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.
156 lock_res_and_lock(lock);
157 fsdb = (struct fs_db *)lock->l_ast_data;
158 lock->l_ast_data = NULL;
159 unlock_res_and_lock(lock);
162 struct lustre_handle lockh;
166 /* clear the bit before lock put */
167 clear_bit(FSDB_REVOKING_LOCK,
171 clear_bit(FSDB_REVOKING_PARAMS,
175 mgs_ir_notify_complete(fsdb);
183 ldlm_lock2handle(lock, &lockh);
184 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
188 RETURN(ldlm_completion_ast(lock, flags, cbdata));
191 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
194 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
197 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
200 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
203 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
206 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
209 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
212 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
215 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb,
216 enum mgs_cfg_type type)
218 ldlm_completion_callback cp = NULL;
219 struct lustre_handle lockh = {
222 struct ldlm_res_id res_id;
223 __u64 flags = LDLM_FL_ATOMIC_CB;
228 LASSERT(fsdb->fsdb_name[0] != '\0');
229 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
232 case MGS_CFG_T_CONFIG:
233 case MGS_CFG_T_NODEMAP:
234 cp = mgs_completion_ast_config;
235 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
238 case MGS_CFG_T_PARAMS:
239 cp = mgs_completion_ast_params;
240 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
243 case MGS_CFG_T_RECOVER:
244 cp = mgs_completion_ast_ir;
246 case MGS_CFG_T_BARRIER:
247 cp = mgs_completion_ast_barrier;
255 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
256 &res_id, LDLM_PLAIN, NULL, LCK_EX,
257 &flags, ldlm_blocking_ast, cp,
258 NULL, fsdb, 0, LVB_T_NONE, NULL,
260 if (rc != ELDLM_OK) {
261 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
262 mgs->mgs_obd->obd_name,
263 le64_to_cpu(res_id.name[0]),
264 le64_to_cpu(res_id.name[1]), rc);
266 if (type == MGS_CFG_T_CONFIG)
267 clear_bit(FSDB_REVOKING_LOCK,
270 if (type == MGS_CFG_T_PARAMS)
271 clear_bit(FSDB_REVOKING_PARAMS,
274 /* lock has been cancelled in completion_ast. */
280 /* Returns: 0 on Success
281 * Returns: 1 means update
282 * Returns: <0 means error
284 static int mgs_check_target(const struct lu_env *env,
285 struct mgs_device *mgs,
286 struct mgs_target_info *mti)
292 rc = mgs_check_index(env, mgs, mti);
294 LCONSOLE_ERROR_MSG(0x13b,
295 "%s claims to have registered, but this MGS does not know about it, preventing registration.\n",
298 } else if (rc == -1) {
299 LCONSOLE_ERROR_MSG(0x13c,
300 "Client log %s-client has disappeared! Regenerating all logs.\n",
302 mti->mti_flags |= LDD_F_WRITECONF;
305 /* Index is correctly marked as used */
312 /* Ensure this is not a failover node that is connecting first*/
313 static int mgs_check_failover_reg(struct mgs_target_info *mti)
319 ptr = mti->mti_params;
320 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
321 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
322 for (i = 0; i < mti->mti_nid_count; i++) {
323 struct lnet_nid nid2;
326 if (target_supports_large_nid(mti)) {
327 rc = libcfs_strnid(&nid2,
328 mti->mti_nidlist[i]);
330 LCONSOLE_WARN("NID %s is unsupported type or improper format\n",
331 libcfs_nidstr(&nid));
335 lnet_nid4_to_nid(mti->mti_nids[i],
339 if (nid_same(&nid, &nid2)) {
340 LCONSOLE_WARN("Denying initial registration attempt from nid %s, specified as failover\n",
341 libcfs_nidstr(&nid));
342 return -EADDRNOTAVAIL;
350 /* Called whenever a target starts up. Flags indicate first connect, etc. */
351 static int mgs_target_reg(struct tgt_session_info *tsi)
353 struct obd_device *obd = tsi->tsi_exp->exp_obd;
354 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
355 struct mgs_target_info *mti, *rep_mti;
356 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
357 struct fs_db *c_fsdb = NULL; /* config fsdb */
358 char barrier_name[20];
364 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
366 return err_serious(rc);
368 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
370 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
372 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
373 RETURN(err_serious(-EFAULT));
376 down_read(&mgs->mgs_barrier_rwsem);
378 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
380 opc = mti->mti_flags & LDD_F_OPC_MASK;
384 if (opc == LDD_F_OPC_READY) {
385 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
386 mti->mti_fsname, mti->mti_stripe_index);
387 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
389 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
390 CERROR("%s: Update IR return failure: rc = %d\n",
391 mti->mti_fsname, rc);
394 GOTO(out_norevoke, rc);
397 /* Do not support unregistering right now. */
398 if (opc != LDD_F_OPC_REG)
399 GOTO(out_norevoke, rc = -EINVAL);
401 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
402 mti->mti_fsname, BARRIER_FILENAME);
403 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
405 CERROR("%s: Can't get db for %s: rc = %d\n",
406 mti->mti_fsname, barrier_name, rc);
408 GOTO(out_norevoke, rc);
411 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
412 mti->mti_fsname, mti->mti_stripe_index);
414 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
415 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
416 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
417 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
