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_param.h>
50 #include "mgs_internal.h"
53 * Regular MGS handlers
55 static int mgs_connect(struct tgt_session_info *tsi)
57 struct ptlrpc_request *req = tgt_ses_req(tsi);
62 rc = tgt_connect(tsi);
66 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
67 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
72 static int mgs_disconnect(struct tgt_session_info *tsi)
78 LASSERT(tsi->tsi_exp);
80 rc = tgt_disconnect(tsi);
82 RETURN(err_serious(rc));
86 static int mgs_exception(struct tgt_session_info *tsi)
90 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
95 static int mgs_set_info(struct tgt_session_info *tsi)
97 struct mgs_thread_info *mgi;
98 struct mgs_send_param *msp, *rep_msp;
99 struct lustre_cfg *lcfg;
104 mgi = mgs_env_info(tsi->tsi_env);
106 RETURN(err_serious(PTR_ERR(mgi)));
108 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
110 RETURN(err_serious(-EFAULT));
112 /* Construct lustre_cfg structure to pass to function mgs_setparam */
113 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
114 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
115 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
117 rc = mgs_setparam(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg,
120 LCONSOLE_WARN("%s: Unable to set parameter %s for %s: %d\n",
121 tgt_name(tsi->tsi_tgt), msp->mgs_param,
122 mgi->mgi_fsname, rc);
126 /* send back the whole msp in the reply */
127 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
131 lustre_cfg_free(lcfg);
141 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
142 void *cbdata, enum ast_type type)
146 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
149 /* l_ast_data is used as a marker to avoid cancel ldlm lock
150 * twice. See LU-2317. */
151 lock_res_and_lock(lock);
152 fsdb = (struct fs_db *)lock->l_ast_data;
153 lock->l_ast_data = NULL;
154 unlock_res_and_lock(lock);
157 struct lustre_handle lockh;
161 /* clear the bit before lock put */
162 clear_bit(FSDB_REVOKING_LOCK,
166 clear_bit(FSDB_REVOKING_PARAMS,
170 mgs_ir_notify_complete(fsdb);
176 ldlm_lock2handle(lock, &lockh);
177 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
181 RETURN(ldlm_completion_ast(lock, flags, cbdata));
184 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
187 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
190 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
193 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
196 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
199 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
202 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
204 ldlm_completion_callback cp = NULL;
205 struct lustre_handle lockh = { 0 };
206 struct ldlm_res_id res_id;
207 __u64 flags = LDLM_FL_ATOMIC_CB;
211 LASSERT(fsdb->fsdb_name[0] != '\0');
212 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
215 case CONFIG_T_CONFIG:
216 cp = mgs_completion_ast_config;
217 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
220 case CONFIG_T_PARAMS:
221 cp = mgs_completion_ast_params;
222 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
225 case CONFIG_T_RECOVER:
226 cp = mgs_completion_ast_ir;
233 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
234 &res_id, LDLM_PLAIN, NULL, LCK_EX,
235 &flags, ldlm_blocking_ast, cp,
236 NULL, fsdb, 0, LVB_T_NONE, NULL,
238 if (rc != ELDLM_OK) {
239 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
240 le64_to_cpu(res_id.name[0]),
241 le64_to_cpu(res_id.name[1]), rc);
243 if (type == CONFIG_T_CONFIG)
244 clear_bit(FSDB_REVOKING_LOCK,
247 if (type == CONFIG_T_PARAMS)
248 clear_bit(FSDB_REVOKING_PARAMS,
251 /* lock has been cancelled in completion_ast. */
260 static int mgs_check_target(const struct lu_env *env,
261 struct mgs_device *mgs,
262 struct mgs_target_info *mti)
267 rc = mgs_check_index(env, mgs, mti);
269 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
270 "this MGS does not know about it, preventing "
271 "registration.\n", mti->mti_svname);
273 } else if (rc == -1) {
274 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
275 "disappeared! Regenerating all logs.\n",
277 mti->mti_flags |= LDD_F_WRITECONF;
280 /* Index is correctly marked as used */
282 /* If the logs don't contain the mti_nids then add
283 them as failover nids */
284 rc = mgs_check_failnid(env, mgs, mti);
290 /* Ensure this is not a failover node that is connecting first*/
291 static int mgs_check_failover_reg(struct mgs_target_info *mti)
297 ptr = mti->mti_params;
298 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
299 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
300 for (i = 0; i < mti->mti_nid_count; i++) {
301 if (nid == mti->mti_nids[i]) {
302 LCONSOLE_WARN("Denying initial registra"
303 "tion attempt from nid %s"
304 ", specified as failover"
305 "\n",libcfs_nid2str(nid));
306 return -EADDRNOTAVAIL;
314 /* Called whenever a target starts up. Flags indicate first connect, etc. */
