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_device *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(lsi->lsi_fstype);
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;
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(mgs);
230 cfs_mutex_init(&mgs->mgs_mutex);
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(mgs);
237 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
242 conf = (typeof(conf)) {
243 .psc_name = LUSTRE_MGS_NAME,
244 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
246 .bc_nbufs = MGS_NBUFS,
247 .bc_buf_size = MGS_BUFSIZE,
248 .bc_req_max_size = MGS_MAXREQSIZE,
249 .bc_rep_max_size = MGS_MAXREPSIZE,
250 .bc_req_portal = MGS_REQUEST_PORTAL,
251 .bc_rep_portal = MGC_REPLY_PORTAL,
254 .tc_thr_name = "ll_mgs",
255 .tc_nthrs_init = MGS_NTHRS_INIT,
256 .tc_nthrs_max = MGS_NTHRS_MAX,
257 .tc_ctx_tags = LCT_MG_THREAD,
260 .so_req_handler = mgs_handle,
261 .so_req_printer = target_print_req,
264 /* Start the service threads */
265 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
266 if (IS_ERR(mgs->mgs_service)) {
267 rc = PTR_ERR(mgs->mgs_service);
268 CERROR("failed to start service: %d\n", rc);
272 ping_evictor_start();
274 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
279 lproc_mgs_cleanup(mgs);
280 obd_llog_finish(obd, 0);
282 /* No extra cleanup needed for llog_init_commit_thread() */
285 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
286 obd->obd_namespace = NULL;
288 fsfilt_put_ops(obd->obd_fsops);
290 server_put_mount(obd->obd_name, mnt);
295 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
297 struct mgs_device *mgs = &obd->u.mgs;
302 case OBD_CLEANUP_EARLY:
304 case OBD_CLEANUP_EXPORTS:
306 ptlrpc_unregister_service(mgs->mgs_service);
307 mgs_cleanup_fsdb_list(mgs);
308 rc = obd_llog_finish(obd, 0);
309 lproc_mgs_cleanup(mgs);
316 * Performs cleanup procedures for passed \a obd given it is mgs obd.
318 static int mgs_cleanup(struct obd_device *obd)
320 struct mgs_device *mgs = &obd->u.mgs;
323 if (mgs->mgs_sb == NULL)
328 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
331 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
332 obd->obd_namespace = NULL;
334 fsfilt_put_ops(obd->obd_fsops);
336 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
340 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
345 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
346 LDLM_FL_BLOCK_CONV))) {
347 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
348 struct lustre_handle lockh;
350 /* clear the bit before lock put */
351 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
353 ldlm_lock2handle(lock, &lockh);
354 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
357 RETURN(ldlm_completion_ast(lock, flags, cbdata));
360 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
365 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
366 LDLM_FL_BLOCK_CONV))) {
369 /* l_ast_data is used as a marker to avoid cancel ldlm lock
370 * twice. See LU-1259. */
371 lock_res_and_lock(lock);
372 fsdb = (struct fs_db *)lock->l_ast_data;
373 lock->l_ast_data = NULL;
374 unlock_res_and_lock(lock);
377 struct lustre_handle lockh;
379 mgs_ir_notify_complete(fsdb);
381 ldlm_lock2handle(lock, &lockh);
382 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
386 RETURN(ldlm_completion_ast(lock, flags, cbdata));
389 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
391 ldlm_completion_callback cp = NULL;
392 struct lustre_handle lockh = { 0 };
393 struct ldlm_res_id res_id;
394 int flags = LDLM_FL_ATOMIC_CB;
398 LASSERT(fsdb->fsdb_name[0] != '\0');
399 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
403 case CONFIG_T_CONFIG:
404 cp = mgs_completion_ast_config;
405 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
408 case CONFIG_T_RECOVER:
409 cp = mgs_completion_ast_ir;
416 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
417 &res_id, LDLM_PLAIN, NULL, LCK_EX,
418 &flags, ldlm_blocking_ast, cp,
419 NULL, fsdb, 0, NULL, &lockh);
420 if (rc != ELDLM_OK) {
421 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
422 le64_to_cpu(res_id.name[0]),
