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/
31 * lustre/mgs/mgs_handler.c
33 * Author: Nathan Rutman <nathan@clusterfs.com>
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mikhail Pershin <tappro@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_MGS
39 #define D_MGS D_CONFIG
41 #include <obd_class.h>
42 #include <lprocfs_status.h>
43 #include <uapi/linux/lustre/lustre_ioctl.h>
44 #include <lustre_ioctl_old.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, 18, 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;
109 mgi = mgs_env_info(tsi->tsi_env);
111 RETURN(err_serious(PTR_ERR(mgi)));
113 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
115 RETURN(err_serious(-EFAULT));
117 param_len = strnlen(msp->mgs_param, sizeof(msp->mgs_param));
118 if (param_len == 0 || param_len == sizeof(msp->mgs_param))
121 /* We only allow '*.lov.stripe{size,count,offset}=*' from an RPC. */
122 s = strchr(msp->mgs_param, '.');
126 if (!str_starts_with(s + 1, "lov.stripesize=") &&
127 !str_starts_with(s + 1, "lov.stripecount=") &&
128 !str_starts_with(s + 1, "lov.stripeoffset="))
132 CDEBUG(D_MGS, "%s: ignoring set info '%s'\n",
133 tgt_name(tsi->tsi_tgt), msp->mgs_param);
145 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
146 void *cbdata, enum ast_type type)
150 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
153 /* l_ast_data is used as a marker to avoid cancel ldlm lock
154 * twice. See LU-2317. */
155 lock_res_and_lock(lock);
156 fsdb = (struct fs_db *)lock->l_ast_data;
157 lock->l_ast_data = NULL;
158 unlock_res_and_lock(lock);
161 struct lustre_handle lockh;
165 /* clear the bit before lock put */
166 clear_bit(FSDB_REVOKING_LOCK,
170 clear_bit(FSDB_REVOKING_PARAMS,
174 mgs_ir_notify_complete(fsdb);
182 ldlm_lock2handle(lock, &lockh);
183 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
187 RETURN(ldlm_completion_ast(lock, flags, cbdata));
190 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
193 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
196 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
199 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
202 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
205 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
208 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
211 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
214 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb,
215 enum mgs_cfg_type type)
217 ldlm_completion_callback cp = NULL;
218 struct lustre_handle lockh = {
221 struct ldlm_res_id res_id;
222 __u64 flags = LDLM_FL_ATOMIC_CB;
226 LASSERT(fsdb->fsdb_name[0] != '\0');
227 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
230 case MGS_CFG_T_CONFIG:
231 case MGS_CFG_T_NODEMAP:
232 cp = mgs_completion_ast_config;
233 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
236 case MGS_CFG_T_PARAMS:
237 cp = mgs_completion_ast_params;
238 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
241 case MGS_CFG_T_RECOVER:
242 cp = mgs_completion_ast_ir;
244 case MGS_CFG_T_BARRIER:
245 cp = mgs_completion_ast_barrier;
253 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
254 &res_id, LDLM_PLAIN, NULL, LCK_EX,
255 &flags, ldlm_blocking_ast, cp,
256 NULL, fsdb, 0, LVB_T_NONE, NULL,
258 if (rc != ELDLM_OK) {
259 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
260 mgs->mgs_obd->obd_name,
261 le64_to_cpu(res_id.name[0]),
262 le64_to_cpu(res_id.name[1]), rc);
264 if (type == MGS_CFG_T_CONFIG)
265 clear_bit(FSDB_REVOKING_LOCK,
268 if (type == MGS_CFG_T_PARAMS)
269 clear_bit(FSDB_REVOKING_PARAMS,
272 /* lock has been cancelled in completion_ast. */
281 static int mgs_check_target(const struct lu_env *env,
282 struct mgs_device *mgs,
283 struct mgs_target_info *mti)
288 rc = mgs_check_index(env, mgs, mti);
290 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
291 "this MGS does not know about it, preventing "
292 "registration.\n", mti->mti_svname);
294 } else if (rc == -1) {
295 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
296 "disappeared! Regenerating all logs.\n",
298 mti->mti_flags |= LDD_F_WRITECONF;
301 /* Index is correctly marked as used */
308 /* Ensure this is not a failover node that is connecting first*/
309 static int mgs_check_failover_reg(struct mgs_target_info *mti)
315 ptr = mti->mti_params;
316 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
317 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
318 for (i = 0; i < mti->mti_nid_count; i++) {
319 struct lnet_nid nid2;
322 if (target_supports_large_nid(mti)) {
323 rc = libcfs_strnid(&nid2, mti->mti_nidlist[i]);
325 LCONSOLE_WARN("NID %s is unsupported type or improper format\n",
326 libcfs_nidstr(&nid));
330 lnet_nid4_to_nid(mti->mti_nids[i], &nid2);
333 if (nid_same(&nid, &nid2)) {
