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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/fid/fid_handler.c
34 * Lustre Sequence Manager
36 * Author: Yury Umanets <umka@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_FID
41 #include <libcfs/libcfs.h>
42 #include <linux/module.h>
44 #include <obd_class.h>
45 #include <dt_object.h>
46 #include <obd_support.h>
47 #include <lustre_req_layout.h>
48 #include <lustre_fid.h>
49 #include "fid_internal.h"
51 static void seq_server_proc_fini(struct lu_server_seq *seq);
53 /* Assigns client to sequence controller node. */
54 int seq_server_set_cli(const struct lu_env *env, struct lu_server_seq *seq,
55 struct lu_client_seq *cli)
61 * Ask client for new range, assign that range to ->seq_space and write
62 * seq state to backing store should be atomic.
64 mutex_lock(&seq->lss_mutex);
67 CDEBUG(D_INFO, "%s: Detached sequence client\n", seq->lss_name);
72 if (seq->lss_cli != NULL) {
73 CDEBUG(D_HA, "%s: Sequence controller is already "
74 "assigned\n", seq->lss_name);
75 GOTO(out_up, rc = -EEXIST);
78 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
79 seq->lss_name, cli->lcs_name);
82 cli->lcs_space.lsr_index = seq->lss_site->ss_node_id;
85 mutex_unlock(&seq->lss_mutex);
88 EXPORT_SYMBOL(seq_server_set_cli);
90 * allocate \a w units of sequence from range \a from.
92 static inline void range_alloc(struct lu_seq_range *to,
93 struct lu_seq_range *from,
96 width = min(lu_seq_range_space(from), width);
97 to->lsr_start = from->lsr_start;
98 to->lsr_end = from->lsr_start + width;
99 from->lsr_start += width;
103 * On controller node, allocate new super sequence for regular sequence server.
104 * As this super sequence controller, this node suppose to maintain fld
106 * \a out range always has currect mds node number of requester.
109 static int __seq_server_alloc_super(struct lu_server_seq *seq,
110 struct lu_seq_range *out,
111 const struct lu_env *env)
113 struct lu_seq_range *space = &seq->lss_space;
117 LASSERT(lu_seq_range_is_sane(space));
119 if (lu_seq_range_is_exhausted(space)) {
120 CERROR("%s: Sequences space is exhausted\n",
124 range_alloc(out, space, seq->lss_width);
127 rc = seq_store_update(env, seq, out, 1 /* sync */);
129 LCONSOLE_INFO("%s: super-sequence allocation rc = %d " DRANGE"\n",
130 seq->lss_name, rc, PRANGE(out));
135 int seq_server_alloc_super(struct lu_server_seq *seq,
136 struct lu_seq_range *out,
137 const struct lu_env *env)
142 mutex_lock(&seq->lss_mutex);
143 rc = __seq_server_alloc_super(seq, out, env);
144 mutex_unlock(&seq->lss_mutex);
149 int seq_server_alloc_spec(struct lu_server_seq *seq,
150 struct lu_seq_range *spec,
151 const struct lu_env *env)
153 struct lu_seq_range *space = &seq->lss_space;
158 * In some cases (like recovery after a disaster)
159 * we may need to allocate sequences manually
160 * Notice some sequences can be lost if requested
161 * range doesn't start at the beginning of current
162 * free space. Also notice it's not possible now
163 * to allocate sequences out of natural order.
165 if (spec->lsr_start >= spec->lsr_end)
167 if (spec->lsr_flags != LU_SEQ_RANGE_MDT &&
168 spec->lsr_flags != LU_SEQ_RANGE_OST)
171 mutex_lock(&seq->lss_mutex);
172 if (spec->lsr_start >= space->lsr_start) {
173 space->lsr_start = spec->lsr_end;
174 rc = seq_store_update(env, seq, spec, 1 /* sync */);
176 LCONSOLE_INFO("%s: "DRANGE" sequences allocated: rc = %d \n",
177 seq->lss_name, PRANGE(spec), rc);
179 mutex_unlock(&seq->lss_mutex);
184 static int __seq_set_init(const struct lu_env *env,
185 struct lu_server_seq *seq)
187 struct lu_seq_range *space = &seq->lss_space;
190 range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
191 range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
193 rc = seq_store_update(env, seq, NULL, 1);
199 * This function implements new seq allocation algorithm using async
200 * updates to seq file on disk. ref bug 18857 for details.
