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 <lustre_param.h>
59 #include <lustre_log.h>
61 #include "mgs_internal.h"
63 /* Establish a connection to the MGS.*/
64 static int mgs_connect(const struct lu_env *env,
65 struct obd_export **exp, struct obd_device *obd,
66 struct obd_uuid *cluuid, struct obd_connect_data *data,
69 struct obd_export *lexp;
70 struct lustre_handle conn = { 0 };
74 if (!exp || !obd || !cluuid)
77 rc = class_connect(&conn, obd, cluuid);
81 lexp = class_conn2export(&conn);
83 GOTO(out, rc = -EFAULT);
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);
97 class_disconnect(lexp);
105 static int mgs_reconnect(const struct lu_env *env,
106 struct obd_export *exp, struct obd_device *obd,
107 struct obd_uuid *cluuid, struct obd_connect_data *data,
112 if (exp == NULL || obd == NULL || cluuid == NULL)
115 mgs_counter_incr(exp, LPROC_MGS_CONNECT);
118 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
119 exp->exp_connect_flags = data->ocd_connect_flags;
120 data->ocd_version = LUSTRE_VERSION_CODE;
123 RETURN(mgs_export_stats_init(obd, exp, localdata));
126 static int mgs_disconnect(struct obd_export *exp)
133 mgs_fsc_cleanup(exp);
135 class_export_get(exp);
136 mgs_counter_incr(exp, LPROC_MGS_DISCONNECT);
138 rc = server_disconnect_export(exp);
139 class_export_put(exp);
143 static int mgs_cleanup(struct obd_device *obd);
144 static int mgs_handle(struct ptlrpc_request *req);
146 static int mgs_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
147 struct obd_device *tgt, int *index)
152 LASSERT(olg == &obd->obd_olg);
153 rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
158 static int mgs_llog_finish(struct obd_device *obd, int count)
160 struct llog_ctxt *ctxt;
164 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
166 rc = llog_cleanup(ctxt);
171 /* Start the MGS obd */
172 static int mgs_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
174 static struct ptlrpc_service_conf conf;
175 struct lprocfs_static_vars lvars;
176 struct mgs_device *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(lsi->lsi_fstype);
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;
206 /* namespace for mgs llog */
207 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
208 LDLM_NAMESPACE_SERVER,
209 LDLM_NAMESPACE_MODEST,
211 if (obd->obd_namespace == NULL)
212 GOTO(err_ops, rc = -ENOMEM);
215 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
216 "mgs_ldlm_client", &obd->obd_ldlm_client);
218 rc = mgs_fs_setup(obd, mnt);
220 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
225 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
229 /* No recovery for MGC's */
230 obd->obd_replayable = 0;
232 /* Internal mgs setup */
233 mgs_init_fsdb_list(mgs);
234 cfs_mutex_init(&mgs->mgs_mutex);
235 mgs->mgs_start_time = cfs_time_current_sec();
238 lprocfs_mgs_init_vars(&lvars);
239 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
240 lproc_mgs_setup(mgs);
241 rc = lprocfs_alloc_md_stats(obd, LPROC_MGS_LAST);
246 conf = (typeof(conf)) {
247 .psc_name = LUSTRE_MGS_NAME,
248 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
250 .bc_nbufs = MGS_NBUFS,
251 .bc_buf_size = MGS_BUFSIZE,
252 .bc_req_max_size = MGS_MAXREQSIZE,
253 .bc_rep_max_size = MGS_MAXREPSIZE,
254 .bc_req_portal = MGS_REQUEST_PORTAL,
255 .bc_rep_portal = MGC_REPLY_PORTAL,
258 .tc_thr_name = "ll_mgs",
259 .tc_nthrs_init = MGS_NTHRS_INIT,
260 .tc_nthrs_max = MGS_NTHRS_MAX,
261 .tc_ctx_tags = LCT_MG_THREAD,
264 .so_req_handler = mgs_handle,
265 .so_req_printer = target_print_req,
268 /* Start the service threads */
269 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
270 if (IS_ERR(mgs->mgs_service)) {
271 rc = PTR_ERR(mgs->mgs_service);
272 CERROR("failed to start service: %d\n", rc);
276 ping_evictor_start();
278 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
283 lproc_mgs_cleanup(mgs);
284 obd_llog_finish(obd, 0);
286 /* No extra cleanup needed for llog_init_commit_thread() */
289 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
290 obd->obd_namespace = NULL;
292 fsfilt_put_ops(obd->obd_fsops);
294 server_put_mount(obd->obd_name, mnt);
299 static int mgs_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
301 struct mgs_device *mgs = &obd->u.mgs;
306 case OBD_CLEANUP_EARLY:
308 case OBD_CLEANUP_EXPORTS:
310 ptlrpc_unregister_service(mgs->mgs_service);
311 mgs_cleanup_fsdb_list(mgs);
312 rc = obd_llog_finish(obd, 0);
313 lproc_mgs_cleanup(mgs);
320 * Performs cleanup procedures for passed \a obd given it is mgs obd.
