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(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 registra"
341 "tion attempt from nid %s"
342 ", specified as failover"
343 "\n",libcfs_nid2str(nid));
344 return -EADDRNOTAVAIL;
352 /* Called whenever a target starts up. Flags indicate first connect, etc. */
353 static int mgs_target_reg(struct tgt_session_info *tsi)
355 struct obd_device *obd = tsi->tsi_exp->exp_obd;
356 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
357 struct mgs_target_info *mti, *rep_mti;
358 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
359 struct fs_db *c_fsdb = NULL; /* config fsdb */
360 char barrier_name[20];
366 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
368 return err_serious(rc);
370 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
372 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
374 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
375 RETURN(err_serious(-EFAULT));
378 down_read(&mgs->mgs_barrier_rwsem);
380 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
382 opc = mti->mti_flags & LDD_F_OPC_MASK;
386 if (opc == LDD_F_OPC_READY) {
387 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
388 mti->mti_fsname, mti->mti_stripe_index);
389 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
391 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
392 CERROR("%s: Update IR return failure: rc = %d\n",
393 mti->mti_fsname, rc);
396 GOTO(out_norevoke, rc);
399 /* Do not support unregistering right now. */
400 if (opc != LDD_F_OPC_REG)
401 GOTO(out_norevoke, rc = -EINVAL);
403 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
404 mti->mti_fsname, BARRIER_FILENAME);
405 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
407 CERROR("%s: Can't get db for %s: rc = %d\n",
408 mti->mti_fsname, barrier_name, rc);
410 GOTO(out_norevoke, rc);
413 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
414 mti->mti_fsname, mti->mti_stripe_index);
416 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
417 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
418 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
419 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
420 LCONSOLE_WARN("%s: the system is in barrier, refuse "
421 "the connection from MDT %s temporary\n",
422 obd->obd_name, mti->mti_svname);
424 GOTO(out_norevoke, rc = -EBUSY);
427 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
428 !b_fsdb->fsdb_barrier_disabled) {
429 LCONSOLE_WARN("%s: the MDT %s does not support write "
430 "barrier, so disable barrier on the "
432 obd->obd_name, mti->mti_svname);
434 b_fsdb->fsdb_barrier_disabled = 1;
438 if (mti->mti_flags & LDD_F_NEED_INDEX)
439 mti->mti_flags |= LDD_F_WRITECONF;
441 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
443 /* We're just here as a startup ping. */
444 CDEBUG(D_MGS, "Server %s is running on %s\n",
445 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
446 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
447 /* above will set appropriate mti flags */
449 /* Nothing wrong, or fatal error */
450 GOTO(out_norevoke, rc);
451 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
452 rc = mgs_check_failover_reg(mti);
454 GOTO(out_norevoke, rc);
457 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
459 if (mti->mti_flags & LDD_F_WRITECONF) {
460 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
461 mti->mti_stripe_index == 0) {
462 mgs_put_fsdb(mgs, b_fsdb);
464 rc = mgs_erase_logs(tsi->tsi_env, mgs,
466 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
467 "request. All servers must be restarted "
468 "in order to regenerate the logs: rc = %d"
469 "\n", obd->obd_name, mti->mti_fsname, rc);
470 if (rc && rc != -ENOENT)
471 GOTO(out_norevoke, rc);
473 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
474 barrier_name, &b_fsdb);
476 CERROR("Can't get db for %s: %d\n",
479 GOTO(out_norevoke, rc);
482 if (!(exp_connect_flags(tsi->tsi_exp) &
483 OBD_CONNECT_BARRIER)) {
484 LCONSOLE_WARN("%s: the MDT %s does not support "
485 "write barrier, disable barrier "
486 "on the whole system.\n",
487 obd->obd_name, mti->mti_svname);
489 b_fsdb->fsdb_barrier_disabled = 1;
491 } else if (mti->mti_flags &
492 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
493 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
494 LCONSOLE_WARN("%s: Regenerating %s log by user "
495 "request: rc = %d\n",
496 obd->obd_name, mti->mti_svname, rc);
498 GOTO(out_norevoke, rc);
501 mti->mti_flags |= LDD_F_UPDATE;
502 /* Erased logs means start from scratch. */
503 mti->mti_flags &= ~LDD_F_UPGRADE14;
506 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
508 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
510 GOTO(out_norevoke, rc);
514 * Log writing contention is handled by the fsdb_mutex.
