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/
31 * lustre/fid/fid_request.c
33 * Lustre Sequence Manager
35 * Author: Yury Umanets <umka@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_FID
40 #include <linux/err.h>
41 #include <linux/module.h>
42 #include <linux/delay.h>
44 #include <obd_class.h>
45 #include <obd_support.h>
46 #include <lustre_fid.h>
48 #include <lustre_mdc.h>
49 #include "fid_internal.h"
51 struct dentry *seq_debugfs_dir;
53 static int seq_client_rpc(struct lu_client_seq *seq,
54 struct lu_seq_range *output, __u32 opc,
57 struct obd_export *exp = seq->lcs_exp;
58 struct ptlrpc_request *req;
59 struct lu_seq_range *out, *in;
61 unsigned int debug_mask;
65 LASSERT(exp != NULL && !IS_ERR(exp));
66 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
67 LUSTRE_MDS_VERSION, SEQ_QUERY);
71 /* Init operation code */
72 op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
75 /* Zero out input range, this is not recovery yet. */
76 in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
77 lu_seq_range_init(in);
79 ptlrpc_request_set_replen(req);
81 in->lsr_index = seq->lcs_space.lsr_index;
82 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
83 fld_range_set_mdt(in);
85 fld_range_set_ost(in);
87 if (opc == SEQ_ALLOC_SUPER) {
88 req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
89 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
98 if (seq->lcs_type == LUSTRE_SEQ_DATA) {
99 req->rq_no_resend = 1;
100 req->rq_no_delay = 1;
102 debug_mask = D_CONSOLE;
104 if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
105 req->rq_reply_portal = MDC_REPLY_PORTAL;
106 req->rq_request_portal = SEQ_METADATA_PORTAL;
108 req->rq_reply_portal = OSC_REPLY_PORTAL;
109 req->rq_request_portal = SEQ_DATA_PORTAL;
115 /* Allow seq client RPC during recovery time. */
116 req->rq_allow_replay = 1;
118 ptlrpc_at_set_req_timeout(req);
120 rc = ptlrpc_queue_wait(req);
125 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
128 if (!lu_seq_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 (lu_seq_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, env);
167 * Check whether the connection to seq controller has been
168 * setup (lcs_exp != NULL)
171 mutex_unlock(&seq->lcs_mutex);
172 RETURN(-EINPROGRESS);
175 rc = seq_client_rpc(seq, &seq->lcs_space,
176 SEQ_ALLOC_SUPER, "super");
178 mutex_unlock(&seq->lcs_mutex);
182 /* Request sequence-controller node to allocate new meta-sequence. */
183 static int seq_client_alloc_meta(const struct lu_env *env,
184 struct lu_client_seq *seq)
190 #ifdef HAVE_SEQ_SERVER
192 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
199 * If meta server return -EINPROGRESS or EAGAIN,
200 * it means meta server might not be ready to
201 * allocate super sequence from sequence controller
204 rc = seq_client_rpc(seq, &seq->lcs_space,
205 SEQ_ALLOC_META, "meta");
206 if (rc == -EINPROGRESS || rc == -EAGAIN)
208 * MDT0 is not ready, let's wait for 2
213 } while (rc == -EINPROGRESS || rc == -EAGAIN);
219 /* Allocate new sequence for client. */
220 static int seq_client_alloc_seq(const struct lu_env *env,
221 struct lu_client_seq *seq, u64 *seqnr)
226 LASSERT(lu_seq_range_is_sane(&seq->lcs_space));
228 if (lu_seq_range_is_exhausted(&seq->lcs_space)) {
229 rc = seq_client_alloc_meta(env, seq);
231 if (rc != -EINPROGRESS)
232 CERROR("%s: Cannot allocate new meta-sequence: rc = %d\n",
236 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
237 seq->lcs_name, PRANGE(&seq->lcs_space));
243 LASSERT(!lu_seq_range_is_exhausted(&seq->lcs_space));
244 *seqnr = seq->lcs_space.lsr_start;
245 seq->lcs_space.lsr_start += 1;
247 CDEBUG(D_INFO, "%s: Allocated sequence [%#llx]\n", seq->lcs_name,
254 * Allocate the whole non-used seq to the caller.
256 * \param[in] env pointer to the thread context
257 * \param[in,out] seq pointer to the client sequence manager
258 * \param[out] seqnr to hold the new allocated sequence
260 * \retval 0 for new sequence allocated.
261 * \retval Negative error number on failure.
263 int seq_client_get_seq(const struct lu_env *env,
264 struct lu_client_seq *seq, u64 *seqnr)
268 LASSERT(seqnr != NULL);
270 mutex_lock(&seq->lcs_mutex);
272 rc = seq_client_alloc_seq(env, seq, seqnr);
274 CERROR("%s: Can't allocate new sequence: rc = %d\n",
277 CDEBUG(D_INFO, "%s: New sequence [0x%16.16llx]\n",
278 seq->lcs_name, *seqnr);
279 seq->lcs_fid.f_seq = *seqnr;
280 seq->lcs_fid.f_ver = 0;
282 * The caller require the whole seq,
283 * so marked this seq to be used
285 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
287 LUSTRE_METADATA_SEQ_MAX_WIDTH;
289 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
291 mutex_unlock(&seq->lcs_mutex);
295 EXPORT_SYMBOL(seq_client_get_seq);
298 * Allocate new fid on passed client @seq and save it to @fid.
