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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/fid/fid_request.c
34 * Lustre Sequence Manager
36 * Author: Yury Umanets <umka@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_FID
41 #include <linux/err.h>
42 #include <linux/module.h>
43 #include <linux/delay.h>
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <lustre_fid.h>
49 #include <lustre_mdc.h>
50 #include "fid_internal.h"
52 struct dentry *seq_debugfs_dir;
54 static int seq_client_rpc(struct lu_client_seq *seq,
55 struct lu_seq_range *output, __u32 opc,
58 struct obd_export *exp = seq->lcs_exp;
59 struct ptlrpc_request *req;
60 struct lu_seq_range *out, *in;
62 unsigned int debug_mask;
66 LASSERT(exp != NULL && !IS_ERR(exp));
67 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
68 LUSTRE_MDS_VERSION, SEQ_QUERY);
72 /* Init operation code */
73 op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
76 /* Zero out input range, this is not recovery yet. */
77 in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
78 lu_seq_range_init(in);
80 ptlrpc_request_set_replen(req);
82 in->lsr_index = seq->lcs_space.lsr_index;
83 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
84 fld_range_set_mdt(in);
86 fld_range_set_ost(in);
88 if (opc == SEQ_ALLOC_SUPER) {
89 req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
90 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
99 if (seq->lcs_type == LUSTRE_SEQ_DATA) {
100 req->rq_no_resend = 1;
101 req->rq_no_delay = 1;
103 debug_mask = D_CONSOLE;
105 if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
106 req->rq_reply_portal = MDC_REPLY_PORTAL;
107 req->rq_request_portal = SEQ_METADATA_PORTAL;
109 req->rq_reply_portal = OSC_REPLY_PORTAL;
110 req->rq_request_portal = SEQ_DATA_PORTAL;
116 /* Allow seq client RPC during recovery time. */
117 req->rq_allow_replay = 1;
119 ptlrpc_at_set_req_timeout(req);
121 rc = ptlrpc_queue_wait(req);
126 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
129 if (!lu_seq_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 (lu_seq_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, env);
168 * Check whether the connection to seq controller has been
169 * setup (lcs_exp != NULL)
172 mutex_unlock(&seq->lcs_mutex);
173 RETURN(-EINPROGRESS);
176 rc = seq_client_rpc(seq, &seq->lcs_space,
177 SEQ_ALLOC_SUPER, "super");
179 mutex_unlock(&seq->lcs_mutex);
183 /* Request sequence-controller node to allocate new meta-sequence. */
184 static int seq_client_alloc_meta(const struct lu_env *env,
185 struct lu_client_seq *seq)
191 #ifdef HAVE_SEQ_SERVER
193 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
205 rc = seq_client_rpc(seq, &seq->lcs_space,
206 SEQ_ALLOC_META, "meta");
207 if (rc == -EINPROGRESS || rc == -EAGAIN)
209 * MDT0 is not ready, let's wait for 2
214 } while (rc == -EINPROGRESS || rc == -EAGAIN);
220 /* Allocate new sequence for client. */
221 static int seq_client_alloc_seq(const struct lu_env *env,
222 struct lu_client_seq *seq, u64 *seqnr)
227 LASSERT(lu_seq_range_is_sane(&seq->lcs_space));
229 if (lu_seq_range_is_exhausted(&seq->lcs_space)) {
230 rc = seq_client_alloc_meta(env, seq);
232 if (rc != -EINPROGRESS)
233 CERROR("%s: Cannot allocate new meta-sequence: rc = %d\n",
237 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
238 seq->lcs_name, PRANGE(&seq->lcs_space));
244 LASSERT(!lu_seq_range_is_exhausted(&seq->lcs_space));
245 *seqnr = seq->lcs_space.lsr_start;
246 seq->lcs_space.lsr_start += 1;
248 CDEBUG(D_INFO, "%s: Allocated sequence [%#llx]\n", seq->lcs_name,
254 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
255 wait_queue_entry_t *link)
257 if (seq->lcs_update) {
258 add_wait_queue(&seq->lcs_waitq, link);
259 set_current_state(TASK_UNINTERRUPTIBLE);
260 mutex_unlock(&seq->lcs_mutex);
264 mutex_lock(&seq->lcs_mutex);
265 remove_wait_queue(&seq->lcs_waitq, link);
266 set_current_state(TASK_RUNNING);
271 mutex_unlock(&seq->lcs_mutex);
276 static void seq_fid_alloc_fini(struct lu_client_seq *seq, __u64 seqnr,
279 LASSERT(seq->lcs_update == 1);
281 mutex_lock(&seq->lcs_mutex);
283 CDEBUG(D_INFO, "%s: New sequence [0x%16.16llx]\n",
284 seq->lcs_name, seqnr);
286 seq->lcs_fid.f_seq = seqnr;
289 * Since the caller require the whole seq,
290 * so marked this seq to be used
292 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
294 LUSTRE_METADATA_SEQ_MAX_WIDTH;
296 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
298 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
300 seq->lcs_fid.f_ver = 0;
304 wake_up_all(&seq->lcs_waitq);
308 * Allocate the whole non-used seq to the caller.