334 LCONSOLE_WARN("Denying initial registration attempt from nid %s, specified as failover\n",
335 libcfs_nidstr(&nid));
336 return -EADDRNOTAVAIL;
344 /* Called whenever a target starts up. Flags indicate first connect, etc. */
345 static int mgs_target_reg(struct tgt_session_info *tsi)
347 struct obd_device *obd = tsi->tsi_exp->exp_obd;
348 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
349 struct mgs_target_info *mti, *rep_mti;
350 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
351 struct fs_db *c_fsdb = NULL; /* config fsdb */
352 char barrier_name[20];
358 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
360 return err_serious(rc);
362 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
364 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
366 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
367 RETURN(err_serious(-EFAULT));
370 down_read(&mgs->mgs_barrier_rwsem);
372 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
374 opc = mti->mti_flags & LDD_F_OPC_MASK;
378 if (opc == LDD_F_OPC_READY) {
379 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
380 mti->mti_fsname, mti->mti_stripe_index);
381 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
383 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
384 CERROR("%s: Update IR return failure: rc = %d\n",
385 mti->mti_fsname, rc);
388 GOTO(out_norevoke, rc);
391 /* Do not support unregistering right now. */
392 if (opc != LDD_F_OPC_REG)
393 GOTO(out_norevoke, rc = -EINVAL);
395 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
396 mti->mti_fsname, BARRIER_FILENAME);
397 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
399 CERROR("%s: Can't get db for %s: rc = %d\n",
400 mti->mti_fsname, barrier_name, rc);
402 GOTO(out_norevoke, rc);
405 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
406 mti->mti_fsname, mti->mti_stripe_index);
408 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
409 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
410 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
411 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
412 LCONSOLE_WARN("%s: the system is in barrier, refuse "
413 "the connection from MDT %s temporary\n",
414 obd->obd_name, mti->mti_svname);
416 GOTO(out_norevoke, rc = -EBUSY);
419 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
420 !b_fsdb->fsdb_barrier_disabled) {
421 LCONSOLE_WARN("%s: the MDT %s does not support write "
422 "barrier, so disable barrier on the "
424 obd->obd_name, mti->mti_svname);
426 b_fsdb->fsdb_barrier_disabled = 1;
430 if (mti->mti_flags & LDD_F_NEED_INDEX)
431 mti->mti_flags |= LDD_F_WRITECONF;
433 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
434 /* We're just here as a startup ping. */
435 CDEBUG(D_MGS, "Server %s is running on %s\n",
436 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
437 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
438 /* above will set appropriate mti flags */
440 /* Nothing wrong, or fatal error */
441 GOTO(out_norevoke, rc);
442 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
443 rc = mgs_check_failover_reg(mti);
445 GOTO(out_norevoke, rc);
448 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
450 if (mti->mti_flags & LDD_F_WRITECONF) {
451 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
452 mti->mti_stripe_index == 0) {
453 mgs_put_fsdb(mgs, b_fsdb);
455 rc = mgs_erase_logs(tsi->tsi_env, mgs,
457 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
458 "request. All servers must be restarted "
459 "in order to regenerate the logs: rc = %d"
460 "\n", obd->obd_name, mti->mti_fsname, rc);
461 if (rc && rc != -ENOENT)
462 GOTO(out_norevoke, rc);
464 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
465 barrier_name, &b_fsdb);
467 CERROR("Can't get db for %s: %d\n",
470 GOTO(out_norevoke, rc);
473 if (!(exp_connect_flags(tsi->tsi_exp) &
474 OBD_CONNECT_BARRIER)) {
475 LCONSOLE_WARN("%s: the MDT %s does not support "
476 "write barrier, disable barrier "
477 "on the whole system.\n",
478 obd->obd_name, mti->mti_svname);
480 b_fsdb->fsdb_barrier_disabled = 1;
482 } else if (mti->mti_flags &
483 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
484 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
485 LCONSOLE_WARN("%s: Regenerating %s log by user "
486 "request: rc = %d\n",
487 obd->obd_name, mti->mti_svname, rc);
489 GOTO(out_norevoke, rc);
492 mti->mti_flags |= LDD_F_UPDATE;
495 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
497 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
499 GOTO(out_norevoke, rc);
503 * Log writing contention is handled by the fsdb_mutex.
505 * It should be alright if someone was reading while we were
506 * updating the logs - if we revoke at the end they will just update
507 * from where they left off.
