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) 2011, 2012, Whamcloud, Inc.
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"
52 /* Establish a connection to the MGS.*/
53 static int mgs_connect(const struct lu_env *env,
54 struct obd_export **exp, struct obd_device *obd,
55 struct obd_uuid *cluuid, struct obd_connect_data *data,
58 struct obd_export *lexp;
59 struct lustre_handle conn = { 0 };
63 if (!exp || !obd || !cluuid)
66 rc = class_connect(&conn, obd, cluuid);
70 lexp = class_conn2export(&conn);
72 GOTO(out, rc = -EFAULT);
74 mgs_counter_incr(lexp, LPROC_MGS_CONNECT);
77 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
78 lexp->exp_connect_flags = data->ocd_connect_flags;
79 data->ocd_version = LUSTRE_VERSION_CODE;
82 rc = mgs_export_stats_init(obd, lexp, localdata);
86 class_disconnect(lexp);
94 static int mgs_reconnect(const struct lu_env *env,
95 struct obd_export *exp, struct obd_device *obd,
96 struct obd_uuid *cluuid, struct obd_connect_data *data,
101 if (exp == NULL || obd == NULL || cluuid == NULL)
104 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
107 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
108 exp->exp_connect_flags = data->ocd_connect_flags;
109 data->ocd_version = LUSTRE_VERSION_CODE;
112 RETURN(mgs_export_stats_init(obd, exp, localdata));
115 static int mgs_disconnect(struct obd_export *exp)
122 mgs_fsc_cleanup(exp);
124 class_export_get(exp);
125 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
127 rc = server_disconnect_export(exp);
128 class_export_put(exp);
132 static int mgs_handle(struct ptlrpc_request *req);
134 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
139 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
140 LDLM_FL_BLOCK_CONV))) {
141 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
142 struct lustre_handle lockh;
144 /* clear the bit before lock put */
145 clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
147 ldlm_lock2handle(lock, &lockh);
148 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
151 RETURN(ldlm_completion_ast(lock, flags, cbdata));
154 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
159 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
160 LDLM_FL_BLOCK_CONV))) {
163 /* l_ast_data is used as a marker to avoid cancel ldlm lock
164 * twice. See LU-1259. */
165 lock_res_and_lock(lock);
166 fsdb = (struct fs_db *)lock->l_ast_data;
167 lock->l_ast_data = NULL;
168 unlock_res_and_lock(lock);
171 struct lustre_handle lockh;
173 mgs_ir_notify_complete(fsdb);
175 ldlm_lock2handle(lock, &lockh);
176 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
180 RETURN(ldlm_completion_ast(lock, flags, cbdata));
183 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
185 ldlm_completion_callback cp = NULL;
186 struct lustre_handle lockh = { 0 };
187 struct ldlm_res_id res_id;
188 __u64 flags = LDLM_FL_ATOMIC_CB;
192 LASSERT(fsdb->fsdb_name[0] != '\0');
193 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
197 case CONFIG_T_CONFIG:
198 cp = mgs_completion_ast_config;
199 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
202 case CONFIG_T_RECOVER:
203 cp = mgs_completion_ast_ir;
210 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
211 &res_id, LDLM_PLAIN, NULL, LCK_EX,
212 &flags, ldlm_blocking_ast, cp,
213 NULL, fsdb, 0, NULL, &lockh);
214 if (rc != ELDLM_OK) {
215 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
216 le64_to_cpu(res_id.name[0]),
217 le64_to_cpu(res_id.name[1]), rc);
219 if (type == CONFIG_T_CONFIG)
220 clear_bit(FSDB_REVOKING_LOCK,
223 /* lock has been cancelled in completion_ast. */
232 static int mgs_check_target(const struct lu_env *env,
233 struct mgs_device *mgs,
234 struct mgs_target_info *mti)
239 rc = mgs_check_index(env, mgs, mti);