516 * It should be alright if someone was reading while we were
517 * updating the logs - if we revoke at the end they will just update
518 * from where they left off.
521 if (mti->mti_flags & LDD_F_UPGRADE14) {
522 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
526 if (mti->mti_flags & LDD_F_UPDATE) {
527 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
528 mti->mti_stripe_index);
530 /* create or update the target log
531 and update the client/mdt logs */
532 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
534 CERROR("Failed to write %s log (%d)\n",
535 mti->mti_svname, rc);
539 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
540 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
542 mti->mti_flags |= LDD_F_REWRITE_LDD;
546 mgs_revoke_lock(mgs, c_fsdb, CONFIG_T_CONFIG);
549 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
551 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
552 mti->mti_fsname, &c_fsdb);
554 CERROR("Fail to get db for %s: %d\n",
555 mti->mti_fsname, rc);
559 memcpy(b_fsdb->fsdb_mdt_index_map,
560 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
561 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
565 up_read(&mgs->mgs_barrier_rwsem);
567 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
568 mti->mti_stripe_index, rc);
569 /* An error flag is set in the mti reply rather than an error code */
571 mti->mti_flags |= LDD_F_ERROR;
573 /* send back the whole mti in the reply */
574 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
577 /* Flush logs to disk */
578 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
580 mgs_put_fsdb(mgs, b_fsdb);
582 mgs_put_fsdb(mgs, c_fsdb);
586 /* Called whenever a target cleans up. */
587 static int mgs_target_del(struct tgt_session_info *tsi)
591 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
596 static int mgs_config_read(struct tgt_session_info *tsi)
598 struct ptlrpc_request *req = tgt_ses_req(tsi);
599 struct mgs_config_body *body;
604 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
606 DEBUG_REQ(D_HA, req, "no mgs_config_body");
607 RETURN(err_serious(-EFAULT));
610 switch (body->mcb_type) {
611 case CONFIG_T_RECOVER:
612 rc = mgs_get_ir_logs(req);
614 case CONFIG_T_NODEMAP:
615 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
617 case CONFIG_T_CONFIG:
628 static int mgs_llog_open(struct tgt_session_info *tsi)
630 struct mgs_thread_info *mgi;
631 struct ptlrpc_request *req = tgt_ses_req(tsi);
637 rc = tgt_llog_open(tsi);
642 * For old clients there is no direct way of knowing which file system
643 * a client is operating at the MGS side. But we need to pick up those
644 * clients so that the MGS can mark the corresponding file system as
645 * non-IR capable because old clients are not ready to be notified.
647 * Therefore we attempt to detect the file systems name by hacking the
648 * llog operation which is currently used by the clients to fetch
649 * configuration logs. At present this is fine because this is the
650 * ONLY llog operation between mgc and the MGS.
652 * If extra llog operation are going to be added, this function needs
655 * When releases prior than 2.0 are not supported, the following code
658 mgi = mgs_env_info(tsi->tsi_env);
660 RETURN(PTR_ERR(mgi));
662 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
664 char *ptr = strrchr(logname, '-');
665 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
667 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
668 if (strcmp(logname, PARAMS_FILENAME) != 0)
669 LCONSOLE_WARN("%s: non-config logname "
671 tgt_name(tsi->tsi_tgt),
673 /* not error, this can be llog test name */
675 strncpy(mgi->mgi_fsname, logname, len);
676 mgi->mgi_fsname[len] = 0;
678 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
680 if (rc && rc != -EEXIST) {
681 LCONSOLE_WARN("%s: Unable to add client %s "
682 "to file system %s: %d\n",
683 tgt_name(tsi->tsi_tgt),
684 libcfs_nid2str(req->rq_peer.nid),
685 mgi->mgi_fsname, rc);
691 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
697 static inline int mgs_init_export(struct obd_export *exp)
699 struct mgs_export_data *data = &exp->u.eu_mgs_data;
701 /* init mgs_export_data for fsc */
702 spin_lock_init(&data->med_lock);
703 INIT_LIST_HEAD(&data->med_clients);
705 spin_lock(&exp->exp_lock);
706 exp->exp_connecting = 1;
707 spin_unlock(&exp->exp_lock);
709 /* self-export doesn't need client data and ldlm initialization */
710 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
711 &exp->exp_client_uuid)))
713 return ldlm_init_export(exp);
716 static inline int mgs_destroy_export(struct obd_export *exp)
720 target_destroy_export(exp);
721 mgs_client_free(exp);
723 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
724 &exp->exp_client_uuid)))
727 ldlm_destroy_export(exp);
732 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
739 for (ptr = arg; *ptr != '\0'; ptr++) {
740 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
744 /* Test for fsname.poolname format.
