1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
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>
42 # define EXPORT_SYMTAB
44 #define DEBUG_SUBSYSTEM S_MGS
45 #define D_MGS D_CONFIG
48 # include <linux/module.h>
49 # include <linux/pagemap.h>
50 # include <linux/miscdevice.h>
51 # include <linux/init.h>
53 # include <liblustre.h>
56 #include <obd_class.h>
57 #include <lustre_dlm.h>
58 #include <lprocfs_status.h>
59 #include <lustre_fsfilt.h>
60 #include <lustre_disk.h>
61 #include "mgs_internal.h"
62 #include <lustre_param.h>
64 /* Establish a connection to the MGS.*/
65 static int mgs_connect(const struct lu_env *env,
66 struct obd_export **exp, struct obd_device *obd,
67 struct obd_uuid *cluuid, struct obd_connect_data *data,
70 struct obd_export *lexp;
71 struct lustre_handle conn = { 0 };
75 if (!exp || !obd || !cluuid)
78 rc = class_connect(&conn, obd, cluuid);
82 lexp = class_conn2export(&conn);
85 mgs_counter_incr(lexp, LPROC_MGS_CONNECT);
88 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
89 lexp->exp_connect_flags = data->ocd_connect_flags;
90 data->ocd_version = LUSTRE_VERSION_CODE;
93 rc = mgs_export_stats_init(obd, lexp, localdata);
96 class_disconnect(lexp);
104 static int mgs_reconnect(const struct lu_env *env,
105 struct obd_export *exp, struct obd_device *obd,
106 struct obd_uuid *cluuid, struct obd_connect_data *data,
111 if (exp == NULL || obd == NULL || cluuid == NULL)
114 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
117 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
118 exp->exp_connect_flags = data->ocd_connect_flags;
119 data->ocd_version = LUSTRE_VERSION_CODE;
122 RETURN(mgs_export_stats_init(obd, exp, localdata));
125 static int mgs_disconnect(struct obd_export *exp)
132 mgs_fsc_cleanup(exp);
134 class_export_get(exp);
135 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
137 rc = server_disconnect_export(exp);
138 class_export_put(exp);
142 static int mgs_cleanup(struct obd_device *obd);
143 static int mgs_handle(struct ptlrpc_request *req);
145 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
146 struct obd_device *tgt, int *index)
151 LASSERT(olg == &obd->obd_olg);
152 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
157 static int mgs_llog_finish(struct obd_device *obd, int count)
159 struct llog_ctxt *ctxt;
163 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
165 rc = llog_cleanup(ctxt);
170 /* Start the MGS obd */
171 static int mgs_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
173 struct lprocfs_static_vars lvars;
174 struct mgs_obd *mgs = &obd->u.mgs;
175 struct lustre_mount_info *lmi;
176 struct lustre_sb_info *lsi;
177 struct vfsmount *mnt;
181 CDEBUG(D_CONFIG, "Starting MGS\n");
184 lmi = server_get_mount(obd->obd_name);
186 RETURN(rc = -EINVAL);
189 lsi = s2lsi(lmi->lmi_sb);
190 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
191 if (IS_ERR(obd->obd_fsops))
192 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
194 if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) {
195 CERROR("%s: Underlying device is marked as read-only. "
196 "Setup failed\n", obd->obd_name);
197 GOTO(err_ops, rc = -EROFS);
200 obd->u.obt.obt_magic = OBT_MAGIC;
201 obd->u.obt.obt_instance = 0;
203 /* namespace for mgs llog */
204 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
205 LDLM_NAMESPACE_SERVER,
206 LDLM_NAMESPACE_MODEST,
208 if (obd->obd_namespace == NULL)
209 GOTO(err_ops, rc = -ENOMEM);
212 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
