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/fid/fid_handler.c
38 * Lustre Sequence Manager
40 * Author: Yury Umanets <umka@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_FID
45 #include <libcfs/libcfs.h>
46 #include <linux/module.h>
48 #include <obd_class.h>
49 #include <dt_object.h>
50 #include <obd_support.h>
51 #include <lustre_req_layout.h>
52 #include <lustre_fid.h>
53 #include "fid_internal.h"
55 static void seq_server_proc_fini(struct lu_server_seq *seq);
57 /* Assigns client to sequence controller node. */
58 int seq_server_set_cli(const struct lu_env *env, struct lu_server_seq *seq,
59 struct lu_client_seq *cli)
65 * Ask client for new range, assign that range to ->seq_space and write
66 * seq state to backing store should be atomic.
68 mutex_lock(&seq->lss_mutex);
71 CDEBUG(D_INFO, "%s: Detached sequence client\n", seq->lss_name);
76 if (seq->lss_cli != NULL) {
77 CDEBUG(D_HA, "%s: Sequence controller is already "
78 "assigned\n", seq->lss_name);
79 GOTO(out_up, rc = -EEXIST);
82 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
83 seq->lss_name, cli->lcs_name);
86 cli->lcs_space.lsr_index = seq->lss_site->ss_node_id;
89 mutex_unlock(&seq->lss_mutex);
92 EXPORT_SYMBOL(seq_server_set_cli);
94 * allocate \a w units of sequence from range \a from.
96 static inline void range_alloc(struct lu_seq_range *to,
97 struct lu_seq_range *from,
100 width = min(range_space(from), width);
101 to->lsr_start = from->lsr_start;
102 to->lsr_end = from->lsr_start + width;
103 from->lsr_start += width;
107 * On controller node, allocate new super sequence for regular sequence server.
108 * As this super sequence controller, this node suppose to maintain fld
110 * \a out range always has currect mds node number of requester.
113 static int __seq_server_alloc_super(struct lu_server_seq *seq,
114 struct lu_seq_range *out,
115 const struct lu_env *env)
117 struct lu_seq_range *space = &seq->lss_space;
121 LASSERT(range_is_sane(space));
123 if (range_is_exhausted(space)) {
124 CERROR("%s: Sequences space is exhausted\n",
128 range_alloc(out, space, seq->lss_width);
131 rc = seq_store_update(env, seq, out, 1 /* sync */);
133 LCONSOLE_INFO("%s: super-sequence allocation rc = %d " DRANGE"\n",
134 seq->lss_name, rc, PRANGE(out));
139 int seq_server_alloc_super(struct lu_server_seq *seq,
140 struct lu_seq_range *out,
141 const struct lu_env *env)
146 mutex_lock(&seq->lss_mutex);
147 rc = __seq_server_alloc_super(seq, out, env);
148 mutex_unlock(&seq->lss_mutex);
153 int seq_server_alloc_spec(struct lu_server_seq *seq,
154 struct lu_seq_range *spec,
155 const struct lu_env *env)
157 struct lu_seq_range *space = &seq->lss_space;
162 * In some cases (like recovery after a disaster)
163 * we may need to allocate sequences manually
164 * Notice some sequences can be lost if requested
165 * range doesn't start at the beginning of current
166 * free space. Also notice it's not possible now
167 * to allocate sequences out of natural order.
169 if (spec->lsr_start >= spec->lsr_end)
171 if (spec->lsr_flags != LU_SEQ_RANGE_MDT &&
172 spec->lsr_flags != LU_SEQ_RANGE_OST)
175 mutex_lock(&seq->lss_mutex);
176 if (spec->lsr_start >= space->lsr_start) {
177 space->lsr_start = spec->lsr_end;
178 rc = seq_store_update(env, seq, spec, 1 /* sync */);
180 LCONSOLE_INFO("%s: "DRANGE" sequences allocated: rc = %d \n",
181 seq->lss_name, PRANGE(spec), rc);
183 mutex_unlock(&seq->lss_mutex);
188 static int __seq_set_init(const struct lu_env *env,
189 struct lu_server_seq *seq)
191 struct lu_seq_range *space = &seq->lss_space;
194 range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
195 range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
197 rc = seq_store_update(env, seq, NULL, 1);
203 * This function implements new seq allocation algorithm using async
204 * updates to seq file on disk. ref bug 18857 for details.
205 * there are four variable to keep track of this process
207 * lss_space; - available lss_space
208 * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
209 * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
212 * when lss_lowater_set reaches the end it is replaced with hiwater one and
213 * a write operation is initiated to allocate new hiwater range.
214 * if last seq write opearion is still not commited, current operation is
215 * flaged as sync write op.