745 * strlen() test if poolname is empty
747 ptr = strchr(arg, '.');
748 if (!ptr || !strlen(ptr))
752 /* Also make sure poolname is not to long. */
753 if (strlen(ptr) > LOV_MAXPOOLNAME)
754 return -ENAMETOOLONG;
755 strncpy(poolname, ptr, strlen(ptr));
757 /* Test if fsname is empty */
758 len = strlen(arg) - strlen(ptr) - 1;
763 if (len > LUSTRE_MAXFSNAME)
764 return -ENAMETOOLONG;
766 strncpy(fsname, arg, len);
771 static int mgs_iocontrol_nodemap(const struct lu_env *env,
772 struct mgs_device *mgs,
773 struct obd_ioctl_data *data)
775 struct lustre_cfg *lcfg = NULL;
778 const char *nodemap_name = NULL;
779 const char *nidstr = NULL;
780 const char *client_idstr = NULL;
781 const char *idtype_str = NULL;
784 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
786 unsigned long client_id;
793 if (data->ioc_type != LUSTRE_CFG_TYPE) {
794 CERROR("%s: unknown cfg record type: %d\n",
795 mgs->mgs_obd->obd_name, data->ioc_type);
796 GOTO(out, rc = -EINVAL);
799 if (data->ioc_plen1 > PAGE_SIZE)
800 GOTO(out, rc = -E2BIG);
802 OBD_ALLOC(lcfg, data->ioc_plen1);
804 GOTO(out, rc = -ENOMEM);
806 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
807 GOTO(out_lcfg, rc = -EFAULT);
809 cmd = lcfg->lcfg_command;
812 case LCFG_NODEMAP_ACTIVATE:
813 if (lcfg->lcfg_bufcount != 2)
814 GOTO(out_lcfg, rc = -EINVAL);
815 param = lustre_cfg_string(lcfg, 1);
816 if (strcmp(param, "1") == 0)
821 case LCFG_NODEMAP_ADD:
822 case LCFG_NODEMAP_DEL:
823 if (lcfg->lcfg_bufcount != 2)
824 GOTO(out_lcfg, rc = -EINVAL);
825 nodemap_name = lustre_cfg_string(lcfg, 1);
826 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
828 case LCFG_NODEMAP_TEST_NID:
829 if (lcfg->lcfg_bufcount != 2)
830 GOTO(out_lcfg, rc = -EINVAL);
831 nidstr = lustre_cfg_string(lcfg, 1);
832 nid = libcfs_str2nid(nidstr);
833 nodemap_test_nid(nid, name_buf, sizeof(name_buf));
834 rc = copy_to_user(data->ioc_pbuf1, name_buf,
835 MIN(data->ioc_plen1, sizeof(name_buf)));
837 GOTO(out_lcfg, rc = -EFAULT);
839 case LCFG_NODEMAP_TEST_ID:
840 if (lcfg->lcfg_bufcount != 4)
841 GOTO(out_lcfg, rc = -EINVAL);
842 nidstr = lustre_cfg_string(lcfg, 1);
843 idtype_str = lustre_cfg_string(lcfg, 2);
844 client_idstr = lustre_cfg_string(lcfg, 3);
846 nid = libcfs_str2nid(nidstr);
847 if (strcmp(idtype_str, "uid") == 0)
848 idtype = NODEMAP_UID;
850 idtype = NODEMAP_GID;
852 rc = kstrtoul(client_idstr, 10, &client_id);
854 GOTO(out_lcfg, rc = -EINVAL);
856 rc = nodemap_test_id(nid, idtype, client_id, &fs_id);
858 GOTO(out_lcfg, rc = -EINVAL);
860 if (data->ioc_plen1 < sizeof(fs_idstr))
861 GOTO(out_lcfg, rc = -EINVAL);
863 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
864 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
865 sizeof(fs_idstr)) != 0)
866 GOTO(out_lcfg, rc = -EINVAL);
868 case LCFG_NODEMAP_ADD_RANGE:
869 case LCFG_NODEMAP_DEL_RANGE:
870 case LCFG_NODEMAP_ADD_UIDMAP:
871 case LCFG_NODEMAP_DEL_UIDMAP:
872 case LCFG_NODEMAP_ADD_GIDMAP:
873 case LCFG_NODEMAP_DEL_GIDMAP:
874 case LCFG_NODEMAP_SET_FILESET:
875 if (lcfg->lcfg_bufcount != 3)
876 GOTO(out_lcfg, rc = -EINVAL);
877 nodemap_name = lustre_cfg_string(lcfg, 1);
878 param = lustre_cfg_string(lcfg, 2);
879 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
881 case LCFG_NODEMAP_ADMIN:
882 case LCFG_NODEMAP_TRUSTED:
883 case LCFG_NODEMAP_DENY_UNKNOWN:
884 case LCFG_NODEMAP_SQUASH_UID:
885 case LCFG_NODEMAP_SQUASH_GID:
886 case LCFG_NODEMAP_MAP_MODE:
887 case LCFG_NODEMAP_AUDIT_MODE:
888 if (lcfg->lcfg_bufcount != 4)
889 GOTO(out_lcfg, rc = -EINVAL);
890 nodemap_name = lustre_cfg_string(lcfg, 1);
891 param = lustre_cfg_string(lcfg, 3);
892 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
899 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
900 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
905 /* revoke nodemap lock */
906 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
