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.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/mgs/mgs_handler.c
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MGS
47 #define D_MGS D_CONFIG
50 # include <linux/module.h>
51 # include <linux/pagemap.h>
52 # include <linux/miscdevice.h>
53 # include <linux/init.h>
55 # include <liblustre.h>
58 #include <obd_class.h>
59 #include <lustre_dlm.h>
60 #include <lprocfs_status.h>
61 #include <lustre_fsfilt.h>
62 #include <lustre_disk.h>
63 #include "mgs_internal.h"
64 #include <lustre_param.h>
66 /* Establish a connection to the MGS.*/
67 static int mgs_connect(const struct lu_env *env,
68 struct obd_export **exp, struct obd_device *obd,
69 struct obd_uuid *cluuid, struct obd_connect_data *data,
72 struct obd_export *lexp;
73 struct lustre_handle conn = { 0 };
77 if (!exp || !obd || !cluuid)
80 rc = class_connect(&conn, obd, cluuid);
84 lexp = class_conn2export(&conn);
87 mgs_counter_incr(lexp, LPROC_MGS_CONNECT);
90 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
91 lexp->exp_connect_flags = data->ocd_connect_flags;
92 data->ocd_version = LUSTRE_VERSION_CODE;
95 rc = mgs_export_stats_init(obd, lexp, localdata);
98 class_disconnect(lexp);
106 static int mgs_reconnect(const struct lu_env *env,
107 struct obd_export *exp, struct obd_device *obd,
108 struct obd_uuid *cluuid, struct obd_connect_data *data,
113 if (exp == NULL || obd == NULL || cluuid == NULL)
116 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
119 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
120 exp->exp_connect_flags = data->ocd_connect_flags;
121 data->ocd_version = LUSTRE_VERSION_CODE;
124 RETURN(mgs_export_stats_init(obd, exp, localdata));
127 static int mgs_disconnect(struct obd_export *exp)
134 mgs_fsc_cleanup(exp);
136 class_export_get(exp);
137 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
139 rc = server_disconnect_export(exp);
140 class_export_put(exp);
144 static int mgs_cleanup(struct obd_device *obd);
145 static int mgs_handle(struct ptlrpc_request *req);
147 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
148 struct obd_device *tgt, int *index)
153 LASSERT(olg == &obd->obd_olg);
154 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
159 static int mgs_llog_finish(struct obd_device *obd, int count)
161 struct llog_ctxt *ctxt;
165 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
167 rc = llog_cleanup(ctxt);
172 /* Start the MGS obd */
173 static int mgs_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
175 struct lprocfs_static_vars lvars;
176 struct mgs_obd *mgs = &obd->u.mgs;
177 struct lustre_mount_info *lmi;
178 struct lustre_sb_info *lsi;
179 struct vfsmount *mnt;
183 CDEBUG(D_CONFIG, "Starting MGS\n");
186 lmi = server_get_mount(obd->obd_name);
188 RETURN(rc = -EINVAL);
191 lsi = s2lsi(lmi->lmi_sb);
192 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
193 if (IS_ERR(obd->obd_fsops))
194 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
196 if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) {
197 CERROR("%s: Underlying device is marked as read-only. "
198 "Setup failed\n", obd->obd_name);
199 GOTO(err_ops, rc = -EROFS);
202 obd->u.obt.obt_magic = OBT_MAGIC;
203 obd->u.obt.obt_instance = 0;
205 /* namespace for mgs llog */
206 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
207 LDLM_NAMESPACE_SERVER,
208 LDLM_NAMESPACE_MODEST,
210 if (obd->obd_namespace == NULL)
211 GOTO(err_ops, rc = -ENOMEM);
214 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
215 "mgs_ldlm_client", &obd->obd_ldlm_client);
217 rc = mgs_fs_setup(obd, mnt);
219 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
224 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
228 /* No recovery for MGC's */
229 obd->obd_replayable = 0;
231 /* Internal mgs setup */
232 mgs_init_fsdb_list(obd);
