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 static struct ptlrpc_service_conf conf;
171 struct lprocfs_static_vars lvars;
172 struct mgs_obd *mgs = &obd->u.mgs;
173 struct lustre_mount_info *lmi;
174 struct lustre_sb_info *lsi;
175 struct vfsmount *mnt;
179 CDEBUG(D_CONFIG, "Starting MGS\n");
182 lmi = server_get_mount(obd->obd_name);
184 RETURN(rc = -EINVAL);
187 lsi = s2lsi(lmi->lmi_sb);
188 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
189 if (IS_ERR(obd->obd_fsops))
190 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
192 if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) {
193 CERROR("%s: Underlying device is marked as read-only. "
194 "Setup failed\n", obd->obd_name);
195 GOTO(err_ops, rc = -EROFS);
198 obd->u.obt.obt_magic = OBT_MAGIC;
199 obd->u.obt.obt_instance = 0;
201 /* namespace for mgs llog */
202 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
203 LDLM_NAMESPACE_SERVER,
204 LDLM_NAMESPACE_MODEST,
206 if (obd->obd_namespace == NULL)
207 GOTO(err_ops, rc = -ENOMEM);
210 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
211 "mgs_ldlm_client", &obd->obd_ldlm_client);
213 rc = mgs_fs_setup(obd, mnt);
215 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
220 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
224 /* No recovery for MGC's */
225 obd->obd_replayable = 0;
227 /* Internal mgs setup */
228 mgs_init_fsdb_list(obd);
229 cfs_mutex_init(&mgs->mgs_mutex);
230 mgs->mgs_start_time = cfs_time_current_sec();
233 lprocfs_mgs_init_vars(&lvars);
234 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
235 lproc_mgs_setup(obd);
236 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
241 conf = (typeof(conf)) {
242 .psc_name = LUSTRE_MGS_NAME,
243 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
245 .bc_nbufs = MGS_NBUFS,
246 .bc_buf_size = MGS_BUFSIZE,
247 .bc_req_max_size = MGS_MAXREQSIZE,
248 .bc_rep_max_size = MGS_MAXREPSIZE,
249 .bc_req_portal = MGS_REQUEST_PORTAL,
250 .bc_rep_portal = MGC_REPLY_PORTAL,
253 .tc_thr_name = "ll_mgs",
254 .tc_nthrs_min = MGS_THREADS_AUTO_MIN,
255 .tc_nthrs_max = MGS_THREADS_AUTO_MAX,
256 .tc_ctx_tags = LCT_MD_THREAD,
259 .so_req_handler = mgs_handle,
260 .so_req_printer = target_print_req,
263 /* Start the service threads */
264 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
265 if (IS_ERR(mgs->mgs_service)) {
266 rc = PTR_ERR(mgs->mgs_service);
267 CERROR("failed to start service: %d\n", rc);
271 ping_evictor_start();
273 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
278 lproc_mgs_cleanup(obd);
279 obd_llog_finish(obd, 0);
281 /* No extra cleanup needed for llog_init_commit_thread() */
284 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
285 obd->obd_namespace = NULL;
287 fsfilt_put_ops(obd->obd_fsops);
289 server_put_mount(obd->obd_name, mnt);
294 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
296 struct mgs_obd *mgs = &obd->u.mgs;
301 case OBD_CLEANUP_EARLY:
303 case OBD_CLEANUP_EXPORTS:
305 ptlrpc_unregister_service(mgs->mgs_service);
306 mgs_cleanup_fsdb_list(obd);
307 rc = obd_llog_finish(obd, 0);
308 lproc_mgs_cleanup(obd);
315 * Performs cleanup procedures for passed \a obd given it is mgs obd.
