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 rc = tgt_connect(tsi);
67 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
68 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
73 static int mgs_disconnect(struct tgt_session_info *tsi)
79 LASSERT(tsi->tsi_exp);
81 rc = tgt_disconnect(tsi);
83 RETURN(err_serious(rc));
87 static int mgs_exception(struct tgt_session_info *tsi)
91 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
96 static int mgs_set_info(struct tgt_session_info *tsi)
98 struct mgs_thread_info *mgi;
99 struct mgs_send_param *msp, *rep_msp;
100 struct lustre_cfg *lcfg;
105 mgi = mgs_env_info(tsi->tsi_env);
107 RETURN(err_serious(PTR_ERR(mgi)));
109 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
111 RETURN(err_serious(-EFAULT));
113 /* Construct lustre_cfg structure to pass to function mgs_setparam */
114 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
115 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
116 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
118 rc = mgs_setparam(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg,
121 LCONSOLE_WARN("%s: Unable to set parameter %s for %s: %d\n",
122 tgt_name(tsi->tsi_tgt), msp->mgs_param,
123 mgi->mgi_fsname, rc);
127 /* send back the whole msp in the reply */
128 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
132 lustre_cfg_free(lcfg);
142 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
143 void *cbdata, enum ast_type type)
147 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
148 LDLM_FL_BLOCK_CONV))) {
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 = (int)(ptr - logname);
544 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
545 LCONSOLE_WARN("%s: non-config logname received: %s\n",
546 tgt_name(tsi->tsi_tgt), logname);
547 /* not error, this can be llog test name */
549 strncpy(mgi->mgi_fsname, logname, len);
550 mgi->mgi_fsname[len] = 0;
552 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
554 if (rc && rc != -EEXIST) {
555 LCONSOLE_WARN("%s: Unable to add client %s "
556 "to file system %s: %d\n",
557 tgt_name(tsi->tsi_tgt),
558 libcfs_nid2str(req->rq_peer.nid),
559 mgi->mgi_fsname, rc);
565 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
571 static inline int mgs_init_export(struct obd_export *exp)
573 struct mgs_export_data *data = &exp->u.eu_mgs_data;
575 /* init mgs_export_data for fsc */
576 spin_lock_init(&data->med_lock);
577 CFS_INIT_LIST_HEAD(&data->med_clients);
579 spin_lock(&exp->exp_lock);
580 exp->exp_connecting = 1;
581 spin_unlock(&exp->exp_lock);
583 /* self-export doesn't need client data and ldlm initialization */
584 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
585 &exp->exp_client_uuid)))
587 return ldlm_init_export(exp);
590 static inline int mgs_destroy_export(struct obd_export *exp)
594 target_destroy_export(exp);
595 mgs_client_free(exp);
597 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
598 &exp->exp_client_uuid)))
601 ldlm_destroy_export(exp);
606 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
611 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
619 strcpy(poolname, ptr);
624 static int mgs_iocontrol_nodemap(const struct lu_env *env,
625 struct mgs_device *mgs,
626 struct obd_ioctl_data *data)
628 struct lustre_cfg *lcfg = NULL;
629 struct lu_nodemap *nodemap;
631 const char *nodemap_name = NULL;
632 const char *nidstr = NULL;
633 const char *client_idstr = NULL;
634 const char *idtype_str = NULL;
645 if (data->ioc_type != LUSTRE_CFG_TYPE) {
646 CERROR("%s: unknown cfg record type: %d\n",
647 mgs->mgs_obd->obd_name, data->ioc_type);
648 GOTO(out, rc = -EINVAL);
651 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
652 GOTO(out, rc = -E2BIG);
654 OBD_ALLOC(lcfg, data->ioc_plen1);
656 GOTO(out, rc = -ENOMEM);
658 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
659 GOTO(out_lcfg, rc = -EFAULT);
661 cmd = lcfg->lcfg_command;
664 case LCFG_NODEMAP_ACTIVATE:
