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.
32 * Copyright (c) 2011, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/fid/fid_handler.c
40 * Lustre Sequence Manager
42 * Author: Yury Umanets <umka@clusterfs.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_FID
51 # include <libcfs/libcfs.h>
52 # include <linux/module.h>
53 #else /* __KERNEL__ */
54 # include <liblustre.h>
58 #include <obd_class.h>
59 #include <lu_target.h>
60 #include <dt_object.h>
61 #include <md_object.h>
62 #include <obd_support.h>
63 #include <lustre_req_layout.h>
64 #include <lustre_fid.h>
65 #include "fid_internal.h"
68 /* Assigns client to sequence controller node. */
69 int seq_server_set_cli(struct lu_server_seq *seq,
70 struct lu_client_seq *cli,
71 const struct lu_env *env)
77 * Ask client for new range, assign that range to ->seq_space and write
78 * seq state to backing store should be atomic.
80 cfs_mutex_lock(&seq->lss_mutex);
83 CDEBUG(D_INFO, "%s: Detached sequence client %s\n",
84 seq->lss_name, cli->lcs_name);
89 if (seq->lss_cli != NULL) {
90 CERROR("%s: Sequence controller is already "
91 "assigned\n", seq->lss_name);
92 GOTO(out_up, rc = -EINVAL);
95 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
96 seq->lss_name, cli->lcs_name);
99 cli->lcs_space.lsr_index = seq->lss_site->ms_node_id;
102 cfs_mutex_unlock(&seq->lss_mutex);
105 EXPORT_SYMBOL(seq_server_set_cli);
107 * allocate \a w units of sequence from range \a from.
109 static inline void range_alloc(struct lu_seq_range *to,
110 struct lu_seq_range *from,
113 width = min(range_space(from), width);
114 to->lsr_start = from->lsr_start;
115 to->lsr_end = from->lsr_start + width;
116 from->lsr_start += width;
120 * On controller node, allocate new super sequence for regular sequence server.
121 * As this super sequence controller, this node suppose to maintain fld
123 * \a out range always has currect mds node number of requester.
126 static int __seq_server_alloc_super(struct lu_server_seq *seq,
127 struct lu_seq_range *out,
128 const struct lu_env *env)
130 struct lu_seq_range *space = &seq->lss_space;
134 LASSERT(range_is_sane(space));
136 if (range_is_exhausted(space)) {
137 CERROR("%s: Sequences space is exhausted\n",
141 range_alloc(out, space, seq->lss_width);
144 rc = seq_store_update(env, seq, out, 1 /* sync */);
146 CDEBUG(D_INFO, "%s: super-sequence allocation rc = %d "
147 DRANGE"\n", seq->lss_name, rc, PRANGE(out));
152 int seq_server_alloc_super(struct lu_server_seq *seq,
153 struct lu_seq_range *out,
154 const struct lu_env *env)
159 cfs_mutex_lock(&seq->lss_mutex);
160 rc = __seq_server_alloc_super(seq, out, env);
161 cfs_mutex_unlock(&seq->lss_mutex);
166 static int __seq_set_init(const struct lu_env *env,
167 struct lu_server_seq *seq)
169 struct lu_seq_range *space = &seq->lss_space;
172 range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
173 range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
175 rc = seq_store_update(env, seq, NULL, 1);
181 * This function implements new seq allocation algorithm using async
182 * updates to seq file on disk. ref bug 18857 for details.
183 * there are four variable to keep track of this process
185 * lss_space; - available lss_space
186 * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
187 * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
190 * when lss_lowater_set reaches the end it is replaced with hiwater one and
191 * a write operation is initiated to allocate new hiwater range.
192 * if last seq write opearion is still not commited, current operation is
193 * flaged as sync write op.
