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 class_export_get(exp);
133 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
135 rc = server_disconnect_export(exp);
137 class_export_put(exp);
141 static int mgs_cleanup(struct obd_device *obd);
142 static int mgs_handle(struct ptlrpc_request *req);
144 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
145 struct obd_device *tgt, int *index)
150 LASSERT(olg == &obd->obd_olg);
151 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
156 static int mgs_llog_finish(struct obd_device *obd, int count)
158 struct llog_ctxt *ctxt;
162 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
164 rc = llog_cleanup(ctxt);
169 /* Start the MGS obd */
170 static int mgs_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
172 struct lprocfs_static_vars lvars;
173 struct mgs_obd *mgs = &obd->u.mgs;
174 struct lustre_mount_info *lmi;
175 struct lustre_sb_info *lsi;
176 struct vfsmount *mnt;
180 CDEBUG(D_CONFIG, "Starting MGS\n");
183 lmi = server_get_mount(obd->obd_name);
185 RETURN(rc = -EINVAL);
188 lsi = s2lsi(lmi->lmi_sb);
189 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
190 if (IS_ERR(obd->obd_fsops))
191 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
193 if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) {
194 CERROR("%s: Underlying device is marked as read-only. "
195 "Setup failed\n", obd->obd_name);
196 GOTO(err_ops, rc = -EROFS);
199 obd->u.obt.obt_magic = OBT_MAGIC;
200 obd->u.obt.obt_instance = 0;
202 /* namespace for mgs llog */
203 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
204 LDLM_NAMESPACE_SERVER,
205 LDLM_NAMESPACE_MODEST,
207 if (obd->obd_namespace == NULL)
208 GOTO(err_ops, rc = -ENOMEM);
211 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
212 "mgs_ldlm_client", &obd->obd_ldlm_client);
214 rc = mgs_fs_setup(obd, mnt);
216 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
221 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
225 /* No recovery for MGC's */
226 obd->obd_replayable = 0;
228 /* Internal mgs setup */
229 mgs_init_fsdb_list(obd);
230 cfs_sema_init(&mgs->mgs_sem, 1);
231 mgs->mgs_start_time = cfs_time_current_sec();
234 lprocfs_mgs_init_vars(&lvars);
235 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
236 lproc_mgs_setup(obd);
237 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
242 /* Start the service threads */
244 ptlrpc_init_svc(MGS_NBUFS, MGS_BUFSIZE, MGS_MAXREQSIZE,
245 MGS_MAXREPSIZE, MGS_REQUEST_PORTAL,
247 mgs_handle, LUSTRE_MGS_NAME,
248 obd->obd_proc_entry, target_print_req,
249 MGS_THREADS_AUTO_MIN, MGS_THREADS_AUTO_MAX,
250 "ll_mgs", LCT_MD_THREAD, NULL);
252 if (!mgs->mgs_service) {
253 CERROR("failed to start service\n");
254 GOTO(err_llog, rc = -ENOMEM);
257 rc = ptlrpc_start_threads(mgs->mgs_service);
259 GOTO(err_thread, rc);
261 ping_evictor_start();
263 LCONSOLE_INFO("MGS %s started\n", obd->obd_name);
268 ptlrpc_unregister_service(mgs->mgs_service);
270 lproc_mgs_cleanup(obd);
271 obd_llog_finish(obd, 0);
273 /* No extra cleanup needed for llog_init_commit_thread() */
276 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
277 obd->obd_namespace = NULL;
279 fsfilt_put_ops(obd->obd_fsops);
281 server_put_mount(obd->obd_name, mnt);
286 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
292 case OBD_CLEANUP_EARLY:
293 case OBD_CLEANUP_EXPORTS:
294 rc = obd_llog_finish(obd, 0);
301 * Performs cleanup procedures for passed \a obd given it is mgs obd.