241 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
242 "this MGS does not know about it, preventing "
243 "registration.\n", mti->mti_svname);
245 } else if (rc == -1) {
246 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
247 "disappeared! Regenerating all logs.\n",
249 mti->mti_flags |= LDD_F_WRITECONF;
252 /* Index is correctly marked as used */
254 /* If the logs don't contain the mti_nids then add
255 them as failover nids */
256 rc = mgs_check_failnid(env, mgs, mti);
262 /* Ensure this is not a failover node that is connecting first*/
263 static int mgs_check_failover_reg(struct mgs_target_info *mti)
269 ptr = mti->mti_params;
270 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
271 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
272 for (i = 0; i < mti->mti_nid_count; i++) {
273 if (nid == mti->mti_nids[i]) {
274 LCONSOLE_WARN("Denying initial registra"
275 "tion attempt from nid %s"
276 ", specified as failover"
277 "\n",libcfs_nid2str(nid));
278 return -EADDRNOTAVAIL;
286 /* Called whenever a target starts up. Flags indicate first connect, etc. */
287 static int mgs_handle_target_reg(struct ptlrpc_request *req)
289 struct obd_device *obd = req->rq_export->exp_obd;
290 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
291 struct lu_env *env = req->rq_svc_thread->t_env;
292 struct mgs_target_info *mti, *rep_mti;
298 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
300 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
302 opc = mti->mti_flags & LDD_F_OPC_MASK;
303 if (opc == LDD_F_OPC_READY) {
304 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
305 mti->mti_fsname, mti->mti_stripe_index);
306 rc = mgs_ir_update(env, mgs, mti);
308 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
309 CERROR("Update IR return with %d(ignore and IR "
312 GOTO(out_nolock, rc);
315 /* Do not support unregistering right now. */
316 if (opc != LDD_F_OPC_REG)
317 GOTO(out_nolock, rc = -EINVAL);
319 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
320 mti->mti_fsname, mti->mti_stripe_index);
322 if (mti->mti_flags & LDD_F_NEED_INDEX)
323 mti->mti_flags |= LDD_F_WRITECONF;
325 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
327 /* We're just here as a startup ping. */
328 CDEBUG(D_MGS, "Server %s is running on %s\n",
329 mti->mti_svname, obd_export_nid2str(req->rq_export));
330 rc = mgs_check_target(env, mgs, mti);
331 /* above will set appropriate mti flags */
333 /* Nothing wrong, or fatal error */
334 GOTO(out_nolock, rc);
335 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
336 rc = mgs_check_failover_reg(mti);
338 GOTO(out_nolock, rc);
341 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
343 if (mti->mti_flags & LDD_F_WRITECONF) {
344 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
345 mti->mti_stripe_index == 0) {
346 rc = mgs_erase_logs(env, mgs, mti->mti_fsname);
347 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
348 "request. All servers must be restarted "
349 "in order to regenerate the logs."
350 "\n", obd->obd_name, mti->mti_fsname);
351 } else if (mti->mti_flags &
352 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
353 rc = mgs_erase_log(env, mgs, mti->mti_svname);
354 LCONSOLE_WARN("%s: Regenerating %s log by user "
356 obd->obd_name, mti->mti_svname);
358 mti->mti_flags |= LDD_F_UPDATE;
359 /* Erased logs means start from scratch. */
360 mti->mti_flags &= ~LDD_F_UPGRADE14;
362 GOTO(out_nolock, rc);
365 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
367 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
368 GOTO(out_nolock, rc);
372 * Log writing contention is handled by the fsdb_mutex.
374 * It should be alright if someone was reading while we were
375 * updating the logs - if we revoke at the end they will just update
376 * from where they left off.