665 if (lcfg->lcfg_bufcount != 2)
666 GOTO(out_lcfg, rc = -EINVAL);
667 param = lustre_cfg_string(lcfg, 1);
668 if (strcmp(param, "1") == 0)
673 case LCFG_NODEMAP_ADD:
674 case LCFG_NODEMAP_DEL:
675 if (lcfg->lcfg_bufcount != 2)
676 GOTO(out_lcfg, rc = -EINVAL);
677 nodemap_name = lustre_cfg_string(lcfg, 1);
678 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
680 case LCFG_NODEMAP_TEST_NID:
681 if (lcfg->lcfg_bufcount != 2)
682 GOTO(out_lcfg, rc = -EINVAL);
683 nidstr = lustre_cfg_string(lcfg, 1);
684 nid = libcfs_str2nid(nidstr);
685 nodemap = nodemap_classify_nid(nid);
686 memset(data->ioc_pbuf1, 0, data->ioc_plen1);
687 if (copy_to_user(data->ioc_pbuf1, nodemap->nm_name,
688 strlen(nodemap->nm_name)) != 0)
689 GOTO(out_lcfg, rc = -EFAULT);
691 case LCFG_NODEMAP_TEST_ID:
692 if (lcfg->lcfg_bufcount != 4)
693 GOTO(out_lcfg, rc = -EINVAL);
694 nidstr = lustre_cfg_string(lcfg, 1);
695 idtype_str = lustre_cfg_string(lcfg, 2);
696 client_idstr = lustre_cfg_string(lcfg, 3);
698 nid = libcfs_str2nid(nidstr);
699 nodemap = nodemap_classify_nid(nid);
700 client_id = simple_strtoul(client_idstr, NULL, 10);
702 if (strcmp(idtype_str, "uid") == 0)
703 idtype = NODEMAP_UID;
705 idtype = NODEMAP_GID;
707 fs_id = nodemap_map_id(nodemap, idtype, NODEMAP_CLIENT_TO_FS,
710 if (data->ioc_plen1 < sizeof(fs_idstr))
711 GOTO(out_lcfg, rc = -EINVAL);
713 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
714 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
715 sizeof(fs_idstr)) != 0)
716 GOTO(out_lcfg, rc = -EINVAL);
718 case LCFG_NODEMAP_ADD_RANGE:
719 case LCFG_NODEMAP_DEL_RANGE:
720 case LCFG_NODEMAP_ADD_UIDMAP:
721 case LCFG_NODEMAP_DEL_UIDMAP:
722 case LCFG_NODEMAP_ADD_GIDMAP:
723 case LCFG_NODEMAP_DEL_GIDMAP:
724 if (lcfg->lcfg_bufcount != 3)
725 GOTO(out_lcfg, rc = -EINVAL);
726 nodemap_name = lustre_cfg_string(lcfg, 1);
727 param = lustre_cfg_string(lcfg, 2);
728 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
730 case LCFG_NODEMAP_ADMIN:
731 case LCFG_NODEMAP_TRUSTED:
732 case LCFG_NODEMAP_SQUASH_UID:
733 case LCFG_NODEMAP_SQUASH_GID:
734 if (lcfg->lcfg_bufcount != 4)
735 GOTO(out_lcfg, rc = -EINVAL);
736 nodemap_name = lustre_cfg_string(lcfg, 1);
737 param = lustre_cfg_string(lcfg, 3);
738 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
745 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
746 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
752 OBD_FREE(lcfg, data->ioc_plen1);
757 static int mgs_iocontrol_pool(const struct lu_env *env,
758 struct mgs_device *mgs,
759 struct obd_ioctl_data *data)
761 struct mgs_thread_info *mgi = mgs_env_info(env);
763 struct lustre_cfg *lcfg = NULL;
764 char *poolname = NULL;
767 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
768 if (poolname == NULL)
771 if (data->ioc_type != LUSTRE_CFG_TYPE) {
772 CERROR("%s: unknown cfg record type: %d\n",
773 mgs->mgs_obd->obd_name, data->ioc_type);
774 GOTO(out_pool, rc = -EINVAL);
777 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
778 GOTO(out_pool, rc = -E2BIG);
780 OBD_ALLOC(lcfg, data->ioc_plen1);
782 GOTO(out_pool, rc = -ENOMEM);
784 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
785 GOTO(out_lcfg, rc = -EFAULT);
787 if (lcfg->lcfg_bufcount < 2)
788 GOTO(out_lcfg, rc = -EFAULT);
790 /* first arg is always <fsname>.<poolname> */
791 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
796 switch (lcfg->lcfg_command) {
798 if (lcfg->lcfg_bufcount != 2)
799 GOTO(out_lcfg, rc = -EINVAL);
800 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
804 if (lcfg->lcfg_bufcount != 3)
805 GOTO(out_lcfg, rc = -EINVAL);
806 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
807 poolname, lustre_cfg_string(lcfg, 2));
810 if (lcfg->lcfg_bufcount != 3)
811 GOTO(out_lcfg, rc = -EINVAL);
812 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
813 poolname, lustre_cfg_string(lcfg, 2));
816 if (lcfg->lcfg_bufcount != 2)
817 GOTO(out_lcfg, rc = -EINVAL);
