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 ptlrpc_at_set_req_timeout(req);
115 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
116 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
118 rc = ptlrpc_queue_wait(req);
120 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
121 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
125 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
128 if (!range_is_sane(output)) {
129 CERROR("%s: Invalid range received from server: "
130 DRANGE"\n", seq->lcs_name, PRANGE(output));
131 GOTO(out_req, rc = -EINVAL);
134 if (range_is_exhausted(output)) {
135 CERROR("%s: Range received from server is exhausted: "
136 DRANGE"]\n", seq->lcs_name, PRANGE(output));
137 GOTO(out_req, rc = -EINVAL);
140 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
141 seq->lcs_name, opcname, PRANGE(output));
145 ptlrpc_req_finished(req);
149 /* Request sequence-controller node to allocate new super-sequence. */
150 int seq_client_alloc_super(struct lu_client_seq *seq,
151 const struct lu_env *env)
156 mutex_lock(&seq->lcs_mutex);
159 #ifdef HAVE_SEQ_SERVER
160 LASSERT(env != NULL);
161 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
167 /* Check whether the connection to seq controller has been
168 * setup (lcs_exp != NULL) */
169 if (seq->lcs_exp == NULL) {
170 mutex_unlock(&seq->lcs_mutex);
171 RETURN(-EINPROGRESS);
174 rc = seq_client_rpc(seq, &seq->lcs_space,
175 SEQ_ALLOC_SUPER, "super");
177 mutex_unlock(&seq->lcs_mutex);
181 /* Request sequence-controller node to allocate new meta-sequence. */
182 static int seq_client_alloc_meta(const struct lu_env *env,
183 struct lu_client_seq *seq)
189 #ifdef HAVE_SEQ_SERVER
190 LASSERT(env != NULL);
191 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
197 /* If meta server return -EINPROGRESS or EAGAIN,
198 * it means meta server might not be ready to
199 * allocate super sequence from sequence controller
201 rc = seq_client_rpc(seq, &seq->lcs_space,
202 SEQ_ALLOC_META, "meta");
203 } while (rc == -EINPROGRESS || rc == -EAGAIN);
209 /* Allocate new sequence for client. */
210 static int seq_client_alloc_seq(const struct lu_env *env,
211 struct lu_client_seq *seq, u64 *seqnr)
216 LASSERT(range_is_sane(&seq->lcs_space));
218 if (range_is_exhausted(&seq->lcs_space)) {
219 rc = seq_client_alloc_meta(env, seq);
221 CERROR("%s: Can't allocate new meta-sequence,"
222 "rc %d\n", seq->lcs_name, rc);
225 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
226 seq->lcs_name, PRANGE(&seq->lcs_space));
232 LASSERT(!range_is_exhausted(&seq->lcs_space));
233 *seqnr = seq->lcs_space.lsr_start;
234 seq->lcs_space.lsr_start += 1;
236 CDEBUG(D_INFO, "%s: Allocated sequence ["LPX64"]\n", seq->lcs_name,
242 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
245 if (seq->lcs_update) {
246 add_wait_queue(&seq->lcs_waitq, link);
247 set_current_state(TASK_UNINTERRUPTIBLE);
248 mutex_unlock(&seq->lcs_mutex);
250 waitq_wait(link, TASK_UNINTERRUPTIBLE);
252 mutex_lock(&seq->lcs_mutex);
253 remove_wait_queue(&seq->lcs_waitq, link);
254 set_current_state(TASK_RUNNING);
258 mutex_unlock(&seq->lcs_mutex);
262 static void seq_fid_alloc_fini(struct lu_client_seq *seq)
264 LASSERT(seq->lcs_update == 1);
265 mutex_lock(&seq->lcs_mutex);
267 wake_up(&seq->lcs_waitq);
271 * Allocate the whole seq to the caller.