379 if (mti->mti_flags & LDD_F_UPGRADE14) {
380 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
384 if (mti->mti_flags & LDD_F_UPDATE) {
385 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
386 mti->mti_stripe_index);
388 /* create or update the target log
389 and update the client/mdt logs */
390 rc = mgs_write_log_target(env, mgs, mti, fsdb);
392 CERROR("Failed to write %s log (%d)\n",
393 mti->mti_svname, rc);
397 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
398 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
400 mti->mti_flags |= LDD_F_REWRITE_LDD;
404 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
407 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
408 mti->mti_stripe_index, rc);
411 /* we need an error flag to tell the target what's going on,
412 * instead of just doing it by error code only. */
413 mti->mti_flags |= LDD_F_ERROR;
415 rc = req_capsule_server_pack(&req->rq_pill);
419 /* send back the whole mti in the reply */
420 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
423 /* Flush logs to disk */
424 dt_sync(req->rq_svc_thread->t_env, mgs->mgs_bottom);
428 static int mgs_set_info_rpc(struct ptlrpc_request *req)
430 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
431 struct lu_env *env = req->rq_svc_thread->t_env;
432 struct mgs_send_param *msp, *rep_msp;
433 struct mgs_thread_info *mgi = mgs_env_info(env);
435 struct lustre_cfg *lcfg;
438 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
441 /* Construct lustre_cfg structure to pass to function mgs_setparam */
442 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
443 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
444 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
446 GOTO(out, rc = PTR_ERR(lcfg));
447 rc = mgs_setparam(env, mgs, lcfg, mgi->mgi_fsname);
449 CERROR("Error %d in setting the parameter %s for fs %s\n",
450 rc, msp->mgs_param, mgi->mgi_fsname);
454 rc = req_capsule_server_pack(&req->rq_pill);
456 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
460 lustre_cfg_free(lcfg);
465 static int mgs_config_read(struct ptlrpc_request *req)
467 struct mgs_config_body *body;
471 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
475 switch (body->mcb_type) {
476 case CONFIG_T_RECOVER:
477 rc = mgs_get_ir_logs(req);
480 case CONFIG_T_CONFIG:
493 * similar as in ost_connect_check_sptlrpc()
495 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
497 struct obd_export *exp = req->rq_export;
498 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
499 struct lu_env *env = req->rq_svc_thread->t_env;
501 struct sptlrpc_flavor flvr;
504 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
505 rc = mgs_find_or_make_fsdb(env, mgs, MGSSELF_NAME, &fsdb);
509 mutex_lock(&fsdb->fsdb_mutex);
510 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
511 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
514 /* by defualt allow any flavors */
515 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
517 mutex_unlock(&fsdb->fsdb_mutex);
519 spin_lock(&exp->exp_lock);
521 exp->exp_sp_peer = req->rq_sp_from;
522 exp->exp_flvr = flvr;
524 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
525 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
526 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
527 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
528 libcfs_nid2str(req->rq_peer.nid));
532 spin_unlock(&exp->exp_lock);
534 if (exp->exp_sp_peer != req->rq_sp_from) {
535 CERROR("RPC source %s doesn't match %s\n",
536 sptlrpc_part2name(req->rq_sp_from),
537 sptlrpc_part2name(exp->exp_sp_peer));
540 rc = sptlrpc_target_export_check(exp, req);
547 /* Called whenever a target cleans up. */
548 /* XXX - Currently unused */
549 static int mgs_handle_target_del(struct ptlrpc_request *req)
552 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
556 /* XXX - Currently unused */
557 static int mgs_handle_exception(struct ptlrpc_request *req)
560 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
565 * For old clients there is no direct way of knowing which filesystems
566 * a client is operating at the MGS side. But we need to pick up those
567 * clients so that the MGS can mark the corresponding filesystem as
568 * non-IR capable because old clients are not ready to be notified.
570 * This is why we have this _hack_ function. We detect the filesystem's
571 * name by hacking llog operation which is currently used by the clients
572 * to fetch configuration logs. At present this is fine because this is
573 * the ONLY llog operation between mgc and the MGS.
575 * If extra llog operation is going to be added, this function needs fixing.
577 * If releases prior than 2.0 are not supported, we can remove this function.