201 * there are four variable to keep track of this process
203 * lss_space; - available lss_space
204 * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
205 * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
208 * when lss_lowater_set reaches the end it is replaced with hiwater one and
209 * a write operation is initiated to allocate new hiwater range.
210 * if last seq write opearion is still not committed, current operation is
211 * flaged as sync write op.
213 static int range_alloc_set(const struct lu_env *env,
214 struct lu_seq_range *out,
215 struct lu_server_seq *seq)
217 struct lu_seq_range *space = &seq->lss_space;
218 struct lu_seq_range *loset = &seq->lss_lowater_set;
219 struct lu_seq_range *hiset = &seq->lss_hiwater_set;
222 if (lu_seq_range_is_zero(loset))
223 __seq_set_init(env, seq);
225 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
226 loset->lsr_start = loset->lsr_end;
228 if (lu_seq_range_is_exhausted(loset)) {
229 /* reached high water mark. */
230 struct lu_device *dev = seq->lss_site->ss_lu->ls_top_dev;
231 int obd_num_clients = dev->ld_obd->obd_num_exports;
234 /* calculate new seq width based on number of clients */
235 set_sz = max(seq->lss_set_width,
236 obd_num_clients * seq->lss_width);
237 set_sz = min(lu_seq_range_space(space), set_sz);
239 /* Switch to hiwater range now */
241 /* allocate new hiwater range */
242 range_alloc(hiset, space, set_sz);
244 /* update ondisk seq with new *space */
245 rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
248 LASSERTF(!lu_seq_range_is_exhausted(loset) ||
249 lu_seq_range_is_sane(loset),
250 DRANGE"\n", PRANGE(loset));
253 range_alloc(out, loset, seq->lss_width);
259 * Check if the sequence server has sequence avaible
261 * Check if the sequence server has sequence avaible, if not, then
262 * allocating super sequence from sequence manager (MDT0).
264 * \param[in] env execution environment
265 * \param[in] seq server sequence
267 * \retval negative errno if allocating new sequence fails
268 * \retval 0 if there is enough sequence or allocating
269 * new sequence succeeds
271 int seq_server_check_and_alloc_super(const struct lu_env *env,
272 struct lu_server_seq *seq)
274 struct lu_seq_range *space = &seq->lss_space;
279 /* Check if available space ends and allocate new super seq */
280 if (lu_seq_range_is_exhausted(space)) {
282 CERROR("%s: No sequence controller is attached.\n",
287 rc = seq_client_alloc_super(seq->lss_cli, env);
289 CDEBUG(D_HA, "%s: Can't allocate super-sequence:"
290 " rc %d\n", seq->lss_name, rc);
294 /* Saving new range to allocation space. */
295 *space = seq->lss_cli->lcs_space;
296 LASSERT(lu_seq_range_is_sane(space));
297 if (seq->lss_cli->lcs_srv == NULL) {
298 struct lu_server_fld *fld;
300 /* Insert it to the local FLDB */
301 fld = seq->lss_site->ss_server_fld;
302 mutex_lock(&fld->lsf_lock);
303 rc = fld_insert_entry(env, fld, space);
304 mutex_unlock(&fld->lsf_lock);
308 if (lu_seq_range_is_zero(&seq->lss_lowater_set))
309 __seq_set_init(env, seq);
313 EXPORT_SYMBOL(seq_server_check_and_alloc_super);
315 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
316 struct lu_seq_range *out,
317 const struct lu_env *env)
319 struct lu_seq_range *space = &seq->lss_space;
324 LASSERT(lu_seq_range_is_sane(space));
326 rc = seq_server_check_and_alloc_super(env, seq);
328 if (rc == -EINPROGRESS) {
331 if (printed++ % 8 == 0)
332 LCONSOLE_INFO("%s: Waiting to contact MDT0000 "
333 "to allocate super-sequence\n",
336 CERROR("%s: Allocated super-sequence failed: rc = %d\n",
342 rc = range_alloc_set(env, out, seq);
344 CERROR("%s: Allocated meta-sequence failed: rc = %d\n",
349 CDEBUG(D_INFO, "%s: Allocated meta-sequence " DRANGE"\n",