317 static int mgs_cleanup(struct obd_device *obd)
319 struct mgs_obd *mgs = &obd->u.mgs;
322 if (mgs->mgs_sb == NULL)
327 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
330 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
331 obd->obd_namespace = NULL;
333 fsfilt_put_ops(obd->obd_fsops);
335 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
339 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
344 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
345 LDLM_FL_BLOCK_CONV))) {
346 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
347 struct lustre_handle lockh;
349 /* clear the bit before lock put */
350 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
352 ldlm_lock2handle(lock, &lockh);
353 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
356 RETURN(ldlm_completion_ast(lock, flags, cbdata));
359 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
364 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
365 LDLM_FL_BLOCK_CONV))) {
368 /* l_ast_data is used as a marker to avoid cancel ldlm lock
369 * twice. See LU-1259. */
370 lock_res_and_lock(lock);
371 fsdb = (struct fs_db *)lock->l_ast_data;
372 lock->l_ast_data = NULL;
373 unlock_res_and_lock(lock);
376 struct lustre_handle lockh;
378 mgs_ir_notify_complete(fsdb);
380 ldlm_lock2handle(lock, &lockh);
381 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
385 RETURN(ldlm_completion_ast(lock, flags, cbdata));
388 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb, int type)
390 ldlm_completion_callback cp = NULL;
391 struct lustre_handle lockh = { 0 };
392 struct ldlm_res_id res_id;
393 int flags = LDLM_FL_ATOMIC_CB;
397 LASSERT(fsdb->fsdb_name[0] != '\0');
398 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
402 case CONFIG_T_CONFIG:
403 cp = mgs_completion_ast_config;
404 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
407 case CONFIG_T_RECOVER:
408 cp = mgs_completion_ast_ir;
415 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
416 LDLM_PLAIN, NULL, LCK_EX, &flags,
417 ldlm_blocking_ast, cp, NULL,
418 fsdb, 0, NULL, &lockh);
419 if (rc != ELDLM_OK) {
420 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
421 le64_to_cpu(res_id.name[0]),
422 le64_to_cpu(res_id.name[1]), rc);
424 if (type == CONFIG_T_CONFIG)
425 cfs_clear_bit(FSDB_REVOKING_LOCK,
428 /* lock has been cancelled in completion_ast. */
437 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
442 rc = mgs_check_index(obd, mti);
444 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
445 "this MGS does not know about it, preventing "
446 "registration.\n", mti->mti_svname);
448 } else if (rc == -1) {
449 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
450 "disappeared! Regenerating all logs.\n",
452 mti->mti_flags |= LDD_F_WRITECONF;
455 /* Index is correctly marked as used */
457 /* If the logs don't contain the mti_nids then add
458 them as failover nids */
459 rc = mgs_check_failnid(obd, mti);
465 /* Ensure this is not a failover node that is connecting first*/
466 static int mgs_check_failover_reg(struct mgs_target_info *mti)
472 ptr = mti->mti_params;
473 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
474 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
475 for (i = 0; i < mti->mti_nid_count; i++) {
476 if (nid == mti->mti_nids[i]) {
477 LCONSOLE_WARN("Denying initial registra"
478 "tion attempt from nid %s"
479 ", specified as failover"
480 "\n",libcfs_nid2str(nid));
481 return -EADDRNOTAVAIL;
489 /* Called whenever a target starts up. Flags indicate first connect, etc. */
490 static int mgs_handle_target_reg(struct ptlrpc_request *req)
492 struct obd_device *obd = req->rq_export->exp_obd;
493 struct mgs_target_info *mti, *rep_mti;
499 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
501 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
503 opc = mti->mti_flags & LDD_F_OPC_MASK;
504 if (opc == LDD_F_OPC_READY) {
505 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
506 mti->mti_fsname, mti->mti_stripe_index);
507 rc = mgs_ir_update(obd, mti);
509 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
510 CERROR("Update IR return with %d(ignore and IR "
513 GOTO(out_nolock, rc);
516 /* Do not support unregistering right now. */
517 if (opc != LDD_F_OPC_REG)
518 GOTO(out_nolock, rc = -EINVAL);
520 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
521 mti->mti_fsname, mti->mti_stripe_index);
523 if (mti->mti_flags & LDD_F_NEED_INDEX)
524 mti->mti_flags |= LDD_F_WRITECONF;
526 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
528 /* We're just here as a startup ping. */
529 CDEBUG(D_MGS, "Server %s is running on %s\n",
530 mti->mti_svname, obd_export_nid2str(req->rq_export));
531 rc = mgs_check_target(obd, mti);
532 /* above will set appropriate mti flags */
534 /* Nothing wrong, or fatal error */
535 GOTO(out_nolock, rc);
537 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
538 && (rc = mgs_check_failover_reg(mti)))
539 GOTO(out_nolock, rc);
542 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
544 if (mti->mti_flags & LDD_F_WRITECONF) {
545 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
546 mti->mti_stripe_index == 0) {
547 rc = mgs_erase_logs(obd, mti->mti_fsname);
548 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
549 "request. All servers must be restarted "
550 "in order to regenerate the logs."
