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_reply_portal = MDC_REPLY_PORTAL;
109 req->rq_request_portal = SEQ_METADATA_PORTAL;
111 req->rq_reply_portal = OSC_REPLY_PORTAL;
112 req->rq_request_portal = SEQ_DATA_PORTAL;
118 ptlrpc_at_set_req_timeout(req);
120 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
121 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
123 rc = ptlrpc_queue_wait(req);
125 if (opc != SEQ_ALLOC_SUPER && seq->lcs_type == LUSTRE_SEQ_METADATA)
126 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
130 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
133 if (!range_is_sane(output)) {
134 CERROR("%s: Invalid range received from server: "
135 DRANGE"\n", seq->lcs_name, PRANGE(output));
136 GOTO(out_req, rc = -EINVAL);
139 if (range_is_exhausted(output)) {
140 CERROR("%s: Range received from server is exhausted: "
141 DRANGE"]\n", seq->lcs_name, PRANGE(output));
142 GOTO(out_req, rc = -EINVAL);
145 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
146 seq->lcs_name, opcname, PRANGE(output));
150 ptlrpc_req_finished(req);
154 /* Request sequence-controller node to allocate new super-sequence. */
155 int seq_client_alloc_super(struct lu_client_seq *seq,
156 const struct lu_env *env)
161 mutex_lock(&seq->lcs_mutex);
164 #ifdef HAVE_SEQ_SERVER
165 LASSERT(env != NULL);
166 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
172 /* Check whether the connection to seq controller has been
173 * setup (lcs_exp != NULL) */
174 if (seq->lcs_exp == NULL) {
175 mutex_unlock(&seq->lcs_mutex);
176 RETURN(-EINPROGRESS);
179 rc = seq_client_rpc(seq, &seq->lcs_space,
180 SEQ_ALLOC_SUPER, "super");
182 mutex_unlock(&seq->lcs_mutex);
186 /* Request sequence-controller node to allocate new meta-sequence. */
187 static int seq_client_alloc_meta(const struct lu_env *env,
188 struct lu_client_seq *seq)
194 #ifdef HAVE_SEQ_SERVER
195 LASSERT(env != NULL);
196 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
202 /* If meta server return -EINPROGRESS or EAGAIN,
203 * it means meta server might not be ready to
204 * allocate super sequence from sequence controller
206 rc = seq_client_rpc(seq, &seq->lcs_space,
207 SEQ_ALLOC_META, "meta");
208 } while (rc == -EINPROGRESS || rc == -EAGAIN);
214 /* Allocate new sequence for client. */
215 static int seq_client_alloc_seq(const struct lu_env *env,
216 struct lu_client_seq *seq, seqno_t *seqnr)
221 LASSERT(range_is_sane(&seq->lcs_space));
223 if (range_is_exhausted(&seq->lcs_space)) {
224 rc = seq_client_alloc_meta(env, seq);
226 CERROR("%s: Can't allocate new meta-sequence,"
227 "rc %d\n", seq->lcs_name, rc);
230 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
231 seq->lcs_name, PRANGE(&seq->lcs_space));
237 LASSERT(!range_is_exhausted(&seq->lcs_space));
238 *seqnr = seq->lcs_space.lsr_start;
239 seq->lcs_space.lsr_start += 1;
241 CDEBUG(D_INFO, "%s: Allocated sequence ["LPX64"]\n", seq->lcs_name,
247 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
250 if (seq->lcs_update) {
251 add_wait_queue(&seq->lcs_waitq, link);
252 set_current_state(TASK_UNINTERRUPTIBLE);
253 mutex_unlock(&seq->lcs_mutex);
255 waitq_wait(link, TASK_UNINTERRUPTIBLE);
257 mutex_lock(&seq->lcs_mutex);
258 remove_wait_queue(&seq->lcs_waitq, link);
259 set_current_state(TASK_RUNNING);
263 mutex_unlock(&seq->lcs_mutex);
267 static void seq_fid_alloc_fini(struct lu_client_seq *seq)
269 LASSERT(seq->lcs_update == 1);
270 mutex_lock(&seq->lcs_mutex);
272 wake_up(&seq->lcs_waitq);
276 * Allocate the whole seq to the caller.