579 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
586 /* XXX: We suppose that llog at mgs is only used for
587 * fetching file system log */
588 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
589 if (logname == NULL) {
590 CERROR("No logname, is llog on MGS used for something else?\n");
594 ptr = strchr(logname, '-');
595 rc = (int)(ptr - logname);
596 if (ptr == NULL || rc >= sizeof(fsname)) {
597 CERROR("Invalid logname received: %s\n", logname);
601 strncpy(fsname, logname, rc);
603 rc = mgs_fsc_attach(req->rq_svc_thread->t_env, req->rq_export, fsname);
604 if (rc < 0 && rc != -EEXIST)
605 CERROR("add fs client %s returns %d\n", fsname, rc);
610 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
611 int mgs_handle(struct ptlrpc_request *req)
613 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
617 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
618 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
619 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
622 LASSERT(current->journal_info == NULL);
623 opc = lustre_msg_get_opc(req->rq_reqmsg);
625 if (opc == SEC_CTX_INIT ||
626 opc == SEC_CTX_INIT_CONT ||
630 if (opc != MGS_CONNECT) {
631 if (!class_connected_export(req->rq_export)) {
632 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
633 req->rq_status = -ENOTCONN;
634 GOTO(out, rc = -ENOTCONN);
640 DEBUG_REQ(D_MGS, req, "connect");
641 /* MGS and MDS have same request format for connect */
642 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
643 rc = target_handle_connect(req);
645 rc = mgs_connect_check_sptlrpc(req);
647 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
648 /* Make clients trying to reconnect after a MGS restart
649 happy; also requires obd_replayable */
650 lustre_msg_add_op_flags(req->rq_repmsg,
651 MSG_CONNECT_RECONNECT);
654 DEBUG_REQ(D_MGS, req, "disconnect");
655 /* MGS and MDS have same request format for disconnect */
656 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
657 rc = target_handle_disconnect(req);
658 req->rq_status = rc; /* superfluous? */
661 DEBUG_REQ(D_MGS, req, "exception");
662 rc = mgs_handle_exception(req);
665 DEBUG_REQ(D_MGS, req, "target add");
666 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
667 rc = mgs_handle_target_reg(req);
670 DEBUG_REQ(D_MGS, req, "target del");
671 rc = mgs_handle_target_del(req);
674 DEBUG_REQ(D_MGS, req, "set_info");
675 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
676 rc = mgs_set_info_rpc(req);
678 case MGS_CONFIG_READ:
679 DEBUG_REQ(D_MGS, req, "read config");
680 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
681 rc = mgs_config_read(req);
684 DEBUG_REQ(D_MGS, req, "enqueue");
685 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
686 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
687 ldlm_server_blocking_ast, NULL);
689 case LDLM_BL_CALLBACK:
690 case LDLM_CP_CALLBACK:
691 DEBUG_REQ(D_MGS, req, "callback");
692 CERROR("callbacks should not happen on MGS\n");
697 DEBUG_REQ(D_INFO, req, "ping");
698 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
699 rc = target_handle_ping(req);
702 DEBUG_REQ(D_MGS, req, "log cancel");
703 rc = -ENOTSUPP; /* la la la */
706 case LLOG_ORIGIN_HANDLE_CREATE:
707 DEBUG_REQ(D_MGS, req, "llog_open");
708 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
709 rc = llog_origin_handle_open(req);
711 (void)mgs_handle_fslog_hack(req);
713 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
714 DEBUG_REQ(D_MGS, req, "llog next block");
715 req_capsule_set(&req->rq_pill,
716 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
717 rc = llog_origin_handle_next_block(req);
719 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
720 DEBUG_REQ(D_MGS, req, "llog prev block");
721 req_capsule_set(&req->rq_pill,
722 &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
723 rc = llog_origin_handle_prev_block(req);
725 case LLOG_ORIGIN_HANDLE_READ_HEADER:
726 DEBUG_REQ(D_MGS, req, "llog read header");
727 req_capsule_set(&req->rq_pill,
728 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
729 rc = llog_origin_handle_read_header(req);
731 case LLOG_ORIGIN_HANDLE_CLOSE:
732 DEBUG_REQ(D_MGS, req, "llog close");
733 rc = llog_origin_handle_close(req);
739 LASSERT(current->journal_info == NULL);
741 DEBUG_REQ(D_MGS, req, "MGS fail to handle opc = %d: rc = %d\n",
744 rc = ptlrpc_error(req);
748 target_send_reply(req, rc, fail);
752 static inline int mgs_init_export(struct obd_export *exp)
754 struct mgs_export_data *data = &exp->u.eu_mgs_data;
756 /* init mgs_export_data for fsc */
757 spin_lock_init(&data->med_lock);
758 CFS_INIT_LIST_HEAD(&data->med_clients);
760 spin_lock(&exp->exp_lock);
761 exp->exp_connecting = 1;
762 spin_unlock(&exp->exp_lock);
764 /* self-export doesn't need client data and ldlm initialization */
765 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
766 &exp->exp_client_uuid)))
768 return ldlm_init_export(exp);
771 static inline int mgs_destroy_export(struct obd_export *exp)
775 target_destroy_export(exp);
776 mgs_client_free(exp);
778 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
779 &exp->exp_client_uuid)))
782 ldlm_destroy_export(exp);
787 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
792 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
800 strcpy(poolname, ptr);
805 static int mgs_iocontrol_pool(const struct lu_env *env,
806 struct mgs_device *mgs,
807 struct obd_ioctl_data *data)
809 struct mgs_thread_info *mgi = mgs_env_info(env);
811 struct lustre_cfg *lcfg = NULL;
812 char *poolname = NULL;
815 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
816 if (poolname == NULL)
819 if (data->ioc_type != LUSTRE_CFG_TYPE) {
820 CERROR("%s: unknown cfg record type: %d\n",
821 mgs->mgs_obd->obd_name, data->ioc_type);
822 GOTO(out_pool, rc = -EINVAL);
825 if (data->ioc_plen1 > CFS_PAGE_SIZE)
826 GOTO(out_pool, rc = -E2BIG);
828 OBD_ALLOC(lcfg, data->ioc_plen1);
830 GOTO(out_pool, rc = -ENOMEM);
832 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
833 GOTO(out_lcfg, rc = -EFAULT);
835 if (lcfg->lcfg_bufcount < 2)
836 GOTO(out_lcfg, rc = -EFAULT);
838 /* first arg is always <fsname>.<poolname> */
839 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
844 switch (lcfg->lcfg_command) {
846 if (lcfg->lcfg_bufcount != 2)
847 GOTO(out_lcfg, rc = -EINVAL);
848 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
852 if (lcfg->lcfg_bufcount != 3)
853 GOTO(out_lcfg, rc = -EINVAL);
854 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
855 poolname, lustre_cfg_string(lcfg, 2));
858 if (lcfg->lcfg_bufcount != 3)
859 GOTO(out_lcfg, rc = -EINVAL);
860 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
861 poolname, lustre_cfg_string(lcfg, 2));
864 if (lcfg->lcfg_bufcount != 2)
865 GOTO(out_lcfg, rc = -EINVAL);
866 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
874 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
875 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
880 OBD_FREE(lcfg, data->ioc_plen1);
882 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
886 /* from mdt_iocontrol */
887 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
888 void *karg, void *uarg)
890 struct mgs_device *mgs = exp2mgs_dev(exp);
891 struct obd_ioctl_data *data = karg;
896 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
898 rc = lu_env_init(&env, LCT_MG_THREAD);
904 case OBD_IOC_PARAM: {
905 struct mgs_thread_info *mgi = mgs_env_info(&env);
906 struct lustre_cfg *lcfg;
908 if (data->ioc_type != LUSTRE_CFG_TYPE) {
909 CERROR("%s: unknown cfg record type: %d\n",
910 mgs->mgs_obd->obd_name, data->ioc_type);
911 GOTO(out, rc = -EINVAL);
914 OBD_ALLOC(lcfg, data->ioc_plen1);
916 GOTO(out, rc = -ENOMEM);
917 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
918 GOTO(out_free, rc = -EFAULT);
920 if (lcfg->lcfg_bufcount < 1)
921 GOTO(out_free, rc = -EINVAL);
923 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
925 CERROR("%s: setparam err: rc = %d\n",
926 exp->exp_obd->obd_name, rc);
928 OBD_FREE(lcfg, data->ioc_plen1);
933 rc = mgs_iocontrol_pool(&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_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);
1006 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1007 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1009 static struct ptlrpc_service_conf conf;