908 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
909 mgs->mgs_obd->obd_name, rc);
911 mgs_revoke_lock(mgs, fsdb, CONFIG_T_NODEMAP);
912 mgs_put_fsdb(mgs, fsdb);
916 OBD_FREE(lcfg, data->ioc_plen1);
921 static int mgs_iocontrol_pool(const struct lu_env *env,
922 struct mgs_device *mgs,
923 struct obd_ioctl_data *data)
925 struct mgs_thread_info *mgi = mgs_env_info(env);
927 struct lustre_cfg *lcfg = NULL;
928 char *poolname = NULL;
931 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
932 if (poolname == NULL)
935 if (data->ioc_type != LUSTRE_CFG_TYPE) {
936 CERROR("%s: unknown cfg record type: %d\n",
937 mgs->mgs_obd->obd_name, data->ioc_type);
938 GOTO(out_pool, rc = -EINVAL);
941 if (data->ioc_plen1 > PAGE_SIZE)
942 GOTO(out_pool, rc = -E2BIG);
944 OBD_ALLOC(lcfg, data->ioc_plen1);
946 GOTO(out_pool, rc = -ENOMEM);
948 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
949 GOTO(out_lcfg, rc = -EFAULT);
951 if (lcfg->lcfg_bufcount < 2)
952 GOTO(out_lcfg, rc = -EFAULT);
954 /* first arg is always <fsname>.<poolname> */
955 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
960 switch (lcfg->lcfg_command) {
962 if (lcfg->lcfg_bufcount != 2)
963 GOTO(out_lcfg, rc = -EINVAL);
964 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
968 if (lcfg->lcfg_bufcount != 3)
969 GOTO(out_lcfg, rc = -EINVAL);
970 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
971 poolname, lustre_cfg_string(lcfg, 2));
974 if (lcfg->lcfg_bufcount != 3)
975 GOTO(out_lcfg, rc = -EINVAL);
976 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
977 poolname, lustre_cfg_string(lcfg, 2));
980 if (lcfg->lcfg_bufcount != 2)
981 GOTO(out_lcfg, rc = -EINVAL);
982 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
990 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
991 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
996 OBD_FREE(lcfg, data->ioc_plen1);
998 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1002 /* from mdt_iocontrol */
1003 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1004 void *karg, void __user *uarg)
1006 struct mgs_device *mgs = exp2mgs_dev(exp);
1007 struct obd_ioctl_data *data = karg;
1012 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1014 rc = lu_env_init(&env, LCT_MG_THREAD);
1020 case OBD_IOC_PARAM: {
1021 struct lustre_cfg *lcfg;
1023 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1024 CERROR("%s: unknown cfg record type: %d\n",
1025 mgs->mgs_obd->obd_name, data->ioc_type);
1026 GOTO(out, rc = -EINVAL);
1029 OBD_ALLOC(lcfg, data->ioc_plen1);
1031 GOTO(out, rc = -ENOMEM);
1032 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1033 GOTO(out_free, rc = -EFAULT);
1035 if (lcfg->lcfg_bufcount < 1)
1036 GOTO(out_free, rc = -EINVAL);
1038 rc = mgs_set_param(&env, mgs, lcfg);
1040 CERROR("%s: setparam err: rc = %d\n",
1041 exp->exp_obd->obd_name, rc);
1043 OBD_FREE(lcfg, data->ioc_plen1);
1047 case OBD_IOC_REPLACE_NIDS: {
1048 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1050 CERROR("%s: no device or fsname specified: rc = %d\n",
1051 exp->exp_obd->obd_name, rc);
1055 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1057 CERROR("%s: device or fsname is too long: rc = %d\n",
1058 exp->exp_obd->obd_name, rc);
1062 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1064 CERROR("%s: device or fsname is not NUL terminated: "
1065 "rc = %d\n", exp->exp_obd->obd_name, rc);
1069 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1071 CERROR("%s: no NIDs specified: rc = %d\n",
1072 exp->exp_obd->obd_name, rc);
1076 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1078 CERROR("%s: NID list is not NUL terminated: "
1079 "rc = %d\n", exp->exp_obd->obd_name, rc);
1083 /* replace nids in llog */
1084 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1087 CERROR("%s: error replacing nids: rc = %d\n",
1088 exp->exp_obd->obd_name, rc);
1093 case OBD_IOC_CLEAR_CONFIGS: {
