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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mgs/mgs_handler.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
39 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40 * Author: Mikhail Pershin <tappro@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_MGS
44 #define D_MGS D_CONFIG
46 #include <obd_class.h>
47 #include <lprocfs_status.h>
48 #include <lustre_ioctl.h>
49 #include <lustre_param.h>
51 #include "mgs_internal.h"
54 * Regular MGS handlers
56 static int mgs_connect(struct tgt_session_info *tsi)
58 struct ptlrpc_request *req = tgt_ses_req(tsi);
63 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_CONNECT_NET, cfs_fail_val);
64 rc = tgt_connect(tsi);
68 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
69 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
74 static int mgs_disconnect(struct tgt_session_info *tsi)
80 LASSERT(tsi->tsi_exp);
82 rc = tgt_disconnect(tsi);
84 RETURN(err_serious(rc));
88 static int mgs_exception(struct tgt_session_info *tsi)
92 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
97 static int mgs_set_info(struct tgt_session_info *tsi)
99 struct mgs_thread_info *mgi;
100 struct mgs_send_param *msp, *rep_msp;
101 struct lustre_cfg *lcfg;
106 mgi = mgs_env_info(tsi->tsi_env);
108 RETURN(err_serious(PTR_ERR(mgi)));
110 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
112 RETURN(err_serious(-EFAULT));
114 /* Construct lustre_cfg structure to pass to function mgs_setparam */
115 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
116 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
117 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
119 rc = mgs_setparam(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg,
122 LCONSOLE_WARN("%s: Unable to set parameter %s for %s: %d\n",
123 tgt_name(tsi->tsi_tgt), msp->mgs_param,
124 mgi->mgi_fsname, rc);
128 /* send back the whole msp in the reply */
129 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
133 lustre_cfg_free(lcfg);
143 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
144 void *cbdata, enum ast_type type)
148 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
151 /* l_ast_data is used as a marker to avoid cancel ldlm lock
152 * twice. See LU-2317. */
153 lock_res_and_lock(lock);
154 fsdb = (struct fs_db *)lock->l_ast_data;
155 lock->l_ast_data = NULL;
156 unlock_res_and_lock(lock);
159 struct lustre_handle lockh;
163 /* clear the bit before lock put */
164 clear_bit(FSDB_REVOKING_LOCK,
168 clear_bit(FSDB_REVOKING_PARAMS,
172 mgs_ir_notify_complete(fsdb);
178 ldlm_lock2handle(lock, &lockh);
179 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
183 RETURN(ldlm_completion_ast(lock, flags, cbdata));
186 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
189 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
192 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
195 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
198 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
201 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
204 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
206 ldlm_completion_callback cp = NULL;
207 struct lustre_handle lockh = { 0 };
208 struct ldlm_res_id res_id;
209 __u64 flags = LDLM_FL_ATOMIC_CB;
213 LASSERT(fsdb->fsdb_name[0] != '\0');
214 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
217 case CONFIG_T_CONFIG:
218 cp = mgs_completion_ast_config;
219 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
222 case CONFIG_T_PARAMS:
223 cp = mgs_completion_ast_params;
224 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
227 case CONFIG_T_RECOVER:
228 cp = mgs_completion_ast_ir;
235 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
236 &res_id, LDLM_PLAIN, NULL, LCK_EX,
237 &flags, ldlm_blocking_ast, cp,
238 NULL, fsdb, 0, LVB_T_NONE, NULL,
240 if (rc != ELDLM_OK) {
241 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
242 le64_to_cpu(res_id.name[0]),
243 le64_to_cpu(res_id.name[1]), rc);
245 if (type == CONFIG_T_CONFIG)
246 clear_bit(FSDB_REVOKING_LOCK,
249 if (type == CONFIG_T_PARAMS)
250 clear_bit(FSDB_REVOKING_PARAMS,
253 /* lock has been cancelled in completion_ast. */
