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
47 # include <linux/module.h>
48 # include <linux/pagemap.h>
49 # include <linux/miscdevice.h>
50 # include <linux/init.h>
52 # include <liblustre.h>
55 #include <obd_class.h>
56 #include <lustre_dlm.h>
57 #include <lprocfs_status.h>
58 #include <lustre_fsfilt.h>
59 #include <lustre_disk.h>
60 #include <lustre_param.h>
61 #include <lustre_log.h>
63 #include "mgs_internal.h"
65 /* Establish a connection to the MGS.*/
66 static int mgs_connect(const struct lu_env *env,
67 struct obd_export **exp, struct obd_device *obd,
68 struct obd_uuid *cluuid, struct obd_connect_data *data,
71 struct obd_export *lexp;
72 struct lustre_handle conn = { 0 };
76 if (!exp || !obd || !cluuid)
79 rc = class_connect(&conn, obd, cluuid);
83 lexp = class_conn2export(&conn);
85 GOTO(out, rc = -EFAULT);
87 mgs_counter_incr(lexp, LPROC_MGS_CONNECT);
90 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
91 lexp->exp_connect_flags = data->ocd_connect_flags;
92 data->ocd_version = LUSTRE_VERSION_CODE;
95 rc = mgs_export_stats_init(obd, lexp, localdata);
99 class_disconnect(lexp);
107 static int mgs_reconnect(const struct lu_env *env,
108 struct obd_export *exp, struct obd_device *obd,
109 struct obd_uuid *cluuid, struct obd_connect_data *data,
114 if (exp == NULL || obd == NULL || cluuid == NULL)
117 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
120 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
121 exp->exp_connect_flags = data->ocd_connect_flags;
122 data->ocd_version = LUSTRE_VERSION_CODE;
125 RETURN(mgs_export_stats_init(obd, exp, localdata));
128 static int mgs_disconnect(struct obd_export *exp)
135 mgs_fsc_cleanup(exp);
137 class_export_get(exp);
138 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
140 rc = server_disconnect_export(exp);
141 class_export_put(exp);
145 static int mgs_handle(struct ptlrpc_request *req);
147 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
148 struct obd_device *tgt, int *index)
153 LASSERT(olg == &obd->obd_olg);
154 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
159 static int mgs_llog_finish(struct obd_device *obd, int count)
161 struct llog_ctxt *ctxt;
165 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
167 rc = llog_cleanup(ctxt);
172 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
177 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
178 LDLM_FL_BLOCK_CONV))) {
179 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
180 struct lustre_handle lockh;
182 /* clear the bit before lock put */
183 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
185 ldlm_lock2handle(lock, &lockh);
186 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
189 RETURN(ldlm_completion_ast(lock, flags, cbdata));
192 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
197 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
198 LDLM_FL_BLOCK_CONV))) {
201 /* l_ast_data is used as a marker to avoid cancel ldlm lock
202 * twice. See LU-1259. */
203 lock_res_and_lock(lock);
204 fsdb = (struct fs_db *)lock->l_ast_data;
205 lock->l_ast_data = NULL;
206 unlock_res_and_lock(lock);
209 struct lustre_handle lockh;
211 mgs_ir_notify_complete(fsdb);
213 ldlm_lock2handle(lock, &lockh);
214 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
218 RETURN(ldlm_completion_ast(lock, flags, cbdata));
221 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
223 ldlm_completion_callback cp = NULL;
224 struct lustre_handle lockh = { 0 };
225 struct ldlm_res_id res_id;
226 int flags = LDLM_FL_ATOMIC_CB;
230 LASSERT(fsdb->fsdb_name[0] != '\0');
231 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
235 case CONFIG_T_CONFIG:
236 cp = mgs_completion_ast_config;
237 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
240 case CONFIG_T_RECOVER:
241 cp = mgs_completion_ast_ir;
248 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
249 &res_id, LDLM_PLAIN, NULL, LCK_EX,
250 &flags, ldlm_blocking_ast, cp,
251 NULL, fsdb, 0, NULL, &lockh);
252 if (rc != ELDLM_OK) {
253 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
254 le64_to_cpu(res_id.name[0]),
