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_request.c
38 * Lustre Sequence Manager
40 * Author: Yury Umanets <umka@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_FID
46 # include <libcfs/libcfs.h>
47 # include <linux/module.h>
48 #else /* __KERNEL__ */
49 # include <liblustre.h>
53 #include <obd_class.h>
54 #include <obd_support.h>
55 #include <lustre_fid.h>
57 #include <lustre_mdc.h>
58 #include "fid_internal.h"
60 static int seq_client_rpc(struct lu_client_seq *seq,
61 struct lu_seq_range *output, __u32 opc,
64 struct obd_export *exp = seq->lcs_exp;
65 struct ptlrpc_request *req;
66 struct lu_seq_range *out, *in;
68 unsigned int debug_mask;
72 LASSERT(exp != NULL && !IS_ERR(exp));
73 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
74 LUSTRE_MDS_VERSION, SEQ_QUERY);
78 /* Init operation code */
79 op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
82 /* Zero out input range, this is not recovery yet. */
83 in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
86 ptlrpc_request_set_replen(req);
88 in->lsr_index = seq->lcs_space.lsr_index;
89 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
90 fld_range_set_mdt(in);
92 fld_range_set_ost(in);
94 if (opc == SEQ_ALLOC_SUPER) {
95 req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
96 req->rq_reply_portal = MDC_REPLY_PORTAL;
97 /* During allocating super sequence for data object,
98 * the current thread might hold the export of MDT0(MDT0
99 * precreating objects on this OST), and it will send the
100 * request to MDT0 here, so we can not keep resending the
101 * request here, otherwise if MDT0 is failed(umounted),
102 * it can not release the export of MDT0 */
103 if (seq->lcs_type == LUSTRE_SEQ_DATA)
104 req->rq_no_delay = req->rq_no_resend = 1;
105 debug_mask = D_CONSOLE;
107 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
108 req->rq_request_portal = SEQ_METADATA_PORTAL;
110 req->rq_request_portal = SEQ_DATA_PORTAL;
114 ptlrpc_at_set_req_timeout(req);
116 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
117 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
119 rc = ptlrpc_queue_wait(req);
121 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
122 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
126 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
129 if (!range_is_sane(output)) {
130 CERROR("%s: Invalid range received from server: "
131 DRANGE"\n", seq->lcs_name, PRANGE(output));
132 GOTO(out_req, rc = -EINVAL);
135 if (range_is_exhausted(output)) {
136 CERROR("%s: Range received from server is exhausted: "
137 DRANGE"]\n", seq->lcs_name, PRANGE(output));
138 GOTO(out_req, rc = -EINVAL);
141 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
142 seq->lcs_name, opcname, PRANGE(output));
146 ptlrpc_req_finished(req);
150 /* Request sequence-controller node to allocate new super-sequence. */
151 int seq_client_alloc_super(struct lu_client_seq *seq,
152 const struct lu_env *env)
157 mutex_lock(&seq->lcs_mutex);
160 #ifdef HAVE_SEQ_SERVER
161 LASSERT(env != NULL);
162 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
168 /* Check whether the connection to seq controller has been
169 * setup (lcs_exp != NULL) */
170 if (seq->lcs_exp == NULL) {
171 mutex_unlock(&seq->lcs_mutex);
172 RETURN(-EINPROGRESS);
175 rc = seq_client_rpc(seq, &seq->lcs_space,
176 SEQ_ALLOC_SUPER, "super");
178 mutex_unlock(&seq->lcs_mutex);
182 /* Request sequence-controller node to allocate new meta-sequence. */
183 static int seq_client_alloc_meta(const struct lu_env *env,
184 struct lu_client_seq *seq)
190 #ifdef HAVE_SEQ_SERVER
191 LASSERT(env != NULL);
192 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
198 /* If meta server return -EINPROGRESS or EAGAIN,
199 * it means meta server might not be ready to
200 * allocate super sequence from sequence controller
202 rc = seq_client_rpc(seq, &seq->lcs_space,
203 SEQ_ALLOC_META, "meta");
204 } while (rc == -EINPROGRESS || rc == -EAGAIN);
210 /* Allocate new sequence for client. */
211 static int seq_client_alloc_seq(const struct lu_env *env,
212 struct lu_client_seq *seq, seqno_t *seqnr)
217 LASSERT(range_is_sane(&seq->lcs_space));
219 if (range_is_exhausted(&seq->lcs_space)) {
220 rc = seq_client_alloc_meta(env, seq);
222 CERROR("%s: Can't allocate new meta-sequence,"
223 "rc %d\n", seq->lcs_name, rc);
226 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
227 seq->lcs_name, PRANGE(&seq->lcs_space));
233 LASSERT(!range_is_exhausted(&seq->lcs_space));
234 *seqnr = seq->lcs_space.lsr_start;
235 seq->lcs_space.lsr_start += 1;
237 CDEBUG(D_INFO, "%s: Allocated sequence ["LPX64"]\n", seq->lcs_name,
243 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
246 if (seq->lcs_update) {
247 add_wait_queue(&seq->lcs_waitq, link);
248 set_current_state(TASK_UNINTERRUPTIBLE);
249 mutex_unlock(&seq->lcs_mutex);
251 waitq_wait(link, TASK_UNINTERRUPTIBLE);
253 mutex_lock(&seq->lcs_mutex);
254 remove_wait_queue(&seq->lcs_waitq, link);
255 set_current_state(TASK_RUNNING);
259 mutex_unlock(&seq->lcs_mutex);
263 static void seq_fid_alloc_fini(struct lu_client_seq *seq)
265 LASSERT(seq->lcs_update == 1);
266 mutex_lock(&seq->lcs_mutex);
268 wake_up(&seq->lcs_waitq);
272 * Allocate the whole seq to the caller.