233 cfs_mutex_init(&mgs->mgs_mutex);
234 mgs->mgs_start_time = cfs_time_current_sec();
237 lprocfs_mgs_init_vars(&lvars);
238 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
239 lproc_mgs_setup(obd);
240 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
245 /* Start the service threads */
247 ptlrpc_init_svc(MGS_NBUFS, MGS_BUFSIZE, MGS_MAXREQSIZE,
248 MGS_MAXREPSIZE, MGS_REQUEST_PORTAL,
250 mgs_handle, LUSTRE_MGS_NAME,
251 obd->obd_proc_entry, target_print_req,
252 MGS_THREADS_AUTO_MIN, MGS_THREADS_AUTO_MAX,
253 "ll_mgs", LCT_MD_THREAD, NULL);
255 if (!mgs->mgs_service) {
256 CERROR("failed to start service\n");
257 GOTO(err_llog, rc = -ENOMEM);
260 rc = ptlrpc_start_threads(mgs->mgs_service);
262 GOTO(err_thread, rc);
264 ping_evictor_start();
266 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
271 ptlrpc_unregister_service(mgs->mgs_service);
273 lproc_mgs_cleanup(obd);
274 obd_llog_finish(obd, 0);
276 /* No extra cleanup needed for llog_init_commit_thread() */
279 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
280 obd->obd_namespace = NULL;
282 fsfilt_put_ops(obd->obd_fsops);
284 server_put_mount(obd->obd_name, mnt);
289 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
291 struct mgs_obd *mgs = &obd->u.mgs;
296 case OBD_CLEANUP_EARLY:
298 case OBD_CLEANUP_EXPORTS:
300 ptlrpc_unregister_service(mgs->mgs_service);
301 mgs_cleanup_fsdb_list(obd);
302 rc = obd_llog_finish(obd, 0);
303 lproc_mgs_cleanup(obd);
310 * Performs cleanup procedures for passed \a obd given it is mgs obd.
312 static int mgs_cleanup(struct obd_device *obd)
314 struct mgs_obd *mgs = &obd->u.mgs;
317 if (mgs->mgs_sb == NULL)
322 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
325 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
326 obd->obd_namespace = NULL;
328 fsfilt_put_ops(obd->obd_fsops);
330 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
334 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
339 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
340 LDLM_FL_BLOCK_CONV))) {
341 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
342 struct lustre_handle lockh;
344 /* clear the bit before lock put */
345 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
347 ldlm_lock2handle(lock, &lockh);
348 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
351 RETURN(ldlm_completion_ast(lock, flags, cbdata));
354 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
359 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
360 LDLM_FL_BLOCK_CONV))) {
363 /* l_ast_data is used as a marker to avoid cancel ldlm lock
364 * twice. See LU-1259. */
365 lock_res_and_lock(lock);
366 fsdb = (struct fs_db *)lock->l_ast_data;
367 lock->l_ast_data = NULL;
368 unlock_res_and_lock(lock);
371 struct lustre_handle lockh;
373 mgs_ir_notify_complete(fsdb);
375 ldlm_lock2handle(lock, &lockh);
376 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
380 RETURN(ldlm_completion_ast(lock, flags, cbdata));
383 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb, int type)
385 ldlm_completion_callback cp = NULL;
386 struct lustre_handle lockh = { 0 };
387 struct ldlm_res_id res_id;
388 int flags = LDLM_FL_ATOMIC_CB;
392 LASSERT(fsdb->fsdb_name[0] != '\0');
393 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
397 case CONFIG_T_CONFIG:
398 cp = mgs_completion_ast_config;
399 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
402 case CONFIG_T_RECOVER:
403 cp = mgs_completion_ast_ir;
410 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
411 LDLM_PLAIN, NULL, LCK_EX, &flags,
412 ldlm_blocking_ast, cp, NULL,
413 fsdb, 0, NULL, &lockh);
414 if (rc != ELDLM_OK) {
415 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
416 le64_to_cpu(res_id.name[0]),
417 le64_to_cpu(res_id.name[1]), rc);
419 if (type == CONFIG_T_CONFIG)
