fld_range_set_type(&res, flags);
497 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
498 if (target->ft_srv != NULL) {
499 LASSERT(env != NULL);
500 rc = fld_server_lookup(env, target->ft_srv, seq, &res);
504 rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
508 *mds = res.lsr_index;
509 fld_cache_insert(fld->lcf_cache, &res);
514 EXPORT_SYMBOL(fld_client_lookup);
516 void fld_client_flush(struct lu_client_fld *fld)
518 fld_cache_flush(fld->lcf_cache);
520 EXPORT_SYMBOL(fld_client_flush);
524 struct proc_dir_entry *fld_type_proc_dir;
526 static int __init fld_mod_init(void)
528 fld_type_proc_dir = lprocfs_seq_register(LUSTRE_FLD_NAME,
531 if (IS_ERR(fld_type_proc_dir))
532 return PTR_ERR(fld_type_proc_dir);
534 #ifdef HAVE_SERVER_SUPPORT
535 fld_server_mod_init();
541 static void __exit fld_mod_exit(void)
543 #ifdef HAVE_SERVER_SUPPORT
544 fld_server_mod_exit();
547 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
548 lprocfs_remove(&fld_type_proc_dir);
549 fld_type_proc_dir = NULL;
553 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
554 MODULE_DESCRIPTION("Lustre FLD");
555 MODULE_LICENSE("GPL");
557 cfs_module(mdd, LUSTRE_VERSION_STRING, fld_mod_init, fld_mod_exit);
558 #endif /* __KERNEL__ */