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_sema_init(&mgs->mgs_sem, 1);
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))) {
361 struct fs_db *fsdb = (struct fs_db *)lock->l_ast_data;
362 struct lustre_handle lockh;
364 mgs_ir_notify_complete(fsdb);
366 ldlm_lock2handle(lock, &lockh);
367 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
370 RETURN(ldlm_completion_ast(lock, flags, cbdata));
373 void mgs_revoke_lock(struct obd_device *obd, struct fs_db *fsdb, int type)
375 ldlm_completion_callback cp = NULL;
376 struct lustre_handle lockh = { 0 };
377 struct ldlm_res_id res_id;
378 int flags = LDLM_FL_ATOMIC_CB;
382 LASSERT(fsdb->fsdb_name[0] != '\0');
383 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
387 case CONFIG_T_CONFIG:
388 cp = mgs_completion_ast_config;
389 if (cfs_test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
392 case CONFIG_T_RECOVER:
393 cp = mgs_completion_ast_ir;
400 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
401 LDLM_PLAIN, NULL, LCK_EX, &flags,
402 ldlm_blocking_ast, cp, NULL,
403 fsdb, 0, NULL, &lockh);
404 if (rc != ELDLM_OK) {
405 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
406 le64_to_cpu(res_id.name[0]),
407 le64_to_cpu(res_id.name[1]), rc);
409 if (type == CONFIG_T_CONFIG)
410 cfs_clear_bit(FSDB_REVOKING_LOCK,
413 /* lock has been cancelled in completion_ast. */
422 static int mgs_check_target(struct obd_device *obd, struct mgs_target_info *mti)
427 rc = mgs_check_index(obd, mti);
429 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
430 "this MGS does not know about it, preventing "
431 "registration.\n", mti->mti_svname);
433 } else if (rc == -1) {
434 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
435 "disappeared! Regenerating all logs.\n",
437 mti->mti_flags |= LDD_F_WRITECONF;
440 /* Index is correctly marked as used */
442 /* If the logs don't contain the mti_nids then add
443 them as failover nids */
444 rc = mgs_check_failnid(obd, mti);
450 /* Ensure this is not a failover node that is connecting first*/
451 static int mgs_check_failover_reg(struct mgs_target_info *mti)
457 ptr = mti->mti_params;
458 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
459 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
460 for (i = 0; i < mti->mti_nid_count; i++) {
461 if (nid == mti->mti_nids[i]) {
462 LCONSOLE_WARN("Denying initial registra"
463 "tion attempt from nid %s"
464 ", specified as failover"
465 "\n",libcfs_nid2str(nid));
466 return -EADDRNOTAVAIL;
474 /* Called whenever a target starts up. Flags indicate first connect, etc. */
475 static int mgs_handle_target_reg(struct ptlrpc_request *req)
477 struct obd_device *obd = req->rq_export->exp_obd;
478 struct mgs_target_info *mti, *rep_mti;
484 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
486 mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
488 opc = mti->mti_flags & LDD_F_OPC_MASK;
489 if (opc == LDD_F_OPC_READY) {
490 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
491 mti->mti_fsname, mti->mti_stripe_index);
492 rc = mgs_ir_update(obd, mti);
494 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
495 CERROR("Update IR return with %d(ignore and IR "
498 GOTO(out_nolock, rc);
501 /* Do not support unregistering right now. */
502 if (opc != LDD_F_OPC_REG)
503 GOTO(out_nolock, rc = -EINVAL);
505 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
506 mti->mti_fsname, mti->mti_stripe_index);
508 if (mti->mti_flags & LDD_F_NEED_INDEX)
509 mti->mti_flags |= LDD_F_WRITECONF;
511 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
513 /* We're just here as a startup ping. */
514 CDEBUG(D_MGS, "Server %s is running on %s\n",
515 mti->mti_svname, obd_export_nid2str(req->rq_export));
516 rc = mgs_check_target(obd, mti);
517 /* above will set appropriate mti flags */
519 /* Nothing wrong, or fatal error */
520 GOTO(out_nolock, rc);
522 if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
523 && (rc = mgs_check_failover_reg(mti)))
524 GOTO(out_nolock, rc);
527 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
529 if (mti->mti_flags & LDD_F_WRITECONF) {
530 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
531 mti->mti_stripe_index == 0) {
532 rc = mgs_erase_logs(obd, mti->mti_fsname);
533 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
534 "request. All servers must be restarted "
535 "in order to regenerate the logs."
