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;
106 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
107 void *cbdata, enum ast_type type)
111 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
114 /* l_ast_data is used as a marker to avoid cancel ldlm lock
115 * twice. See LU-2317. */
116 lock_res_and_lock(lock);
117 fsdb = (struct fs_db *)lock->l_ast_data;
118 lock->l_ast_data = NULL;
119 unlock_res_and_lock(lock);
122 struct lustre_handle lockh;
126 /* clear the bit before lock put */
127 clear_bit(FSDB_REVOKING_LOCK,
131 clear_bit(FSDB_REVOKING_PARAMS,
135 mgs_ir_notify_complete(fsdb);
143 ldlm_lock2handle(lock, &lockh);
144 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
148 RETURN(ldlm_completion_ast(lock, flags, cbdata));
151 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
154 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
157 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
160 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
163 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
166 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
169 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
172 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
175 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
177 ldlm_completion_callback cp = NULL;
178 struct lustre_handle lockh = {
181 struct ldlm_res_id res_id;
182 __u64 flags = LDLM_FL_ATOMIC_CB;
186 LASSERT(fsdb->fsdb_name[0] != '\0');
187 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
190 case CONFIG_T_CONFIG:
191 case CONFIG_T_NODEMAP:
192 cp = mgs_completion_ast_config;
193 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
196 case CONFIG_T_PARAMS:
197 cp = mgs_completion_ast_params;
198 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
201 case CONFIG_T_RECOVER:
202 cp = mgs_completion_ast_ir;
204 case CONFIG_T_BARRIER:
205 cp = mgs_completion_ast_barrier;
213 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
214 &res_id, LDLM_PLAIN, NULL, LCK_EX,
215 &flags, ldlm_blocking_ast, cp,
216 NULL, fsdb, 0, LVB_T_NONE, NULL,
218 if (rc != ELDLM_OK) {
219 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
220 mgs->mgs_obd->obd_name,
221 le64_to_cpu(res_id.name[0]),
222 le64_to_cpu(res_id.name[1]), rc);
224 if (type == CONFIG_T_CONFIG)
225 clear_bit(FSDB_REVOKING_LOCK,
228 if (type == CONFIG_T_PARAMS)
229 clear_bit(FSDB_REVOKING_PARAMS,
232 /* lock has been cancelled in completion_ast. */
241 static int mgs_check_target(const struct lu_env *env,
242 struct mgs_device *mgs,
243 struct mgs_target_info *mti)
248 rc = mgs_check_index(env, mgs, mti);
250 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
251 "this MGS does not know about it, preventing "
252 "registration.\n", mti->mti_svname);
254 } else if (rc == -1) {
255 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
256 "disappeared! Regenerating all logs.\n",
258 mti->mti_flags |= LDD_F_WRITECONF;
261 /* Index is correctly marked as used */
268 /* Ensure this is not a failover node that is connecting first*/
269 static int mgs_check_failover_reg(struct mgs_target_info *mti)
275 ptr = mti->mti_params;
276 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
277 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
278 for (i = 0; i < mti->mti_nid_count; i++) {
279 if (nid == mti->mti_nids[i]) {
280 LCONSOLE_WARN("Denying initial registration attempt from nid %s, specified as failover\n",
281 libcfs_nid2str(nid));
282 return -EADDRNOTAVAIL;
290 /* Called whenever a target starts up. Flags indicate first connect, etc. */
291 static int mgs_target_reg(struct tgt_session_info *tsi)
293 struct obd_device *obd = tsi->tsi_exp->exp_obd;
294 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
295 struct mgs_target_info *mti, *rep_mti;
296 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
297 struct fs_db *c_fsdb = NULL; /* config fsdb */
298 char barrier_name[20];
304 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
306 return err_serious(rc);
308 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
310 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
312 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
313 RETURN(err_serious(-EFAULT));
316 down_read(&mgs->mgs_barrier_rwsem);
318 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
320 opc = mti->mti_flags & LDD_F_OPC_MASK;
324 if (opc == LDD_F_OPC_READY) {
325 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
326 mti->mti_fsname, mti->mti_stripe_index);
327 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
329 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
330 CERROR("%s: Update IR return failure: rc = %d\n",
331 mti->mti_fsname, rc);
334 GOTO(out_norevoke, rc);
337 /* Do not support unregistering right now. */
338 if (opc != LDD_F_OPC_REG)
339 GOTO(out_norevoke, rc = -EINVAL);
341 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
342 mti->mti_fsname, BARRIER_FILENAME);
343 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
345 CERROR("%s: Can't get db for %s: rc = %d\n",
346 mti->mti_fsname, barrier_name, rc);
348 GOTO(out_norevoke, rc);
351 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
352 mti->mti_fsname, mti->mti_stripe_index);
354 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
355 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
