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 <asm/div64.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 /* TODO: these 3 functions are copies of flow-control code from mdc_lib.c
64 * It should be common thing. The same about mdc RPC lock */
65 static int fld_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
69 client_obd_list_lock(&cli->cl_loi_list_lock);
70 rc = cfs_list_empty(&mcw->mcw_entry);
71 client_obd_list_unlock(&cli->cl_loi_list_lock);
75 static void fld_enter_request(struct client_obd *cli)
77 struct mdc_cache_waiter mcw;
78 struct l_wait_info lwi = { 0 };
80 client_obd_list_lock(&cli->cl_loi_list_lock);
81 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
82 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
83 init_waitqueue_head(&mcw.mcw_waitq);
84 client_obd_list_unlock(&cli->cl_loi_list_lock);
85 l_wait_event(mcw.mcw_waitq, fld_req_avail(cli, &mcw), &lwi);
87 cli->cl_r_in_flight++;
88 client_obd_list_unlock(&cli->cl_loi_list_lock);
92 static void fld_exit_request(struct client_obd *cli)
95 struct mdc_cache_waiter *mcw;
97 client_obd_list_lock(&cli->cl_loi_list_lock);
98 cli->cl_r_in_flight--;
99 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
101 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
102 /* No free request slots anymore */
106 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
107 cfs_list_del_init(&mcw->mcw_entry);
108 cli->cl_r_in_flight++;
109 wake_up(&mcw->mcw_waitq);
111 client_obd_list_unlock(&cli->cl_loi_list_lock);
114 static int fld_rrb_hash(struct lu_client_fld *fld,
117 LASSERT(fld->lcf_count > 0);
118 return do_div(seq, fld->lcf_count);
121 static struct lu_fld_target *
122 fld_rrb_scan(struct lu_client_fld *fld, seqno_t seq)
124 struct lu_fld_target *target;
128 /* Because almost all of special sequence located in MDT0,
129 * it should go to index 0 directly, instead of calculating
130 * hash again, and also if other MDTs is not being connected,
131 * the fld lookup requests(for seq on MDT0) should not be
132 * blocked because of other MDTs */
133 if (fid_seq_is_norm(seq))
134 hash = fld_rrb_hash(fld, seq);
138 cfs_list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
139 if (target->ft_idx == hash)
143 CERROR("%s: Can't find target by hash %d (seq "LPX64"). "
144 "Targets (%d):\n", fld->lcf_name, hash, seq,
147 cfs_list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
148 const char *srv_name = target->ft_srv != NULL ?
149 target->ft_srv->lsf_name : "<null>";
150 const char *exp_name = target->ft_exp != NULL ?
151 (char *)target->ft_exp->exp_obd->obd_uuid.uuid :
154 CERROR(" exp: 0x%p (%s), srv: 0x%p (%s), idx: "LPU64"\n",
155 target->ft_exp, exp_name, target->ft_srv,
156 srv_name, target->ft_idx);
160 * If target is not found, there is logical error anyway, so here is
161 * LBUG() to catch this situation.
167 struct lu_fld_hash fld_hash[] = {
170 .fh_hash_func = fld_rrb_hash,
171 .fh_scan_func = fld_rrb_scan
178 static struct lu_fld_target *
179 fld_client_get_target(struct lu_client_fld *fld, seqno_t seq)
181 struct lu_fld_target *target;
184 LASSERT(fld->lcf_hash != NULL);
186 spin_lock(&fld->lcf_lock);
187 target = fld->lcf_hash->fh_scan_func(fld, seq);
188 spin_unlock(&fld->lcf_lock);
190 if (target != NULL) {
191 CDEBUG(D_INFO, "%s: Found target (idx "LPU64
192 ") by seq "LPX64"\n", fld->lcf_name,
193 target->ft_idx, seq);
200 * Add export to FLD. This is usually done by CMM and LMV as they are main users
203 int fld_client_add_target(struct lu_client_fld *fld,
204 struct lu_fld_target *tar)
207 struct lu_fld_target *target, *tmp;