213 "mgs_ldlm_client", &obd->obd_ldlm_client);
215 rc = mgs_fs_setup(obd, mnt);
217 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
222 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
226 /* No recovery for MGC's */
227 obd->obd_replayable = 0;
229 /* Internal mgs setup */
230 mgs_init_fsdb_list(obd);
231 cfs_sema_init(&mgs->mgs_sem, 1);
232 mgs->mgs_start_time = cfs_time_current_sec();
235 lprocfs_mgs_init_vars(&lvars);
236 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
237 lproc_mgs_setup(obd);
238 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
243 /* Start the service threads */
245 ptlrpc_init_svc(MGS_NBUFS, MGS_BUFSIZE, MGS_MAXREQSIZE,
246 MGS_MAXREPSIZE, MGS_REQUEST_PORTAL,
248 mgs_handle, LUSTRE_MGS_NAME,
249 obd->obd_proc_entry, target_print_req,
250 MGS_THREADS_AUTO_MIN, MGS_THREADS_AUTO_MAX,
251 "ll_mgs", LCT_MD_THREAD, NULL);
253 if (!mgs->mgs_service) {
254 CERROR("failed to start service\n");
255 GOTO(err_llog, rc = -ENOMEM);
258 rc = ptlrpc_start_threads(mgs->mgs_service);
260 GOTO(err_thread, rc);
262 ping_evictor_start();
264 LCONSOLE_INFO("MGS %s started\n", obd->obd_name);
269 ptlrpc_unregister_service(mgs->mgs_service);
271 lproc_mgs_cleanup(obd);
272 obd_llog_finish(obd, 0);
274 /* No extra cleanup needed for llog_init_commit_thread() */
277 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
278 obd->obd_namespace = NULL;
280 fsfilt_put_ops(obd->obd_fsops);
282 server_put_mount(obd->obd_name, mnt);
287 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
293 case OBD_CLEANUP_EARLY:
294 case OBD_CLEANUP_EXPORTS:
295 rc = obd_llog_finish(obd, 0);
302 * Performs cleanup procedures for passed \a obd given it is mgs obd.
304 static int mgs_cleanup(struct obd_device *obd)
306 struct mgs_obd *mgs = &obd->u.mgs;
309 if (mgs->mgs_sb == NULL)
314 ptlrpc_unregister_service(mgs->mgs_service);
316 mgs_cleanup_fsdb_list(obd);
317 lproc_mgs_cleanup(obd);
320 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
323 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
324 obd->obd_namespace = NULL;
326 fsfilt_put_ops(obd->obd_fsops);
328 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
332 /* similar to filter_prepare_destroy */
333 int mgs_get_lock(struct obd_device *obd, struct ldlm_res_id *res,
334 struct lustre_handle *lockh)
339 rc = ldlm_cli_enqueue_local(obd->obd_namespace, res,
340 LDLM_PLAIN, NULL, LCK_EX,
341 &flags, ldlm_blocking_ast,
342 ldlm_completion_ast, NULL,
343 NULL, 0, NULL, lockh);
345 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
346 le64_to_cpu(res->name[0]), le64_to_cpu(res->name[1]),
352 int mgs_put_lock(struct lustre_handle *lockh)
355 ldlm_lock_decref_and_cancel(lockh, LCK_EX);
359 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb)
361 struct lustre_handle lockh;
362 struct ldlm_res_id res_id;
367 LASSERT(fsdb->fsdb_name[0] != '\0');
368 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, CONFIG_T_CONFIG);
371 bit = FSDB_REVOKING_LOCK;
372 if (!rc && cfs_test_and_set_bit(bit, &fsdb->fsdb_flags) == 0) {
373 lockrc = mgs_get_lock(obd, &res_id, &lockh);
374 /* clear the bit before lock put */
375 cfs_clear_bit(bit, &fsdb->fsdb_flags);
377 if (lockrc != ELDLM_OK)
378 CERROR("lock error %d for fs %s\n",
379 lockrc, fsdb->fsdb_name);
381 mgs_put_lock(&lockh);
388 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
393 rc = mgs_check_index(obd, mti);
395 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
396 "this MGS does not know about it, preventing "
397 "registration.\n", mti->mti_svname);
