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 <md_object.h>
51 #include <obd_support.h>
52 #include <lustre_req_layout.h>
53 #include <lustre_fid.h>
54 #include "fid_internal.h"
56 static void seq_server_proc_fini(struct lu_server_seq *seq);
58 /* Assigns client to sequence controller node. */
59 int seq_server_set_cli(const struct lu_env *env, struct lu_server_seq *seq,
60 struct lu_client_seq *cli)
66 * Ask client for new range, assign that range to ->seq_space and write
67 * seq state to backing store should be atomic.
69 mutex_lock(&seq->lss_mutex);
72 CDEBUG(D_INFO, "%s: Detached sequence client\n", seq->lss_name);
77 if (seq->lss_cli != NULL) {
78 CDEBUG(D_HA, "%s: Sequence controller is already "
79 "assigned\n", seq->lss_name);
80 GOTO(out_up, rc = -EEXIST);
83 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
84 seq->lss_name, cli->lcs_name);
87 cli->lcs_space.lsr_index = seq->lss_site->ss_node_id;
90 mutex_unlock(&seq->lss_mutex);
93 EXPORT_SYMBOL(seq_server_set_cli);
95 * allocate \a w units of sequence from range \a from.
97 static inline void range_alloc(struct lu_seq_range *to,
98 struct lu_seq_range *from,
101 width = min(range_space(from), width);
102 to->lsr_start = from->lsr_start;
103 to->lsr_end = from->lsr_start + width;
104 from->lsr_start += width;
108 * On controller node, allocate new super sequence for regular sequence server.
109 * As this super sequence controller, this node suppose to maintain fld
111 * \a out range always has currect mds node number of requester.
114 static int __seq_server_alloc_super(struct lu_server_seq *seq,
115 struct lu_seq_range *out,
116 const struct lu_env *env)
118 struct lu_seq_range *space = &seq->lss_space;
122 LASSERT(range_is_sane(space));
124 if (range_is_exhausted(space)) {
125 CERROR("%s: Sequences space is exhausted\n",
129 range_alloc(out, space, seq->lss_width);
132 rc = seq_store_update(env, seq, out, 1 /* sync */);
134 LCONSOLE_INFO("%s: super-sequence allocation rc = %d " DRANGE"\n",
135 seq->lss_name, rc, PRANGE(out));
140 int seq_server_alloc_super(struct lu_server_seq *seq,
141 struct lu_seq_range *out,
142 const struct lu_env *env)
147 mutex_lock(&seq->lss_mutex);
148 rc = __seq_server_alloc_super(seq, out, env);
149 mutex_unlock(&seq->lss_mutex);
154 int seq_server_alloc_spec(struct lu_server_seq *seq,
155 struct lu_seq_range *spec,
156 const struct lu_env *env)
158 struct lu_seq_range *space = &seq->lss_space;
163 * In some cases (like recovery after a disaster)
164 * we may need to allocate sequences manually
165 * Notice some sequences can be lost if requested
166 * range doesn't start at the beginning of current
167 * free space. Also notice it's not possible now
168 * to allocate sequences out of natural order.
170 if (spec->lsr_start >= spec->lsr_end)
172 if (spec->lsr_flags != LU_SEQ_RANGE_MDT &&
173 spec->lsr_flags != LU_SEQ_RANGE_OST)
176 mutex_lock(&seq->lss_mutex);
177 if (spec->lsr_start >= space->lsr_start) {
178 space->lsr_start = spec->lsr_end;
179 rc = seq_store_update(env, seq, spec, 1 /* sync */);
181 LCONSOLE_INFO("%s: "DRANGE" sequences allocated: rc = %d \n",
182 seq->lss_name, PRANGE(spec), rc);
184 mutex_unlock(&seq->lss_mutex);
189 static int __seq_set_init(const struct lu_env *env,
190 struct lu_server_seq *seq)
192 struct lu_seq_range *space = &seq->lss_space;
195 range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
196 range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
198 rc = seq_store_update(env, seq, NULL, 1);
204 * This function implements new seq allocation algorithm using async
205 * updates to seq file on disk. ref bug 18857 for details.
206 * there are four variable to keep track of this process
208 * lss_space; - available lss_space
209 * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
210 * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
213 * when lss_lowater_set reaches the end it is replaced with hiwater one and
214 * a write operation is initiated to allocate new hiwater range.
215 * if last seq write opearion is still not commited, current operation is
216 * flaged as sync write op.