255 le64_to_cpu(res_id.name[1]), rc);
257 if (type == CONFIG_T_CONFIG)
258 cfs_clear_bit(FSDB_REVOKING_LOCK,
261 /* lock has been cancelled in completion_ast. */
270 static int mgs_check_target(const struct lu_env *env,
271 struct mgs_device *mgs,
272 struct mgs_target_info *mti)
277 rc = mgs_check_index(env, mgs, mti);
279 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
280 "this MGS does not know about it, preventing "
281 "registration.\n", mti->mti_svname);
283 } else if (rc == -1) {
284 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
285 "disappeared! Regenerating all logs.\n",
287 mti->mti_flags |= LDD_F_WRITECONF;
290 /* Index is correctly marked as used */
292 /* If the logs don't contain the mti_nids then add
293 them as failover nids */
294 rc = mgs_check_failnid(env, mgs, mti);
300 /* Ensure this is not a failover node that is connecting first*/
301 static int mgs_check_failover_reg(struct mgs_target_info *mti)
307 ptr = mti->mti_params;
308 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
309 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
310 for (i = 0; i < mti->mti_nid_count; i++) {
311 if (nid == mti->mti_nids[i]) {
312 LCONSOLE_WARN("Denying initial registra"
313 "tion attempt from nid %s"
314 ", specified as failover"
315 "\n",libcfs_nid2str(nid));
316 return -EADDRNOTAVAIL;
324 /* Called whenever a target starts up. Flags indicate first connect, etc. */
325 static int mgs_handle_target_reg(struct ptlrpc_request *req)
327 struct obd_device *obd = req->rq_export->exp_obd;
328 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
329 struct lu_env *env = req->rq_svc_thread->t_env;
330 struct mgs_target_info *mti, *rep_mti;
336 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
338 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
340 opc = mti->mti_flags & LDD_F_OPC_MASK;
341 if (opc == LDD_F_OPC_READY) {
342 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
343 mti->mti_fsname, mti->mti_stripe_index);
344 rc = mgs_ir_update(env, mgs, mti);
346 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
347 CERROR("Update IR return with %d(ignore and IR "
350 GOTO(out_nolock, rc);
353 /* Do not support unregistering right now. */
354 if (opc != LDD_F_OPC_REG)
355 GOTO(out_nolock, rc = -EINVAL);
357 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
358 mti->mti_fsname, mti->mti_stripe_index);
360 if (mti->mti_flags & LDD_F_NEED_INDEX)
361 mti->mti_flags |= LDD_F_WRITECONF;
363 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
365 /* We're just here as a startup ping. */
366 CDEBUG(D_MGS, "Server %s is running on %s\n",
367 mti->mti_svname, obd_export_nid2str(req->rq_export));
368 rc = mgs_check_target(env, mgs, mti);
369 /* above will set appropriate mti flags */
371 /* Nothing wrong, or fatal error */
372 GOTO(out_nolock, rc);
374 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
375 && (rc = mgs_check_failover_reg(mti)))
376 GOTO(out_nolock, rc);
379 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
381 if (mti->mti_flags & LDD_F_WRITECONF) {
382 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
383 mti->mti_stripe_index == 0) {
384 rc = mgs_erase_logs(env, mgs, mti->mti_fsname);
385 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
386 "request. All servers must be restarted "
387 "in order to regenerate the logs."
388 "\n", obd->obd_name, mti->mti_fsname);
389 } else if (mti->mti_flags &
390 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
391 rc = mgs_erase_log(env, mgs, mti->mti_svname);
392 LCONSOLE_WARN("%s: Regenerating %s log by user "
394 obd->obd_name, mti->mti_svname);
396 mti->mti_flags |= LDD_F_UPDATE;
397 /* Erased logs means start from scratch. */
398 mti->mti_flags &= ~LDD_F_UPGRADE14;
400 GOTO(out_nolock, rc);
403 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
405 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
406 GOTO(out_nolock, rc);
410 * Log writing contention is handled by the fsdb_mutex.
412 * It should be alright if someone was reading while we were
413 * updating the logs - if we revoke at the end they will just update
414 * from where they left off.
