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>
41 #define DEBUG_SUBSYSTEM S_MGS
42 #define D_MGS D_CONFIG
45 # include <linux/module.h>
46 # include <linux/pagemap.h>
47 # include <linux/miscdevice.h>
48 # include <linux/init.h>
50 # include <liblustre.h>
53 #include <obd_class.h>
54 #include <lustre_dlm.h>
55 #include <lprocfs_status.h>
56 #include <lustre_fsfilt.h>
57 #include <lustre_disk.h>
58 #include "mgs_internal.h"
59 #include <lustre_param.h>
61 /* Establish a connection to the MGS.*/
62 static int mgs_connect(const struct lu_env *env,
63 struct obd_export **exp, struct obd_device *obd,
64 struct obd_uuid *cluuid, struct obd_connect_data *data,
67 struct obd_export *lexp;
68 struct lustre_handle conn = { 0 };
72 if (!exp || !obd || !cluuid)
75 rc = class_connect(&conn, obd, cluuid);
79 lexp = class_conn2export(&conn);
82 mgs_counter_incr(lexp, LPROC_MGS_CONNECT);
85 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
86 lexp->exp_connect_flags = data->ocd_connect_flags;
87 data->ocd_version = LUSTRE_VERSION_CODE;
90 rc = mgs_export_stats_init(obd, lexp, localdata);
93 class_disconnect(lexp);
101 static int mgs_reconnect(const struct lu_env *env,
102 struct obd_export *exp, struct obd_device *obd,
103 struct obd_uuid *cluuid, struct obd_connect_data *data,
108 if (exp == NULL || obd == NULL || cluuid == NULL)
111 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
114 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
115 exp->exp_connect_flags = data->ocd_connect_flags;
116 data->ocd_version = LUSTRE_VERSION_CODE;
119 RETURN(mgs_export_stats_init(obd, exp, localdata));
122 static int mgs_disconnect(struct obd_export *exp)
129 mgs_fsc_cleanup(exp);
131 class_export_get(exp);
132 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
134 rc = server_disconnect_export(exp);
135 class_export_put(exp);
139 static int mgs_cleanup(struct obd_device *obd);
140 static int mgs_handle(struct ptlrpc_request *req);
142 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
143 struct obd_device *tgt, int *index)
148 LASSERT(olg == &obd->obd_olg);
149 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
154 static int mgs_llog_finish(struct obd_device *obd, int count)
156 struct llog_ctxt *ctxt;
160 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
162 rc = llog_cleanup(ctxt);
167 /* Start the MGS obd */
168 static int mgs_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
170 struct lprocfs_static_vars lvars;
171 struct mgs_obd *mgs = &obd->u.mgs;
172 struct lustre_mount_info *lmi;
173 struct lustre_sb_info *lsi;
174 struct vfsmount *mnt;
178 CDEBUG(D_CONFIG, "Starting MGS\n");
181 lmi = server_get_mount(obd->obd_name);
183 RETURN(rc = -EINVAL);
186 lsi = s2lsi(lmi->lmi_sb);
187 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
188 if (IS_ERR(obd->obd_fsops))
189 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
191 if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) {
192 CERROR("%s: Underlying device is marked as read-only. "
193 "Setup failed\n", obd->obd_name);
194 GOTO(err_ops, rc = -EROFS);
197 obd->u.obt.obt_magic = OBT_MAGIC;
198 obd->u.obt.obt_instance = 0;
200 /* namespace for mgs llog */
201 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
202 LDLM_NAMESPACE_SERVER,
203 LDLM_NAMESPACE_MODEST,
205 if (obd->obd_namespace == NULL)
206 GOTO(err_ops, rc = -ENOMEM);
209 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
210 "mgs_ldlm_client", &obd->obd_ldlm_client);
212 rc = mgs_fs_setup(obd, mnt);
214 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
219 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
223 /* No recovery for MGC's */
224 obd->obd_replayable = 0;
226 /* Internal mgs setup */
227 mgs_init_fsdb_list(obd);
228 cfs_mutex_init(&mgs->mgs_mutex);