262 static int mgs_check_target(const struct lu_env *env,
263 struct mgs_device *mgs,
264 struct mgs_target_info *mti)
269 rc = mgs_check_index(env, mgs, mti);
271 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
272 "this MGS does not know about it, preventing "
273 "registration.\n", mti->mti_svname);
275 } else if (rc == -1) {
276 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
277 "disappeared! Regenerating all logs.\n",
279 mti->mti_flags |= LDD_F_WRITECONF;
282 /* Index is correctly marked as used */
284 /* If the logs don't contain the mti_nids then add
285 them as failover nids */
286 rc = mgs_check_failnid(env, mgs, mti);
292 /* Ensure this is not a failover node that is connecting first*/
293 static int mgs_check_failover_reg(struct mgs_target_info *mti)
299 ptr = mti->mti_params;
300 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
301 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
302 for (i = 0; i < mti->mti_nid_count; i++) {
303 if (nid == mti->mti_nids[i]) {
304 LCONSOLE_WARN("Denying initial registra"
305 "tion attempt from nid %s"
306 ", specified as failover"
307 "\n",libcfs_nid2str(nid));
308 return -EADDRNOTAVAIL;
316 /* Called whenever a target starts up. Flags indicate first connect, etc. */
317 static int mgs_target_reg(struct tgt_session_info *tsi)
319 struct obd_device *obd = tsi->tsi_exp->exp_obd;
320 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
321 struct mgs_target_info *mti, *rep_mti;
328 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
330 return err_serious(rc);
332 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
334 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
336 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
337 RETURN(err_serious(-EFAULT));
340 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
342 opc = mti->mti_flags & LDD_F_OPC_MASK;
346 if (opc == LDD_F_OPC_READY) {
347 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
348 mti->mti_fsname, mti->mti_stripe_index);
349 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
351 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
352 CERROR("Update IR return with %d(ignore and IR "
355 GOTO(out_nolock, rc);
358 /* Do not support unregistering right now. */
359 if (opc != LDD_F_OPC_REG)
360 GOTO(out_nolock, rc = -EINVAL);
362 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
363 mti->mti_fsname, mti->mti_stripe_index);
365 if (mti->mti_flags & LDD_F_NEED_INDEX)
366 mti->mti_flags |= LDD_F_WRITECONF;
368 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
370 /* We're just here as a startup ping. */
371 CDEBUG(D_MGS, "Server %s is running on %s\n",
372 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
373 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
374 /* above will set appropriate mti flags */
376 /* Nothing wrong, or fatal error */
377 GOTO(out_nolock, rc);
378 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
379 rc = mgs_check_failover_reg(mti);
381 GOTO(out_nolock, rc);
384 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
386 if (mti->mti_flags & LDD_F_WRITECONF) {
387 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
388 mti->mti_stripe_index == 0) {
389 rc = mgs_erase_logs(tsi->tsi_env, mgs,
391 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
392 "request. All servers must be restarted "
393 "in order to regenerate the logs."
394 "\n", obd->obd_name, mti->mti_fsname);
395 } else if (mti->mti_flags &
396 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
397 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
398 LCONSOLE_WARN("%s: Regenerating %s log by user "
400 obd->obd_name, mti->mti_svname);
402 mti->mti_flags |= LDD_F_UPDATE;
403 /* Erased logs means start from scratch. */
404 mti->mti_flags &= ~LDD_F_UPGRADE14;
406 GOTO(out_nolock, rc);
409 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &fsdb);
411 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
412 GOTO(out_nolock, rc);
416 * Log writing contention is handled by the fsdb_mutex.
418 * It should be alright if someone was reading while we were
419 * updating the logs - if we revoke at the end they will just update
420 * from where they left off.
