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) 2010, 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/mgs/mgs_handler.c
34 * Author: Nathan Rutman <nathan@clusterfs.com>
35 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
36 * Author: Mikhail Pershin <tappro@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_MGS
40 #define D_MGS D_CONFIG
42 #include <obd_class.h>
43 #include <lprocfs_status.h>
44 #include <uapi/linux/lustre/lustre_ioctl.h>
45 #include <uapi/linux/lustre/lustre_param.h>
46 #include <uapi/linux/lustre/lustre_barrier_user.h>
48 #include "mgs_internal.h"
51 * Regular MGS handlers
53 static int mgs_connect(struct tgt_session_info *tsi)
55 struct ptlrpc_request *req = tgt_ses_req(tsi);
60 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_CONNECT_NET, cfs_fail_val);
61 rc = tgt_connect(tsi);
65 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
66 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
71 static int mgs_disconnect(struct tgt_session_info *tsi)
77 LASSERT(tsi->tsi_exp);
79 rc = tgt_disconnect(tsi);
81 RETURN(err_serious(rc));
85 static int mgs_exception(struct tgt_session_info *tsi)
89 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
94 static inline bool str_starts_with(const char *str, const char *prefix)
96 return strncmp(str, prefix, strlen(prefix)) == 0;
99 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
100 static int mgs_set_info(struct tgt_session_info *tsi)
102 struct mgs_thread_info *mgi;
103 struct mgs_send_param *msp, *rep_msp;
104 struct lustre_cfg *lcfg;
111 mgi = mgs_env_info(tsi->tsi_env);
113 RETURN(err_serious(PTR_ERR(mgi)));
115 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
117 RETURN(err_serious(-EFAULT));
119 param_len = strnlen(msp->mgs_param, sizeof(msp->mgs_param));
120 if (param_len == 0 || param_len == sizeof(msp->mgs_param))
123 /* We only allow '*.lov.stripe{size,count,offset}=*' from an RPC. */
124 s = strchr(msp->mgs_param, '.');
128 if (!str_starts_with(s + 1, "lov.stripesize=") &&
129 !str_starts_with(s + 1, "lov.stripecount=") &&
130 !str_starts_with(s + 1, "lov.stripeoffset="))
133 /* Construct lustre_cfg structure to pass to function mgs_set_param */
134 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
135 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
136 OBD_ALLOC(lcfg, lustre_cfg_len(mgi->mgi_bufs.lcfg_bufcount,
137 mgi->mgi_bufs.lcfg_buflen));
140 lustre_cfg_init(lcfg, LCFG_PARAM, &mgi->mgi_bufs);
142 rc = mgs_set_param(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg);
144 LCONSOLE_WARN("%s: Unable to set parameter %s: %d\n",
145 tgt_name(tsi->tsi_tgt), msp->mgs_param, rc);
149 /* send back the whole msp in the reply */
150 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
154 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
166 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
167 void *cbdata, enum ast_type type)
171 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
174 /* l_ast_data is used as a marker to avoid cancel ldlm lock
175 * twice. See LU-2317. */
176 lock_res_and_lock(lock);
177 fsdb = (struct fs_db *)lock->l_ast_data;
178 lock->l_ast_data = NULL;
179 unlock_res_and_lock(lock);
182 struct lustre_handle lockh;
186 /* clear the bit before lock put */
187 clear_bit(FSDB_REVOKING_LOCK,
191 clear_bit(FSDB_REVOKING_PARAMS,
195 mgs_ir_notify_complete(fsdb);
203 ldlm_lock2handle(lock, &lockh);
204 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
208 RETURN(ldlm_completion_ast(lock, flags, cbdata));
211 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
214 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
217 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
220 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
223 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
226 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
229 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
232 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
235 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
237 ldlm_completion_callback cp = NULL;
238 struct lustre_handle lockh = {
241 struct ldlm_res_id res_id;
242 __u64 flags = LDLM_FL_ATOMIC_CB;
246 LASSERT(fsdb->fsdb_name[0] != '\0');
247 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
250 case CONFIG_T_CONFIG:
251 case CONFIG_T_NODEMAP:
252 cp = mgs_completion_ast_config;
253 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
256 case CONFIG_T_PARAMS:
257 cp = mgs_completion_ast_params;
258 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
261 case CONFIG_T_RECOVER:
262 cp = mgs_completion_ast_ir;
264 case CONFIG_T_BARRIER:
265 cp = mgs_completion_ast_barrier;
273 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
274 &res_id, LDLM_PLAIN, NULL, LCK_EX,
275 &flags, ldlm_blocking_ast, cp,
276 NULL, fsdb, 0, LVB_T_NONE, NULL,
278 if (rc != ELDLM_OK) {
279 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
280 mgs->mgs_obd->obd_name,
281 le64_to_cpu(res_id.name[0]),
282 le64_to_cpu(res_id.name[1]), rc);
284 if (type == CONFIG_T_CONFIG)
285 clear_bit(FSDB_REVOKING_LOCK,
288 if (type == CONFIG_T_PARAMS)
289 clear_bit(FSDB_REVOKING_PARAMS,
292 /* lock has been cancelled in completion_ast. */
301 static int mgs_check_target(const struct lu_env *env,
302 struct mgs_device *mgs,
303 struct mgs_target_info *mti)
308 rc = mgs_check_index(env, mgs, mti);