551 "\n", obd->obd_name, mti->mti_fsname);
552 } else if (mti->mti_flags &
553 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
554 rc = mgs_erase_log(obd, mti->mti_svname);
555 LCONSOLE_WARN("%s: Regenerating %s log by user "
557 obd->obd_name, mti->mti_svname);
559 mti->mti_flags |= LDD_F_UPDATE;
560 /* Erased logs means start from scratch. */
561 mti->mti_flags &= ~LDD_F_UPGRADE14;
564 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
566 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
567 GOTO(out_nolock, rc);
571 * Log writing contention is handled by the fsdb_mutex.
573 * It should be alright if someone was reading while we were
574 * updating the logs - if we revoke at the end they will just update
575 * from where they left off.
579 if (mti->mti_flags & LDD_F_UPGRADE14) {
580 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
582 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
586 /* We're good to go */
587 mti->mti_flags |= LDD_F_UPDATE;
591 if (mti->mti_flags & LDD_F_UPDATE) {
592 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
593 mti->mti_stripe_index);
595 /* create or update the target log
596 and update the client/mdt logs */
597 rc = mgs_write_log_target(obd, mti, fsdb);
599 CERROR("Failed to write %s log (%d)\n",
600 mti->mti_svname, rc);
604 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
605 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
607 mti->mti_flags |= LDD_F_REWRITE_LDD;
611 mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
614 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
615 mti->mti_stripe_index, rc);
618 /* we need an error flag to tell the target what's going on,
619 * instead of just doing it by error code only. */
620 mti->mti_flags |= LDD_F_ERROR;
622 rc = req_capsule_server_pack(&req->rq_pill);
626 /* send back the whole mti in the reply */
627 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
630 /* Flush logs to disk */
631 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
635 static int mgs_set_info_rpc(struct ptlrpc_request *req)
637 struct obd_device *obd = req->rq_export->exp_obd;
638 struct mgs_send_param *msp, *rep_msp;
640 struct lustre_cfg_bufs bufs;
641 struct lustre_cfg *lcfg;
642 char fsname[MTI_NAME_MAXLEN];
645 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
648 /* Construct lustre_cfg structure to pass to function mgs_setparam */
649 lustre_cfg_bufs_reset(&bufs, NULL);
650 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
651 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
652 rc = mgs_setparam(obd, lcfg, fsname);
654 CERROR("Error %d in setting the parameter %s for fs %s\n",
655 rc, msp->mgs_param, fsname);
659 lustre_cfg_free(lcfg);
661 rc = req_capsule_server_pack(&req->rq_pill);
663 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
669 static int mgs_config_read(struct ptlrpc_request *req)
671 struct mgs_config_body *body;
675 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
679 switch (body->mcb_type) {
680 case CONFIG_T_RECOVER:
681 rc = mgs_get_ir_logs(req);
684 case CONFIG_T_CONFIG:
697 * similar as in ost_connect_check_sptlrpc()
699 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
701 struct obd_export *exp = req->rq_export;
702 struct obd_device *obd = exp->exp_obd;
704 struct sptlrpc_flavor flvr;
707 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
708 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
712 cfs_mutex_lock(&fsdb->fsdb_mutex);
713 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
714 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
717 /* by defualt allow any flavors */
718 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
720 cfs_mutex_unlock(&fsdb->fsdb_mutex);
722 cfs_spin_lock(&exp->exp_lock);
724 exp->exp_sp_peer = req->rq_sp_from;
725 exp->exp_flvr = flvr;