350 seq->lss_name, PRANGE(out));
355 int seq_server_alloc_meta(struct lu_server_seq *seq,
356 struct lu_seq_range *out,
357 const struct lu_env *env)
362 mutex_lock(&seq->lss_mutex);
363 rc = __seq_server_alloc_meta(seq, out, env);
364 mutex_unlock(&seq->lss_mutex);
368 EXPORT_SYMBOL(seq_server_alloc_meta);
370 static int seq_server_handle(struct lu_site *site,
371 const struct lu_env *env,
372 __u32 opc, struct lu_seq_range *out)
375 struct seq_server_site *ss_site;
376 struct dt_device *dev;
379 ss_site = lu_site2seq(site);
383 if (!ss_site->ss_server_seq) {
384 CERROR("Sequence server is not "
389 dev = lu2dt_dev(ss_site->ss_server_seq->lss_obj->do_lu.lo_dev);
393 rc = seq_server_alloc_meta(ss_site->ss_server_seq, out, env);
395 case SEQ_ALLOC_SUPER:
396 if (!ss_site->ss_control_seq) {
397 CERROR("Sequence controller is not "
402 dev = lu2dt_dev(ss_site->ss_control_seq->lss_obj->do_lu.lo_dev);
406 rc = seq_server_alloc_super(ss_site->ss_control_seq, out, env);
416 static int seq_handler(struct tgt_session_info *tsi)
418 struct lu_seq_range *out, *tmp;
419 struct lu_site *site;
425 LASSERT(!(lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY));
426 site = tsi->tsi_exp->exp_obd->obd_lu_dev->ld_site;
427 LASSERT(site != NULL);
429 opc = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_OPC);
431 out = req_capsule_server_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
433 RETURN(err_serious(-EPROTO));
435 tmp = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
437 /* seq client passed mdt id, we need to pass that using out
440 out->lsr_index = tmp->lsr_index;
441 out->lsr_flags = tmp->lsr_flags;
442 rc = seq_server_handle(site, tsi->tsi_env, *opc, out);
444 rc = err_serious(-EPROTO);
450 struct tgt_handler seq_handlers[] = {
451 TGT_SEQ_HDL(HABEO_REFERO, SEQ_QUERY, seq_handler),
453 EXPORT_SYMBOL(seq_handlers);
455 /* context key constructor/destructor: seq_key_init, seq_key_fini */
456 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
458 /* context key: seq_thread_key */
459 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD | LCT_DT_THREAD);
461 extern const struct file_operations seq_fld_proc_seq_fops;
463 static int seq_server_proc_init(struct lu_server_seq *seq)
465 #ifdef CONFIG_PROC_FS
469 seq->lss_proc_dir = lprocfs_register(seq->lss_name,
472 if (IS_ERR(seq->lss_proc_dir)) {
473 rc = PTR_ERR(seq->lss_proc_dir);
477 rc = lprocfs_add_vars(seq->lss_proc_dir, seq_server_proc_list, seq);
479 CERROR("%s: Can't init sequence manager "
480 "proc, rc %d\n", seq->lss_name, rc);
481 GOTO(out_cleanup, rc);
484 if (seq->lss_type == LUSTRE_SEQ_CONTROLLER) {
485 rc = lprocfs_seq_create(seq->lss_proc_dir, "fldb", 0644,
486 &seq_fld_proc_seq_fops, seq);
488 CERROR("%s: Can't create fldb for sequence manager "
489 "proc: rc = %d\n", seq->lss_name, rc);
490 GOTO(out_cleanup, rc);
497 seq_server_proc_fini(seq);
499 #else /* !CONFIG_PROC_FS */
501 #endif /* CONFIG_PROC_FS */
504 static void seq_server_proc_fini(struct lu_server_seq *seq)
506 #ifdef CONFIG_PROC_FS
508 if (seq->lss_proc_dir != NULL) {
509 if (!IS_ERR(seq->lss_proc_dir))
510 lprocfs_remove(&seq->lss_proc_dir);
511 seq->lss_proc_dir = NULL;
514 #endif /* CONFIG_PROC_FS */
517 int seq_server_init(const struct lu_env *env,
518 struct lu_server_seq *seq,
519 struct dt_device *dev,
521 enum lu_mgr_type type,
522 struct seq_server_site *ss)
524 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
527 LASSERT(dev != NULL);
528 LASSERT(prefix != NULL);
530 LASSERT(ss->ss_lu != NULL);
532 /* A compile-time check for FIDs that used to be in lustre_idl.h
533 * but is moved here to remove CLASSERT/LASSERT in that header.