229 mgs->mgs_start_time = cfs_time_current_sec();
232 lprocfs_mgs_init_vars(&lvars);
233 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
234 lproc_mgs_setup(obd);
235 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
240 /* Start the service threads */
242 ptlrpc_init_svc(MGS_NBUFS, MGS_BUFSIZE, MGS_MAXREQSIZE,
243 MGS_MAXREPSIZE, MGS_REQUEST_PORTAL,
245 mgs_handle, LUSTRE_MGS_NAME,
246 obd->obd_proc_entry, target_print_req,
247 MGS_THREADS_AUTO_MIN, MGS_THREADS_AUTO_MAX,
248 "ll_mgs", LCT_MD_THREAD, NULL);
250 if (!mgs->mgs_service) {
251 CERROR("failed to start service\n");
252 GOTO(err_llog, rc = -ENOMEM);
255 rc = ptlrpc_start_threads(mgs->mgs_service);
257 GOTO(err_thread, rc);
259 ping_evictor_start();
261 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
266 ptlrpc_unregister_service(mgs->mgs_service);
268 lproc_mgs_cleanup(obd);
269 obd_llog_finish(obd, 0);
271 /* No extra cleanup needed for llog_init_commit_thread() */
274 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
275 obd->obd_namespace = NULL;
277 fsfilt_put_ops(obd->obd_fsops);
279 server_put_mount(obd->obd_name, mnt);
284 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
286 struct mgs_obd *mgs = &obd->u.mgs;
291 case OBD_CLEANUP_EARLY:
293 case OBD_CLEANUP_EXPORTS:
295 ptlrpc_unregister_service(mgs->mgs_service);
296 mgs_cleanup_fsdb_list(obd);
297 rc = obd_llog_finish(obd, 0);
298 lproc_mgs_cleanup(obd);
305 * Performs cleanup procedures for passed \a obd given it is mgs obd.
307 static int mgs_cleanup(struct obd_device *obd)
309 struct mgs_obd *mgs = &obd->u.mgs;
312 if (mgs->mgs_sb == NULL)
317 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
320 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
321 obd->obd_namespace = NULL;
323 fsfilt_put_ops(obd->obd_fsops);
325 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
329 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
334 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
335 LDLM_FL_BLOCK_CONV))) {
336 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
337 struct lustre_handle lockh;
339 /* clear the bit before lock put */
340 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
342 ldlm_lock2handle(lock, &lockh);
343 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
346 RETURN(ldlm_completion_ast(lock, flags, cbdata));
349 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
354 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
355 LDLM_FL_BLOCK_CONV))) {
358 /* l_ast_data is used as a marker to avoid cancel ldlm lock
359 * twice. See LU-1259. */
360 lock_res_and_lock(lock);
361 fsdb = (struct fs_db *)lock->l_ast_data;
362 lock->l_ast_data = NULL;
363 unlock_res_and_lock(lock);
366 struct lustre_handle lockh;
368 mgs_ir_notify_complete(fsdb);
370 ldlm_lock2handle(lock, &lockh);
371 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
375 RETURN(ldlm_completion_ast(lock, flags, cbdata));
378 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb, int type)
380 ldlm_completion_callback cp = NULL;
381 struct lustre_handle lockh = { 0 };
382 struct ldlm_res_id res_id;
383 int flags = LDLM_FL_ATOMIC_CB;
387 LASSERT(fsdb->fsdb_name[0] != '\0');
388 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
392 case CONFIG_T_CONFIG:
393 cp = mgs_completion_ast_config;
394 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
397 case CONFIG_T_RECOVER:
398 cp = mgs_completion_ast_ir;
405 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
406 LDLM_PLAIN, NULL, LCK_EX, &flags,
407 ldlm_blocking_ast, cp, NULL,
408 fsdb, 0, NULL, &lockh);
409 if (rc != ELDLM_OK) {
410 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
411 le64_to_cpu(res_id.name[0]),
412 le64_to_cpu(res_id.name[1]), rc);
414 if (type == CONFIG_T_CONFIG)
415 cfs_clear_bit(FSDB_REVOKING_LOCK,