356 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
357 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
358 LCONSOLE_WARN("%s: the system is in barrier, refuse "
359 "the connection from MDT %s temporary\n",
360 obd->obd_name, mti->mti_svname);
362 GOTO(out_norevoke, rc = -EBUSY);
365 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
366 !b_fsdb->fsdb_barrier_disabled) {
367 LCONSOLE_WARN("%s: the MDT %s does not support write "
368 "barrier, so disable barrier on the "
370 obd->obd_name, mti->mti_svname);
372 b_fsdb->fsdb_barrier_disabled = 1;
376 if (mti->mti_flags & LDD_F_NEED_INDEX)
377 mti->mti_flags |= LDD_F_WRITECONF;
379 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
380 /* We're just here as a startup ping. */
381 CDEBUG(D_MGS, "Server %s is running on %s\n",
382 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
383 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
384 /* above will set appropriate mti flags */
386 /* Nothing wrong, or fatal error */
387 GOTO(out_norevoke, rc);
388 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
389 rc = mgs_check_failover_reg(mti);
391 GOTO(out_norevoke, rc);
394 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
396 if (mti->mti_flags & LDD_F_WRITECONF) {
397 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
398 mti->mti_stripe_index == 0) {
399 mgs_put_fsdb(mgs, b_fsdb);
401 rc = mgs_erase_logs(tsi->tsi_env, mgs,
403 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
404 "request. All servers must be restarted "
405 "in order to regenerate the logs: rc = %d"
406 "\n", obd->obd_name, mti->mti_fsname, rc);
407 if (rc && rc != -ENOENT)
408 GOTO(out_norevoke, rc);
410 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
411 barrier_name, &b_fsdb);
413 CERROR("Can't get db for %s: %d\n",
416 GOTO(out_norevoke, rc);
419 if (!(exp_connect_flags(tsi->tsi_exp) &
420 OBD_CONNECT_BARRIER)) {
421 LCONSOLE_WARN("%s: the MDT %s does not support "
422 "write barrier, disable barrier "
423 "on the whole system.\n",
424 obd->obd_name, mti->mti_svname);
426 b_fsdb->fsdb_barrier_disabled = 1;
428 } else if (mti->mti_flags &
429 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
430 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
431 LCONSOLE_WARN("%s: Regenerating %s log by user "
432 "request: rc = %d\n",
433 obd->obd_name, mti->mti_svname, rc);
435 GOTO(out_norevoke, rc);
438 mti->mti_flags |= LDD_F_UPDATE;
441 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
443 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
445 GOTO(out_norevoke, rc);
449 * Log writing contention is handled by the fsdb_mutex.
451 * It should be alright if someone was reading while we were
452 * updating the logs - if we revoke at the end they will just update
453 * from where they left off.
455 if (mti->mti_flags & LDD_F_UPDATE) {
456 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
457 mti->mti_stripe_index);
459 /* create or update the target log
460 and update the client/mdt logs */
461 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
463 CERROR("Failed to write %s log (%d)\n",
464 mti->mti_svname, rc);
468 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
469 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
470 mti->mti_flags |= LDD_F_REWRITE_LDD;
474 mgs_revoke_lock(mgs, c_fsdb, CONFIG_T_CONFIG);
477 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
479 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
480 mti->mti_fsname, &c_fsdb);
482 CERROR("Fail to get db for %s: %d\n",
483 mti->mti_fsname, rc);
487 memcpy(b_fsdb->fsdb_mdt_index_map,
488 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
489 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
493 up_read(&mgs->mgs_barrier_rwsem);
495 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
496 mti->mti_stripe_index, rc);
497 /* An error flag is set in the mti reply rather than an error code */
499 mti->mti_flags |= LDD_F_ERROR;
501 /* send back the whole mti in the reply */
502 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
505 /* Flush logs to disk */
506 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
508 mgs_put_fsdb(mgs, b_fsdb);
510 mgs_put_fsdb(mgs, c_fsdb);
514 /* Called whenever a target cleans up. */
515 static int mgs_target_del(struct tgt_session_info *tsi)
519 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
524 static int mgs_config_read(struct tgt_session_info *tsi)
526 struct ptlrpc_request *req = tgt_ses_req(tsi);
527 struct mgs_config_body *body;
532 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
534 DEBUG_REQ(D_HA, req, "no mgs_config_body");
535 RETURN(err_serious(-EFAULT));
538 switch (body->mcb_type) {
539 case CONFIG_T_RECOVER:
540 rc = mgs_get_ir_logs(req);
542 case CONFIG_T_NODEMAP:
543 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
545 case CONFIG_T_CONFIG:
556 static int mgs_llog_open(struct tgt_session_info *tsi)
558 struct mgs_thread_info *mgi;
559 struct ptlrpc_request *req = tgt_ses_req(tsi);
565 rc = tgt_llog_open(tsi);
570 * For old clients there is no direct way of knowing which file system
571 * a client is operating at the MGS side. But we need to pick up those
572 * clients so that the MGS can mark the corresponding file system as
573 * non-IR capable because old clients are not ready to be notified.