210 LASSERT(tar != NULL);
211 name = fld_target_name(tar);
212 LASSERT(name != NULL);
213 LASSERT(tar->ft_srv != NULL || tar->ft_exp != NULL);
215 if (fld->lcf_flags != LUSTRE_FLD_INIT) {
216 CERROR("%s: Attempt to add target %s (idx "LPU64") "
217 "on fly - skip it\n", fld->lcf_name, name,
221 CDEBUG(D_INFO, "%s: Adding target %s (idx "
222 LPU64")\n", fld->lcf_name, name, tar->ft_idx);
225 OBD_ALLOC_PTR(target);
229 spin_lock(&fld->lcf_lock);
230 cfs_list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
231 if (tmp->ft_idx == tar->ft_idx) {
232 spin_unlock(&fld->lcf_lock);
233 OBD_FREE_PTR(target);
234 CERROR("Target %s exists in FLD and known as %s:#"LPU64"\n",
235 name, fld_target_name(tmp), tmp->ft_idx);
240 target->ft_exp = tar->ft_exp;
241 if (target->ft_exp != NULL)
242 class_export_get(target->ft_exp);
243 target->ft_srv = tar->ft_srv;
244 target->ft_idx = tar->ft_idx;
246 cfs_list_add_tail(&target->ft_chain,
250 spin_unlock(&fld->lcf_lock);
254 EXPORT_SYMBOL(fld_client_add_target);
256 /* Remove export from FLD */
257 int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
259 struct lu_fld_target *target, *tmp;
262 spin_lock(&fld->lcf_lock);
263 cfs_list_for_each_entry_safe(target, tmp,
264 &fld->lcf_targets, ft_chain) {
265 if (target->ft_idx == idx) {
267 cfs_list_del(&target->ft_chain);
268 spin_unlock(&fld->lcf_lock);
270 if (target->ft_exp != NULL)
271 class_export_put(target->ft_exp);
273 OBD_FREE_PTR(target);
277 spin_unlock(&fld->lcf_lock);
280 EXPORT_SYMBOL(fld_client_del_target);
283 static int fld_client_proc_init(struct lu_client_fld *fld)
288 fld->lcf_proc_dir = lprocfs_seq_register(fld->lcf_name,
291 if (IS_ERR(fld->lcf_proc_dir)) {
292 CERROR("%s: LProcFS failed in fld-init\n",
294 rc = PTR_ERR(fld->lcf_proc_dir);
298 rc = lprocfs_seq_add_vars(fld->lcf_proc_dir,
299 fld_client_proc_list, fld);
301 CERROR("%s: Can't init FLD proc, rc %d\n",
303 GOTO(out_cleanup, rc);
309 fld_client_proc_fini(fld);
313 void fld_client_proc_fini(struct lu_client_fld *fld)
316 if (fld->lcf_proc_dir) {
317 if (!IS_ERR(fld->lcf_proc_dir))
318 lprocfs_remove(&fld->lcf_proc_dir);
319 fld->lcf_proc_dir = NULL;
324 static int fld_client_proc_init(struct lu_client_fld *fld)
329 void fld_client_proc_fini(struct lu_client_fld *fld)
335 EXPORT_SYMBOL(fld_client_proc_fini);
337 static inline int hash_is_sane(int hash)
339 return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
342 int fld_client_init(struct lu_client_fld *fld,
343 const char *prefix, int hash)
345 int cache_size, cache_threshold;
349 LASSERT(fld != NULL);
351 snprintf(fld->lcf_name, sizeof(fld->lcf_name),
354 if (!hash_is_sane(hash)) {
355 CERROR("%s: Wrong hash function %#x\n",
356 fld->lcf_name, hash);
361 spin_lock_init(&fld->lcf_lock);
362 fld->lcf_hash = &fld_hash[hash];
363 fld->lcf_flags = LUSTRE_FLD_INIT;
364 CFS_INIT_LIST_HEAD(&fld->lcf_targets);
366 cache_size = FLD_CLIENT_CACHE_SIZE /
367 sizeof(struct fld_cache_entry);
369 cache_threshold = cache_size *
370 FLD_CLIENT_CACHE_THRESHOLD / 100;
372 fld->lcf_cache = fld_cache_init(fld->lcf_name,
373 cache_size, cache_threshold);
374 if (IS_ERR(fld->lcf_cache)) {
375 rc = PTR_ERR(fld->lcf_cache);
376 fld->lcf_cache = NULL;
380 rc = fld_client_proc_init(fld);
386 fld_client_fini(fld);
388 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
389 fld->lcf_name, fld->lcf_hash->fh_name);
392 EXPORT_SYMBOL(fld_client_init);
394 void fld_client_fini(struct lu_client_fld *fld)
396 struct lu_fld_target *target, *tmp;
399 spin_lock(&fld->lcf_lock);
400 cfs_list_for_each_entry_safe(target, tmp,