423 if (mti->mti_flags & LDD_F_UPGRADE14) {
424 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
428 if (mti->mti_flags & LDD_F_UPDATE) {
429 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
430 mti->mti_stripe_index);
432 /* create or update the target log
433 and update the client/mdt logs */
434 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, fsdb);
436 CERROR("Failed to write %s log (%d)\n",
437 mti->mti_svname, rc);
441 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
442 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
444 mti->mti_flags |= LDD_F_REWRITE_LDD;
448 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
451 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
452 mti->mti_stripe_index, rc);
453 /* An error flag is set in the mti reply rather than an error code */
455 mti->mti_flags |= LDD_F_ERROR;
457 /* send back the whole mti in the reply */
458 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
461 /* Flush logs to disk */
462 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
466 /* Called whenever a target cleans up. */
467 static int mgs_target_del(struct tgt_session_info *tsi)
471 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
476 static int mgs_config_read(struct tgt_session_info *tsi)
478 struct ptlrpc_request *req = tgt_ses_req(tsi);
479 struct mgs_config_body *body;
484 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
486 DEBUG_REQ(D_HA, req, "no mgs_config_body");
487 RETURN(err_serious(-EFAULT));
490 switch (body->mcb_type) {
491 case CONFIG_T_RECOVER:
492 rc = mgs_get_ir_logs(req);
494 case CONFIG_T_CONFIG:
505 static int mgs_llog_open(struct tgt_session_info *tsi)
507 struct mgs_thread_info *mgi;
508 struct ptlrpc_request *req = tgt_ses_req(tsi);
514 rc = tgt_llog_open(tsi);
519 * For old clients there is no direct way of knowing which file system
520 * a client is operating at the MGS side. But we need to pick up those
521 * clients so that the MGS can mark the corresponding file system as
522 * non-IR capable because old clients are not ready to be notified.
524 * Therefore we attempt to detect the file systems name by hacking the
525 * llog operation which is currently used by the clients to fetch
526 * configuration logs. At present this is fine because this is the
527 * ONLY llog operation between mgc and the MGS.
529 * If extra llog operation are going to be added, this function needs
532 * When releases prior than 2.0 are not supported, the following code
535 mgi = mgs_env_info(tsi->tsi_env);
537 RETURN(PTR_ERR(mgi));
539 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
541 char *ptr = strchr(logname, '-');
542 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
544 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
545 if (strcmp(logname, PARAMS_FILENAME) != 0)
546 LCONSOLE_WARN("%s: non-config logname "
548 tgt_name(tsi->tsi_tgt),
550 /* not error, this can be llog test name */
552 strncpy(mgi->mgi_fsname, logname, len);
553 mgi->mgi_fsname[len] = 0;
555 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
557 if (rc && rc != -EEXIST) {
558 LCONSOLE_WARN("%s: Unable to add client %s "
559 "to file system %s: %d\n",
560 tgt_name(tsi->tsi_tgt),
561 libcfs_nid2str(req->rq_peer.nid),
562 mgi->mgi_fsname, rc);
568 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
574 static inline int mgs_init_export(struct obd_export *exp)
576 struct mgs_export_data *data = &exp->u.eu_mgs_data;
578 /* init mgs_export_data for fsc */
579 spin_lock_init(&data->med_lock);
580 INIT_LIST_HEAD(&data->med_clients);
582 spin_lock(&exp->exp_lock);
583 exp->exp_connecting = 1;
584 spin_unlock(&exp->exp_lock);
586 /* self-export doesn't need client data and ldlm initialization */
587 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
588 &exp->exp_client_uuid)))
590 return ldlm_init_export(exp);
593 static inline int mgs_destroy_export(struct obd_export *exp)
597 target_destroy_export(exp);
598 mgs_client_free(exp);
600 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
601 &exp->exp_client_uuid)))
604 ldlm_destroy_export(exp);
609 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