536 "\n", obd->obd_name, mti->mti_fsname);
537 } else if (mti->mti_flags &
538 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
539 rc = mgs_erase_log(obd, mti->mti_svname);
540 LCONSOLE_WARN("%s: Regenerating %s log by user "
542 obd->obd_name, mti->mti_svname);
544 mti->mti_flags |= LDD_F_UPDATE;
545 /* Erased logs means start from scratch. */
546 mti->mti_flags &= ~LDD_F_UPGRADE14;
549 rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
551 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
552 GOTO(out_nolock, rc);
556 * Log writing contention is handled by the fsdb_sem.
558 * It should be alright if someone was reading while we were
559 * updating the logs - if we revoke at the end they will just update
560 * from where they left off.
564 if (mti->mti_flags & LDD_F_UPGRADE14) {
565 rc = mgs_upgrade_sv_14(obd, mti, fsdb);
567 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
571 /* We're good to go */
572 mti->mti_flags |= LDD_F_UPDATE;
576 if (mti->mti_flags & LDD_F_UPDATE) {
577 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
578 mti->mti_stripe_index);
580 /* create or update the target log
581 and update the client/mdt logs */
582 rc = mgs_write_log_target(obd, mti, fsdb);
584 CERROR("Failed to write %s log (%d)\n",
585 mti->mti_svname, rc);
589 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
590 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
592 mti->mti_flags |= LDD_F_REWRITE_LDD;
596 mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
599 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
600 mti->mti_stripe_index, rc);
603 /* we need an error flag to tell the target what's going on,
604 * instead of just doing it by error code only. */
605 mti->mti_flags |= LDD_F_ERROR;
607 rc = req_capsule_server_pack(&req->rq_pill);
611 /* send back the whole mti in the reply */
612 rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
615 /* Flush logs to disk */
616 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
620 static int mgs_set_info_rpc(struct ptlrpc_request *req)
622 struct obd_device *obd = req->rq_export->exp_obd;
623 struct mgs_send_param *msp, *rep_msp;
625 struct lustre_cfg_bufs bufs;
626 struct lustre_cfg *lcfg;
627 char fsname[MTI_NAME_MAXLEN];
630 msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
633 /* Construct lustre_cfg structure to pass to function mgs_setparam */
634 lustre_cfg_bufs_reset(&bufs, NULL);
635 lustre_cfg_bufs_set_string(&bufs, 1, msp->mgs_param);
636 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
637 rc = mgs_setparam(obd, lcfg, fsname);
639 CERROR("Error %d in setting the parameter %s for fs %s\n",
640 rc, msp->mgs_param, fsname);
644 lustre_cfg_free(lcfg);
646 rc = req_capsule_server_pack(&req->rq_pill);
648 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
654 static int mgs_config_read(struct ptlrpc_request *req)
656 struct mgs_config_body *body;
660 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
664 switch (body->mcb_type) {
665 case CONFIG_T_RECOVER:
666 rc = mgs_get_ir_logs(req);
669 case CONFIG_T_CONFIG:
682 * similar as in ost_connect_check_sptlrpc()
684 static int mgs_connect_check_sptlrpc(struct ptlrpc_request *req)
686 struct obd_export *exp = req->rq_export;
687 struct obd_device *obd = exp->exp_obd;
689 struct sptlrpc_flavor flvr;
692 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
693 rc = mgs_find_or_make_fsdb(obd, MGSSELF_NAME, &fsdb);
697 cfs_down(&fsdb->fsdb_sem);
698 if (sptlrpc_rule_set_choose(&fsdb->fsdb_srpc_gen,
699 LUSTRE_SP_MGC, LUSTRE_SP_MGS,
702 /* by defualt allow any flavors */
703 flvr.sf_rpc = SPTLRPC_FLVR_ANY;
705 cfs_up(&fsdb->fsdb_sem);
707 cfs_spin_lock(&exp->exp_lock);
709 exp->exp_sp_peer = req->rq_sp_from;