399 } else if (rc == -1) {
400 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
401 "disappeared! Regenerating all logs.\n",
403 mti->mti_flags |= LDD_F_WRITECONF;
406 /* Index is correctly marked as used */
408 /* If the logs don't contain the mti_nids then add
409 them as failover nids */
410 rc = mgs_check_failnid(obd, mti);
416 /* Ensure this is not a failover node that is connecting first*/
417 static int mgs_check_failover_reg(struct mgs_target_info *mti)
423 ptr = mti->mti_params;
424 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
425 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
426 for (i = 0; i < mti->mti_nid_count; i++) {
427 if (nid == mti->mti_nids[i]) {
428 LCONSOLE_WARN("Denying initial registra"
429 "tion attempt from nid %s"
430 ", specified as failover"
431 "\n",libcfs_nid2str(nid));
432 return -EADDRNOTAVAIL;
440 /* Called whenever a target starts up. Flags indicate first connect, etc. */
441 static int mgs_handle_target_reg(struct ptlrpc_request *req)
443 struct obd_device *obd = req->rq_export->exp_obd;
444 struct mgs_target_info *mti, *rep_mti;
450 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
452 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
454 opc = mti->mti_flags & LDD_F_OPC_MASK;
455 if (opc == LDD_F_OPC_READY) {
456 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
457 mti->mti_fsname, mti->mti_stripe_index);
458 rc = mgs_ir_update(obd, mti);
460 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
461 CERROR("Update IR return with %d(ignore and IR "
464 GOTO(out_nolock, rc);
467 /* Do not support unregistering right now. */
468 if (opc != LDD_F_OPC_REG)
469 GOTO(out_nolock, rc = -EINVAL);
471 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
472 mti->mti_fsname, mti->mti_stripe_index);
474 if (mti->mti_flags & LDD_F_NEED_INDEX)
475 mti->mti_flags |= LDD_F_WRITECONF;
477 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
479 /* We're just here as a startup ping. */
480 CDEBUG(D_MGS, "Server %s is running on %s\n",
481 mti->mti_svname, obd_export_nid2str(req->rq_export));
482 rc = mgs_check_target(obd, mti);
483 /* above will set appropriate mti flags */
485 /* Nothing wrong, or fatal error */
486 GOTO(out_nolock, rc);
488 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
489 && (rc = mgs_check_failover_reg(mti)))
490 GOTO(out_nolock, rc);
493 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
495 if (mti->mti_flags & LDD_F_WRITECONF) {
496 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
497 mti->mti_stripe_index == 0) {
498 rc = mgs_erase_logs(obd, mti->mti_fsname);
499 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
500 "request. All servers must be restarted "
501 "in order to regenerate the logs."
502 "\n", obd->obd_name, mti->mti_fsname);
503 } else if (mti->mti_flags &
504 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
505 rc = mgs_erase_log(obd, mti->mti_svname);
506 LCONSOLE_WARN("%s: Regenerating %s log by user "
508 obd->obd_name, mti->mti_svname);
510 mti->mti_flags |= LDD_F_UPDATE;
511 /* Erased logs means start from scratch. */
512 mti->mti_flags &= ~LDD_F_UPGRADE14;
515 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
517 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
518 GOTO(out_nolock, rc);
522 * Log writing contention is handled by the fsdb_sem.
524 * It should be alright if someone was reading while we were
525 * updating the logs - if we revoke at the end they will just update
526 * from where they left off.
