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 CDEBUG(D_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)
289 struct mgs_obd *mgs = &obd->u.mgs;
294 case OBD_CLEANUP_EARLY:
296 case OBD_CLEANUP_EXPORTS:
298 ptlrpc_unregister_service(mgs->mgs_service);
299 mgs_cleanup_fsdb_list(obd);
300 rc = obd_llog_finish(obd, 0);
301 lproc_mgs_cleanup(obd);
308 * Performs cleanup procedures for passed \a obd given it is mgs obd.
310 static int mgs_cleanup(struct obd_device *obd)
312 struct mgs_obd *mgs = &obd->u.mgs;
315 if (mgs->mgs_sb == NULL)
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 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
337 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
338 LDLM_FL_BLOCK_CONV))) {
339 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
340 struct lustre_handle lockh;
342 /* clear the bit before lock put */
343 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
345 ldlm_lock2handle(lock, &lockh);
346 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
349 RETURN(ldlm_completion_ast(lock, flags, cbdata));
352 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
357 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
358 LDLM_FL_BLOCK_CONV))) {
359 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
360 struct lustre_handle lockh;
362 mgs_ir_notify_complete(fsdb);
364 ldlm_lock2handle(lock, &lockh);
365 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
368 RETURN(ldlm_completion_ast(lock, flags, cbdata));
371 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb, int type)
373 ldlm_completion_callback cp = NULL;
374 struct lustre_handle lockh = { 0 };
375 struct ldlm_res_id res_id;
376 int flags = LDLM_FL_ATOMIC_CB;
380 LASSERT(fsdb->fsdb_name[0] != '\0');
381 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
385 case CONFIG_T_CONFIG:
386 cp = mgs_completion_ast_config;
387 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
390 case CONFIG_T_RECOVER:
391 cp = mgs_completion_ast_ir;
398 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
399 LDLM_PLAIN, NULL, LCK_EX, &flags,
400 ldlm_blocking_ast, cp, NULL,
401 fsdb, 0, NULL, &lockh);
402 if (rc != ELDLM_OK) {
403 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
404 le64_to_cpu(res_id.name[0]),
405 le64_to_cpu(res_id.name[1]), rc);
407 if (type == CONFIG_T_CONFIG)
408 cfs_clear_bit(FSDB_REVOKING_LOCK,
411 /* lock has been cancelled in completion_ast. */
420 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
425 rc = mgs_check_index(obd, mti);
427 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
428 "this MGS does not know about it, preventing "
429 "registration.\n", mti->mti_svname);
431 } else if (rc == -1) {
432 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
433 "disappeared! Regenerating all logs.\n",
435 mti->mti_flags |= LDD_F_WRITECONF;
438 /* Index is correctly marked as used */
440 /* If the logs don't contain the mti_nids then add
441 them as failover nids */
442 rc = mgs_check_failnid(obd, mti);
448 /* Ensure this is not a failover node that is connecting first*/
449 static int mgs_check_failover_reg(struct mgs_target_info *mti)
455 ptr = mti->mti_params;
456 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
457 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
458 for (i = 0; i < mti->mti_nid_count; i++) {
459 if (nid == mti->mti_nids[i]) {
460 LCONSOLE_WARN("Denying initial registra"
461 "tion attempt from nid %s"
462 ", specified as failover"
463 "\n",libcfs_nid2str(nid));
464 return -EADDRNOTAVAIL;
472 /* Called whenever a target starts up. Flags indicate first connect, etc. */
473 static int mgs_handle_target_reg(struct ptlrpc_request *req)
475 struct obd_device *obd = req->rq_export->exp_obd;
476 struct mgs_target_info *mti, *rep_mti;
482 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
484 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
486 opc = mti->mti_flags & LDD_F_OPC_MASK;
487 if (opc == LDD_F_OPC_READY) {
488 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
489 mti->mti_fsname, mti->mti_stripe_index);
490 rc = mgs_ir_update(obd, mti);