417 if (mti->mti_flags & LDD_F_UPGRADE14) {
418 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
422 if (mti->mti_flags & LDD_F_UPDATE) {
423 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
424 mti->mti_stripe_index);
426 /* create or update the target log
427 and update the client/mdt logs */
428 rc = mgs_write_log_target(env, mgs, mti, fsdb);
430 CERROR("Failed to write %s log (%d)\n",
431 mti->mti_svname, rc);
435 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
436 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
438 mti->mti_flags |= LDD_F_REWRITE_LDD;
442 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
445 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
446 mti->mti_stripe_index, rc);
449 /* we need an error flag to tell the target what's going on,
450 * instead of just doing it by error code only. */
451 mti->mti_flags |= LDD_F_ERROR;
453 rc = req_capsule_server_pack(&req->rq_pill);
457 /* send back the whole mti in the reply */
458 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
461 /* Flush logs to disk */
462 dt_sync(req->rq_svc_thread->t_env, mgs->mgs_bottom);
466 static int mgs_set_info_rpc(struct ptlrpc_request *req)
468 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
469 struct lu_env *env = req->rq_svc_thread->t_env;
470 struct mgs_send_param *msp, *rep_msp;
471 struct mgs_thread_info *mgi = mgs_env_info(env);
473 struct lustre_cfg *lcfg;
476 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
479 /* Construct lustre_cfg structure to pass to function mgs_setparam */
480 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
481 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
482 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
484 GOTO(out, rc = PTR_ERR(lcfg));
485 rc = mgs_setparam(env, mgs, lcfg, mgi->mgi_fsname);
487 CERROR("Error %d in setting the parameter %s for fs %s\n",
488 rc, msp->mgs_param, mgi->mgi_fsname);
492 rc = req_capsule_server_pack(&req->rq_pill);
494 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
498 lustre_cfg_free(lcfg);
503 static int mgs_config_read(struct ptlrpc_request *req)
505 struct mgs_config_body *body;
509 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
513 switch (body->mcb_type) {
514 case CONFIG_T_RECOVER:
515 rc = mgs_get_ir_logs(req);
518 case CONFIG_T_CONFIG:
531 * similar as in ost_connect_check_sptlrpc()
533 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
535 struct obd_export *exp = req->rq_export;
536 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
537 struct lu_env *env = req->rq_svc_thread->t_env;
539 struct sptlrpc_flavor flvr;
542 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
543 rc = mgs_find_or_make_fsdb(env, mgs, MGSSELF_NAME, &fsdb);
547 cfs_mutex_lock(&fsdb->fsdb_mutex);
548 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
549 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
552 /* by defualt allow any flavors */
553 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
555 cfs_mutex_unlock(&fsdb->fsdb_mutex);
557 cfs_spin_lock(&exp->exp_lock);
559 exp->exp_sp_peer = req->rq_sp_from;
560 exp->exp_flvr = flvr;
562 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
563 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
564 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
565 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
566 libcfs_nid2str(req->rq_peer.nid));
570 cfs_spin_unlock(&exp->exp_lock);
572 if (exp->exp_sp_peer != req->rq_sp_from) {
573 CERROR("RPC source %s doesn't match %s\n",
574 sptlrpc_part2name(req->rq_sp_from),
575 sptlrpc_part2name(exp->exp_sp_peer));
578 rc = sptlrpc_target_export_check(exp, req);
585 /* Called whenever a target cleans up. */
586 /* XXX - Currently unused */
587 static int mgs_handle_target_del(struct ptlrpc_request *req)
590 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
594 /* XXX - Currently unused */
595 static int mgs_handle_exception(struct ptlrpc_request *req)
598 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
603 * For old clients there is no direct way of knowing which filesystems
604 * a client is operating at the MGS side. But we need to pick up those
605 * clients so that the MGS can mark the corresponding filesystem as
606 * non-IR capable because old clients are not ready to be notified.
608 * This is why we have this _hack_ function. We detect the filesystem's
609 * name by hacking llog operation which is currently used by the clients
610 * to fetch configuration logs. At present this is fine because this is
611 * the ONLY llog operation between mgc and the MGS.
613 * If extra llog operation is going to be added, this function needs fixing.
615 * If releases prior than 2.0 are not supported, we can remove this function.