530 if (mti->mti_flags & LDD_F_UPGRADE14) {
531 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
533 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
537 /* We're good to go */
538 mti->mti_flags |= LDD_F_UPDATE;
542 if (mti->mti_flags & LDD_F_UPDATE) {
543 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
544 mti->mti_stripe_index);
546 /* create or update the target log
547 and update the client/mdt logs */
548 rc = mgs_write_log_target(obd, mti, fsdb);
550 CERROR("Failed to write %s log (%d)\n",
551 mti->mti_svname, rc);
555 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
556 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
558 mti->mti_flags |= LDD_F_REWRITE_LDD;
562 mgs_revoke_lock(obd, fsdb);
565 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
566 mti->mti_stripe_index, rc);
569 /* we need an error flag to tell the target what's going on,
570 * instead of just doing it by error code only. */
571 mti->mti_flags |= LDD_F_ERROR;
573 rc = req_capsule_server_pack(&req->rq_pill);
577 /* send back the whole mti in the reply */
578 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
581 /* Flush logs to disk */
582 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
586 static int mgs_set_info_rpc(struct ptlrpc_request *req)
588 struct obd_device *obd = req->rq_export->exp_obd;
589 struct mgs_send_param *msp, *rep_msp;
591 struct lustre_cfg_bufs bufs;
592 struct lustre_cfg *lcfg;
593 char fsname[MTI_NAME_MAXLEN];
596 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
599 /* Construct lustre_cfg structure to pass to function mgs_setparam */
600 lustre_cfg_bufs_reset(&bufs, NULL);
601 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
602 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
603 rc = mgs_setparam(obd, lcfg, fsname);
605 CERROR("Error %d in setting the parameter %s for fs %s\n",
606 rc, msp->mgs_param, fsname);
610 lustre_cfg_free(lcfg);
612 rc = req_capsule_server_pack(&req->rq_pill);
614 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
620 static int mgs_config_read(struct ptlrpc_request *req)
622 struct mgs_config_body *body;
626 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
630 switch (body->mcb_type) {
631 case CONFIG_T_RECOVER:
632 rc = mgs_get_ir_logs(req);
635 case CONFIG_T_CONFIG:
648 * similar as in ost_connect_check_sptlrpc()
650 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
652 struct obd_export *exp = req->rq_export;
653 struct obd_device *obd = exp->exp_obd;
655 struct sptlrpc_flavor flvr;
658 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
659 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
663 cfs_down(&fsdb->fsdb_sem);
664 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
665 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
668 /* by defualt allow any flavors */
669 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
671 cfs_up(&fsdb->fsdb_sem);
673 cfs_spin_lock(&exp->exp_lock);
675 exp->exp_sp_peer = req->rq_sp_from;
676 exp->exp_flvr = flvr;
678 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
679 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
680 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
681 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
682 libcfs_nid2str(req->rq_peer.nid));
686 cfs_spin_unlock(&exp->exp_lock);
688 if (exp->exp_sp_peer != req->rq_sp_from) {
689 CERROR("RPC source %s doesn't match %s\n",
690 sptlrpc_part2name(req->rq_sp_from),
691 sptlrpc_part2name(exp->exp_sp_peer));
694 rc = sptlrpc_target_export_check(exp, req);
701 /* Called whenever a target cleans up. */
702 /* XXX - Currently unused */
703 static int mgs_handle_target_del(struct ptlrpc_request *req)
706 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
710 /* XXX - Currently unused */
711 static int mgs_handle_exception(struct ptlrpc_request *req)
714 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
718 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
725 /* XXX: We suppose that llog at mgs is only used for
726 * fetching file system log */
727 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
728 if (logname == NULL) {
729 CERROR("No logname, is llog on MGS used for something else?\n");
733 ptr = strchr(logname, '-');
734 rc = (int)(ptr - logname);
735 if (ptr == NULL || rc >= sizeof(fsname)) {
736 CERROR("Invalid logname received: %s\n", logname);
740 strncpy(fsname, logname, rc);
742 rc = mgs_fsc_attach(req->rq_export, fsname);
743 if (rc < 0 && rc != -EEXIST)
744 CERROR("add fs client %s returns %d\n", fsname, rc);