418 /* lock has been cancelled in completion_ast. */
427 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
432 rc = mgs_check_index(obd, mti);
434 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
435 "this MGS does not know about it, preventing "
436 "registration.\n", mti->mti_svname);
438 } else if (rc == -1) {
439 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
440 "disappeared! Regenerating all logs.\n",
442 mti->mti_flags |= LDD_F_WRITECONF;
445 /* Index is correctly marked as used */
447 /* If the logs don't contain the mti_nids then add
448 them as failover nids */
449 rc = mgs_check_failnid(obd, mti);
455 /* Ensure this is not a failover node that is connecting first*/
456 static int mgs_check_failover_reg(struct mgs_target_info *mti)
462 ptr = mti->mti_params;
463 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
464 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
465 for (i = 0; i < mti->mti_nid_count; i++) {
466 if (nid == mti->mti_nids[i]) {
467 LCONSOLE_WARN("Denying initial registra"
468 "tion attempt from nid %s"
469 ", specified as failover"
470 "\n",libcfs_nid2str(nid));
471 return -EADDRNOTAVAIL;
479 /* Called whenever a target starts up. Flags indicate first connect, etc. */
480 static int mgs_handle_target_reg(struct ptlrpc_request *req)
482 struct obd_device *obd = req->rq_export->exp_obd;
483 struct mgs_target_info *mti, *rep_mti;
489 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
491 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
493 opc = mti->mti_flags & LDD_F_OPC_MASK;
494 if (opc == LDD_F_OPC_READY) {
495 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
496 mti->mti_fsname, mti->mti_stripe_index);
497 rc = mgs_ir_update(obd, mti);
499 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
500 CERROR("Update IR return with %d(ignore and IR "
503 GOTO(out_nolock, rc);
506 /* Do not support unregistering right now. */
507 if (opc != LDD_F_OPC_REG)
508 GOTO(out_nolock, rc = -EINVAL);
510 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
511 mti->mti_fsname, mti->mti_stripe_index);
513 if (mti->mti_flags & LDD_F_NEED_INDEX)
514 mti->mti_flags |= LDD_F_WRITECONF;
516 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
518 /* We're just here as a startup ping. */
519 CDEBUG(D_MGS, "Server %s is running on %s\n",
520 mti->mti_svname, obd_export_nid2str(req->rq_export));
521 rc = mgs_check_target(obd, mti);
522 /* above will set appropriate mti flags */
524 /* Nothing wrong, or fatal error */
525 GOTO(out_nolock, rc);
527 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
528 && (rc = mgs_check_failover_reg(mti)))
529 GOTO(out_nolock, rc);
532 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
534 if (mti->mti_flags & LDD_F_WRITECONF) {
535 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
536 mti->mti_stripe_index == 0) {
537 rc = mgs_erase_logs(obd, mti->mti_fsname);
538 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
539 "request. All servers must be restarted "
540 "in order to regenerate the logs."
541 "\n", obd->obd_name, mti->mti_fsname);
542 } else if (mti->mti_flags &
543 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
544 rc = mgs_erase_log(obd, mti->mti_svname);
545 LCONSOLE_WARN("%s: Regenerating %s log by user "
547 obd->obd_name, mti->mti_svname);
549 mti->mti_flags |= LDD_F_UPDATE;
550 /* Erased logs means start from scratch. */
551 mti->mti_flags &= ~LDD_F_UPGRADE14;
554 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
556 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
557 GOTO(out_nolock, rc);
561 * Log writing contention is handled by the fsdb_mutex.
563 * It should be alright if someone was reading while we were
564 * updating the logs - if we revoke at the end they will just update
565 * from where they left off.