727 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
728 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
729 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
730 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
731 libcfs_nid2str(req->rq_peer.nid));
735 cfs_spin_unlock(&exp->exp_lock);
737 if (exp->exp_sp_peer != req->rq_sp_from) {
738 CERROR("RPC source %s doesn't match %s\n",
739 sptlrpc_part2name(req->rq_sp_from),
740 sptlrpc_part2name(exp->exp_sp_peer));
743 rc = sptlrpc_target_export_check(exp, req);
750 /* Called whenever a target cleans up. */
751 /* XXX - Currently unused */
752 static int mgs_handle_target_del(struct ptlrpc_request *req)
755 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
759 /* XXX - Currently unused */
760 static int mgs_handle_exception(struct ptlrpc_request *req)
763 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
768 * For old clients there is no direct way of knowing which filesystems
769 * a client is operating at the MGS side. But we need to pick up those
770 * clients so that the MGS can mark the corresponding filesystem as
771 * non-IR capable because old clients are not ready to be notified.
773 * This is why we have this _hack_ function. We detect the filesystem's
774 * name by hacking llog operation which is currently used by the clients
775 * to fetch configuration logs. At present this is fine because this is
776 * the ONLY llog operation between mgc and the MGS.
778 * If extra llog operation is going to be added, this function needs fixing.
780 * If releases prior than 2.0 are not supported, we can remove this function.
782 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
789 /* XXX: We suppose that llog at mgs is only used for
790 * fetching file system log */
791 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
792 if (logname == NULL) {
793 CERROR("No logname, is llog on MGS used for something else?\n");
797 ptr = strchr(logname, '-');
798 rc = (int)(ptr - logname);
799 if (ptr == NULL || rc >= sizeof(fsname)) {
800 CERROR("Invalid logname received: %s\n", logname);
804 strncpy(fsname, logname, rc);
806 rc = mgs_fsc_attach(req->rq_export, fsname);
807 if (rc < 0 && rc != -EEXIST)
808 CERROR("add fs client %s returns %d\n", fsname, rc);
813 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
814 int mgs_handle(struct ptlrpc_request *req)
816 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
820 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
821 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
822 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
825 LASSERT(current->journal_info == NULL);
826 opc = lustre_msg_get_opc(req->rq_reqmsg);
828 if (opc == SEC_CTX_INIT ||
829 opc == SEC_CTX_INIT_CONT ||
833 if (opc != MGS_CONNECT) {
834 if (!class_connected_export(req->rq_export)) {
835 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
836 req->rq_status = -ENOTCONN;
837 GOTO(out, rc = -ENOTCONN);
843 DEBUG_REQ(D_MGS, req, "connect");
844 /* MGS and MDS have same request format for connect */
845 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
846 rc = target_handle_connect(req);
848 rc = mgs_connect_check_sptlrpc(req);
850 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
851 /* Make clients trying to reconnect after a MGS restart
852 happy; also requires obd_replayable */
853 lustre_msg_add_op_flags(req->rq_repmsg,
854 MSG_CONNECT_RECONNECT);
857 DEBUG_REQ(D_MGS, req, "disconnect");
858 /* MGS and MDS have same request format for disconnect */
859 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
860 rc = target_handle_disconnect(req);
861 req->rq_status = rc; /* superfluous? */
864 DEBUG_REQ(D_MGS, req, "exception");
865 rc = mgs_handle_exception(req);
868 DEBUG_REQ(D_MGS, req, "target add");