492 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
493 CERROR("Update IR return with %d(ignore and IR "
496 GOTO(out_nolock, rc);
499 /* Do not support unregistering right now. */
500 if (opc != LDD_F_OPC_REG)
501 GOTO(out_nolock, rc = -EINVAL);
503 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
504 mti->mti_fsname, mti->mti_stripe_index);
506 if (mti->mti_flags & LDD_F_NEED_INDEX)
507 mti->mti_flags |= LDD_F_WRITECONF;
509 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
511 /* We're just here as a startup ping. */
512 CDEBUG(D_MGS, "Server %s is running on %s\n",
513 mti->mti_svname, obd_export_nid2str(req->rq_export));
514 rc = mgs_check_target(obd, mti);
515 /* above will set appropriate mti flags */
517 /* Nothing wrong, or fatal error */
518 GOTO(out_nolock, rc);
520 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
521 && (rc = mgs_check_failover_reg(mti)))
522 GOTO(out_nolock, rc);
525 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
527 if (mti->mti_flags & LDD_F_WRITECONF) {
528 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
529 mti->mti_stripe_index == 0) {
530 rc = mgs_erase_logs(obd, mti->mti_fsname);
531 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
532 "request. All servers must be restarted "
533 "in order to regenerate the logs."
534 "\n", obd->obd_name, mti->mti_fsname);
535 } else if (mti->mti_flags &
536 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
537 rc = mgs_erase_log(obd, mti->mti_svname);
538 LCONSOLE_WARN("%s: Regenerating %s log by user "
540 obd->obd_name, mti->mti_svname);
542 mti->mti_flags |= LDD_F_UPDATE;
543 /* Erased logs means start from scratch. */
544 mti->mti_flags &= ~LDD_F_UPGRADE14;
547 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
549 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
550 GOTO(out_nolock, rc);
554 * Log writing contention is handled by the fsdb_sem.
556 * It should be alright if someone was reading while we were
557 * updating the logs - if we revoke at the end they will just update
558 * from where they left off.
562 if (mti->mti_flags & LDD_F_UPGRADE14) {
563 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
565 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
569 /* We're good to go */
570 mti->mti_flags |= LDD_F_UPDATE;
574 if (mti->mti_flags & LDD_F_UPDATE) {
575 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
576 mti->mti_stripe_index);
578 /* create or update the target log
579 and update the client/mdt logs */
580 rc = mgs_write_log_target(obd, mti, fsdb);
582 CERROR("Failed to write %s log (%d)\n",
583 mti->mti_svname, rc);
587 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
588 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
590 mti->mti_flags |= LDD_F_REWRITE_LDD;
594 mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
597 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
598 mti->mti_stripe_index, rc);
601 /* we need an error flag to tell the target what's going on,
602 * instead of just doing it by error code only. */
603 mti->mti_flags |= LDD_F_ERROR;
605 rc = req_capsule_server_pack(&req->rq_pill);
609 /* send back the whole mti in the reply */
610 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
613 /* Flush logs to disk */
614 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
618 static int mgs_set_info_rpc(struct ptlrpc_request *req)
620 struct obd_device *obd = req->rq_export->exp_obd;
621 struct mgs_send_param *msp, *rep_msp;
623 struct lustre_cfg_bufs bufs;
624 struct lustre_cfg *lcfg;
625 char fsname[MTI_NAME_MAXLEN];
628 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
631 /* Construct lustre_cfg structure to pass to function mgs_setparam */
632 lustre_cfg_bufs_reset(&bufs, NULL);
633 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
634 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
635 rc = mgs_setparam(obd, lcfg, fsname);
637 CERROR("Error %d in setting the parameter %s for fs %s\n",
638 rc, msp->mgs_param, fsname);
642 lustre_cfg_free(lcfg);
644 rc = req_capsule_server_pack(&req->rq_pill);
646 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