509 if (mti->mti_flags & LDD_F_UPDATE) {
510 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
511 mti->mti_stripe_index);
513 /* create or update the target log
514 and update the client/mdt logs */
515 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
517 CERROR("Failed to write %s log (%d)\n",
518 mti->mti_svname, rc);
522 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
523 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
524 mti->mti_flags |= LDD_F_REWRITE_LDD;
528 mgs_revoke_lock(mgs, c_fsdb, MGS_CFG_T_CONFIG);
531 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
533 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
534 mti->mti_fsname, &c_fsdb);
536 CERROR("Fail to get db for %s: %d\n",
537 mti->mti_fsname, rc);
541 memcpy(b_fsdb->fsdb_mdt_index_map,
542 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
543 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
547 up_read(&mgs->mgs_barrier_rwsem);
549 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
550 mti->mti_stripe_index, rc);
551 /* An error flag is set in the mti reply rather than an error code */
553 mti->mti_flags |= LDD_F_ERROR;
555 /* send back the whole mti in the reply */
556 if (target_supports_large_nid(mti)) {
557 size_t len = offsetof(struct mgs_target_info, mti_nidlist);
560 mti_len = mti->mti_nid_count * LNET_NIDSTR_SIZE;
561 err = req_capsule_server_grow(tsi->tsi_pill,
562 &RMF_MGS_TARGET_INFO,
567 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
569 if (target_supports_large_nid(mti))
570 memcpy(rep_mti->mti_nidlist, mti->mti_nidlist, mti_len);
572 /* Flush logs to disk */
573 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
575 mgs_put_fsdb(mgs, b_fsdb);
577 mgs_put_fsdb(mgs, c_fsdb);
581 /* Called whenever a target cleans up. */
582 static int mgs_target_del(struct tgt_session_info *tsi)
586 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
591 static int mgs_config_read(struct tgt_session_info *tsi)
593 struct ptlrpc_request *req = tgt_ses_req(tsi);
594 struct mgs_config_body *body;
599 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
601 DEBUG_REQ(D_HA, req, "no mgs_config_body");
602 RETURN(err_serious(-EFAULT));
605 switch (body->mcb_type) {
606 case MGS_CFG_T_RECOVER:
607 rc = mgs_get_ir_logs(req);
609 case MGS_CFG_T_NODEMAP:
610 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
612 case MGS_CFG_T_CONFIG:
623 static int mgs_llog_open(struct tgt_session_info *tsi)
625 struct mgs_thread_info *mgi;
626 struct ptlrpc_request *req = tgt_ses_req(tsi);
632 rc = tgt_llog_open(tsi);
637 * For old clients there is no direct way of knowing which file system
638 * a client is operating at the MGS side. But we need to pick up those
639 * clients so that the MGS can mark the corresponding file system as
640 * non-IR capable because old clients are not ready to be notified.
642 * Therefore we attempt to detect the file systems name by hacking the
643 * llog operation which is currently used by the clients to fetch
644 * configuration logs. At present this is fine because this is the
645 * ONLY llog operation between mgc and the MGS.
647 * If extra llog operation are going to be added, this function needs
650 * When releases prior than 2.0 are not supported, the following code
653 mgi = mgs_env_info(tsi->tsi_env);
655 RETURN(PTR_ERR(mgi));
657 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
659 char *ptr = strrchr(logname, '-');
660 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
662 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
663 if (strcmp(logname, PARAMS_FILENAME) != 0)
664 LCONSOLE_WARN("%s: non-config logname "
666 tgt_name(tsi->tsi_tgt),
668 /* not error, this can be llog test name */
670 strncpy(mgi->mgi_fsname, logname, len);
671 mgi->mgi_fsname[len] = 0;
673 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
675 if (rc && rc != -EEXIST) {
676 LCONSOLE_WARN("%s: Unable to add client %s "
677 "to file system %s: %d\n",
678 tgt_name(tsi->tsi_tgt),
679 libcfs_nidstr(&req->rq_peer.nid),
680 mgi->mgi_fsname, rc);
686 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
692 static inline int mgs_init_export(struct obd_export *exp)
694 struct mgs_export_data *data = &exp->u.eu_mgs_data;
696 /* init mgs_export_data for fsc */
697 spin_lock_init(&data->med_lock);
698 INIT_LIST_HEAD(&data->med_clients);
700 spin_lock(&exp->exp_lock);
701 exp->exp_connecting = 1;
702 spin_unlock(&exp->exp_lock);
704 /* self-export doesn't need client data and ldlm initialization */
705 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
706 &exp->exp_client_uuid)))
708 return ldlm_init_export(exp);
711 static inline int mgs_destroy_export(struct obd_export *exp)
715 target_destroy_export(exp);
716 mgs_client_free(exp);
718 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
719 &exp->exp_client_uuid)))
722 ldlm_destroy_export(exp);
727 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
734 for (ptr = arg; *ptr != '\0'; ptr++) {
735 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
739 /* Test for fsname.poolname format.