315 static int mgs_target_reg(struct tgt_session_info *tsi)
317 struct obd_device *obd = tsi->tsi_exp->exp_obd;
318 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
319 struct mgs_target_info *mti, *rep_mti;
326 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
328 return err_serious(rc);
330 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
332 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
334 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
335 RETURN(err_serious(-EFAULT));
338 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
340 opc = mti->mti_flags & LDD_F_OPC_MASK;
344 if (opc == LDD_F_OPC_READY) {
345 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
346 mti->mti_fsname, mti->mti_stripe_index);
347 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
349 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
350 CERROR("Update IR return with %d(ignore and IR "
353 GOTO(out_nolock, rc);
356 /* Do not support unregistering right now. */
357 if (opc != LDD_F_OPC_REG)
358 GOTO(out_nolock, rc = -EINVAL);
360 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
361 mti->mti_fsname, mti->mti_stripe_index);
363 if (mti->mti_flags & LDD_F_NEED_INDEX)
364 mti->mti_flags |= LDD_F_WRITECONF;
366 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
368 /* We're just here as a startup ping. */
369 CDEBUG(D_MGS, "Server %s is running on %s\n",
370 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
371 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
372 /* above will set appropriate mti flags */
374 /* Nothing wrong, or fatal error */
375 GOTO(out_nolock, rc);
376 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
377 rc = mgs_check_failover_reg(mti);
379 GOTO(out_nolock, rc);
382 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
384 if (mti->mti_flags & LDD_F_WRITECONF) {
385 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
386 mti->mti_stripe_index == 0) {
387 rc = mgs_erase_logs(tsi->tsi_env, mgs,
389 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
390 "request. All servers must be restarted "
391 "in order to regenerate the logs."
392 "\n", obd->obd_name, mti->mti_fsname);
393 } else if (mti->mti_flags &
394 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
395 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
396 LCONSOLE_WARN("%s: Regenerating %s log by user "
398 obd->obd_name, mti->mti_svname);
400 mti->mti_flags |= LDD_F_UPDATE;
401 /* Erased logs means start from scratch. */
402 mti->mti_flags &= ~LDD_F_UPGRADE14;
404 GOTO(out_nolock, rc);
407 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &fsdb);
409 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
410 GOTO(out_nolock, rc);
414 * Log writing contention is handled by the fsdb_mutex.
416 * It should be alright if someone was reading while we were
417 * updating the logs - if we revoke at the end they will just update
418 * from where they left off.
421 if (mti->mti_flags & LDD_F_UPGRADE14) {
422 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
426 if (mti->mti_flags & LDD_F_UPDATE) {
427 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
428 mti->mti_stripe_index);
430 /* create or update the target log
431 and update the client/mdt logs */
432 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, fsdb);
434 CERROR("Failed to write %s log (%d)\n",
435 mti->mti_svname, rc);
439 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
440 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
442 mti->mti_flags |= LDD_F_REWRITE_LDD;
446 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
449 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
450 mti->mti_stripe_index, rc);
451 /* An error flag is set in the mti reply rather than an error code */
453 mti->mti_flags |= LDD_F_ERROR;
455 /* send back the whole mti in the reply */
456 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
459 /* Flush logs to disk */
460 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
464 /* Called whenever a target cleans up. */
465 static int mgs_target_del(struct tgt_session_info *tsi)
469 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
474 static int mgs_config_read(struct tgt_session_info *tsi)
476 struct ptlrpc_request *req = tgt_ses_req(tsi);
477 struct mgs_config_body *body;
482 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
484 DEBUG_REQ(D_HA, req, "no mgs_config_body");
485 RETURN(err_serious(-EFAULT));
488 switch (body->mcb_type) {
489 case CONFIG_T_RECOVER:
490 rc = mgs_get_ir_logs(req);
492 case CONFIG_T_CONFIG:
503 static int mgs_llog_open(struct tgt_session_info *tsi)
505 struct mgs_thread_info *mgi;
506 struct ptlrpc_request *req = tgt_ses_req(tsi);
512 rc = tgt_llog_open(tsi);
517 * For old clients there is no direct way of knowing which file system
518 * a client is operating at the MGS side. But we need to pick up those
519 * clients so that the MGS can mark the corresponding file system as
520 * non-IR capable because old clients are not ready to be notified.