310 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
311 "this MGS does not know about it, preventing "
312 "registration.\n", mti->mti_svname);
314 } else if (rc == -1) {
315 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
316 "disappeared! Regenerating all logs.\n",
318 mti->mti_flags |= LDD_F_WRITECONF;
321 /* Index is correctly marked as used */
328 /* Ensure this is not a failover node that is connecting first*/
329 static int mgs_check_failover_reg(struct mgs_target_info *mti)
335 ptr = mti->mti_params;
336 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
337 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
338 for (i = 0; i < mti->mti_nid_count; i++) {
339 if (nid == mti->mti_nids[i]) {
340 LCONSOLE_WARN("Denying initial registration attempt from nid %s, specified as failover\n",
341 libcfs_nid2str(nid));
342 return -EADDRNOTAVAIL;
350 /* Called whenever a target starts up. Flags indicate first connect, etc. */
351 static int mgs_target_reg(struct tgt_session_info *tsi)
353 struct obd_device *obd = tsi->tsi_exp->exp_obd;
354 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
355 struct mgs_target_info *mti, *rep_mti;
356 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
357 struct fs_db *c_fsdb = NULL; /* config fsdb */
358 char barrier_name[20];
364 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
366 return err_serious(rc);
368 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
370 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
372 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
373 RETURN(err_serious(-EFAULT));
376 down_read(&mgs->mgs_barrier_rwsem);
378 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
380 opc = mti->mti_flags & LDD_F_OPC_MASK;
384 if (opc == LDD_F_OPC_READY) {
385 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
386 mti->mti_fsname, mti->mti_stripe_index);
387 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
389 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
390 CERROR("%s: Update IR return failure: rc = %d\n",
391 mti->mti_fsname, rc);
394 GOTO(out_norevoke, rc);
397 /* Do not support unregistering right now. */
398 if (opc != LDD_F_OPC_REG)
399 GOTO(out_norevoke, rc = -EINVAL);
401 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
402 mti->mti_fsname, BARRIER_FILENAME);
403 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
405 CERROR("%s: Can't get db for %s: rc = %d\n",
406 mti->mti_fsname, barrier_name, rc);
408 GOTO(out_norevoke, rc);
411 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
412 mti->mti_fsname, mti->mti_stripe_index);
414 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
415 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
416 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
417 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
418 LCONSOLE_WARN("%s: the system is in barrier, refuse "
419 "the connection from MDT %s temporary\n",
420 obd->obd_name, mti->mti_svname);
422 GOTO(out_norevoke, rc = -EBUSY);
425 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
426 !b_fsdb->fsdb_barrier_disabled) {
427 LCONSOLE_WARN("%s: the MDT %s does not support write "
428 "barrier, so disable barrier on the "
430 obd->obd_name, mti->mti_svname);
432 b_fsdb->fsdb_barrier_disabled = 1;
436 if (mti->mti_flags & LDD_F_NEED_INDEX)
437 mti->mti_flags |= LDD_F_WRITECONF;
439 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
440 /* We're just here as a startup ping. */
441 CDEBUG(D_MGS, "Server %s is running on %s\n",
442 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
443 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
444 /* above will set appropriate mti flags */
446 /* Nothing wrong, or fatal error */
447 GOTO(out_norevoke, rc);
448 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
449 rc = mgs_check_failover_reg(mti);
451 GOTO(out_norevoke, rc);
454 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
456 if (mti->mti_flags & LDD_F_WRITECONF) {
457 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
458 mti->mti_stripe_index == 0) {
459 mgs_put_fsdb(mgs, b_fsdb);
461 rc = mgs_erase_logs(tsi->tsi_env, mgs,
463 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
464 "request. All servers must be restarted "
465 "in order to regenerate the logs: rc = %d"
466 "\n", obd->obd_name, mti->mti_fsname, rc);
467 if (rc && rc != -ENOENT)
468 GOTO(out_norevoke, rc);
470 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
471 barrier_name, &b_fsdb);
473 CERROR("Can't get db for %s: %d\n",
476 GOTO(out_norevoke, rc);
479 if (!(exp_connect_flags(tsi->tsi_exp) &
480 OBD_CONNECT_BARRIER)) {
481 LCONSOLE_WARN("%s: the MDT %s does not support "
482 "write barrier, disable barrier "
483 "on the whole system.\n",
484 obd->obd_name, mti->mti_svname);
486 b_fsdb->fsdb_barrier_disabled = 1;
488 } else if (mti->mti_flags &
489 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
490 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
491 LCONSOLE_WARN("%s: Regenerating %s log by user "
492 "request: rc = %d\n",
493 obd->obd_name, mti->mti_svname, rc);
495 GOTO(out_norevoke, rc);
498 mti->mti_flags |= LDD_F_UPDATE;
501 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
503 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
505 GOTO(out_norevoke, rc);
509 * Log writing contention is handled by the fsdb_mutex.