303 static int mgs_cleanup(struct obd_device *obd)
305 struct mgs_obd *mgs = &obd->u.mgs;
308 if (mgs->mgs_sb == NULL)
313 ptlrpc_unregister_service(mgs->mgs_service);
315 mgs_cleanup_fsdb_list(obd);
316 lproc_mgs_cleanup(obd);
319 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
322 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
323 obd->obd_namespace = NULL;
325 fsfilt_put_ops(obd->obd_fsops);
327 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
331 /* similar to filter_prepare_destroy */
332 int mgs_get_lock(struct obd_device *obd, struct ldlm_res_id *res,
333 struct lustre_handle *lockh)
338 rc = ldlm_cli_enqueue_local(obd->obd_namespace, res,
339 LDLM_PLAIN, NULL, LCK_EX,
340 &flags, ldlm_blocking_ast,
341 ldlm_completion_ast, NULL,
342 NULL, 0, NULL, lockh);
344 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
345 le64_to_cpu(res->name[0]), le64_to_cpu(res->name[1]),
351 int mgs_put_lock(struct lustre_handle *lockh)
354 ldlm_lock_decref_and_cancel(lockh, LCK_EX);
358 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb)
360 struct lustre_handle lockh;
361 struct ldlm_res_id res_id;
366 LASSERT(fsdb->fsdb_name[0] != '\0');
367 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, CONFIG_T_CONFIG);
370 bit = FSDB_REVOKING_LOCK;
371 if (!rc && cfs_test_and_set_bit(bit, &fsdb->fsdb_flags) == 0) {
372 lockrc = mgs_get_lock(obd, &res_id, &lockh);
373 /* clear the bit before lock put */
374 cfs_clear_bit(bit, &fsdb->fsdb_flags);
376 if (lockrc != ELDLM_OK)
377 CERROR("lock error %d for fs %s\n",
378 lockrc, fsdb->fsdb_name);
380 mgs_put_lock(&lockh);
387 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
392 rc = mgs_check_index(obd, mti);
394 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
395 "this MGS does not know about it, preventing "
396 "registration.\n", mti->mti_svname);
398 } else if (rc == -1) {
399 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
400 "disappeared! Regenerating all logs.\n",
402 mti->mti_flags |= LDD_F_WRITECONF;
405 /* Index is correctly marked as used */
407 /* If the logs don't contain the mti_nids then add
408 them as failover nids */
409 rc = mgs_check_failnid(obd, mti);
415 /* Ensure this is not a failover node that is connecting first*/
416 static int mgs_check_failover_reg(struct mgs_target_info *mti)
422 ptr = mti->mti_params;
423 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
424 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
425 for (i = 0; i < mti->mti_nid_count; i++) {
426 if (nid == mti->mti_nids[i]) {
427 LCONSOLE_WARN("Denying initial registra"
428 "tion attempt from nid %s"
429 ", specified as failover"
430 "\n",libcfs_nid2str(nid));
431 return -EADDRNOTAVAIL;
439 /* Called whenever a target starts up. Flags indicate first connect, etc. */
440 static int mgs_handle_target_reg(struct ptlrpc_request *req)
442 struct obd_device *obd = req->rq_export->exp_obd;
443 struct mgs_target_info *mti, *rep_mti;
449 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
451 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
453 opc = mti->mti_flags & LDD_F_OPC_MASK;
454 if (opc == LDD_F_OPC_READY) {
455 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
456 mti->mti_fsname, mti->mti_stripe_index);
457 rc = mgs_ir_update(obd, mti);
459 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
460 CERROR("Update IR return with %d(ignore and IR "
463 GOTO(out_nolock, rc);
466 /* Do not support unregistering right now. */
467 if (opc != LDD_F_OPC_REG)
468 GOTO(out_nolock, rc = -EINVAL);
470 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
471 mti->mti_fsname, mti->mti_stripe_index);
473 if (mti->mti_flags & LDD_F_NEED_INDEX)
474 mti->mti_flags |= LDD_F_WRITECONF;
476 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
478 /* We're just here as a startup ping. */
479 CDEBUG(D_MGS, "Server %s is running on %s\n",
480 mti->mti_svname, obd_export_nid2str(req->rq_export));
481 rc = mgs_check_target(obd, mti);
482 /* above will set appropriate mti flags */
484 /* Nothing wrong, or fatal error */
485 GOTO(out_nolock, rc);
487 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
488 && (rc = mgs_check_failover_reg(mti)))
489 GOTO(out_nolock, rc);
492 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
494 if (mti->mti_flags & LDD_F_WRITECONF) {
495 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
496 mti->mti_stripe_index == 0) {
497 rc = mgs_erase_logs(obd, mti->mti_fsname);
498 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
499 "request. All servers must be restarted "
500 "in order to regenerate the logs."