617 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
624 /* XXX: We suppose that llog at mgs is only used for
625 * fetching file system log */
626 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
627 if (logname == NULL) {
628 CERROR("No logname, is llog on MGS used for something else?\n");
632 ptr = strchr(logname, '-');
633 rc = (int)(ptr - logname);
634 if (ptr == NULL || rc >= sizeof(fsname)) {
635 CERROR("Invalid logname received: %s\n", logname);
639 strncpy(fsname, logname, rc);
641 rc = mgs_fsc_attach(req->rq_svc_thread->t_env, req->rq_export, fsname);
642 if (rc < 0 && rc != -EEXIST)
643 CERROR("add fs client %s returns %d\n", fsname, rc);
648 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
649 int mgs_handle(struct ptlrpc_request *req)
651 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
655 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
656 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
657 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
660 LASSERT(current->journal_info == NULL);
661 opc = lustre_msg_get_opc(req->rq_reqmsg);
663 if (opc == SEC_CTX_INIT ||
664 opc == SEC_CTX_INIT_CONT ||
668 if (opc != MGS_CONNECT) {
669 if (!class_connected_export(req->rq_export)) {
670 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
671 req->rq_status = -ENOTCONN;
672 GOTO(out, rc = -ENOTCONN);
678 DEBUG_REQ(D_MGS, req, "connect");
679 /* MGS and MDS have same request format for connect */
680 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
681 rc = target_handle_connect(req);
683 rc = mgs_connect_check_sptlrpc(req);
685 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
686 /* Make clients trying to reconnect after a MGS restart
687 happy; also requires obd_replayable */
688 lustre_msg_add_op_flags(req->rq_repmsg,
689 MSG_CONNECT_RECONNECT);
692 DEBUG_REQ(D_MGS, req, "disconnect");
693 /* MGS and MDS have same request format for disconnect */
694 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
695 rc = target_handle_disconnect(req);
696 req->rq_status = rc; /* superfluous? */
699 DEBUG_REQ(D_MGS, req, "exception");
700 rc = mgs_handle_exception(req);
703 DEBUG_REQ(D_MGS, req, "target add");
704 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
705 rc = mgs_handle_target_reg(req);
708 DEBUG_REQ(D_MGS, req, "target del");
709 rc = mgs_handle_target_del(req);
712 DEBUG_REQ(D_MGS, req, "set_info");
713 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
714 rc = mgs_set_info_rpc(req);
716 case MGS_CONFIG_READ:
717 DEBUG_REQ(D_MGS, req, "read config");
718 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
719 rc = mgs_config_read(req);
722 DEBUG_REQ(D_MGS, req, "enqueue");
723 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
724 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
725 ldlm_server_blocking_ast, NULL);
727 case LDLM_BL_CALLBACK:
728 case LDLM_CP_CALLBACK:
729 DEBUG_REQ(D_MGS, req, "callback");
730 CERROR("callbacks should not happen on MGS\n");
735 DEBUG_REQ(D_INFO, req, "ping");
736 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
737 rc = target_handle_ping(req);
740 DEBUG_REQ(D_MGS, req, "log cancel");
741 rc = -ENOTSUPP; /* la la la */
744 case LLOG_ORIGIN_HANDLE_CREATE:
745 DEBUG_REQ(D_MGS, req, "llog_open");
746 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
747 rc = llog_origin_handle_open(req);
749 (void)mgs_handle_fslog_hack(req);
751 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
752 DEBUG_REQ(D_MGS, req, "llog next block");
753 req_capsule_set(&req->rq_pill,
754 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
755 rc = llog_origin_handle_next_block(req);
757 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
758 DEBUG_REQ(D_MGS, req, "llog prev block");
759 req_capsule_set(&req->rq_pill,
760 &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
761 rc = llog_origin_handle_prev_block(req);
763 case LLOG_ORIGIN_HANDLE_READ_HEADER:
764 DEBUG_REQ(D_MGS, req, "llog read header");
765 req_capsule_set(&req->rq_pill,
766 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
767 rc = llog_origin_handle_read_header(req);
769 case LLOG_ORIGIN_HANDLE_CLOSE:
770 DEBUG_REQ(D_MGS, req, "llog close");
771 rc = llog_origin_handle_close(req);
777 LASSERT(current->journal_info == NULL);
779 DEBUG_REQ(D_MGS, req, "MGS fail to handle opc = %d: rc = %d\n",
782 rc = ptlrpc_error(req);
786 target_send_reply(req, rc, fail);