1094 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1096 CERROR("%s: no device or fsname specified: rc = %d\n",
1097 exp->exp_obd->obd_name, rc);
1101 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1103 CERROR("%s: device or fsname is too long: rc = %d\n",
1104 exp->exp_obd->obd_name, rc);
1108 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1110 CERROR("%s: device or fsname is not NUL terminated: "
1111 "rc = %d\n", exp->exp_obd->obd_name, rc);
1115 /* remove records marked SKIP from config logs */
1116 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1118 CERROR("%s: error clearing config log: rc = %d\n",
1119 exp->exp_obd->obd_name, rc);
1125 rc = mgs_iocontrol_pool(&env, mgs, data);
1128 case OBD_IOC_BARRIER:
1129 rc = mgs_iocontrol_barrier(&env, mgs, data);
1132 case OBD_IOC_NODEMAP:
1133 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1136 case OBD_IOC_LCFG_FORK:
1137 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1141 case OBD_IOC_LCFG_ERASE:
1142 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1145 case OBD_IOC_CATLOGLIST:
1146 rc = mgs_list_logs(&env, mgs, data);
1148 case OBD_IOC_LLOG_CANCEL:
1149 case OBD_IOC_LLOG_REMOVE:
1150 case OBD_IOC_LLOG_CHECK:
1151 case OBD_IOC_LLOG_INFO:
1152 case OBD_IOC_LLOG_PRINT: {
1153 struct llog_ctxt *ctxt;
1155 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1156 rc = llog_ioctl(&env, ctxt, cmd, data);
1157 llog_ctxt_put(ctxt);
1162 CERROR("%s: unknown command %#x\n",
1163 mgs->mgs_obd->obd_name, cmd);
1172 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1174 struct obd_connect_data *data = NULL;
1175 struct obd_device *obd;
1179 OBD_ALLOC_PTR(data);
1183 obd = class_name2obd(nextdev);
1185 CERROR("can't locate next device: %s\n", nextdev);
1186 GOTO(out, rc = -ENOTCONN);
1189 data->ocd_version = LUSTRE_VERSION_CODE;
1191 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1192 &obd->obd_uuid, data, NULL);
1194 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1198 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1199 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1200 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1206 static struct tgt_handler mgs_mgs_handlers[] = {
1207 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1208 0, MGS_CONNECT, mgs_connect,
1209 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1210 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1211 0, MGS_DISCONNECT, mgs_disconnect,
1212 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1213 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1214 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
1215 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
1217 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
1218 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1219 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
1222 static struct tgt_handler mgs_obd_handlers[] = {
1223 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1226 static struct tgt_handler mgs_dlm_handlers[] = {
1227 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1228 .th_name = "LDLM_ENQUEUE",
1229 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1230 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1231 * * is safe. If such tests will be needed we have to distinguish
1232 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1233 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1235 .th_opc = LDLM_ENQUEUE,
1236 .th_flags = HABEO_CLAVIS,
1237 .th_act = tgt_enqueue,
1238 .th_fmt = &RQF_LDLM_ENQUEUE,
1239 .th_version = LUSTRE_DLM_VERSION,
1243 static struct tgt_handler mgs_llog_handlers[] = {
1244 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1245 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1246 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1247 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1248 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1251 static struct tgt_opc_slice mgs_common_slice[] = {