740 * strlen() test if poolname is empty
742 ptr = strchr(arg, '.');
743 if (!ptr || !strlen(ptr))
747 /* Check pool name validity. */
748 if (ptr[0] == '\0' || lov_pool_is_reserved(ptr))
750 /* Also make sure poolname is not to long. */
751 if (strlen(ptr) > LOV_MAXPOOLNAME)
752 return -ENAMETOOLONG;
753 strlcpy(poolname, ptr, LOV_MAXPOOLNAME + 1);
755 /* Test if fsname is empty */
756 len = strlen(arg) - strlen(ptr) - 1;
761 if (len > LUSTRE_MAXFSNAME)
762 return -ENAMETOOLONG;
764 strncpy(fsname, arg, len);
769 static int mgs_iocontrol_nodemap(const struct lu_env *env,
770 struct mgs_device *mgs,
771 struct obd_ioctl_data *data)
773 struct lustre_cfg *lcfg = NULL;
776 const char *nodemap_name = NULL;
777 const char *nidstr = NULL;
778 const char *client_idstr = NULL;
779 const char *idtype_str = NULL;
782 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
784 unsigned long client_id;
791 if (data->ioc_type != LUSTRE_CFG_TYPE) {
792 CERROR("%s: unknown cfg record type: %d\n",
793 mgs->mgs_obd->obd_name, data->ioc_type);
794 GOTO(out, rc = -EINVAL);
797 if (data->ioc_plen1 > PAGE_SIZE)
798 GOTO(out, rc = -E2BIG);
800 OBD_ALLOC(lcfg, data->ioc_plen1);
802 GOTO(out, rc = -ENOMEM);
804 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
805 GOTO(out_lcfg, rc = -EFAULT);
807 cmd = lcfg->lcfg_command;
810 case LCFG_NODEMAP_ACTIVATE:
811 if (lcfg->lcfg_bufcount != 2)
812 GOTO(out_lcfg, rc = -EINVAL);
813 param = lustre_cfg_string(lcfg, 1);
814 if (strcmp(param, "1") == 0)
819 case LCFG_NODEMAP_ADD:
820 case LCFG_NODEMAP_DEL:
821 if (lcfg->lcfg_bufcount != 2)
822 GOTO(out_lcfg, rc = -EINVAL);
823 nodemap_name = lustre_cfg_string(lcfg, 1);
824 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
826 case LCFG_NODEMAP_TEST_NID:
827 if (lcfg->lcfg_bufcount != 2)
828 GOTO(out_lcfg, rc = -EINVAL);
829 nidstr = lustre_cfg_string(lcfg, 1);
830 rc = libcfs_strnid(&nid, nidstr);
834 nodemap_test_nid(&nid, name_buf, sizeof(name_buf));
835 rc = copy_to_user(data->ioc_pbuf1, name_buf,
836 min_t(size_t, data->ioc_plen1,
839 GOTO(out_lcfg, rc = -EFAULT);
841 case LCFG_NODEMAP_TEST_ID:
842 if (lcfg->lcfg_bufcount != 4)
843 GOTO(out_lcfg, rc = -EINVAL);
844 nidstr = lustre_cfg_string(lcfg, 1);
845 idtype_str = lustre_cfg_string(lcfg, 2);
846 client_idstr = lustre_cfg_string(lcfg, 3);
848 rc = libcfs_strnid(&nid, nidstr);
852 if (strcmp(idtype_str, "uid") == 0)
853 idtype = NODEMAP_UID;
854 else if (strcmp(idtype_str, "gid") == 0)
855 idtype = NODEMAP_GID;
856 else if (strcmp(idtype_str, "projid") == 0)
857 idtype = NODEMAP_PROJID;
859 GOTO(out_lcfg, rc = -EINVAL);
861 rc = kstrtoul(client_idstr, 10, &client_id);
863 GOTO(out_lcfg, rc = -EINVAL);
865 rc = nodemap_test_id(&nid, idtype, client_id, &fs_id);
867 GOTO(out_lcfg, rc = -EINVAL);
869 if (data->ioc_plen1 < sizeof(fs_idstr))
870 GOTO(out_lcfg, rc = -EINVAL);
872 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
873 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
874 sizeof(fs_idstr)) != 0)
875 GOTO(out_lcfg, rc = -EINVAL);
877 case LCFG_NODEMAP_ADD_RANGE:
878 case LCFG_NODEMAP_DEL_RANGE:
879 case LCFG_NODEMAP_ADD_UIDMAP:
880 case LCFG_NODEMAP_DEL_UIDMAP:
881 case LCFG_NODEMAP_ADD_GIDMAP:
882 case LCFG_NODEMAP_DEL_GIDMAP:
883 case LCFG_NODEMAP_ADD_PROJIDMAP:
884 case LCFG_NODEMAP_DEL_PROJIDMAP:
885 case LCFG_NODEMAP_SET_FILESET:
886 case LCFG_NODEMAP_SET_SEPOL:
887 if (lcfg->lcfg_bufcount != 3)
888 GOTO(out_lcfg, rc = -EINVAL);
889 nodemap_name = lustre_cfg_string(lcfg, 1);
890 param = lustre_cfg_string(lcfg, 2);
891 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
893 case LCFG_NODEMAP_ADMIN:
894 case LCFG_NODEMAP_TRUSTED:
895 case LCFG_NODEMAP_DENY_UNKNOWN:
896 case LCFG_NODEMAP_SQUASH_UID:
897 case LCFG_NODEMAP_SQUASH_GID:
898 case LCFG_NODEMAP_SQUASH_PROJID:
899 case LCFG_NODEMAP_MAP_MODE:
900 case LCFG_NODEMAP_AUDIT_MODE:
901 case LCFG_NODEMAP_FORBID_ENCRYPT:
902 case LCFG_NODEMAP_READONLY_MOUNT:
903 case LCFG_NODEMAP_RBAC:
904 if (lcfg->lcfg_bufcount != 4)
905 GOTO(out_lcfg, rc = -EINVAL);
906 nodemap_name = lustre_cfg_string(lcfg, 1);
907 param = lustre_cfg_string(lcfg, 3);
908 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
915 CDEBUG_LIMIT(rc == -EEXIST ? D_INFO : D_ERROR,
916 "%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
917 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
922 /* revoke nodemap lock */
923 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
925 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
926 mgs->mgs_obd->obd_name, rc);
928 mgs_revoke_lock(mgs, fsdb, MGS_CFG_T_NODEMAP);
929 mgs_put_fsdb(mgs, fsdb);
933 OBD_FREE(lcfg, data->ioc_plen1);
938 static int mgs_iocontrol_pool(const struct lu_env *env,
939 struct mgs_device *mgs,
940 struct obd_ioctl_data *data)