869 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
870 rc = mgs_handle_target_reg(req);
873 DEBUG_REQ(D_MGS, req, "target del");
874 rc = mgs_handle_target_del(req);
877 DEBUG_REQ(D_MGS, req, "set_info");
878 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
879 rc = mgs_set_info_rpc(req);
881 case MGS_CONFIG_READ:
882 DEBUG_REQ(D_MGS, req, "read config");
883 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
884 rc = mgs_config_read(req);
887 DEBUG_REQ(D_MGS, req, "enqueue");
888 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
889 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
890 ldlm_server_blocking_ast, NULL);
892 case LDLM_BL_CALLBACK:
893 case LDLM_CP_CALLBACK:
894 DEBUG_REQ(D_MGS, req, "callback");
895 CERROR("callbacks should not happen on MGS\n");
900 DEBUG_REQ(D_INFO, req, "ping");
901 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
902 rc = target_handle_ping(req);
905 DEBUG_REQ(D_MGS, req, "log cancel");
906 rc = -ENOTSUPP; /* la la la */
909 case LLOG_ORIGIN_HANDLE_CREATE:
910 DEBUG_REQ(D_MGS, req, "llog_init");
911 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
912 rc = llog_origin_handle_create(req);
914 (void)mgs_handle_fslog_hack(req);
916 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
917 DEBUG_REQ(D_MGS, req, "llog next block");
918 req_capsule_set(&req->rq_pill,
919 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
920 rc = llog_origin_handle_next_block(req);
922 case LLOG_ORIGIN_HANDLE_READ_HEADER:
923 DEBUG_REQ(D_MGS, req, "llog read header");
924 req_capsule_set(&req->rq_pill,
925 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
926 rc = llog_origin_handle_read_header(req);
928 case LLOG_ORIGIN_HANDLE_CLOSE:
929 DEBUG_REQ(D_MGS, req, "llog close");
930 rc = llog_origin_handle_close(req);
933 DEBUG_REQ(D_MGS, req, "llog catinfo");
934 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
935 rc = llog_catinfo(req);
938 req->rq_status = -ENOTSUPP;
939 rc = ptlrpc_error(req);
943 LASSERT(current->journal_info == NULL);
946 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
949 target_send_reply(req, rc, fail);
953 static inline int mgs_init_export(struct obd_export *exp)
955 struct mgs_export_data *data = &exp->u.eu_mgs_data;
957 /* init mgs_export_data for fsc */
958 cfs_spin_lock_init(&data->med_lock);
959 CFS_INIT_LIST_HEAD(&data->med_clients);
961 cfs_spin_lock(&exp->exp_lock);
962 exp->exp_connecting = 1;
963 cfs_spin_unlock(&exp->exp_lock);
965 /* self-export doesn't need client data and ldlm initialization */
966 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
967 &exp->exp_client_uuid)))
969 return ldlm_init_export(exp);
972 static inline int mgs_destroy_export(struct obd_export *exp)
976 target_destroy_export(exp);
977 mgs_client_free(exp);
979 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
980 &exp->exp_client_uuid)))
983 ldlm_destroy_export(exp);
988 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
993 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
1001 strcpy(poolname, ptr);
1006 static int mgs_iocontrol_pool(struct obd_device *obd,
1007 struct obd_ioctl_data *data)
1010 struct lustre_cfg *lcfg = NULL;
1011 struct llog_rec_hdr rec;
1012 char *fsname = NULL;
1013 char *poolname = NULL;
1016 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
1020 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1021 if (poolname == NULL) {
1025 rec.lrh_len = llog_data_len(data->ioc_plen1);
1027 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1028 rec.lrh_type = OBD_CFG_REC;
1030 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1035 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1040 OBD_ALLOC(lcfg, data->ioc_plen1);
1042 GOTO(out_pool, rc = -ENOMEM);
1044 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1045 GOTO(out_pool, rc = -EFAULT);