790 static inline int mgs_init_export(struct obd_export *exp)
792 struct mgs_export_data *data = &exp->u.eu_mgs_data;
794 /* init mgs_export_data for fsc */
795 cfs_spin_lock_init(&data->med_lock);
796 CFS_INIT_LIST_HEAD(&data->med_clients);
798 cfs_spin_lock(&exp->exp_lock);
799 exp->exp_connecting = 1;
800 cfs_spin_unlock(&exp->exp_lock);
802 /* self-export doesn't need client data and ldlm initialization */
803 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
804 &exp->exp_client_uuid)))
806 return ldlm_init_export(exp);
809 static inline int mgs_destroy_export(struct obd_export *exp)
813 target_destroy_export(exp);
814 mgs_client_free(exp);
816 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
817 &exp->exp_client_uuid)))
820 ldlm_destroy_export(exp);
825 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
830 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
838 strcpy(poolname, ptr);
843 static int mgs_iocontrol_pool(const struct lu_env *env,
844 struct mgs_device *mgs,
845 struct obd_ioctl_data *data)
847 struct mgs_thread_info *mgi = mgs_env_info(env);
849 struct lustre_cfg *lcfg = NULL;
850 struct llog_rec_hdr rec;
851 char *poolname = NULL;
854 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
855 if (poolname == NULL)
857 rec.lrh_len = llog_data_len(data->ioc_plen1);
859 if (data->ioc_type == LUSTRE_CFG_TYPE) {
860 rec.lrh_type = OBD_CFG_REC;
862 CERROR("unknown cfg record type:%d \n", data->ioc_type);
863 GOTO(out_pool, rc = -EINVAL);
866 if (data->ioc_plen1 > CFS_PAGE_SIZE)
867 GOTO(out_pool, rc = -E2BIG);
869 OBD_ALLOC(lcfg, data->ioc_plen1);
871 GOTO(out_pool, rc = -ENOMEM);
873 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
874 GOTO(out_lcfg, rc = -EFAULT);
876 if (lcfg->lcfg_bufcount < 2)
877 GOTO(out_lcfg, rc = -EFAULT);
879 /* first arg is always <fsname>.<poolname> */
880 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
885 switch (lcfg->lcfg_command) {
887 if (lcfg->lcfg_bufcount != 2)
888 GOTO(out_lcfg, rc = -EINVAL);
889 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
893 if (lcfg->lcfg_bufcount != 3)
894 GOTO(out_lcfg, rc = -EINVAL);
895 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
896 poolname, lustre_cfg_string(lcfg, 2));
899 if (lcfg->lcfg_bufcount != 3)
900 GOTO(out_lcfg, rc = -EINVAL);
901 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
902 poolname, lustre_cfg_string(lcfg, 2));
905 if (lcfg->lcfg_bufcount != 2)
906 GOTO(out_lcfg, rc = -EINVAL);
907 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
915 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
916 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
921 OBD_FREE(lcfg, data->ioc_plen1);
923 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
927 /* from mdt_iocontrol */
928 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
929 void *karg, void *uarg)
931 struct mgs_device *mgs = exp2mgs_dev(exp);
932 struct obd_ioctl_data *data = karg;
933 struct lvfs_run_ctxt saved;
938 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
940 rc = lu_env_init(&env, LCT_MG_THREAD);
946 case OBD_IOC_PARAM: {
947 struct mgs_thread_info *mgi = mgs_env_info(&env);
948 struct lustre_cfg *lcfg;
949 struct llog_rec_hdr rec;
951 rec.lrh_len = llog_data_len(data->ioc_plen1);
953 if (data->ioc_type == LUSTRE_CFG_TYPE) {
954 rec.lrh_type = OBD_CFG_REC;
956 CERROR("unknown cfg record type:%d \n", data->ioc_type);
957 GOTO(out, rc = -EINVAL);
960 OBD_ALLOC(lcfg, data->ioc_plen1);
962 GOTO(out, rc = -ENOMEM);
963 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
964 GOTO(out_free, rc = -EFAULT);
966 if (lcfg->lcfg_bufcount < 1)
967 GOTO(out_free, rc = -EINVAL);
969 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
971 CERROR("%s: setparam err: rc = %d\n",
972 exp->exp_obd->obd_name, rc);
974 OBD_FREE(lcfg, data->ioc_plen1);
979 rc = mgs_iocontrol_pool(&env, mgs, data);
982 case OBD_IOC_DUMP_LOG: {
983 struct llog_ctxt *ctxt;
984 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
985 push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
986 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
987 pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
993 case OBD_IOC_LLOG_CHECK:
994 case OBD_IOC_LLOG_INFO:
995 case OBD_IOC_LLOG_PRINT: {
996 struct llog_ctxt *ctxt;
997 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
999 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1000 rc = llog_ioctl(ctxt, cmd, data);