569 if (mti->mti_flags & LDD_F_UPGRADE14) {
570 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
572 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
576 /* We're good to go */
577 mti->mti_flags |= LDD_F_UPDATE;
581 if (mti->mti_flags & LDD_F_UPDATE) {
582 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
583 mti->mti_stripe_index);
585 /* create or update the target log
586 and update the client/mdt logs */
587 rc = mgs_write_log_target(obd, mti, fsdb);
589 CERROR("Failed to write %s log (%d)\n",
590 mti->mti_svname, rc);
594 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
595 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
597 mti->mti_flags |= LDD_F_REWRITE_LDD;
601 mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
604 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
605 mti->mti_stripe_index, rc);
608 /* we need an error flag to tell the target what's going on,
609 * instead of just doing it by error code only. */
610 mti->mti_flags |= LDD_F_ERROR;
612 rc = req_capsule_server_pack(&req->rq_pill);
616 /* send back the whole mti in the reply */
617 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
620 /* Flush logs to disk */
621 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
625 static int mgs_set_info_rpc(struct ptlrpc_request *req)
627 struct obd_device *obd = req->rq_export->exp_obd;
628 struct mgs_send_param *msp, *rep_msp;
630 struct lustre_cfg_bufs bufs;
631 struct lustre_cfg *lcfg;
632 char fsname[MTI_NAME_MAXLEN];
635 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
638 /* Construct lustre_cfg structure to pass to function mgs_setparam */
639 lustre_cfg_bufs_reset(&bufs, NULL);
640 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
641 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
642 rc = mgs_setparam(obd, lcfg, fsname);
644 CERROR("Error %d in setting the parameter %s for fs %s\n",
645 rc, msp->mgs_param, fsname);
649 lustre_cfg_free(lcfg);
651 rc = req_capsule_server_pack(&req->rq_pill);
653 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
659 static int mgs_config_read(struct ptlrpc_request *req)
661 struct mgs_config_body *body;
665 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
669 switch (body->mcb_type) {
670 case CONFIG_T_RECOVER:
671 rc = mgs_get_ir_logs(req);
674 case CONFIG_T_CONFIG:
687 * similar as in ost_connect_check_sptlrpc()
689 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
691 struct obd_export *exp = req->rq_export;
692 struct obd_device *obd = exp->exp_obd;
694 struct sptlrpc_flavor flvr;
697 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
698 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
702 cfs_mutex_lock(&fsdb->fsdb_mutex);
703 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
704 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
707 /* by defualt allow any flavors */
708 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
710 cfs_mutex_unlock(&fsdb->fsdb_mutex);
712 cfs_spin_lock(&exp->exp_lock);
714 exp->exp_sp_peer = req->rq_sp_from;
715 exp->exp_flvr = flvr;
717 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
718 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
719 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
720 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
721 libcfs_nid2str(req->rq_peer.nid));
725 cfs_spin_unlock(&exp->exp_lock);
727 if (exp->exp_sp_peer != req->rq_sp_from) {
728 CERROR("RPC source %s doesn't match %s\n",
729 sptlrpc_part2name(req->rq_sp_from),
730 sptlrpc_part2name(exp->exp_sp_peer));
733 rc = sptlrpc_target_export_check(exp, req);
740 /* Called whenever a target cleans up. */
741 /* XXX - Currently unused */
742 static int mgs_handle_target_del(struct ptlrpc_request *req)
745 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
749 /* XXX - Currently unused */
750 static int mgs_handle_exception(struct ptlrpc_request *req)
753 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
758 * For old clients there is no direct way of knowing which filesystems
759 * a client is operating at the MGS side. But we need to pick up those
760 * clients so that the MGS can mark the corresponding filesystem as
761 * non-IR capable because old clients are not ready to be notified.
763 * This is why we have this _hack_ function. We detect the filesystem's
764 * name by hacking llog operation which is currently used by the clients
765 * to fetch configuration logs. At present this is fine because this is
766 * the ONLY llog operation between mgc and the MGS.
768 * If extra llog operation is going to be added, this function needs fixing.
770 * If releases prior than 2.0 are not supported, we can remove this function.
772 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
779 /* XXX: We suppose that llog at mgs is only used for
780 * fetching file system log */
