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, 2014, 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
45 #include <linux/module.h>
46 #include <libcfs/libcfs.h>
48 #include <obd_class.h>
49 #include <obd_support.h>
50 #include <lustre_fid.h>
52 #include <lustre_mdc.h>
53 #include "fid_internal.h"
55 static int seq_client_rpc(struct lu_client_seq *seq,
56 struct lu_seq_range *output, __u32 opc,
59 struct obd_export *exp = seq->lcs_exp;
60 struct ptlrpc_request *req;
61 struct lu_seq_range *out, *in;
63 unsigned int debug_mask;
67 LASSERT(exp != NULL && !IS_ERR(exp));
68 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
69 LUSTRE_MDS_VERSION, SEQ_QUERY);
73 /* Init operation code */
74 op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
77 /* Zero out input range, this is not recovery yet. */
78 in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
81 ptlrpc_request_set_replen(req);
83 in->lsr_index = seq->lcs_space.lsr_index;
84 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
85 fld_range_set_mdt(in);
87 fld_range_set_ost(in);
89 if (opc == SEQ_ALLOC_SUPER) {
90 req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
91 req->rq_reply_portal = MDC_REPLY_PORTAL;
92 /* During allocating super sequence for data object,
93 * the current thread might hold the export of MDT0(MDT0
94 * precreating objects on this OST), and it will send the
95 * request to MDT0 here, so we can not keep resending the
96 * request here, otherwise if MDT0 is failed(umounted),
97 * it can not release the export of MDT0 */
98 if (seq->lcs_type == LUSTRE_SEQ_DATA)
99 req->rq_no_delay = req->rq_no_resend = 1;
100 debug_mask = D_CONSOLE;
102 if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
103 req->rq_reply_portal = MDC_REPLY_PORTAL;
104 req->rq_request_portal = SEQ_METADATA_PORTAL;
106 req->rq_reply_portal = OSC_REPLY_PORTAL;
107 req->rq_request_portal = SEQ_DATA_PORTAL;
113 /* Allow seq client RPC during recovery time. */
114 req->rq_allow_replay = 1;
116 ptlrpc_at_set_req_timeout(req);
118 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
119 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
121 rc = ptlrpc_queue_wait(req);
123 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
124 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
128 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
131 if (!range_is_sane(output)) {
132 CERROR("%s: Invalid range received from server: "
133 DRANGE"\n", seq->lcs_name, PRANGE(output));
134 GOTO(out_req, rc = -EINVAL);
137 if (range_is_exhausted(output)) {
138 CERROR("%s: Range received from server is exhausted: "
139 DRANGE"]\n", seq->lcs_name, PRANGE(output));
140 GOTO(out_req, rc = -EINVAL);
143 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
144 seq->lcs_name, opcname, PRANGE(output));
148 ptlrpc_req_finished(req);
152 /* Request sequence-controller node to allocate new super-sequence. */
153 int seq_client_alloc_super(struct lu_client_seq *seq,
154 const struct lu_env *env)
159 mutex_lock(&seq->lcs_mutex);
162 #ifdef HAVE_SEQ_SERVER
163 LASSERT(env != NULL);
164 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
170 /* Check whether the connection to seq controller has been
171 * setup (lcs_exp != NULL) */
172 if (seq->lcs_exp == NULL) {
173 mutex_unlock(&seq->lcs_mutex);
174 RETURN(-EINPROGRESS);
177 rc = seq_client_rpc(seq, &seq->lcs_space,
178 SEQ_ALLOC_SUPER, "super");
180 mutex_unlock(&seq->lcs_mutex);
184 /* Request sequence-controller node to allocate new meta-sequence. */
185 static int seq_client_alloc_meta(const struct lu_env *env,
186 struct lu_client_seq *seq)
192 #ifdef HAVE_SEQ_SERVER
193 LASSERT(env != NULL);
194 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
200 /* If meta server return -EINPROGRESS or EAGAIN,
201 * it means meta server might not be ready to
202 * allocate super sequence from sequence controller
204 rc = seq_client_rpc(seq, &seq->lcs_space,
205 SEQ_ALLOC_META, "meta");
206 } while (rc == -EINPROGRESS || rc == -EAGAIN);
212 /* Allocate new sequence for client. */
213 static int seq_client_alloc_seq(const struct lu_env *env,
214 struct lu_client_seq *seq, u64 *seqnr)
219 LASSERT(range_is_sane(&seq->lcs_space));
221 if (range_is_exhausted(&seq->lcs_space)) {
222 rc = seq_client_alloc_meta(env, seq);
224 CERROR("%s: Can't allocate new meta-sequence,"
225 "rc %d\n", seq->lcs_name, rc);
228 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
229 seq->lcs_name, PRANGE(&seq->lcs_space));
235 LASSERT(!range_is_exhausted(&seq->lcs_space));
236 *seqnr = seq->lcs_space.lsr_start;
237 seq->lcs_space.lsr_start += 1;
239 CDEBUG(D_INFO, "%s: Allocated sequence ["LPX64"]\n", seq->lcs_name,
245 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
248 if (seq->lcs_update) {
249 add_wait_queue(&seq->lcs_waitq, link);
250 set_current_state(TASK_UNINTERRUPTIBLE);
251 mutex_unlock(&seq->lcs_mutex);
255 mutex_lock(&seq->lcs_mutex);
256 remove_wait_queue(&seq->lcs_waitq, link);
257 set_current_state(TASK_RUNNING);
261 mutex_unlock(&seq->lcs_mutex);
265 static void seq_fid_alloc_fini(struct lu_client_seq *seq)
267 LASSERT(seq->lcs_update == 1);
268 mutex_lock(&seq->lcs_mutex);
270 wake_up(&seq->lcs_waitq);
274 * Allocate the whole seq to the caller.
