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,
66 LASSERT(fld->lcf_count > 0);
67 return do_div(seq, fld->lcf_count);
70 static struct lu_fld_target *
71 fld_rrb_scan(struct lu_client_fld *fld, seqno_t seq)
73 struct lu_fld_target *target;
77 /* Because almost all of special sequence located in MDT0,
78 * it should go to index 0 directly, instead of calculating
79 * hash again, and also if other MDTs is not being connected,
80 * the fld lookup requests(for seq on MDT0) should not be
81 * blocked because of other MDTs */
82 if (fid_seq_is_norm(seq))
83 hash = fld_rrb_hash(fld, seq);
87 cfs_list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
88 if (target->ft_idx == hash)
92 CERROR("%s: Can't find target by hash %d (seq "LPX64"). "
93 "Targets (%d):\n", fld->lcf_name, hash, seq,
96 cfs_list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
97 const char *srv_name = target->ft_srv != NULL ?
98 target->ft_srv->lsf_name : "<null>";
99 const char *exp_name = target->ft_exp != NULL ?
100 (char *)target->ft_exp->exp_obd->obd_uuid.uuid :
103 CERROR(" exp: 0x%p (%s), srv: 0x%p (%s), idx: "LPU64"\n",
104 target->ft_exp, exp_name, target->ft_srv,
105 srv_name, target->ft_idx);
109 * If target is not found, there is logical error anyway, so here is
110 * LBUG() to catch this situation.
116 struct lu_fld_hash fld_hash[] = {
119 .fh_hash_func = fld_rrb_hash,
120 .fh_scan_func = fld_rrb_scan
127 static struct lu_fld_target *
128 fld_client_get_target(struct lu_client_fld *fld, seqno_t seq)
130 struct lu_fld_target *target;
133 LASSERT(fld->lcf_hash != NULL);
135 spin_lock(&fld->lcf_lock);
136 target = fld->lcf_hash->fh_scan_func(fld, seq);
137 spin_unlock(&fld->lcf_lock);
139 if (target != NULL) {
140 CDEBUG(D_INFO, "%s: Found target (idx "LPU64
141 ") by seq "LPX64"\n", fld->lcf_name,
142 target->ft_idx, seq);
149 * Add export to FLD. This is usually done by CMM and LMV as they are main users
152 int fld_client_add_target(struct lu_client_fld *fld,
153 struct lu_fld_target *tar)
156 struct lu_fld_target *target, *tmp;
159 LASSERT(tar != NULL);
160 name = fld_target_name(tar);
161 LASSERT(name != NULL);
162 LASSERT(tar->ft_srv != NULL || tar->ft_exp != NULL);
164 if (fld->lcf_flags != LUSTRE_FLD_INIT) {
165 CERROR("%s: Attempt to add target %s (idx "LPU64") "
166 "on fly - skip it\n", fld->lcf_name, name,
170 CDEBUG(D_INFO, "%s: Adding target %s (idx "
171 LPU64")\n", fld->lcf_name, name, tar->ft_idx);
174 OBD_ALLOC_PTR(target);
178 spin_lock(&fld->lcf_lock);
179 cfs_list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
180 if (tmp->ft_idx == tar->ft_idx) {
181 spin_unlock(&fld->lcf_lock);
182 OBD_FREE_PTR(target);
183 CERROR("Target %s exists in FLD and known as %s:#"LPU64"\n",
184 name, fld_target_name(tmp), tmp->ft_idx);
189 target->ft_exp = tar->ft_exp;
190 if (target->ft_exp != NULL)
191 class_export_get(target->ft_exp);
192 target->ft_srv = tar->ft_srv;
193 target->ft_idx = tar->ft_idx;
195 cfs_list_add_tail(&target->ft_chain,
199 spin_unlock(&fld->lcf_lock);
203 EXPORT_SYMBOL(fld_client_add_target);
205 /* Remove export from FLD */
206 int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
208 struct lu_fld_target *target, *tmp;
211 spin_lock(&fld->lcf_lock);
212 cfs_list_for_each_entry_safe(target, tmp,
213 &fld->lcf_targets, ft_chain) {
214 if (target->ft_idx == idx) {