1001 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1002 llog_ctxt_put(ctxt);
1008 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1017 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1019 struct obd_connect_data *data = NULL;
1020 struct obd_device *obd;
1024 OBD_ALLOC_PTR(data);
1028 obd = class_name2obd(nextdev);
1030 CERROR("can't locate next device: %s\n", nextdev);
1031 GOTO(out, rc = -ENOTCONN);
1034 data->ocd_version = LUSTRE_VERSION_CODE;
1036 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1037 &obd->obd_uuid, data, NULL);
1039 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1043 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1044 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1045 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1052 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1053 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1055 static struct ptlrpc_service_conf conf;
1056 struct lprocfs_static_vars lvars = { 0 };
1057 struct obd_device *obd;
1058 struct lustre_mount_info *lmi;
1062 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1066 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1068 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1072 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1075 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1077 obd->u.obt.obt_magic = OBT_MAGIC;
1078 obd->u.obt.obt_instance = 0;
1080 /* namespace for mgs llog */
1081 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1082 LDLM_NAMESPACE_SERVER,
1083 LDLM_NAMESPACE_MODEST,
1085 if (obd->obd_namespace == NULL)
1086 GOTO(err_ops, rc = -ENOMEM);
1089 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1090 "mgs_ldlm_client", &obd->obd_ldlm_client);
1092 rc = mgs_fs_setup(env, mgs);
1094 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1099 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1103 /* No recovery for MGC's */
1104 obd->obd_replayable = 0;
1106 /* Internal mgs setup */
1107 mgs_init_fsdb_list(mgs);
1108 cfs_mutex_init(&mgs->mgs_mutex);
1109 mgs->mgs_start_time = cfs_time_current_sec();
1110 cfs_spin_lock_init(&mgs->mgs_lock);
1113 lprocfs_mgs_init_vars(&lvars);
1114 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
1115 lproc_mgs_setup(mgs);
1116 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
1121 conf = (typeof(conf)) {
1122 .psc_name = LUSTRE_MGS_NAME,
1123 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1125 .bc_nbufs = MGS_NBUFS,
1126 .bc_buf_size = MGS_BUFSIZE,
1127 .bc_req_max_size = MGS_MAXREQSIZE,
1128 .bc_rep_max_size = MGS_MAXREPSIZE,
1129 .bc_req_portal = MGS_REQUEST_PORTAL,
1130 .bc_rep_portal = MGC_REPLY_PORTAL,
1133 .tc_thr_name = "ll_mgs",
1134 .tc_nthrs_init = MGS_NTHRS_INIT,
1135 .tc_nthrs_max = MGS_NTHRS_MAX,
1136 .tc_ctx_tags = LCT_MG_THREAD,
1139 .so_req_handler = mgs_handle,
1140 .so_req_printer = target_print_req,
1143 /* Start the service threads */
1144 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1145 if (IS_ERR(mgs->mgs_service)) {
1146 rc = PTR_ERR(mgs->mgs_service);
1147 CERROR("failed to start service: %d\n", rc);
1148 GOTO(err_lproc, rc);
1151 ping_evictor_start();
1153 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1155 /* device stack is not yet fully setup to keep no objects behind */
1156 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1160 lproc_mgs_cleanup(mgs);
1163 /* No extra cleanup needed for llog_init_commit_thread() */
1164 mgs_fs_cleanup(env, mgs);
1166 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1167 obd->obd_namespace = NULL;
1169 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1170 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1171 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1172 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1175 obd_disconnect(mgs->mgs_bottom_exp);
1179 static struct lu_device *mgs_device_free(const struct lu_env *env,
1180 struct lu_device *lu)
1182 struct mgs_device *mgs = lu2mgs_dev(lu);
1185 dt_device_fini(&mgs->mgs_dt_dev);
1190 static int mgs_process_config(const struct lu_env *env,
1191 struct lu_device *dev,
1192 struct lustre_cfg *lcfg)
1198 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1199 const struct lu_object_conf *unused)
1201 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1202 struct lu_device *under;
1203 struct lu_object *below;