511 * It should be alright if someone was reading while we were
512 * updating the logs - if we revoke at the end they will just update
513 * from where they left off.
515 if (mti->mti_flags & LDD_F_UPDATE) {
516 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
517 mti->mti_stripe_index);
519 /* create or update the target log
520 and update the client/mdt logs */
521 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
523 CERROR("Failed to write %s log (%d)\n",
524 mti->mti_svname, rc);
528 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
529 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
530 mti->mti_flags |= LDD_F_REWRITE_LDD;
534 mgs_revoke_lock(mgs, c_fsdb, CONFIG_T_CONFIG);
537 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
539 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
540 mti->mti_fsname, &c_fsdb);
542 CERROR("Fail to get db for %s: %d\n",
543 mti->mti_fsname, rc);
547 memcpy(b_fsdb->fsdb_mdt_index_map,
548 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
549 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
553 up_read(&mgs->mgs_barrier_rwsem);
555 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
556 mti->mti_stripe_index, rc);
557 /* An error flag is set in the mti reply rather than an error code */
559 mti->mti_flags |= LDD_F_ERROR;
561 /* send back the whole mti in the reply */
562 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
565 /* Flush logs to disk */
566 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
568 mgs_put_fsdb(mgs, b_fsdb);
570 mgs_put_fsdb(mgs, c_fsdb);
574 /* Called whenever a target cleans up. */
575 static int mgs_target_del(struct tgt_session_info *tsi)
579 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
584 static int mgs_config_read(struct tgt_session_info *tsi)
586 struct ptlrpc_request *req = tgt_ses_req(tsi);
587 struct mgs_config_body *body;
592 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
594 DEBUG_REQ(D_HA, req, "no mgs_config_body");
595 RETURN(err_serious(-EFAULT));
598 switch (body->mcb_type) {
599 case CONFIG_T_RECOVER:
600 rc = mgs_get_ir_logs(req);
602 case CONFIG_T_NODEMAP:
603 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
605 case CONFIG_T_CONFIG:
616 static int mgs_llog_open(struct tgt_session_info *tsi)
618 struct mgs_thread_info *mgi;
619 struct ptlrpc_request *req = tgt_ses_req(tsi);
625 rc = tgt_llog_open(tsi);
630 * For old clients there is no direct way of knowing which file system
631 * a client is operating at the MGS side. But we need to pick up those
632 * clients so that the MGS can mark the corresponding file system as
633 * non-IR capable because old clients are not ready to be notified.
635 * Therefore we attempt to detect the file systems name by hacking the
636 * llog operation which is currently used by the clients to fetch
637 * configuration logs. At present this is fine because this is the
638 * ONLY llog operation between mgc and the MGS.
640 * If extra llog operation are going to be added, this function needs
643 * When releases prior than 2.0 are not supported, the following code
646 mgi = mgs_env_info(tsi->tsi_env);
648 RETURN(PTR_ERR(mgi));
650 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
652 char *ptr = strrchr(logname, '-');
653 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
655 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
656 if (strcmp(logname, PARAMS_FILENAME) != 0)
657 LCONSOLE_WARN("%s: non-config logname "
659 tgt_name(tsi->tsi_tgt),
661 /* not error, this can be llog test name */
663 strncpy(mgi->mgi_fsname, logname, len);
664 mgi->mgi_fsname[len] = 0;
666 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
668 if (rc && rc != -EEXIST) {
669 LCONSOLE_WARN("%s: Unable to add client %s "
670 "to file system %s: %d\n",
671 tgt_name(tsi->tsi_tgt),
672 libcfs_nid2str(req->rq_peer.nid),
673 mgi->mgi_fsname, rc);
679 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
685 static inline int mgs_init_export(struct obd_export *exp)
687 struct mgs_export_data *data = &exp->u.eu_mgs_data;
689 /* init mgs_export_data for fsc */
690 spin_lock_init(&data->med_lock);
691 INIT_LIST_HEAD(&data->med_clients);
693 spin_lock(&exp->exp_lock);
694 exp->exp_connecting = 1;
695 spin_unlock(&exp->exp_lock);
697 /* self-export doesn't need client data and ldlm initialization */
698 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
699 &exp->exp_client_uuid)))
701 return ldlm_init_export(exp);
704 static inline int mgs_destroy_export(struct obd_export *exp)
708 target_destroy_export(exp);
709 mgs_client_free(exp);
711 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
712 &exp->exp_client_uuid)))
715 ldlm_destroy_export(exp);
720 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
727 for (ptr = arg; *ptr != '\0'; ptr++) {
728 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
732 /* Test for fsname.poolname format.