423 le64_to_cpu(res_id.name[1]), rc);
425 if (type == CONFIG_T_CONFIG)
426 cfs_clear_bit(FSDB_REVOKING_LOCK,
429 /* lock has been cancelled in completion_ast. */
438 static int mgs_check_target(const struct lu_env *env,
439 struct mgs_device *mgs,
440 struct mgs_target_info *mti)
445 rc = mgs_check_index(env, mgs, mti);
447 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
448 "this MGS does not know about it, preventing "
449 "registration.\n", mti->mti_svname);
451 } else if (rc == -1) {
452 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
453 "disappeared! Regenerating all logs.\n",
455 mti->mti_flags |= LDD_F_WRITECONF;
458 /* Index is correctly marked as used */
460 /* If the logs don't contain the mti_nids then add
461 them as failover nids */
462 rc = mgs_check_failnid(env, mgs, mti);
468 /* Ensure this is not a failover node that is connecting first*/
469 static int mgs_check_failover_reg(struct mgs_target_info *mti)
475 ptr = mti->mti_params;
476 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
477 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
478 for (i = 0; i < mti->mti_nid_count; i++) {
479 if (nid == mti->mti_nids[i]) {
480 LCONSOLE_WARN("Denying initial registra"
481 "tion attempt from nid %s"
482 ", specified as failover"
483 "\n",libcfs_nid2str(nid));
484 return -EADDRNOTAVAIL;
492 /* Called whenever a target starts up. Flags indicate first connect, etc. */
493 static int mgs_handle_target_reg(struct ptlrpc_request *req)
495 struct obd_device *obd = req->rq_export->exp_obd;
496 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
497 struct lu_env *env = req->rq_svc_thread->t_env;
498 struct mgs_target_info *mti, *rep_mti;
504 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
506 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
508 opc = mti->mti_flags & LDD_F_OPC_MASK;
509 if (opc == LDD_F_OPC_READY) {
510 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
511 mti->mti_fsname, mti->mti_stripe_index);
512 rc = mgs_ir_update(env, mgs, mti);
514 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
515 CERROR("Update IR return with %d(ignore and IR "
518 GOTO(out_nolock, rc);
521 /* Do not support unregistering right now. */
522 if (opc != LDD_F_OPC_REG)
523 GOTO(out_nolock, rc = -EINVAL);
525 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
526 mti->mti_fsname, mti->mti_stripe_index);
528 if (mti->mti_flags & LDD_F_NEED_INDEX)
529 mti->mti_flags |= LDD_F_WRITECONF;
531 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
533 /* We're just here as a startup ping. */
534 CDEBUG(D_MGS, "Server %s is running on %s\n",
535 mti->mti_svname, obd_export_nid2str(req->rq_export));
536 rc = mgs_check_target(env, mgs, mti);
537 /* above will set appropriate mti flags */
539 /* Nothing wrong, or fatal error */
540 GOTO(out_nolock, rc);
542 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
543 && (rc = mgs_check_failover_reg(mti)))
544 GOTO(out_nolock, rc);
547 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
549 if (mti->mti_flags & LDD_F_WRITECONF) {
550 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
551 mti->mti_stripe_index == 0) {
552 rc = mgs_erase_logs(env, mgs, mti->mti_fsname);
553 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
554 "request. All servers must be restarted "
555 "in order to regenerate the logs."
556 "\n", obd->obd_name, mti->mti_fsname);
557 } else if (mti->mti_flags &
558 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
559 rc = mgs_erase_log(env, mgs, mti->mti_svname);
560 LCONSOLE_WARN("%s: Regenerating %s log by user "
562 obd->obd_name, mti->mti_svname);
564 mti->mti_flags |= LDD_F_UPDATE;
565 /* Erased logs means start from scratch. */
566 mti->mti_flags &= ~LDD_F_UPGRADE14;
569 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
571 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
572 GOTO(out_nolock, rc);
576 * Log writing contention is handled by the fsdb_mutex.
578 * It should be alright if someone was reading while we were
579 * updating the logs - if we revoke at the end they will just update
580 * from where they left off.
