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 2008 Sun Microsystems, Inc. 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/fid/fid_handler.c
38 * Lustre Sequence Manager
40 * Author: Yury Umanets <umka@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_FID
49 # include <libcfs/libcfs.h>
50 # include <linux/module.h>
51 #else /* __KERNEL__ */
52 # include <liblustre.h>
56 #include <obd_class.h>
57 #include <dt_object.h>
58 #include <md_object.h>
59 #include <obd_support.h>
60 #include <lustre_req_layout.h>
61 #include <lustre_fid.h>
62 #include "fid_internal.h"
65 /* Assigns client to sequence controller node. */
66 int seq_server_set_cli(struct lu_server_seq *seq,
67 struct lu_client_seq *cli,
68 const struct lu_env *env)
74 * Ask client for new range, assign that range to ->seq_space and write
75 * seq state to backing store should be atomic.
80 CDEBUG(D_INFO, "%s: Detached sequence client %s\n",
81 seq->lss_name, cli->lcs_name);
86 if (seq->lss_cli != NULL) {
87 CERROR("%s: Sequence controller is already "
88 "assigned\n", seq->lss_name);
89 GOTO(out_up, rc = -EINVAL);
92 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
93 seq->lss_name, cli->lcs_name);
101 EXPORT_SYMBOL(seq_server_set_cli);
104 * On controller node, allocate new super sequence for regular sequence server.
106 static int __seq_server_alloc_super(struct lu_server_seq *seq,
108 struct lu_range *out,
109 const struct lu_env *env)
111 struct lu_range *space = &seq->lss_space;
115 LASSERT(range_is_sane(space));
118 CDEBUG(D_INFO, "%s: Input seq range: "
119 DRANGE"\n", seq->lss_name, PRANGE(in));
121 if (in->lr_end > space->lr_start)
122 space->lr_start = in->lr_end;
125 CDEBUG(D_INFO, "%s: Recovered space: "DRANGE"\n",
126 seq->lss_name, PRANGE(space));
128 if (range_space(space) < seq->lss_width) {
129 CWARN("%s: Sequences space to be exhausted soon. "
130 "Only "LPU64" sequences left\n", seq->lss_name,
133 space->lr_start = space->lr_end;
134 } else if (range_is_exhausted(space)) {
135 CERROR("%s: Sequences space is exhausted\n",
139 range_alloc(out, space, seq->lss_width);
143 rc = seq_store_write(seq, env);
145 CERROR("%s: Can't write space data, rc %d\n",
150 CDEBUG(D_INFO, "%s: Allocated super-sequence "
151 DRANGE"\n", seq->lss_name, PRANGE(out));
156 int seq_server_alloc_super(struct lu_server_seq *seq,
158 struct lu_range *out,
159 const struct lu_env *env)
165 rc = __seq_server_alloc_super(seq, in, out, env);
171 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
173 struct lu_range *out,
174 const struct lu_env *env)
176 struct lu_range *space = &seq->lss_space;
180 LASSERT(range_is_sane(space));
183 * This is recovery case. Adjust super range if input range looks like
184 * it is allocated from new super.
187 CDEBUG(D_INFO, "%s: Input seq range: "
188 DRANGE"\n", seq->lss_name, PRANGE(in));
190 if (range_is_exhausted(space)) {
192 * Server cannot send empty range to client, this is why
193 * we check here that range from client is "newer" than
196 LASSERT(in->lr_end > space->lr_start);
199 * Start is set to end of last allocated, because it
200 * *is* already allocated so we take that into account
201 * and do not use for other allocations.
203 space->lr_start = in->lr_end;
206 * End is set to in->lr_start + super sequence
207 * allocation unit. That is because in->lr_start is
208 * first seq in new allocated range from controller
211 space->lr_end = in->lr_start + LUSTRE_SEQ_SUPER_WIDTH;
214 CERROR("%s: No sequence controller "
215 "is attached.\n", seq->lss_name);
220 * Let controller know that this is recovery and last
221 * obtained range from it was @space.
223 rc = seq_client_replay_super(seq->lss_cli, space, env);
225 CERROR("%s: Can't replay super-sequence, "
226 "rc %d\n", seq->lss_name, rc);
231 * Update super start by end from client's range. Super
232 * end should not be changed if range was not exhausted.