216 cfs_list_del(&target->ft_chain);
217 spin_unlock(&fld->lcf_lock);
219 if (target->ft_exp != NULL)
220 class_export_put(target->ft_exp);
222 OBD_FREE_PTR(target);
226 spin_unlock(&fld->lcf_lock);
229 EXPORT_SYMBOL(fld_client_del_target);
232 static int fld_client_proc_init(struct lu_client_fld *fld)
237 fld->lcf_proc_dir = lprocfs_seq_register(fld->lcf_name,
240 if (IS_ERR(fld->lcf_proc_dir)) {
241 CERROR("%s: LProcFS failed in fld-init\n",
243 rc = PTR_ERR(fld->lcf_proc_dir);
247 rc = lprocfs_seq_add_vars(fld->lcf_proc_dir,
248 fld_client_proc_list, fld);
250 CERROR("%s: Can't init FLD proc, rc %d\n",
252 GOTO(out_cleanup, rc);
258 fld_client_proc_fini(fld);
262 void fld_client_proc_fini(struct lu_client_fld *fld)
265 if (fld->lcf_proc_dir) {
266 if (!IS_ERR(fld->lcf_proc_dir))
267 lprocfs_remove(&fld->lcf_proc_dir);
268 fld->lcf_proc_dir = NULL;
273 static int fld_client_proc_init(struct lu_client_fld *fld)
278 void fld_client_proc_fini(struct lu_client_fld *fld)
284 EXPORT_SYMBOL(fld_client_proc_fini);
286 static inline int hash_is_sane(int hash)
288 return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
291 int fld_client_init(struct lu_client_fld *fld,
292 const char *prefix, int hash)
294 int cache_size, cache_threshold;
298 LASSERT(fld != NULL);
300 snprintf(fld->lcf_name, sizeof(fld->lcf_name),
303 if (!hash_is_sane(hash)) {
304 CERROR("%s: Wrong hash function %#x\n",
305 fld->lcf_name, hash);
310 spin_lock_init(&fld->lcf_lock);
311 fld->lcf_hash = &fld_hash[hash];
312 fld->lcf_flags = LUSTRE_FLD_INIT;
313 CFS_INIT_LIST_HEAD(&fld->lcf_targets);
315 cache_size = FLD_CLIENT_CACHE_SIZE /
316 sizeof(struct fld_cache_entry);
318 cache_threshold = cache_size *
319 FLD_CLIENT_CACHE_THRESHOLD / 100;
321 fld->lcf_cache = fld_cache_init(fld->lcf_name,
322 cache_size, cache_threshold);
323 if (IS_ERR(fld->lcf_cache)) {
324 rc = PTR_ERR(fld->lcf_cache);
325 fld->lcf_cache = NULL;
329 rc = fld_client_proc_init(fld);
335 fld_client_fini(fld);
337 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
338 fld->lcf_name, fld->lcf_hash->fh_name);
341 EXPORT_SYMBOL(fld_client_init);
343 void fld_client_fini(struct lu_client_fld *fld)
345 struct lu_fld_target *target, *tmp;
348 spin_lock(&fld->lcf_lock);
349 cfs_list_for_each_entry_safe(target, tmp,
350 &fld->lcf_targets, ft_chain) {
352 cfs_list_del(&target->ft_chain);
353 if (target->ft_exp != NULL)
354 class_export_put(target->ft_exp);
355 OBD_FREE_PTR(target);
357 spin_unlock(&fld->lcf_lock);
359 if (fld->lcf_cache != NULL) {
360 if (!IS_ERR(fld->lcf_cache))
361 fld_cache_fini(fld->lcf_cache);
362 fld->lcf_cache = NULL;
367 EXPORT_SYMBOL(fld_client_fini);
369 int fld_client_rpc(struct obd_export *exp,
370 struct lu_seq_range *range, __u32 fld_op,
371 struct ptlrpc_request **reqp)
373 struct ptlrpc_request *req = NULL;
374 struct lu_seq_range *prange;
377 struct obd_import *imp;
380 LASSERT(exp != NULL);
383 imp = class_exp2cliimp(exp);
386 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
387 LUSTRE_MDS_VERSION, FLD_QUERY);
391 /* XXX: only needed when talking to old server(< 2.6), it should
392 * be removed when < 2.6 server is not supported */
393 op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
396 /* For MDS_MDS seq lookup, it will always use LWP connection,
397 * but LWP will be evicted after restart, so cause the error.