522 * Therefore we attempt to detect the file systems name by hacking the
523 * llog operation which is currently used by the clients to fetch
524 * configuration logs. At present this is fine because this is the
525 * ONLY llog operation between mgc and the MGS.
527 * If extra llog operation are going to be added, this function needs
530 * When releases prior than 2.0 are not supported, the following code
533 mgi = mgs_env_info(tsi->tsi_env);
535 RETURN(PTR_ERR(mgi));
537 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
539 char *ptr = strchr(logname, '-');
540 int len = (int)(ptr - logname);
542 if ((ptr == NULL && strcmp(logname, PARAMS_FILENAME) != 0) ||
543 len >= sizeof(mgi->mgi_fsname)) {
544 LCONSOLE_WARN("%s: non-config logname received: %s\n",
545 tgt_name(tsi->tsi_tgt), logname);
546 /* not error, this can be llog test name */
548 strncpy(mgi->mgi_fsname, logname, len);
549 mgi->mgi_fsname[len] = 0;
551 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
553 if (rc && rc != -EEXIST) {
554 LCONSOLE_WARN("%s: Unable to add client %s "
555 "to file system %s: %d\n",
556 tgt_name(tsi->tsi_tgt),
557 libcfs_nid2str(req->rq_peer.nid),
558 mgi->mgi_fsname, rc);
564 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
570 static inline int mgs_init_export(struct obd_export *exp)
572 struct mgs_export_data *data = &exp->u.eu_mgs_data;
574 /* init mgs_export_data for fsc */
575 spin_lock_init(&data->med_lock);
576 CFS_INIT_LIST_HEAD(&data->med_clients);
578 spin_lock(&exp->exp_lock);
579 exp->exp_connecting = 1;
580 spin_unlock(&exp->exp_lock);
582 /* self-export doesn't need client data and ldlm initialization */
583 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
584 &exp->exp_client_uuid)))
586 return ldlm_init_export(exp);
589 static inline int mgs_destroy_export(struct obd_export *exp)
593 target_destroy_export(exp);
594 mgs_client_free(exp);
596 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
597 &exp->exp_client_uuid)))
600 ldlm_destroy_export(exp);
605 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
610 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
618 strcpy(poolname, ptr);
623 static int mgs_iocontrol_pool(const struct lu_env *env,
624 struct mgs_device *mgs,
625 struct obd_ioctl_data *data)
627 struct mgs_thread_info *mgi = mgs_env_info(env);
629 struct lustre_cfg *lcfg = NULL;
630 char *poolname = NULL;
633 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
634 if (poolname == NULL)
637 if (data->ioc_type != LUSTRE_CFG_TYPE) {
638 CERROR("%s: unknown cfg record type: %d\n",
639 mgs->mgs_obd->obd_name, data->ioc_type);
640 GOTO(out_pool, rc = -EINVAL);
643 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
644 GOTO(out_pool, rc = -E2BIG);
646 OBD_ALLOC(lcfg, data->ioc_plen1);
648 GOTO(out_pool, rc = -ENOMEM);
650 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
651 GOTO(out_lcfg, rc = -EFAULT);
653 if (lcfg->lcfg_bufcount < 2)
654 GOTO(out_lcfg, rc = -EFAULT);
656 /* first arg is always <fsname>.<poolname> */
657 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
662 switch (lcfg->lcfg_command) {
664 if (lcfg->lcfg_bufcount != 2)
665 GOTO(out_lcfg, rc = -EINVAL);
666 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
670 if (lcfg->lcfg_bufcount != 3)
671 GOTO(out_lcfg, rc = -EINVAL);
672 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
673 poolname, lustre_cfg_string(lcfg, 2));
676 if (lcfg->lcfg_bufcount != 3)
677 GOTO(out_lcfg, rc = -EINVAL);
678 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
679 poolname, lustre_cfg_string(lcfg, 2));
682 if (lcfg->lcfg_bufcount != 2)
683 GOTO(out_lcfg, rc = -EINVAL);
684 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
692 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