310 * \param[in] env pointer to the thread context
311 * \param[in,out] seq pointer to the client sequence manager
312 * \param[out] seqnr to hold the new allocated sequence
314 * \retval 0 for new sequence allocated.
315 * \retval Negative error number on failure.
317 int seq_client_get_seq(const struct lu_env *env,
318 struct lu_client_seq *seq, u64 *seqnr)
320 wait_queue_entry_t link;
323 LASSERT(seqnr != NULL);
325 mutex_lock(&seq->lcs_mutex);
328 /* To guarantee that we can get a whole non-used sequence. */
329 while (seq_fid_alloc_prep(seq, &link) != 0)
332 rc = seq_client_alloc_seq(env, seq, seqnr);
333 seq_fid_alloc_fini(seq, rc ? 0 : *seqnr, true);
335 CERROR("%s: Can't allocate new sequence: rc = %d\n",
337 mutex_unlock(&seq->lcs_mutex);
341 EXPORT_SYMBOL(seq_client_get_seq);
344 * Allocate new fid on passed client @seq and save it to @fid.
346 * \param[in] env pointer to the thread context
347 * \param[in,out] seq pointer to the client sequence manager
348 * \param[out] fid to hold the new allocated fid
350 * \retval 1 for notify the caller that sequence switch
351 * is performed to allow it to setup FLD for it.
352 * \retval 0 for new FID allocated in current sequence.
353 * \retval Negative error number on failure.
355 int seq_client_alloc_fid(const struct lu_env *env,
356 struct lu_client_seq *seq, struct lu_fid *fid)
358 wait_queue_entry_t link;
362 LASSERT(seq != NULL);
363 LASSERT(fid != NULL);
366 mutex_lock(&seq->lcs_mutex);
368 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
369 seq->lcs_fid.f_oid = seq->lcs_width;
374 if (unlikely(!fid_is_zero(&seq->lcs_fid) &&
375 fid_oid(&seq->lcs_fid) < seq->lcs_width)) {
376 /* Just bump last allocated fid and return to caller. */
377 seq->lcs_fid.f_oid++;
383 * Release seq::lcs_mutex via seq_fid_alloc_prep() to avoid
384 * deadlock during seq_client_alloc_seq().
386 rc = seq_fid_alloc_prep(seq, &link);
390 rc = seq_client_alloc_seq(env, seq, &seqnr);
391 /* Re-take seq::lcs_mutex via seq_fid_alloc_fini(). */
392 seq_fid_alloc_fini(seq, rc ? 0 : seqnr, false);
394 if (rc != -EINPROGRESS)
395 CERROR("%s: Can't allocate new sequence: rc = %d\n",
397 mutex_unlock(&seq->lcs_mutex);
407 mutex_unlock(&seq->lcs_mutex);
409 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
413 EXPORT_SYMBOL(seq_client_alloc_fid);
416 * Finish the current sequence due to disconnect.