322 static int mgs_cleanup(struct obd_device *obd)
324 struct mgs_device *mgs = &obd->u.mgs;
327 if (mgs->mgs_sb == NULL)
332 server_put_mount(obd->obd_name, mgs->mgs_vfsmnt);
335 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
336 obd->obd_namespace = NULL;
338 fsfilt_put_ops(obd->obd_fsops);
340 LCONSOLE_INFO("%s has stopped.\n", obd->obd_name);
344 static int mgs_completion_ast_config(struct ldlm_lock *lock, int flags,
349 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
350 LDLM_FL_BLOCK_CONV))) {
351 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
352 struct lustre_handle lockh;
354 /* clear the bit before lock put */
355 cfs_clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
357 ldlm_lock2handle(lock, &lockh);
358 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
361 RETURN(ldlm_completion_ast(lock, flags, cbdata));
364 static int mgs_completion_ast_ir(struct ldlm_lock *lock, int flags,
369 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
370 LDLM_FL_BLOCK_CONV))) {
373 /* l_ast_data is used as a marker to avoid cancel ldlm lock
374 * twice. See LU-1259. */
375 lock_res_and_lock(lock);
376 fsdb = (struct fs_db *)lock->l_ast_data;
377 lock->l_ast_data = NULL;
378 unlock_res_and_lock(lock);
381 struct lustre_handle lockh;
383 mgs_ir_notify_complete(fsdb);
385 ldlm_lock2handle(lock, &lockh);
386 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
390 RETURN(ldlm_completion_ast(lock, flags, cbdata));
393 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
395 ldlm_completion_callback cp = NULL;
396 struct lustre_handle lockh = { 0 };
397 struct ldlm_res_id res_id;
398 int flags = LDLM_FL_ATOMIC_CB;
402 LASSERT(fsdb->fsdb_name[0] != '\0');
403 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
407 case CONFIG_T_CONFIG:
408 cp = mgs_completion_ast_config;
409 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
412 case CONFIG_T_RECOVER:
413 cp = mgs_completion_ast_ir;
420 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
421 &res_id, LDLM_PLAIN, NULL, LCK_EX,
422 &flags, ldlm_blocking_ast, cp,
423 NULL, fsdb, 0, NULL, &lockh);
424 if (rc != ELDLM_OK) {
425 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
426 le64_to_cpu(res_id.name[0]),
427 le64_to_cpu(res_id.name[1]), rc);
429 if (type == CONFIG_T_CONFIG)
430 cfs_clear_bit(FSDB_REVOKING_LOCK,
433 /* lock has been cancelled in completion_ast. */
442 static int mgs_check_target(const struct lu_env *env,
443 struct mgs_device *mgs,
444 struct mgs_target_info *mti)
449 rc = mgs_check_index(env, mgs, mti);
451 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
452 "this MGS does not know about it, preventing "
453 "registration.\n", mti->mti_svname);
455 } else if (rc == -1) {
456 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
457 "disappeared! Regenerating all logs.\n",
459 mti->mti_flags |= LDD_F_WRITECONF;
462 /* Index is correctly marked as used */
464 /* If the logs don't contain the mti_nids then add
465 them as failover nids */
466 rc = mgs_check_failnid(env, mgs, mti);
472 /* Ensure this is not a failover node that is connecting first*/
473 static int mgs_check_failover_reg(struct mgs_target_info *mti)
479 ptr = mti->mti_params;
480 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
481 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
482 for (i = 0; i < mti->mti_nid_count; i++) {
483 if (nid == mti->mti_nids[i]) {
484 LCONSOLE_WARN("Denying initial registra"
485 "tion attempt from nid %s"
486 ", specified as failover"
487 "\n",libcfs_nid2str(nid));
488 return -EADDRNOTAVAIL;
496 /* Called whenever a target starts up. Flags indicate first connect, etc. */
497 static int mgs_handle_target_reg(struct ptlrpc_request *req)
499 struct obd_device *obd = req->rq_export->exp_obd;
500 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
501 struct lu_env *env = req->rq_svc_thread->t_env;
502 struct mgs_target_info *mti, *rep_mti;
508 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
510 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
512 opc = mti->mti_flags & LDD_F_OPC_MASK;
513 if (opc == LDD_F_OPC_READY) {
514 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
515 mti->mti_fsname, mti->mti_stripe_index);
516 rc = mgs_ir_update(env, mgs, mti);