501 "\n", obd->obd_name, mti->mti_fsname);
502 } else if (mti->mti_flags &
503 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
504 rc = mgs_erase_log(obd, mti->mti_svname);
505 LCONSOLE_WARN("%s: Regenerating %s log by user "
507 obd->obd_name, mti->mti_svname);
509 mti->mti_flags |= LDD_F_UPDATE;
510 /* Erased logs means start from scratch. */
511 mti->mti_flags &= ~LDD_F_UPGRADE14;
514 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
516 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
517 GOTO(out_nolock, rc);
521 * Log writing contention is handled by the fsdb_sem.
523 * It should be alright if someone was reading while we were
524 * updating the logs - if we revoke at the end they will just update
525 * from where they left off.
529 if (mti->mti_flags & LDD_F_UPGRADE14) {
530 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
532 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
536 /* We're good to go */
537 mti->mti_flags |= LDD_F_UPDATE;
541 if (mti->mti_flags & LDD_F_UPDATE) {
542 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
543 mti->mti_stripe_index);
545 /* create or update the target log
546 and update the client/mdt logs */
547 rc = mgs_write_log_target(obd, mti, fsdb);
549 CERROR("Failed to write %s log (%d)\n",
550 mti->mti_svname, rc);
554 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
555 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
557 mti->mti_flags |= LDD_F_REWRITE_LDD;
561 mgs_revoke_lock(obd, fsdb);
564 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
565 mti->mti_stripe_index, rc);
568 /* we need an error flag to tell the target what's going on,
569 * instead of just doing it by error code only. */
570 mti->mti_flags |= LDD_F_ERROR;
572 rc = req_capsule_server_pack(&req->rq_pill);
576 /* send back the whole mti in the reply */
577 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
580 /* Flush logs to disk */
581 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
585 static int mgs_set_info_rpc(struct ptlrpc_request *req)
587 struct obd_device *obd = req->rq_export->exp_obd;
588 struct mgs_send_param *msp, *rep_msp;
590 struct lustre_cfg_bufs bufs;
591 struct lustre_cfg *lcfg;
592 char fsname[MTI_NAME_MAXLEN];
595 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
598 /* Construct lustre_cfg structure to pass to function mgs_setparam */
599 lustre_cfg_bufs_reset(&bufs, NULL);
600 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
601 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
602 rc = mgs_setparam(obd, lcfg, fsname);
604 CERROR("Error %d in setting the parameter %s for fs %s\n",
605 rc, msp->mgs_param, fsname);
609 lustre_cfg_free(lcfg);
611 rc = req_capsule_server_pack(&req->rq_pill);
613 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
619 static int mgs_config_read(struct ptlrpc_request *req)
621 struct mgs_config_body *body;
625 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
629 switch (body->mcb_type) {
630 case CONFIG_T_RECOVER:
631 rc = mgs_get_ir_logs(req);
634 case CONFIG_T_CONFIG:
647 * similar as in ost_connect_check_sptlrpc()
649 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
651 struct obd_export *exp = req->rq_export;
652 struct obd_device *obd = exp->exp_obd;
654 struct sptlrpc_flavor flvr;
657 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
658 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
662 cfs_down(&fsdb->fsdb_sem);
663 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
664 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
667 /* by defualt allow any flavors */
668 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
670 cfs_up(&fsdb->fsdb_sem);
672 cfs_spin_lock(&exp->exp_lock);
674 exp->exp_sp_peer = req->rq_sp_from;
675 exp->exp_flvr = flvr;
677 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
678 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
679 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
680 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
681 libcfs_nid2str(req->rq_peer.nid));
685 cfs_spin_unlock(&exp->exp_lock);
687 if (exp->exp_sp_peer != req->rq_sp_from) {
688 CERROR("RPC source %s doesn't match %s\n",
689 sptlrpc_part2name(req->rq_sp_from),
690 sptlrpc_part2name(exp->exp_sp_peer));
693 rc = sptlrpc_target_export_check(exp, req);
700 /* Called whenever a target cleans up. */
701 /* XXX - Currently unused */
702 static int mgs_handle_target_del(struct ptlrpc_request *req)