274 int seq_client_get_seq(const struct lu_env *env,
275 struct lu_client_seq *seq, seqno_t *seqnr)
280 LASSERT(seqnr != NULL);
281 mutex_lock(&seq->lcs_mutex);
282 init_waitqueue_entry_current(&link);
285 rc = seq_fid_alloc_prep(seq, &link);
290 rc = seq_client_alloc_seq(env, seq, seqnr);
292 CERROR("%s: Can't allocate new sequence, "
293 "rc %d\n", seq->lcs_name, rc);
294 seq_fid_alloc_fini(seq);
295 mutex_unlock(&seq->lcs_mutex);
299 CDEBUG(D_INFO, "%s: allocate sequence "
300 "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
302 /* Since the caller require the whole seq,
303 * so marked this seq to be used */
304 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
305 seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
307 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
309 seq->lcs_fid.f_seq = *seqnr;
310 seq->lcs_fid.f_ver = 0;
312 * Inform caller that sequence switch is performed to allow it
313 * to setup FLD for it.
315 seq_fid_alloc_fini(seq);
316 mutex_unlock(&seq->lcs_mutex);
320 EXPORT_SYMBOL(seq_client_get_seq);
322 /* Allocate new fid on passed client @seq and save it to @fid. */
323 int seq_client_alloc_fid(const struct lu_env *env,
324 struct lu_client_seq *seq, struct lu_fid *fid)
330 LASSERT(seq != NULL);
331 LASSERT(fid != NULL);
333 init_waitqueue_entry_current(&link);
334 mutex_lock(&seq->lcs_mutex);
336 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
337 seq->lcs_fid.f_oid = seq->lcs_width;
342 if (!fid_is_zero(&seq->lcs_fid) &&
343 fid_oid(&seq->lcs_fid) < seq->lcs_width) {
344 /* Just bump last allocated fid and return to caller. */
345 seq->lcs_fid.f_oid += 1;
350 rc = seq_fid_alloc_prep(seq, &link);
354 rc = seq_client_alloc_seq(env, seq, &seqnr);
356 CERROR("%s: Can't allocate new sequence, "
357 "rc %d\n", seq->lcs_name, rc);
358 seq_fid_alloc_fini(seq);
359 mutex_unlock(&seq->lcs_mutex);
363 CDEBUG(D_INFO, "%s: Switch to sequence "
364 "[0x%16.16"LPF64"x]\n", seq->lcs_name, seqnr);
366 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
367 seq->lcs_fid.f_seq = seqnr;
368 seq->lcs_fid.f_ver = 0;
371 * Inform caller that sequence switch is performed to allow it
372 * to setup FLD for it.
376 seq_fid_alloc_fini(seq);
381 mutex_unlock(&seq->lcs_mutex);
383 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
386 EXPORT_SYMBOL(seq_client_alloc_fid);
389 * Finish the current sequence due to disconnect.
