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/module.h>
42 #include <libcfs/libcfs.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 static int seq_client_rpc(struct lu_client_seq *seq,
52 struct lu_seq_range *output, __u32 opc,
55 struct obd_export *exp = seq->lcs_exp;
56 struct ptlrpc_request *req;
57 struct lu_seq_range *out, *in;
59 unsigned int debug_mask;
63 LASSERT(exp != NULL && !IS_ERR(exp));
64 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
65 LUSTRE_MDS_VERSION, SEQ_QUERY);
69 /* Init operation code */
70 op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
73 /* Zero out input range, this is not recovery yet. */
74 in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
75 lu_seq_range_init(in);
77 ptlrpc_request_set_replen(req);
79 in->lsr_index = seq->lcs_space.lsr_index;
80 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
81 fld_range_set_mdt(in);
83 fld_range_set_ost(in);
85 if (opc == SEQ_ALLOC_SUPER) {
86 req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
87 req->rq_reply_portal = MDC_REPLY_PORTAL;
88 /* During allocating super sequence for data object,
89 * the current thread might hold the export of MDT0(MDT0
90 * precreating objects on this OST), and it will send the
91 * request to MDT0 here, so we can not keep resending the
92 * request here, otherwise if MDT0 is failed(umounted),
93 * it can not release the export of MDT0 */
94 if (seq->lcs_type == LUSTRE_SEQ_DATA)
95 req->rq_no_delay = req->rq_no_resend = 1;
96 debug_mask = D_CONSOLE;
98 if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
99 req->rq_reply_portal = MDC_REPLY_PORTAL;
100 req->rq_request_portal = SEQ_METADATA_PORTAL;
102 req->rq_reply_portal = OSC_REPLY_PORTAL;
103 req->rq_request_portal = SEQ_DATA_PORTAL;
109 /* Allow seq client RPC during recovery time. */
110 req->rq_allow_replay = 1;
112 ptlrpc_at_set_req_timeout(req);
114 rc = ptlrpc_queue_wait(req);
119 out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
122 if (!lu_seq_range_is_sane(output)) {
123 CERROR("%s: Invalid range received from server: "
124 DRANGE"\n", seq->lcs_name, PRANGE(output));
125 GOTO(out_req, rc = -EINVAL);
128 if (lu_seq_range_is_exhausted(output)) {
129 CERROR("%s: Range received from server is exhausted: "
130 DRANGE"]\n", seq->lcs_name, PRANGE(output));
131 GOTO(out_req, rc = -EINVAL);
134 CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
135 seq->lcs_name, opcname, PRANGE(output));
139 ptlrpc_req_finished(req);
143 /* Request sequence-controller node to allocate new super-sequence. */
144 int seq_client_alloc_super(struct lu_client_seq *seq,
145 const struct lu_env *env)
150 mutex_lock(&seq->lcs_mutex);
153 #ifdef HAVE_SEQ_SERVER
154 LASSERT(env != NULL);
155 rc = seq_server_alloc_super(seq->lcs_srv, &seq->lcs_space,
161 /* Check whether the connection to seq controller has been
162 * setup (lcs_exp != NULL) */
163 if (seq->lcs_exp == NULL) {
164 mutex_unlock(&seq->lcs_mutex);
165 RETURN(-EINPROGRESS);
168 rc = seq_client_rpc(seq, &seq->lcs_space,
169 SEQ_ALLOC_SUPER, "super");
171 mutex_unlock(&seq->lcs_mutex);
175 /* Request sequence-controller node to allocate new meta-sequence. */
176 static int seq_client_alloc_meta(const struct lu_env *env,
177 struct lu_client_seq *seq)
183 #ifdef HAVE_SEQ_SERVER
184 LASSERT(env != NULL);
185 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
191 /* If meta server return -EINPROGRESS or EAGAIN,
192 * it means meta server might not be ready to
193 * allocate super sequence from sequence controller
195 rc = seq_client_rpc(seq, &seq->lcs_space,
196 SEQ_ALLOC_META, "meta");
197 if (rc == -EINPROGRESS || rc == -EAGAIN) {
198 wait_queue_head_t waitq;
199 struct l_wait_info lwi;
201 /* MDT0 is not ready, let's wait for 2
202 * seconds and retry. */
203 init_waitqueue_head(&waitq);
204 lwi = LWI_TIMEOUT(cfs_time_seconds(2), NULL,
206 l_wait_event(waitq, 0, &lwi);
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, u64 *seqnr)
221 LASSERT(lu_seq_range_is_sane(&seq->lcs_space));
223 if (lu_seq_range_is_exhausted(&seq->lcs_space)) {
224 rc = seq_client_alloc_meta(env, seq);
226 if (rc != -EINPROGRESS)
227 CERROR("%s: Can't allocate new meta-sequence,"
228 "rc = %d\n", seq->lcs_name, rc);
231 CDEBUG(D_INFO, "%s: New range - "DRANGE"\n",