518 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
519 CERROR("Update IR return with %d(ignore and IR "
522 GOTO(out_nolock, rc);
525 /* Do not support unregistering right now. */
526 if (opc != LDD_F_OPC_REG)
527 GOTO(out_nolock, rc = -EINVAL);
529 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
530 mti->mti_fsname, mti->mti_stripe_index);
532 if (mti->mti_flags & LDD_F_NEED_INDEX)
533 mti->mti_flags |= LDD_F_WRITECONF;
535 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
537 /* We're just here as a startup ping. */
538 CDEBUG(D_MGS, "Server %s is running on %s\n",
539 mti->mti_svname, obd_export_nid2str(req->rq_export));
540 rc = mgs_check_target(env, mgs, mti);
541 /* above will set appropriate mti flags */
543 /* Nothing wrong, or fatal error */
544 GOTO(out_nolock, rc);
546 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
547 && (rc = mgs_check_failover_reg(mti)))
548 GOTO(out_nolock, rc);
551 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
553 if (mti->mti_flags & LDD_F_WRITECONF) {
554 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
555 mti->mti_stripe_index == 0) {
556 rc = mgs_erase_logs(env, mgs, mti->mti_fsname);
557 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
558 "request. All servers must be restarted "
559 "in order to regenerate the logs."
560 "\n", obd->obd_name, mti->mti_fsname);
561 } else if (mti->mti_flags &
562 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
563 rc = mgs_erase_log(env, mgs, mti->mti_svname);
564 LCONSOLE_WARN("%s: Regenerating %s log by user "
566 obd->obd_name, mti->mti_svname);
568 mti->mti_flags |= LDD_F_UPDATE;
569 /* Erased logs means start from scratch. */
570 mti->mti_flags &= ~LDD_F_UPGRADE14;
572 GOTO(out_nolock, rc);
575 rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
577 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
578 GOTO(out_nolock, rc);
582 * Log writing contention is handled by the fsdb_mutex.
584 * It should be alright if someone was reading while we were
585 * updating the logs - if we revoke at the end they will just update
586 * from where they left off.
589 if (mti->mti_flags & LDD_F_UPGRADE14) {
590 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
594 if (mti->mti_flags & LDD_F_UPDATE) {
595 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
596 mti->mti_stripe_index);
598 /* create or update the target log
599 and update the client/mdt logs */
600 rc = mgs_write_log_target(env, mgs, mti, fsdb);
602 CERROR("Failed to write %s log (%d)\n",
603 mti->mti_svname, rc);
607 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
608 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
610 mti->mti_flags |= LDD_F_REWRITE_LDD;
614 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
617 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
618 mti->mti_stripe_index, rc);
621 /* we need an error flag to tell the target what's going on,
622 * instead of just doing it by error code only. */
623 mti->mti_flags |= LDD_F_ERROR;
625 rc = req_capsule_server_pack(&req->rq_pill);
629 /* send back the whole mti in the reply */
630 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
633 /* Flush logs to disk */
634 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
638 static int mgs_set_info_rpc(struct ptlrpc_request *req)
640 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
641 struct lu_env *env = req->rq_svc_thread->t_env;
642 struct mgs_send_param *msp, *rep_msp;
643 struct mgs_thread_info *mgi = mgs_env_info(env);
645 struct lustre_cfg *lcfg;
648 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
651 /* Construct lustre_cfg structure to pass to function mgs_setparam */
652 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
653 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
654 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
656 GOTO(out, rc = PTR_ERR(lcfg));
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 rc = req_capsule_server_pack(&req->rq_pill);
666 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
670 lustre_cfg_free(lcfg);
675 static int mgs_config_read(struct ptlrpc_request *req)
677 struct mgs_config_body *body;
681 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
685 switch (body->mcb_type) {