614 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
622 strcpy(poolname, ptr);
627 static int mgs_iocontrol_nodemap(const struct lu_env *env,
628 struct mgs_device *mgs,
629 struct obd_ioctl_data *data)
631 struct lustre_cfg *lcfg = NULL;
632 struct lu_nodemap *nodemap;
634 const char *nodemap_name = NULL;
635 const char *nidstr = NULL;
636 const char *client_idstr = NULL;
637 const char *idtype_str = NULL;
648 if (data->ioc_type != LUSTRE_CFG_TYPE) {
649 CERROR("%s: unknown cfg record type: %d\n",
650 mgs->mgs_obd->obd_name, data->ioc_type);
651 GOTO(out, rc = -EINVAL);
654 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
655 GOTO(out, rc = -E2BIG);
657 OBD_ALLOC(lcfg, data->ioc_plen1);
659 GOTO(out, rc = -ENOMEM);
661 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
662 GOTO(out_lcfg, rc = -EFAULT);
664 cmd = lcfg->lcfg_command;
667 case LCFG_NODEMAP_ACTIVATE:
668 if (lcfg->lcfg_bufcount != 2)
669 GOTO(out_lcfg, rc = -EINVAL);
670 param = lustre_cfg_string(lcfg, 1);
671 if (strcmp(param, "1") == 0)
676 case LCFG_NODEMAP_ADD:
677 case LCFG_NODEMAP_DEL:
678 if (lcfg->lcfg_bufcount != 2)
679 GOTO(out_lcfg, rc = -EINVAL);
680 nodemap_name = lustre_cfg_string(lcfg, 1);
681 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
683 case LCFG_NODEMAP_TEST_NID:
684 if (lcfg->lcfg_bufcount != 2)
685 GOTO(out_lcfg, rc = -EINVAL);
686 nidstr = lustre_cfg_string(lcfg, 1);
687 nid = libcfs_str2nid(nidstr);
688 nodemap = nodemap_classify_nid(nid);
689 memset(data->ioc_pbuf1, 0, data->ioc_plen1);
690 if (copy_to_user(data->ioc_pbuf1, nodemap->nm_name,
691 strlen(nodemap->nm_name)) != 0)
692 GOTO(out_lcfg, rc = -EFAULT);
694 case LCFG_NODEMAP_TEST_ID:
695 if (lcfg->lcfg_bufcount != 4)
696 GOTO(out_lcfg, rc = -EINVAL);
697 nidstr = lustre_cfg_string(lcfg, 1);
698 idtype_str = lustre_cfg_string(lcfg, 2);
699 client_idstr = lustre_cfg_string(lcfg, 3);
701 nid = libcfs_str2nid(nidstr);
702 nodemap = nodemap_classify_nid(nid);
703 client_id = simple_strtoul(client_idstr, NULL, 10);
705 if (strcmp(idtype_str, "uid") == 0)
706 idtype = NODEMAP_UID;
708 idtype = NODEMAP_GID;
710 fs_id = nodemap_map_id(nodemap, idtype, NODEMAP_CLIENT_TO_FS,
713 if (data->ioc_plen1 < sizeof(fs_idstr))
714 GOTO(out_lcfg, rc = -EINVAL);
716 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
717 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
718 sizeof(fs_idstr)) != 0)
719 GOTO(out_lcfg, rc = -EINVAL);
721 case LCFG_NODEMAP_ADD_RANGE:
722 case LCFG_NODEMAP_DEL_RANGE:
723 case LCFG_NODEMAP_ADD_UIDMAP:
724 case LCFG_NODEMAP_DEL_UIDMAP:
725 case LCFG_NODEMAP_ADD_GIDMAP:
726 case LCFG_NODEMAP_DEL_GIDMAP:
727 if (lcfg->lcfg_bufcount != 3)
728 GOTO(out_lcfg, rc = -EINVAL);
729 nodemap_name = lustre_cfg_string(lcfg, 1);
730 param = lustre_cfg_string(lcfg, 2);
731 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
733 case LCFG_NODEMAP_ADMIN:
734 case LCFG_NODEMAP_TRUSTED:
735 case LCFG_NODEMAP_SQUASH_UID:
736 case LCFG_NODEMAP_SQUASH_GID:
737 if (lcfg->lcfg_bufcount != 4)
738 GOTO(out_lcfg, rc = -EINVAL);
739 nodemap_name = lustre_cfg_string(lcfg, 1);
740 param = lustre_cfg_string(lcfg, 3);
741 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
748 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
749 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
755 OBD_FREE(lcfg, data->ioc_plen1);
760 static int mgs_iocontrol_pool(const struct lu_env *env,
761 struct mgs_device *mgs,
762 struct obd_ioctl_data *data)
764 struct mgs_thread_info *mgi = mgs_env_info(env);
766 struct lustre_cfg *lcfg = NULL;
767 char *poolname = NULL;
770 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
771 if (poolname == NULL)
774 if (data->ioc_type != LUSTRE_CFG_TYPE) {
775 CERROR("%s: unknown cfg record type: %d\n",
776 mgs->mgs_obd->obd_name, data->ioc_type);
777 GOTO(out_pool, rc = -EINVAL);
780 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
781 GOTO(out_pool, rc = -E2BIG);
783 OBD_ALLOC(lcfg, data->ioc_plen1);
785 GOTO(out_pool, rc = -ENOMEM);
787 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
788 GOTO(out_lcfg, rc = -EFAULT);
790 if (lcfg->lcfg_bufcount < 2)
791 GOTO(out_lcfg, rc = -EFAULT);
793 /* first arg is always <fsname>.<poolname> */
