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 %s\n",
73 seq->lss_name, cli->lcs_name);
78 if (seq->lss_cli != NULL) {
79 CDEBUG(D_HA, "%s: Sequence controller is already "
80 "assigned\n", seq->lss_name);
81 GOTO(out_up, rc = -EEXIST);
84 CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
85 seq->lss_name, cli->lcs_name);
88 cli->lcs_space.lsr_index = seq->lss_site->ss_node_id;
91 mutex_unlock(&seq->lss_mutex);
94 EXPORT_SYMBOL(seq_server_set_cli);
96 * allocate \a w units of sequence from range \a from.
98 static inline void range_alloc(struct lu_seq_range *to,
99 struct lu_seq_range *from,
102 width = min(range_space(from), width);
103 to->lsr_start = from->lsr_start;
104 to->lsr_end = from->lsr_start + width;
105 from->lsr_start += width;
109 * On controller node, allocate new super sequence for regular sequence server.
110 * As this super sequence controller, this node suppose to maintain fld
112 * \a out range always has currect mds node number of requester.
115 static int __seq_server_alloc_super(struct lu_server_seq *seq,
116 struct lu_seq_range *out,
117 const struct lu_env *env)
119 struct lu_seq_range *space = &seq->lss_space;
123 LASSERT(range_is_sane(space));
125 if (range_is_exhausted(space)) {
126 CERROR("%s: Sequences space is exhausted\n",
130 range_alloc(out, space, seq->lss_width);
133 rc = seq_store_update(env, seq, out, 1 /* sync */);
135 LCONSOLE_INFO("%s: super-sequence allocation rc = %d " DRANGE"\n",
136 seq->lss_name, rc, PRANGE(out));
141 int seq_server_alloc_super(struct lu_server_seq *seq,
142 struct lu_seq_range *out,
143 const struct lu_env *env)
148 mutex_lock(&seq->lss_mutex);
149 rc = __seq_server_alloc_super(seq, out, env);
150 mutex_unlock(&seq->lss_mutex);
155 int seq_server_alloc_spec(struct lu_server_seq *seq,
156 struct lu_seq_range *spec,
157 const struct lu_env *env)
159 struct lu_seq_range *space = &seq->lss_space;
164 * In some cases (like recovery after a disaster)
165 * we may need to allocate sequences manually
166 * Notice some sequences can be lost if requested
167 * range doesn't start at the beginning of current
168 * free space. Also notice it's not possible now
169 * to allocate sequences out of natural order.
171 if (spec->lsr_start >= spec->lsr_end)
173 if (spec->lsr_flags != LU_SEQ_RANGE_MDT &&
174 spec->lsr_flags != LU_SEQ_RANGE_OST)
177 mutex_lock(&seq->lss_mutex);
178 if (spec->lsr_start >= space->lsr_start) {
179 space->lsr_start = spec->lsr_end;
180 rc = seq_store_update(env, seq, spec, 1 /* sync */);
182 LCONSOLE_INFO("%s: "DRANGE" sequences allocated: rc = %d \n",
183 seq->lss_name, PRANGE(spec), rc);
185 mutex_unlock(&seq->lss_mutex);
190 static int __seq_set_init(const struct lu_env *env,
191 struct lu_server_seq *seq)
193 struct lu_seq_range *space = &seq->lss_space;
196 range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
197 range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
199 rc = seq_store_update(env, seq, NULL, 1);
205 * This function implements new seq allocation algorithm using async
206 * updates to seq file on disk. ref bug 18857 for details.
207 * there are four variable to keep track of this process
209 * lss_space; - available lss_space
210 * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
211 * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
214 * when lss_lowater_set reaches the end it is replaced with hiwater one and
215 * a write operation is initiated to allocate new hiwater range.
216 * if last seq write opearion is still not commited, current operation is
217 * flaged as sync write op.