420 cfs_clear_bit(FSDB_REVOKING_LOCK,
423 /* lock has been cancelled in completion_ast. */
432 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
437 rc = mgs_check_index(obd, mti);
439 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
440 "this MGS does not know about it, preventing "
441 "registration.\n", mti->mti_svname);
443 } else if (rc == -1) {
444 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
445 "disappeared! Regenerating all logs.\n",
447 mti->mti_flags |= LDD_F_WRITECONF;
450 /* Index is correctly marked as used */
452 /* If the logs don't contain the mti_nids then add
453 them as failover nids */
454 rc = mgs_check_failnid(obd, mti);
460 /* Ensure this is not a failover node that is connecting first*/
461 static int mgs_check_failover_reg(struct mgs_target_info *mti)
467 ptr = mti->mti_params;
468 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
469 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
470 for (i = 0; i < mti->mti_nid_count; i++) {
471 if (nid == mti->mti_nids[i]) {
472 LCONSOLE_WARN("Denying initial registra"
473 "tion attempt from nid %s"
474 ", specified as failover"
475 "\n",libcfs_nid2str(nid));
476 return -EADDRNOTAVAIL;
484 /* Called whenever a target starts up. Flags indicate first connect, etc. */
485 static int mgs_handle_target_reg(struct ptlrpc_request *req)
487 struct obd_device *obd = req->rq_export->exp_obd;
488 struct mgs_target_info *mti, *rep_mti;
494 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
496 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
498 opc = mti->mti_flags & LDD_F_OPC_MASK;
499 if (opc == LDD_F_OPC_READY) {
500 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
501 mti->mti_fsname, mti->mti_stripe_index);
502 rc = mgs_ir_update(obd, mti);
504 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
505 CERROR("Update IR return with %d(ignore and IR "
508 GOTO(out_nolock, rc);
511 /* Do not support unregistering right now. */
512 if (opc != LDD_F_OPC_REG)
513 GOTO(out_nolock, rc = -EINVAL);
515 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
516 mti->mti_fsname, mti->mti_stripe_index);
518 if (mti->mti_flags & LDD_F_NEED_INDEX)
519 mti->mti_flags |= LDD_F_WRITECONF;
521 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
523 /* We're just here as a startup ping. */
524 CDEBUG(D_MGS, "Server %s is running on %s\n",
525 mti->mti_svname, obd_export_nid2str(req->rq_export));
526 rc = mgs_check_target(obd, mti);
527 /* above will set appropriate mti flags */
529 /* Nothing wrong, or fatal error */
530 GOTO(out_nolock, rc);
532 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
533 && (rc = mgs_check_failover_reg(mti)))
534 GOTO(out_nolock, rc);
537 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
539 if (mti->mti_flags & LDD_F_WRITECONF) {
540 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
541 mti->mti_stripe_index == 0) {
542 rc = mgs_erase_logs(obd, mti->mti_fsname);
543 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
544 "request. All servers must be restarted "
545 "in order to regenerate the logs."
546 "\n", obd->obd_name, mti->mti_fsname);
547 } else if (mti->mti_flags &
548 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
549 rc = mgs_erase_log(obd, mti->mti_svname);
550 LCONSOLE_WARN("%s: Regenerating %s log by user "
552 obd->obd_name, mti->mti_svname);
554 mti->mti_flags |= LDD_F_UPDATE;
555 /* Erased logs means start from scratch. */
556 mti->mti_flags &= ~LDD_F_UPGRADE14;
559 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
561 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
562 GOTO(out_nolock, rc);
566 * Log writing contention is handled by the fsdb_mutex.
568 * It should be alright if someone was reading while we were
569 * updating the logs - if we revoke at the end they will just update
570 * from where they left off.