418 LCONSOLE_WARN("%s: the system is in barrier, refuse the connection from MDT %s temporary\n",
419 obd->obd_name, mti->mti_svname);
421 GOTO(out_norevoke, rc = -EBUSY);
424 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
425 !b_fsdb->fsdb_barrier_disabled) {
426 LCONSOLE_WARN("%s: the MDT %s does not support write barrier, so disable barrier on the whole system.\n",
427 obd->obd_name, mti->mti_svname);
429 b_fsdb->fsdb_barrier_disabled = 1;
433 if (mti->mti_flags & LDD_F_NEED_INDEX)
434 mti->mti_flags |= LDD_F_WRITECONF;
436 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
437 /* We're just here as a startup ping. */
438 CDEBUG(D_MGS, "Server %s is running on %s\n",
439 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
440 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
441 /* above will set appropriate mti flags */
443 /* Nothing wrong, or fatal error */
444 GOTO(out_norevoke, rc);
445 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
446 rc = mgs_check_failover_reg(mti);
448 GOTO(out_norevoke, rc);
451 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
453 if (mti->mti_flags & LDD_F_WRITECONF) {
454 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
455 mti->mti_stripe_index == 0) {
456 mgs_put_fsdb(mgs, b_fsdb);
458 rc = mgs_erase_logs(tsi->tsi_env, mgs,
460 LCONSOLE_WARN("%s: Logs for fs %s were removed by user request. All servers must be restarted in order to regenerate the logs: rc = %d\n",
461 obd->obd_name, mti->mti_fsname, rc);
462 if (rc && rc != -ENOENT)
463 GOTO(out_norevoke, rc);
465 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
466 barrier_name, &b_fsdb);
468 CERROR("Can't get db for %s: %d\n",
471 GOTO(out_norevoke, rc);
474 if (!(exp_connect_flags(tsi->tsi_exp) &
475 OBD_CONNECT_BARRIER)) {
476 LCONSOLE_WARN("%s: the MDT %s does not support write barrier, disable barrier on the whole system.\n",
477 obd->obd_name, mti->mti_svname);
479 b_fsdb->fsdb_barrier_disabled = 1;
481 } else if (mti->mti_flags &
482 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
483 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
484 LCONSOLE_WARN("%s: Regenerating %s log by user request: rc = %d\n",
485 obd->obd_name, mti->mti_svname, rc);
487 GOTO(out_norevoke, rc);
490 mti->mti_flags |= LDD_F_UPDATE;
493 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
495 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
497 GOTO(out_norevoke, rc);
501 * Log writing contention is handled by the fsdb_mutex.
503 * It should be alright if someone was reading while we were
504 * updating the logs - if we revoke at the end they will just update
505 * from where they left off.
507 if (mti->mti_flags & LDD_F_UPDATE) {
508 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
509 mti->mti_stripe_index);
511 /* create/update target log and update the client/mdt logs */
512 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
514 CERROR("Failed to write %s log (%d)\n",
515 mti->mti_svname, rc);
519 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
520 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
521 mti->mti_flags |= LDD_F_REWRITE_LDD;
525 mgs_revoke_lock(mgs, c_fsdb, MGS_CFG_T_CONFIG);
528 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
530 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
531 mti->mti_fsname, &c_fsdb);
533 CERROR("Fail to get db for %s: %d\n",
534 mti->mti_fsname, rc);
538 memcpy(b_fsdb->fsdb_mdt_index_map,
539 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
540 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
544 up_read(&mgs->mgs_barrier_rwsem);
546 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
547 mti->mti_stripe_index, rc);
548 /* An error flag is set in the mti reply rather than an error code */
550 mti->mti_flags |= LDD_F_ERROR;
552 /* send back the whole mti in the reply */
553 if (target_supports_large_nid(mti)) {
554 size_t len = offsetof(struct mgs_target_info, mti_nidlist);
557 mti_len = mti->mti_nid_count * LNET_NIDSTR_SIZE;
558 err = req_capsule_server_grow(tsi->tsi_pill,
559 &RMF_MGS_TARGET_INFO,
564 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
566 if (target_supports_large_nid(mti))
567 memcpy(rep_mti->mti_nidlist, mti->mti_nidlist, mti_len);
569 /* Flush logs to disk */
570 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
572 mgs_put_fsdb(mgs, b_fsdb);
574 mgs_put_fsdb(mgs, c_fsdb);
578 /* Called whenever a target cleans up. */
579 static int mgs_target_del(struct tgt_session_info *tsi)
583 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
588 static int mgs_config_read(struct tgt_session_info *tsi)
590 struct ptlrpc_request *req = tgt_ses_req(tsi);
591 struct mgs_config_body *body;
596 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
598 DEBUG_REQ(D_HA, req, "no mgs_config_body");
599 RETURN(err_serious(-EFAULT));
602 switch (body->mcb_type) {
603 case MGS_CFG_T_RECOVER:
604 rc = mgs_get_ir_logs(req);
606 case MGS_CFG_T_NODEMAP:
607 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
609 case MGS_CFG_T_CONFIG:
620 static int mgs_llog_open(struct tgt_session_info *tsi)
622 struct mgs_thread_info *mgi;
623 struct ptlrpc_request *req = tgt_ses_req(tsi);
629 rc = tgt_llog_open(tsi);
634 * For old clients there is no direct way of knowing which file system
635 * a client is operating at the MGS side. But we need to pick up those
636 * clients so that the MGS can mark the corresponding file system as
637 * non-IR capable because old clients are not ready to be notified.