218 static int range_alloc_set(const struct lu_env *env,
219 struct lu_seq_range *out,
220 struct lu_server_seq *seq)
222 struct lu_seq_range *space = &seq->lss_space;
223 struct lu_seq_range *loset = &seq->lss_lowater_set;
224 struct lu_seq_range *hiset = &seq->lss_hiwater_set;
227 if (range_is_zero(loset))
228 __seq_set_init(env, seq);
230 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
231 loset->lsr_start = loset->lsr_end;
233 if (range_is_exhausted(loset)) {
234 /* reached high water mark. */
235 struct lu_device *dev = seq->lss_site->ss_lu->ls_top_dev;
236 int obd_num_clients = dev->ld_obd->obd_num_exports;
239 /* calculate new seq width based on number of clients */
240 set_sz = max(seq->lss_set_width,
241 obd_num_clients * seq->lss_width);
242 set_sz = min(range_space(space), set_sz);
244 /* Switch to hiwater range now */
246 /* allocate new hiwater range */
247 range_alloc(hiset, space, set_sz);
249 /* update ondisk seq with new *space */
250 rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
253 LASSERTF(!range_is_exhausted(loset) || range_is_sane(loset),
254 DRANGE"\n", PRANGE(loset));
257 range_alloc(out, loset, seq->lss_width);
262 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
263 struct lu_seq_range *out,
264 const struct lu_env *env)
266 struct lu_seq_range *space = &seq->lss_space;
271 LASSERT(range_is_sane(space));
273 /* Check if available space ends and allocate new super seq */
274 if (range_is_exhausted(space)) {
276 CERROR("%s: No sequence controller is attached.\n",
281 rc = seq_client_alloc_super(seq->lss_cli, env);
283 CERROR("%s: Can't allocate super-sequence, rc %d\n",
288 /* Saving new range to allocation space. */
289 *space = seq->lss_cli->lcs_space;
290 LASSERT(range_is_sane(space));
291 if (seq->lss_cli->lcs_srv == NULL) {
292 struct lu_server_fld *fld;
294 /* Insert it to the local FLDB */
295 fld = seq->lss_site->ss_server_fld;
296 mutex_lock(&fld->lsf_lock);
297 rc = fld_insert_entry(env, fld, space);
298 mutex_unlock(&fld->lsf_lock);
302 rc = range_alloc_set(env, out, seq);
304 CERROR("%s: Allocated meta-sequence failed: rc = %d\n",
309 CDEBUG(D_INFO, "%s: Allocated meta-sequence " DRANGE"\n",
310 seq->lss_name, PRANGE(out));
315 int seq_server_alloc_meta(struct lu_server_seq *seq,
316 struct lu_seq_range *out,
317 const struct lu_env *env)
322 mutex_lock(&seq->lss_mutex);
323 rc = __seq_server_alloc_meta(seq, out, env);
324 mutex_unlock(&seq->lss_mutex);
328 EXPORT_SYMBOL(seq_server_alloc_meta);
330 static int seq_server_handle(struct lu_site *site,
331 const struct lu_env *env,
332 __u32 opc, struct lu_seq_range *out)
335 struct seq_server_site *ss_site;
338 ss_site = lu_site2seq(site);
342 if (!ss_site->ss_server_seq) {
343 CERROR("Sequence server is not "
347 rc = seq_server_alloc_meta(ss_site->ss_server_seq, out, env);
349 case SEQ_ALLOC_SUPER:
350 if (!ss_site->ss_control_seq) {
351 CERROR("Sequence controller is not "
355 rc = seq_server_alloc_super(ss_site->ss_control_seq, out, env);
365 static int seq_handler(struct tgt_session_info *tsi)
367 struct lu_seq_range *out, *tmp;
368 struct lu_site *site;
374 LASSERT(!(lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY));
375 site = tsi->tsi_exp->exp_obd->obd_lu_dev->ld_site;
376 LASSERT(site != NULL);
378 opc = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_OPC);
380 out = req_capsule_server_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
382 RETURN(err_serious(-EPROTO));
384 tmp = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
386 /* seq client passed mdt id, we need to pass that using out
389 out->lsr_index = tmp->lsr_index;
390 out->lsr_flags = tmp->lsr_flags;
391 rc = seq_server_handle(site, tsi->tsi_env, *opc, out);
393 rc = err_serious(-EPROTO);
399 struct tgt_handler seq_handlers[] = {
400 TGT_SEQ_HDL(HABEO_REFERO, SEQ_QUERY, seq_handler),
402 EXPORT_SYMBOL(seq_handlers);
404 /* context key constructor/destructor: seq_key_init, seq_key_fini */