1253 .tos_opc_start = MGS_FIRST_OPC,
1254 .tos_opc_end = MGS_LAST_OPC,
1255 .tos_hs = mgs_mgs_handlers
1258 .tos_opc_start = OBD_FIRST_OPC,
1259 .tos_opc_end = OBD_LAST_OPC,
1260 .tos_hs = mgs_obd_handlers
1263 .tos_opc_start = LDLM_FIRST_OPC,
1264 .tos_opc_end = LDLM_LAST_OPC,
1265 .tos_hs = mgs_dlm_handlers
1268 .tos_opc_start = LLOG_FIRST_OPC,
1269 .tos_opc_end = LLOG_LAST_OPC,
1270 .tos_hs = mgs_llog_handlers
1273 .tos_opc_start = SEC_FIRST_OPC,
1274 .tos_opc_end = SEC_LAST_OPC,
1275 .tos_hs = tgt_sec_ctx_handlers
1282 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1283 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1285 struct ptlrpc_service_conf conf;
1286 struct obd_device *obd;
1287 struct lustre_mount_info *lmi;
1288 struct llog_ctxt *ctxt;
1293 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1297 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1299 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1303 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1306 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1308 obd->u.obt.obt_magic = OBT_MAGIC;
1309 obd->u.obt.obt_instance = 0;
1311 /* namespace for mgs llog */
1312 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1313 LDLM_NAMESPACE_SERVER,
1314 LDLM_NAMESPACE_MODEST,
1316 if (obd->obd_namespace == NULL)
1317 GOTO(err_ops, rc = -ENOMEM);
1319 /* No recovery for MGCs */
1320 obd->obd_replayable = 0;
1322 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1323 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1324 OBD_FAIL_MGS_ALL_REPLY_NET);
1328 rc = mgs_fs_setup(env, mgs);
1330 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1335 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1336 obd, &llog_osd_ops);
1340 /* XXX: we need this trick till N:1 stack is supported
1341 * set "current" directory for named llogs */
1342 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1344 ctxt->loc_dir = mgs->mgs_configs_dir;
1345 llog_ctxt_put(ctxt);
1347 /* Internal mgs setup */
1348 mgs_init_fsdb_list(mgs);
1349 mutex_init(&mgs->mgs_mutex);
1350 mgs->mgs_start_time = ktime_get_real_seconds();
1351 spin_lock_init(&mgs->mgs_lock);
1352 mutex_init(&mgs->mgs_health_mutex);
1353 init_rwsem(&mgs->mgs_barrier_rwsem);
1355 rc = mgs_lcfg_rename(env, mgs);
1359 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1361 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1366 /* Setup params fsdb and log, so that other servers can make a local
1367 * copy successfully when they are mounted. See LU-4783 */
1368 rc = mgs_params_fsdb_setup(env, mgs);
1370 /* params fsdb and log can be setup later */
1371 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1372 obd->obd_name, PARAMS_FILENAME, rc);
1374 /* Setup _mgs fsdb, useful for srpc */
1375 mgs__mgs_fsdb_setup(env, mgs);
1377 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1378 "mgs_ldlm_client", &obd->obd_ldlm_client);
1380 conf = (typeof(conf)) {
1381 .psc_name = LUSTRE_MGS_NAME,
1382 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1384 .bc_nbufs = MGS_NBUFS,
1385 .bc_buf_size = MGS_BUFSIZE,
1386 .bc_req_max_size = MGS_MAXREQSIZE,
1387 .bc_rep_max_size = MGS_MAXREPSIZE,
1388 .bc_req_portal = MGS_REQUEST_PORTAL,
1389 .bc_rep_portal = MGC_REPLY_PORTAL,
1392 .tc_thr_name = "ll_mgs",
1393 .tc_nthrs_init = MGS_NTHRS_INIT,
1394 .tc_nthrs_max = MGS_NTHRS_MAX,
1395 .tc_ctx_tags = LCT_MG_THREAD,
1398 .so_req_handler = tgt_request_handle,
1399 .so_req_printer = target_print_req,
1403 /* Start the service threads */
1404 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1405 obd->obd_proc_entry);
1406 if (IS_ERR(mgs->mgs_service)) {
1407 rc = PTR_ERR(mgs->mgs_service);
1408 CERROR("failed to start mgs service: %d\n", rc);
1409 mgs->mgs_service = NULL;
1410 GOTO(err_lproc, rc);
1413 ping_evictor_start();