749 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
750 int mgs_handle(struct ptlrpc_request *req)
752 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
756 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
757 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
758 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
761 LASSERT(current->journal_info == NULL);
762 opc = lustre_msg_get_opc(req->rq_reqmsg);
764 if (opc == SEC_CTX_INIT ||
765 opc == SEC_CTX_INIT_CONT ||
769 if (opc != MGS_CONNECT) {
770 if (!class_connected_export(req->rq_export)) {
771 CERROR("lustre_mgs: operation %d on unconnected MGS\n",
773 req->rq_status = -ENOTCONN;
774 GOTO(out, rc = -ENOTCONN);
780 DEBUG_REQ(D_MGS, req, "connect");
781 /* MGS and MDS have same request format for connect */
782 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
783 rc = target_handle_connect(req);
785 rc = mgs_connect_check_sptlrpc(req);
787 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
788 /* Make clients trying to reconnect after a MGS restart
789 happy; also requires obd_replayable */
790 lustre_msg_add_op_flags(req->rq_repmsg,
791 MSG_CONNECT_RECONNECT);
794 DEBUG_REQ(D_MGS, req, "disconnect");
795 /* MGS and MDS have same request format for disconnect */
796 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
797 rc = target_handle_disconnect(req);
798 req->rq_status = rc; /* superfluous? */
801 DEBUG_REQ(D_MGS, req, "exception");
802 rc = mgs_handle_exception(req);
805 DEBUG_REQ(D_MGS, req, "target add");
806 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
807 rc = mgs_handle_target_reg(req);
810 DEBUG_REQ(D_MGS, req, "target del");
811 rc = mgs_handle_target_del(req);
814 DEBUG_REQ(D_MGS, req, "set_info");
815 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
816 rc = mgs_set_info_rpc(req);
818 case MGS_CONFIG_READ:
819 DEBUG_REQ(D_MGS, req, "read config");
820 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
821 rc = mgs_config_read(req);
824 DEBUG_REQ(D_MGS, req, "enqueue");
825 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
826 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
827 ldlm_server_blocking_ast, NULL);
829 case LDLM_BL_CALLBACK:
830 case LDLM_CP_CALLBACK:
831 DEBUG_REQ(D_MGS, req, "callback");
832 CERROR("callbacks should not happen on MGS\n");
837 DEBUG_REQ(D_INFO, req, "ping");
838 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
839 rc = target_handle_ping(req);
842 DEBUG_REQ(D_MGS, req, "log cancel");
843 rc = -ENOTSUPP; /* la la la */
846 case LLOG_ORIGIN_HANDLE_CREATE:
847 DEBUG_REQ(D_MGS, req, "llog_init");
848 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
849 rc = llog_origin_handle_create(req);
851 (void)mgs_handle_fslog_hack(req);
853 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
854 DEBUG_REQ(D_MGS, req, "llog next block");
855 req_capsule_set(&req->rq_pill,
856 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
857 rc = llog_origin_handle_next_block(req);
859 case LLOG_ORIGIN_HANDLE_READ_HEADER:
860 DEBUG_REQ(D_MGS, req, "llog read header");
861 req_capsule_set(&req->rq_pill,
862 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
863 rc = llog_origin_handle_read_header(req);
865 case LLOG_ORIGIN_HANDLE_CLOSE:
866 DEBUG_REQ(D_MGS, req, "llog close");
867 rc = llog_origin_handle_close(req);
870 DEBUG_REQ(D_MGS, req, "llog catinfo");
871 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
872 rc = llog_catinfo(req);
875 req->rq_status = -ENOTSUPP;
876 rc = ptlrpc_error(req);
880 LASSERT(current->journal_info == NULL);
883 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
886 target_send_reply(req, rc, fail);
890 static inline int mgs_init_export(struct obd_export *exp)
892 struct mgs_export_data *data = &exp->u.eu_mgs_data;
894 /* init mgs_export_data for fsc */
895 cfs_spin_lock_init(&data->med_lock);
896 CFS_INIT_LIST_HEAD(&data->med_clients);
898 cfs_spin_lock(&exp->exp_lock);
899 exp->exp_connecting = 1;
900 cfs_spin_unlock(&exp->exp_lock);
902 /* self-export doesn't need client data and ldlm initialization */
903 if (unlikely(exp == exp->exp_obd->obd_self_export))
905 return ldlm_init_export(exp);
908 static inline int mgs_destroy_export(struct obd_export *exp)
912 target_destroy_export(exp);
913 mgs_client_free(exp);
915 if (unlikely(exp == exp->exp_obd->obd_self_export))
918 ldlm_destroy_export(exp);
923 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
928 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
936 strcpy(poolname, ptr);