818 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
826 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
827 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
832 OBD_FREE(lcfg, data->ioc_plen1);
834 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
838 /* from mdt_iocontrol */
839 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
840 void *karg, void *uarg)
842 struct mgs_device *mgs = exp2mgs_dev(exp);
843 struct obd_ioctl_data *data = karg;
848 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
850 rc = lu_env_init(&env, LCT_MG_THREAD);
856 case OBD_IOC_PARAM: {
857 struct mgs_thread_info *mgi = mgs_env_info(&env);
858 struct lustre_cfg *lcfg;
860 if (data->ioc_type != LUSTRE_CFG_TYPE) {
861 CERROR("%s: unknown cfg record type: %d\n",
862 mgs->mgs_obd->obd_name, data->ioc_type);
863 GOTO(out, rc = -EINVAL);
866 OBD_ALLOC(lcfg, data->ioc_plen1);
868 GOTO(out, rc = -ENOMEM);
869 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
870 GOTO(out_free, rc = -EFAULT);
872 if (lcfg->lcfg_bufcount < 1)
873 GOTO(out_free, rc = -EINVAL);
875 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
877 CERROR("%s: setparam err: rc = %d\n",
878 exp->exp_obd->obd_name, rc);
880 OBD_FREE(lcfg, data->ioc_plen1);
884 case OBD_IOC_REPLACE_NIDS: {
885 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
886 CERROR("No device name specified!\n");
891 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
892 CERROR("Device name is not NUL terminated!\n");
897 if (data->ioc_plen1 > MTI_NAME_MAXLEN) {
898 CERROR("Device name is too long\n");
903 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
904 CERROR("No NIDs were specified!\n");
909 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
910 CERROR("NID list is not NUL terminated!\n");
915 /* replace nids in llog */
916 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
919 CERROR("%s: error replacing nids: rc = %d\n",
920 exp->exp_obd->obd_name, rc);
926 rc = mgs_iocontrol_pool(&env, mgs, data);
929 case OBD_IOC_NODEMAP:
930 rc = mgs_iocontrol_nodemap(&env, mgs, data);
933 case OBD_IOC_DUMP_LOG: {
934 struct llog_ctxt *ctxt;
936 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
937 rc = class_config_dump_llog(&env, ctxt, data->ioc_inlbuf1,
944 case OBD_IOC_CATLOGLIST:
945 rc = mgs_list_logs(&env, mgs, data);
947 case OBD_IOC_LLOG_CANCEL:
948 case OBD_IOC_LLOG_REMOVE:
949 case OBD_IOC_LLOG_CHECK:
950 case OBD_IOC_LLOG_INFO:
951 case OBD_IOC_LLOG_PRINT: {
952 struct llog_ctxt *ctxt;
954 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
955 rc = llog_ioctl(&env, ctxt, cmd, data);
961 CERROR("%s: unknown command %#x\n",
962 mgs->mgs_obd->obd_name, cmd);
971 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
973 struct obd_connect_data *data = NULL;
974 struct obd_device *obd;
982 obd = class_name2obd(nextdev);
984 CERROR("can't locate next device: %s\n", nextdev);
985 GOTO(out, rc = -ENOTCONN);
988 data->ocd_version = LUSTRE_VERSION_CODE;
990 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
991 &obd->obd_uuid, data, NULL);
993 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
997 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
998 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
999 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1005 static struct tgt_handler mgs_mgs_handlers[] = {
1006 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1007 0, MGS_CONNECT, mgs_connect,
1008 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1009 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1010 0, MGS_DISCONNECT, mgs_disconnect,
1011 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1012 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1013 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
1014 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
1015 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1016 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