1415 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1417 /* device stack is not yet fully setup to keep no objects behind */
1418 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1421 mgs_params_fsdb_cleanup(env, mgs);
1422 lproc_mgs_cleanup(mgs);
1424 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1426 ctxt->loc_dir = NULL;
1427 llog_cleanup(env, ctxt);
1430 tgt_fini(env, &mgs->mgs_lut);
1432 /* No extra cleanup needed for llog_init_commit_thread() */
1433 mgs_fs_cleanup(env, mgs);
1435 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1436 obd->obd_namespace = NULL;
1438 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1439 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1440 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1441 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1444 obd_disconnect(mgs->mgs_bottom_exp);
1447 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1451 static struct lu_device *mgs_device_free(const struct lu_env *env,
1452 struct lu_device *lu)
1454 struct mgs_device *mgs = lu2mgs_dev(lu);
1457 dt_device_fini(&mgs->mgs_dt_dev);
1462 static int mgs_process_config(const struct lu_env *env,
1463 struct lu_device *dev,
1464 struct lustre_cfg *lcfg)
1470 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1471 const struct lu_object_conf *unused)
1473 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1474 struct lu_device *under;
1475 struct lu_object *below;
1479 /* do no set .do_ops as mgs calls to bottom osd directly */
1481 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1482 PFID(lu_object_fid(o)));
1484 under = &d->mgs_bottom->dd_lu_dev;
1485 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1487 lu_object_add(o, below);
1494 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1496 struct mgs_object *obj = lu2mgs_obj(o);
1497 struct lu_object_header *h = o->lo_header;
1499 dt_object_fini(&obj->mgo_obj);
1500 lu_object_header_fini(h);
1504 static int mgs_object_print(const struct lu_env *env, void *cookie,
1505 lu_printer_t p, const struct lu_object *l)
1507 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1509 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1512 static struct lu_object_operations mgs_lu_obj_ops = {
1513 .loo_object_init = mgs_object_init,
1514 .loo_object_free = mgs_object_free,
1515 .loo_object_print = mgs_object_print,
1518 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1519 const struct lu_object_header *hdr,
1520 struct lu_device *d)
1522 struct lu_object_header *h;
1523 struct mgs_object *o;
1524 struct lu_object *l;
1526 LASSERT(hdr == NULL);
1530 l = &o->mgo_obj.do_lu;
1533 lu_object_header_init(h);
1534 dt_object_init(&o->mgo_obj, h, d);
1535 lu_object_add_top(h, l);
1537 l->lo_ops = &mgs_lu_obj_ops;
1545 const struct lu_device_operations mgs_lu_ops = {
1546 .ldo_object_alloc = mgs_object_alloc,
1547 .ldo_process_config = mgs_process_config,
1550 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1551 struct lu_device_type *type,
1552 struct lustre_cfg *lcfg)
1554 struct mgs_device *mgs;
1555 struct lu_device *ludev;
1559 ludev = ERR_PTR(-ENOMEM);
1563 ludev = mgs2lu_dev(mgs);
1564 dt_device_init(&mgs->mgs_dt_dev, type);
1565 rc = mgs_init0(env, mgs, type, lcfg);
1567 mgs_device_free(env, ludev);
1568 ludev = ERR_PTR(rc);
1574 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1575 struct lu_device *d)
1577 struct mgs_device *mgs = lu2mgs_dev(d);
1578 struct obd_device *obd = mgs->mgs_obd;
1579 struct llog_ctxt *ctxt;
1583 LASSERT(mgs->mgs_bottom);
1585 class_disconnect_exports(obd);
1587 ping_evictor_stop();
1589 mutex_lock(&mgs->mgs_health_mutex);
1590 ptlrpc_unregister_service(mgs->mgs_service);
1591 mutex_unlock(&mgs->mgs_health_mutex);
1593 mgs_params_fsdb_cleanup(env, mgs);
1594 mgs_cleanup_fsdb_list(mgs);
1596 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1597 obd_exports_barrier(obd);