401 &fld->lcf_targets, ft_chain) {
403 cfs_list_del(&target->ft_chain);
404 if (target->ft_exp != NULL)
405 class_export_put(target->ft_exp);
406 OBD_FREE_PTR(target);
408 spin_unlock(&fld->lcf_lock);
410 if (fld->lcf_cache != NULL) {
411 if (!IS_ERR(fld->lcf_cache))
412 fld_cache_fini(fld->lcf_cache);
413 fld->lcf_cache = NULL;
418 EXPORT_SYMBOL(fld_client_fini);
420 int fld_client_rpc(struct obd_export *exp,
421 struct lu_seq_range *range, __u32 fld_op,
422 struct ptlrpc_request **reqp)
424 struct ptlrpc_request *req = NULL;
425 struct lu_seq_range *prange;
428 struct obd_import *imp;
431 LASSERT(exp != NULL);
433 imp = class_exp2cliimp(exp);
436 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
437 LUSTRE_MDS_VERSION, FLD_QUERY);
441 /* XXX: only needed when talking to old server(< 2.6), it should
442 * be removed when < 2.6 server is not supported */
443 op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
446 if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS)
447 req->rq_allow_replay = 1;
450 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
451 LUSTRE_MDS_VERSION, FLD_READ);
455 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
456 RCL_SERVER, PAGE_CACHE_SIZE);
466 prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
468 ptlrpc_request_set_replen(req);
469 req->rq_request_portal = FLD_REQUEST_PORTAL;
470 req->rq_reply_portal = MDC_REPLY_PORTAL;
471 ptlrpc_at_set_req_timeout(req);
473 fld_enter_request(&exp->exp_obd->u.cli);
474 rc = ptlrpc_queue_wait(req);
475 fld_exit_request(&exp->exp_obd->u.cli);
479 if (fld_op == FLD_QUERY) {
480 prange = req_capsule_server_get(&req->rq_pill,
483 GOTO(out_req, rc = -EFAULT);
489 if (rc != 0 || reqp == NULL) {
490 ptlrpc_req_finished(req);
500 int fld_client_lookup(struct lu_client_fld *fld, seqno_t seq, mdsno_t *mds,
501 __u32 flags, const struct lu_env *env)
503 struct lu_seq_range res = { 0 };
504 struct lu_fld_target *target;
508 fld->lcf_flags |= LUSTRE_FLD_RUN;
510 rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
512 *mds = res.lsr_index;
516 /* Can not find it in the cache */
517 target = fld_client_get_target(fld, seq);
518 LASSERT(target != NULL);
520 CDEBUG(D_INFO, "%s: Lookup fld entry (seq: "LPX64") on "
521 "target %s (idx "LPU64")\n", fld->lcf_name, seq,
522 fld_target_name(target), target->ft_idx);
525 fld_range_set_type(&res, flags);
527 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
528 if (target->ft_srv != NULL) {
529 LASSERT(env != NULL);
530 rc = fld_server_lookup(env, target->ft_srv, seq, &res);
534 rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
538 *mds = res.lsr_index;
539 fld_cache_insert(fld->lcf_cache, &res);
544 EXPORT_SYMBOL(fld_client_lookup);
546 void fld_client_flush(struct lu_client_fld *fld)
548 fld_cache_flush(fld->lcf_cache);
550 EXPORT_SYMBOL(fld_client_flush);
554 struct proc_dir_entry *fld_type_proc_dir;
556 static int __init fld_mod_init(void)
558 fld_type_proc_dir = lprocfs_seq_register(LUSTRE_FLD_NAME,
561 if (IS_ERR(fld_type_proc_dir))
562 return PTR_ERR(fld_type_proc_dir);
564 #ifdef HAVE_SERVER_SUPPORT
565 fld_server_mod_init();
571 static void __exit fld_mod_exit(void)
573 #ifdef HAVE_SERVER_SUPPORT
574 fld_server_mod_exit();
577 if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
578 lprocfs_remove(&fld_type_proc_dir);
579 fld_type_proc_dir = NULL;
583 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
584 MODULE_DESCRIPTION("Lustre FLD");
585 MODULE_LICENSE("GPL");
587 cfs_module(mdd, LUSTRE_VERSION_STRING, fld_mod_init, fld_mod_exit);
588 #endif /* __KERNEL__ */