1019 static struct tgt_handler mgs_obd_handlers[] = {
1020 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1023 static struct tgt_handler mgs_dlm_handlers[] = {
1024 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1025 .th_name = "LDLM_ENQUEUE",
1026 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1027 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1028 * * is safe. If such tests will be needed we have to distinguish
1029 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1030 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1032 .th_opc = LDLM_ENQUEUE,
1033 .th_flags = HABEO_CLAVIS,
1034 .th_act = tgt_enqueue,
1035 .th_fmt = &RQF_LDLM_ENQUEUE,
1036 .th_version = LUSTRE_DLM_VERSION,
1040 static struct tgt_handler mgs_llog_handlers[] = {
1041 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1042 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1043 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1044 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1045 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1048 static struct tgt_opc_slice mgs_common_slice[] = {
1050 .tos_opc_start = MGS_FIRST_OPC,
1051 .tos_opc_end = MGS_LAST_OPC,
1052 .tos_hs = mgs_mgs_handlers
1055 .tos_opc_start = OBD_FIRST_OPC,
1056 .tos_opc_end = OBD_LAST_OPC,
1057 .tos_hs = mgs_obd_handlers
1060 .tos_opc_start = LDLM_FIRST_OPC,
1061 .tos_opc_end = LDLM_LAST_OPC,
1062 .tos_hs = mgs_dlm_handlers
1065 .tos_opc_start = LLOG_FIRST_OPC,
1066 .tos_opc_end = LLOG_LAST_OPC,
1067 .tos_hs = mgs_llog_handlers
1070 .tos_opc_start = SEC_FIRST_OPC,
1071 .tos_opc_end = SEC_LAST_OPC,
1072 .tos_hs = tgt_sec_ctx_handlers
1079 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1080 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1082 struct ptlrpc_service_conf conf;
1083 struct obd_device *obd;
1084 struct lustre_mount_info *lmi;
1085 struct llog_ctxt *ctxt;
1086 struct fs_db *fsdb = NULL;
1091 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1095 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1097 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1101 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1104 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1106 obd->u.obt.obt_magic = OBT_MAGIC;
1107 obd->u.obt.obt_instance = 0;
1109 /* namespace for mgs llog */
1110 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1111 LDLM_NAMESPACE_SERVER,
1112 LDLM_NAMESPACE_MODEST,
1114 if (obd->obd_namespace == NULL)
1115 GOTO(err_ops, rc = -ENOMEM);
1117 /* No recovery for MGCs */
1118 obd->obd_replayable = 0;
1120 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1121 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1122 OBD_FAIL_MGS_ALL_REPLY_NET);
1126 rc = mgs_fs_setup(env, mgs);
1128 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1133 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1134 obd, &llog_osd_ops);
1138 /* XXX: we need this trick till N:1 stack is supported
1139 * set "current" directory for named llogs */
1140 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1142 ctxt->loc_dir = mgs->mgs_configs_dir;
1143 llog_ctxt_put(ctxt);
1145 /* Internal mgs setup */
1146 mgs_init_fsdb_list(mgs);
1147 mutex_init(&mgs->mgs_mutex);
1148 mgs->mgs_start_time = cfs_time_current_sec();
1149 spin_lock_init(&mgs->mgs_lock);
1151 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1153 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1158 /* Setup params fsdb and log, so that other servers can make a local
1159 * copy successfully when they are mounted. See LU-4783 */
1160 rc = mgs_params_fsdb_setup(env, mgs, fsdb);
1162 /* params fsdb and log can be setup later */
1163 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1164 obd->obd_name, PARAMS_FILENAME, rc);
1166 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1167 "mgs_ldlm_client", &obd->obd_ldlm_client);
1169 conf = (typeof(conf)) {
1170 .psc_name = LUSTRE_MGS_NAME,