705 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
709 /* XXX - Currently unused */
710 static int mgs_handle_exception(struct ptlrpc_request *req)
713 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
717 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
718 int mgs_handle(struct ptlrpc_request *req)
720 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
724 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
725 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
726 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
729 LASSERT(current->journal_info == NULL);
730 opc = lustre_msg_get_opc(req->rq_reqmsg);
732 if (opc == SEC_CTX_INIT ||
733 opc == SEC_CTX_INIT_CONT ||
737 if (opc != MGS_CONNECT) {
738 if (!class_connected_export(req->rq_export)) {
739 CERROR("lustre_mgs: operation %d on unconnected MGS\n",
741 req->rq_status = -ENOTCONN;
742 GOTO(out, rc = -ENOTCONN);
748 DEBUG_REQ(D_MGS, req, "connect");
749 /* MGS and MDS have same request format for connect */
750 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
751 rc = target_handle_connect(req);
753 rc = mgs_connect_check_sptlrpc(req);
755 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
756 /* Make clients trying to reconnect after a MGS restart
757 happy; also requires obd_replayable */
758 lustre_msg_add_op_flags(req->rq_repmsg,
759 MSG_CONNECT_RECONNECT);
762 DEBUG_REQ(D_MGS, req, "disconnect");
763 /* MGS and MDS have same request format for disconnect */
764 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
765 rc = target_handle_disconnect(req);
766 req->rq_status = rc; /* superfluous? */
769 DEBUG_REQ(D_MGS, req, "exception");
770 rc = mgs_handle_exception(req);
773 DEBUG_REQ(D_MGS, req, "target add");
774 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
775 rc = mgs_handle_target_reg(req);
778 DEBUG_REQ(D_MGS, req, "target del");
779 rc = mgs_handle_target_del(req);
782 DEBUG_REQ(D_MGS, req, "set_info");
783 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
784 rc = mgs_set_info_rpc(req);
786 case MGS_CONFIG_READ:
787 DEBUG_REQ(D_MGS, req, "read config");
788 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
789 rc = mgs_config_read(req);
792 DEBUG_REQ(D_MGS, req, "enqueue");
793 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
794 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
795 ldlm_server_blocking_ast, NULL);
797 case LDLM_BL_CALLBACK:
798 case LDLM_CP_CALLBACK:
799 DEBUG_REQ(D_MGS, req, "callback");
800 CERROR("callbacks should not happen on MGS\n");
805 DEBUG_REQ(D_INFO, req, "ping");
806 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
807 rc = target_handle_ping(req);
810 DEBUG_REQ(D_MGS, req, "log cancel");
811 rc = -ENOTSUPP; /* la la la */
814 case LLOG_ORIGIN_HANDLE_CREATE:
815 DEBUG_REQ(D_MGS, req, "llog_init");
816 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
817 rc = llog_origin_handle_create(req);
819 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
820 DEBUG_REQ(D_MGS, req, "llog next block");
821 req_capsule_set(&req->rq_pill,
822 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
823 rc = llog_origin_handle_next_block(req);
825 case LLOG_ORIGIN_HANDLE_READ_HEADER:
826 DEBUG_REQ(D_MGS, req, "llog read header");
827 req_capsule_set(&req->rq_pill,
828 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
829 rc = llog_origin_handle_read_header(req);
831 case LLOG_ORIGIN_HANDLE_CLOSE:
832 DEBUG_REQ(D_MGS, req, "llog close");
833 rc = llog_origin_handle_close(req);
836 DEBUG_REQ(D_MGS, req, "llog catinfo");
837 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
838 rc = llog_catinfo(req);
841 req->rq_status = -ENOTSUPP;
842 rc = ptlrpc_error(req);
846 LASSERT(current->journal_info == NULL);
849 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
852 target_send_reply(req, rc, fail);
856 static inline int mgs_init_export(struct obd_export *exp)
858 cfs_spin_lock(&exp->exp_lock);
859 exp->exp_connecting = 1;
860 cfs_spin_unlock(&exp->exp_lock);
862 return ldlm_init_export(exp);
865 static inline int mgs_destroy_export(struct obd_export *exp)
869 target_destroy_export(exp);
870 mgs_client_free(exp);
871 ldlm_destroy_export(exp);
876 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
881 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
889 strcpy(poolname, ptr);
894 static int mgs_iocontrol_pool(struct obd_device *obd,
895 struct obd_ioctl_data *data)