300 * \param[in] env pointer to the thread context
301 * \param[in,out] seq pointer to the client sequence manager
302 * \param[out] fid to hold the new allocated fid
304 * \retval 1 for notify the caller that sequence switch
305 * is performed to allow it to setup FLD for it.
306 * \retval 0 for new FID allocated in current sequence.
307 * \retval Negative error number on failure.
309 int seq_client_alloc_fid(const struct lu_env *env,
310 struct lu_client_seq *seq, struct lu_fid *fid)
315 LASSERT(seq != NULL);
316 LASSERT(fid != NULL);
318 mutex_lock(&seq->lcs_mutex);
320 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
321 seq->lcs_fid.f_oid = seq->lcs_width;
323 if (unlikely(!fid_is_zero(&seq->lcs_fid) &&
324 fid_oid(&seq->lcs_fid) < seq->lcs_width)) {
325 /* Just bump last allocated fid and return to caller. */
326 seq->lcs_fid.f_oid++;
331 rc = seq_client_alloc_seq(env, seq, &seqnr);
333 if (rc != -EINPROGRESS)
334 CERROR("%s: Can't allocate new sequence: rc = %d\n",
337 CDEBUG(D_INFO, "%s: New sequence [0x%16.16llx]\n",
338 seq->lcs_name, seqnr);
340 seq->lcs_fid.f_seq = seqnr;
341 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
342 seq->lcs_fid.f_ver = 0;
349 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name,
352 mutex_unlock(&seq->lcs_mutex);
356 EXPORT_SYMBOL(seq_client_alloc_fid);
359 * Finish the current sequence due to disconnect.
360 * See mdc_import_event()
362 void seq_client_flush(struct lu_client_seq *seq)
364 LASSERT(seq != NULL);
365 mutex_lock(&seq->lcs_mutex);
367 fid_zero(&seq->lcs_fid);
369 * this id shld not be used for seq range allocation.
370 * set to -1 for dgb check.
372 seq->lcs_space.lsr_index = -1;
374 lu_seq_range_init(&seq->lcs_space);
375 mutex_unlock(&seq->lcs_mutex);
377 EXPORT_SYMBOL(seq_client_flush);
379 static void seq_client_debugfs_fini(struct lu_client_seq *seq)
381 debugfs_remove_recursive(seq->lcs_debugfs_entry);
384 static void seq_client_debugfs_init(struct lu_client_seq *seq)
386 seq->lcs_debugfs_entry = debugfs_create_dir(seq->lcs_name,
389 ldebugfs_add_vars(seq->lcs_debugfs_entry,
390 seq_client_debugfs_list, seq);
393 void seq_client_fini(struct lu_client_seq *seq)
397 seq_client_debugfs_fini(seq);
400 class_export_put(seq->lcs_exp);
407 EXPORT_SYMBOL(seq_client_fini);
409 void seq_client_init(struct lu_client_seq *seq,
410 struct obd_export *exp,
411 enum lu_cli_type type,
413 struct lu_server_seq *srv)
417 LASSERT(seq != NULL);
418 LASSERT(prefix != NULL);
421 seq->lcs_type = type;
423 mutex_init(&seq->lcs_mutex);
424 if (type == LUSTRE_SEQ_METADATA)
425 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
427 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
429 /* Make sure that things are clear before work is started. */
430 seq_client_flush(seq);
433 seq->lcs_exp = class_export_get(exp);
435 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
438 seq_client_debugfs_init(seq);
440 EXPORT_SYMBOL(seq_client_init);
442 int client_fid_init(struct obd_device *obd,
443 struct obd_export *exp, enum lu_cli_type type)
445 struct client_obd *cli = &obd->u.cli;
450 down_write(&cli->cl_seq_rwsem);
451 OBD_ALLOC_PTR(cli->cl_seq);
453 GOTO(out, rc = -ENOMEM);
455 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
457 GOTO(out, rc = -ENOMEM);
459 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
461 /* Init client side sequence-manager */
462 seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
463 OBD_FREE(prefix, MAX_OBD_NAME + 5);
466 if (rc && cli->cl_seq) {
467 OBD_FREE_PTR(cli->cl_seq);
470 up_write(&cli->cl_seq_rwsem);
474 EXPORT_SYMBOL(client_fid_init);
476 int client_fid_fini(struct obd_device *obd)
478 struct client_obd *cli = &obd->u.cli;
481 down_write(&cli->cl_seq_rwsem);
483 seq_client_fini(cli->cl_seq);
484 OBD_FREE_PTR(cli->cl_seq);
487 up_write(&cli->cl_seq_rwsem);
491 EXPORT_SYMBOL(client_fid_fini);
493 static int __init fid_init(void)
496 #ifdef HAVE_SERVER_SUPPORT
497 int rc = fid_server_mod_init();
502 de = debugfs_create_dir(LUSTRE_SEQ_NAME,
503 debugfs_lustre_root);
505 seq_debugfs_dir = de;
506 return PTR_ERR_OR_ZERO(de);
509 static void __exit fid_exit(void)
511 # ifdef HAVE_SERVER_SUPPORT
512 fid_server_mod_exit();
514 debugfs_remove_recursive(seq_debugfs_dir);
517 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
518 MODULE_DESCRIPTION("Lustre File IDentifier");
519 MODULE_VERSION(LUSTRE_VERSION_STRING);
520 MODULE_LICENSE("GPL");
522 module_init(fid_init);
523 module_exit(fid_exit);