219 static int range_alloc_set(const struct lu_env *env,
220 struct lu_seq_range *out,
221 struct lu_server_seq *seq)
223 struct lu_seq_range *space = &seq->lss_space;
224 struct lu_seq_range *loset = &seq->lss_lowater_set;
225 struct lu_seq_range *hiset = &seq->lss_hiwater_set;
228 if (range_is_zero(loset))
229 __seq_set_init(env, seq);
231 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
232 loset->lsr_start = loset->lsr_end;
234 if (range_is_exhausted(loset)) {
235 /* reached high water mark. */
236 struct lu_device *dev = seq->lss_site->ss_lu->ls_top_dev;
237 int obd_num_clients = dev->ld_obd->obd_num_exports;
240 /* calculate new seq width based on number of clients */
241 set_sz = max(seq->lss_set_width,
242 obd_num_clients * seq->lss_width);
243 set_sz = min(range_space(space), set_sz);
245 /* Switch to hiwater range now */
247 /* allocate new hiwater range */
248 range_alloc(hiset, space, set_sz);
250 /* update ondisk seq with new *space */
251 rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
254 LASSERTF(!range_is_exhausted(loset) || range_is_sane(loset),
255 DRANGE"\n", PRANGE(loset));
258 range_alloc(out, loset, seq->lss_width);
263 static int __seq_server_alloc_meta(struct lu_server_seq *seq,
264 struct lu_seq_range *out,
265 const struct lu_env *env)
267 struct lu_seq_range *space = &seq->lss_space;
272 LASSERT(range_is_sane(space));
274 /* Check if available space ends and allocate new super seq */
275 if (range_is_exhausted(space)) {
277 CERROR("%s: No sequence controller is attached.\n",
282 rc = seq_client_alloc_super(seq->lss_cli, env);
284 CERROR("%s: Can't allocate super-sequence, rc %d\n",
289 /* Saving new range to allocation space. */
290 *space = seq->lss_cli->lcs_space;
291 LASSERT(range_is_sane(space));
292 if (seq->lss_cli->lcs_srv == NULL) {
293 struct lu_server_fld *fld;
295 /* Insert it to the local FLDB */
296 fld = seq->lss_site->ss_server_fld;
297 mutex_lock(&fld->lsf_lock);
298 rc = fld_insert_entry(env, fld, space);
299 mutex_unlock(&fld->lsf_lock);
303 rc = range_alloc_set(env, out, seq);
305 CERROR("%s: Allocated meta-sequence failed: rc = %d\n",
310 CDEBUG(D_INFO, "%s: Allocated meta-sequence " DRANGE"\n",
311 seq->lss_name, PRANGE(out));
316 int seq_server_alloc_meta(struct lu_server_seq *seq,
317 struct lu_seq_range *out,
318 const struct lu_env *env)
323 mutex_lock(&seq->lss_mutex);
324 rc = __seq_server_alloc_meta(seq, out, env);
325 mutex_unlock(&seq->lss_mutex);
329 EXPORT_SYMBOL(seq_server_alloc_meta);
331 static int seq_server_handle(struct lu_site *site,
332 const struct lu_env *env,
333 __u32 opc, struct lu_seq_range *out)
336 struct seq_server_site *ss_site;
339 ss_site = lu_site2seq(site);
343 if (!ss_site->ss_server_seq) {
344 CERROR("Sequence server is not "
348 rc = seq_server_alloc_meta(ss_site->ss_server_seq, out, env);
350 case SEQ_ALLOC_SUPER:
351 if (!ss_site->ss_control_seq) {
352 CERROR("Sequence controller is not "
356 rc = seq_server_alloc_super(ss_site->ss_control_seq, out, env);
366 static int seq_handler(struct tgt_session_info *tsi)
368 struct lu_seq_range *out, *tmp;
369 struct lu_site *site;
375 LASSERT(!(lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY));
376 site = tsi->tsi_exp->exp_obd->obd_lu_dev->ld_site;
377 LASSERT(site != NULL);
379 opc = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_OPC);
381 out = req_capsule_server_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
383 RETURN(err_serious(-EPROTO));
385 tmp = req_capsule_client_get(tsi->tsi_pill, &RMF_SEQ_RANGE);
387 /* seq client passed mdt id, we need to pass that using out
390 out->lsr_index = tmp->lsr_index;
391 out->lsr_flags = tmp->lsr_flags;
392 rc = seq_server_handle(site, tsi->tsi_env, *opc, out);
394 rc = err_serious(-EPROTO);
400 struct tgt_handler seq_handlers[] = {
401 TGT_SEQ_HDL(HABEO_REFERO, SEQ_QUERY, seq_handler),
403 EXPORT_SYMBOL(seq_handlers);
405 /* context key constructor/destructor: seq_key_init, seq_key_fini */