693 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
698 OBD_FREE(lcfg, data->ioc_plen1);
700 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
704 /* from mdt_iocontrol */
705 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
706 void *karg, void *uarg)
708 struct mgs_device *mgs = exp2mgs_dev(exp);
709 struct obd_ioctl_data *data = karg;
714 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
716 rc = lu_env_init(&env, LCT_MG_THREAD);
722 case OBD_IOC_PARAM: {
723 struct mgs_thread_info *mgi = mgs_env_info(&env);
724 struct lustre_cfg *lcfg;
726 if (data->ioc_type != LUSTRE_CFG_TYPE) {
727 CERROR("%s: unknown cfg record type: %d\n",
728 mgs->mgs_obd->obd_name, data->ioc_type);
729 GOTO(out, rc = -EINVAL);
732 OBD_ALLOC(lcfg, data->ioc_plen1);
734 GOTO(out, rc = -ENOMEM);
735 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
736 GOTO(out_free, rc = -EFAULT);
738 if (lcfg->lcfg_bufcount < 1)
739 GOTO(out_free, rc = -EINVAL);
741 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
743 CERROR("%s: setparam err: rc = %d\n",
744 exp->exp_obd->obd_name, rc);
746 OBD_FREE(lcfg, data->ioc_plen1);
750 case OBD_IOC_REPLACE_NIDS: {
751 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
752 CERROR("No device name specified!\n");
757 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
758 CERROR("Device name is not NUL terminated!\n");
763 if (data->ioc_plen1 > MTI_NAME_MAXLEN) {
764 CERROR("Device name is too long\n");
769 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
770 CERROR("No NIDs were specified!\n");
775 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
776 CERROR("NID list is not NUL terminated!\n");
781 /* replace nids in llog */
782 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
785 CERROR("%s: error replacing nids: rc = %d\n",
786 exp->exp_obd->obd_name, rc);
792 rc = mgs_iocontrol_pool(&env, mgs, data);
795 case OBD_IOC_DUMP_LOG: {
796 struct llog_ctxt *ctxt;
798 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
799 rc = class_config_dump_llog(&env, ctxt, data->ioc_inlbuf1,
806 case OBD_IOC_LLOG_CANCEL:
807 case OBD_IOC_LLOG_REMOVE:
808 case OBD_IOC_LLOG_CHECK:
809 case OBD_IOC_LLOG_INFO:
810 case OBD_IOC_LLOG_PRINT: {
811 struct llog_ctxt *ctxt;
813 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
814 rc = llog_ioctl(&env, ctxt, cmd, data);
820 CERROR("%s: unknown command %#x\n",
821 mgs->mgs_obd->obd_name, cmd);
830 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
832 struct obd_connect_data *data = NULL;
833 struct obd_device *obd;
841 obd = class_name2obd(nextdev);
843 CERROR("can't locate next device: %s\n", nextdev);
844 GOTO(out, rc = -ENOTCONN);
847 data->ocd_version = LUSTRE_VERSION_CODE;
849 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
850 &obd->obd_uuid, data, NULL);
852 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
856 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
857 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
858 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
864 static struct tgt_handler mgs_mgs_handlers[] = {
865 TGT_RPC_HANDLER(MGS_FIRST_OPC,
866 0, MGS_CONNECT, mgs_connect,
867 &RQF_CONNECT, LUSTRE_OBD_VERSION),
868 TGT_RPC_HANDLER(MGS_FIRST_OPC,
869 0, MGS_DISCONNECT, mgs_disconnect,
870 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
871 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
872 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
873 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
874 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
875 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
878 static struct tgt_handler mgs_obd_handlers[] = {
879 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
882 static struct tgt_handler mgs_dlm_handlers[] = {
883 TGT_DLM_HDL(HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