942 struct mgs_thread_info *mgi = mgs_env_info(env);
944 struct lustre_cfg *lcfg = NULL;
945 char *poolname = NULL;
948 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
949 if (poolname == NULL)
952 if (data->ioc_type != LUSTRE_CFG_TYPE) {
953 CERROR("%s: unknown cfg record type: %d\n",
954 mgs->mgs_obd->obd_name, data->ioc_type);
955 GOTO(out_pool, rc = -EINVAL);
958 if (data->ioc_plen1 > PAGE_SIZE)
959 GOTO(out_pool, rc = -E2BIG);
961 OBD_ALLOC(lcfg, data->ioc_plen1);
963 GOTO(out_pool, rc = -ENOMEM);
965 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
966 GOTO(out_lcfg, rc = -EFAULT);
968 if (lcfg->lcfg_bufcount < 2)
969 GOTO(out_lcfg, rc = -EFAULT);
971 /* first arg is always <fsname>.<poolname> */
972 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
977 switch (lcfg->lcfg_command) {
979 if (lcfg->lcfg_bufcount != 2)
980 GOTO(out_lcfg, rc = -EINVAL);
981 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
985 if (lcfg->lcfg_bufcount != 3)
986 GOTO(out_lcfg, rc = -EINVAL);
987 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
988 poolname, lustre_cfg_string(lcfg, 2));
991 if (lcfg->lcfg_bufcount != 3)
992 GOTO(out_lcfg, rc = -EINVAL);
993 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
994 poolname, lustre_cfg_string(lcfg, 2));
997 if (lcfg->lcfg_bufcount != 2)
998 GOTO(out_lcfg, rc = -EINVAL);
999 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
1007 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1008 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
1013 OBD_FREE(lcfg, data->ioc_plen1);
1015 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1019 /* from mdt_iocontrol */
1020 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1021 void *karg, void __user *uarg)
1023 struct obd_device *obd = exp->exp_obd;
1024 struct mgs_device *mgs = exp2mgs_dev(exp);
1025 struct obd_ioctl_data *data;
1030 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
1031 obd->obd_name, cmd, len, karg, uarg);
1032 if (unlikely(karg == NULL))
1033 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc));
1036 rc = lu_env_init(&env, LCT_MG_THREAD);
1042 case OBD_IOC_PARAM: {
1043 struct lustre_cfg *lcfg;
1045 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1046 CERROR("%s: unknown cfg record type '%x': rc = %d\n",
1047 obd->obd_name, data->ioc_type, rc);
1051 OBD_ALLOC(lcfg, data->ioc_plen1);
1053 GOTO(out, rc = -ENOMEM);
1054 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1055 GOTO(out_free, rc = -EFAULT);
1057 if (lcfg->lcfg_bufcount < 1)
1060 rc = mgs_set_param(&env, mgs, lcfg);
1062 CERROR("%s: setparam err: rc = %d\n",
1065 OBD_FREE(lcfg, data->ioc_plen1);
1068 case OBD_IOC_REPLACE_NIDS:
1069 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1070 CERROR("%s: no device or fsname specified: rc = %d\n",
1075 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1077 CERROR("%s: device or fsname is too long: rc = %d\n",
1082 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1083 CERROR("%s: unterminated device or fsname: rc = %d\n",
1088 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1089 CERROR("%s: no NIDs specified: rc = %d\n",
1094 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1095 CERROR("%s: NID list is not NUL terminated: rc = %d\n",
1100 /* replace nids in llog */
1101 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1104 CERROR("%s: error replacing NIDs for '%s': rc = %d\n",
1105 obd->obd_name, data->ioc_inlbuf1, rc);
1108 case OBD_IOC_CLEAR_CONFIGS:
1109 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1110 CERROR("%s: no device or fsname specified: rc = %d\n",
1115 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1117 CERROR("%s: device or fsname is too long: rc = %d\n",
1122 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1123 CERROR("%s: unterminated device or fsname: rc = %d\n",
1128 /* remove records marked SKIP from config logs */
1129 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1131 CERROR("%s: error clearing config log: rc = %d\n",
1136 rc = mgs_iocontrol_pool(&env, mgs, data);
1138 #ifdef OBD_IOC_BARRIER
1139 case_OBD_IOC_DEPRECATED_FT(OBD_IOC_BARRIER, obd->obd_name, 2, 17);
1141 case OBD_IOC_BARRIER_V2:
1142 rc = mgs_iocontrol_barrier(&env, mgs, data);
1144 case OBD_IOC_NODEMAP:
1145 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1147 case OBD_IOC_LCFG_FORK:
1148 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1151 case OBD_IOC_LCFG_ERASE:
1152 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1154 case OBD_IOC_CATLOGLIST:
1155 rc = mgs_list_logs(&env, mgs, data);
1157 case OBD_IOC_LLOG_CANCEL:
1158 case OBD_IOC_LLOG_REMOVE:
1159 case OBD_IOC_LLOG_CHECK:
1160 case OBD_IOC_LLOG_INFO:
1161 case OBD_IOC_LLOG_PRINT: {
1162 struct llog_ctxt *ctxt;