575 * Therefore we attempt to detect the file systems name by hacking the
576 * llog operation which is currently used by the clients to fetch
577 * configuration logs. At present this is fine because this is the
578 * ONLY llog operation between mgc and the MGS.
580 * If extra llog operation are going to be added, this function needs
583 * When releases prior than 2.0 are not supported, the following code
586 mgi = mgs_env_info(tsi->tsi_env);
588 RETURN(PTR_ERR(mgi));
590 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
592 char *ptr = strrchr(logname, '-');
593 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
595 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
596 if (strcmp(logname, PARAMS_FILENAME) != 0)
597 LCONSOLE_WARN("%s: non-config logname "
599 tgt_name(tsi->tsi_tgt),
601 /* not error, this can be llog test name */
603 strncpy(mgi->mgi_fsname, logname, len);
604 mgi->mgi_fsname[len] = 0;
606 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
608 if (rc && rc != -EEXIST) {
609 LCONSOLE_WARN("%s: Unable to add client %s "
610 "to file system %s: %d\n",
611 tgt_name(tsi->tsi_tgt),
612 libcfs_nid2str(req->rq_peer.nid),
613 mgi->mgi_fsname, rc);
619 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
625 static inline int mgs_init_export(struct obd_export *exp)
627 struct mgs_export_data *data = &exp->u.eu_mgs_data;
629 /* init mgs_export_data for fsc */
630 spin_lock_init(&data->med_lock);
631 INIT_LIST_HEAD(&data->med_clients);
633 spin_lock(&exp->exp_lock);
634 exp->exp_connecting = 1;
635 spin_unlock(&exp->exp_lock);
637 /* self-export doesn't need client data and ldlm initialization */
638 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
639 &exp->exp_client_uuid)))
641 return ldlm_init_export(exp);
644 static inline int mgs_destroy_export(struct obd_export *exp)
648 target_destroy_export(exp);
649 mgs_client_free(exp);
651 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
652 &exp->exp_client_uuid)))
655 ldlm_destroy_export(exp);
660 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
667 for (ptr = arg; *ptr != '\0'; ptr++) {
668 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
672 /* Test for fsname.poolname format.
673 * strlen() test if poolname is empty
675 ptr = strchr(arg, '.');
676 if (!ptr || !strlen(ptr))
680 /* Also make sure poolname is not to long. */
681 if (strlen(ptr) > LOV_MAXPOOLNAME)
682 return -ENAMETOOLONG;
683 strlcpy(poolname, ptr, LOV_MAXPOOLNAME + 1);
685 /* Test if fsname is empty */
686 len = strlen(arg) - strlen(ptr) - 1;
691 if (len > LUSTRE_MAXFSNAME)
692 return -ENAMETOOLONG;
694 strncpy(fsname, arg, len);
699 static int mgs_iocontrol_nodemap(const struct lu_env *env,
700 struct mgs_device *mgs,
701 struct obd_ioctl_data *data)
703 struct lustre_cfg *lcfg = NULL;
706 const char *nodemap_name = NULL;
707 const char *nidstr = NULL;
708 const char *client_idstr = NULL;
709 const char *idtype_str = NULL;
712 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
714 unsigned long client_id;
721 if (data->ioc_type != LUSTRE_CFG_TYPE) {
722 CERROR("%s: unknown cfg record type: %d\n",
723 mgs->mgs_obd->obd_name, data->ioc_type);
724 GOTO(out, rc = -EINVAL);
727 if (data->ioc_plen1 > PAGE_SIZE)
728 GOTO(out, rc = -E2BIG);
730 OBD_ALLOC(lcfg, data->ioc_plen1);
732 GOTO(out, rc = -ENOMEM);
734 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
735 GOTO(out_lcfg, rc = -EFAULT);
737 cmd = lcfg->lcfg_command;
740 case LCFG_NODEMAP_ACTIVATE:
741 if (lcfg->lcfg_bufcount != 2)
742 GOTO(out_lcfg, rc = -EINVAL);
743 param = lustre_cfg_string(lcfg, 1);
744 if (strcmp(param, "1") == 0)
749 case LCFG_NODEMAP_ADD:
750 case LCFG_NODEMAP_DEL:
751 if (lcfg->lcfg_bufcount != 2)
752 GOTO(out_lcfg, rc = -EINVAL);
753 nodemap_name = lustre_cfg_string(lcfg, 1);
754 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
756 case LCFG_NODEMAP_TEST_NID:
757 if (lcfg->lcfg_bufcount != 2)
758 GOTO(out_lcfg, rc = -EINVAL);
759 nidstr = lustre_cfg_string(lcfg, 1);
760 nid = libcfs_str2nid(nidstr);
761 nodemap_test_nid(nid, name_buf, sizeof(name_buf));
762 rc = copy_to_user(data->ioc_pbuf1, name_buf,
763 min_t(size_t, data->ioc_plen1,
766 GOTO(out_lcfg, rc = -EFAULT);
768 case LCFG_NODEMAP_TEST_ID:
769 if (lcfg->lcfg_bufcount != 4)