652 static int mgs_config_read(struct ptlrpc_request *req)
654 struct mgs_config_body *body;
658 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
662 switch (body->mcb_type) {
663 case CONFIG_T_RECOVER:
664 rc = mgs_get_ir_logs(req);
667 case CONFIG_T_CONFIG:
680 * similar as in ost_connect_check_sptlrpc()
682 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
684 struct obd_export *exp = req->rq_export;
685 struct obd_device *obd = exp->exp_obd;
687 struct sptlrpc_flavor flvr;
690 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
691 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
695 cfs_down(&fsdb->fsdb_sem);
696 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
697 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
700 /* by defualt allow any flavors */
701 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
703 cfs_up(&fsdb->fsdb_sem);
705 cfs_spin_lock(&exp->exp_lock);
707 exp->exp_sp_peer = req->rq_sp_from;
708 exp->exp_flvr = flvr;
710 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
711 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
712 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
713 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
714 libcfs_nid2str(req->rq_peer.nid));
718 cfs_spin_unlock(&exp->exp_lock);
720 if (exp->exp_sp_peer != req->rq_sp_from) {
721 CERROR("RPC source %s doesn't match %s\n",
722 sptlrpc_part2name(req->rq_sp_from),
723 sptlrpc_part2name(exp->exp_sp_peer));
726 rc = sptlrpc_target_export_check(exp, req);
733 /* Called whenever a target cleans up. */
734 /* XXX - Currently unused */
735 static int mgs_handle_target_del(struct ptlrpc_request *req)
738 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
742 /* XXX - Currently unused */
743 static int mgs_handle_exception(struct ptlrpc_request *req)
746 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
751 * For old clients there is no direct way of knowing which filesystems
752 * a client is operating at the MGS side. But we need to pick up those
753 * clients so that the MGS can mark the corresponding filesystem as
754 * non-IR capable because old clients are not ready to be notified.
756 * This is why we have this _hack_ function. We detect the filesystem's
757 * name by hacking llog operation which is currently used by the clients
758 * to fetch configuration logs. At present this is fine because this is
759 * the ONLY llog operation between mgc and the MGS.
761 * If extra llog operation is going to be added, this function needs fixing.
763 * If releases prior than 2.0 are not supported, we can remove this function.
765 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
772 /* XXX: We suppose that llog at mgs is only used for
773 * fetching file system log */
774 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
775 if (logname == NULL) {
776 CERROR("No logname, is llog on MGS used for something else?\n");
780 ptr = strchr(logname, '-');
781 rc = (int)(ptr - logname);
782 if (ptr == NULL || rc >= sizeof(fsname)) {
783 CERROR("Invalid logname received: %s\n", logname);
787 strncpy(fsname, logname, rc);
789 rc = mgs_fsc_attach(req->rq_export, fsname);
790 if (rc < 0 && rc != -EEXIST)
791 CERROR("add fs client %s returns %d\n", fsname, rc);
796 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
797 int mgs_handle(struct ptlrpc_request *req)
799 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
803 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
804 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
805 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
808 LASSERT(current->journal_info == NULL);
809 opc = lustre_msg_get_opc(req->rq_reqmsg);
811 if (opc == SEC_CTX_INIT ||
812 opc == SEC_CTX_INIT_CONT ||
816 if (opc != MGS_CONNECT) {
817 if (!class_connected_export(req->rq_export)) {
818 CERROR("lustre_mgs: operation %d on unconnected MGS\n",
820 req->rq_status = -ENOTCONN;
821 GOTO(out, rc = -ENOTCONN);
827 DEBUG_REQ(D_MGS, req, "connect");
828 /* MGS and MDS have same request format for connect */
829 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
830 rc = target_handle_connect(req);