794 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
799 switch (lcfg->lcfg_command) {
801 if (lcfg->lcfg_bufcount != 2)
802 GOTO(out_lcfg, rc = -EINVAL);
803 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
807 if (lcfg->lcfg_bufcount != 3)
808 GOTO(out_lcfg, rc = -EINVAL);
809 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
810 poolname, lustre_cfg_string(lcfg, 2));
813 if (lcfg->lcfg_bufcount != 3)
814 GOTO(out_lcfg, rc = -EINVAL);
815 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
816 poolname, lustre_cfg_string(lcfg, 2));
819 if (lcfg->lcfg_bufcount != 2)
820 GOTO(out_lcfg, rc = -EINVAL);
821 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
829 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
830 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
835 OBD_FREE(lcfg, data->ioc_plen1);
837 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
841 /* from mdt_iocontrol */
842 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
843 void *karg, void *uarg)
845 struct mgs_device *mgs = exp2mgs_dev(exp);
846 struct obd_ioctl_data *data = karg;
851 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
853 rc = lu_env_init(&env, LCT_MG_THREAD);
859 case OBD_IOC_PARAM: {
860 struct mgs_thread_info *mgi = mgs_env_info(&env);
861 struct lustre_cfg *lcfg;
863 if (data->ioc_type != LUSTRE_CFG_TYPE) {
864 CERROR("%s: unknown cfg record type: %d\n",
865 mgs->mgs_obd->obd_name, data->ioc_type);
866 GOTO(out, rc = -EINVAL);
869 OBD_ALLOC(lcfg, data->ioc_plen1);
871 GOTO(out, rc = -ENOMEM);
872 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
873 GOTO(out_free, rc = -EFAULT);
875 if (lcfg->lcfg_bufcount < 1)
876 GOTO(out_free, rc = -EINVAL);
878 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
880 CERROR("%s: setparam err: rc = %d\n",
881 exp->exp_obd->obd_name, rc);
883 OBD_FREE(lcfg, data->ioc_plen1);
887 case OBD_IOC_REPLACE_NIDS: {
888 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
889 CERROR("No device name specified!\n");
894 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
895 CERROR("Device name is not NUL terminated!\n");
900 if (data->ioc_plen1 > MTI_NAME_MAXLEN) {
901 CERROR("Device name is too long\n");
906 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
907 CERROR("No NIDs were specified!\n");
912 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
913 CERROR("NID list is not NUL terminated!\n");
918 /* replace nids in llog */
919 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
922 CERROR("%s: error replacing nids: rc = %d\n",
923 exp->exp_obd->obd_name, rc);
929 rc = mgs_iocontrol_pool(&env, mgs, data);
932 case OBD_IOC_NODEMAP:
933 rc = mgs_iocontrol_nodemap(&env, mgs, data);
936 case OBD_IOC_DUMP_LOG: {
937 struct llog_ctxt *ctxt;
939 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
940 rc = class_config_dump_llog(&env, ctxt, data->ioc_inlbuf1,
947 case OBD_IOC_CATLOGLIST:
948 rc = mgs_list_logs(&env, mgs, data);
950 case OBD_IOC_LLOG_CANCEL:
951 case OBD_IOC_LLOG_REMOVE:
952 case OBD_IOC_LLOG_CHECK:
953 case OBD_IOC_LLOG_INFO:
954 case OBD_IOC_LLOG_PRINT: {
955 struct llog_ctxt *ctxt;
957 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
958 rc = llog_ioctl(&env, ctxt, cmd, data);
964 CERROR("%s: unknown command %#x\n",
965 mgs->mgs_obd->obd_name, cmd);
974 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
976 struct obd_connect_data *data = NULL;
977 struct obd_device *obd;
985 obd = class_name2obd(nextdev);
987 CERROR("can't locate next device: %s\n", nextdev);
988 GOTO(out, rc = -ENOTCONN);
991 data->ocd_version = LUSTRE_VERSION_CODE;
993 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
994 &obd->obd_uuid, data, NULL);
996 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1000 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1001 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1002 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1008 static struct tgt_handler mgs_mgs_handlers[] = {
1009 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1010 0, MGS_CONNECT, mgs_connect,
1011 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1012 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1013 0, MGS_DISCONNECT, mgs_disconnect,