1171 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1173 .bc_nbufs = MGS_NBUFS,
1174 .bc_buf_size = MGS_BUFSIZE,
1175 .bc_req_max_size = MGS_MAXREQSIZE,
1176 .bc_rep_max_size = MGS_MAXREPSIZE,
1177 .bc_req_portal = MGS_REQUEST_PORTAL,
1178 .bc_rep_portal = MGC_REPLY_PORTAL,
1181 .tc_thr_name = "ll_mgs",
1182 .tc_nthrs_init = MGS_NTHRS_INIT,
1183 .tc_nthrs_max = MGS_NTHRS_MAX,
1184 .tc_ctx_tags = LCT_MG_THREAD,
1187 .so_req_handler = tgt_request_handle,
1188 .so_req_printer = target_print_req,
1192 /* Start the service threads */
1193 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1194 if (IS_ERR(mgs->mgs_service)) {
1195 rc = PTR_ERR(mgs->mgs_service);
1196 CERROR("failed to start mgs service: %d\n", rc);
1197 mgs->mgs_service = NULL;
1198 GOTO(err_lproc, rc);
1201 ping_evictor_start();
1203 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1205 /* device stack is not yet fully setup to keep no objects behind */
1206 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1209 mgs_params_fsdb_cleanup(env, mgs);
1210 lproc_mgs_cleanup(mgs);
1212 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1214 ctxt->loc_dir = NULL;
1215 llog_cleanup(env, ctxt);
1218 tgt_fini(env, &mgs->mgs_lut);
1220 /* No extra cleanup needed for llog_init_commit_thread() */
1221 mgs_fs_cleanup(env, mgs);
1223 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1224 obd->obd_namespace = NULL;
1226 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1227 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1228 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1229 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1232 obd_disconnect(mgs->mgs_bottom_exp);
1235 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1239 static struct lu_device *mgs_device_free(const struct lu_env *env,
1240 struct lu_device *lu)
1242 struct mgs_device *mgs = lu2mgs_dev(lu);
1245 dt_device_fini(&mgs->mgs_dt_dev);
1250 static int mgs_process_config(const struct lu_env *env,
1251 struct lu_device *dev,
1252 struct lustre_cfg *lcfg)
1258 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1259 const struct lu_object_conf *unused)
1261 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1262 struct lu_device *under;
1263 struct lu_object *below;
1267 /* do no set .do_ops as mgs calls to bottom osd directly */
1269 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1270 PFID(lu_object_fid(o)));
1272 under = &d->mgs_bottom->dd_lu_dev;
1273 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1275 lu_object_add(o, below);
1282 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1284 struct mgs_object *obj = lu2mgs_obj(o);
1285 struct lu_object_header *h = o->lo_header;
1287 dt_object_fini(&obj->mgo_obj);
1288 lu_object_header_fini(h);
1292 static int mgs_object_print(const struct lu_env *env, void *cookie,
1293 lu_printer_t p, const struct lu_object *l)
1295 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1297 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1300 struct lu_object_operations mgs_lu_obj_ops = {
1301 .loo_object_init = mgs_object_init,
1302 .loo_object_free = mgs_object_free,
1303 .loo_object_print = mgs_object_print,
1306 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1307 const struct lu_object_header *hdr,
1308 struct lu_device *d)
1310 struct lu_object_header *h;
1311 struct mgs_object *o;
1312 struct lu_object *l;
1314 LASSERT(hdr == NULL);
1318 l = &o->mgo_obj.do_lu;
1321 lu_object_header_init(h);
1322 dt_object_init(&o->mgo_obj, h, d);
1323 lu_object_add_top(h, l);
1325 l->lo_ops = &mgs_lu_obj_ops;
1333 const struct lu_device_operations mgs_lu_ops = {
1334 .ldo_object_alloc = mgs_object_alloc,
1335 .ldo_process_config = mgs_process_config,
1338 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1339 struct lu_device_type *type,
1340 struct lustre_cfg *lcfg)
1342 struct mgs_device *mgs;
1343 struct lu_device *ludev;
1347 ludev = ERR_PTR(-ENOMEM);
1351 ludev = mgs2lu_dev(mgs);