195 static int range_alloc_set(const struct lu_env *env,
196 struct lu_seq_range *out,
197 struct lu_server_seq *seq)
199 struct lu_seq_range *space = &seq->lss_space;
200 struct lu_seq_range *loset = &seq->lss_lowater_set;
201 struct lu_seq_range *hiset = &seq->lss_hiwater_set;
204 if (range_is_zero(loset))
205 __seq_set_init(env, seq);
207 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
208 loset->lsr_start = loset->lsr_end;
210 if (range_is_exhausted(loset)) {
211 /* reached high water mark. */
212 struct lu_device *dev = seq->lss_site->ms_lu.ls_top_dev;
213 int obd_num_clients = dev->ld_obd->obd_num_exports;
216 /* calculate new seq width based on number of clients */
217 set_sz = max(seq->lss_set_width,
218 obd_num_clients * seq->lss_width);
219 set_sz = min(range_space(space), set_sz);
221 /* Switch to hiwater range now */
223 /* allocate new hiwater range */
224 range_alloc(hiset, space, set_sz);
226 /* update ondisk seq with new *space */
227 rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
230 LASSERTF(!range_is_exhausted(loset) || range_is_sane(loset),
231 DRANGE"\n", PRANGE(loset));
234 range_alloc(out, loset, seq->lss_width);
239 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
240 struct lu_seq_range *out,
241 const struct lu_env *env)
243 struct lu_seq_range *space = &seq->lss_space;
248 LASSERT(range_is_sane(space));
250 /* Check if available space ends and allocate new super seq */
251 if (range_is_exhausted(space)) {
253 CERROR("%s: No sequence controller is attached.\n",
258 rc = seq_client_alloc_super(seq->lss_cli, env);
260 CERROR("%s: Can't allocate super-sequence, rc %d\n",
265 /* Saving new range to allocation space. */
266 *space = seq->lss_cli->lcs_space;
267 LASSERT(range_is_sane(space));
270 rc = range_alloc_set(env, out, seq);
272 CDEBUG(D_INFO, "%s: Allocated meta-sequence "
273 DRANGE"\n", seq->lss_name, PRANGE(out));
279 int seq_server_alloc_meta(struct lu_server_seq *seq,
280 struct lu_seq_range *out,
281 const struct lu_env *env)
286 cfs_mutex_lock(&seq->lss_mutex);
287 rc = __seq_server_alloc_meta(seq, out, env);
288 cfs_mutex_unlock(&seq->lss_mutex);
292 EXPORT_SYMBOL(seq_server_alloc_meta);
294 static int seq_server_handle(struct lu_site *site,
295 const struct lu_env *env,
296 __u32 opc, struct lu_seq_range *out)
299 struct md_site *mite;
302 mite = lu_site2md(site);
305 if (!mite->ms_server_seq) {
306 CERROR("Sequence server is not "
310 rc = seq_server_alloc_meta(mite->ms_server_seq, out, env);
312 case SEQ_ALLOC_SUPER:
313 if (!mite->ms_control_seq) {
314 CERROR("Sequence controller is not "
318 rc = seq_server_alloc_super(mite->ms_control_seq, out, env);
328 static int seq_req_handle(struct ptlrpc_request *req,
329 const struct lu_env *env,
330 struct seq_thread_info *info)
332 struct lu_seq_range *out, *tmp;
333 struct lu_site *site;
338 LASSERT(!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY));
339 site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
340 LASSERT(site != NULL);
342 rc = req_capsule_server_pack(info->sti_pill);
344 RETURN(err_serious(rc));
346 opc = req_capsule_client_get(info->sti_pill, &RMF_SEQ_OPC);
348 out = req_capsule_server_get(info->sti_pill, &RMF_SEQ_RANGE);
350 RETURN(err_serious(-EPROTO));
352 tmp = req_capsule_client_get(info->sti_pill, &RMF_SEQ_RANGE);
354 /* seq client passed mdt id, we need to pass that using out
357 out->lsr_index = tmp->lsr_index;
358 out->lsr_flags = tmp->lsr_flags;
359 rc = seq_server_handle(site, env, *opc, out);
361 rc = err_serious(-EPROTO);
366 /* context key constructor/destructor: seq_key_init, seq_key_fini */
367 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
369 /* context key: seq_thread_key */
370 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD);
372 static void seq_thread_info_init(struct ptlrpc_request *req,
373 struct seq_thread_info *info)
375 info->sti_pill = &req->rq_pill;
376 /* Init request capsule */
377 req_capsule_init(info->sti_pill, req, RCL_SERVER);
378 req_capsule_set(info->sti_pill, &RQF_SEQ_QUERY);
381 static void seq_thread_info_fini(struct seq_thread_info *info)
383 req_capsule_fini(info->sti_pill);
386 static int seq_handle(struct ptlrpc_request *req)
388 const struct lu_env *env;
389 struct seq_thread_info *info;
392 env = req->rq_svc_thread->t_env;
393 LASSERT(env != NULL);
395 info = lu_context_key_get(&env->le_ctx, &seq_thread_key);
396 LASSERT(info != NULL);
398 seq_thread_info_init(req, info);
399 rc = seq_req_handle(req, env, info);
400 /* XXX: we don't need replay but MDT assign transno in any case,
401 * remove it manually before reply*/
402 lustre_msg_set_transno(req->rq_repmsg, 0);
403 seq_thread_info_fini(info);
409 * Entry point for handling FLD RPCs called from MDT.