1164 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1165 rc = llog_ioctl(&env, ctxt, cmd, data);
1166 llog_ctxt_put(ctxt);
1170 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
1178 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1180 struct obd_connect_data *data = NULL;
1181 struct obd_device *obd;
1185 OBD_ALLOC_PTR(data);
1189 obd = class_name2obd(nextdev);
1191 CERROR("can't locate next device: %s\n", nextdev);
1192 GOTO(out, rc = -ENOTCONN);
1195 data->ocd_version = LUSTRE_VERSION_CODE;
1197 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1198 &obd->obd_uuid, data, NULL);
1200 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1204 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1205 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1206 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1212 static struct tgt_handler mgs_mgs_handlers[] = {
1213 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1214 0, MGS_CONNECT, mgs_connect,
1215 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1216 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1217 0, MGS_DISCONNECT, mgs_disconnect,
1218 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1219 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1220 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 18, 53, 0)
1221 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_SET_INFO, mgs_set_info),
1223 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1224 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1225 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1228 static struct tgt_handler mgs_obd_handlers[] = {
1229 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1232 static struct tgt_handler mgs_dlm_handlers[] = {
1233 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1234 .th_name = "LDLM_ENQUEUE",
1235 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1236 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1237 * * is safe. If such tests will be needed we have to distinguish
1238 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1239 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1241 .th_opc = LDLM_ENQUEUE,
1242 .th_flags = HAS_KEY,
1243 .th_act = tgt_enqueue,
1244 .th_fmt = &RQF_LDLM_ENQUEUE,
1245 .th_version = LUSTRE_DLM_VERSION,
1249 static struct tgt_handler mgs_llog_handlers[] = {
1250 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1251 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1252 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1253 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1256 static struct tgt_opc_slice mgs_common_slice[] = {
1258 .tos_opc_start = MGS_FIRST_OPC,
1259 .tos_opc_end = MGS_LAST_OPC,
1260 .tos_hs = mgs_mgs_handlers
1263 .tos_opc_start = OBD_FIRST_OPC,
1264 .tos_opc_end = OBD_LAST_OPC,
1265 .tos_hs = mgs_obd_handlers
1268 .tos_opc_start = LDLM_FIRST_OPC,
1269 .tos_opc_end = LDLM_LAST_OPC,
1270 .tos_hs = mgs_dlm_handlers
1273 .tos_opc_start = LLOG_FIRST_OPC,
1274 .tos_opc_end = LLOG_LAST_OPC,
1275 .tos_hs = mgs_llog_handlers
1278 .tos_opc_start = SEC_FIRST_OPC,
1279 .tos_opc_end = SEC_LAST_OPC,
1280 .tos_hs = tgt_sec_ctx_handlers
1287 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1288 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1290 struct ptlrpc_service_conf conf;
1291 struct obd_device *obd;
1292 struct lustre_mount_info *lmi;
1293 struct llog_ctxt *ctxt;
1298 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1302 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1304 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1308 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1311 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1315 /* namespace for mgs llog */
1316 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1317 LDLM_NAMESPACE_SERVER,
1318 LDLM_NAMESPACE_MODEST,
1320 if (IS_ERR(obd->obd_namespace)) {
1321 rc = PTR_ERR(obd->obd_namespace);
1322 CERROR("%s: unable to create server namespace: rc = %d\n",
1324 obd->obd_namespace = NULL;
1328 /* No recovery for MGCs */
1329 obd->obd_replayable = 0;
1331 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1332 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1333 OBD_FAIL_MGS_ALL_REPLY_NET);
1337 rc = mgs_fs_setup(env, mgs);
1339 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1344 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1345 obd, &llog_osd_ops);
1349 /* XXX: we need this trick till N:1 stack is supported
1350 * set "current" directory for named llogs */
1351 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1353 ctxt->loc_dir = mgs->mgs_configs_dir;
1354 llog_ctxt_put(ctxt);
1356 /* Internal mgs setup */
1357 mgs_init_fsdb_list(mgs);
1358 mutex_init(&mgs->mgs_mutex);
1359 mgs->mgs_start_time = ktime_get_real_seconds();