733 * strlen() test if poolname is empty
735 ptr = strchr(arg, '.');
736 if (!ptr || !strlen(ptr))
740 /* Also make sure poolname is not to long. */
741 if (strlen(ptr) > LOV_MAXPOOLNAME)
742 return -ENAMETOOLONG;
743 strncpy(poolname, ptr, LOV_MAXPOOLNAME);
745 /* Test if fsname is empty */
746 len = strlen(arg) - strlen(ptr) - 1;
751 if (len > LUSTRE_MAXFSNAME)
752 return -ENAMETOOLONG;
754 strncpy(fsname, arg, len);
759 static int mgs_iocontrol_nodemap(const struct lu_env *env,
760 struct mgs_device *mgs,
761 struct obd_ioctl_data *data)
763 struct lustre_cfg *lcfg = NULL;
766 const char *nodemap_name = NULL;
767 const char *nidstr = NULL;
768 const char *client_idstr = NULL;
769 const char *idtype_str = NULL;
772 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
774 unsigned long client_id;
781 if (data->ioc_type != LUSTRE_CFG_TYPE) {
782 CERROR("%s: unknown cfg record type: %d\n",
783 mgs->mgs_obd->obd_name, data->ioc_type);
784 GOTO(out, rc = -EINVAL);
787 if (data->ioc_plen1 > PAGE_SIZE)
788 GOTO(out, rc = -E2BIG);
790 OBD_ALLOC(lcfg, data->ioc_plen1);
792 GOTO(out, rc = -ENOMEM);
794 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
795 GOTO(out_lcfg, rc = -EFAULT);
797 cmd = lcfg->lcfg_command;
800 case LCFG_NODEMAP_ACTIVATE:
801 if (lcfg->lcfg_bufcount != 2)
802 GOTO(out_lcfg, rc = -EINVAL);
803 param = lustre_cfg_string(lcfg, 1);
804 if (strcmp(param, "1") == 0)
809 case LCFG_NODEMAP_ADD:
810 case LCFG_NODEMAP_DEL:
811 if (lcfg->lcfg_bufcount != 2)
812 GOTO(out_lcfg, rc = -EINVAL);
813 nodemap_name = lustre_cfg_string(lcfg, 1);
814 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
816 case LCFG_NODEMAP_TEST_NID:
817 if (lcfg->lcfg_bufcount != 2)
818 GOTO(out_lcfg, rc = -EINVAL);
819 nidstr = lustre_cfg_string(lcfg, 1);
820 nid = libcfs_str2nid(nidstr);
821 nodemap_test_nid(nid, name_buf, sizeof(name_buf));
822 rc = copy_to_user(data->ioc_pbuf1, name_buf,
823 MIN(data->ioc_plen1, sizeof(name_buf)));
825 GOTO(out_lcfg, rc = -EFAULT);
827 case LCFG_NODEMAP_TEST_ID:
828 if (lcfg->lcfg_bufcount != 4)
829 GOTO(out_lcfg, rc = -EINVAL);
830 nidstr = lustre_cfg_string(lcfg, 1);
831 idtype_str = lustre_cfg_string(lcfg, 2);
832 client_idstr = lustre_cfg_string(lcfg, 3);
834 nid = libcfs_str2nid(nidstr);
835 if (strcmp(idtype_str, "uid") == 0)
836 idtype = NODEMAP_UID;
838 idtype = NODEMAP_GID;
840 rc = kstrtoul(client_idstr, 10, &client_id);
842 GOTO(out_lcfg, rc = -EINVAL);
844 rc = nodemap_test_id(nid, idtype, client_id, &fs_id);
846 GOTO(out_lcfg, rc = -EINVAL);
848 if (data->ioc_plen1 < sizeof(fs_idstr))
849 GOTO(out_lcfg, rc = -EINVAL);
851 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
852 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
853 sizeof(fs_idstr)) != 0)
854 GOTO(out_lcfg, rc = -EINVAL);
856 case LCFG_NODEMAP_ADD_RANGE:
857 case LCFG_NODEMAP_DEL_RANGE:
858 case LCFG_NODEMAP_ADD_UIDMAP:
859 case LCFG_NODEMAP_DEL_UIDMAP:
860 case LCFG_NODEMAP_ADD_GIDMAP:
861 case LCFG_NODEMAP_DEL_GIDMAP:
862 case LCFG_NODEMAP_SET_FILESET:
863 case LCFG_NODEMAP_SET_SEPOL:
864 if (lcfg->lcfg_bufcount != 3)
865 GOTO(out_lcfg, rc = -EINVAL);
866 nodemap_name = lustre_cfg_string(lcfg, 1);
867 param = lustre_cfg_string(lcfg, 2);
868 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
870 case LCFG_NODEMAP_ADMIN:
871 case LCFG_NODEMAP_TRUSTED:
872 case LCFG_NODEMAP_DENY_UNKNOWN:
873 case LCFG_NODEMAP_SQUASH_UID:
874 case LCFG_NODEMAP_SQUASH_GID:
875 case LCFG_NODEMAP_MAP_MODE:
876 case LCFG_NODEMAP_AUDIT_MODE:
877 if (lcfg->lcfg_bufcount != 4)
878 GOTO(out_lcfg, rc = -EINVAL);
879 nodemap_name = lustre_cfg_string(lcfg, 1);
880 param = lustre_cfg_string(lcfg, 3);
881 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
888 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
889 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
894 /* revoke nodemap lock */
895 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
897 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
898 mgs->mgs_obd->obd_name, rc);
900 mgs_revoke_lock(mgs, fsdb, CONFIG_T_NODEMAP);
901 mgs_put_fsdb(mgs, fsdb);
905 OBD_FREE(lcfg, data->ioc_plen1);
910 static int mgs_iocontrol_pool(const struct lu_env *env,
911 struct mgs_device *mgs,
912 struct obd_ioctl_data *data)
914 struct mgs_thread_info *mgi = mgs_env_info(env);
916 struct lustre_cfg *lcfg = NULL;
917 char *poolname = NULL;
920 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
921 if (poolname == NULL)
924 if (data->ioc_type != LUSTRE_CFG_TYPE) {
925 CERROR("%s: unknown cfg record type: %d\n",
926 mgs->mgs_obd->obd_name, data->ioc_type);
927 GOTO(out_pool, rc = -EINVAL);
930 if (data->ioc_plen1 > PAGE_SIZE)
931 GOTO(out_pool, rc = -E2BIG);