686 case CONFIG_T_RECOVER:
687 rc = mgs_get_ir_logs(req);
690 case CONFIG_T_CONFIG:
703 * similar as in ost_connect_check_sptlrpc()
705 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
707 struct obd_export *exp = req->rq_export;
708 struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
709 struct lu_env *env = req->rq_svc_thread->t_env;
711 struct sptlrpc_flavor flvr;
714 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
715 rc = mgs_find_or_make_fsdb(env, mgs, MGSSELF_NAME, &fsdb);
719 cfs_mutex_lock(&fsdb->fsdb_mutex);
720 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
721 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
724 /* by defualt allow any flavors */
725 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
727 cfs_mutex_unlock(&fsdb->fsdb_mutex);
729 cfs_spin_lock(&exp->exp_lock);
731 exp->exp_sp_peer = req->rq_sp_from;
732 exp->exp_flvr = flvr;
734 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
735 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
736 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
737 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
738 libcfs_nid2str(req->rq_peer.nid));
742 cfs_spin_unlock(&exp->exp_lock);
744 if (exp->exp_sp_peer != req->rq_sp_from) {
745 CERROR("RPC source %s doesn't match %s\n",
746 sptlrpc_part2name(req->rq_sp_from),
747 sptlrpc_part2name(exp->exp_sp_peer));
750 rc = sptlrpc_target_export_check(exp, req);
757 /* Called whenever a target cleans up. */
758 /* XXX - Currently unused */
759 static int mgs_handle_target_del(struct ptlrpc_request *req)
762 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
766 /* XXX - Currently unused */
767 static int mgs_handle_exception(struct ptlrpc_request *req)
770 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
775 * For old clients there is no direct way of knowing which filesystems
776 * a client is operating at the MGS side. But we need to pick up those
777 * clients so that the MGS can mark the corresponding filesystem as
778 * non-IR capable because old clients are not ready to be notified.
780 * This is why we have this _hack_ function. We detect the filesystem's
781 * name by hacking llog operation which is currently used by the clients
782 * to fetch configuration logs. At present this is fine because this is
783 * the ONLY llog operation between mgc and the MGS.
785 * If extra llog operation is going to be added, this function needs fixing.
787 * If releases prior than 2.0 are not supported, we can remove this function.
789 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
796 /* XXX: We suppose that llog at mgs is only used for
797 * fetching file system log */
798 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
799 if (logname == NULL) {
800 CERROR("No logname, is llog on MGS used for something else?\n");
804 ptr = strchr(logname, '-');
805 rc = (int)(ptr - logname);
806 if (ptr == NULL || rc >= sizeof(fsname)) {
807 CERROR("Invalid logname received: %s\n", logname);
811 strncpy(fsname, logname, rc);
813 rc = mgs_fsc_attach(req->rq_svc_thread->t_env, req->rq_export, fsname);
814 if (rc < 0 && rc != -EEXIST)
815 CERROR("add fs client %s returns %d\n", fsname, rc);
820 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
821 int mgs_handle(struct ptlrpc_request *req)
823 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
827 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
828 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
829 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
832 LASSERT(current->journal_info == NULL);
833 opc = lustre_msg_get_opc(req->rq_reqmsg);
835 if (opc == SEC_CTX_INIT ||
836 opc == SEC_CTX_INIT_CONT ||
840 if (opc != MGS_CONNECT) {
841 if (!class_connected_export(req->rq_export)) {
842 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
843 req->rq_status = -ENOTCONN;
844 GOTO(out, rc = -ENOTCONN);
850 DEBUG_REQ(D_MGS, req, "connect");
851 /* MGS and MDS have same request format for connect */
852 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
853 rc = target_handle_connect(req);
855 rc = mgs_connect_check_sptlrpc(req);
857 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
858 /* Make clients trying to reconnect after a MGS restart