1047 if (lcfg->lcfg_bufcount < 2) {
1048 GOTO(out_pool, rc = -EFAULT);
1051 /* first arg is always <fsname>.<poolname> */
1052 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
1055 switch (lcfg->lcfg_command) {
1056 case LCFG_POOL_NEW: {
1057 if (lcfg->lcfg_bufcount != 2)
1059 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
1063 case LCFG_POOL_ADD: {
1064 if (lcfg->lcfg_bufcount != 3)
1066 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1067 lustre_cfg_string(lcfg, 2));
1070 case LCFG_POOL_REM: {
1071 if (lcfg->lcfg_bufcount != 3)
1073 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1074 lustre_cfg_string(lcfg, 2));
1077 case LCFG_POOL_DEL: {
1078 if (lcfg->lcfg_bufcount != 2)
1080 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1091 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1092 rc, lcfg->lcfg_command, fsname, poolname);
1098 OBD_FREE(lcfg, data->ioc_plen1);
1101 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1103 if (poolname != NULL)
1104 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1109 /* from mdt_iocontrol */
1110 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1111 void *karg, void *uarg)
1113 struct obd_device *obd = exp->exp_obd;
1114 struct obd_ioctl_data *data = karg;
1115 struct lvfs_run_ctxt saved;
1119 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1123 case OBD_IOC_PARAM: {
1124 struct lustre_cfg *lcfg;
1125 struct llog_rec_hdr rec;
1126 char fsname[MTI_NAME_MAXLEN];
1128 rec.lrh_len = llog_data_len(data->ioc_plen1);
1130 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1131 rec.lrh_type = OBD_CFG_REC;
1133 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1137 OBD_ALLOC(lcfg, data->ioc_plen1);
1140 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1141 GOTO(out_free, rc = -EFAULT);
1143 if (lcfg->lcfg_bufcount < 1)
1144 GOTO(out_free, rc = -EINVAL);
1146 rc = mgs_setparam(obd, lcfg, fsname);
1148 CERROR("setparam err %d\n", rc);
1152 OBD_FREE(lcfg, data->ioc_plen1);
1156 case OBD_IOC_POOL: {
1157 RETURN(mgs_iocontrol_pool(obd, data));
1160 case OBD_IOC_DUMP_LOG: {
1161 struct llog_ctxt *ctxt;
1162 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1163 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1164 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1165 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1166 llog_ctxt_put(ctxt);
1171 case OBD_IOC_LLOG_CHECK:
1172 case OBD_IOC_LLOG_INFO:
1173 case OBD_IOC_LLOG_PRINT: {
1174 struct llog_ctxt *ctxt;
1175 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1177 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1178 rc = llog_ioctl(ctxt, cmd, data);
1179 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1180 llog_ctxt_put(ctxt);
1186 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1192 /* use obd ops to offer management infrastructure */
1193 static struct obd_ops mgs_obd_ops = {
1194 .o_owner = THIS_MODULE,
1195 .o_connect = mgs_connect,
1196 .o_reconnect = mgs_reconnect,
1197 .o_disconnect = mgs_disconnect,
1198 .o_setup = mgs_setup,
1199 .o_precleanup = mgs_precleanup,
1200 .o_cleanup = mgs_cleanup,
1201 .o_init_export = mgs_init_export,
1202 .o_destroy_export = mgs_destroy_export,
1203 .o_iocontrol = mgs_iocontrol,
1204 .o_llog_init = mgs_llog_init,
1205 .o_llog_finish = mgs_llog_finish
1208 static int __init mgs_init(void)
1210 struct lprocfs_static_vars lvars;
1212 lprocfs_mgs_init_vars(&lvars);
1213 class_register_type(&mgs_obd_ops, NULL,
1214 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1219 static void /*__exit*/ mgs_exit(void)
1221 class_unregister_type(LUSTRE_MGS_NAME);
1224 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1225 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1226 MODULE_LICENSE("GPL");
1228 module_init(mgs_init);
1229 module_exit(mgs_exit);