574 if (mti->mti_flags & LDD_F_UPGRADE14) {
575 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
577 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
581 /* We're good to go */
582 mti->mti_flags |= LDD_F_UPDATE;
586 if (mti->mti_flags & LDD_F_UPDATE) {
587 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
588 mti->mti_stripe_index);
590 /* create or update the target log
591 and update the client/mdt logs */
592 rc = mgs_write_log_target(obd, mti, fsdb);
594 CERROR("Failed to write %s log (%d)\n",
595 mti->mti_svname, rc);
599 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
600 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
602 mti->mti_flags |= LDD_F_REWRITE_LDD;
606 mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
609 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
610 mti->mti_stripe_index, rc);
613 /* we need an error flag to tell the target what's going on,
614 * instead of just doing it by error code only. */
615 mti->mti_flags |= LDD_F_ERROR;
617 rc = req_capsule_server_pack(&req->rq_pill);
621 /* send back the whole mti in the reply */
622 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
625 /* Flush logs to disk */
626 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
630 static int mgs_set_info_rpc(struct ptlrpc_request *req)
632 struct obd_device *obd = req->rq_export->exp_obd;
633 struct mgs_send_param *msp, *rep_msp;
635 struct lustre_cfg_bufs bufs;
636 struct lustre_cfg *lcfg;
637 char fsname[MTI_NAME_MAXLEN];
640 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
643 /* Construct lustre_cfg structure to pass to function mgs_setparam */
644 lustre_cfg_bufs_reset(&bufs, NULL);
645 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
646 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
647 rc = mgs_setparam(obd, lcfg, fsname);
649 CERROR("Error %d in setting the parameter %s for fs %s\n",
650 rc, msp->mgs_param, fsname);
654 lustre_cfg_free(lcfg);
656 rc = req_capsule_server_pack(&req->rq_pill);
658 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
664 static int mgs_config_read(struct ptlrpc_request *req)
666 struct mgs_config_body *body;
670 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
674 switch (body->mcb_type) {
675 case CONFIG_T_RECOVER:
676 rc = mgs_get_ir_logs(req);
679 case CONFIG_T_CONFIG:
692 * similar as in ost_connect_check_sptlrpc()
694 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
696 struct obd_export *exp = req->rq_export;
697 struct obd_device *obd = exp->exp_obd;
699 struct sptlrpc_flavor flvr;
702 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
703 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
707 cfs_mutex_lock(&fsdb->fsdb_mutex);
708 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
709 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
712 /* by defualt allow any flavors */
713 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
715 cfs_mutex_unlock(&fsdb->fsdb_mutex);
717 cfs_spin_lock(&exp->exp_lock);
719 exp->exp_sp_peer = req->rq_sp_from;
720 exp->exp_flvr = flvr;
722 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
723 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
724 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
725 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
726 libcfs_nid2str(req->rq_peer.nid));
730 cfs_spin_unlock(&exp->exp_lock);
732 if (exp->exp_sp_peer != req->rq_sp_from) {
733 CERROR("RPC source %s doesn't match %s\n",
734 sptlrpc_part2name(req->rq_sp_from),
735 sptlrpc_part2name(exp->exp_sp_peer));
738 rc = sptlrpc_target_export_check(exp, req);
745 /* Called whenever a target cleans up. */
746 /* XXX - Currently unused */
747 static int mgs_handle_target_del(struct ptlrpc_request *req)
750 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
754 /* XXX - Currently unused */
755 static int mgs_handle_exception(struct ptlrpc_request *req)
758 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
763 * For old clients there is no direct way of knowing which filesystems
764 * a client is operating at the MGS side. But we need to pick up those
765 * clients so that the MGS can mark the corresponding filesystem as
766 * non-IR capable because old clients are not ready to be notified.
768 * This is why we have this _hack_ function. We detect the filesystem's
769 * name by hacking llog operation which is currently used by the clients
770 * to fetch configuration logs. At present this is fine because this is
771 * the ONLY llog operation between mgc and the MGS.
773 * If extra llog operation is going to be added, this function needs fixing.
775 * If releases prior than 2.0 are not supported, we can remove this function.
777 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
784 /* XXX: We suppose that llog at mgs is only used for
785 * fetching file system log */