933 OBD_ALLOC(lcfg, data->ioc_plen1);
935 GOTO(out_pool, rc = -ENOMEM);
937 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
938 GOTO(out_lcfg, rc = -EFAULT);
940 if (lcfg->lcfg_bufcount < 2)
941 GOTO(out_lcfg, rc = -EFAULT);
943 /* first arg is always <fsname>.<poolname> */
944 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
949 switch (lcfg->lcfg_command) {
951 if (lcfg->lcfg_bufcount != 2)
952 GOTO(out_lcfg, rc = -EINVAL);
953 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
957 if (lcfg->lcfg_bufcount != 3)
958 GOTO(out_lcfg, rc = -EINVAL);
959 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
960 poolname, lustre_cfg_string(lcfg, 2));
963 if (lcfg->lcfg_bufcount != 3)
964 GOTO(out_lcfg, rc = -EINVAL);
965 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
966 poolname, lustre_cfg_string(lcfg, 2));
969 if (lcfg->lcfg_bufcount != 2)
970 GOTO(out_lcfg, rc = -EINVAL);
971 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
979 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
980 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
985 OBD_FREE(lcfg, data->ioc_plen1);
987 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
991 /* from mdt_iocontrol */
992 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
993 void *karg, void __user *uarg)
995 struct mgs_device *mgs = exp2mgs_dev(exp);
996 struct obd_ioctl_data *data = karg;
1001 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1003 rc = lu_env_init(&env, LCT_MG_THREAD);
1009 case OBD_IOC_PARAM: {
1010 struct lustre_cfg *lcfg;
1012 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1013 CERROR("%s: unknown cfg record type: %d\n",
1014 mgs->mgs_obd->obd_name, data->ioc_type);
1015 GOTO(out, rc = -EINVAL);
1018 OBD_ALLOC(lcfg, data->ioc_plen1);
1020 GOTO(out, rc = -ENOMEM);
1021 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1022 GOTO(out_free, rc = -EFAULT);
1024 if (lcfg->lcfg_bufcount < 1)
1025 GOTO(out_free, rc = -EINVAL);
1027 rc = mgs_set_param(&env, mgs, lcfg);
1029 CERROR("%s: setparam err: rc = %d\n",
1030 exp->exp_obd->obd_name, rc);
1032 OBD_FREE(lcfg, data->ioc_plen1);
1036 case OBD_IOC_REPLACE_NIDS: {
1037 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1039 CERROR("%s: no device or fsname specified: rc = %d\n",
1040 exp->exp_obd->obd_name, rc);
1044 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1046 CERROR("%s: device or fsname is too long: rc = %d\n",
1047 exp->exp_obd->obd_name, rc);
1051 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1053 CERROR("%s: device or fsname is not NUL terminated: "
1054 "rc = %d\n", exp->exp_obd->obd_name, rc);
1058 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1060 CERROR("%s: no NIDs specified: rc = %d\n",
1061 exp->exp_obd->obd_name, rc);
1065 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1067 CERROR("%s: NID list is not NUL terminated: "
1068 "rc = %d\n", exp->exp_obd->obd_name, rc);
1072 /* replace nids in llog */
1073 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1076 CERROR("%s: error replacing nids: rc = %d\n",
1077 exp->exp_obd->obd_name, rc);
1082 case OBD_IOC_CLEAR_CONFIGS: {
1083 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1085 CERROR("%s: no device or fsname specified: rc = %d\n",
1086 exp->exp_obd->obd_name, rc);
1090 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1092 CERROR("%s: device or fsname is too long: rc = %d\n",
1093 exp->exp_obd->obd_name, rc);
1097 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1099 CERROR("%s: device or fsname is not NUL terminated: "
1100 "rc = %d\n", exp->exp_obd->obd_name, rc);
1104 /* remove records marked SKIP from config logs */
1105 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1107 CERROR("%s: error clearing config log: rc = %d\n",
1108 exp->exp_obd->obd_name, rc);
1114 rc = mgs_iocontrol_pool(&env, mgs, data);
1117 case OBD_IOC_BARRIER:
1118 rc = mgs_iocontrol_barrier(&env, mgs, data);
1121 case OBD_IOC_NODEMAP:
1122 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1125 case OBD_IOC_LCFG_FORK:
1126 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1130 case OBD_IOC_LCFG_ERASE:
1131 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1134 case OBD_IOC_CATLOGLIST:
1135 rc = mgs_list_logs(&env, mgs, data);
1137 case OBD_IOC_LLOG_CANCEL:
1138 case OBD_IOC_LLOG_REMOVE:
1139 case OBD_IOC_LLOG_CHECK:
1140 case OBD_IOC_LLOG_INFO:
1141 case OBD_IOC_LLOG_PRINT: {
1142 struct llog_ctxt *ctxt;
1144 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1145 rc = llog_ioctl(&env, ctxt, cmd, data);
1146 llog_ctxt_put(ctxt);
1151 CERROR("%s: unknown command %#x\n",
1152 mgs->mgs_obd->obd_name, cmd);