859 happy; also requires obd_replayable */
860 lustre_msg_add_op_flags(req->rq_repmsg,
861 MSG_CONNECT_RECONNECT);
864 DEBUG_REQ(D_MGS, req, "disconnect");
865 /* MGS and MDS have same request format for disconnect */
866 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
867 rc = target_handle_disconnect(req);
868 req->rq_status = rc; /* superfluous? */
871 DEBUG_REQ(D_MGS, req, "exception");
872 rc = mgs_handle_exception(req);
875 DEBUG_REQ(D_MGS, req, "target add");
876 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
877 rc = mgs_handle_target_reg(req);
880 DEBUG_REQ(D_MGS, req, "target del");
881 rc = mgs_handle_target_del(req);
884 DEBUG_REQ(D_MGS, req, "set_info");
885 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
886 rc = mgs_set_info_rpc(req);
888 case MGS_CONFIG_READ:
889 DEBUG_REQ(D_MGS, req, "read config");
890 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
891 rc = mgs_config_read(req);
894 DEBUG_REQ(D_MGS, req, "enqueue");
895 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
896 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
897 ldlm_server_blocking_ast, NULL);
899 case LDLM_BL_CALLBACK:
900 case LDLM_CP_CALLBACK:
901 DEBUG_REQ(D_MGS, req, "callback");
902 CERROR("callbacks should not happen on MGS\n");
907 DEBUG_REQ(D_INFO, req, "ping");
908 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
909 rc = target_handle_ping(req);
912 DEBUG_REQ(D_MGS, req, "log cancel");
913 rc = -ENOTSUPP; /* la la la */
916 case LLOG_ORIGIN_HANDLE_CREATE:
917 DEBUG_REQ(D_MGS, req, "llog_open");
918 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
919 rc = llog_origin_handle_open(req);
921 (void)mgs_handle_fslog_hack(req);
923 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
924 DEBUG_REQ(D_MGS, req, "llog next block");
925 req_capsule_set(&req->rq_pill,
926 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
927 rc = llog_origin_handle_next_block(req);
929 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
930 DEBUG_REQ(D_MGS, req, "llog prev block");
931 req_capsule_set(&req->rq_pill,
932 &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
933 rc = llog_origin_handle_prev_block(req);
935 case LLOG_ORIGIN_HANDLE_READ_HEADER:
936 DEBUG_REQ(D_MGS, req, "llog read header");
937 req_capsule_set(&req->rq_pill,
938 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
939 rc = llog_origin_handle_read_header(req);
941 case LLOG_ORIGIN_HANDLE_CLOSE:
942 DEBUG_REQ(D_MGS, req, "llog close");
943 rc = llog_origin_handle_close(req);
949 LASSERT(current->journal_info == NULL);
951 DEBUG_REQ(D_MGS, req, "MGS fail to handle opc = %d: rc = %d\n",
954 rc = ptlrpc_error(req);
958 target_send_reply(req, rc, fail);
962 static inline int mgs_init_export(struct obd_export *exp)
964 struct mgs_export_data *data = &exp->u.eu_mgs_data;
966 /* init mgs_export_data for fsc */
967 cfs_spin_lock_init(&data->med_lock);
968 CFS_INIT_LIST_HEAD(&data->med_clients);
970 cfs_spin_lock(&exp->exp_lock);
971 exp->exp_connecting = 1;
972 cfs_spin_unlock(&exp->exp_lock);
974 /* self-export doesn't need client data and ldlm initialization */
975 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
976 &exp->exp_client_uuid)))
978 return ldlm_init_export(exp);
981 static inline int mgs_destroy_export(struct obd_export *exp)
985 target_destroy_export(exp);
986 mgs_client_free(exp);
988 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
989 &exp->exp_client_uuid)))
992 ldlm_destroy_export(exp);
997 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
1002 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
1010 strcpy(poolname, ptr);
1015 static int mgs_iocontrol_pool(const struct lu_env *env,
1016 struct mgs_device *mgs,
1017 struct obd_ioctl_data *data)
1019 struct mgs_thread_info *mgi = mgs_env_info(env);
1021 struct lustre_cfg *lcfg = NULL;
1022 struct llog_rec_hdr rec;
1023 char *poolname = NULL;
1026 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1027 if (poolname == NULL)
1029 rec.lrh_len = llog_data_len(data->ioc_plen1);
1031 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1032 rec.lrh_type = OBD_CFG_REC;
1034 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1035 GOTO(out_pool, rc = -EINVAL);
1038 if (data->ioc_plen1 > CFS_PAGE_SIZE)