1014 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1015 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1016 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
1017 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
1018 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1019 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
1022 static struct tgt_handler mgs_obd_handlers[] = {
1023 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1026 static struct tgt_handler mgs_dlm_handlers[] = {
1027 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1028 .th_name = "LDLM_ENQUEUE",
1029 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1030 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1031 * * is safe. If such tests will be needed we have to distinguish
1032 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1033 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1035 .th_opc = LDLM_ENQUEUE,
1036 .th_flags = HABEO_CLAVIS,
1037 .th_act = tgt_enqueue,
1038 .th_fmt = &RQF_LDLM_ENQUEUE,
1039 .th_version = LUSTRE_DLM_VERSION,
1043 static struct tgt_handler mgs_llog_handlers[] = {
1044 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1045 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1046 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1047 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1048 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1051 static struct tgt_opc_slice mgs_common_slice[] = {
1053 .tos_opc_start = MGS_FIRST_OPC,
1054 .tos_opc_end = MGS_LAST_OPC,
1055 .tos_hs = mgs_mgs_handlers
1058 .tos_opc_start = OBD_FIRST_OPC,
1059 .tos_opc_end = OBD_LAST_OPC,
1060 .tos_hs = mgs_obd_handlers
1063 .tos_opc_start = LDLM_FIRST_OPC,
1064 .tos_opc_end = LDLM_LAST_OPC,
1065 .tos_hs = mgs_dlm_handlers
1068 .tos_opc_start = LLOG_FIRST_OPC,
1069 .tos_opc_end = LLOG_LAST_OPC,
1070 .tos_hs = mgs_llog_handlers
1073 .tos_opc_start = SEC_FIRST_OPC,
1074 .tos_opc_end = SEC_LAST_OPC,
1075 .tos_hs = tgt_sec_ctx_handlers
1082 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1083 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1085 struct ptlrpc_service_conf conf;
1086 struct obd_device *obd;
1087 struct lustre_mount_info *lmi;
1088 struct llog_ctxt *ctxt;
1089 struct fs_db *fsdb = NULL;
1094 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1098 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1100 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1104 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1107 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1109 obd->u.obt.obt_magic = OBT_MAGIC;
1110 obd->u.obt.obt_instance = 0;
1112 /* namespace for mgs llog */
1113 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1114 LDLM_NAMESPACE_SERVER,
1115 LDLM_NAMESPACE_MODEST,
1117 if (obd->obd_namespace == NULL)
1118 GOTO(err_ops, rc = -ENOMEM);
1120 /* No recovery for MGCs */
1121 obd->obd_replayable = 0;
1123 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1124 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1125 OBD_FAIL_MGS_ALL_REPLY_NET);
1129 rc = mgs_fs_setup(env, mgs);
1131 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1136 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1137 obd, &llog_osd_ops);
1141 /* XXX: we need this trick till N:1 stack is supported
1142 * set "current" directory for named llogs */
1143 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1145 ctxt->loc_dir = mgs->mgs_configs_dir;
1146 llog_ctxt_put(ctxt);
1148 /* Internal mgs setup */
1149 mgs_init_fsdb_list(mgs);
1150 mutex_init(&mgs->mgs_mutex);
1151 mgs->mgs_start_time = cfs_time_current_sec();
1152 spin_lock_init(&mgs->mgs_lock);
1154 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1156 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1161 /* Setup params fsdb and log, so that other servers can make a local
1162 * copy successfully when they are mounted. See LU-4783 */
1163 rc = mgs_params_fsdb_setup(env, mgs, fsdb);
1165 /* params fsdb and log can be setup later */
1166 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1167 obd->obd_name, PARAMS_FILENAME, rc);