639 * Therefore we attempt to detect the file systems name by hacking the
640 * llog operation which is currently used by the clients to fetch
641 * configuration logs. At present this is fine because this is the
642 * ONLY llog operation between mgc and the MGS.
644 * If extra llog operation are going to be added, this function needs
647 * When releases prior than 2.0 are not supported, the following code
650 mgi = mgs_env_info(tsi->tsi_env);
652 RETURN(PTR_ERR(mgi));
654 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
656 char *ptr = strrchr(logname, '-');
657 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
659 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
660 if (strcmp(logname, PARAMS_FILENAME) != 0)
661 LCONSOLE_WARN("%s: non-config logname received: %s\n",
662 tgt_name(tsi->tsi_tgt),
664 /* not error, this can be llog test name */
666 strncpy(mgi->mgi_fsname, logname, len);
667 mgi->mgi_fsname[len] = 0;
669 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
671 if (rc && rc != -EEXIST) {
672 LCONSOLE_WARN("%s: Unable to add client %s to file system %s: %d\n",
673 tgt_name(tsi->tsi_tgt),
674 libcfs_nidstr(&req->rq_peer.nid),
675 mgi->mgi_fsname, rc);
681 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
687 static inline int mgs_init_export(struct obd_export *exp)
689 struct mgs_export_data *data = &exp->u.eu_mgs_data;
691 /* init mgs_export_data for fsc */
692 spin_lock_init(&data->med_lock);
693 INIT_LIST_HEAD(&data->med_clients);
695 spin_lock(&exp->exp_lock);
696 exp->exp_connecting = 1;
697 spin_unlock(&exp->exp_lock);
699 /* self-export doesn't need client data and ldlm initialization */
700 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
701 &exp->exp_client_uuid)))
703 return ldlm_init_export(exp);
706 static inline int mgs_destroy_export(struct obd_export *exp)
710 target_destroy_export(exp);
711 mgs_client_free(exp);
713 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
714 &exp->exp_client_uuid)))
717 ldlm_destroy_export(exp);
722 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
729 for (ptr = arg; *ptr != '\0'; ptr++) {
730 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
734 /* Test for fsname.poolname format. strlen test if poolname is empty */
735 ptr = strchr(arg, '.');
736 if (!ptr || !strlen(ptr))
740 /* Check pool name validity. */
741 if (ptr[0] == '\0' || lov_pool_is_reserved(ptr))
743 /* Also make sure poolname is not to long. */
744 if (strlen(ptr) > LOV_MAXPOOLNAME)
745 return -ENAMETOOLONG;
746 strscpy(poolname, ptr, LOV_MAXPOOLNAME + 1);
748 /* Test if fsname is empty */
749 len = strlen(arg) - strlen(ptr) - 1;
754 if (len > LUSTRE_MAXFSNAME)
755 return -ENAMETOOLONG;
757 strncpy(fsname, arg, len);
762 static int mgs_iocontrol_nodemap(const struct lu_env *env,
763 struct mgs_device *mgs,
764 struct obd_ioctl_data *data)
766 struct lustre_cfg *lcfg = NULL;
769 const char *nodemap_name = NULL;
770 const char *nidstr = NULL;
771 const char *client_idstr = NULL;
772 const char *idtype_str = NULL;
775 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
777 unsigned long client_id;
784 if (data->ioc_type != LUSTRE_CFG_TYPE) {
785 CERROR("%s: unknown cfg record type: %d\n",
786 mgs->mgs_obd->obd_name, data->ioc_type);
787 GOTO(out, rc = -EINVAL);
790 if (data->ioc_plen1 > PAGE_SIZE)
791 GOTO(out, rc = -E2BIG);
793 OBD_ALLOC(lcfg, data->ioc_plen1);
795 GOTO(out, rc = -ENOMEM);
797 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
798 GOTO(out_lcfg, rc = -EFAULT);
800 cmd = lcfg->lcfg_command;
803 case LCFG_NODEMAP_ACTIVATE:
804 if (lcfg->lcfg_bufcount != 2)
805 GOTO(out_lcfg, rc = -EINVAL);
806 param = lustre_cfg_string(lcfg, 1);
807 if (strcmp(param, "1") == 0)
812 case LCFG_NODEMAP_ADD:
813 case LCFG_NODEMAP_DEL:
814 if (lcfg->lcfg_bufcount != 2)
815 GOTO(out_lcfg, rc = -EINVAL);
816 nodemap_name = lustre_cfg_string(lcfg, 1);
817 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
819 case LCFG_NODEMAP_TEST_NID:
820 if (lcfg->lcfg_bufcount != 2)
821 GOTO(out_lcfg, rc = -EINVAL);
822 nidstr = lustre_cfg_string(lcfg, 1);
823 rc = libcfs_strnid(&nid, nidstr);
827 nodemap_test_nid(&nid, name_buf, sizeof(name_buf));
828 rc = copy_to_user(data->ioc_pbuf1, name_buf,
829 min_t(size_t, data->ioc_plen1,
832 GOTO(out_lcfg, rc = -EFAULT);
834 case LCFG_NODEMAP_TEST_ID:
835 if (lcfg->lcfg_bufcount != 4)
836 GOTO(out_lcfg, rc = -EINVAL);
837 nidstr = lustre_cfg_string(lcfg, 1);
838 idtype_str = lustre_cfg_string(lcfg, 2);
839 client_idstr = lustre_cfg_string(lcfg, 3);
841 rc = libcfs_strnid(&nid, nidstr);
845 if (strcmp(idtype_str, "uid") == 0)
846 idtype = NODEMAP_UID;
847 else if (strcmp(idtype_str, "gid") == 0)
848 idtype = NODEMAP_GID;
849 else if (strcmp(idtype_str, "projid") == 0)
850 idtype = NODEMAP_PROJID;
852 GOTO(out_lcfg, rc = -EINVAL);
854 rc = kstrtoul(client_idstr, 10, &client_id);
856 GOTO(out_lcfg, rc = -EINVAL);
858 rc = nodemap_test_id(&nid, idtype, client_id, &fs_id);
860 GOTO(out_lcfg, rc = -EINVAL);
862 if (data->ioc_plen1 < sizeof(fs_idstr))
863 GOTO(out_lcfg, rc = -EINVAL);
865 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
866 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
867 sizeof(fs_idstr)) != 0)
868 GOTO(out_lcfg, rc = -EINVAL);
870 case LCFG_NODEMAP_ADD_RANGE:
871 case LCFG_NODEMAP_DEL_RANGE:
872 case LCFG_NODEMAP_ADD_UIDMAP:
873 case LCFG_NODEMAP_DEL_UIDMAP:
874 case LCFG_NODEMAP_ADD_GIDMAP:
875 case LCFG_NODEMAP_DEL_GIDMAP:
876 case LCFG_NODEMAP_ADD_PROJIDMAP:
877 case LCFG_NODEMAP_DEL_PROJIDMAP:
878 case LCFG_NODEMAP_SET_FILESET:
879 case LCFG_NODEMAP_SET_SEPOL:
880 if (lcfg->lcfg_bufcount != 3)
881 GOTO(out_lcfg, rc = -EINVAL);
882 nodemap_name = lustre_cfg_string(lcfg, 1);
883 param = lustre_cfg_string(lcfg, 2);
884 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
886 case LCFG_NODEMAP_ADMIN:
887 case LCFG_NODEMAP_TRUSTED:
888 case LCFG_NODEMAP_DENY_UNKNOWN:
889 case LCFG_NODEMAP_SQUASH_UID:
890 case LCFG_NODEMAP_SQUASH_GID:
891 case LCFG_NODEMAP_SQUASH_PROJID:
892 case LCFG_NODEMAP_MAP_MODE:
893 case LCFG_NODEMAP_AUDIT_MODE:
894 case LCFG_NODEMAP_FORBID_ENCRYPT:
895 case LCFG_NODEMAP_READONLY_MOUNT:
896 case LCFG_NODEMAP_RBAC:
897 if (lcfg->lcfg_bufcount != 4)
898 GOTO(out_lcfg, rc = -EINVAL);
899 nodemap_name = lustre_cfg_string(lcfg, 1);
900 param = lustre_cfg_string(lcfg, 3);
901 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
908 CDEBUG_LIMIT(rc == -EEXIST ? D_INFO : D_ERROR,
909 "%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
910 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
915 /* revoke nodemap lock */
916 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
918 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
919 mgs->mgs_obd->obd_name, rc);
921 mgs_revoke_lock(mgs, fsdb, MGS_CFG_T_NODEMAP);
922 mgs_put_fsdb(mgs, fsdb);
926 OBD_FREE(lcfg, data->ioc_plen1);
931 static int mgs_iocontrol_pool(const struct lu_env *env,
932 struct mgs_device *mgs,
933 struct obd_ioctl_data *data)
935 struct mgs_thread_info *mgi = mgs_env_info(env);
937 struct lustre_cfg *lcfg = NULL;
938 char *poolname = NULL;
942 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
943 if (poolname == NULL)
946 if (data->ioc_type != LUSTRE_CFG_TYPE) {
947 CERROR("%s: unknown cfg record type: %d\n",
948 mgs->mgs_obd->obd_name, data->ioc_type);
949 GOTO(out_pool, rc = -EINVAL);
952 if (data->ioc_plen1 > PAGE_SIZE)
953 GOTO(out_pool, rc = -E2BIG);
955 OBD_ALLOC(lcfg, data->ioc_plen1);
957 GOTO(out_pool, rc = -ENOMEM);
959 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
960 GOTO(out_lcfg, rc = -EFAULT);
962 if (lcfg->lcfg_bufcount < 2)
963 GOTO(out_lcfg, rc = -EFAULT);
965 /* first arg is always <fsname>.<poolname> */
966 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
971 switch (lcfg->lcfg_command) {
973 if (lcfg->lcfg_bufcount != 2)
974 GOTO(out_lcfg, rc = -EINVAL);
975 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
979 if (lcfg->lcfg_bufcount != 3)
980 GOTO(out_lcfg, rc = -EINVAL);
981 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
982 poolname, lustre_cfg_string(lcfg, 2));
985 if (lcfg->lcfg_bufcount != 3)
986 GOTO(out_lcfg, rc = -EINVAL);
987 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
988 poolname, lustre_cfg_string(lcfg, 2));
991 if (lcfg->lcfg_bufcount != 2)
992 GOTO(out_lcfg, rc = -EINVAL);
993 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
1001 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1002 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
1007 OBD_FREE(lcfg, data->ioc_plen1);
1009 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1013 /* from mdt_iocontrol */