584 if (mti->mti_flags & LDD_F_UPGRADE14) {
585 rc = mgs_upgrade_sv_14(env, mgs, mti, fsdb);
587 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
591 /* We're good to go */
592 mti->mti_flags |= LDD_F_UPDATE;
596 if (mti->mti_flags & LDD_F_UPDATE) {
597 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
598 mti->mti_stripe_index);
600 /* create or update the target log
601 and update the client/mdt logs */
602 rc = mgs_write_log_target(env, mgs, mti, fsdb);
604 CERROR("Failed to write %s log (%d)\n",
605 mti->mti_svname, rc);
609 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
610 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
612 mti->mti_flags |= LDD_F_REWRITE_LDD;
616 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
619 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
620 mti->mti_stripe_index, rc);
623 /* we need an error flag to tell the target what's going on,
624 * instead of just doing it by error code only. */
625 mti->mti_flags |= LDD_F_ERROR;
627 rc = req_capsule_server_pack(&req->rq_pill);
631 /* send back the whole mti in the reply */
632 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
635 /* Flush logs to disk */
636 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
640 static int mgs_set_info_rpc(struct ptlrpc_request *req)
642 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
643 struct lu_env *env = req->rq_svc_thread->t_env;
644 struct mgs_send_param *msp, *rep_msp;
645 struct mgs_thread_info *mgi = mgs_env_info(env);
647 struct lustre_cfg *lcfg;
650 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
653 /* Construct lustre_cfg structure to pass to function mgs_setparam */
654 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
655 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
656 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
657 rc = mgs_setparam(env, mgs, lcfg, mgi->mgi_fsname);
659 CERROR("Error %d in setting the parameter %s for fs %s\n",
660 rc, msp->mgs_param, mgi->mgi_fsname);
664 lustre_cfg_free(lcfg);
666 rc = req_capsule_server_pack(&req->rq_pill);
668 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
674 static int mgs_config_read(struct ptlrpc_request *req)
676 struct mgs_config_body *body;
680 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
684 switch (body->mcb_type) {
685 case CONFIG_T_RECOVER:
686 rc = mgs_get_ir_logs(req);
689 case CONFIG_T_CONFIG:
702 * similar as in ost_connect_check_sptlrpc()
704 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
706 struct obd_export *exp = req->rq_export;
707 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
708 struct lu_env *env = req->rq_svc_thread->t_env;
710 struct sptlrpc_flavor flvr;
713 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
714 rc = mgs_find_or_make_fsdb(env, mgs, MGSSELF_NAME, &fsdb);
718 cfs_mutex_lock(&fsdb->fsdb_mutex);
719 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
720 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
723 /* by defualt allow any flavors */
724 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
726 cfs_mutex_unlock(&fsdb->fsdb_mutex);
728 cfs_spin_lock(&exp->exp_lock);
730 exp->exp_sp_peer = req->rq_sp_from;
731 exp->exp_flvr = flvr;
733 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
734 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
735 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
736 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
737 libcfs_nid2str(req->rq_peer.nid));
741 cfs_spin_unlock(&exp->exp_lock);
743 if (exp->exp_sp_peer != req->rq_sp_from) {
744 CERROR("RPC source %s doesn't match %s\n",
745 sptlrpc_part2name(req->rq_sp_from),
746 sptlrpc_part2name(exp->exp_sp_peer));
749 rc = sptlrpc_target_export_check(exp, req);
756 /* Called whenever a target cleans up. */
757 /* XXX - Currently unused */
758 static int mgs_handle_target_del(struct ptlrpc_request *req)
761 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
765 /* XXX - Currently unused */
766 static int mgs_handle_exception(struct ptlrpc_request *req)
769 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
774 * For old clients there is no direct way of knowing which filesystems
775 * a client is operating at the MGS side. But we need to pick up those
776 * clients so that the MGS can mark the corresponding filesystem as
777 * non-IR capable because old clients are not ready to be notified.
779 * This is why we have this _hack_ function. We detect the filesystem's
780 * name by hacking llog operation which is currently used by the clients
781 * to fetch configuration logs. At present this is fine because this is
782 * the ONLY llog operation between mgc and the MGS.
784 * If extra llog operation is going to be added, this function needs fixing.
786 * If releases prior than 2.0 are not supported, we can remove this function.