278 int seq_client_get_seq(const struct lu_env *env,
279 struct lu_client_seq *seq, seqno_t *seqnr)
284 LASSERT(seqnr != NULL);
285 mutex_lock(&seq->lcs_mutex);
286 init_waitqueue_entry_current(&link);
289 rc = seq_fid_alloc_prep(seq, &link);
294 rc = seq_client_alloc_seq(env, seq, seqnr);
296 CERROR("%s: Can't allocate new sequence, "
297 "rc %d\n", seq->lcs_name, rc);
298 seq_fid_alloc_fini(seq);
299 mutex_unlock(&seq->lcs_mutex);
303 CDEBUG(D_INFO, "%s: allocate sequence "
304 "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
306 /* Since the caller require the whole seq,
307 * so marked this seq to be used */
308 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
309 seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
311 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
313 seq->lcs_fid.f_seq = *seqnr;
314 seq->lcs_fid.f_ver = 0;
316 * Inform caller that sequence switch is performed to allow it
317 * to setup FLD for it.
319 seq_fid_alloc_fini(seq);
320 mutex_unlock(&seq->lcs_mutex);
324 EXPORT_SYMBOL(seq_client_get_seq);
326 /* Allocate new fid on passed client @seq and save it to @fid. */
327 int seq_client_alloc_fid(const struct lu_env *env,
328 struct lu_client_seq *seq, struct lu_fid *fid)
334 LASSERT(seq != NULL);
335 LASSERT(fid != NULL);
337 init_waitqueue_entry_current(&link);
338 mutex_lock(&seq->lcs_mutex);
340 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
341 seq->lcs_fid.f_oid = seq->lcs_width;
346 if (!fid_is_zero(&seq->lcs_fid) &&
347 fid_oid(&seq->lcs_fid) < seq->lcs_width) {
348 /* Just bump last allocated fid and return to caller. */
349 seq->lcs_fid.f_oid += 1;
354 rc = seq_fid_alloc_prep(seq, &link);
358 rc = seq_client_alloc_seq(env, seq, &seqnr);
360 CERROR("%s: Can't allocate new sequence, "
361 "rc %d\n", seq->lcs_name, rc);
362 seq_fid_alloc_fini(seq);
363 mutex_unlock(&seq->lcs_mutex);
367 CDEBUG(D_INFO, "%s: Switch to sequence "
368 "[0x%16.16"LPF64"x]\n", seq->lcs_name, seqnr);
370 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
371 seq->lcs_fid.f_seq = seqnr;
372 seq->lcs_fid.f_ver = 0;
375 * Inform caller that sequence switch is performed to allow it
376 * to setup FLD for it.
380 seq_fid_alloc_fini(seq);
385 mutex_unlock(&seq->lcs_mutex);
387 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
390 EXPORT_SYMBOL(seq_client_alloc_fid);
393 * Finish the current sequence due to disconnect.
394 * See mdc_import_event()
396 void seq_client_flush(struct lu_client_seq *seq)
400 LASSERT(seq != NULL);
401 init_waitqueue_entry_current(&link);
402 mutex_lock(&seq->lcs_mutex);
404 while (seq->lcs_update) {
405 add_wait_queue(&seq->lcs_waitq, &link);
406 set_current_state(TASK_UNINTERRUPTIBLE);
407 mutex_unlock(&seq->lcs_mutex);
409 waitq_wait(&link, TASK_UNINTERRUPTIBLE);
411 mutex_lock(&seq->lcs_mutex);
412 remove_wait_queue(&seq->lcs_waitq, &link);
413 set_current_state(TASK_RUNNING);
416 fid_zero(&seq->lcs_fid);
418 * this id shld not be used for seq range allocation.
419 * set to -1 for dgb check.