1010 struct lprocfs_static_vars lvars = { 0 };
1011 struct obd_device *obd;
1012 struct lustre_mount_info *lmi;
1013 struct llog_ctxt *ctxt;
1018 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1022 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1024 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1028 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1031 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1033 obd->u.obt.obt_magic = OBT_MAGIC;
1034 obd->u.obt.obt_instance = 0;
1036 /* namespace for mgs llog */
1037 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1038 LDLM_NAMESPACE_SERVER,
1039 LDLM_NAMESPACE_MODEST,
1041 if (obd->obd_namespace == NULL)
1042 GOTO(err_ops, rc = -ENOMEM);
1045 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1046 "mgs_ldlm_client", &obd->obd_ldlm_client);
1048 rc = mgs_fs_setup(env, mgs);
1050 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1055 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1056 obd, &llog_osd_ops);
1060 /* XXX: we need this trick till N:1 stack is supported
1061 * set "current" directory for named llogs */
1062 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1064 ctxt->loc_dir = mgs->mgs_configs_dir;
1065 llog_ctxt_put(ctxt);
1067 /* No recovery for MGC's */
1068 obd->obd_replayable = 0;
1070 /* Internal mgs setup */
1071 mgs_init_fsdb_list(mgs);
1072 mutex_init(&mgs->mgs_mutex);
1073 mgs->mgs_start_time = cfs_time_current_sec();
1074 spin_lock_init(&mgs->mgs_lock);
1077 lprocfs_mgs_init_vars(&lvars);
1078 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
1079 lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1080 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
1085 conf = (typeof(conf)) {
1086 .psc_name = LUSTRE_MGS_NAME,
1087 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1089 .bc_nbufs = MGS_NBUFS,
1090 .bc_buf_size = MGS_BUFSIZE,
1091 .bc_req_max_size = MGS_MAXREQSIZE,
1092 .bc_rep_max_size = MGS_MAXREPSIZE,
1093 .bc_req_portal = MGS_REQUEST_PORTAL,
1094 .bc_rep_portal = MGC_REPLY_PORTAL,
1097 .tc_thr_name = "ll_mgs",
1098 .tc_nthrs_init = MGS_NTHRS_INIT,
1099 .tc_nthrs_max = MGS_NTHRS_MAX,
1100 .tc_ctx_tags = LCT_MG_THREAD,
1103 .so_req_handler = mgs_handle,
1104 .so_req_printer = target_print_req,
1107 /* Start the service threads */
1108 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1109 if (IS_ERR(mgs->mgs_service)) {
1110 rc = PTR_ERR(mgs->mgs_service);
1111 CERROR("failed to start service: %d\n", rc);
1112 GOTO(err_lproc, rc);
1115 ping_evictor_start();
1117 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1119 /* device stack is not yet fully setup to keep no objects behind */
1120 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1124 lproc_mgs_cleanup(mgs);
1126 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1128 ctxt->loc_dir = NULL;
1129 llog_cleanup(env, ctxt);
1132 /* No extra cleanup needed for llog_init_commit_thread() */
1133 mgs_fs_cleanup(env, mgs);
1135 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1136 obd->obd_namespace = NULL;
1138 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1139 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1140 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1141 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1144 obd_disconnect(mgs->mgs_bottom_exp);
1148 static struct lu_device *mgs_device_free(const struct lu_env *env,
1149 struct lu_device *lu)
1151 struct mgs_device *mgs = lu2mgs_dev(lu);
1154 dt_device_fini(&mgs->mgs_dt_dev);
1159 static int mgs_process_config(const struct lu_env *env,
1160 struct lu_device *dev,
1161 struct lustre_cfg *lcfg)
1167 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1168 const struct lu_object_conf *unused)
1170 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1171 struct lu_device *under;
1172 struct lu_object *below;
1176 /* do no set .do_ops as mgs calls to bottom osd directly */
1178 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1179 PFID(lu_object_fid(o)));
1181 under = &d->mgs_bottom->dd_lu_dev;
1182 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1184 lu_object_add(o, below);