788 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
795 /* XXX: We suppose that llog at mgs is only used for
796 * fetching file system log */
797 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
798 if (logname == NULL) {
799 CERROR("No logname, is llog on MGS used for something else?\n");
803 ptr = strchr(logname, '-');
804 rc = (int)(ptr - logname);
805 if (ptr == NULL || rc >= sizeof(fsname)) {
806 CERROR("Invalid logname received: %s\n", logname);
810 strncpy(fsname, logname, rc);
812 rc = mgs_fsc_attach(req->rq_svc_thread->t_env, req->rq_export, fsname);
813 if (rc < 0 && rc != -EEXIST)
814 CERROR("add fs client %s returns %d\n", fsname, rc);
819 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
820 int mgs_handle(struct ptlrpc_request *req)
822 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
826 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
827 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
828 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
831 LASSERT(current->journal_info == NULL);
832 opc = lustre_msg_get_opc(req->rq_reqmsg);
834 if (opc == SEC_CTX_INIT ||
835 opc == SEC_CTX_INIT_CONT ||
839 if (opc != MGS_CONNECT) {
840 if (!class_connected_export(req->rq_export)) {
841 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
842 req->rq_status = -ENOTCONN;
843 GOTO(out, rc = -ENOTCONN);
849 DEBUG_REQ(D_MGS, req, "connect");
850 /* MGS and MDS have same request format for connect */
851 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
852 rc = target_handle_connect(req);
854 rc = mgs_connect_check_sptlrpc(req);
856 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
857 /* Make clients trying to reconnect after a MGS restart
858 happy; also requires obd_replayable */
859 lustre_msg_add_op_flags(req->rq_repmsg,
860 MSG_CONNECT_RECONNECT);
863 DEBUG_REQ(D_MGS, req, "disconnect");
864 /* MGS and MDS have same request format for disconnect */
865 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
866 rc = target_handle_disconnect(req);
867 req->rq_status = rc; /* superfluous? */
870 DEBUG_REQ(D_MGS, req, "exception");
871 rc = mgs_handle_exception(req);
874 DEBUG_REQ(D_MGS, req, "target add");
875 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
876 rc = mgs_handle_target_reg(req);
879 DEBUG_REQ(D_MGS, req, "target del");
880 rc = mgs_handle_target_del(req);
883 DEBUG_REQ(D_MGS, req, "set_info");
884 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
885 rc = mgs_set_info_rpc(req);
887 case MGS_CONFIG_READ:
888 DEBUG_REQ(D_MGS, req, "read config");
889 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
890 rc = mgs_config_read(req);
893 DEBUG_REQ(D_MGS, req, "enqueue");
894 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
895 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
896 ldlm_server_blocking_ast, NULL);
898 case LDLM_BL_CALLBACK:
899 case LDLM_CP_CALLBACK:
900 DEBUG_REQ(D_MGS, req, "callback");
901 CERROR("callbacks should not happen on MGS\n");
906 DEBUG_REQ(D_INFO, req, "ping");
907 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
908 rc = target_handle_ping(req);
911 DEBUG_REQ(D_MGS, req, "log cancel");
912 rc = -ENOTSUPP; /* la la la */
915 case LLOG_ORIGIN_HANDLE_CREATE:
916 DEBUG_REQ(D_MGS, req, "llog_open");
917 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
918 rc = llog_origin_handle_open(req);
920 (void)mgs_handle_fslog_hack(req);
922 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
923 DEBUG_REQ(D_MGS, req, "llog next block");
924 req_capsule_set(&req->rq_pill,
925 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
926 rc = llog_origin_handle_next_block(req);
928 case LLOG_ORIGIN_HANDLE_READ_HEADER:
929 DEBUG_REQ(D_MGS, req, "llog read header");
930 req_capsule_set(&req->rq_pill,
931 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
932 rc = llog_origin_handle_read_header(req);
934 case LLOG_ORIGIN_HANDLE_CLOSE:
935 DEBUG_REQ(D_MGS, req, "llog close");
936 rc = llog_origin_handle_close(req);
942 LASSERT(current->journal_info == NULL);
944 DEBUG_REQ(D_MGS, req, "MGS fail to handle opc = %d: rc = %d\n",
947 rc = ptlrpc_error(req);
951 target_send_reply(req, rc, fail);
955 static inline int mgs_init_export(struct obd_export *exp)
957 struct mgs_export_data *data = &exp->u.eu_mgs_data;
959 /* init mgs_export_data for fsc */
960 cfs_spin_lock_init(&data->med_lock);
961 CFS_INIT_LIST_HEAD(&data->med_clients);
963 cfs_spin_lock(&exp->exp_lock);
964 exp->exp_connecting = 1;
965 cfs_spin_unlock(&exp->exp_lock);
967 /* self-export doesn't need client data and ldlm initialization */
968 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
969 &exp->exp_client_uuid)))
971 return ldlm_init_export(exp);
974 static inline int mgs_destroy_export(struct obd_export *exp)
978 target_destroy_export(exp);
979 mgs_client_free(exp);
981 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
982 &exp->exp_client_uuid)))
985 ldlm_destroy_export(exp);
990 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