234 if (in->lr_end > space->lr_start)
235 space->lr_start = in->lr_end;
240 CDEBUG(D_INFO, "%s: Recovered space: "DRANGE"\n",
241 seq->lss_name, PRANGE(space));
244 * XXX: Avoid cascading RPCs using kind of async preallocation
245 * when meta-sequence is close to exhausting.
247 if (range_is_exhausted(space)) {
249 CERROR("%s: No sequence controller "
250 "is attached.\n", seq->lss_name);
254 rc = seq_client_alloc_super(seq->lss_cli, env);
256 CERROR("%s: Can't allocate super-sequence, "
257 "rc %d\n", seq->lss_name, rc);
261 /* Saving new range to allocation space. */
262 *space = seq->lss_cli->lcs_space;
263 LASSERT(range_is_sane(space));
266 range_alloc(out, space, seq->lss_width);
269 rc = seq_store_write(seq, env);
271 CERROR("%s: Can't write space data, rc %d\n",
276 CDEBUG(D_INFO, "%s: Allocated meta-sequence "
277 DRANGE"\n", seq->lss_name, PRANGE(out));
283 int seq_server_alloc_meta(struct lu_server_seq *seq,
285 struct lu_range *out,
286 const struct lu_env *env)
292 rc = __seq_server_alloc_meta(seq, in, out, env);
297 EXPORT_SYMBOL(seq_server_alloc_meta);
299 static int seq_server_handle(struct lu_site *site,
300 const struct lu_env *env,
301 __u32 opc, struct lu_range *in,
302 struct lu_range *out)
305 struct md_site *mite;
308 mite = lu_site2md(site);
311 if (!mite->ms_server_seq) {
312 CERROR("Sequence server is not "
316 rc = seq_server_alloc_meta(mite->ms_server_seq,
319 case SEQ_ALLOC_SUPER:
320 if (!mite->ms_control_seq) {
321 CERROR("Sequence controller is not "
325 rc = seq_server_alloc_super(mite->ms_control_seq,
336 static int seq_req_handle(struct ptlrpc_request *req,
337 const struct lu_env *env,
338 struct seq_thread_info *info)
340 struct lu_range *out, *in = NULL;
341 struct lu_site *site;
346 site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
347 LASSERT(site != NULL);
349 rc = req_capsule_server_pack(info->sti_pill);
351 RETURN(err_serious(rc));
353 opc = req_capsule_client_get(info->sti_pill, &RMF_SEQ_OPC);
355 out = req_capsule_server_get(info->sti_pill, &RMF_SEQ_RANGE);
357 RETURN(err_serious(-EPROTO));
359 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
360 in = req_capsule_client_get(info->sti_pill,
363 LASSERT(!range_is_zero(in) && range_is_sane(in));
366 rc = seq_server_handle(site, env, *opc, in, out);
368 rc = err_serious(-EPROTO);
373 /* context key constructor/destructor: seq_key_init, seq_key_fini */
374 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
376 /* context key: seq_thread_key */
377 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD);
379 static void seq_thread_info_init(struct ptlrpc_request *req,
380 struct seq_thread_info *info)
382 info->sti_pill = &req->rq_pill;
383 /* Init request capsule */
384 req_capsule_init(info->sti_pill, req, RCL_SERVER);
385 req_capsule_set(info->sti_pill, &RQF_SEQ_QUERY);
388 static void seq_thread_info_fini(struct seq_thread_info *info)
390 req_capsule_fini(info->sti_pill);
393 static int seq_handle(struct ptlrpc_request *req)
395 const struct lu_env *env;
396 struct seq_thread_info *info;
399 env = req->rq_svc_thread->t_env;
400 LASSERT(env != NULL);
402 info = lu_context_key_get(&env->le_ctx, &seq_thread_key);
403 LASSERT(info != NULL);
405 seq_thread_info_init(req, info);
406 rc = seq_req_handle(req, env, info);
407 seq_thread_info_fini(info);
413 * Entry point for handling FLD RPCs called from MDT.