832 rc = mgs_connect_check_sptlrpc(req);
834 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
835 /* Make clients trying to reconnect after a MGS restart
836 happy; also requires obd_replayable */
837 lustre_msg_add_op_flags(req->rq_repmsg,
838 MSG_CONNECT_RECONNECT);
841 DEBUG_REQ(D_MGS, req, "disconnect");
842 /* MGS and MDS have same request format for disconnect */
843 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
844 rc = target_handle_disconnect(req);
845 req->rq_status = rc; /* superfluous? */
848 DEBUG_REQ(D_MGS, req, "exception");
849 rc = mgs_handle_exception(req);
852 DEBUG_REQ(D_MGS, req, "target add");
853 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
854 rc = mgs_handle_target_reg(req);
857 DEBUG_REQ(D_MGS, req, "target del");
858 rc = mgs_handle_target_del(req);
861 DEBUG_REQ(D_MGS, req, "set_info");
862 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
863 rc = mgs_set_info_rpc(req);
865 case MGS_CONFIG_READ:
866 DEBUG_REQ(D_MGS, req, "read config");
867 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
868 rc = mgs_config_read(req);
871 DEBUG_REQ(D_MGS, req, "enqueue");
872 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
873 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
874 ldlm_server_blocking_ast, NULL);
876 case LDLM_BL_CALLBACK:
877 case LDLM_CP_CALLBACK:
878 DEBUG_REQ(D_MGS, req, "callback");
879 CERROR("callbacks should not happen on MGS\n");
884 DEBUG_REQ(D_INFO, req, "ping");
885 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
886 rc = target_handle_ping(req);
889 DEBUG_REQ(D_MGS, req, "log cancel");
890 rc = -ENOTSUPP; /* la la la */
893 case LLOG_ORIGIN_HANDLE_CREATE:
894 DEBUG_REQ(D_MGS, req, "llog_init");
895 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
896 rc = llog_origin_handle_create(req);
898 (void)mgs_handle_fslog_hack(req);
900 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
901 DEBUG_REQ(D_MGS, req, "llog next block");
902 req_capsule_set(&req->rq_pill,
903 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
904 rc = llog_origin_handle_next_block(req);
906 case LLOG_ORIGIN_HANDLE_READ_HEADER:
907 DEBUG_REQ(D_MGS, req, "llog read header");
908 req_capsule_set(&req->rq_pill,
909 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
910 rc = llog_origin_handle_read_header(req);
912 case LLOG_ORIGIN_HANDLE_CLOSE:
913 DEBUG_REQ(D_MGS, req, "llog close");
914 rc = llog_origin_handle_close(req);
917 DEBUG_REQ(D_MGS, req, "llog catinfo");
918 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
919 rc = llog_catinfo(req);
922 req->rq_status = -ENOTSUPP;
923 rc = ptlrpc_error(req);
927 LASSERT(current->journal_info == NULL);
930 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
933 target_send_reply(req, rc, fail);
937 static inline int mgs_init_export(struct obd_export *exp)
939 struct mgs_export_data *data = &exp->u.eu_mgs_data;
941 /* init mgs_export_data for fsc */
942 cfs_spin_lock_init(&data->med_lock);
943 CFS_INIT_LIST_HEAD(&data->med_clients);
945 cfs_spin_lock(&exp->exp_lock);
946 exp->exp_connecting = 1;
947 cfs_spin_unlock(&exp->exp_lock);
949 /* self-export doesn't need client data and ldlm initialization */
950 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
951 &exp->exp_client_uuid)))
953 return ldlm_init_export(exp);
956 static inline int mgs_destroy_export(struct obd_export *exp)
960 target_destroy_export(exp);
961 mgs_client_free(exp);
963 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
964 &exp->exp_client_uuid)))
967 ldlm_destroy_export(exp);
972 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
977 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
985 strcpy(poolname, ptr);
990 static int mgs_iocontrol_pool(struct obd_device *obd,
991 struct obd_ioctl_data *data)
994 struct lustre_cfg *lcfg = NULL;
995 struct llog_rec_hdr rec;
997 char *poolname = NULL;
1000 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
1004 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1005 if (poolname == NULL) {
1009 rec.lrh_len = llog_data_len(data->ioc_plen1);
1011 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1012 rec.lrh_type = OBD_CFG_REC;