710 exp->exp_flvr = flvr;
712 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
713 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
714 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
715 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
716 libcfs_nid2str(req->rq_peer.nid));
720 cfs_spin_unlock(&exp->exp_lock);
722 if (exp->exp_sp_peer != req->rq_sp_from) {
723 CERROR("RPC source %s doesn't match %s\n",
724 sptlrpc_part2name(req->rq_sp_from),
725 sptlrpc_part2name(exp->exp_sp_peer));
728 rc = sptlrpc_target_export_check(exp, req);
735 /* Called whenever a target cleans up. */
736 /* XXX - Currently unused */
737 static int mgs_handle_target_del(struct ptlrpc_request *req)
740 mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_DEL);
744 /* XXX - Currently unused */
745 static int mgs_handle_exception(struct ptlrpc_request *req)
748 mgs_counter_incr(req->rq_export, LPROC_MGS_EXCEPTION);
753 * For old clients there is no direct way of knowing which filesystems
754 * a client is operating at the MGS side. But we need to pick up those
755 * clients so that the MGS can mark the corresponding filesystem as
756 * non-IR capable because old clients are not ready to be notified.
758 * This is why we have this _hack_ function. We detect the filesystem's
759 * name by hacking llog operation which is currently used by the clients
760 * to fetch configuration logs. At present this is fine because this is
761 * the ONLY llog operation between mgc and the MGS.
763 * If extra llog operation is going to be added, this function needs fixing.
765 * If releases prior than 2.0 are not supported, we can remove this function.
767 static int mgs_handle_fslog_hack(struct ptlrpc_request *req)
774 /* XXX: We suppose that llog at mgs is only used for
775 * fetching file system log */
776 logname = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
777 if (logname == NULL) {
778 CERROR("No logname, is llog on MGS used for something else?\n");
782 ptr = strchr(logname, '-');
783 rc = (int)(ptr - logname);
784 if (ptr == NULL || rc >= sizeof(fsname)) {
785 CERROR("Invalid logname received: %s\n", logname);
789 strncpy(fsname, logname, rc);
791 rc = mgs_fsc_attach(req->rq_export, fsname);
792 if (rc < 0 && rc != -EEXIST)
793 CERROR("add fs client %s returns %d\n", fsname, rc);
798 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
799 int mgs_handle(struct ptlrpc_request *req)
801 int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
805 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
806 CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
807 if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
810 LASSERT(current->journal_info == NULL);
811 opc = lustre_msg_get_opc(req->rq_reqmsg);
813 if (opc == SEC_CTX_INIT ||
814 opc == SEC_CTX_INIT_CONT ||
818 if (opc != MGS_CONNECT) {
819 if (!class_connected_export(req->rq_export)) {
820 DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
821 req->rq_status = -ENOTCONN;
822 GOTO(out, rc = -ENOTCONN);
828 DEBUG_REQ(D_MGS, req, "connect");
829 /* MGS and MDS have same request format for connect */
830 req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
831 rc = target_handle_connect(req);
833 rc = mgs_connect_check_sptlrpc(req);
835 if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
836 /* Make clients trying to reconnect after a MGS restart
837 happy; also requires obd_replayable */
838 lustre_msg_add_op_flags(req->rq_repmsg,
839 MSG_CONNECT_RECONNECT);
842 DEBUG_REQ(D_MGS, req, "disconnect");
843 /* MGS and MDS have same request format for disconnect */
844 req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
845 rc = target_handle_disconnect(req);