770 GOTO(out_lcfg, rc = -EINVAL);
771 nidstr = lustre_cfg_string(lcfg, 1);
772 idtype_str = lustre_cfg_string(lcfg, 2);
773 client_idstr = lustre_cfg_string(lcfg, 3);
775 nid = libcfs_str2nid(nidstr);
776 if (strcmp(idtype_str, "uid") == 0)
777 idtype = NODEMAP_UID;
779 idtype = NODEMAP_GID;
781 rc = kstrtoul(client_idstr, 10, &client_id);
783 GOTO(out_lcfg, rc = -EINVAL);
785 rc = nodemap_test_id(nid, idtype, client_id, &fs_id);
787 GOTO(out_lcfg, rc = -EINVAL);
789 if (data->ioc_plen1 < sizeof(fs_idstr))
790 GOTO(out_lcfg, rc = -EINVAL);
792 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
793 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
794 sizeof(fs_idstr)) != 0)
795 GOTO(out_lcfg, rc = -EINVAL);
797 case LCFG_NODEMAP_ADD_RANGE:
798 case LCFG_NODEMAP_DEL_RANGE:
799 case LCFG_NODEMAP_ADD_UIDMAP:
800 case LCFG_NODEMAP_DEL_UIDMAP:
801 case LCFG_NODEMAP_ADD_GIDMAP:
802 case LCFG_NODEMAP_DEL_GIDMAP:
803 case LCFG_NODEMAP_SET_FILESET:
804 case LCFG_NODEMAP_SET_SEPOL:
805 if (lcfg->lcfg_bufcount != 3)
806 GOTO(out_lcfg, rc = -EINVAL);
807 nodemap_name = lustre_cfg_string(lcfg, 1);
808 param = lustre_cfg_string(lcfg, 2);
809 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
811 case LCFG_NODEMAP_ADMIN:
812 case LCFG_NODEMAP_TRUSTED:
813 case LCFG_NODEMAP_DENY_UNKNOWN:
814 case LCFG_NODEMAP_SQUASH_UID:
815 case LCFG_NODEMAP_SQUASH_GID:
816 case LCFG_NODEMAP_MAP_MODE:
817 case LCFG_NODEMAP_AUDIT_MODE:
818 if (lcfg->lcfg_bufcount != 4)
819 GOTO(out_lcfg, rc = -EINVAL);
820 nodemap_name = lustre_cfg_string(lcfg, 1);
821 param = lustre_cfg_string(lcfg, 3);
822 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
829 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
830 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
835 /* revoke nodemap lock */
836 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
838 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
839 mgs->mgs_obd->obd_name, rc);
841 mgs_revoke_lock(mgs, fsdb, CONFIG_T_NODEMAP);
842 mgs_put_fsdb(mgs, fsdb);
846 OBD_FREE(lcfg, data->ioc_plen1);
851 static int mgs_iocontrol_pool(const struct lu_env *env,
852 struct mgs_device *mgs,
853 struct obd_ioctl_data *data)
855 struct mgs_thread_info *mgi = mgs_env_info(env);
857 struct lustre_cfg *lcfg = NULL;
858 char *poolname = NULL;
861 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
862 if (poolname == NULL)
865 if (data->ioc_type != LUSTRE_CFG_TYPE) {
866 CERROR("%s: unknown cfg record type: %d\n",
867 mgs->mgs_obd->obd_name, data->ioc_type);
868 GOTO(out_pool, rc = -EINVAL);
871 if (data->ioc_plen1 > PAGE_SIZE)
872 GOTO(out_pool, rc = -E2BIG);
874 OBD_ALLOC(lcfg, data->ioc_plen1);
876 GOTO(out_pool, rc = -ENOMEM);
878 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
879 GOTO(out_lcfg, rc = -EFAULT);
881 if (lcfg->lcfg_bufcount < 2)
882 GOTO(out_lcfg, rc = -EFAULT);
884 /* first arg is always <fsname>.<poolname> */
885 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
890 switch (lcfg->lcfg_command) {
892 if (lcfg->lcfg_bufcount != 2)
893 GOTO(out_lcfg, rc = -EINVAL);
894 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
898 if (lcfg->lcfg_bufcount != 3)
899 GOTO(out_lcfg, rc = -EINVAL);
900 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
901 poolname, lustre_cfg_string(lcfg, 2));
904 if (lcfg->lcfg_bufcount != 3)
905 GOTO(out_lcfg, rc = -EINVAL);
906 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
907 poolname, lustre_cfg_string(lcfg, 2));
910 if (lcfg->lcfg_bufcount != 2)
911 GOTO(out_lcfg, rc = -EINVAL);
912 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
920 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
921 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
926 OBD_FREE(lcfg, data->ioc_plen1);
928 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
932 /* from mdt_iocontrol */
933 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
934 void *karg, void __user *uarg)
936 struct mgs_device *mgs = exp2mgs_dev(exp);
937 struct obd_ioctl_data *data = karg;
942 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
944 rc = lu_env_init(&env, LCT_MG_THREAD);
950 case OBD_IOC_PARAM: {
951 struct lustre_cfg *lcfg;
953 if (data->ioc_type != LUSTRE_CFG_TYPE) {
954 CERROR("%s: unknown cfg record type: %d\n",