786 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
787 if (logname == NULL) {
788 CERROR("No logname, is llog on MGS used for something else?\n");
792 ptr = strchr(logname, '-');
793 rc = (int)(ptr - logname);
794 if (ptr == NULL || rc >= sizeof(fsname)) {
795 CERROR("Invalid logname received: %s\n", logname);
799 strncpy(fsname, logname, rc);
801 rc = mgs_fsc_attach(req->rq_export, fsname);
802 if (rc < 0 && rc != -EEXIST)
803 CERROR("add fs client %s returns %d\n", fsname, rc);
808 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
809 int mgs_handle(struct ptlrpc_request *req)
811 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
815 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
816 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
817 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
820 LASSERT(current->journal_info == NULL);
821 opc = lustre_msg_get_opc(req->rq_reqmsg);
823 if (opc == SEC_CTX_INIT ||
824 opc == SEC_CTX_INIT_CONT ||
828 if (opc != MGS_CONNECT) {
829 if (!class_connected_export(req->rq_export)) {
830 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
831 req->rq_status = -ENOTCONN;
832 GOTO(out, rc = -ENOTCONN);
838 DEBUG_REQ(D_MGS, req, "connect");
839 /* MGS and MDS have same request format for connect */
840 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
841 rc = target_handle_connect(req);
843 rc = mgs_connect_check_sptlrpc(req);
845 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
846 /* Make clients trying to reconnect after a MGS restart
847 happy; also requires obd_replayable */
848 lustre_msg_add_op_flags(req->rq_repmsg,
849 MSG_CONNECT_RECONNECT);
852 DEBUG_REQ(D_MGS, req, "disconnect");
853 /* MGS and MDS have same request format for disconnect */
854 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
855 rc = target_handle_disconnect(req);
856 req->rq_status = rc; /* superfluous? */
859 DEBUG_REQ(D_MGS, req, "exception");
860 rc = mgs_handle_exception(req);
863 DEBUG_REQ(D_MGS, req, "target add");
864 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
865 rc = mgs_handle_target_reg(req);
868 DEBUG_REQ(D_MGS, req, "target del");
869 rc = mgs_handle_target_del(req);
872 DEBUG_REQ(D_MGS, req, "set_info");
873 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
874 rc = mgs_set_info_rpc(req);
876 case MGS_CONFIG_READ:
877 DEBUG_REQ(D_MGS, req, "read config");
878 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
879 rc = mgs_config_read(req);
882 DEBUG_REQ(D_MGS, req, "enqueue");
883 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
884 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
885 ldlm_server_blocking_ast, NULL);
887 case LDLM_BL_CALLBACK:
888 case LDLM_CP_CALLBACK:
889 DEBUG_REQ(D_MGS, req, "callback");
890 CERROR("callbacks should not happen on MGS\n");
895 DEBUG_REQ(D_INFO, req, "ping");
896 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
897 rc = target_handle_ping(req);
900 DEBUG_REQ(D_MGS, req, "log cancel");
901 rc = -ENOTSUPP; /* la la la */
904 case LLOG_ORIGIN_HANDLE_CREATE:
905 DEBUG_REQ(D_MGS, req, "llog_init");
906 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
907 rc = llog_origin_handle_create(req);
909 (void)mgs_handle_fslog_hack(req);
911 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
912 DEBUG_REQ(D_MGS, req, "llog next block");
913 req_capsule_set(&req->rq_pill,
914 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
915 rc = llog_origin_handle_next_block(req);
917 case LLOG_ORIGIN_HANDLE_READ_HEADER:
918 DEBUG_REQ(D_MGS, req, "llog read header");
919 req_capsule_set(&req->rq_pill,
920 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
921 rc = llog_origin_handle_read_header(req);
923 case LLOG_ORIGIN_HANDLE_CLOSE:
924 DEBUG_REQ(D_MGS, req, "llog close");
925 rc = llog_origin_handle_close(req);
928 DEBUG_REQ(D_MGS, req, "llog catinfo");
929 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
930 rc = llog_catinfo(req);
933 req->rq_status = -ENOTSUPP;
934 rc = ptlrpc_error(req);
938 LASSERT(current->journal_info == NULL);
941 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
944 target_send_reply(req, rc, fail);
948 static inline int mgs_init_export(struct obd_export *exp)
950 struct mgs_export_data *data = &exp->u.eu_mgs_data;
952 /* init mgs_export_data for fsc */
953 cfs_spin_lock_init(&data->med_lock);
954 CFS_INIT_LIST_HEAD(&data->med_clients);
956 cfs_spin_lock(&exp->exp_lock);
957 exp->exp_connecting = 1;
958 cfs_spin_unlock(&exp->exp_lock);
960 /* self-export doesn't need client data and ldlm initialization */
961 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
962 &exp->exp_client_uuid)))
964 return ldlm_init_export(exp);
967 static inline int mgs_destroy_export(struct obd_export *exp)
971 target_destroy_export(exp);
972 mgs_client_free(exp);
974 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