415 int seq_query(struct com_thread_info *info)
417 return seq_handle(info->cti_pill->rc_req);
419 EXPORT_SYMBOL(seq_query);
421 static void seq_server_proc_fini(struct lu_server_seq *seq);
424 static int seq_server_proc_init(struct lu_server_seq *seq)
429 seq->lss_proc_dir = lprocfs_register(seq->lss_name,
432 if (IS_ERR(seq->lss_proc_dir)) {
433 rc = PTR_ERR(seq->lss_proc_dir);
437 rc = lprocfs_add_vars(seq->lss_proc_dir,
438 seq_server_proc_list, seq);
440 CERROR("%s: Can't init sequence manager "
441 "proc, rc %d\n", seq->lss_name, rc);
442 GOTO(out_cleanup, rc);
448 seq_server_proc_fini(seq);
452 static void seq_server_proc_fini(struct lu_server_seq *seq)
455 if (seq->lss_proc_dir != NULL) {
456 if (!IS_ERR(seq->lss_proc_dir))
457 lprocfs_remove(&seq->lss_proc_dir);
458 seq->lss_proc_dir = NULL;
463 static int seq_server_proc_init(struct lu_server_seq *seq)
468 static void seq_server_proc_fini(struct lu_server_seq *seq)
474 int seq_server_init(struct lu_server_seq *seq,
475 struct dt_device *dev,
477 enum lu_mgr_type type,
478 const struct lu_env *env)
480 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
483 LASSERT(dev != NULL);
484 LASSERT(prefix != NULL);
487 seq->lss_type = type;
488 range_init(&seq->lss_space);
489 sema_init(&seq->lss_sem, 1);
491 seq->lss_width = is_srv ?
492 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
494 snprintf(seq->lss_name, sizeof(seq->lss_name),
495 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
497 rc = seq_store_init(seq, env, dev);
501 /* Request backing store for saved sequence info. */
502 rc = seq_store_read(seq, env);
503 if (rc == -ENODATA) {
505 /* Nothing is read, init by default value. */
506 seq->lss_space = is_srv ?
507 LUSTRE_SEQ_ZERO_RANGE:
508 LUSTRE_SEQ_SPACE_RANGE;
510 CDEBUG(D_INFO, "%s: No data found "
511 "on store. Initialize space\n",
514 /* Save default controller value to store. */
515 rc = seq_store_write(seq, env);
517 CERROR("%s: Can't write space data, "
518 "rc %d\n", seq->lss_name, rc);
521 CERROR("%s: Can't read space data, rc %d\n",
527 LASSERT(range_is_sane(&seq->lss_space));
529 LASSERT(!range_is_zero(&seq->lss_space) &&
530 range_is_sane(&seq->lss_space));
533 rc = seq_server_proc_init(seq);
540 seq_server_fini(seq, env);
543 EXPORT_SYMBOL(seq_server_init);
545 void seq_server_fini(struct lu_server_seq *seq,
546 const struct lu_env *env)
550 seq_server_proc_fini(seq);
551 seq_store_fini(seq, env);
555 EXPORT_SYMBOL(seq_server_fini);
557 cfs_proc_dir_entry_t *seq_type_proc_dir = NULL;
559 static struct lu_local_obj_desc llod_seq_srv = {
560 .llod_name = LUSTRE_SEQ_SRV_NAME,
561 .llod_oid = FID_SEQ_SRV_OID,
565 static struct lu_local_obj_desc llod_seq_ctl = {
566 .llod_name = LUSTRE_SEQ_CTL_NAME,
567 .llod_oid = FID_SEQ_CTL_OID,
571 static int __init fid_mod_init(void)
573 seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
576 if (IS_ERR(seq_type_proc_dir))
577 return PTR_ERR(seq_type_proc_dir);
579 llo_local_obj_register(&llod_seq_srv);
580 llo_local_obj_register(&llod_seq_ctl);
582 LU_CONTEXT_KEY_INIT(&seq_thread_key);
583 lu_context_key_register(&seq_thread_key);
587 static void __exit fid_mod_exit(void)
589 llo_local_obj_unregister(&llod_seq_srv);
590 llo_local_obj_unregister(&llod_seq_ctl);
592 lu_context_key_degister(&seq_thread_key);
593 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
594 lprocfs_remove(&seq_type_proc_dir);
595 seq_type_proc_dir = NULL;
599 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
600 MODULE_DESCRIPTION("Lustre FID Module");
601 MODULE_LICENSE("GPL");
603 cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);