886 static struct tgt_handler mgs_llog_handlers[] = {
887 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
888 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
889 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
890 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
891 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
894 static struct tgt_opc_slice mgs_common_slice[] = {
896 .tos_opc_start = MGS_FIRST_OPC,
897 .tos_opc_end = MGS_LAST_OPC,
898 .tos_hs = mgs_mgs_handlers
901 .tos_opc_start = OBD_FIRST_OPC,
902 .tos_opc_end = OBD_LAST_OPC,
903 .tos_hs = mgs_obd_handlers
906 .tos_opc_start = LDLM_FIRST_OPC,
907 .tos_opc_end = LDLM_LAST_OPC,
908 .tos_hs = mgs_dlm_handlers
911 .tos_opc_start = LLOG_FIRST_OPC,
912 .tos_opc_end = LLOG_LAST_OPC,
913 .tos_hs = mgs_llog_handlers
916 .tos_opc_start = SEC_FIRST_OPC,
917 .tos_opc_end = SEC_LAST_OPC,
918 .tos_hs = tgt_sec_ctx_handlers
925 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
926 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
928 struct ptlrpc_service_conf conf;
929 struct obd_device *obd;
930 struct lustre_mount_info *lmi;
931 struct llog_ctxt *ctxt;
932 struct fs_db *fsdb = NULL;
937 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
941 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
943 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
947 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
950 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
952 obd->u.obt.obt_magic = OBT_MAGIC;
953 obd->u.obt.obt_instance = 0;
955 /* namespace for mgs llog */
956 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
957 LDLM_NAMESPACE_SERVER,
958 LDLM_NAMESPACE_MODEST,
960 if (obd->obd_namespace == NULL)
961 GOTO(err_ops, rc = -ENOMEM);
963 /* No recovery for MGCs */
964 obd->obd_replayable = 0;
966 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
967 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
968 OBD_FAIL_MGS_ALL_REPLY_NET);
972 rc = mgs_fs_setup(env, mgs);
974 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
979 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
984 /* XXX: we need this trick till N:1 stack is supported
985 * set "current" directory for named llogs */
986 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
988 ctxt->loc_dir = mgs->mgs_configs_dir;
991 /* Internal mgs setup */
992 mgs_init_fsdb_list(mgs);
993 mutex_init(&mgs->mgs_mutex);
994 mgs->mgs_start_time = cfs_time_current_sec();
995 spin_lock_init(&mgs->mgs_lock);
997 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
999 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1004 /* Setup params fsdb and log, so that other servers can make a local
1005 * copy successfully when they are mounted. See LU-4783 */
1006 rc = mgs_params_fsdb_setup(env, mgs, fsdb);
1008 /* params fsdb and log can be setup later */
1009 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1010 obd->obd_name, PARAMS_FILENAME, rc);
1012 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1013 "mgs_ldlm_client", &obd->obd_ldlm_client);
1015 conf = (typeof(conf)) {
1016 .psc_name = LUSTRE_MGS_NAME,
1017 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1019 .bc_nbufs = MGS_NBUFS,
1020 .bc_buf_size = MGS_BUFSIZE,
1021 .bc_req_max_size = MGS_MAXREQSIZE,
1022 .bc_rep_max_size = MGS_MAXREPSIZE,
1023 .bc_req_portal = MGS_REQUEST_PORTAL,
1024 .bc_rep_portal = MGC_REPLY_PORTAL,
1027 .tc_thr_name = "ll_mgs",
1028 .tc_nthrs_init = MGS_NTHRS_INIT,
1029 .tc_nthrs_max = MGS_NTHRS_MAX,
1030 .tc_ctx_tags = LCT_MG_THREAD,
1033 .so_req_handler = tgt_request_handle,
1034 .so_req_printer = target_print_req,
1038 /* Start the service threads */
1039 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1040 if (IS_ERR(mgs->mgs_service)) {