276 int seq_client_get_seq(const struct lu_env *env,
277 struct lu_client_seq *seq, u64 *seqnr)
282 LASSERT(seqnr != NULL);
283 mutex_lock(&seq->lcs_mutex);
284 init_waitqueue_entry(&link, current);
287 rc = seq_fid_alloc_prep(seq, &link);
292 rc = seq_client_alloc_seq(env, seq, seqnr);
294 CERROR("%s: Can't allocate new sequence, "
295 "rc %d\n", seq->lcs_name, rc);
296 seq_fid_alloc_fini(seq);
297 mutex_unlock(&seq->lcs_mutex);
301 CDEBUG(D_INFO, "%s: allocate sequence "
302 "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
304 /* Since the caller require the whole seq,
305 * so marked this seq to be used */
306 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
307 seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
309 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
311 seq->lcs_fid.f_seq = *seqnr;
312 seq->lcs_fid.f_ver = 0;
314 * Inform caller that sequence switch is performed to allow it
315 * to setup FLD for it.
317 seq_fid_alloc_fini(seq);
318 mutex_unlock(&seq->lcs_mutex);
322 EXPORT_SYMBOL(seq_client_get_seq);
324 /* Allocate new fid on passed client @seq and save it to @fid. */
325 int seq_client_alloc_fid(const struct lu_env *env,
326 struct lu_client_seq *seq, struct lu_fid *fid)
332 LASSERT(seq != NULL);
333 LASSERT(fid != NULL);
335 init_waitqueue_entry(&link, current);
336 mutex_lock(&seq->lcs_mutex);
338 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
339 seq->lcs_fid.f_oid = seq->lcs_width;
344 if (!fid_is_zero(&seq->lcs_fid) &&
345 fid_oid(&seq->lcs_fid) < seq->lcs_width) {
346 /* Just bump last allocated fid and return to caller. */
347 seq->lcs_fid.f_oid += 1;
352 rc = seq_fid_alloc_prep(seq, &link);
356 rc = seq_client_alloc_seq(env, seq, &seqnr);
358 CERROR("%s: Can't allocate new sequence, "
359 "rc %d\n", seq->lcs_name, rc);
360 seq_fid_alloc_fini(seq);
361 mutex_unlock(&seq->lcs_mutex);
365 CDEBUG(D_INFO, "%s: Switch to sequence "
366 "[0x%16.16"LPF64"x]\n", seq->lcs_name, seqnr);
368 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
369 seq->lcs_fid.f_seq = seqnr;
370 seq->lcs_fid.f_ver = 0;
373 * Inform caller that sequence switch is performed to allow it
374 * to setup FLD for it.
378 seq_fid_alloc_fini(seq);
383 mutex_unlock(&seq->lcs_mutex);
385 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
388 EXPORT_SYMBOL(seq_client_alloc_fid);
391 * Finish the current sequence due to disconnect.
392 * See mdc_import_event()
394 void seq_client_flush(struct lu_client_seq *seq)
398 LASSERT(seq != NULL);
399 init_waitqueue_entry(&link, current);
400 mutex_lock(&seq->lcs_mutex);
402 while (seq->lcs_update) {
403 add_wait_queue(&seq->lcs_waitq, &link);
404 set_current_state(TASK_UNINTERRUPTIBLE);
405 mutex_unlock(&seq->lcs_mutex);
409 mutex_lock(&seq->lcs_mutex);
410 remove_wait_queue(&seq->lcs_waitq, &link);
411 set_current_state(TASK_RUNNING);
414 fid_zero(&seq->lcs_fid);
416 * this id shld not be used for seq range allocation.
417 * set to -1 for dgb check.
420 seq->lcs_space.lsr_index = -1;
422 range_init(&seq->lcs_space);
423 mutex_unlock(&seq->lcs_mutex);
425 EXPORT_SYMBOL(seq_client_flush);
427 static void seq_client_proc_fini(struct lu_client_seq *seq)
429 #ifdef CONFIG_PROC_FS
431 if (seq->lcs_proc_dir) {
432 if (!IS_ERR(seq->lcs_proc_dir))
433 lprocfs_remove(&seq->lcs_proc_dir);
434 seq->lcs_proc_dir = NULL;
437 #endif /* CONFIG_PROC_FS */
440 static int seq_client_proc_init(struct lu_client_seq *seq)
442 #ifdef CONFIG_PROC_FS
446 seq->lcs_proc_dir = lprocfs_register(seq->lcs_name, seq_type_proc_dir,
448 if (IS_ERR(seq->lcs_proc_dir)) {
449 CERROR("%s: LProcFS failed in seq-init\n",
451 rc = PTR_ERR(seq->lcs_proc_dir);
455 rc = lprocfs_add_vars(seq->lcs_proc_dir, 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);
468 #else /* !CONFIG_PROC_FS */
470 #endif /* CONFIG_PROC_FS */
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);
574 struct proc_dir_entry *seq_type_proc_dir;
576 static int __init fid_mod_init(void)
578 seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
581 if (IS_ERR(seq_type_proc_dir))
582 return PTR_ERR(seq_type_proc_dir);
584 # ifdef HAVE_SERVER_SUPPORT
585 fid_server_mod_init();
591 static void __exit fid_mod_exit(void)
593 # ifdef HAVE_SERVER_SUPPORT
594 fid_server_mod_exit();
597 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
598 lprocfs_remove(&seq_type_proc_dir);
599 seq_type_proc_dir = NULL;
603 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
604 MODULE_DESCRIPTION("Lustre FID Module");
605 MODULE_VERSION(LUSTRE_VERSION_STRING);
606 MODULE_LICENSE("GPL");
608 module_init(fid_mod_init);
609 module_exit(fid_mod_exit);