232 seq->lcs_name, PRANGE(&seq->lcs_space));
238 LASSERT(!lu_seq_range_is_exhausted(&seq->lcs_space));
239 *seqnr = seq->lcs_space.lsr_start;
240 seq->lcs_space.lsr_start += 1;
242 CDEBUG(D_INFO, "%s: Allocated sequence [%#llx]\n", seq->lcs_name,
248 static int seq_fid_alloc_prep(struct lu_client_seq *seq,
251 if (seq->lcs_update) {
252 add_wait_queue(&seq->lcs_waitq, link);
253 set_current_state(TASK_UNINTERRUPTIBLE);
254 mutex_unlock(&seq->lcs_mutex);
258 mutex_lock(&seq->lcs_mutex);
259 remove_wait_queue(&seq->lcs_waitq, link);
260 set_current_state(TASK_RUNNING);
265 mutex_unlock(&seq->lcs_mutex);
270 static void seq_fid_alloc_fini(struct lu_client_seq *seq, __u64 seqnr,
273 LASSERT(seq->lcs_update == 1);
275 mutex_lock(&seq->lcs_mutex);
277 CDEBUG(D_INFO, "%s: New sequence [0x%16.16llx]\n",
278 seq->lcs_name, seqnr);
280 seq->lcs_fid.f_seq = seqnr;
282 /* Since the caller require the whole seq,
283 * so marked this seq to be used */
284 if (seq->lcs_type == LUSTRE_SEQ_METADATA)
286 LUSTRE_METADATA_SEQ_MAX_WIDTH;
288 seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
290 seq->lcs_fid.f_oid = LUSTRE_FID_INIT_OID;
292 seq->lcs_fid.f_ver = 0;
296 wake_up_all(&seq->lcs_waitq);
300 * Allocate the whole non-used seq to the caller.
302 * \param[in] env pointer to the thread context
303 * \param[in,out] seq pointer to the client sequence manager
304 * \param[out] seqnr to hold the new allocated sequence
306 * \retval 0 for new sequence allocated.
307 * \retval Negative error number on failure.
309 int seq_client_get_seq(const struct lu_env *env,
310 struct lu_client_seq *seq, u64 *seqnr)
315 LASSERT(seqnr != NULL);
317 mutex_lock(&seq->lcs_mutex);
318 init_waitqueue_entry(&link, current);
320 /* To guarantee that we can get a whole non-used sequence. */
321 while (seq_fid_alloc_prep(seq, &link) != 0);
323 rc = seq_client_alloc_seq(env, seq, seqnr);
324 seq_fid_alloc_fini(seq, rc ? 0 : *seqnr, true);
326 CERROR("%s: Can't allocate new sequence: rc = %d\n",
328 mutex_unlock(&seq->lcs_mutex);
332 EXPORT_SYMBOL(seq_client_get_seq);
335 * Allocate new fid on passed client @seq and save it to @fid.
337 * \param[in] env pointer to the thread context
338 * \param[in,out] seq pointer to the client sequence manager
339 * \param[out] fid to hold the new allocated fid
341 * \retval 1 for notify the caller that sequence switch
342 * is performed to allow it to setup FLD for it.
343 * \retval 0 for new FID allocated in current sequence.
344 * \retval Negative error number on failure.
346 int seq_client_alloc_fid(const struct lu_env *env,
347 struct lu_client_seq *seq, struct lu_fid *fid)
353 LASSERT(seq != NULL);
354 LASSERT(fid != NULL);
356 init_waitqueue_entry(&link, current);
357 mutex_lock(&seq->lcs_mutex);
359 if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
360 seq->lcs_fid.f_oid = seq->lcs_width;
365 if (unlikely(!fid_is_zero(&seq->lcs_fid) &&
366 fid_oid(&seq->lcs_fid) < seq->lcs_width)) {
367 /* Just bump last allocated fid and return to caller. */
368 seq->lcs_fid.f_oid++;
373 /* Release seq::lcs_mutex via seq_fid_alloc_prep() to avoid
374 * deadlock during seq_client_alloc_seq(). */
375 rc = seq_fid_alloc_prep(seq, &link);
379 rc = seq_client_alloc_seq(env, seq, &seqnr);
380 /* Re-take seq::lcs_mutex via seq_fid_alloc_fini(). */
381 seq_fid_alloc_fini(seq, rc ? 0 : seqnr, false);
383 if (rc != -EINPROGRESS)
384 CERROR("%s: Can't allocate new sequence: "
385 "rc = %d\n", seq->lcs_name, rc);
386 mutex_unlock(&seq->lcs_mutex);
396 mutex_unlock(&seq->lcs_mutex);
398 CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name, PFID(fid));
402 EXPORT_SYMBOL(seq_client_alloc_fid);
405 * Finish the current sequence due to disconnect.
406 * See mdc_import_event()
408 void seq_client_flush(struct lu_client_seq *seq)
412 LASSERT(seq != NULL);
413 init_waitqueue_entry(&link, current);
414 mutex_lock(&seq->lcs_mutex);
416 while (seq->lcs_update) {
417 add_wait_queue(&seq->lcs_waitq, &link);
418 set_current_state(TASK_UNINTERRUPTIBLE);
419 mutex_unlock(&seq->lcs_mutex);
423 mutex_lock(&seq->lcs_mutex);
424 remove_wait_queue(&seq->lcs_waitq, &link);
425 set_current_state(TASK_RUNNING);
428 fid_zero(&seq->lcs_fid);
430 * this id shld not be used for seq range allocation.