995 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
1003 strcpy(poolname, ptr);
1008 static int mgs_iocontrol_pool(const struct lu_env *env,
1009 struct mgs_device *mgs,
1010 struct obd_ioctl_data *data)
1012 struct mgs_thread_info *mgi = mgs_env_info(env);
1014 struct lustre_cfg *lcfg = NULL;
1015 struct llog_rec_hdr rec;
1016 char *poolname = NULL;
1019 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1020 if (poolname == NULL)
1022 rec.lrh_len = llog_data_len(data->ioc_plen1);
1024 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1025 rec.lrh_type = OBD_CFG_REC;
1027 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1032 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1037 OBD_ALLOC(lcfg, data->ioc_plen1);
1039 GOTO(out_pool, rc = -ENOMEM);
1041 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1042 GOTO(out_pool, rc = -EFAULT);
1044 if (lcfg->lcfg_bufcount < 2) {
1045 GOTO(out_pool, rc = -EFAULT);
1048 /* first arg is always <fsname>.<poolname> */
1049 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
1052 switch (lcfg->lcfg_command) {
1054 if (lcfg->lcfg_bufcount != 2)
1056 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
1060 if (lcfg->lcfg_bufcount != 3)
1062 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
1063 poolname, lustre_cfg_string(lcfg, 2));
1066 if (lcfg->lcfg_bufcount != 3)
1068 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
1069 poolname, lustre_cfg_string(lcfg, 2));
1072 if (lcfg->lcfg_bufcount != 2)
1074 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
1082 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1083 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
1089 OBD_FREE(lcfg, data->ioc_plen1);
1091 if (poolname != NULL)
1092 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1097 /* from mdt_iocontrol */
1098 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1099 void *karg, void *uarg)
1101 struct mgs_device *mgs = exp2mgs_dev(exp);
1102 struct obd_ioctl_data *data = karg;
1103 struct lvfs_run_ctxt saved;
1108 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1110 rc = lu_env_init(&env, LCT_MG_THREAD);
1116 case OBD_IOC_PARAM: {
1117 struct mgs_thread_info *mgi = mgs_env_info(&env);
1118 struct lustre_cfg *lcfg;
1119 struct llog_rec_hdr rec;
1121 rec.lrh_len = llog_data_len(data->ioc_plen1);
1123 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1124 rec.lrh_type = OBD_CFG_REC;
1126 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1127 GOTO(out, rc = -EINVAL);
1130 OBD_ALLOC(lcfg, data->ioc_plen1);
1132 GOTO(out, rc = -ENOMEM);
1133 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1134 GOTO(out_free, rc = -EFAULT);
1136 if (lcfg->lcfg_bufcount < 1)
1137 GOTO(out_free, rc = -EINVAL);
1139 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
1141 CERROR("setparam err %d\n", rc);
1145 OBD_FREE(lcfg, data->ioc_plen1);
1150 rc = mgs_iocontrol_pool(&env, mgs, data);
1153 case OBD_IOC_DUMP_LOG: {
1154 struct llog_ctxt *ctxt;
1155 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1156 push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
1157 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1158 pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
1159 llog_ctxt_put(ctxt);
1164 case OBD_IOC_LLOG_CHECK:
1165 case OBD_IOC_LLOG_INFO:
1166 case OBD_IOC_LLOG_PRINT: {
1167 struct llog_ctxt *ctxt;
1168 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1170 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1171 rc = llog_ioctl(ctxt, cmd, data);
1172 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1173 llog_ctxt_put(ctxt);
1179 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1188 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1189 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1191 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1193 /* use obd ops to offer management infrastructure */
1194 static struct obd_ops mgs_obd_ops = {
1195 .o_owner = THIS_MODULE,
1196 .o_connect = mgs_connect,
1197 .o_reconnect = mgs_reconnect,
1198 .o_disconnect = mgs_disconnect,
1199 .o_setup = mgs_setup,
1200 .o_precleanup = mgs_precleanup,
1201 .o_cleanup = mgs_cleanup,
1202 .o_init_export = mgs_init_export,
1203 .o_destroy_export = mgs_destroy_export,
1204 .o_iocontrol = mgs_iocontrol,
1205 .o_llog_init = mgs_llog_init,
1206 .o_llog_finish = mgs_llog_finish
1209 static int __init mgs_init(void)
1211 struct lprocfs_static_vars lvars;
1214 lprocfs_mgs_init_vars(&lvars);
1215 class_register_type(&mgs_obd_ops, NULL,
1216 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1218 LU_CONTEXT_KEY_INIT(&mgs_thread_key);
1219 rc = lu_context_key_register(&mgs_thread_key);
1221 class_unregister_type(LUSTRE_MGS_NAME);
1226 static void /*__exit*/ mgs_exit(void)
1228 class_unregister_type(LUSTRE_MGS_NAME);
1229 lu_context_key_degister(&mgs_thread_key);
1232 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1233 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1234 MODULE_LICENSE("GPL");
1236 module_init(mgs_init);
1237 module_exit(mgs_exit);