1161 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1163 struct obd_connect_data *data = NULL;
1164 struct obd_device *obd;
1168 OBD_ALLOC_PTR(data);
1172 obd = class_name2obd(nextdev);
1174 CERROR("can't locate next device: %s\n", nextdev);
1175 GOTO(out, rc = -ENOTCONN);
1178 data->ocd_version = LUSTRE_VERSION_CODE;
1180 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1181 &obd->obd_uuid, data, NULL);
1183 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1187 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1188 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1189 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1195 static struct tgt_handler mgs_mgs_handlers[] = {
1196 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1197 0, MGS_CONNECT, mgs_connect,
1198 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1199 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1200 0, MGS_DISCONNECT, mgs_disconnect,
1201 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1202 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1203 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
1204 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_SET_INFO, mgs_set_info),
1206 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1207 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1208 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1211 static struct tgt_handler mgs_obd_handlers[] = {
1212 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1215 static struct tgt_handler mgs_dlm_handlers[] = {
1216 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1217 .th_name = "LDLM_ENQUEUE",
1218 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1219 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1220 * * is safe. If such tests will be needed we have to distinguish
1221 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1222 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1224 .th_opc = LDLM_ENQUEUE,
1225 .th_flags = HAS_KEY,
1226 .th_act = tgt_enqueue,
1227 .th_fmt = &RQF_LDLM_ENQUEUE,
1228 .th_version = LUSTRE_DLM_VERSION,
1232 static struct tgt_handler mgs_llog_handlers[] = {
1233 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1234 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1235 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1236 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1239 static struct tgt_opc_slice mgs_common_slice[] = {
1241 .tos_opc_start = MGS_FIRST_OPC,
1242 .tos_opc_end = MGS_LAST_OPC,
1243 .tos_hs = mgs_mgs_handlers
1246 .tos_opc_start = OBD_FIRST_OPC,
1247 .tos_opc_end = OBD_LAST_OPC,
1248 .tos_hs = mgs_obd_handlers
1251 .tos_opc_start = LDLM_FIRST_OPC,
1252 .tos_opc_end = LDLM_LAST_OPC,
1253 .tos_hs = mgs_dlm_handlers
1256 .tos_opc_start = LLOG_FIRST_OPC,
1257 .tos_opc_end = LLOG_LAST_OPC,
1258 .tos_hs = mgs_llog_handlers
1261 .tos_opc_start = SEC_FIRST_OPC,
1262 .tos_opc_end = SEC_LAST_OPC,
1263 .tos_hs = tgt_sec_ctx_handlers
1270 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1271 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1273 struct ptlrpc_service_conf conf;
1274 struct obd_device *obd;
1275 struct lustre_mount_info *lmi;
1276 struct llog_ctxt *ctxt;
1281 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1285 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1287 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1291 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1294 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1296 obd->u.obt.obt_magic = OBT_MAGIC;
1297 obd->u.obt.obt_instance = 0;
1299 /* namespace for mgs llog */
1300 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1301 LDLM_NAMESPACE_SERVER,
1302 LDLM_NAMESPACE_MODEST,
1304 if (obd->obd_namespace == NULL)
1305 GOTO(err_ops, rc = -ENOMEM);
1307 /* No recovery for MGCs */
1308 obd->obd_replayable = 0;
1310 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1311 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1312 OBD_FAIL_MGS_ALL_REPLY_NET);
1316 rc = mgs_fs_setup(env, mgs);
1318 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1323 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1324 obd, &llog_osd_ops);
1328 /* XXX: we need this trick till N:1 stack is supported
1329 * set "current" directory for named llogs */
1330 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1332 ctxt->loc_dir = mgs->mgs_configs_dir;
1333 llog_ctxt_put(ctxt);
1335 /* Internal mgs setup */
1336 mgs_init_fsdb_list(mgs);
1337 mutex_init(&mgs->mgs_mutex);
1338 mgs->mgs_start_time = ktime_get_real_seconds();
1339 spin_lock_init(&mgs->mgs_lock);
1340 mutex_init(&mgs->mgs_health_mutex);
1341 init_rwsem(&mgs->mgs_barrier_rwsem);