955 mgs->mgs_obd->obd_name, data->ioc_type);
956 GOTO(out, rc = -EINVAL);
959 OBD_ALLOC(lcfg, data->ioc_plen1);
961 GOTO(out, rc = -ENOMEM);
962 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
963 GOTO(out_free, rc = -EFAULT);
965 if (lcfg->lcfg_bufcount < 1)
966 GOTO(out_free, rc = -EINVAL);
968 rc = mgs_set_param(&env, mgs, lcfg);
970 CERROR("%s: setparam err: rc = %d\n",
971 exp->exp_obd->obd_name, rc);
973 OBD_FREE(lcfg, data->ioc_plen1);
977 case OBD_IOC_REPLACE_NIDS: {
978 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
980 CERROR("%s: no device or fsname specified: rc = %d\n",
981 exp->exp_obd->obd_name, rc);
985 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
987 CERROR("%s: device or fsname is too long: rc = %d\n",
988 exp->exp_obd->obd_name, rc);
992 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
994 CERROR("%s: device or fsname is not NUL terminated: "
995 "rc = %d\n", exp->exp_obd->obd_name, rc);
999 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1001 CERROR("%s: no NIDs specified: rc = %d\n",
1002 exp->exp_obd->obd_name, rc);
1006 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1008 CERROR("%s: NID list is not NUL terminated: "
1009 "rc = %d\n", exp->exp_obd->obd_name, rc);
1013 /* replace nids in llog */
1014 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1017 CERROR("%s: error replacing nids: rc = %d\n",
1018 exp->exp_obd->obd_name, rc);
1023 case OBD_IOC_CLEAR_CONFIGS: {
1024 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1026 CERROR("%s: no device or fsname specified: rc = %d\n",
1027 exp->exp_obd->obd_name, rc);
1031 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1033 CERROR("%s: device or fsname is too long: rc = %d\n",
1034 exp->exp_obd->obd_name, rc);
1038 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1040 CERROR("%s: device or fsname is not NUL terminated: "
1041 "rc = %d\n", exp->exp_obd->obd_name, rc);
1045 /* remove records marked SKIP from config logs */
1046 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1048 CERROR("%s: error clearing config log: rc = %d\n",
1049 exp->exp_obd->obd_name, rc);
1055 rc = mgs_iocontrol_pool(&env, mgs, data);
1058 case OBD_IOC_BARRIER:
1059 rc = mgs_iocontrol_barrier(&env, mgs, data);
1062 case OBD_IOC_NODEMAP:
1063 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1066 case OBD_IOC_LCFG_FORK:
1067 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1071 case OBD_IOC_LCFG_ERASE:
1072 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1075 case OBD_IOC_CATLOGLIST:
1076 rc = mgs_list_logs(&env, mgs, data);
1078 case OBD_IOC_LLOG_CANCEL:
1079 case OBD_IOC_LLOG_REMOVE:
1080 case OBD_IOC_LLOG_CHECK:
1081 case OBD_IOC_LLOG_INFO:
1082 case OBD_IOC_LLOG_PRINT: {
1083 struct llog_ctxt *ctxt;
1085 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1086 rc = llog_ioctl(&env, ctxt, cmd, data);
1087 llog_ctxt_put(ctxt);
1092 CERROR("%s: unknown command %#x\n",
1093 mgs->mgs_obd->obd_name, cmd);
1102 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1104 struct obd_connect_data *data = NULL;
1105 struct obd_device *obd;
1109 OBD_ALLOC_PTR(data);
1113 obd = class_name2obd(nextdev);
1115 CERROR("can't locate next device: %s\n", nextdev);
1116 GOTO(out, rc = -ENOTCONN);
1119 data->ocd_version = LUSTRE_VERSION_CODE;
1121 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1122 &obd->obd_uuid, data, NULL);
1124 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1128 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1129 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1130 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1136 static struct tgt_handler mgs_mgs_handlers[] = {
1137 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1138 0, MGS_CONNECT, mgs_connect,
1139 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1140 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1141 0, MGS_DISCONNECT, mgs_disconnect,
1142 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1143 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1144 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1145 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1146 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1149 static struct tgt_handler mgs_obd_handlers[] = {