406 LU_KEY_INIT_FINI(seq, struct seq_thread_info);
408 /* context key: seq_thread_key */
409 LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD | LCT_DT_THREAD);
411 extern const struct file_operations seq_fld_proc_seq_fops;
413 static int seq_server_proc_init(struct lu_server_seq *seq)
419 seq->lss_proc_dir = lprocfs_seq_register(seq->lss_name,
422 if (IS_ERR(seq->lss_proc_dir)) {
423 rc = PTR_ERR(seq->lss_proc_dir);
427 rc = lprocfs_seq_add_vars(seq->lss_proc_dir,
428 seq_server_proc_list, seq);
430 CERROR("%s: Can't init sequence manager "
431 "proc, rc %d\n", seq->lss_name, rc);
432 GOTO(out_cleanup, rc);
435 if (seq->lss_type == LUSTRE_SEQ_CONTROLLER) {
436 rc = lprocfs_seq_create(seq->lss_proc_dir, "fldb", 0644,
437 &seq_fld_proc_seq_fops, seq);
439 CERROR("%s: Can't create fldb for sequence manager "
440 "proc: rc = %d\n", seq->lss_name, rc);
441 GOTO(out_cleanup, rc);
448 seq_server_proc_fini(seq);
455 static void seq_server_proc_fini(struct lu_server_seq *seq)
459 if (seq->lss_proc_dir != NULL) {
460 if (!IS_ERR(seq->lss_proc_dir))
461 lprocfs_remove(&seq->lss_proc_dir);
462 seq->lss_proc_dir = NULL;
468 int seq_server_init(const struct lu_env *env,
469 struct lu_server_seq *seq,
470 struct dt_device *dev,
472 enum lu_mgr_type type,
473 struct seq_server_site *ss)
475 int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
478 LASSERT(dev != NULL);
479 LASSERT(prefix != NULL);
481 LASSERT(ss->ss_lu != NULL);
483 /* A compile-time check for FIDs that used to be in lustre_idl.h
484 * but is moved here to remove CLASSERT/LASSERT in that header.
485 * Check all lu_fid fields are converted in fid_cpu_to_le() and friends
486 * and that there is no padding added by compiler to the struct. */
490 CLASSERT(sizeof(tst) == sizeof(tst.f_seq) +
491 sizeof(tst.f_oid) + sizeof(tst.f_ver));
495 seq->lss_type = type;
497 range_init(&seq->lss_space);
499 range_init(&seq->lss_lowater_set);
500 range_init(&seq->lss_hiwater_set);
501 seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
503 mutex_init(&seq->lss_mutex);
505 seq->lss_width = is_srv ?
506 LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
508 snprintf(seq->lss_name, sizeof(seq->lss_name),
509 "%s-%s", (is_srv ? "srv" : "ctl"), prefix);
511 rc = seq_store_init(seq, env, dev);
514 /* Request backing store for saved sequence info. */
515 rc = seq_store_read(seq, env);
516 if (rc == -ENODATA) {
518 /* Nothing is read, init by default value. */
519 seq->lss_space = is_srv ?
520 LUSTRE_SEQ_ZERO_RANGE:
521 LUSTRE_SEQ_SPACE_RANGE;
523 seq->lss_space.lsr_index = ss->ss_node_id;
524 LCONSOLE_INFO("%s: No data found "
525 "on store. Initialize space\n",
528 rc = seq_store_update(env, seq, NULL, 0);
530 CERROR("%s: Can't write space data, "
531 "rc %d\n", seq->lss_name, rc);
534 CERROR("%s: Can't read space data, rc %d\n",
540 LASSERT(range_is_sane(&seq->lss_space));
542 LASSERT(!range_is_zero(&seq->lss_space) &&
543 range_is_sane(&seq->lss_space));
546 rc = seq_server_proc_init(seq);
553 seq_server_fini(seq, env);
556 EXPORT_SYMBOL(seq_server_init);
558 void seq_server_fini(struct lu_server_seq *seq,
559 const struct lu_env *env)
563 seq_server_proc_fini(seq);
564 seq_store_fini(seq, env);
568 EXPORT_SYMBOL(seq_server_fini);
570 int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss)
575 if (ss->ss_server_seq) {
576 seq_server_fini(ss->ss_server_seq, env);
577 OBD_FREE_PTR(ss->ss_server_seq);
578 ss->ss_server_seq = NULL;
581 if (ss->ss_control_seq) {
582 seq_server_fini(ss->ss_control_seq, env);
583 OBD_FREE_PTR(ss->ss_control_seq);
584 ss->ss_control_seq = NULL;
587 if (ss->ss_client_seq) {
588 seq_client_fini(ss->ss_client_seq);
589 OBD_FREE_PTR(ss->ss_client_seq);
590 ss->ss_client_seq = NULL;
595 EXPORT_SYMBOL(seq_site_fini);
597 int fid_server_mod_init(void)
599 LU_CONTEXT_KEY_INIT(&seq_thread_key);
600 return lu_context_key_register(&seq_thread_key);
603 void fid_server_mod_exit(void)
605 lu_context_key_degister(&seq_thread_key);