217 static int range_alloc_set(const struct lu_env *env,
218 struct lu_seq_range *out,
219 struct lu_server_seq *seq)
221 struct lu_seq_range *space = &seq->lss_space;
222 struct lu_seq_range *loset = &seq->lss_lowater_set;
223 struct lu_seq_range *hiset = &seq->lss_hiwater_set;
226 if (range_is_zero(loset))
227 __seq_set_init(env, seq);
229 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
230 loset->lsr_start = loset->lsr_end;
232 if (range_is_exhausted(loset)) {
233 /* reached high water mark. */
234 struct lu_device *dev = seq->lss_site->ss_lu->ls_top_dev;
235 int obd_num_clients = dev->ld_obd->obd_num_exports;
238 /* calculate new seq width based on number of clients */
239 set_sz = max(seq->lss_set_width,
240 obd_num_clients * seq->lss_width);
241 set_sz = min(range_space(space), set_sz);
243 /* Switch to hiwater range now */
245 /* allocate new hiwater range */
246 range_alloc(hiset, space, set_sz);
248 /* update ondisk seq with new *space */
249 rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
252 LASSERTF(!range_is_exhausted(loset) || range_is_sane(loset),
253 DRANGE"\n", PRANGE(loset));
256 range_alloc(out, loset, seq->lss_width);
261 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
262 struct lu_seq_range *out,
263 const struct lu_env *env)
265 struct lu_seq_range *space = &seq->lss_space;
270 LASSERT(range_is_sane(space));
272 /* Check if available space ends and allocate new super seq */
273 if (range_is_exhausted(space)) {
275 CERROR("%s: No sequence controller is attached.\n",
280 rc = seq_client_alloc_super(seq->lss_cli, env);
282 CERROR("%s: Can't allocate super-sequence, rc %d\n",
287 /* Saving new range to allocation space. */
288 *space = seq->lss_cli->lcs_space;
289 LASSERT(range_is_sane(space));
290 if (seq->lss_cli->lcs_srv == NULL) {
291 struct lu_server_fld *fld;
293 /* Insert it to the local FLDB */
294 fld = seq->lss_site->ss_server_fld;
295 mutex_lock(&fld->lsf_lock);
296 rc = fld_insert_entry(env, fld, space);
297 mutex_unlock(&fld->lsf_lock);
301 rc = range_alloc_set(env, out, seq);
303 CERROR("%s: Allocated meta-sequence failed: rc = %d\n",
308 CDEBUG(D_INFO, "%s: Allocated meta-sequence " DRANGE"\n",
309 seq->lss_name, PRANGE(out));
314 int seq_server_alloc_meta(struct lu_server_seq *seq,
315 struct lu_seq_range *out,
316 const struct lu_env *env)
321 mutex_lock(&seq->lss_mutex);
322 rc = __seq_server_alloc_meta(seq, out, env);
323 mutex_unlock(&seq->lss_mutex);
327 EXPORT_SYMBOL(seq_server_alloc_meta);
329 static int seq_server_handle(struct lu_site *site,
330 const struct lu_env *env,
331 __u32 opc, struct lu_seq_range *out)
334 struct seq_server_site *ss_site;
337 ss_site = lu_site2seq(site);
341 if (!ss_site->ss_server_seq) {
342 CERROR("Sequence server is not "
346 rc = seq_server_alloc_meta(ss_site->ss_server_seq, out, env);
348 case SEQ_ALLOC_SUPER:
349 if (!ss_site->ss_control_seq) {
350 CERROR("Sequence controller is not "
354 rc = seq_server_alloc_super(ss_site->ss_control_seq, out, env);
364 static int seq_handler(struct tgt_session_info *tsi)
366 struct lu_seq_range *out, *tmp;
367 struct lu_site *site;
373 LASSERT(!(lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY));
374 site = tsi->tsi_exp->exp_obd->obd_lu_dev->ld_site;
375 LASSERT(site != NULL);
377 opc = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_OPC);
379 out = req_capsule_server_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
381 RETURN(err_serious(-EPROTO));
383 tmp = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
385 /* seq client passed mdt id, we need to pass that using out
388 out->lsr_index = tmp->lsr_index;
389 out->lsr_flags = tmp->lsr_flags;
390 rc = seq_server_handle(site, tsi->tsi_env, *opc, out);
392 rc = err_serious(-EPROTO);
398 struct tgt_handler seq_handlers[] = {
399 TGT_SEQ_HDL(HABEO_REFERO, SEQ_QUERY, seq_handler),
401 EXPORT_SYMBOL(seq_handlers);
403 /* context key constructor/destructor: seq_key_init, seq_key_fini */