975 &exp->exp_client_uuid)))
978 ldlm_destroy_export(exp);
983 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
988 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
996 strcpy(poolname, ptr);
1001 static int mgs_iocontrol_pool(struct obd_device *obd,
1002 struct obd_ioctl_data *data)
1005 struct lustre_cfg *lcfg = NULL;
1006 struct llog_rec_hdr rec;
1007 char *fsname = NULL;
1008 char *poolname = NULL;
1011 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
1015 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1016 if (poolname == NULL) {
1020 rec.lrh_len = llog_data_len(data->ioc_plen1);
1022 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1023 rec.lrh_type = OBD_CFG_REC;
1025 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1030 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1035 OBD_ALLOC(lcfg, data->ioc_plen1);
1037 GOTO(out_pool, rc = -ENOMEM);
1039 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1040 GOTO(out_pool, rc = -EFAULT);
1042 if (lcfg->lcfg_bufcount < 2) {
1043 GOTO(out_pool, rc = -EFAULT);
1046 /* first arg is always <fsname>.<poolname> */
1047 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
1050 switch (lcfg->lcfg_command) {
1051 case LCFG_POOL_NEW: {
1052 if (lcfg->lcfg_bufcount != 2)
1054 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
1058 case LCFG_POOL_ADD: {
1059 if (lcfg->lcfg_bufcount != 3)
1061 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1062 lustre_cfg_string(lcfg, 2));
1065 case LCFG_POOL_REM: {
1066 if (lcfg->lcfg_bufcount != 3)
1068 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1069 lustre_cfg_string(lcfg, 2));
1072 case LCFG_POOL_DEL: {
1073 if (lcfg->lcfg_bufcount != 2)
1075 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1086 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1087 rc, lcfg->lcfg_command, fsname, poolname);
1093 OBD_FREE(lcfg, data->ioc_plen1);
1096 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1098 if (poolname != NULL)
1099 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1104 /* from mdt_iocontrol */
1105 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1106 void *karg, void *uarg)
1108 struct obd_device *obd = exp->exp_obd;
1109 struct obd_ioctl_data *data = karg;
1110 struct lvfs_run_ctxt saved;
1114 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1118 case OBD_IOC_PARAM: {
1119 struct lustre_cfg *lcfg;
1120 struct llog_rec_hdr rec;
1121 char fsname[MTI_NAME_MAXLEN];
1123 rec.lrh_len = llog_data_len(data->ioc_plen1);
1125 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1126 rec.lrh_type = OBD_CFG_REC;
1128 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1132 OBD_ALLOC(lcfg, data->ioc_plen1);
1135 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1136 GOTO(out_free, rc = -EFAULT);
1138 if (lcfg->lcfg_bufcount < 1)
1139 GOTO(out_free, rc = -EINVAL);
1141 rc = mgs_setparam(obd, lcfg, fsname);
1143 CERROR("setparam err %d\n", rc);
1147 OBD_FREE(lcfg, data->ioc_plen1);
1151 case OBD_IOC_POOL: {
1152 RETURN(mgs_iocontrol_pool(obd, data));
1155 case OBD_IOC_DUMP_LOG: {
1156 struct llog_ctxt *ctxt;
1157 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1158 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1159 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1160 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1161 llog_ctxt_put(ctxt);
1166 case OBD_IOC_LLOG_CHECK:
1167 case OBD_IOC_LLOG_INFO:
1168 case OBD_IOC_LLOG_PRINT: {
1169 struct llog_ctxt *ctxt;
1170 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1172 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1173 rc = llog_ioctl(ctxt, cmd, data);
1174 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1175 llog_ctxt_put(ctxt);
1181 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1187 /* use obd ops to offer management infrastructure */
1188 static struct obd_ops mgs_obd_ops = {
1189 .o_owner = THIS_MODULE,
1190 .o_connect = mgs_connect,
1191 .o_reconnect = mgs_reconnect,
1192 .o_disconnect = mgs_disconnect,
1193 .o_setup = mgs_setup,
1194 .o_precleanup = mgs_precleanup,
1195 .o_cleanup = mgs_cleanup,
1196 .o_init_export = mgs_init_export,
1197 .o_destroy_export = mgs_destroy_export,
1198 .o_iocontrol = mgs_iocontrol,
1199 .o_llog_init = mgs_llog_init,
1200 .o_llog_finish = mgs_llog_finish
1203 static int __init mgs_init(void)
1205 struct lprocfs_static_vars lvars;
1207 lprocfs_mgs_init_vars(&lvars);
1208 class_register_type(&mgs_obd_ops, NULL,
1209 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1214 static void /*__exit*/ mgs_exit(void)
1216 class_unregister_type(LUSTRE_MGS_NAME);
1219 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1220 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1221 MODULE_LICENSE("GPL");
1223 module_init(mgs_init);
1224 module_exit(mgs_exit);