1598 obd_zombie_barrier();
1600 tgt_fini(env, &mgs->mgs_lut);
1601 lproc_mgs_cleanup(mgs);
1603 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1605 ctxt->loc_dir = NULL;
1606 llog_cleanup(env, ctxt);
1609 mgs_fs_cleanup(env, mgs);
1611 ldlm_namespace_free_post(obd->obd_namespace);
1612 obd->obd_namespace = NULL;
1614 lu_site_purge(env, d->ld_site, ~0);
1615 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1616 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1617 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1620 LASSERT(mgs->mgs_bottom_exp);
1621 obd_disconnect(mgs->mgs_bottom_exp);
1623 server_put_mount(obd->obd_name, true);
1628 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1629 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1631 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1633 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1635 static struct lu_device_type_operations mgs_device_type_ops = {
1636 .ldto_init = mgs_type_init,
1637 .ldto_fini = mgs_type_fini,
1639 .ldto_start = mgs_type_start,
1640 .ldto_stop = mgs_type_stop,
1642 .ldto_device_alloc = mgs_device_alloc,
1643 .ldto_device_free = mgs_device_free,
1645 .ldto_device_fini = mgs_device_fini
1648 static struct lu_device_type mgs_device_type = {
1649 .ldt_tags = LU_DEVICE_DT,
1650 .ldt_name = LUSTRE_MGS_NAME,
1651 .ldt_ops = &mgs_device_type_ops,
1652 .ldt_ctx_tags = LCT_MG_THREAD
1655 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1656 struct obd_device *obd, struct obd_uuid *cluuid,
1657 struct obd_connect_data *data, void *localdata)
1661 if (exp == NULL || obd == NULL || cluuid == NULL)
1664 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1667 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1669 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1670 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1672 exp->exp_connect_data = *data;
1673 data->ocd_version = LUSTRE_VERSION_CODE;
1676 RETURN(mgs_export_stats_init(obd, exp, localdata));
1679 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1680 struct obd_device *obd, struct obd_uuid *cluuid,
1681 struct obd_connect_data *data, void *localdata)
1683 struct obd_export *lexp;
1684 struct lustre_handle conn = {
1691 if (exp == NULL || obd == NULL || cluuid == NULL)
1694 rc = class_connect(&conn, obd, cluuid);
1698 lexp = class_conn2export(&conn);
1702 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1704 GOTO(out_disconnect, rc);
1711 class_disconnect(lexp);
1716 static int mgs_obd_disconnect(struct obd_export *exp)
1724 mgs_fsc_cleanup(exp);
1726 class_export_get(exp);
1727 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1729 rc = server_disconnect_export(exp);
1730 class_export_put(exp);
1734 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1736 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1739 mutex_lock(&mgs->mgs_health_mutex);
1740 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1741 mutex_unlock(&mgs->mgs_health_mutex);
1743 return rc != 0 ? 1 : 0;
1746 /* use obd ops to offer management infrastructure */
1747 static struct obd_ops mgs_obd_device_ops = {
1748 .o_owner = THIS_MODULE,
1749 .o_connect = mgs_obd_connect,
1750 .o_reconnect = mgs_obd_reconnect,
1751 .o_disconnect = mgs_obd_disconnect,
1752 .o_init_export = mgs_init_export,
1753 .o_destroy_export = mgs_destroy_export,
1754 .o_iocontrol = mgs_iocontrol,
1755 .o_health_check = mgs_health_check,
1758 static int __init mgs_init(void)
1760 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1761 LUSTRE_MGS_NAME, &mgs_device_type);
1764 static void __exit mgs_exit(void)
1766 class_unregister_type(LUSTRE_MGS_NAME);
1769 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1770 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1771 MODULE_VERSION(LUSTRE_VERSION_STRING);
1772 MODULE_LICENSE("GPL");
1774 module_init(mgs_init);
1775 module_exit(mgs_exit);