273 int seq_client_get_seq(const struct lu_env *env,
274 struct lu_client_seq *seq, u64 *seqnr)
279 LASSERT(seqnr != NULL);
280 mutex_lock(&seq->lcs_mutex);
281 init_waitqueue_entry_current(&link);
284 rc = seq_fid_alloc_prep(seq, &link);
289 rc = seq_client_alloc_seq(env, seq, seqnr);
291 CERROR("%s: Can't allocate new sequence, "
292 "rc %d\n", seq->lcs_name, rc);
293 seq_fid_alloc_fini(seq);
294 mutex_unlock(&seq->lcs_mutex);
298 CDEBUG(D_INFO, "%s: allocate sequence "
299 "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
301 /* Since the caller require the whole seq,
302 * so marked this seq to be used */
303 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
304 seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
306 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
308 seq->lcs_fid.f_seq = *seqnr;
309 seq->lcs_fid.f_ver = 0;
311 * Inform caller that sequence switch is performed to allow it
312 * to setup FLD for it.
314 seq_fid_alloc_fini(seq);
315 mutex_unlock(&seq->lcs_mutex);
319 EXPORT_SYMBOL(seq_client_get_seq);
321 /* Allocate new fid on passed client @seq and save it to @fid. */
322 int seq_client_alloc_fid(const struct lu_env *env,
323 struct lu_client_seq *seq, struct lu_fid *fid)
329 LASSERT(seq != NULL);
330 LASSERT(fid != NULL);
332 init_waitqueue_entry_current(&link);
333 mutex_lock(&seq->lcs_mutex);
335 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
336 seq->lcs_fid.f_oid = seq->lcs_width;
341 if (!fid_is_zero(&seq->lcs_fid) &&
342 fid_oid(&seq->lcs_fid) < seq->lcs_width) {
343 /* Just bump last allocated fid and return to caller. */
344 seq->lcs_fid.f_oid += 1;
349 rc = seq_fid_alloc_prep(seq, &link);
353 rc = seq_client_alloc_seq(env, seq, &seqnr);
355 CERROR("%s: Can't allocate new sequence, "
356 "rc %d\n", seq->lcs_name, rc);
357 seq_fid_alloc_fini(seq);
358 mutex_unlock(&seq->lcs_mutex);
362 CDEBUG(D_INFO, "%s: Switch to sequence "
363 "[0x%16.16"LPF64"x]\n", seq->lcs_name, seqnr);
365 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
366 seq->lcs_fid.f_seq = seqnr;
367 seq->lcs_fid.f_ver = 0;
370 * Inform caller that sequence switch is performed to allow it
371 * to setup FLD for it.
375 seq_fid_alloc_fini(seq);
380 mutex_unlock(&seq->lcs_mutex);
382 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
385 EXPORT_SYMBOL(seq_client_alloc_fid);
388 * Finish the current sequence due to disconnect.
389 * See mdc_import_event()
391 void seq_client_flush(struct lu_client_seq *seq)
395 LASSERT(seq != NULL);
396 init_waitqueue_entry_current(&link);
397 mutex_lock(&seq->lcs_mutex);
399 while (seq->lcs_update) {
400 add_wait_queue(&seq->lcs_waitq, &link);
401 set_current_state(TASK_UNINTERRUPTIBLE);
402 mutex_unlock(&seq->lcs_mutex);
404 waitq_wait(&link, TASK_UNINTERRUPTIBLE);
406 mutex_lock(&seq->lcs_mutex);
407 remove_wait_queue(&seq->lcs_waitq, &link);
408 set_current_state(TASK_RUNNING);
411 fid_zero(&seq->lcs_fid);
413 * this id shld not be used for seq range allocation.
414 * set to -1 for dgb check.