411 int seq_query(struct com_thread_info *info)
413 return seq_handle(info->cti_pill->rc_req);
415 EXPORT_SYMBOL(seq_query);
417 static void seq_server_proc_fini(struct lu_server_seq *seq);
420 static int seq_server_proc_init(struct lu_server_seq *seq)
425 seq->lss_proc_dir = lprocfs_register(seq->lss_name,
428 if (IS_ERR(seq->lss_proc_dir)) {
429 rc = PTR_ERR(seq->lss_proc_dir);
433 rc = lprocfs_add_vars(seq->lss_proc_dir,
434 seq_server_proc_list, seq);
436 CERROR("%s: Can't init sequence manager "
437 "proc, rc %d\n", seq->lss_name, rc);
438 GOTO(out_cleanup, rc);
444 seq_server_proc_fini(seq);
448 static void seq_server_proc_fini(struct lu_server_seq *seq)
451 if (seq->lss_proc_dir != NULL) {
452 if (!IS_ERR(seq->lss_proc_dir))
453 lprocfs_remove(&seq->lss_proc_dir);
454 seq->lss_proc_dir = NULL;
459 static int seq_server_proc_init(struct lu_server_seq *seq)
464 static void seq_server_proc_fini(struct lu_server_seq *seq)
471 int seq_server_init(struct lu_server_seq *seq,
472 struct dt_device *dev,
474 enum lu_mgr_type type,
476 const struct lu_env *env)
478 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
481 LASSERT(dev != NULL);
482 LASSERT(prefix != NULL);
485 seq->lss_type = type;
487 range_init(&seq->lss_space);
489 range_init(&seq->lss_lowater_set);
490 range_init(&seq->lss_hiwater_set);
491 seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
493 cfs_mutex_init(&seq->lss_mutex);
495 seq->lss_width = is_srv ?
496 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
498 snprintf(seq->lss_name, sizeof(seq->lss_name),
499 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
501 rc = seq_store_init(seq, env, dev);
504 /* Request backing store for saved sequence info. */
505 rc = seq_store_read(seq, env);
506 if (rc == -ENODATA) {
508 /* Nothing is read, init by default value. */
509 seq->lss_space = is_srv ?
510 LUSTRE_SEQ_ZERO_RANGE:
511 LUSTRE_SEQ_SPACE_RANGE;
513 seq->lss_space.lsr_index = ms->ms_node_id;
514 CDEBUG(D_INFO, "%s: No data found "
515 "on store. Initialize space\n",
518 rc = seq_store_update(env, seq, NULL, 0);
520 CERROR("%s: Can't write space data, "
521 "rc %d\n", seq->lss_name, rc);
524 CERROR("%s: Can't read space data, rc %d\n",
530 LASSERT(range_is_sane(&seq->lss_space));
532 LASSERT(!range_is_zero(&seq->lss_space) &&
533 range_is_sane(&seq->lss_space));
536 rc = seq_server_proc_init(seq);
543 seq_server_fini(seq, env);
546 EXPORT_SYMBOL(seq_server_init);
548 void seq_server_fini(struct lu_server_seq *seq,
549 const struct lu_env *env)
553 seq_server_proc_fini(seq);
554 seq_store_fini(seq, env);
558 EXPORT_SYMBOL(seq_server_fini);
560 cfs_proc_dir_entry_t *seq_type_proc_dir = NULL;
562 static struct lu_local_obj_desc llod_seq_srv = {
563 .llod_name = LUSTRE_SEQ_SRV_NAME,
564 .llod_oid = FID_SEQ_SRV_OID,
568 static struct lu_local_obj_desc llod_seq_ctl = {
569 .llod_name = LUSTRE_SEQ_CTL_NAME,
570 .llod_oid = FID_SEQ_CTL_OID,
574 static int __init fid_mod_init(void)
576 seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
579 if (IS_ERR(seq_type_proc_dir))
580 return PTR_ERR(seq_type_proc_dir);
582 llo_local_obj_register(&llod_seq_srv);
583 llo_local_obj_register(&llod_seq_ctl);
585 LU_CONTEXT_KEY_INIT(&seq_thread_key);
586 lu_context_key_register(&seq_thread_key);
590 static void __exit fid_mod_exit(void)
592 llo_local_obj_unregister(&llod_seq_srv);
593 llo_local_obj_unregister(&llod_seq_ctl);
595 lu_context_key_degister(&seq_thread_key);
596 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
597 lprocfs_remove(&seq_type_proc_dir);
598 seq_type_proc_dir = NULL;
602 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
603 MODULE_DESCRIPTION("Lustre FID Module");
604 MODULE_LICENSE("GPL");
606 cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);