846 req->rq_status = rc; /* superfluous? */
849 DEBUG_REQ(D_MGS, req, "exception");
850 rc = mgs_handle_exception(req);
853 DEBUG_REQ(D_MGS, req, "target add");
854 req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
855 rc = mgs_handle_target_reg(req);
858 DEBUG_REQ(D_MGS, req, "target del");
859 rc = mgs_handle_target_del(req);
862 DEBUG_REQ(D_MGS, req, "set_info");
863 req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
864 rc = mgs_set_info_rpc(req);
866 case MGS_CONFIG_READ:
867 DEBUG_REQ(D_MGS, req, "read config");
868 req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
869 rc = mgs_config_read(req);
872 DEBUG_REQ(D_MGS, req, "enqueue");
873 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
874 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
875 ldlm_server_blocking_ast, NULL);
877 case LDLM_BL_CALLBACK:
878 case LDLM_CP_CALLBACK:
879 DEBUG_REQ(D_MGS, req, "callback");
880 CERROR("callbacks should not happen on MGS\n");
885 DEBUG_REQ(D_INFO, req, "ping");
886 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
887 rc = target_handle_ping(req);
890 DEBUG_REQ(D_MGS, req, "log cancel");
891 rc = -ENOTSUPP; /* la la la */
894 case LLOG_ORIGIN_HANDLE_CREATE:
895 DEBUG_REQ(D_MGS, req, "llog_init");
896 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
897 rc = llog_origin_handle_create(req);
899 (void)mgs_handle_fslog_hack(req);
901 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
902 DEBUG_REQ(D_MGS, req, "llog next block");
903 req_capsule_set(&req->rq_pill,
904 &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
905 rc = llog_origin_handle_next_block(req);
907 case LLOG_ORIGIN_HANDLE_READ_HEADER:
908 DEBUG_REQ(D_MGS, req, "llog read header");
909 req_capsule_set(&req->rq_pill,
910 &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
911 rc = llog_origin_handle_read_header(req);
913 case LLOG_ORIGIN_HANDLE_CLOSE:
914 DEBUG_REQ(D_MGS, req, "llog close");
915 rc = llog_origin_handle_close(req);
918 DEBUG_REQ(D_MGS, req, "llog catinfo");
919 req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
920 rc = llog_catinfo(req);
923 req->rq_status = -ENOTSUPP;
924 rc = ptlrpc_error(req);
928 LASSERT(current->journal_info == NULL);
931 CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
934 target_send_reply(req, rc, fail);
938 static inline int mgs_init_export(struct obd_export *exp)
940 struct mgs_export_data *data = &exp->u.eu_mgs_data;
942 /* init mgs_export_data for fsc */
943 cfs_spin_lock_init(&data->med_lock);
944 CFS_INIT_LIST_HEAD(&data->med_clients);
946 cfs_spin_lock(&exp->exp_lock);
947 exp->exp_connecting = 1;
948 cfs_spin_unlock(&exp->exp_lock);
950 /* self-export doesn't need client data and ldlm initialization */
951 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
952 &exp->exp_client_uuid)))
954 return ldlm_init_export(exp);
957 static inline int mgs_destroy_export(struct obd_export *exp)
961 target_destroy_export(exp);
962 mgs_client_free(exp);
964 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
965 &exp->exp_client_uuid)))
968 ldlm_destroy_export(exp);
973 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
978 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
986 strcpy(poolname, ptr);
991 static int mgs_iocontrol_pool(struct obd_device *obd,
992 struct obd_ioctl_data *data)
995 struct lustre_cfg *lcfg = NULL;
996 struct llog_rec_hdr rec;
998 char *poolname = NULL;
1001 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
1005 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
1006 if (poolname == NULL) {
1010 rec.lrh_len = llog_data_len(data->ioc_plen1);
1012 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1013 rec.lrh_type = OBD_CFG_REC;