1169 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1170 "mgs_ldlm_client", &obd->obd_ldlm_client);
1172 conf = (typeof(conf)) {
1173 .psc_name = LUSTRE_MGS_NAME,
1174 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1176 .bc_nbufs = MGS_NBUFS,
1177 .bc_buf_size = MGS_BUFSIZE,
1178 .bc_req_max_size = MGS_MAXREQSIZE,
1179 .bc_rep_max_size = MGS_MAXREPSIZE,
1180 .bc_req_portal = MGS_REQUEST_PORTAL,
1181 .bc_rep_portal = MGC_REPLY_PORTAL,
1184 .tc_thr_name = "ll_mgs",
1185 .tc_nthrs_init = MGS_NTHRS_INIT,
1186 .tc_nthrs_max = MGS_NTHRS_MAX,
1187 .tc_ctx_tags = LCT_MG_THREAD,
1190 .so_req_handler = tgt_request_handle,
1191 .so_req_printer = target_print_req,
1195 /* Start the service threads */
1196 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1197 if (IS_ERR(mgs->mgs_service)) {
1198 rc = PTR_ERR(mgs->mgs_service);
1199 CERROR("failed to start mgs service: %d\n", rc);
1200 mgs->mgs_service = NULL;
1201 GOTO(err_lproc, rc);
1204 ping_evictor_start();
1206 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1208 /* device stack is not yet fully setup to keep no objects behind */
1209 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1212 mgs_params_fsdb_cleanup(env, mgs);
1213 lproc_mgs_cleanup(mgs);
1215 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1217 ctxt->loc_dir = NULL;
1218 llog_cleanup(env, ctxt);
1221 tgt_fini(env, &mgs->mgs_lut);
1223 /* No extra cleanup needed for llog_init_commit_thread() */
1224 mgs_fs_cleanup(env, mgs);
1226 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1227 obd->obd_namespace = NULL;
1229 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1230 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1231 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1232 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1235 obd_disconnect(mgs->mgs_bottom_exp);
1238 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1242 static struct lu_device *mgs_device_free(const struct lu_env *env,
1243 struct lu_device *lu)
1245 struct mgs_device *mgs = lu2mgs_dev(lu);
1248 dt_device_fini(&mgs->mgs_dt_dev);
1253 static int mgs_process_config(const struct lu_env *env,
1254 struct lu_device *dev,
1255 struct lustre_cfg *lcfg)
1261 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1262 const struct lu_object_conf *unused)
1264 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1265 struct lu_device *under;
1266 struct lu_object *below;
1270 /* do no set .do_ops as mgs calls to bottom osd directly */
1272 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1273 PFID(lu_object_fid(o)));
1275 under = &d->mgs_bottom->dd_lu_dev;
1276 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1278 lu_object_add(o, below);
1285 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1287 struct mgs_object *obj = lu2mgs_obj(o);
1288 struct lu_object_header *h = o->lo_header;
1290 dt_object_fini(&obj->mgo_obj);
1291 lu_object_header_fini(h);
1295 static int mgs_object_print(const struct lu_env *env, void *cookie,
1296 lu_printer_t p, const struct lu_object *l)
1298 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1300 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1303 struct lu_object_operations mgs_lu_obj_ops = {
1304 .loo_object_init = mgs_object_init,
1305 .loo_object_free = mgs_object_free,
1306 .loo_object_print = mgs_object_print,
1309 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1310 const struct lu_object_header *hdr,
1311 struct lu_device *d)
1313 struct lu_object_header *h;
1314 struct mgs_object *o;
1315 struct lu_object *l;
1317 LASSERT(hdr == NULL);
1321 l = &o->mgo_obj.do_lu;
1324 lu_object_header_init(h);
1325 dt_object_init(&o->mgo_obj, h, d);
1326 lu_object_add_top(h, l);
1328 l->lo_ops = &mgs_lu_obj_ops;
1336 const struct lu_device_operations mgs_lu_ops = {
1337 .ldo_object_alloc = mgs_object_alloc,
1338 .ldo_process_config = mgs_process_config,
1341 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1342 struct lu_device_type *type,
1343 struct lustre_cfg *lcfg)
1345 struct mgs_device *mgs;
1346 struct lu_device *ludev;
1350 ludev = ERR_PTR(-ENOMEM);
1354 ludev = mgs2lu_dev(mgs);