422 seq->lcs_space.lsr_index = -1;
424 range_init(&seq->lcs_space);
425 mutex_unlock(&seq->lcs_mutex);
427 EXPORT_SYMBOL(seq_client_flush);
429 static void seq_client_proc_fini(struct lu_client_seq *seq)
433 if (seq->lcs_proc_dir) {
434 if (!IS_ERR(seq->lcs_proc_dir))
435 lprocfs_remove(&seq->lcs_proc_dir);
436 seq->lcs_proc_dir = NULL;
442 static int seq_client_proc_init(struct lu_client_seq *seq)
448 seq->lcs_proc_dir = lprocfs_seq_register(seq->lcs_name,
451 if (IS_ERR(seq->lcs_proc_dir)) {
452 CERROR("%s: LProcFS failed in seq-init\n",
454 rc = PTR_ERR(seq->lcs_proc_dir);
458 rc = lprocfs_seq_add_vars(seq->lcs_proc_dir,
459 seq_client_proc_list, seq);
461 CERROR("%s: Can't init sequence manager "
462 "proc, rc %d\n", seq->lcs_name, rc);
463 GOTO(out_cleanup, rc);
469 seq_client_proc_fini(seq);
477 int seq_client_init(struct lu_client_seq *seq,
478 struct obd_export *exp,
479 enum lu_cli_type type,
481 struct lu_server_seq *srv)
486 LASSERT(seq != NULL);
487 LASSERT(prefix != NULL);
490 seq->lcs_type = type;
492 mutex_init(&seq->lcs_mutex);
493 if (type == LUSTRE_SEQ_METADATA)
494 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
496 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
498 init_waitqueue_head(&seq->lcs_waitq);
499 /* Make sure that things are clear before work is started. */
500 seq_client_flush(seq);
503 seq->lcs_exp = class_export_get(exp);
505 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
508 rc = seq_client_proc_init(seq);
510 seq_client_fini(seq);
513 EXPORT_SYMBOL(seq_client_init);
515 void seq_client_fini(struct lu_client_seq *seq)
519 seq_client_proc_fini(seq);
521 if (seq->lcs_exp != NULL) {
522 class_export_put(seq->lcs_exp);
529 EXPORT_SYMBOL(seq_client_fini);
531 int client_fid_init(struct obd_device *obd,
532 struct obd_export *exp, enum lu_cli_type type)
534 struct client_obd *cli = &obd->u.cli;
539 OBD_ALLOC_PTR(cli->cl_seq);
540 if (cli->cl_seq == NULL)
543 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
545 GOTO(out_free_seq, rc = -ENOMEM);
547 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
549 /* Init client side sequence-manager */
550 rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
551 OBD_FREE(prefix, MAX_OBD_NAME + 5);
553 GOTO(out_free_seq, rc);
557 OBD_FREE_PTR(cli->cl_seq);
561 EXPORT_SYMBOL(client_fid_init);
563 int client_fid_fini(struct obd_device *obd)
565 struct client_obd *cli = &obd->u.cli;
568 if (cli->cl_seq != NULL) {
569 seq_client_fini(cli->cl_seq);
570 OBD_FREE_PTR(cli->cl_seq);
576 EXPORT_SYMBOL(client_fid_fini);
579 struct proc_dir_entry *seq_type_proc_dir;
581 static int __init fid_mod_init(void)
583 seq_type_proc_dir = lprocfs_seq_register(LUSTRE_SEQ_NAME,
586 if (IS_ERR(seq_type_proc_dir))
587 return PTR_ERR(seq_type_proc_dir);
589 # ifdef HAVE_SERVER_SUPPORT
590 fid_server_mod_init();
596 static void __exit fid_mod_exit(void)
598 # ifdef HAVE_SERVER_SUPPORT
599 fid_server_mod_exit();
602 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
603 lprocfs_remove(&seq_type_proc_dir);
604 seq_type_proc_dir = NULL;
608 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
609 MODULE_DESCRIPTION("Lustre FID Module");
610 MODULE_LICENSE("GPL");
612 cfs_module(fid, "0.1.0", fid_mod_init, fid_mod_exit);
613 #endif /* __KERNEL__ */