1150 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1153 static struct tgt_handler mgs_dlm_handlers[] = {
1154 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1155 .th_name = "LDLM_ENQUEUE",
1156 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1157 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1158 * * is safe. If such tests will be needed we have to distinguish
1159 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1160 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1162 .th_opc = LDLM_ENQUEUE,
1163 .th_flags = HAS_KEY,
1164 .th_act = tgt_enqueue,
1165 .th_fmt = &RQF_LDLM_ENQUEUE,
1166 .th_version = LUSTRE_DLM_VERSION,
1170 static struct tgt_handler mgs_llog_handlers[] = {
1171 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1172 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1173 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1174 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1177 static struct tgt_opc_slice mgs_common_slice[] = {
1179 .tos_opc_start = MGS_FIRST_OPC,
1180 .tos_opc_end = MGS_LAST_OPC,
1181 .tos_hs = mgs_mgs_handlers
1184 .tos_opc_start = OBD_FIRST_OPC,
1185 .tos_opc_end = OBD_LAST_OPC,
1186 .tos_hs = mgs_obd_handlers
1189 .tos_opc_start = LDLM_FIRST_OPC,
1190 .tos_opc_end = LDLM_LAST_OPC,
1191 .tos_hs = mgs_dlm_handlers
1194 .tos_opc_start = LLOG_FIRST_OPC,
1195 .tos_opc_end = LLOG_LAST_OPC,
1196 .tos_hs = mgs_llog_handlers
1199 .tos_opc_start = SEC_FIRST_OPC,
1200 .tos_opc_end = SEC_LAST_OPC,
1201 .tos_hs = tgt_sec_ctx_handlers
1208 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1209 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1211 struct ptlrpc_service_conf conf;
1212 struct obd_device *obd;
1213 struct lustre_mount_info *lmi;
1214 struct llog_ctxt *ctxt;
1219 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1223 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1225 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1229 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1232 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1234 obd->u.obt.obt_magic = OBT_MAGIC;
1235 obd->u.obt.obt_instance = 0;
1237 /* namespace for mgs llog */
1238 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1239 LDLM_NAMESPACE_SERVER,
1240 LDLM_NAMESPACE_MODEST,
1242 if (obd->obd_namespace == NULL)
1243 GOTO(err_ops, rc = -ENOMEM);
1245 /* No recovery for MGCs */
1246 obd->obd_replayable = 0;
1248 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1249 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1250 OBD_FAIL_MGS_ALL_REPLY_NET);
1254 rc = mgs_fs_setup(env, mgs);
1256 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1261 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1262 obd, &llog_osd_ops);
1266 /* XXX: we need this trick till N:1 stack is supported
1267 * set "current" directory for named llogs */
1268 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1270 ctxt->loc_dir = mgs->mgs_configs_dir;
1271 llog_ctxt_put(ctxt);
1273 /* Internal mgs setup */
1274 mgs_init_fsdb_list(mgs);
1275 mutex_init(&mgs->mgs_mutex);
1276 mgs->mgs_start_time = ktime_get_real_seconds();
1277 spin_lock_init(&mgs->mgs_lock);
1278 mutex_init(&mgs->mgs_health_mutex);
1279 init_rwsem(&mgs->mgs_barrier_rwsem);
1281 rc = mgs_lcfg_rename(env, mgs);
1285 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1287 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1292 /* Setup params fsdb and log, so that other servers can make a local
1293 * copy successfully when they are mounted. See LU-4783 */
1294 rc = mgs_params_fsdb_setup(env, mgs);
1296 /* params fsdb and log can be setup later */
1297 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1298 obd->obd_name, PARAMS_FILENAME, rc);
1300 /* Setup _mgs fsdb, useful for srpc */
1301 mgs__mgs_fsdb_setup(env, mgs);
1303 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1304 "mgs_ldlm_client", &obd->obd_ldlm_client);
1306 conf = (typeof(conf)) {
1307 .psc_name = LUSTRE_MGS_NAME,
1308 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1310 .bc_nbufs = MGS_NBUFS,
1311 .bc_buf_size = MGS_BUFSIZE,
1312 .bc_req_max_size = MGS_MAXREQSIZE,
1313 .bc_rep_max_size = MGS_MAXREPSIZE,
1314 .bc_req_portal = MGS_REQUEST_PORTAL,