1355 dt_device_init(&mgs->mgs_dt_dev, type);
1356 rc = mgs_init0(env, mgs, type, lcfg);
1358 mgs_device_free(env, ludev);
1359 ludev = ERR_PTR(rc);
1365 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1366 struct lu_device *d)
1368 struct mgs_device *mgs = lu2mgs_dev(d);
1369 struct obd_device *obd = mgs->mgs_obd;
1370 struct llog_ctxt *ctxt;
1374 LASSERT(mgs->mgs_bottom);
1376 class_disconnect_exports(obd);
1378 ping_evictor_stop();
1380 ptlrpc_unregister_service(mgs->mgs_service);
1382 obd_exports_barrier(obd);
1383 obd_zombie_barrier();
1385 tgt_fini(env, &mgs->mgs_lut);
1386 mgs_params_fsdb_cleanup(env, mgs);
1387 mgs_cleanup_fsdb_list(mgs);
1388 lproc_mgs_cleanup(mgs);
1390 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1392 ctxt->loc_dir = NULL;
1393 llog_cleanup(env, ctxt);
1396 mgs_fs_cleanup(env, mgs);
1398 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1399 obd->obd_namespace = NULL;
1401 lu_site_purge(env, d->ld_site, ~0);
1402 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1403 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1404 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1407 LASSERT(mgs->mgs_bottom_exp);
1408 obd_disconnect(mgs->mgs_bottom_exp);
1410 server_put_mount(obd->obd_name, true);
1415 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1416 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1418 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1420 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1422 static struct lu_device_type_operations mgs_device_type_ops = {
1423 .ldto_init = mgs_type_init,
1424 .ldto_fini = mgs_type_fini,
1426 .ldto_start = mgs_type_start,
1427 .ldto_stop = mgs_type_stop,
1429 .ldto_device_alloc = mgs_device_alloc,
1430 .ldto_device_free = mgs_device_free,
1432 .ldto_device_fini = mgs_device_fini
1435 static struct lu_device_type mgs_device_type = {
1436 .ldt_tags = LU_DEVICE_DT,
1437 .ldt_name = LUSTRE_MGS_NAME,
1438 .ldt_ops = &mgs_device_type_ops,
1439 .ldt_ctx_tags = LCT_MG_THREAD
1442 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1443 struct obd_device *obd, struct obd_uuid *cluuid,
1444 struct obd_connect_data *data, void *localdata)
1446 struct obd_export *lexp;
1447 struct lustre_handle conn = { 0 };
1452 if (exp == NULL || obd == NULL || cluuid == NULL)
1455 rc = class_connect(&conn, obd, cluuid);
1459 lexp = class_conn2export(&conn);
1464 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1465 data->ocd_version = LUSTRE_VERSION_CODE;
1466 lexp->exp_connect_data = *data;
1469 tgt_counter_incr(lexp, LPROC_MGS_CONNECT);
1471 rc = mgs_export_stats_init(obd, lexp, localdata);
1473 class_disconnect(lexp);
1480 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1481 struct obd_device *obd, struct obd_uuid *cluuid,
1482 struct obd_connect_data *data, void *localdata)
1486 if (exp == NULL || obd == NULL || cluuid == NULL)
1489 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1492 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1493 data->ocd_version = LUSTRE_VERSION_CODE;
1494 exp->exp_connect_data = *data;
1497 RETURN(mgs_export_stats_init(obd, exp, localdata));
1500 static int mgs_obd_disconnect(struct obd_export *exp)
1508 mgs_fsc_cleanup(exp);
1510 class_export_get(exp);
1511 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1513 rc = server_disconnect_export(exp);
1514 class_export_put(exp);
1518 /* use obd ops to offer management infrastructure */
1519 static struct obd_ops mgs_obd_device_ops = {
1520 .o_owner = THIS_MODULE,
1521 .o_connect = mgs_obd_connect,
1522 .o_reconnect = mgs_obd_reconnect,
1523 .o_disconnect = mgs_obd_disconnect,
1524 .o_init_export = mgs_init_export,
1525 .o_destroy_export = mgs_destroy_export,
1526 .o_iocontrol = mgs_iocontrol,
1529 static int __init mgs_init(void)
1531 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1532 LUSTRE_MGS_NAME, &mgs_device_type);
1535 static void /*__exit*/ mgs_exit(void)
1537 class_unregister_type(LUSTRE_MGS_NAME);
1540 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1541 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1542 MODULE_LICENSE("GPL");
1544 module_init(mgs_init);
1545 module_exit(mgs_exit);