1360 spin_lock_init(&mgs->mgs_lock);
1361 mutex_init(&mgs->mgs_health_mutex);
1362 init_rwsem(&mgs->mgs_barrier_rwsem);
1364 rc = mgs_lcfg_rename(env, mgs);
1368 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1370 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1375 /* Setup params fsdb and log, so that other servers can make a local
1376 * copy successfully when they are mounted. See LU-4783 */
1377 rc = mgs_params_fsdb_setup(env, mgs);
1379 /* params fsdb and log can be setup later */
1380 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1381 obd->obd_name, PARAMS_FILENAME, rc);
1383 /* Setup _mgs fsdb, useful for srpc */
1384 mgs__mgs_fsdb_setup(env, mgs);
1386 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1387 "mgs_ldlm_client", &obd->obd_ldlm_client);
1389 conf = (typeof(conf)) {
1390 .psc_name = LUSTRE_MGS_NAME,
1391 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1393 .bc_nbufs = MGS_NBUFS,
1394 .bc_buf_size = MGS_BUFSIZE,
1395 .bc_req_max_size = MGS_MAXREQSIZE,
1396 .bc_rep_max_size = MGS_MAXREPSIZE,
1397 .bc_req_portal = MGS_REQUEST_PORTAL,
1398 .bc_rep_portal = MGC_REPLY_PORTAL,
1401 .tc_thr_name = "ll_mgs",
1402 .tc_nthrs_init = MGS_NTHRS_INIT,
1403 .tc_nthrs_max = MGS_NTHRS_MAX,
1404 .tc_ctx_tags = LCT_MG_THREAD,
1407 .so_req_handler = tgt_request_handle,
1408 .so_req_printer = target_print_req,
1412 /* Start the service threads */
1413 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1414 obd->obd_debugfs_entry);
1415 if (IS_ERR(mgs->mgs_service)) {
1416 rc = PTR_ERR(mgs->mgs_service);
1417 CERROR("failed to start mgs service: %d\n", rc);
1418 mgs->mgs_service = NULL;
1419 GOTO(err_lproc, rc);
1422 ping_evictor_start();
1424 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1426 /* device stack is not yet fully setup to keep no objects behind */
1427 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1430 mgs_params_fsdb_cleanup(env, mgs);
1431 lproc_mgs_cleanup(mgs);
1433 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1435 ctxt->loc_dir = NULL;
1436 llog_cleanup(env, ctxt);
1439 tgt_fini(env, &mgs->mgs_lut);
1441 /* No extra cleanup needed for llog_init_commit_thread() */
1442 mgs_fs_cleanup(env, mgs);
1444 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1445 obd->obd_namespace = NULL;
1447 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1448 lu_site_print(env, mgs2lu_dev(mgs)->ld_site,
1449 &mgs2lu_dev(mgs)->ld_site->ls_obj_hash.nelems,
1450 D_OTHER, lu_cdebug_printer);
1451 obd_disconnect(mgs->mgs_bottom_exp);
1454 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1458 static struct lu_device *mgs_device_free(const struct lu_env *env,
1459 struct lu_device *lu)
1461 struct mgs_device *mgs = lu2mgs_dev(lu);
1464 dt_device_fini(&mgs->mgs_dt_dev);
1469 static int mgs_process_config(const struct lu_env *env,
1470 struct lu_device *dev,
1471 struct lustre_cfg *lcfg)
1477 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1478 const struct lu_object_conf *unused)
1480 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1481 struct lu_device *under;
1482 struct lu_object *below;
1486 /* do no set .do_ops as mgs calls to bottom osd directly */
1488 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1489 PFID(lu_object_fid(o)));
1491 under = &d->mgs_bottom->dd_lu_dev;
1492 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1494 lu_object_add(o, below);
1501 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1503 struct mgs_object *obj = lu2mgs_obj(o);
1504 struct lu_object_header *h = o->lo_header;
1506 dt_object_fini(&obj->mgo_obj);
1507 lu_object_header_fini(h);
1508 OBD_FREE_PRE(obj, sizeof(*obj), "kfreed");
1509 kfree_rcu(obj, mgo_header.loh_rcu);
1512 static int mgs_object_print(const struct lu_env *env, void *cookie,
1513 lu_printer_t p, const struct lu_object *l)
1515 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1517 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1520 static const struct lu_object_operations mgs_lu_obj_ops = {
1521 .loo_object_init = mgs_object_init,
1522 .loo_object_free = mgs_object_free,
1523 .loo_object_print = mgs_object_print,
1526 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1527 const struct lu_object_header *hdr,
1528 struct lu_device *d)
1530 struct lu_object_header *h;
1531 struct mgs_object *o;
1532 struct lu_object *l;
1534 LASSERT(hdr == NULL);
1538 l = &o->mgo_obj.do_lu;
1541 lu_object_header_init(h);
1542 dt_object_init(&o->mgo_obj, h, d);
1543 lu_object_add_top(h, l);
1545 l->lo_ops = &mgs_lu_obj_ops;
1553 const struct lu_device_operations mgs_lu_ops = {
1554 .ldo_object_alloc = mgs_object_alloc,
1555 .ldo_process_config = mgs_process_config,
1558 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1559 struct lu_device_type *type,