1014 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1015 void *karg, void __user *uarg)
1017 struct obd_device *obd = exp->exp_obd;
1018 struct mgs_device *mgs = exp2mgs_dev(exp);
1019 struct obd_ioctl_data *data;
1024 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
1025 obd->obd_name, cmd, len, karg, uarg);
1026 if (unlikely(karg == NULL))
1027 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc));
1030 rc = lu_env_init(&env, LCT_MG_THREAD);
1036 case OBD_IOC_PARAM: {
1037 struct lustre_cfg *lcfg;
1039 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1040 CERROR("%s: unknown cfg record type '%x': rc = %d\n",
1041 obd->obd_name, data->ioc_type, rc);
1045 OBD_ALLOC(lcfg, data->ioc_plen1);
1047 GOTO(out, rc = -ENOMEM);
1048 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1049 GOTO(out_free, rc = -EFAULT);
1051 if (lcfg->lcfg_bufcount < 1)
1054 rc = mgs_set_param(&env, mgs, lcfg);
1056 CERROR("%s: setparam err: rc = %d\n",
1059 OBD_FREE(lcfg, data->ioc_plen1);
1062 case OBD_IOC_REPLACE_NIDS:
1063 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1064 CERROR("%s: no device or fsname specified: rc = %d\n",
1069 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1071 CERROR("%s: device or fsname is too long: rc = %d\n",
1076 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1077 CERROR("%s: unterminated device or fsname: rc = %d\n",
1082 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1083 CERROR("%s: no NIDs specified: rc = %d\n",
1088 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1089 CERROR("%s: NID list is not NUL terminated: rc = %d\n",
1094 /* replace nids in llog */
1095 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1098 CERROR("%s: error replacing NIDs for '%s': rc = %d\n",
1099 obd->obd_name, data->ioc_inlbuf1, rc);
1102 case OBD_IOC_CLEAR_CONFIGS:
1103 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1104 CERROR("%s: no device or fsname specified: rc = %d\n",
1109 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1111 CERROR("%s: device or fsname is too long: rc = %d\n",
1116 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1117 CERROR("%s: unterminated device or fsname: rc = %d\n",
1122 /* remove records marked SKIP from config logs */
1123 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1125 CERROR("%s: error clearing config log: rc = %d\n",
1130 rc = mgs_iocontrol_pool(&env, mgs, data);
1132 #ifdef OBD_IOC_BARRIER
1133 case_OBD_IOC_DEPRECATED_FT(OBD_IOC_BARRIER, obd->obd_name, 2, 17);
1135 case OBD_IOC_BARRIER_V2:
1136 rc = mgs_iocontrol_barrier(&env, mgs, data);
1138 case OBD_IOC_NODEMAP:
1139 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1141 case OBD_IOC_LCFG_FORK:
1142 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1145 case OBD_IOC_LCFG_ERASE:
1146 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1148 case OBD_IOC_CATLOGLIST:
1149 rc = mgs_list_logs(&env, mgs, data);
1151 case OBD_IOC_LLOG_CANCEL:
1152 case OBD_IOC_LLOG_REMOVE:
1153 case OBD_IOC_LLOG_CHECK:
1154 case OBD_IOC_LLOG_INFO:
1155 case OBD_IOC_LLOG_PRINT: {
1156 struct llog_ctxt *ctxt;
1158 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1159 rc = llog_ioctl(&env, ctxt, cmd, data);
1160 llog_ctxt_put(ctxt);
1164 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
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;
1180 OBD_ALLOC_PTR(data);
1184 obd = class_name2obd(nextdev);
1186 CERROR("can't locate next device: %s\n", nextdev);
1187 GOTO(out, rc = -ENOTCONN);
1190 data->ocd_version = LUSTRE_VERSION_CODE;
1192 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1193 &obd->obd_uuid, data, NULL);
1195 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1199 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1200 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1201 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1207 static struct tgt_handler mgs_mgs_handlers[] = {
1208 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1209 0, MGS_CONNECT, mgs_connect,
1210 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1211 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1212 0, MGS_DISCONNECT, mgs_disconnect,
1213 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1214 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1215 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 18, 53, 0)