534 * Check all lu_fid fields are converted in fid_cpu_to_le() and friends
535 * and that there is no padding added by compiler to the struct. */
539 CLASSERT(sizeof(tst) == sizeof(tst.f_seq) +
540 sizeof(tst.f_oid) + sizeof(tst.f_ver));
544 seq->lss_type = type;
546 lu_seq_range_init(&seq->lss_space);
548 lu_seq_range_init(&seq->lss_lowater_set);
549 lu_seq_range_init(&seq->lss_hiwater_set);
550 seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
552 mutex_init(&seq->lss_mutex);
554 seq->lss_width = is_srv ?
555 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
557 snprintf(seq->lss_name, sizeof(seq->lss_name),
558 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
560 rc = seq_store_init(seq, env, dev);
563 /* Request backing store for saved sequence info. */
564 rc = seq_store_read(seq, env);
565 if (rc == -ENODATA) {
567 /* Nothing is read, init by default value. */
568 seq->lss_space = is_srv ?
569 LUSTRE_SEQ_ZERO_RANGE:
570 LUSTRE_SEQ_SPACE_RANGE;
572 seq->lss_space.lsr_index = ss->ss_node_id;
573 LCONSOLE_INFO("%s: No data found "
574 "on store. Initialize space\n",
577 rc = seq_store_update(env, seq, NULL, 0);
579 CERROR("%s: Can't write space data, "
580 "rc %d\n", seq->lss_name, rc);
583 CERROR("%s: Can't read space data, rc %d\n",
589 LASSERT(lu_seq_range_is_sane(&seq->lss_space));
591 LASSERT(!lu_seq_range_is_zero(&seq->lss_space) &&
592 lu_seq_range_is_sane(&seq->lss_space));
595 rc = seq_server_proc_init(seq);
602 seq_server_fini(seq, env);
605 EXPORT_SYMBOL(seq_server_init);
607 void seq_server_fini(struct lu_server_seq *seq,
608 const struct lu_env *env)
612 seq_server_proc_fini(seq);
613 seq_store_fini(seq, env);
617 EXPORT_SYMBOL(seq_server_fini);
619 int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss)
624 if (ss->ss_server_seq) {
625 seq_server_fini(ss->ss_server_seq, env);
626 OBD_FREE_PTR(ss->ss_server_seq);
627 ss->ss_server_seq = NULL;
630 if (ss->ss_control_seq) {
631 seq_server_fini(ss->ss_control_seq, env);
632 OBD_FREE_PTR(ss->ss_control_seq);
633 ss->ss_control_seq = NULL;
636 if (ss->ss_client_seq) {
637 seq_client_fini(ss->ss_client_seq);
638 OBD_FREE_PTR(ss->ss_client_seq);
639 ss->ss_client_seq = NULL;
644 EXPORT_SYMBOL(seq_site_fini);
646 int fid_server_mod_init(void)
648 LU_CONTEXT_KEY_INIT(&seq_thread_key);
649 return lu_context_key_register(&seq_thread_key);
652 void fid_server_mod_exit(void)
654 lu_context_key_degister(&seq_thread_key);