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;
91 /* During allocating super sequence for data object,
92 * the current thread might hold the export of MDT0(MDT0
93 * precreating objects on this OST), and it will send the
94 * request to MDT0 here, so we can not keep resending the
95 * request here, otherwise if MDT0 is failed(umounted),
96 * it can not release the export of MDT0 */
97 if (seq->lcs_type == LUSTRE_SEQ_DATA)
98 req->rq_no_delay = req->rq_no_resend = 1;
99 debug_mask = D_CONSOLE;
101 if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
102 req->rq_reply_portal = MDC_REPLY_PORTAL;
103 req->rq_request_portal = SEQ_METADATA_PORTAL;
105 req->rq_reply_portal = OSC_REPLY_PORTAL;
106 req->rq_request_portal = SEQ_DATA_PORTAL;
112 /* Allow seq client RPC during recovery time. */
113 req->rq_allow_replay = 1;
115 ptlrpc_at_set_req_timeout(req);
117 rc = ptlrpc_queue_wait(req);
122 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
125 if (!lu_seq_range_is_sane(output)) {
126 CERROR("%s: Invalid range received from server: "
127 DRANGE"\n", seq->lcs_name, PRANGE(output));
128 GOTO(out_req, rc = -EINVAL);
131 if (lu_seq_range_is_exhausted(output)) {
132 CERROR("%s: Range received from server is exhausted: "
133 DRANGE"]\n", seq->lcs_name, PRANGE(output));
134 GOTO(out_req, rc = -EINVAL);
137 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
138 seq->lcs_name, opcname, PRANGE(output));
142 ptlrpc_req_finished(req);
146 /* Request sequence-controller node to allocate new super-sequence. */
147 int seq_client_alloc_super(struct lu_client_seq *seq,
148 const struct lu_env *env)
153 mutex_lock(&seq->lcs_mutex);
156 #ifdef HAVE_SEQ_SERVER
157 LASSERT(env != NULL);
158 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
164 /* Check whether the connection to seq controller has been
165 * setup (lcs_exp != NULL) */
166 if (seq->lcs_exp == NULL) {
167 mutex_unlock(&seq->lcs_mutex);
168 RETURN(-EINPROGRESS);
171 rc = seq_client_rpc(seq, &seq->lcs_space,
172 SEQ_ALLOC_SUPER, "super");
174 mutex_unlock(&seq->lcs_mutex);
178 /* Request sequence-controller node to allocate new meta-sequence. */
179 static int seq_client_alloc_meta(const struct lu_env *env,
180 struct lu_client_seq *seq)
186 #ifdef HAVE_SEQ_SERVER
188 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
194 /* If meta server return -EINPROGRESS or EAGAIN,
195 * it means meta server might not be ready to
196 * allocate super sequence from sequence controller
198 rc = seq_client_rpc(seq, &seq->lcs_space,
199 SEQ_ALLOC_META, "meta");
200 if (rc == -EINPROGRESS || rc == -EAGAIN)
201 /* MDT0 is not ready, let's wait for 2
202 * seconds and retry. */
205 } while (rc == -EINPROGRESS || rc == -EAGAIN);
211 /* Allocate new sequence for client. */
212 static int seq_client_alloc_seq(const struct lu_env *env,
213 struct lu_client_seq *seq, u64 *seqnr)
218 LASSERT(lu_seq_range_is_sane(&seq->lcs_space));
220 if (lu_seq_range_is_exhausted(&seq->lcs_space)) {
221 rc = seq_client_alloc_meta(env, seq);
223 if (rc != -EINPROGRESS)
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(!lu_seq_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 [%#llx]\n", seq->lcs_name,
245 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
246 wait_queue_entry_t *link)
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);
262 mutex_unlock(&seq->lcs_mutex);
267 static void seq_fid_alloc_fini(struct lu_client_seq *seq, __u64 seqnr,
270 LASSERT(seq->lcs_update == 1);
272 mutex_lock(&seq->lcs_mutex);
274 CDEBUG(D_INFO, "%s: New sequence [0x%16.16llx]\n",
275 seq->lcs_name, seqnr);
277 seq->lcs_fid.f_seq = seqnr;
279 /* Since the caller require the whole seq,
280 * so marked this seq to be used */
281 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
283 LUSTRE_METADATA_SEQ_MAX_WIDTH;
285 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
287 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
289 seq->lcs_fid.f_ver = 0;
293 wake_up_all(&seq->lcs_waitq);
297 * Allocate the whole non-used seq to the caller.