431 * set to -1 for dgb check.
434 seq->lcs_space.lsr_index = -1;
436 lu_seq_range_init(&seq->lcs_space);
437 mutex_unlock(&seq->lcs_mutex);
439 EXPORT_SYMBOL(seq_client_flush);
441 static void seq_client_proc_fini(struct lu_client_seq *seq)
443 #ifdef CONFIG_PROC_FS
445 if (seq->lcs_proc_dir) {
446 if (!IS_ERR(seq->lcs_proc_dir))
447 lprocfs_remove(&seq->lcs_proc_dir);
448 seq->lcs_proc_dir = NULL;
451 #endif /* CONFIG_PROC_FS */
454 static int seq_client_proc_init(struct lu_client_seq *seq)
456 #ifdef CONFIG_PROC_FS
460 seq->lcs_proc_dir = lprocfs_register(seq->lcs_name, seq_type_proc_dir,
462 if (IS_ERR(seq->lcs_proc_dir)) {
463 CERROR("%s: LProcFS failed in seq-init\n",
465 rc = PTR_ERR(seq->lcs_proc_dir);
469 rc = lprocfs_add_vars(seq->lcs_proc_dir, seq_client_proc_list, seq);
471 CERROR("%s: Can't init sequence manager "
472 "proc, rc %d\n", seq->lcs_name, rc);
473 GOTO(out_cleanup, rc);
479 seq_client_proc_fini(seq);
482 #else /* !CONFIG_PROC_FS */
484 #endif /* CONFIG_PROC_FS */
487 int seq_client_init(struct lu_client_seq *seq,
488 struct obd_export *exp,
489 enum lu_cli_type type,
491 struct lu_server_seq *srv)
496 LASSERT(seq != NULL);
497 LASSERT(prefix != NULL);
500 seq->lcs_type = type;
502 mutex_init(&seq->lcs_mutex);
503 if (type == LUSTRE_SEQ_METADATA)
504 seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
506 seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
508 init_waitqueue_head(&seq->lcs_waitq);
509 /* Make sure that things are clear before work is started. */
510 seq_client_flush(seq);
513 seq->lcs_exp = class_export_get(exp);
515 snprintf(seq->lcs_name, sizeof(seq->lcs_name),
518 rc = seq_client_proc_init(seq);
520 seq_client_fini(seq);
523 EXPORT_SYMBOL(seq_client_init);
525 void seq_client_fini(struct lu_client_seq *seq)
529 seq_client_proc_fini(seq);
531 if (seq->lcs_exp != NULL) {
532 class_export_put(seq->lcs_exp);
539 EXPORT_SYMBOL(seq_client_fini);
541 int client_fid_init(struct obd_device *obd,
542 struct obd_export *exp, enum lu_cli_type type)
544 struct client_obd *cli = &obd->u.cli;
549 down_write(&cli->cl_seq_rwsem);
550 OBD_ALLOC_PTR(cli->cl_seq);
552 GOTO(out, rc = -ENOMEM);
554 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
556 GOTO(out, rc = -ENOMEM);
558 snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
560 /* Init client side sequence-manager */
561 rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
562 OBD_FREE(prefix, MAX_OBD_NAME + 5);
567 if (rc && cli->cl_seq) {
568 OBD_FREE_PTR(cli->cl_seq);
571 up_write(&cli->cl_seq_rwsem);
575 EXPORT_SYMBOL(client_fid_init);
577 int client_fid_fini(struct obd_device *obd)
579 struct client_obd *cli = &obd->u.cli;
582 down_write(&cli->cl_seq_rwsem);
584 seq_client_fini(cli->cl_seq);
585 OBD_FREE_PTR(cli->cl_seq);
588 up_write(&cli->cl_seq_rwsem);
592 EXPORT_SYMBOL(client_fid_fini);
594 struct proc_dir_entry *seq_type_proc_dir;
596 static int __init fid_init(void)
598 seq_type_proc_dir = lprocfs_register(LUSTRE_SEQ_NAME,
601 if (IS_ERR(seq_type_proc_dir))
602 return PTR_ERR(seq_type_proc_dir);
604 # ifdef HAVE_SERVER_SUPPORT
605 fid_server_mod_init();
611 static void __exit fid_exit(void)
613 # ifdef HAVE_SERVER_SUPPORT
614 fid_server_mod_exit();
617 if (seq_type_proc_dir != NULL && !IS_ERR(seq_type_proc_dir)) {
618 lprocfs_remove(&seq_type_proc_dir);
619 seq_type_proc_dir = NULL;
623 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
624 MODULE_DESCRIPTION("Lustre File IDentifier");
625 MODULE_VERSION(LUSTRE_VERSION_STRING);
626 MODULE_LICENSE("GPL");
628 module_init(fid_init);
629 module_exit(fid_exit);