405 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
407 /* context key: seq_thread_key */
408 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD | LCT_DT_THREAD);
410 extern const struct file_operations seq_fld_proc_seq_fops;
412 static int seq_server_proc_init(struct lu_server_seq *seq)
418 seq->lss_proc_dir = lprocfs_seq_register(seq->lss_name,
421 if (IS_ERR(seq->lss_proc_dir)) {
422 rc = PTR_ERR(seq->lss_proc_dir);
426 rc = lprocfs_seq_add_vars(seq->lss_proc_dir,
427 seq_server_proc_list, seq);
429 CERROR("%s: Can't init sequence manager "
430 "proc, rc %d\n", seq->lss_name, rc);
431 GOTO(out_cleanup, rc);
434 if (seq->lss_type == LUSTRE_SEQ_CONTROLLER) {
435 rc = lprocfs_seq_create(seq->lss_proc_dir, "fldb", 0644,
436 &seq_fld_proc_seq_fops, seq);
438 CERROR("%s: Can't create fldb for sequence manager "
439 "proc: rc = %d\n", seq->lss_name, rc);
440 GOTO(out_cleanup, rc);
447 seq_server_proc_fini(seq);
454 static void seq_server_proc_fini(struct lu_server_seq *seq)
458 if (seq->lss_proc_dir != NULL) {
459 if (!IS_ERR(seq->lss_proc_dir))
460 lprocfs_remove(&seq->lss_proc_dir);
461 seq->lss_proc_dir = NULL;
467 int seq_server_init(const struct lu_env *env,
468 struct lu_server_seq *seq,
469 struct dt_device *dev,
471 enum lu_mgr_type type,
472 struct seq_server_site *ss)
474 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
477 LASSERT(dev != NULL);
478 LASSERT(prefix != NULL);
480 LASSERT(ss->ss_lu != NULL);
482 /* A compile-time check for FIDs that used to be in lustre_idl.h
483 * but is moved here to remove CLASSERT/LASSERT in that header.
484 * Check all lu_fid fields are converted in fid_cpu_to_le() and friends
485 * and that there is no padding added by compiler to the struct. */
489 CLASSERT(sizeof(tst) == sizeof(tst.f_seq) +
490 sizeof(tst.f_oid) + sizeof(tst.f_ver));
494 seq->lss_type = type;
496 range_init(&seq->lss_space);
498 range_init(&seq->lss_lowater_set);
499 range_init(&seq->lss_hiwater_set);
500 seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
502 mutex_init(&seq->lss_mutex);
504 seq->lss_width = is_srv ?
505 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
507 snprintf(seq->lss_name, sizeof(seq->lss_name),
508 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
510 rc = seq_store_init(seq, env, dev);
513 /* Request backing store for saved sequence info. */
514 rc = seq_store_read(seq, env);
515 if (rc == -ENODATA) {
517 /* Nothing is read, init by default value. */
518 seq->lss_space = is_srv ?
519 LUSTRE_SEQ_ZERO_RANGE:
520 LUSTRE_SEQ_SPACE_RANGE;
522 seq->lss_space.lsr_index = ss->ss_node_id;
523 LCONSOLE_INFO("%s: No data found "
524 "on store. Initialize space\n",
527 rc = seq_store_update(env, seq, NULL, 0);
529 CERROR("%s: Can't write space data, "
530 "rc %d\n", seq->lss_name, rc);
533 CERROR("%s: Can't read space data, rc %d\n",
539 LASSERT(range_is_sane(&seq->lss_space));
541 LASSERT(!range_is_zero(&seq->lss_space) &&
542 range_is_sane(&seq->lss_space));
545 rc = seq_server_proc_init(seq);
552 seq_server_fini(seq, env);
555 EXPORT_SYMBOL(seq_server_init);
557 void seq_server_fini(struct lu_server_seq *seq,
558 const struct lu_env *env)
562 seq_server_proc_fini(seq);
563 seq_store_fini(seq, env);
567 EXPORT_SYMBOL(seq_server_fini);
569 int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss)
574 if (ss->ss_server_seq) {
575 seq_server_fini(ss->ss_server_seq, env);
576 OBD_FREE_PTR(ss->ss_server_seq);
577 ss->ss_server_seq = NULL;
580 if (ss->ss_control_seq) {
581 seq_server_fini(ss->ss_control_seq, env);
582 OBD_FREE_PTR(ss->ss_control_seq);
583 ss->ss_control_seq = NULL;
586 if (ss->ss_client_seq) {
587 seq_client_fini(ss->ss_client_seq);
588 OBD_FREE_PTR(ss->ss_client_seq);
589 ss->ss_client_seq = NULL;
594 EXPORT_SYMBOL(seq_site_fini);
596 int fid_server_mod_init(void)
598 LU_CONTEXT_KEY_INIT(&seq_thread_key);
599 return lu_context_key_register(&seq_thread_key);
602 void fid_server_mod_exit(void)
604 lu_context_key_degister(&seq_thread_key);