299 * \param[in] env pointer to the thread context
300 * \param[in,out] seq pointer to the client sequence manager
301 * \param[out] seqnr to hold the new allocated sequence
303 * \retval 0 for new sequence allocated.
304 * \retval Negative error number on failure.
306 int seq_client_get_seq(const struct lu_env *env,
307 struct lu_client_seq *seq, u64 *seqnr)
309 wait_queue_entry_t link;
312 LASSERT(seqnr != NULL);
314 mutex_lock(&seq->lcs_mutex);
315 init_waitqueue_entry(&link, current);
317 /* To guarantee that we can get a whole non-used sequence. */
318 while (seq_fid_alloc_prep(seq, &link) != 0);
320 rc = seq_client_alloc_seq(env, seq, seqnr);
321 seq_fid_alloc_fini(seq, rc ? 0 : *seqnr, true);
323 CERROR("%s: Can't allocate new sequence: rc = %d\n",
325 mutex_unlock(&seq->lcs_mutex);
329 EXPORT_SYMBOL(seq_client_get_seq);
332 * Allocate new fid on passed client @seq and save it to @fid.
334 * \param[in] env pointer to the thread context
335 * \param[in,out] seq pointer to the client sequence manager
336 * \param[out] fid to hold the new allocated fid
338 * \retval 1 for notify the caller that sequence switch
339 * is performed to allow it to setup FLD for it.
340 * \retval 0 for new FID allocated in current sequence.
341 * \retval Negative error number on failure.
343 int seq_client_alloc_fid(const struct lu_env *env,
344 struct lu_client_seq *seq, struct lu_fid *fid)
346 wait_queue_entry_t link;
350 LASSERT(seq != NULL);
351 LASSERT(fid != NULL);
353 init_waitqueue_entry(&link, current);
354 mutex_lock(&seq->lcs_mutex);
356 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
357 seq->lcs_fid.f_oid = seq->lcs_width;
362 if (unlikely(!fid_is_zero(&seq->lcs_fid) &&
363 fid_oid(&seq->lcs_fid) < seq->lcs_width)) {
364 /* Just bump last allocated fid and return to caller. */
365 seq->lcs_fid.f_oid++;
370 /* Release seq::lcs_mutex via seq_fid_alloc_prep() to avoid
371 * deadlock during seq_client_alloc_seq(). */
372 rc = seq_fid_alloc_prep(seq, &link);
376 rc = seq_client_alloc_seq(env, seq, &seqnr);
377 /* Re-take seq::lcs_mutex via seq_fid_alloc_fini(). */
378 seq_fid_alloc_fini(seq, rc ? 0 : seqnr, false);
380 if (rc != -EINPROGRESS)
381 CERROR("%s: Can't allocate new sequence: "
382 "rc = %d\n", seq->lcs_name, rc);
383 mutex_unlock(&seq->lcs_mutex);
393 mutex_unlock(&seq->lcs_mutex);
395 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
399 EXPORT_SYMBOL(seq_client_alloc_fid);
402 * Finish the current sequence due to disconnect.
403 * See mdc_import_event()
405 void seq_client_flush(struct lu_client_seq *seq)
407 wait_queue_entry_t link;
409 LASSERT(seq != NULL);
410 init_waitqueue_entry(&link, current);
411 mutex_lock(&seq->lcs_mutex);
413 while (seq->lcs_update) {
414 add_wait_queue(&seq->lcs_waitq, &link);
415 set_current_state(TASK_UNINTERRUPTIBLE);
416 mutex_unlock(&seq->lcs_mutex);
420 mutex_lock(&seq->lcs_mutex);
421 remove_wait_queue(&seq->lcs_waitq, &link);
422 set_current_state(TASK_RUNNING);
425 fid_zero(&seq->lcs_fid);
427 * this id shld not be used for seq range allocation.