1039 GOTO(out_pool, rc = -E2BIG);
1041 OBD_ALLOC(lcfg, data->ioc_plen1);
1043 GOTO(out_pool, rc = -ENOMEM);
1045 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1046 GOTO(out_lcfg, rc = -EFAULT);
1048 if (lcfg->lcfg_bufcount < 2)
1049 GOTO(out_lcfg, rc = -EFAULT);
1051 /* first arg is always <fsname>.<poolname> */
1052 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
1057 switch (lcfg->lcfg_command) {
1059 if (lcfg->lcfg_bufcount != 2)
1060 GOTO(out_lcfg, rc = -EINVAL);
1061 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
1065 if (lcfg->lcfg_bufcount != 3)
1066 GOTO(out_lcfg, rc = -EINVAL);
1067 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
1068 poolname, lustre_cfg_string(lcfg, 2));
1071 if (lcfg->lcfg_bufcount != 3)
1072 GOTO(out_lcfg, rc = -EINVAL);
1073 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
1074 poolname, lustre_cfg_string(lcfg, 2));
1077 if (lcfg->lcfg_bufcount != 2)
1078 GOTO(out_lcfg, rc = -EINVAL);
1079 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
1087 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1088 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
1093 OBD_FREE(lcfg, data->ioc_plen1);
1095 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1099 /* from mdt_iocontrol */
1100 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1101 void *karg, void *uarg)
1103 struct mgs_device *mgs = exp2mgs_dev(exp);
1104 struct obd_ioctl_data *data = karg;
1105 struct lvfs_run_ctxt saved;
1110 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1112 rc = lu_env_init(&env, LCT_MG_THREAD);
1118 case OBD_IOC_PARAM: {
1119 struct mgs_thread_info *mgi = mgs_env_info(&env);
1120 struct lustre_cfg *lcfg;
1121 struct llog_rec_hdr rec;
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);
1129 GOTO(out, rc = -EINVAL);
1132 OBD_ALLOC(lcfg, data->ioc_plen1);
1134 GOTO(out, rc = -ENOMEM);
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(&env, mgs, lcfg, mgi->mgi_fsname);
1143 CERROR("%s: setparam err: rc = %d\n",
1144 exp->exp_obd->obd_name, rc);
1146 OBD_FREE(lcfg, data->ioc_plen1);
1151 rc = mgs_iocontrol_pool(&env, mgs, data);
1154 case OBD_IOC_DUMP_LOG: {
1155 struct llog_ctxt *ctxt;
1156 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1157 push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
1158 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1159 pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL);
1160 llog_ctxt_put(ctxt);
1165 case OBD_IOC_LLOG_CHECK:
1166 case OBD_IOC_LLOG_INFO:
1167 case OBD_IOC_LLOG_PRINT: {
1168 struct llog_ctxt *ctxt;
1169 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1171 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1172 rc = llog_ioctl(ctxt, cmd, data);
1173 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1174 llog_ctxt_put(ctxt);
1180 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1189 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1190 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1192 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1194 /* use obd ops to offer management infrastructure */
1195 static struct obd_ops mgs_obd_ops = {
1196 .o_owner = THIS_MODULE,
1197 .o_connect = mgs_connect,
1198 .o_reconnect = mgs_reconnect,
1199 .o_disconnect = mgs_disconnect,
1200 .o_setup = mgs_setup,
1201 .o_precleanup = mgs_precleanup,
1202 .o_cleanup = mgs_cleanup,
1203 .o_init_export = mgs_init_export,
1204 .o_destroy_export = mgs_destroy_export,
1205 .o_iocontrol = mgs_iocontrol,
1206 .o_llog_init = mgs_llog_init,
1207 .o_llog_finish = mgs_llog_finish
1210 static int __init mgs_init(void)
1212 struct lprocfs_static_vars lvars;
1215 lprocfs_mgs_init_vars(&lvars);
1216 class_register_type(&mgs_obd_ops, NULL,
1217 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1219 LU_CONTEXT_KEY_INIT(&mgs_thread_key);
1220 rc = lu_context_key_register(&mgs_thread_key);
1222 class_unregister_type(LUSTRE_MGS_NAME);
1227 static void /*__exit*/ mgs_exit(void)
1229 class_unregister_type(LUSTRE_MGS_NAME);
1230 lu_context_key_degister(&mgs_thread_key);
1233 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1234 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1235 MODULE_LICENSE("GPL");
1237 module_init(mgs_init);
1238 module_exit(mgs_exit);