781 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
782 if (logname == NULL) {
783 CERROR("No logname, is llog on MGS used for something else?\n");
787 ptr = strchr(logname, '-');
788 rc = (int)(ptr - logname);
789 if (ptr == NULL || rc >= sizeof(fsname)) {
790 CERROR("Invalid logname received: %s\n", logname);
794 strncpy(fsname, logname, rc);
796 rc = mgs_fsc_attach(req->rq_export, fsname);
797 if (rc < 0 && rc != -EEXIST)
798 CERROR("add fs client %s returns %d\n", fsname, rc);
803 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
804 int mgs_handle(struct ptlrpc_request *req)
806 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
810 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
811 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
812 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
815 LASSERT(current->journal_info == NULL);
816 opc = lustre_msg_get_opc(req->rq_reqmsg);
818 if (opc == SEC_CTX_INIT ||
819 opc == SEC_CTX_INIT_CONT ||
823 if (opc != MGS_CONNECT) {
824 if (!class_connected_export(req->rq_export)) {
825 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
826 req->rq_status = -ENOTCONN;
827 GOTO(out, rc = -ENOTCONN);
833 DEBUG_REQ(D_MGS, req, "connect");
834 /* MGS and MDS have same request format for connect */
835 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
836 rc = target_handle_connect(req);
838 rc = mgs_connect_check_sptlrpc(req);
840 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
841 /* Make clients trying to reconnect after a MGS restart
842 happy; also requires obd_replayable */
843 lustre_msg_add_op_flags(req->rq_repmsg,
844 MSG_CONNECT_RECONNECT);
847 DEBUG_REQ(D_MGS, req, "disconnect");
848 /* MGS and MDS have same request format for disconnect */
849 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
850 rc = target_handle_disconnect(req);
851 req->rq_status = rc; /* superfluous? */
854 DEBUG_REQ(D_MGS, req, "exception");
855 rc = mgs_handle_exception(req);
858 DEBUG_REQ(D_MGS, req, "target add");
859 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
860 rc = mgs_handle_target_reg(req);
863 DEBUG_REQ(D_MGS, req, "target del");
864 rc = mgs_handle_target_del(req);
867 DEBUG_REQ(D_MGS, req, "set_info");
868 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
869 rc = mgs_set_info_rpc(req);
871 case MGS_CONFIG_READ:
872 DEBUG_REQ(D_MGS, req, "read config");
873 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
874 rc = mgs_config_read(req);
877 DEBUG_REQ(D_MGS, req, "enqueue");
878 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
879 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
880 ldlm_server_blocking_ast, NULL);
882 case LDLM_BL_CALLBACK:
883 case LDLM_CP_CALLBACK:
884 DEBUG_REQ(D_MGS, req, "callback");
885 CERROR("callbacks should not happen on MGS\n");
890 DEBUG_REQ(D_INFO, req, "ping");
891 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
892 rc = target_handle_ping(req);
895 DEBUG_REQ(D_MGS, req, "log cancel");
896 rc = -ENOTSUPP; /* la la la */
899 case LLOG_ORIGIN_HANDLE_CREATE:
900 DEBUG_REQ(D_MGS, req, "llog_init");
901 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
902 rc = llog_origin_handle_create(req);
904 (void)mgs_handle_fslog_hack(req);
906 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
907 DEBUG_REQ(D_MGS, req, "llog next block");
908 req_capsule_set(&req->rq_pill,
909 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
910 rc = llog_origin_handle_next_block(req);
912 case LLOG_ORIGIN_HANDLE_READ_HEADER:
913 DEBUG_REQ(D_MGS, req, "llog read header");
914 req_capsule_set(&req->rq_pill,
915 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
916 rc = llog_origin_handle_read_header(req);
918 case LLOG_ORIGIN_HANDLE_CLOSE:
919 DEBUG_REQ(D_MGS, req, "llog close");
920 rc = llog_origin_handle_close(req);
923 DEBUG_REQ(D_MGS, req, "llog catinfo");
924 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
925 rc = llog_catinfo(req);
928 req->rq_status = -ENOTSUPP;
929 rc = ptlrpc_error(req);
933 LASSERT(current->journal_info == NULL);
936 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
939 target_send_reply(req, rc, fail);
943 static inline int mgs_init_export(struct obd_export *exp)
945 struct mgs_export_data *data = &exp->u.eu_mgs_data;
947 /* init mgs_export_data for fsc */
948 cfs_spin_lock_init(&data->med_lock);
949 CFS_INIT_LIST_HEAD(&data->med_clients);
951 cfs_spin_lock(&exp->exp_lock);
952 exp->exp_connecting = 1;
953 cfs_spin_unlock(&exp->exp_lock);
955 /* self-export doesn't need client data and ldlm initialization */
956 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
957 &exp->exp_client_uuid)))
959 return ldlm_init_export(exp);
962 static inline int mgs_destroy_export(struct obd_export *exp)
966 target_destroy_export(exp);
967 mgs_client_free(exp);
969 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