428 * set to -1 for dgb check.
431 seq->lcs_space.lsr_index = -1;
433 lu_seq_range_init(&seq->lcs_space);
434 mutex_unlock(&seq->lcs_mutex);
436 EXPORT_SYMBOL(seq_client_flush);
438 static void seq_client_debugfs_fini(struct lu_client_seq *seq)
440 debugfs_remove_recursive(seq->lcs_debugfs_entry);
443 static int seq_client_debugfs_init(struct lu_client_seq *seq)
447 seq->lcs_debugfs_entry = ldebugfs_register(seq->lcs_name,
450 if (IS_ERR_OR_NULL(seq->lcs_debugfs_entry)) {
451 CERROR("%s: LdebugFS failed in seq-init\n", seq->lcs_name);
452 rc = seq->lcs_debugfs_entry ? PTR_ERR(seq->lcs_debugfs_entry)
454 seq->lcs_debugfs_entry = NULL;
458 rc = ldebugfs_add_vars(seq->lcs_debugfs_entry,
459 seq_client_debugfs_list, seq);
461 CERROR("%s: Can't init sequence manager debugfs, rc %d\n",
463 GOTO(out_cleanup, rc);
469 seq_client_debugfs_fini(seq);
473 void seq_client_fini(struct lu_client_seq *seq)
477 seq_client_debugfs_fini(seq);
479 if (seq->lcs_exp != NULL) {
480 class_export_put(seq->lcs_exp);
487 EXPORT_SYMBOL(seq_client_fini);
489 int seq_client_init(struct lu_client_seq *seq,
490 struct obd_export *exp,
491 enum lu_cli_type type,
493 struct lu_server_seq *srv)
498 LASSERT(seq != NULL);
499 LASSERT(prefix != NULL);
502 seq->lcs_type = type;
504 mutex_init(&seq->lcs_mutex);
505 if (type == LUSTRE_SEQ_METADATA)
506 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
508 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
510 init_waitqueue_head(&seq->lcs_waitq);
511 /* Make sure that things are clear before work is started. */
512 seq_client_flush(seq);
515 seq->lcs_exp = class_export_get(exp);
517 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
520 rc = seq_client_debugfs_init(seq);
522 seq_client_fini(seq);
525 EXPORT_SYMBOL(seq_client_init);
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 down_write(&cli->cl_seq_rwsem);
536 OBD_ALLOC_PTR(cli->cl_seq);
538 GOTO(out, rc = -ENOMEM);
540 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
542 GOTO(out, rc = -ENOMEM);
544 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
546 /* Init client side sequence-manager */
547 rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
548 OBD_FREE(prefix, MAX_OBD_NAME + 5);
553 if (rc && cli->cl_seq) {
554 OBD_FREE_PTR(cli->cl_seq);
557 up_write(&cli->cl_seq_rwsem);
561 EXPORT_SYMBOL(client_fid_init);
563 int client_fid_fini(struct obd_device *obd)
565 struct client_obd *cli = &obd->u.cli;
568 down_write(&cli->cl_seq_rwsem);
570 seq_client_fini(cli->cl_seq);
571 OBD_FREE_PTR(cli->cl_seq);
574 up_write(&cli->cl_seq_rwsem);
578 EXPORT_SYMBOL(client_fid_fini);
580 static int __init fid_init(void)
583 #ifdef HAVE_SERVER_SUPPORT
584 int rc = fid_server_mod_init();
589 de = ldebugfs_register(LUSTRE_SEQ_NAME,
593 seq_debugfs_dir = de;
594 return PTR_ERR_OR_ZERO(de);
597 static void __exit fid_exit(void)
599 # ifdef HAVE_SERVER_SUPPORT
600 fid_server_mod_exit();
602 debugfs_remove_recursive(seq_debugfs_dir);
605 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
606 MODULE_DESCRIPTION("Lustre File IDentifier");
607 MODULE_VERSION(LUSTRE_VERSION_STRING);
608 MODULE_LICENSE("GPL");
610 module_init(fid_init);
611 module_exit(fid_exit);