404 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
406 /* context key: seq_thread_key */
407 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD | LCT_DT_THREAD);
409 extern const struct file_operations seq_fld_proc_seq_fops;
411 static int seq_server_proc_init(struct lu_server_seq *seq)
417 seq->lss_proc_dir = lprocfs_seq_register(seq->lss_name,
420 if (IS_ERR(seq->lss_proc_dir)) {
421 rc = PTR_ERR(seq->lss_proc_dir);
425 rc = lprocfs_seq_add_vars(seq->lss_proc_dir,
426 seq_server_proc_list, seq);
428 CERROR("%s: Can't init sequence manager "
429 "proc, rc %d\n", seq->lss_name, rc);
430 GOTO(out_cleanup, rc);
433 if (seq->lss_type == LUSTRE_SEQ_CONTROLLER) {
434 rc = lprocfs_seq_create(seq->lss_proc_dir, "fldb", 0644,
435 &seq_fld_proc_seq_fops, seq);
437 CERROR("%s: Can't create fldb for sequence manager "
438 "proc: rc = %d\n", seq->lss_name, rc);
439 GOTO(out_cleanup, rc);
446 seq_server_proc_fini(seq);
453 static void seq_server_proc_fini(struct lu_server_seq *seq)
457 if (seq->lss_proc_dir != NULL) {
458 if (!IS_ERR(seq->lss_proc_dir))
459 lprocfs_remove(&seq->lss_proc_dir);
460 seq->lss_proc_dir = NULL;
466 int seq_server_init(const struct lu_env *env,
467 struct lu_server_seq *seq,
468 struct dt_device *dev,
470 enum lu_mgr_type type,
471 struct seq_server_site *ss)
473 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
476 LASSERT(dev != NULL);
477 LASSERT(prefix != NULL);
479 LASSERT(ss->ss_lu != NULL);
481 /* A compile-time check for FIDs that used to be in lustre_idl.h
482 * but is moved here to remove CLASSERT/LASSERT in that header.
483 * Check all lu_fid fields are converted in fid_cpu_to_le() and friends
484 * and that there is no padding added by compiler to the struct. */
488 CLASSERT(sizeof(tst) == sizeof(tst.f_seq) +
489 sizeof(tst.f_oid) + sizeof(tst.f_ver));
493 seq->lss_type = type;
495 range_init(&seq->lss_space);
497 range_init(&seq->lss_lowater_set);
498 range_init(&seq->lss_hiwater_set);
499 seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
501 mutex_init(&seq->lss_mutex);
503 seq->lss_width = is_srv ?
504 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
506 snprintf(seq->lss_name, sizeof(seq->lss_name),
507 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
509 rc = seq_store_init(seq, env, dev);
512 /* Request backing store for saved sequence info. */
513 rc = seq_store_read(seq, env);
514 if (rc == -ENODATA) {
516 /* Nothing is read, init by default value. */
517 seq->lss_space = is_srv ?
518 LUSTRE_SEQ_ZERO_RANGE:
519 LUSTRE_SEQ_SPACE_RANGE;
521 seq->lss_space.lsr_index = ss->ss_node_id;
522 LCONSOLE_INFO("%s: No data found "
523 "on store. Initialize space\n",
526 rc = seq_store_update(env, seq, NULL, 0);
528 CERROR("%s: Can't write space data, "
529 "rc %d\n", seq->lss_name, rc);
532 CERROR("%s: Can't read space data, rc %d\n",
538 LASSERT(range_is_sane(&seq->lss_space));
540 LASSERT(!range_is_zero(&seq->lss_space) &&
541 range_is_sane(&seq->lss_space));
544 rc = seq_server_proc_init(seq);
551 seq_server_fini(seq, env);
554 EXPORT_SYMBOL(seq_server_init);
556 void seq_server_fini(struct lu_server_seq *seq,
557 const struct lu_env *env)
561 seq_server_proc_fini(seq);
562 seq_store_fini(seq, env);
566 EXPORT_SYMBOL(seq_server_fini);
568 int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss)
573 if (ss->ss_server_seq) {
574 seq_server_fini(ss->ss_server_seq, env);
575 OBD_FREE_PTR(ss->ss_server_seq);
576 ss->ss_server_seq = NULL;
579 if (ss->ss_control_seq) {
580 seq_server_fini(ss->ss_control_seq, env);
581 OBD_FREE_PTR(ss->ss_control_seq);
582 ss->ss_control_seq = NULL;
585 if (ss->ss_client_seq) {
586 seq_client_fini(ss->ss_client_seq);
587 OBD_FREE_PTR(ss->ss_client_seq);
588 ss->ss_client_seq = NULL;
593 EXPORT_SYMBOL(seq_site_fini);
595 int fid_server_mod_init(void)
597 LU_CONTEXT_KEY_INIT(&seq_thread_key);
598 return lu_context_key_register(&seq_thread_key);
601 void fid_server_mod_exit(void)
603 lu_context_key_degister(&seq_thread_key);