1191 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1193 struct mgs_object *obj = lu2mgs_obj(o);
1194 struct lu_object_header *h = o->lo_header;
1196 dt_object_fini(&obj->mgo_obj);
1197 lu_object_header_fini(h);
1201 static int mgs_object_print(const struct lu_env *env, void *cookie,
1202 lu_printer_t p, const struct lu_object *l)
1204 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1206 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1209 struct lu_object_operations mgs_lu_obj_ops = {
1210 .loo_object_init = mgs_object_init,
1211 .loo_object_free = mgs_object_free,
1212 .loo_object_print = mgs_object_print,
1215 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1216 const struct lu_object_header *hdr,
1217 struct lu_device *d)
1219 struct lu_object_header *h;
1220 struct mgs_object *o;
1221 struct lu_object *l;
1223 LASSERT(hdr == NULL);
1227 l = &o->mgo_obj.do_lu;
1230 lu_object_header_init(h);
1231 dt_object_init(&o->mgo_obj, h, d);
1232 lu_object_add_top(h, l);
1234 l->lo_ops = &mgs_lu_obj_ops;
1242 const struct lu_device_operations mgs_lu_ops = {
1243 .ldo_object_alloc = mgs_object_alloc,
1244 .ldo_process_config = mgs_process_config,
1247 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1248 struct lu_device_type *type,
1249 struct lustre_cfg *lcfg)
1251 struct mgs_device *mgs;
1252 struct lu_device *ludev;
1256 ludev = ERR_PTR(-ENOMEM);
1260 ludev = mgs2lu_dev(mgs);
1261 dt_device_init(&mgs->mgs_dt_dev, type);
1262 rc = mgs_init0(env, mgs, type, lcfg);
1264 mgs_device_free(env, ludev);
1265 ludev = ERR_PTR(rc);
1271 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1272 struct lu_device *d)
1274 struct mgs_device *mgs = lu2mgs_dev(d);
1275 struct obd_device *obd = mgs->mgs_obd;
1276 struct llog_ctxt *ctxt;
1280 LASSERT(mgs->mgs_bottom);
1282 ping_evictor_stop();
1284 ptlrpc_unregister_service(mgs->mgs_service);
1286 obd_exports_barrier(obd);
1287 obd_zombie_barrier();
1289 mgs_cleanup_fsdb_list(mgs);
1290 lproc_mgs_cleanup(mgs);
1292 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1294 ctxt->loc_dir = NULL;
1295 llog_cleanup(env, ctxt);
1298 mgs_fs_cleanup(env, mgs);
1300 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1301 obd->obd_namespace = NULL;
1303 lu_site_purge(env, d->ld_site, ~0);
1304 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1305 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1306 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1309 LASSERT(mgs->mgs_bottom_exp);
1310 obd_disconnect(mgs->mgs_bottom_exp);
1312 server_put_mount(obd->obd_name, NULL);
1317 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1318 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1320 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1322 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1324 static struct lu_device_type_operations mgs_device_type_ops = {
1325 .ldto_init = mgs_type_init,
1326 .ldto_fini = mgs_type_fini,
1328 .ldto_start = mgs_type_start,
1329 .ldto_stop = mgs_type_stop,
1331 .ldto_device_alloc = mgs_device_alloc,
1332 .ldto_device_free = mgs_device_free,
1334 .ldto_device_fini = mgs_device_fini
1337 static struct lu_device_type mgs_device_type = {
1338 .ldt_tags = LU_DEVICE_DT,
1339 .ldt_name = LUSTRE_MGS_NAME,
1340 .ldt_ops = &mgs_device_type_ops,
1341 .ldt_ctx_tags = LCT_MG_THREAD
1345 /* use obd ops to offer management infrastructure */
1346 static struct obd_ops mgs_obd_ops = {
1347 .o_owner = THIS_MODULE,
1348 .o_connect = mgs_connect,
1349 .o_reconnect = mgs_reconnect,
1350 .o_disconnect = mgs_disconnect,
1351 .o_init_export = mgs_init_export,
1352 .o_destroy_export = mgs_destroy_export,
1353 .o_iocontrol = mgs_iocontrol,
1356 static int __init mgs_init(void)
1358 struct lprocfs_static_vars lvars;
1360 lprocfs_mgs_init_vars(&lvars);
1361 class_register_type(&mgs_obd_ops, NULL, lvars.module_vars,
1362 LUSTRE_MGS_NAME, &mgs_device_type);
1367 static void /*__exit*/ mgs_exit(void)
1369 class_unregister_type(LUSTRE_MGS_NAME);
1372 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1373 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1374 MODULE_LICENSE("GPL");
1376 module_init(mgs_init);
1377 module_exit(mgs_exit);