1315 .bc_rep_portal = MGC_REPLY_PORTAL,
1318 .tc_thr_name = "ll_mgs",
1319 .tc_nthrs_init = MGS_NTHRS_INIT,
1320 .tc_nthrs_max = MGS_NTHRS_MAX,
1321 .tc_ctx_tags = LCT_MG_THREAD,
1324 .so_req_handler = tgt_request_handle,
1325 .so_req_printer = target_print_req,
1329 /* Start the service threads */
1330 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1331 obd->obd_debugfs_entry);
1332 if (IS_ERR(mgs->mgs_service)) {
1333 rc = PTR_ERR(mgs->mgs_service);
1334 CERROR("failed to start mgs service: %d\n", rc);
1335 mgs->mgs_service = NULL;
1336 GOTO(err_lproc, rc);
1339 ping_evictor_start();
1341 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1343 /* device stack is not yet fully setup to keep no objects behind */
1344 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1347 mgs_params_fsdb_cleanup(env, mgs);
1348 lproc_mgs_cleanup(mgs);
1350 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1352 ctxt->loc_dir = NULL;
1353 llog_cleanup(env, ctxt);
1356 tgt_fini(env, &mgs->mgs_lut);
1358 /* No extra cleanup needed for llog_init_commit_thread() */
1359 mgs_fs_cleanup(env, mgs);
1361 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1362 obd->obd_namespace = NULL;
1364 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1365 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1366 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1367 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1370 obd_disconnect(mgs->mgs_bottom_exp);
1373 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1377 static struct lu_device *mgs_device_free(const struct lu_env *env,
1378 struct lu_device *lu)
1380 struct mgs_device *mgs = lu2mgs_dev(lu);
1383 dt_device_fini(&mgs->mgs_dt_dev);
1388 static int mgs_process_config(const struct lu_env *env,
1389 struct lu_device *dev,
1390 struct lustre_cfg *lcfg)
1396 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1397 const struct lu_object_conf *unused)
1399 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1400 struct lu_device *under;
1401 struct lu_object *below;
1405 /* do no set .do_ops as mgs calls to bottom osd directly */
1407 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1408 PFID(lu_object_fid(o)));
1410 under = &d->mgs_bottom->dd_lu_dev;
1411 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1413 lu_object_add(o, below);
1420 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1422 struct mgs_object *obj = lu2mgs_obj(o);
1423 struct lu_object_header *h = o->lo_header;
1425 dt_object_fini(&obj->mgo_obj);
1426 lu_object_header_fini(h);
1427 OBD_FREE_PRE(obj, sizeof(*obj), "kfreed");
1428 kfree_rcu(obj, mgo_header.loh_rcu);
1431 static int mgs_object_print(const struct lu_env *env, void *cookie,
1432 lu_printer_t p, const struct lu_object *l)
1434 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1436 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1439 static struct lu_object_operations mgs_lu_obj_ops = {
1440 .loo_object_init = mgs_object_init,
1441 .loo_object_free = mgs_object_free,
1442 .loo_object_print = mgs_object_print,
1445 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1446 const struct lu_object_header *hdr,
1447 struct lu_device *d)
1449 struct lu_object_header *h;
1450 struct mgs_object *o;
1451 struct lu_object *l;
1453 LASSERT(hdr == NULL);
1457 l = &o->mgo_obj.do_lu;
1460 lu_object_header_init(h);
1461 dt_object_init(&o->mgo_obj, h, d);
1462 lu_object_add_top(h, l);
1464 l->lo_ops = &mgs_lu_obj_ops;
1472 const struct lu_device_operations mgs_lu_ops = {
1473 .ldo_object_alloc = mgs_object_alloc,
1474 .ldo_process_config = mgs_process_config,
1477 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1478 struct lu_device_type *type,
1479 struct lustre_cfg *lcfg)
1481 struct mgs_device *mgs;
1482 struct lu_device *ludev;
1486 ludev = ERR_PTR(-ENOMEM);
1490 ludev = mgs2lu_dev(mgs);
1491 dt_device_init(&mgs->mgs_dt_dev, type);
1492 rc = mgs_init0(env, mgs, type, lcfg);
1494 mgs_device_free(env, ludev);
1495 ludev = ERR_PTR(rc);
1501 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1502 struct lu_device *d)
1504 struct mgs_device *mgs = lu2mgs_dev(d);
1505 struct obd_device *obd = mgs->mgs_obd;
1506 struct llog_ctxt *ctxt;
1510 LASSERT(mgs->mgs_bottom);