398 * so we will set no_delay for seq lookup request, once the
399 * request fails because of the eviction. always retry here */
400 if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS) {
401 req->rq_allow_replay = 1;
402 req->rq_no_delay = 1;
406 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
407 LUSTRE_MDS_VERSION, FLD_READ);
411 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
412 RCL_SERVER, PAGE_CACHE_SIZE);
422 prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
424 ptlrpc_request_set_replen(req);
425 req->rq_request_portal = FLD_REQUEST_PORTAL;
426 req->rq_reply_portal = MDC_REPLY_PORTAL;
427 ptlrpc_at_set_req_timeout(req);
429 obd_get_request_slot(&exp->exp_obd->u.cli);
430 rc = ptlrpc_queue_wait(req);
431 obd_put_request_slot(&exp->exp_obd->u.cli);
433 if (rc == -EWOULDBLOCK) {
434 /* For no_delay req(see above), EWOULDBLOCK means the
435 * connection is being evicted, but this seq lookup
436 * should not return error, since it would cause
437 * unecessary failure of the application, instead
438 * it should retry here */
439 ptlrpc_req_finished(req);
446 if (fld_op == FLD_QUERY) {
447 prange = req_capsule_server_get(&req->rq_pill,
450 GOTO(out_req, rc = -EFAULT);
456 if (rc != 0 || reqp == NULL) {
457 ptlrpc_req_finished(req);
467 int fld_client_lookup(struct lu_client_fld *fld, seqno_t seq, mdsno_t *mds,
468 __u32 flags, const struct lu_env *env)
470 struct lu_seq_range res = { 0 };
471 struct lu_fld_target *target;
475 fld->lcf_flags |= LUSTRE_FLD_RUN;
477 rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
479 *mds = res.lsr_index;
483 /* Can not find it in the cache */
484 target = fld_client_get_target(fld, seq);
485 LASSERT(target != NULL);
487 CDEBUG(D_INFO, "%s: Lookup fld entry (seq: "LPX64") on "
488 "target %s (idx "LPU64")\n", fld->lcf_name, seq,
489 fld_target_name(target), target->ft_idx);
492 fld_range_set_type(&res, flags);
494 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
495 if (target->ft_srv != NULL) {
496 LASSERT(env != NULL);
497 rc = fld_server_lookup(env, target->ft_srv, seq, &res);
501 rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
505 *mds = res.lsr_index;
506 fld_cache_insert(fld->lcf_cache, &res);
511 EXPORT_SYMBOL(fld_client_lookup);
513 void fld_client_flush(struct lu_client_fld *fld)
515 fld_cache_flush(fld->lcf_cache);
517 EXPORT_SYMBOL(fld_client_flush);
521 struct proc_dir_entry *fld_type_proc_dir;
523 static int __init fld_mod_init(void)
525 fld_type_proc_dir = lprocfs_seq_register(LUSTRE_FLD_NAME,
528 if (IS_ERR(fld_type_proc_dir))
529 return PTR_ERR(fld_type_proc_dir);
531 #ifdef HAVE_SERVER_SUPPORT
532 fld_server_mod_init();
538 static void __exit fld_mod_exit(void)
540 #ifdef HAVE_SERVER_SUPPORT
541 fld_server_mod_exit();
544 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
545 lprocfs_remove(&fld_type_proc_dir);
546 fld_type_proc_dir = NULL;
550 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
551 MODULE_DESCRIPTION("Lustre FLD");
552 MODULE_LICENSE("GPL");
554 cfs_module(mdd, LUSTRE_VERSION_STRING, fld_mod_init, fld_mod_exit);
555 #endif /* __KERNEL__ */