1216 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_SET_INFO, mgs_set_info),
1218 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1219 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1220 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1223 static struct tgt_handler mgs_obd_handlers[] = {
1224 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1227 static struct tgt_handler mgs_dlm_handlers[] = {
1228 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1229 .th_name = "LDLM_ENQUEUE",
1230 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1231 * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1232 * is safe. If such tests will be needed we have to distinguish
1233 * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1234 * instead of common OBD_FAIL_LDLM_ENQUEUE_NET
1237 .th_opc = LDLM_ENQUEUE,
1238 .th_flags = HAS_KEY,
1239 .th_act = tgt_enqueue,
1240 .th_fmt = &RQF_LDLM_ENQUEUE,
1241 .th_version = LUSTRE_DLM_VERSION,
1245 static struct tgt_handler mgs_llog_handlers[] = {
1246 TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1247 TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1248 TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1249 TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1252 static struct tgt_opc_slice mgs_common_slice[] = {
1254 .tos_opc_start = MGS_FIRST_OPC,
1255 .tos_opc_end = MGS_LAST_OPC,
1256 .tos_hs = mgs_mgs_handlers
1259 .tos_opc_start = OBD_FIRST_OPC,
1260 .tos_opc_end = OBD_LAST_OPC,
1261 .tos_hs = mgs_obd_handlers
1264 .tos_opc_start = LDLM_FIRST_OPC,
1265 .tos_opc_end = LDLM_LAST_OPC,
1266 .tos_hs = mgs_dlm_handlers
1269 .tos_opc_start = LLOG_FIRST_OPC,
1270 .tos_opc_end = LLOG_LAST_OPC,
1271 .tos_hs = mgs_llog_handlers
1274 .tos_opc_start = SEC_FIRST_OPC,
1275 .tos_opc_end = SEC_LAST_OPC,
1276 .tos_hs = tgt_sec_ctx_handlers
1283 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1284 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1286 struct ptlrpc_service_conf conf;
1287 struct obd_device *obd;
1288 struct lustre_mount_info *lmi;
1289 struct llog_ctxt *ctxt;
1294 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1298 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1300 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1304 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1307 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1311 /* namespace for mgs llog */
1312 obd->obd_namespace = ldlm_namespace_new(obd, "MGS",
1313 LDLM_NAMESPACE_SERVER,
1314 LDLM_NAMESPACE_MODEST,
1316 if (IS_ERR(obd->obd_namespace)) {
1317 rc = PTR_ERR(obd->obd_namespace);
1318 CERROR("%s: unable to create server namespace: rc = %d\n",
1320 obd->obd_namespace = NULL;
1324 /* No recovery for MGCs */
1325 obd->obd_replayable = 0;
1327 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1328 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1329 OBD_FAIL_MGS_ALL_REPLY_NET);
1333 rc = mgs_fs_setup(env, mgs);
1335 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1340 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1341 obd, &llog_osd_ops);
1345 /* XXX: we need this trick till N:1 stack is supported
1346 * set "current" directory for named llogs
1348 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1350 ctxt->loc_dir = mgs->mgs_configs_dir;
1351 llog_ctxt_put(ctxt);
1353 /* Internal mgs setup */
1354 mgs_init_fsdb_list(mgs);
1355 mutex_init(&mgs->mgs_mutex);
1356 mgs->mgs_start_time = ktime_get_real_seconds();
1357 spin_lock_init(&mgs->mgs_lock);
1358 mutex_init(&mgs->mgs_health_mutex);
1359 init_rwsem(&mgs->mgs_barrier_rwsem);
1361 rc = mgs_lcfg_rename(env, mgs);
1365 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1367 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1372 /* Setup params fsdb and log, so that other servers can make a local
1373 * copy successfully when they are mounted. See LU-4783
1375 rc = mgs_params_fsdb_setup(env, mgs);
1377 /* params fsdb and log can be setup later */
1378 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1379 obd->obd_name, PARAMS_FILENAME, rc);
1381 /* Setup _mgs fsdb, useful for srpc */
1382 mgs__mgs_fsdb_setup(env, mgs);
1384 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1385 "mgs_ldlm_client", &obd->obd_ldlm_client);
1387 conf = (typeof(conf)) {
1388 .psc_name = LUSTRE_MGS_NAME,