1560 struct lustre_cfg *lcfg)
1562 struct mgs_device *mgs;
1563 struct lu_device *ludev;
1567 ludev = ERR_PTR(-ENOMEM);
1571 ludev = mgs2lu_dev(mgs);
1572 dt_device_init(&mgs->mgs_dt_dev, type);
1573 rc = mgs_init0(env, mgs, type, lcfg);
1575 mgs_device_free(env, ludev);
1576 ludev = ERR_PTR(rc);
1582 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1583 struct lu_device *d)
1585 struct mgs_device *mgs = lu2mgs_dev(d);
1586 struct obd_device *obd = mgs->mgs_obd;
1587 struct llog_ctxt *ctxt;
1591 LASSERT(mgs->mgs_bottom);
1593 class_disconnect_exports(obd);
1595 ping_evictor_stop();
1597 mutex_lock(&mgs->mgs_health_mutex);
1598 ptlrpc_unregister_service(mgs->mgs_service);
1599 mutex_unlock(&mgs->mgs_health_mutex);
1601 mgs_params_fsdb_cleanup(env, mgs);
1602 mgs_cleanup_fsdb_list(mgs);
1604 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1605 obd_exports_barrier(obd);
1606 obd_zombie_barrier();
1608 tgt_fini(env, &mgs->mgs_lut);
1609 lproc_mgs_cleanup(mgs);
1611 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1613 ctxt->loc_dir = NULL;
1614 llog_cleanup(env, ctxt);
1617 mgs_fs_cleanup(env, mgs);
1619 ldlm_namespace_free_post(obd->obd_namespace);
1620 obd->obd_namespace = NULL;
1622 lu_site_purge(env, d->ld_site, ~0);
1623 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
1624 D_OTHER, lu_cdebug_printer);
1625 LASSERT(mgs->mgs_bottom_exp);
1626 obd_disconnect(mgs->mgs_bottom_exp);
1628 server_put_mount(obd->obd_name, true);
1633 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1634 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1636 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1638 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1640 static const struct lu_device_type_operations mgs_device_type_ops = {
1641 .ldto_init = mgs_type_init,
1642 .ldto_fini = mgs_type_fini,
1644 .ldto_start = mgs_type_start,
1645 .ldto_stop = mgs_type_stop,
1647 .ldto_device_alloc = mgs_device_alloc,
1648 .ldto_device_free = mgs_device_free,
1650 .ldto_device_fini = mgs_device_fini
1653 static struct lu_device_type mgs_device_type = {
1654 .ldt_tags = LU_DEVICE_DT,
1655 .ldt_name = LUSTRE_MGS_NAME,
1656 .ldt_ops = &mgs_device_type_ops,
1657 .ldt_ctx_tags = LCT_MG_THREAD
1660 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1661 struct obd_device *obd, struct obd_uuid *cluuid,
1662 struct obd_connect_data *data, void *localdata)
1666 if (exp == NULL || obd == NULL || cluuid == NULL)
1669 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1672 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1674 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1675 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1677 exp->exp_connect_data = *data;
1678 data->ocd_version = LUSTRE_VERSION_CODE;
1681 RETURN(mgs_export_stats_init(obd, exp, localdata));
1684 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1685 struct obd_device *obd, struct obd_uuid *cluuid,
1686 struct obd_connect_data *data, void *localdata)
1688 struct obd_export *lexp;
1689 struct lustre_handle conn = {
1696 if (exp == NULL || obd == NULL || cluuid == NULL)
1699 rc = class_connect(&conn, obd, cluuid);
1703 lexp = class_conn2export(&conn);
1707 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1709 GOTO(out_disconnect, rc);
1716 class_disconnect(lexp);
1721 static int mgs_obd_disconnect(struct obd_export *exp)
1729 mgs_fsc_cleanup(exp);
1731 class_export_get(exp);
1732 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1734 rc = server_disconnect_export(exp);
1735 class_export_put(exp);
1739 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1741 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1744 mutex_lock(&mgs->mgs_health_mutex);
1745 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1746 mutex_unlock(&mgs->mgs_health_mutex);
1748 return rc != 0 ? 1 : 0;
1751 /* use obd ops to offer management infrastructure */
1752 static const struct obd_ops mgs_obd_device_ops = {
1753 .o_owner = THIS_MODULE,
1754 .o_connect = mgs_obd_connect,
1755 .o_reconnect = mgs_obd_reconnect,
1756 .o_disconnect = mgs_obd_disconnect,
1757 .o_init_export = mgs_init_export,
1758 .o_destroy_export = mgs_destroy_export,
1759 .o_iocontrol = mgs_iocontrol,
1760 .o_health_check = mgs_health_check,
1763 static int __init mgs_init(void)
1767 rc = libcfs_setup();
1771 return class_register_type(&mgs_obd_device_ops, NULL, true,
1772 LUSTRE_MGS_NAME, &mgs_device_type);
1775 static void __exit mgs_exit(void)
1777 class_unregister_type(LUSTRE_MGS_NAME);
1780 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1781 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1782 MODULE_VERSION(LUSTRE_VERSION_STRING);
1783 MODULE_LICENSE("GPL");
1785 module_init(mgs_init);
1786 module_exit(mgs_exit);