390 * See mdc_import_event()
392 void seq_client_flush(struct lu_client_seq *seq)
396 LASSERT(seq != NULL);
397 init_waitqueue_entry_current(&link);
398 mutex_lock(&seq->lcs_mutex);
400 while (seq->lcs_update) {
401 add_wait_queue(&seq->lcs_waitq, &link);
402 set_current_state(TASK_UNINTERRUPTIBLE);
403 mutex_unlock(&seq->lcs_mutex);
405 waitq_wait(&link, TASK_UNINTERRUPTIBLE);
407 mutex_lock(&seq->lcs_mutex);
408 remove_wait_queue(&seq->lcs_waitq, &link);
409 set_current_state(TASK_RUNNING);
412 fid_zero(&seq->lcs_fid);
414 * this id shld not be used for seq range allocation.
415 * set to -1 for dgb check.
418 seq->lcs_space.lsr_index = -1;
420 range_init(&seq->lcs_space);
421 mutex_unlock(&seq->lcs_mutex);
423 EXPORT_SYMBOL(seq_client_flush);
425 static void seq_client_proc_fini(struct lu_client_seq *seq)
429 if (seq->lcs_proc_dir) {
430 if (!IS_ERR(seq->lcs_proc_dir))
431 lprocfs_remove(&seq->lcs_proc_dir);
432 seq->lcs_proc_dir = NULL;
438 static int seq_client_proc_init(struct lu_client_seq *seq)
444 seq->lcs_proc_dir = lprocfs_seq_register(seq->lcs_name,
447 if (IS_ERR(seq->lcs_proc_dir)) {
448 CERROR("%s: LProcFS failed in seq-init\n",
450 rc = PTR_ERR(seq->lcs_proc_dir);
454 rc = lprocfs_seq_add_vars(seq->lcs_proc_dir,
455 seq_client_proc_list, seq);
457 CERROR("%s: Can't init sequence manager "
458 "proc, rc %d\n", seq->lcs_name, rc);
459 GOTO(out_cleanup, rc);
465 seq_client_proc_fini(seq);
473 int seq_client_init(struct lu_client_seq *seq,
474 struct obd_export *exp,
475 enum lu_cli_type type,
477 struct lu_server_seq *srv)
482 LASSERT(seq != NULL);
483 LASSERT(prefix != NULL);
486 seq->lcs_type = type;
488 mutex_init(&seq->lcs_mutex);
489 if (type == LUSTRE_SEQ_METADATA)
490 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
492 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
494 init_waitqueue_head(&seq->lcs_waitq);
495 /* Make sure that things are clear before work is started. */
496 seq_client_flush(seq);
499 seq->lcs_exp = class_export_get(exp);
501 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
504 rc = seq_client_proc_init(seq);
506 seq_client_fini(seq);
509 EXPORT_SYMBOL(seq_client_init);
511 void seq_client_fini(struct lu_client_seq *seq)
515 seq_client_proc_fini(seq);
517 if (seq->lcs_exp != NULL) {
518 class_export_put(seq->lcs_exp);
525 EXPORT_SYMBOL(seq_client_fini);
527 int client_fid_init(struct obd_device *obd,
528 struct obd_export *exp, enum lu_cli_type type)
530 struct client_obd *cli = &obd->u.cli;
535 OBD_ALLOC_PTR(cli->cl_seq);
536 if (cli->cl_seq == NULL)
539 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
541 GOTO(out_free_seq, rc = -ENOMEM);
543 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
545 /* Init client side sequence-manager */
546 rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
547 OBD_FREE(prefix, MAX_OBD_NAME + 5);
549 GOTO(out_free_seq, rc);
553 OBD_FREE_PTR(cli->cl_seq);
557 EXPORT_SYMBOL(client_fid_init);
559 int client_fid_fini(struct obd_device *obd)
561 struct client_obd *cli = &obd->u.cli;
564 if (cli->cl_seq != NULL) {
565 seq_client_fini(cli->cl_seq);
566 OBD_FREE_PTR(cli->cl_seq);
572 EXPORT_SYMBOL(client_fid_fini);
575 struct proc_dir_entry *seq_type_proc_dir;
577 static int __init fid_mod_init(void)
579 seq_type_proc_dir = lprocfs_seq_register(LUSTRE_SEQ_NAME,
582 if (IS_ERR(seq_type_proc_dir))
583 return PTR_ERR(seq_type_proc_dir);
585 # ifdef HAVE_SERVER_SUPPORT
586 fid_server_mod_init();
592 static void __exit fid_mod_exit(void)
594 # ifdef HAVE_SERVER_SUPPORT
595 fid_server_mod_exit();
598 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
599 lprocfs_remove(&seq_type_proc_dir);
600 seq_type_proc_dir = NULL;
604 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
605 MODULE_DESCRIPTION("Lustre FID Module");
606 MODULE_LICENSE("GPL");
608 cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);
609 #endif /* __KERNEL__ */