1343 rc = mgs_lcfg_rename(env, mgs);
1347 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1349 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1354 /* Setup params fsdb and log, so that other servers can make a local
1355 * copy successfully when they are mounted. See LU-4783 */
1356 rc = mgs_params_fsdb_setup(env, mgs);
1358 /* params fsdb and log can be setup later */
1359 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1360 obd->obd_name, PARAMS_FILENAME, rc);
1362 /* Setup _mgs fsdb, useful for srpc */
1363 mgs__mgs_fsdb_setup(env, mgs);
1365 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1366 "mgs_ldlm_client", &obd->obd_ldlm_client);
1368 conf = (typeof(conf)) {
1369 .psc_name = LUSTRE_MGS_NAME,
1370 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1372 .bc_nbufs = MGS_NBUFS,
1373 .bc_buf_size = MGS_BUFSIZE,
1374 .bc_req_max_size = MGS_MAXREQSIZE,
1375 .bc_rep_max_size = MGS_MAXREPSIZE,
1376 .bc_req_portal = MGS_REQUEST_PORTAL,
1377 .bc_rep_portal = MGC_REPLY_PORTAL,
1380 .tc_thr_name = "ll_mgs",
1381 .tc_nthrs_init = MGS_NTHRS_INIT,
1382 .tc_nthrs_max = MGS_NTHRS_MAX,
1383 .tc_ctx_tags = LCT_MG_THREAD,
1386 .so_req_handler = tgt_request_handle,
1387 .so_req_printer = target_print_req,
1391 /* Start the service threads */
1392 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1393 obd->obd_debugfs_entry);
1394 if (IS_ERR(mgs->mgs_service)) {
1395 rc = PTR_ERR(mgs->mgs_service);
1396 CERROR("failed to start mgs service: %d\n", rc);
1397 mgs->mgs_service = NULL;
1398 GOTO(err_lproc, rc);
1401 ping_evictor_start();
1403 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1405 /* device stack is not yet fully setup to keep no objects behind */
1406 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1409 mgs_params_fsdb_cleanup(env, mgs);
1410 lproc_mgs_cleanup(mgs);
1412 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1414 ctxt->loc_dir = NULL;
1415 llog_cleanup(env, ctxt);
1418 tgt_fini(env, &mgs->mgs_lut);
1420 /* No extra cleanup needed for llog_init_commit_thread() */
1421 mgs_fs_cleanup(env, mgs);
1423 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1424 obd->obd_namespace = NULL;
1426 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1427 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1428 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1429 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1432 obd_disconnect(mgs->mgs_bottom_exp);
1435 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1439 static struct lu_device *mgs_device_free(const struct lu_env *env,
1440 struct lu_device *lu)
1442 struct mgs_device *mgs = lu2mgs_dev(lu);
1445 dt_device_fini(&mgs->mgs_dt_dev);
1450 static int mgs_process_config(const struct lu_env *env,
1451 struct lu_device *dev,
1452 struct lustre_cfg *lcfg)
1458 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1459 const struct lu_object_conf *unused)
1461 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1462 struct lu_device *under;
1463 struct lu_object *below;
1467 /* do no set .do_ops as mgs calls to bottom osd directly */
1469 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1470 PFID(lu_object_fid(o)));
1472 under = &d->mgs_bottom->dd_lu_dev;
1473 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1475 lu_object_add(o, below);
1482 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1484 struct mgs_object *obj = lu2mgs_obj(o);
1485 struct lu_object_header *h = o->lo_header;
1487 dt_object_fini(&obj->mgo_obj);
1488 lu_object_header_fini(h);
1492 static int mgs_object_print(const struct lu_env *env, void *cookie,
1493 lu_printer_t p, const struct lu_object *l)
1495 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1497 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1500 static struct lu_object_operations mgs_lu_obj_ops = {
1501 .loo_object_init = mgs_object_init,
1502 .loo_object_free = mgs_object_free,
1503 .loo_object_print = mgs_object_print,
1506 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1507 const struct lu_object_header *hdr,
1508 struct lu_device *d)
1510 struct lu_object_header *h;
1511 struct mgs_object *o;
1512 struct lu_object *l;
1514 LASSERT(hdr == NULL);
1518 l = &o->mgo_obj.do_lu;
1521 lu_object_header_init(h);
1522 dt_object_init(&o->mgo_obj, h, d);
1523 lu_object_add_top(h, l);
1525 l->lo_ops = &mgs_lu_obj_ops;
1533 const struct lu_device_operations mgs_lu_ops = {
1534 .ldo_object_alloc = mgs_object_alloc,
1535 .ldo_process_config = mgs_process_config,
1538 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1539 struct lu_device_type *type,
1540 struct lustre_cfg *lcfg)
1542 struct mgs_device *mgs;
1543 struct lu_device *ludev;
1547 ludev = ERR_PTR(-ENOMEM);