970 &exp->exp_client_uuid)))
973 ldlm_destroy_export(exp);
978 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
983 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
991 strcpy(poolname, ptr);
996 static int mgs_iocontrol_pool(struct obd_device *obd,
997 struct obd_ioctl_data *data)
1000 struct lustre_cfg *lcfg = NULL;
1001 struct llog_rec_hdr rec;
1002 char *fsname = NULL;
1003 char *poolname = NULL;
1006 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
1010 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1011 if (poolname == NULL) {
1015 rec.lrh_len = llog_data_len(data->ioc_plen1);
1017 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1018 rec.lrh_type = OBD_CFG_REC;
1020 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1025 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1030 OBD_ALLOC(lcfg, data->ioc_plen1);
1032 GOTO(out_pool, rc = -ENOMEM);
1034 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1035 GOTO(out_pool, rc = -EFAULT);
1037 if (lcfg->lcfg_bufcount < 2) {
1038 GOTO(out_pool, rc = -EFAULT);
1041 /* first arg is always <fsname>.<poolname> */
1042 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
1045 switch (lcfg->lcfg_command) {
1046 case LCFG_POOL_NEW: {
1047 if (lcfg->lcfg_bufcount != 2)
1049 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
1053 case LCFG_POOL_ADD: {
1054 if (lcfg->lcfg_bufcount != 3)
1056 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1057 lustre_cfg_string(lcfg, 2));
1060 case LCFG_POOL_REM: {
1061 if (lcfg->lcfg_bufcount != 3)
1063 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1064 lustre_cfg_string(lcfg, 2));
1067 case LCFG_POOL_DEL: {
1068 if (lcfg->lcfg_bufcount != 2)
1070 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1081 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1082 rc, lcfg->lcfg_command, fsname, poolname);
1088 OBD_FREE(lcfg, data->ioc_plen1);
1091 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1093 if (poolname != NULL)
1094 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1099 /* from mdt_iocontrol */
1100 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1101 void *karg, void *uarg)
1103 struct obd_device *obd = exp->exp_obd;
1104 struct obd_ioctl_data *data = karg;
1105 struct lvfs_run_ctxt saved;
1109 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1113 case OBD_IOC_PARAM: {
1114 struct lustre_cfg *lcfg;
1115 struct llog_rec_hdr rec;
1116 char fsname[MTI_NAME_MAXLEN];
1118 rec.lrh_len = llog_data_len(data->ioc_plen1);
1120 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1121 rec.lrh_type = OBD_CFG_REC;
1123 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1127 OBD_ALLOC(lcfg, data->ioc_plen1);
1130 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1131 GOTO(out_free, rc = -EFAULT);
1133 if (lcfg->lcfg_bufcount < 1)
1134 GOTO(out_free, rc = -EINVAL);
1136 rc = mgs_setparam(obd, lcfg, fsname);
1138 CERROR("setparam err %d\n", rc);
1142 OBD_FREE(lcfg, data->ioc_plen1);
1146 case OBD_IOC_POOL: {
1147 RETURN(mgs_iocontrol_pool(obd, data));
1150 case OBD_IOC_DUMP_LOG: {
1151 struct llog_ctxt *ctxt;
1152 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1153 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1154 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1155 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1156 llog_ctxt_put(ctxt);
1161 case OBD_IOC_LLOG_CHECK:
1162 case OBD_IOC_LLOG_INFO:
1163 case OBD_IOC_LLOG_PRINT: {
1164 struct llog_ctxt *ctxt;
1165 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1167 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1168 rc = llog_ioctl(ctxt, cmd, data);
1169 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1170 llog_ctxt_put(ctxt);
1176 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1182 /* use obd ops to offer management infrastructure */
1183 static struct obd_ops mgs_obd_ops = {
1184 .o_owner = THIS_MODULE,
1185 .o_connect = mgs_connect,
1186 .o_reconnect = mgs_reconnect,
1187 .o_disconnect = mgs_disconnect,
1188 .o_setup = mgs_setup,
1189 .o_precleanup = mgs_precleanup,
1190 .o_cleanup = mgs_cleanup,
1191 .o_init_export = mgs_init_export,
1192 .o_destroy_export = mgs_destroy_export,
1193 .o_iocontrol = mgs_iocontrol,
1194 .o_llog_init = mgs_llog_init,
1195 .o_llog_finish = mgs_llog_finish
1198 static int __init mgs_init(void)
1200 struct lprocfs_static_vars lvars;
1202 lprocfs_mgs_init_vars(&lvars);
1203 class_register_type(&mgs_obd_ops, NULL,
1204 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1209 static void /*__exit*/ mgs_exit(void)
1211 class_unregister_type(LUSTRE_MGS_NAME);
1214 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1215 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1216 MODULE_LICENSE("GPL");
1218 module_init(mgs_init);
1219 module_exit(mgs_exit);