1014 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1019 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1024 OBD_ALLOC(lcfg, data->ioc_plen1);
1026 GOTO(out_pool, rc = -ENOMEM);
1028 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1029 GOTO(out_pool, rc = -EFAULT);
1031 if (lcfg->lcfg_bufcount < 2) {
1032 GOTO(out_pool, rc = -EFAULT);
1035 /* first arg is always <fsname>.<poolname> */
1036 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
1039 switch (lcfg->lcfg_command) {
1040 case LCFG_POOL_NEW: {
1041 if (lcfg->lcfg_bufcount != 2)
1043 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
1047 case LCFG_POOL_ADD: {
1048 if (lcfg->lcfg_bufcount != 3)
1050 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1051 lustre_cfg_string(lcfg, 2));
1054 case LCFG_POOL_REM: {
1055 if (lcfg->lcfg_bufcount != 3)
1057 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1058 lustre_cfg_string(lcfg, 2));
1061 case LCFG_POOL_DEL: {
1062 if (lcfg->lcfg_bufcount != 2)
1064 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1075 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1076 rc, lcfg->lcfg_command, fsname, poolname);
1082 OBD_FREE(lcfg, data->ioc_plen1);
1085 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1087 if (poolname != NULL)
1088 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1093 /* from mdt_iocontrol */
1094 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1095 void *karg, void *uarg)
1097 struct obd_device *obd = exp->exp_obd;
1098 struct obd_ioctl_data *data = karg;
1099 struct lvfs_run_ctxt saved;
1103 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1107 case OBD_IOC_PARAM: {
1108 struct lustre_cfg *lcfg;
1109 struct llog_rec_hdr rec;
1110 char fsname[MTI_NAME_MAXLEN];
1112 rec.lrh_len = llog_data_len(data->ioc_plen1);
1114 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1115 rec.lrh_type = OBD_CFG_REC;
1117 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1121 OBD_ALLOC(lcfg, data->ioc_plen1);
1124 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1125 GOTO(out_free, rc = -EFAULT);
1127 if (lcfg->lcfg_bufcount < 1)
1128 GOTO(out_free, rc = -EINVAL);
1130 rc = mgs_setparam(obd, lcfg, fsname);
1132 CERROR("setparam err %d\n", rc);
1136 OBD_FREE(lcfg, data->ioc_plen1);
1140 case OBD_IOC_POOL: {
1141 RETURN(mgs_iocontrol_pool(obd, data));
1144 case OBD_IOC_DUMP_LOG: {
1145 struct llog_ctxt *ctxt;
1146 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1147 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1148 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1149 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1150 llog_ctxt_put(ctxt);
1155 case OBD_IOC_LLOG_CHECK:
1156 case OBD_IOC_LLOG_INFO:
1157 case OBD_IOC_LLOG_PRINT: {
1158 struct llog_ctxt *ctxt;
1159 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1161 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1162 rc = llog_ioctl(ctxt, cmd, data);
1163 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1164 llog_ctxt_put(ctxt);
1170 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1176 /* use obd ops to offer management infrastructure */
1177 static struct obd_ops mgs_obd_ops = {
1178 .o_owner = THIS_MODULE,
1179 .o_connect = mgs_connect,
1180 .o_reconnect = mgs_reconnect,
1181 .o_disconnect = mgs_disconnect,
1182 .o_setup = mgs_setup,
1183 .o_precleanup = mgs_precleanup,
1184 .o_cleanup = mgs_cleanup,
1185 .o_init_export = mgs_init_export,
1186 .o_destroy_export = mgs_destroy_export,
1187 .o_iocontrol = mgs_iocontrol,
1188 .o_llog_init = mgs_llog_init,
1189 .o_llog_finish = mgs_llog_finish
1192 static int __init mgs_init(void)
1194 struct lprocfs_static_vars lvars;
1196 lprocfs_mgs_init_vars(&lvars);
1197 class_register_type(&mgs_obd_ops, NULL,
1198 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1203 static void /*__exit*/ mgs_exit(void)
1205 class_unregister_type(LUSTRE_MGS_NAME);
1208 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1209 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1210 MODULE_LICENSE("GPL");
1212 module_init(mgs_init);
1213 module_exit(mgs_exit);