1041 rc = PTR_ERR(mgs->mgs_service);
1042 CERROR("failed to start mgs service: %d\n", rc);
1043 mgs->mgs_service = NULL;
1044 GOTO(err_lproc, rc);
1047 ping_evictor_start();
1049 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1051 /* device stack is not yet fully setup to keep no objects behind */
1052 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1055 mgs_params_fsdb_cleanup(env, mgs);
1056 lproc_mgs_cleanup(mgs);
1058 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1060 ctxt->loc_dir = NULL;
1061 llog_cleanup(env, ctxt);
1064 tgt_fini(env, &mgs->mgs_lut);
1066 /* No extra cleanup needed for llog_init_commit_thread() */
1067 mgs_fs_cleanup(env, mgs);
1069 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1070 obd->obd_namespace = NULL;
1072 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1073 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1074 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1075 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1078 obd_disconnect(mgs->mgs_bottom_exp);
1081 server_put_mount(lustre_cfg_string(lcfg, 0));
1085 static struct lu_device *mgs_device_free(const struct lu_env *env,
1086 struct lu_device *lu)
1088 struct mgs_device *mgs = lu2mgs_dev(lu);
1091 dt_device_fini(&mgs->mgs_dt_dev);
1096 static int mgs_process_config(const struct lu_env *env,
1097 struct lu_device *dev,
1098 struct lustre_cfg *lcfg)
1104 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1105 const struct lu_object_conf *unused)
1107 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1108 struct lu_device *under;
1109 struct lu_object *below;
1113 /* do no set .do_ops as mgs calls to bottom osd directly */
1115 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1116 PFID(lu_object_fid(o)));
1118 under = &d->mgs_bottom->dd_lu_dev;
1119 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1121 lu_object_add(o, below);
1128 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1130 struct mgs_object *obj = lu2mgs_obj(o);
1131 struct lu_object_header *h = o->lo_header;
1133 dt_object_fini(&obj->mgo_obj);
1134 lu_object_header_fini(h);
1138 static int mgs_object_print(const struct lu_env *env, void *cookie,
1139 lu_printer_t p, const struct lu_object *l)
1141 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1143 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1146 struct lu_object_operations mgs_lu_obj_ops = {
1147 .loo_object_init = mgs_object_init,
1148 .loo_object_free = mgs_object_free,
1149 .loo_object_print = mgs_object_print,
1152 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1153 const struct lu_object_header *hdr,
1154 struct lu_device *d)
1156 struct lu_object_header *h;
1157 struct mgs_object *o;
1158 struct lu_object *l;
1160 LASSERT(hdr == NULL);
1164 l = &o->mgo_obj.do_lu;
1167 lu_object_header_init(h);
1168 dt_object_init(&o->mgo_obj, h, d);
1169 lu_object_add_top(h, l);
1171 l->lo_ops = &mgs_lu_obj_ops;
1179 const struct lu_device_operations mgs_lu_ops = {
1180 .ldo_object_alloc = mgs_object_alloc,
1181 .ldo_process_config = mgs_process_config,
1184 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1185 struct lu_device_type *type,
1186 struct lustre_cfg *lcfg)
1188 struct mgs_device *mgs;
1189 struct lu_device *ludev;
1193 ludev = ERR_PTR(-ENOMEM);
1197 ludev = mgs2lu_dev(mgs);
1198 dt_device_init(&mgs->mgs_dt_dev, type);
1199 rc = mgs_init0(env, mgs, type, lcfg);
1201 mgs_device_free(env, ludev);
1202 ludev = ERR_PTR(rc);
1208 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1209 struct lu_device *d)
1211 struct mgs_device *mgs = lu2mgs_dev(d);
1212 struct obd_device *obd = mgs->mgs_obd;
1213 struct llog_ctxt *ctxt;
1217 LASSERT(mgs->mgs_bottom);
1219 class_disconnect_exports(obd);
1221 ping_evictor_stop();