1352 dt_device_init(&mgs->mgs_dt_dev, type);
1353 rc = mgs_init0(env, mgs, type, lcfg);
1355 mgs_device_free(env, ludev);
1356 ludev = ERR_PTR(rc);
1362 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1363 struct lu_device *d)
1365 struct mgs_device *mgs = lu2mgs_dev(d);
1366 struct obd_device *obd = mgs->mgs_obd;
1367 struct llog_ctxt *ctxt;
1371 LASSERT(mgs->mgs_bottom);
1373 class_disconnect_exports(obd);
1375 ping_evictor_stop();
1377 ptlrpc_unregister_service(mgs->mgs_service);
1379 obd_exports_barrier(obd);
1380 obd_zombie_barrier();
1382 tgt_fini(env, &mgs->mgs_lut);
1383 mgs_params_fsdb_cleanup(env, mgs);
1384 mgs_cleanup_fsdb_list(mgs);
1385 lproc_mgs_cleanup(mgs);
1387 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1389 ctxt->loc_dir = NULL;
1390 llog_cleanup(env, ctxt);
1393 mgs_fs_cleanup(env, mgs);
1395 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1396 obd->obd_namespace = NULL;
1398 lu_site_purge(env, d->ld_site, ~0);
1399 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1400 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1401 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1404 LASSERT(mgs->mgs_bottom_exp);
1405 obd_disconnect(mgs->mgs_bottom_exp);
1407 server_put_mount(obd->obd_name, true);
1412 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1413 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1415 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1417 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1419 static struct lu_device_type_operations mgs_device_type_ops = {
1420 .ldto_init = mgs_type_init,
1421 .ldto_fini = mgs_type_fini,
1423 .ldto_start = mgs_type_start,
1424 .ldto_stop = mgs_type_stop,
1426 .ldto_device_alloc = mgs_device_alloc,
1427 .ldto_device_free = mgs_device_free,
1429 .ldto_device_fini = mgs_device_fini
1432 static struct lu_device_type mgs_device_type = {
1433 .ldt_tags = LU_DEVICE_DT,
1434 .ldt_name = LUSTRE_MGS_NAME,
1435 .ldt_ops = &mgs_device_type_ops,
1436 .ldt_ctx_tags = LCT_MG_THREAD
1439 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1440 struct obd_device *obd, struct obd_uuid *cluuid,
1441 struct obd_connect_data *data, void *localdata)
1443 struct obd_export *lexp;
1444 struct lustre_handle conn = { 0 };
1449 if (exp == NULL || obd == NULL || cluuid == NULL)
1452 rc = class_connect(&conn, obd, cluuid);
1456 lexp = class_conn2export(&conn);
1461 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1462 data->ocd_version = LUSTRE_VERSION_CODE;
1463 lexp->exp_connect_data = *data;
1466 tgt_counter_incr(lexp, LPROC_MGS_CONNECT);
1468 rc = mgs_export_stats_init(obd, lexp, localdata);
1470 class_disconnect(lexp);
1477 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1478 struct obd_device *obd, struct obd_uuid *cluuid,
1479 struct obd_connect_data *data, void *localdata)
1483 if (exp == NULL || obd == NULL || cluuid == NULL)
1486 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1489 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1490 data->ocd_version = LUSTRE_VERSION_CODE;
1491 exp->exp_connect_data = *data;
1494 RETURN(mgs_export_stats_init(obd, exp, localdata));
1497 static int mgs_obd_disconnect(struct obd_export *exp)
1505 mgs_fsc_cleanup(exp);
1507 class_export_get(exp);
1508 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1510 rc = server_disconnect_export(exp);
1511 class_export_put(exp);
1515 /* use obd ops to offer management infrastructure */
1516 static struct obd_ops mgs_obd_device_ops = {
1517 .o_owner = THIS_MODULE,
1518 .o_connect = mgs_obd_connect,
1519 .o_reconnect = mgs_obd_reconnect,
1520 .o_disconnect = mgs_obd_disconnect,
1521 .o_init_export = mgs_init_export,
1522 .o_destroy_export = mgs_destroy_export,
1523 .o_iocontrol = mgs_iocontrol,
1526 static int __init mgs_init(void)
1528 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1529 #ifndef HAVE_ONLY_PROCFS_SEQ
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);