1207 /* do no set .do_ops as mgs calls to bottom osd directly */
1209 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1210 PFID(lu_object_fid(o)));
1212 under = &d->mgs_bottom->dd_lu_dev;
1213 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1215 lu_object_add(o, below);
1222 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1224 struct mgs_object *obj = lu2mgs_obj(o);
1225 struct lu_object_header *h = o->lo_header;
1227 dt_object_fini(&obj->mgo_obj);
1228 lu_object_header_fini(h);
1232 static int mgs_object_print(const struct lu_env *env, void *cookie,
1233 lu_printer_t p, const struct lu_object *l)
1235 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1237 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1240 struct lu_object_operations mgs_lu_obj_ops = {
1241 .loo_object_init = mgs_object_init,
1242 .loo_object_free = mgs_object_free,
1243 .loo_object_print = mgs_object_print,
1246 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1247 const struct lu_object_header *hdr,
1248 struct lu_device *d)
1250 struct lu_object_header *h;
1251 struct mgs_object *o;
1252 struct lu_object *l;
1254 LASSERT(hdr == NULL);
1258 l = &o->mgo_obj.do_lu;
1261 lu_object_header_init(h);
1262 dt_object_init(&o->mgo_obj, h, d);
1263 lu_object_add_top(h, l);
1265 l->lo_ops = &mgs_lu_obj_ops;
1273 const struct lu_device_operations mgs_lu_ops = {
1274 .ldo_object_alloc = mgs_object_alloc,
1275 .ldo_process_config = mgs_process_config,
1278 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1279 struct lu_device_type *type,
1280 struct lustre_cfg *lcfg)
1282 struct mgs_device *mgs;
1283 struct lu_device *ludev;
1287 ludev = ERR_PTR(-ENOMEM);
1291 ludev = mgs2lu_dev(mgs);
1292 dt_device_init(&mgs->mgs_dt_dev, type);
1293 rc = mgs_init0(env, mgs, type, lcfg);
1295 mgs_device_free(env, ludev);
1296 ludev = ERR_PTR(rc);
1302 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1303 struct lu_device *d)
1305 struct mgs_device *mgs = lu2mgs_dev(d);
1306 struct obd_device *obd = mgs->mgs_obd;
1309 LASSERT(mgs->mgs_bottom);
1311 ping_evictor_stop();
1313 ptlrpc_unregister_service(mgs->mgs_service);
1315 obd_exports_barrier(obd);
1316 obd_zombie_barrier();
1318 mgs_cleanup_fsdb_list(mgs);
1319 lproc_mgs_cleanup(mgs);
1321 obd_llog_finish(obd, 0);
1323 mgs_fs_cleanup(env, mgs);
1325 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1326 obd->obd_namespace = NULL;
1328 lu_site_purge(env, d->ld_site, ~0);
1329 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1330 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1331 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1334 LASSERT(mgs->mgs_bottom_exp);
1335 obd_disconnect(mgs->mgs_bottom_exp);
1337 server_put_mount(obd->obd_name, NULL);
1342 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1343 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1345 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1347 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1349 static struct lu_device_type_operations mgs_device_type_ops = {
1350 .ldto_init = mgs_type_init,
1351 .ldto_fini = mgs_type_fini,
1353 .ldto_start = mgs_type_start,
1354 .ldto_stop = mgs_type_stop,
1356 .ldto_device_alloc = mgs_device_alloc,
1357 .ldto_device_free = mgs_device_free,
1359 .ldto_device_fini = mgs_device_fini
1362 static struct lu_device_type mgs_device_type = {
1363 .ldt_tags = LU_DEVICE_DT,
1364 .ldt_name = LUSTRE_MGS_NAME,
1365 .ldt_ops = &mgs_device_type_ops,
1366 .ldt_ctx_tags = LCT_MG_THREAD
1370 /* use obd ops to offer management infrastructure */
1371 static struct obd_ops mgs_obd_ops = {
1372 .o_owner = THIS_MODULE,
1373 .o_connect = mgs_connect,
1374 .o_reconnect = mgs_reconnect,
1375 .o_disconnect = mgs_disconnect,
1376 .o_init_export = mgs_init_export,
1377 .o_destroy_export = mgs_destroy_export,
1378 .o_iocontrol = mgs_iocontrol,
1379 .o_llog_init = mgs_llog_init,
1380 .o_llog_finish = mgs_llog_finish
1383 static int __init mgs_init(void)
1385 struct lprocfs_static_vars lvars;
1387 lprocfs_mgs_init_vars(&lvars);
1388 class_register_type(&mgs_obd_ops, NULL, lvars.module_vars,
1389 LUSTRE_MGS_NAME, &mgs_device_type);
1394 static void /*__exit*/ mgs_exit(void)
1396 class_unregister_type(LUSTRE_MGS_NAME);
1399 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1400 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1401 MODULE_LICENSE("GPL");
1403 module_init(mgs_init);
1404 module_exit(mgs_exit);