1223 ptlrpc_unregister_service(mgs->mgs_service);
1225 obd_exports_barrier(obd);
1226 obd_zombie_barrier();
1228 tgt_fini(env, &mgs->mgs_lut);
1229 mgs_params_fsdb_cleanup(env, mgs);
1230 mgs_cleanup_fsdb_list(mgs);
1231 lproc_mgs_cleanup(mgs);
1233 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1235 ctxt->loc_dir = NULL;
1236 llog_cleanup(env, ctxt);
1239 mgs_fs_cleanup(env, mgs);
1241 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1242 obd->obd_namespace = NULL;
1244 lu_site_purge(env, d->ld_site, ~0);
1245 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1246 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1247 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1250 LASSERT(mgs->mgs_bottom_exp);
1251 obd_disconnect(mgs->mgs_bottom_exp);
1253 server_put_mount(obd->obd_name);
1258 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1259 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1261 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1263 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1265 static struct lu_device_type_operations mgs_device_type_ops = {
1266 .ldto_init = mgs_type_init,
1267 .ldto_fini = mgs_type_fini,
1269 .ldto_start = mgs_type_start,
1270 .ldto_stop = mgs_type_stop,
1272 .ldto_device_alloc = mgs_device_alloc,
1273 .ldto_device_free = mgs_device_free,
1275 .ldto_device_fini = mgs_device_fini
1278 static struct lu_device_type mgs_device_type = {
1279 .ldt_tags = LU_DEVICE_DT,
1280 .ldt_name = LUSTRE_MGS_NAME,
1281 .ldt_ops = &mgs_device_type_ops,
1282 .ldt_ctx_tags = LCT_MG_THREAD
1285 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1286 struct obd_device *obd, struct obd_uuid *cluuid,
1287 struct obd_connect_data *data, void *localdata)
1289 struct obd_export *lexp;
1290 struct lustre_handle conn = { 0 };
1295 if (exp == NULL || obd == NULL || cluuid == NULL)
1298 rc = class_connect(&conn, obd, cluuid);
1302 lexp = class_conn2export(&conn);
1307 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1308 data->ocd_version = LUSTRE_VERSION_CODE;
1309 lexp->exp_connect_data = *data;
1312 tgt_counter_incr(lexp, LPROC_MGS_CONNECT);
1314 rc = mgs_export_stats_init(obd, lexp, localdata);
1316 class_disconnect(lexp);
1323 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1324 struct obd_device *obd, struct obd_uuid *cluuid,
1325 struct obd_connect_data *data, void *localdata)
1329 if (exp == NULL || obd == NULL || cluuid == NULL)
1332 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1335 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1336 data->ocd_version = LUSTRE_VERSION_CODE;
1337 exp->exp_connect_data = *data;
1340 RETURN(mgs_export_stats_init(obd, exp, localdata));
1343 static int mgs_obd_disconnect(struct obd_export *exp)
1351 mgs_fsc_cleanup(exp);
1353 class_export_get(exp);
1354 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1356 rc = server_disconnect_export(exp);
1357 class_export_put(exp);
1361 /* use obd ops to offer management infrastructure */
1362 static struct obd_ops mgs_obd_device_ops = {
1363 .o_owner = THIS_MODULE,
1364 .o_connect = mgs_obd_connect,
1365 .o_reconnect = mgs_obd_reconnect,
1366 .o_disconnect = mgs_obd_disconnect,
1367 .o_init_export = mgs_init_export,
1368 .o_destroy_export = mgs_destroy_export,
1369 .o_iocontrol = mgs_iocontrol,
1372 static int __init mgs_init(void)
1374 struct lprocfs_static_vars lvars;
1376 lprocfs_mgs_init_vars(&lvars);
1377 class_register_type(&mgs_obd_device_ops, NULL, lvars.module_vars,
1378 LUSTRE_MGS_NAME, &mgs_device_type);
1383 static void /*__exit*/ mgs_exit(void)
1385 class_unregister_type(LUSTRE_MGS_NAME);
1388 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1389 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1390 MODULE_LICENSE("GPL");
1392 module_init(mgs_init);
1393 module_exit(mgs_exit);