941 static int mgs_iocontrol_pool(struct obd_device *obd,
942 struct obd_ioctl_data *data)
945 struct lustre_cfg *lcfg = NULL;
946 struct llog_rec_hdr rec;
948 char *poolname = NULL;
951 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
955 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
956 if (poolname == NULL) {
960 rec.lrh_len = llog_data_len(data->ioc_plen1);
962 if (data->ioc_type == LUSTRE_CFG_TYPE) {
963 rec.lrh_type = OBD_CFG_REC;
965 CERROR("unknown cfg record type:%d \n", data->ioc_type);
970 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
975 OBD_ALLOC(lcfg, data->ioc_plen1);
977 GOTO(out_pool, rc = -ENOMEM);
979 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
980 GOTO(out_pool, rc = -EFAULT);
982 if (lcfg->lcfg_bufcount < 2) {
983 GOTO(out_pool, rc = -EFAULT);
986 /* first arg is always <fsname>.<poolname> */
987 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
990 switch (lcfg->lcfg_command) {
991 case LCFG_POOL_NEW: {
992 if (lcfg->lcfg_bufcount != 2)
994 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
998 case LCFG_POOL_ADD: {
999 if (lcfg->lcfg_bufcount != 3)
1001 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1002 lustre_cfg_string(lcfg, 2));
1005 case LCFG_POOL_REM: {
1006 if (lcfg->lcfg_bufcount != 3)
1008 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1009 lustre_cfg_string(lcfg, 2));
1012 case LCFG_POOL_DEL: {
1013 if (lcfg->lcfg_bufcount != 2)
1015 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1026 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1027 rc, lcfg->lcfg_command, fsname, poolname);
1033 OBD_FREE(lcfg, data->ioc_plen1);
1036 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1038 if (poolname != NULL)
1039 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1044 /* from mdt_iocontrol */
1045 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1046 void *karg, void *uarg)
1048 struct obd_device *obd = exp->exp_obd;
1049 struct obd_ioctl_data *data = karg;
1050 struct lvfs_run_ctxt saved;
1054 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1058 case OBD_IOC_PARAM: {
1059 struct lustre_cfg *lcfg;
1060 struct llog_rec_hdr rec;
1061 char fsname[MTI_NAME_MAXLEN];
1063 rec.lrh_len = llog_data_len(data->ioc_plen1);
1065 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1066 rec.lrh_type = OBD_CFG_REC;
1068 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1072 OBD_ALLOC(lcfg, data->ioc_plen1);
1075 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1076 GOTO(out_free, rc = -EFAULT);
1078 if (lcfg->lcfg_bufcount < 1)
1079 GOTO(out_free, rc = -EINVAL);
1081 rc = mgs_setparam(obd, lcfg, fsname);
1083 CERROR("setparam err %d\n", rc);
1087 OBD_FREE(lcfg, data->ioc_plen1);
1091 case OBD_IOC_POOL: {
1092 RETURN(mgs_iocontrol_pool(obd, data));
1095 case OBD_IOC_DUMP_LOG: {
1096 struct llog_ctxt *ctxt;
1097 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1098 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1099 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1100 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1101 llog_ctxt_put(ctxt);
1106 case OBD_IOC_LLOG_CHECK:
1107 case OBD_IOC_LLOG_INFO:
1108 case OBD_IOC_LLOG_PRINT: {
1109 struct llog_ctxt *ctxt;
1110 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1112 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1113 rc = llog_ioctl(ctxt, cmd, data);
1114 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1115 llog_ctxt_put(ctxt);
1121 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1127 /* use obd ops to offer management infrastructure */
1128 static struct obd_ops mgs_obd_ops = {
1129 .o_owner = THIS_MODULE,
1130 .o_connect = mgs_connect,
1131 .o_reconnect = mgs_reconnect,
1132 .o_disconnect = mgs_disconnect,
1133 .o_setup = mgs_setup,
1134 .o_precleanup = mgs_precleanup,
1135 .o_cleanup = mgs_cleanup,
1136 .o_init_export = mgs_init_export,
1137 .o_destroy_export = mgs_destroy_export,
1138 .o_iocontrol = mgs_iocontrol,
1139 .o_llog_init = mgs_llog_init,
1140 .o_llog_finish = mgs_llog_finish
1143 static int __init mgs_init(void)
1145 struct lprocfs_static_vars lvars;
1147 lprocfs_mgs_init_vars(&lvars);
1148 class_register_type(&mgs_obd_ops, NULL,
1149 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1154 static void /*__exit*/ mgs_exit(void)
1156 class_unregister_type(LUSTRE_MGS_NAME);
1159 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1160 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1161 MODULE_LICENSE("GPL");
1163 module_init(mgs_init);
1164 module_exit(mgs_exit);