1389 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1391 .bc_nbufs = MGS_NBUFS,
1392 .bc_buf_size = MGS_BUFSIZE,
1393 .bc_req_max_size = MGS_MAXREQSIZE,
1394 .bc_rep_max_size = MGS_MAXREPSIZE,
1395 .bc_req_portal = MGS_REQUEST_PORTAL,
1396 .bc_rep_portal = MGC_REPLY_PORTAL,
1399 .tc_thr_name = "ll_mgs",
1400 .tc_nthrs_init = MGS_NTHRS_INIT,
1401 .tc_nthrs_max = MGS_NTHRS_MAX,
1402 .tc_ctx_tags = LCT_MG_THREAD,
1405 .so_req_handler = tgt_request_handle,
1406 .so_req_printer = target_print_req,
1410 /* Start the service threads */
1411 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1412 obd->obd_debugfs_entry);
1413 if (IS_ERR(mgs->mgs_service)) {
1414 rc = PTR_ERR(mgs->mgs_service);
1415 CERROR("failed to start mgs service: %d\n", rc);
1416 mgs->mgs_service = NULL;
1417 GOTO(err_lproc, rc);
1420 ping_evictor_start();
1422 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1424 /* device stack is not yet fully setup to keep no objects behind */
1425 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1428 mgs_params_fsdb_cleanup(env, mgs);
1429 lproc_mgs_cleanup(mgs);
1431 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1433 ctxt->loc_dir = NULL;
1434 llog_cleanup(env, ctxt);
1437 tgt_fini(env, &mgs->mgs_lut);
1439 /* No extra cleanup needed for llog_init_commit_thread() */
1440 mgs_fs_cleanup(env, mgs);
1442 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1443 obd->obd_namespace = NULL;
1445 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1446 lu_site_print(env, mgs2lu_dev(mgs)->ld_site,
1447 &mgs2lu_dev(mgs)->ld_site->ls_obj_hash.nelems,
1448 D_OTHER, lu_cdebug_printer);
1449 obd_disconnect(mgs->mgs_bottom_exp);
1452 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1456 static struct lu_device *mgs_device_free(const struct lu_env *env,
1457 struct lu_device *lu)
1459 struct mgs_device *mgs = lu2mgs_dev(lu);
1463 dt_device_fini(&mgs->mgs_dt_dev);
1468 static int mgs_process_config(const struct lu_env *env,
1469 struct lu_device *dev,
1470 struct lustre_cfg *lcfg)
1476 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1477 const struct lu_object_conf *unused)
1479 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1480 struct lu_device *under;
1481 struct lu_object *below;
1486 /* do no set .do_ops as mgs calls to bottom osd directly */
1487 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1488 PFID(lu_object_fid(o)));
1490 under = &d->mgs_bottom->dd_lu_dev;
1491 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1493 lu_object_add(o, below);
1500 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1502 struct mgs_object *obj = lu2mgs_obj(o);
1503 struct lu_object_header *h = o->lo_header;
1505 dt_object_fini(&obj->mgo_obj);
1506 lu_object_header_fini(h);
1507 OBD_FREE_PRE(obj, sizeof(*obj), "kfreed");
1508 kfree_rcu(obj, mgo_header.loh_rcu);
1511 static int mgs_object_print(const struct lu_env *env, void *cookie,
1512 lu_printer_t p, const struct lu_object *l)
1514 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1516 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1519 static const struct lu_object_operations mgs_lu_obj_ops = {
1520 .loo_object_init = mgs_object_init,
1521 .loo_object_free = mgs_object_free,
1522 .loo_object_print = mgs_object_print,
1525 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1526 const struct lu_object_header *hdr,
1527 struct lu_device *d)
1529 struct lu_object_header *h;
1530 struct mgs_object *o;
1531 struct lu_object *l;
1533 LASSERT(hdr == NULL);
1537 l = &o->mgo_obj.do_lu;
1540 lu_object_header_init(h);
1541 dt_object_init(&o->mgo_obj, h, d);
1542 lu_object_add_top(h, l);
1544 l->lo_ops = &mgs_lu_obj_ops;
1552 const struct lu_device_operations mgs_lu_ops = {
1553 .ldo_object_alloc = mgs_object_alloc,
1554 .ldo_process_config = mgs_process_config,
1557 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1558 struct lu_device_type *type,
1559 struct lustre_cfg *lcfg)
1561 struct mgs_device *mgs;
1562 struct lu_device *ludev;
1566 ludev = ERR_PTR(-ENOMEM);
1570 ludev = mgs2lu_dev(mgs);
1571 dt_device_init(&mgs->mgs_dt_dev, type);
1572 rc = mgs_init0(env, mgs, type, lcfg);
1574 mgs_device_free(env, ludev);
1575 ludev = ERR_PTR(rc);
1581 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1582 struct lu_device *d)
1584 struct mgs_device *mgs = lu2mgs_dev(d);