417 seq->lcs_space.lsr_index = -1;
419 range_init(&seq->lcs_space);
420 mutex_unlock(&seq->lcs_mutex);
422 EXPORT_SYMBOL(seq_client_flush);
424 static void seq_client_proc_fini(struct lu_client_seq *seq)
426 #ifdef CONFIG_PROC_FS
428 if (seq->lcs_proc_dir) {
429 if (!IS_ERR(seq->lcs_proc_dir))
430 lprocfs_remove(&seq->lcs_proc_dir);
431 seq->lcs_proc_dir = NULL;
434 #endif /* CONFIG_PROC_FS */
437 static int seq_client_proc_init(struct lu_client_seq *seq)
439 #ifdef CONFIG_PROC_FS
443 seq->lcs_proc_dir = lprocfs_register(seq->lcs_name, seq_type_proc_dir,
445 if (IS_ERR(seq->lcs_proc_dir)) {
446 CERROR("%s: LProcFS failed in seq-init\n",
448 rc = PTR_ERR(seq->lcs_proc_dir);
452 rc = lprocfs_add_vars(seq->lcs_proc_dir, seq_client_proc_list, seq);
454 CERROR("%s: Can't init sequence manager "
455 "proc, rc %d\n", seq->lcs_name, rc);
456 GOTO(out_cleanup, rc);
462 seq_client_proc_fini(seq);
465 #else /* !CONFIG_PROC_FS */
467 #endif /* CONFIG_PROC_FS */
470 int seq_client_init(struct lu_client_seq *seq,
471 struct obd_export *exp,
472 enum lu_cli_type type,
474 struct lu_server_seq *srv)
479 LASSERT(seq != NULL);
480 LASSERT(prefix != NULL);
483 seq->lcs_type = type;
485 mutex_init(&seq->lcs_mutex);
486 if (type == LUSTRE_SEQ_METADATA)
487 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
489 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
491 init_waitqueue_head(&seq->lcs_waitq);
492 /* Make sure that things are clear before work is started. */
493 seq_client_flush(seq);
496 seq->lcs_exp = class_export_get(exp);
498 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
501 rc = seq_client_proc_init(seq);
503 seq_client_fini(seq);
506 EXPORT_SYMBOL(seq_client_init);
508 void seq_client_fini(struct lu_client_seq *seq)
512 seq_client_proc_fini(seq);
514 if (seq->lcs_exp != NULL) {
515 class_export_put(seq->lcs_exp);
522 EXPORT_SYMBOL(seq_client_fini);
524 int client_fid_init(struct obd_device *obd,
525 struct obd_export *exp, enum lu_cli_type type)
527 struct client_obd *cli = &obd->u.cli;
532 OBD_ALLOC_PTR(cli->cl_seq);
533 if (cli->cl_seq == NULL)
536 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
538 GOTO(out_free_seq, rc = -ENOMEM);
540 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
542 /* Init client side sequence-manager */
543 rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
544 OBD_FREE(prefix, MAX_OBD_NAME + 5);
546 GOTO(out_free_seq, rc);
550 OBD_FREE_PTR(cli->cl_seq);
554 EXPORT_SYMBOL(client_fid_init);
556 int client_fid_fini(struct obd_device *obd)
558 struct client_obd *cli = &obd->u.cli;
561 if (cli->cl_seq != NULL) {
562 seq_client_fini(cli->cl_seq);
563 OBD_FREE_PTR(cli->cl_seq);
569 EXPORT_SYMBOL(client_fid_fini);
571 struct proc_dir_entry *seq_type_proc_dir;
573 static int __init fid_mod_init(void)
575 seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
578 if (IS_ERR(seq_type_proc_dir))
579 return PTR_ERR(seq_type_proc_dir);
581 # ifdef HAVE_SERVER_SUPPORT
582 fid_server_mod_init();
588 static void __exit fid_mod_exit(void)
590 # ifdef HAVE_SERVER_SUPPORT
591 fid_server_mod_exit();
594 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
595 lprocfs_remove(&seq_type_proc_dir);
596 seq_type_proc_dir = NULL;
600 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
601 MODULE_DESCRIPTION("Lustre FID Module");
602 MODULE_LICENSE("GPL");
604 cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);