1551 ludev = mgs2lu_dev(mgs);
1552 dt_device_init(&mgs->mgs_dt_dev, type);
1553 rc = mgs_init0(env, mgs, type, lcfg);
1555 mgs_device_free(env, ludev);
1556 ludev = ERR_PTR(rc);
1562 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1563 struct lu_device *d)
1565 struct mgs_device *mgs = lu2mgs_dev(d);
1566 struct obd_device *obd = mgs->mgs_obd;
1567 struct llog_ctxt *ctxt;
1571 LASSERT(mgs->mgs_bottom);
1573 class_disconnect_exports(obd);
1575 ping_evictor_stop();
1577 mutex_lock(&mgs->mgs_health_mutex);
1578 ptlrpc_unregister_service(mgs->mgs_service);
1579 mutex_unlock(&mgs->mgs_health_mutex);
1581 mgs_params_fsdb_cleanup(env, mgs);
1582 mgs_cleanup_fsdb_list(mgs);
1584 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1585 obd_exports_barrier(obd);
1586 obd_zombie_barrier();
1588 tgt_fini(env, &mgs->mgs_lut);
1589 lproc_mgs_cleanup(mgs);
1591 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1593 ctxt->loc_dir = NULL;
1594 llog_cleanup(env, ctxt);
1597 mgs_fs_cleanup(env, mgs);
1599 ldlm_namespace_free_post(obd->obd_namespace);
1600 obd->obd_namespace = NULL;
1602 lu_site_purge(env, d->ld_site, ~0);
1603 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1604 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1605 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1608 LASSERT(mgs->mgs_bottom_exp);
1609 obd_disconnect(mgs->mgs_bottom_exp);
1611 server_put_mount(obd->obd_name, true);
1616 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1617 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1619 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1621 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1623 static struct lu_device_type_operations mgs_device_type_ops = {
1624 .ldto_init = mgs_type_init,
1625 .ldto_fini = mgs_type_fini,
1627 .ldto_start = mgs_type_start,
1628 .ldto_stop = mgs_type_stop,
1630 .ldto_device_alloc = mgs_device_alloc,
1631 .ldto_device_free = mgs_device_free,
1633 .ldto_device_fini = mgs_device_fini
1636 static struct lu_device_type mgs_device_type = {
1637 .ldt_tags = LU_DEVICE_DT,
1638 .ldt_name = LUSTRE_MGS_NAME,
1639 .ldt_ops = &mgs_device_type_ops,
1640 .ldt_ctx_tags = LCT_MG_THREAD
1643 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1644 struct obd_device *obd, struct obd_uuid *cluuid,
1645 struct obd_connect_data *data, void *localdata)
1649 if (exp == NULL || obd == NULL || cluuid == NULL)
1652 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1655 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1657 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1658 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1660 exp->exp_connect_data = *data;
1661 data->ocd_version = LUSTRE_VERSION_CODE;
1664 RETURN(mgs_export_stats_init(obd, exp, localdata));
1667 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1668 struct obd_device *obd, struct obd_uuid *cluuid,
1669 struct obd_connect_data *data, void *localdata)
1671 struct obd_export *lexp;
1672 struct lustre_handle conn = {
1679 if (exp == NULL || obd == NULL || cluuid == NULL)
1682 rc = class_connect(&conn, obd, cluuid);
1686 lexp = class_conn2export(&conn);
1690 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1692 GOTO(out_disconnect, rc);
1699 class_disconnect(lexp);
1704 static int mgs_obd_disconnect(struct obd_export *exp)
1712 mgs_fsc_cleanup(exp);
1714 class_export_get(exp);
1715 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1717 rc = server_disconnect_export(exp);
1718 class_export_put(exp);
1722 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1724 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1727 mutex_lock(&mgs->mgs_health_mutex);
1728 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1729 mutex_unlock(&mgs->mgs_health_mutex);
1731 return rc != 0 ? 1 : 0;
1734 /* use obd ops to offer management infrastructure */
1735 static const struct obd_ops mgs_obd_device_ops = {
1736 .o_owner = THIS_MODULE,
1737 .o_connect = mgs_obd_connect,
1738 .o_reconnect = mgs_obd_reconnect,
1739 .o_disconnect = mgs_obd_disconnect,
1740 .o_init_export = mgs_init_export,
1741 .o_destroy_export = mgs_destroy_export,
1742 .o_iocontrol = mgs_iocontrol,
1743 .o_health_check = mgs_health_check,
1746 static int __init mgs_init(void)
1748 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1749 LUSTRE_MGS_NAME, &mgs_device_type);
1752 static void __exit mgs_exit(void)
1754 class_unregister_type(LUSTRE_MGS_NAME);
1757 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1758 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1759 MODULE_VERSION(LUSTRE_VERSION_STRING);
1760 MODULE_LICENSE("GPL");
1762 module_init(mgs_init);
1763 module_exit(mgs_exit);