898 struct lustre_cfg *lcfg = NULL;
899 struct llog_rec_hdr rec;
901 char *poolname = NULL;
904 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
908 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
909 if (poolname == NULL) {
913 rec.lrh_len = llog_data_len(data->ioc_plen1);
915 if (data->ioc_type == LUSTRE_CFG_TYPE) {
916 rec.lrh_type = OBD_CFG_REC;
918 CERROR("unknown cfg record type:%d \n", data->ioc_type);
923 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
928 OBD_ALLOC(lcfg, data->ioc_plen1);
930 GOTO(out_pool, rc = -ENOMEM);
932 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
933 GOTO(out_pool, rc = -EFAULT);
935 if (lcfg->lcfg_bufcount < 2) {
936 GOTO(out_pool, rc = -EFAULT);
939 /* first arg is always <fsname>.<poolname> */
940 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
943 switch (lcfg->lcfg_command) {
944 case LCFG_POOL_NEW: {
945 if (lcfg->lcfg_bufcount != 2)
947 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
951 case LCFG_POOL_ADD: {
952 if (lcfg->lcfg_bufcount != 3)
954 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
955 lustre_cfg_string(lcfg, 2));
958 case LCFG_POOL_REM: {
959 if (lcfg->lcfg_bufcount != 3)
961 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
962 lustre_cfg_string(lcfg, 2));
965 case LCFG_POOL_DEL: {
966 if (lcfg->lcfg_bufcount != 2)
968 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
979 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
980 rc, lcfg->lcfg_command, fsname, poolname);
986 OBD_FREE(lcfg, data->ioc_plen1);
989 OBD_FREE(fsname, MTI_NAME_MAXLEN);
991 if (poolname != NULL)
992 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
997 /* from mdt_iocontrol */
998 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
999 void *karg, void *uarg)
1001 struct obd_device *obd = exp->exp_obd;
1002 struct obd_ioctl_data *data = karg;
1003 struct lvfs_run_ctxt saved;
1007 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1011 case OBD_IOC_PARAM: {
1012 struct lustre_cfg *lcfg;
1013 struct llog_rec_hdr rec;
1014 char fsname[MTI_NAME_MAXLEN];
1016 rec.lrh_len = llog_data_len(data->ioc_plen1);
1018 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1019 rec.lrh_type = OBD_CFG_REC;
1021 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1025 OBD_ALLOC(lcfg, data->ioc_plen1);
1028 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1029 GOTO(out_free, rc = -EFAULT);
1031 if (lcfg->lcfg_bufcount < 1)
1032 GOTO(out_free, rc = -EINVAL);
1034 rc = mgs_setparam(obd, lcfg, fsname);
1036 CERROR("setparam err %d\n", rc);
1040 OBD_FREE(lcfg, data->ioc_plen1);
1044 case OBD_IOC_POOL: {
1045 RETURN(mgs_iocontrol_pool(obd, data));
1048 case OBD_IOC_DUMP_LOG: {
1049 struct llog_ctxt *ctxt;
1050 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1051 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1052 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1053 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1054 llog_ctxt_put(ctxt);
1059 case OBD_IOC_LLOG_CHECK:
1060 case OBD_IOC_LLOG_INFO:
1061 case OBD_IOC_LLOG_PRINT: {
1062 struct llog_ctxt *ctxt;
1063 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1065 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1066 rc = llog_ioctl(ctxt, cmd, data);
1067 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1068 llog_ctxt_put(ctxt);
1074 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1080 /* use obd ops to offer management infrastructure */
1081 static struct obd_ops mgs_obd_ops = {
1082 .o_owner = THIS_MODULE,
1083 .o_connect = mgs_connect,
1084 .o_reconnect = mgs_reconnect,
1085 .o_disconnect = mgs_disconnect,
1086 .o_setup = mgs_setup,
1087 .o_precleanup = mgs_precleanup,
1088 .o_cleanup = mgs_cleanup,
1089 .o_init_export = mgs_init_export,
1090 .o_destroy_export = mgs_destroy_export,
1091 .o_iocontrol = mgs_iocontrol,
1092 .o_llog_init = mgs_llog_init,
1093 .o_llog_finish = mgs_llog_finish
1096 static int __init mgs_init(void)
1098 struct lprocfs_static_vars lvars;
1100 lprocfs_mgs_init_vars(&lvars);
1101 class_register_type(&mgs_obd_ops, NULL,
1102 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1107 static void /*__exit*/ mgs_exit(void)
1109 class_unregister_type(LUSTRE_MGS_NAME);
1112 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1113 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1114 MODULE_LICENSE("GPL");
1116 module_init(mgs_init);
1117 module_exit(mgs_exit);