417 * See mdc_import_event()
419 void seq_client_flush(struct lu_client_seq *seq)
421 wait_queue_entry_t link;
423 LASSERT(seq != NULL);
425 mutex_lock(&seq->lcs_mutex);
427 while (seq->lcs_update) {
428 add_wait_queue(&seq->lcs_waitq, &link);
429 set_current_state(TASK_UNINTERRUPTIBLE);
430 mutex_unlock(&seq->lcs_mutex);
434 mutex_lock(&seq->lcs_mutex);
435 remove_wait_queue(&seq->lcs_waitq, &link);
436 set_current_state(TASK_RUNNING);
439 fid_zero(&seq->lcs_fid);
441 * this id shld not be used for seq range allocation.
442 * set to -1 for dgb check.
444 seq->lcs_space.lsr_index = -1;
446 lu_seq_range_init(&seq->lcs_space);
447 mutex_unlock(&seq->lcs_mutex);
449 EXPORT_SYMBOL(seq_client_flush);
451 static void seq_client_debugfs_fini(struct lu_client_seq *seq)
453 debugfs_remove_recursive(seq->lcs_debugfs_entry);
456 static void seq_client_debugfs_init(struct lu_client_seq *seq)
458 seq->lcs_debugfs_entry = debugfs_create_dir(seq->lcs_name,
461 ldebugfs_add_vars(seq->lcs_debugfs_entry,
462 seq_client_debugfs_list, seq);
465 void seq_client_fini(struct lu_client_seq *seq)
469 seq_client_debugfs_fini(seq);
472 class_export_put(seq->lcs_exp);
479 EXPORT_SYMBOL(seq_client_fini);
481 void seq_client_init(struct lu_client_seq *seq,
482 struct obd_export *exp,
483 enum lu_cli_type type,
485 struct lu_server_seq *srv)
489 LASSERT(seq != NULL);
490 LASSERT(prefix != NULL);
493 seq->lcs_type = type;
495 mutex_init(&seq->lcs_mutex);
496 if (type == LUSTRE_SEQ_METADATA)
497 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
499 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
501 init_waitqueue_head(&seq->lcs_waitq);
502 /* Make sure that things are clear before work is started. */
503 seq_client_flush(seq);
506 seq->lcs_exp = class_export_get(exp);
508 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
511 seq_client_debugfs_init(seq);
513 EXPORT_SYMBOL(seq_client_init);
515 int client_fid_init(struct obd_device *obd,
516 struct obd_export *exp, enum lu_cli_type type)
518 struct client_obd *cli = &obd->u.cli;
523 down_write(&cli->cl_seq_rwsem);
524 OBD_ALLOC_PTR(cli->cl_seq);
526 GOTO(out, rc = -ENOMEM);
528 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
530 GOTO(out, rc = -ENOMEM);
532 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
534 /* Init client side sequence-manager */
535 seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
536 OBD_FREE(prefix, MAX_OBD_NAME + 5);
539 if (rc && cli->cl_seq) {
540 OBD_FREE_PTR(cli->cl_seq);
543 up_write(&cli->cl_seq_rwsem);
547 EXPORT_SYMBOL(client_fid_init);
549 int client_fid_fini(struct obd_device *obd)
551 struct client_obd *cli = &obd->u.cli;
554 down_write(&cli->cl_seq_rwsem);
556 seq_client_fini(cli->cl_seq);
557 OBD_FREE_PTR(cli->cl_seq);
560 up_write(&cli->cl_seq_rwsem);
564 EXPORT_SYMBOL(client_fid_fini);
566 static int __init fid_init(void)
569 #ifdef HAVE_SERVER_SUPPORT
570 int rc = fid_server_mod_init();
575 de = debugfs_create_dir(LUSTRE_SEQ_NAME,
576 debugfs_lustre_root);
578 seq_debugfs_dir = de;
579 return PTR_ERR_OR_ZERO(de);
582 static void __exit fid_exit(void)
584 # ifdef HAVE_SERVER_SUPPORT
585 fid_server_mod_exit();
587 debugfs_remove_recursive(seq_debugfs_dir);
590 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
591 MODULE_DESCRIPTION("Lustre File IDentifier");
592 MODULE_VERSION(LUSTRE_VERSION_STRING);
593 MODULE_LICENSE("GPL");
595 module_init(fid_init);
596 module_exit(fid_exit);