1585 struct obd_device *obd = mgs->mgs_obd;
1586 struct llog_ctxt *ctxt;
1590 LASSERT(mgs->mgs_bottom);
1592 class_disconnect_exports(obd);
1594 ping_evictor_stop();
1596 mutex_lock(&mgs->mgs_health_mutex);
1597 ptlrpc_unregister_service(mgs->mgs_service);
1598 mutex_unlock(&mgs->mgs_health_mutex);
1600 mgs_params_fsdb_cleanup(env, mgs);
1601 mgs_cleanup_fsdb_list(mgs);
1603 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1604 obd_exports_barrier(obd);
1605 obd_zombie_barrier();
1607 tgt_fini(env, &mgs->mgs_lut);
1608 lproc_mgs_cleanup(mgs);
1610 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1612 ctxt->loc_dir = NULL;
1613 llog_cleanup(env, ctxt);
1616 mgs_fs_cleanup(env, mgs);
1618 ldlm_namespace_free_post(obd->obd_namespace);
1619 obd->obd_namespace = NULL;
1621 lu_site_purge(env, d->ld_site, ~0);
1622 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
1623 D_OTHER, lu_cdebug_printer);
1624 LASSERT(mgs->mgs_bottom_exp);
1625 obd_disconnect(mgs->mgs_bottom_exp);
1627 server_put_mount(obd->obd_name, true);
1632 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1633 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1635 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1637 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1639 static const struct lu_device_type_operations mgs_device_type_ops = {
1640 .ldto_init = mgs_type_init,
1641 .ldto_fini = mgs_type_fini,
1643 .ldto_start = mgs_type_start,
1644 .ldto_stop = mgs_type_stop,
1646 .ldto_device_alloc = mgs_device_alloc,
1647 .ldto_device_free = mgs_device_free,
1649 .ldto_device_fini = mgs_device_fini
1652 static struct lu_device_type mgs_device_type = {
1653 .ldt_tags = LU_DEVICE_DT,
1654 .ldt_name = LUSTRE_MGS_NAME,
1655 .ldt_ops = &mgs_device_type_ops,
1656 .ldt_ctx_tags = LCT_MG_THREAD
1659 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1660 struct obd_device *obd, struct obd_uuid *cluuid,
1661 struct obd_connect_data *data, void *localdata)
1665 if (exp == NULL || obd == NULL || cluuid == NULL)
1668 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1671 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1673 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1674 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1676 exp->exp_connect_data = *data;
1677 data->ocd_version = LUSTRE_VERSION_CODE;
1680 RETURN(mgs_export_stats_init(obd, exp, localdata));
1683 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1684 struct obd_device *obd, struct obd_uuid *cluuid,
1685 struct obd_connect_data *data, void *localdata)
1687 struct obd_export *lexp;
1688 struct lustre_handle conn = {
1695 if (exp == NULL || obd == NULL || cluuid == NULL)
1698 rc = class_connect(&conn, obd, cluuid);
1702 lexp = class_conn2export(&conn);
1706 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1708 GOTO(out_disconnect, rc);
1715 class_disconnect(lexp);
1720 static int mgs_obd_disconnect(struct obd_export *exp)
1728 mgs_fsc_cleanup(exp);
1730 class_export_get(exp);
1731 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1733 rc = server_disconnect_export(exp);
1734 class_export_put(exp);
1738 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1740 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1743 mutex_lock(&mgs->mgs_health_mutex);
1744 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1745 mutex_unlock(&mgs->mgs_health_mutex);
1747 return rc != 0 ? 1 : 0;
1750 /* use obd ops to offer management infrastructure */
1751 static const struct obd_ops mgs_obd_device_ops = {
1752 .o_owner = THIS_MODULE,
1753 .o_connect = mgs_obd_connect,
1754 .o_reconnect = mgs_obd_reconnect,
1755 .o_disconnect = mgs_obd_disconnect,
1756 .o_init_export = mgs_init_export,
1757 .o_destroy_export = mgs_destroy_export,
1758 .o_iocontrol = mgs_iocontrol,
1759 .o_health_check = mgs_health_check,
1762 static int __init mgs_init(void)
1766 rc = libcfs_setup();
1770 return class_register_type(&mgs_obd_device_ops, NULL, true,
1771 LUSTRE_MGS_NAME, &mgs_device_type);
1774 static void __exit mgs_exit(void)
1776 class_unregister_type(LUSTRE_MGS_NAME);
1779 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1780 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1781 MODULE_VERSION(LUSTRE_VERSION_STRING);
1782 MODULE_LICENSE("GPL");
1784 module_init(mgs_init);
1785 module_exit(mgs_exit);