1512 class_disconnect_exports(obd);
1514 ping_evictor_stop();
1516 mutex_lock(&mgs->mgs_health_mutex);
1517 ptlrpc_unregister_service(mgs->mgs_service);
1518 mutex_unlock(&mgs->mgs_health_mutex);
1520 mgs_params_fsdb_cleanup(env, mgs);
1521 mgs_cleanup_fsdb_list(mgs);
1523 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1524 obd_exports_barrier(obd);
1525 obd_zombie_barrier();
1527 tgt_fini(env, &mgs->mgs_lut);
1528 lproc_mgs_cleanup(mgs);
1530 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1532 ctxt->loc_dir = NULL;
1533 llog_cleanup(env, ctxt);
1536 mgs_fs_cleanup(env, mgs);
1538 ldlm_namespace_free_post(obd->obd_namespace);
1539 obd->obd_namespace = NULL;
1541 lu_site_purge(env, d->ld_site, ~0);
1542 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1543 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1544 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1547 LASSERT(mgs->mgs_bottom_exp);
1548 obd_disconnect(mgs->mgs_bottom_exp);
1550 server_put_mount(obd->obd_name, true);
1555 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1556 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1558 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1560 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1562 static struct lu_device_type_operations mgs_device_type_ops = {
1563 .ldto_init = mgs_type_init,
1564 .ldto_fini = mgs_type_fini,
1566 .ldto_start = mgs_type_start,
1567 .ldto_stop = mgs_type_stop,
1569 .ldto_device_alloc = mgs_device_alloc,
1570 .ldto_device_free = mgs_device_free,
1572 .ldto_device_fini = mgs_device_fini
1575 static struct lu_device_type mgs_device_type = {
1576 .ldt_tags = LU_DEVICE_DT,
1577 .ldt_name = LUSTRE_MGS_NAME,
1578 .ldt_ops = &mgs_device_type_ops,
1579 .ldt_ctx_tags = LCT_MG_THREAD
1582 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1583 struct obd_device *obd, struct obd_uuid *cluuid,
1584 struct obd_connect_data *data, void *localdata)
1588 if (exp == NULL || obd == NULL || cluuid == NULL)
1591 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1594 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1596 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1597 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1599 exp->exp_connect_data = *data;
1600 data->ocd_version = LUSTRE_VERSION_CODE;
1603 RETURN(mgs_export_stats_init(obd, exp, localdata));
1606 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1607 struct obd_device *obd, struct obd_uuid *cluuid,
1608 struct obd_connect_data *data, void *localdata)
1610 struct obd_export *lexp;
1611 struct lustre_handle conn = {
1618 if (exp == NULL || obd == NULL || cluuid == NULL)
1621 rc = class_connect(&conn, obd, cluuid);
1625 lexp = class_conn2export(&conn);
1629 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1631 GOTO(out_disconnect, rc);
1638 class_disconnect(lexp);
1643 static int mgs_obd_disconnect(struct obd_export *exp)
1651 mgs_fsc_cleanup(exp);
1653 class_export_get(exp);
1654 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1656 rc = server_disconnect_export(exp);
1657 class_export_put(exp);
1661 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1663 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1666 mutex_lock(&mgs->mgs_health_mutex);
1667 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1668 mutex_unlock(&mgs->mgs_health_mutex);
1670 return rc != 0 ? 1 : 0;
1673 /* use obd ops to offer management infrastructure */
1674 static const struct obd_ops mgs_obd_device_ops = {
1675 .o_owner = THIS_MODULE,
1676 .o_connect = mgs_obd_connect,
1677 .o_reconnect = mgs_obd_reconnect,
1678 .o_disconnect = mgs_obd_disconnect,
1679 .o_init_export = mgs_init_export,
1680 .o_destroy_export = mgs_destroy_export,
1681 .o_iocontrol = mgs_iocontrol,
1682 .o_health_check = mgs_health_check,
1685 static int __init mgs_init(void)
1687 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1688 LUSTRE_MGS_NAME, &mgs_device_type);
1691 static void __exit mgs_exit(void)
1693 class_unregister_type(LUSTRE_MGS_NAME);
1696 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1697 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1698 MODULE_VERSION(LUSTRE_VERSION_STRING);
1699 MODULE_LICENSE("GPL");
1701 module_init(mgs_init);
1702 module_exit(mgs_exit);