1015 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1020 if (data->ioc_plen1 > CFS_PAGE_SIZE) {
1025 OBD_ALLOC(lcfg, data->ioc_plen1);
1027 GOTO(out_pool, rc = -ENOMEM);
1029 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1030 GOTO(out_pool, rc = -EFAULT);
1032 if (lcfg->lcfg_bufcount < 2) {
1033 GOTO(out_pool, rc = -EFAULT);
1036 /* first arg is always <fsname>.<poolname> */
1037 mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), fsname,
1040 switch (lcfg->lcfg_command) {
1041 case LCFG_POOL_NEW: {
1042 if (lcfg->lcfg_bufcount != 2)
1044 rc = mgs_pool_cmd(obd, LCFG_POOL_NEW, fsname,
1048 case LCFG_POOL_ADD: {
1049 if (lcfg->lcfg_bufcount != 3)
1051 rc = mgs_pool_cmd(obd, LCFG_POOL_ADD, fsname, poolname,
1052 lustre_cfg_string(lcfg, 2));
1055 case LCFG_POOL_REM: {
1056 if (lcfg->lcfg_bufcount != 3)
1058 rc = mgs_pool_cmd(obd, LCFG_POOL_REM, fsname, poolname,
1059 lustre_cfg_string(lcfg, 2));
1062 case LCFG_POOL_DEL: {
1063 if (lcfg->lcfg_bufcount != 2)
1065 rc = mgs_pool_cmd(obd, LCFG_POOL_DEL, fsname,
1076 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
1077 rc, lcfg->lcfg_command, fsname, poolname);
1083 OBD_FREE(lcfg, data->ioc_plen1);
1086 OBD_FREE(fsname, MTI_NAME_MAXLEN);
1088 if (poolname != NULL)
1089 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
1094 /* from mdt_iocontrol */
1095 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1096 void *karg, void *uarg)
1098 struct obd_device *obd = exp->exp_obd;
1099 struct obd_ioctl_data *data = karg;
1100 struct lvfs_run_ctxt saved;
1104 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1108 case OBD_IOC_PARAM: {
1109 struct lustre_cfg *lcfg;
1110 struct llog_rec_hdr rec;
1111 char fsname[MTI_NAME_MAXLEN];
1113 rec.lrh_len = llog_data_len(data->ioc_plen1);
1115 if (data->ioc_type == LUSTRE_CFG_TYPE) {
1116 rec.lrh_type = OBD_CFG_REC;
1118 CERROR("unknown cfg record type:%d \n", data->ioc_type);
1122 OBD_ALLOC(lcfg, data->ioc_plen1);
1125 if (cfs_copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1126 GOTO(out_free, rc = -EFAULT);
1128 if (lcfg->lcfg_bufcount < 1)
1129 GOTO(out_free, rc = -EINVAL);
1131 rc = mgs_setparam(obd, lcfg, fsname);
1133 CERROR("setparam err %d\n", rc);
1137 OBD_FREE(lcfg, data->ioc_plen1);
1141 case OBD_IOC_POOL: {
1142 RETURN(mgs_iocontrol_pool(obd, data));
1145 case OBD_IOC_DUMP_LOG: {
1146 struct llog_ctxt *ctxt;
1147 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1148 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1149 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
1150 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1151 llog_ctxt_put(ctxt);
1156 case OBD_IOC_LLOG_CHECK:
1157 case OBD_IOC_LLOG_INFO:
1158 case OBD_IOC_LLOG_PRINT: {
1159 struct llog_ctxt *ctxt;
1160 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
1162 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1163 rc = llog_ioctl(ctxt, cmd, data);
1164 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
1165 llog_ctxt_put(ctxt);
1171 CDEBUG(D_INFO, "unknown command %x\n", cmd);
1177 /* use obd ops to offer management infrastructure */
1178 static struct obd_ops mgs_obd_ops = {
1179 .o_owner = THIS_MODULE,
1180 .o_connect = mgs_connect,
1181 .o_reconnect = mgs_reconnect,
1182 .o_disconnect = mgs_disconnect,
1183 .o_setup = mgs_setup,
1184 .o_precleanup = mgs_precleanup,
1185 .o_cleanup = mgs_cleanup,
1186 .o_init_export = mgs_init_export,
1187 .o_destroy_export = mgs_destroy_export,
1188 .o_iocontrol = mgs_iocontrol,
1189 .o_llog_init = mgs_llog_init,
1190 .o_llog_finish = mgs_llog_finish
1193 static int __init mgs_init(void)
1195 struct lprocfs_static_vars lvars;
1197 lprocfs_mgs_init_vars(&lvars);
1198 class_register_type(&mgs_obd_ops, NULL,
1199 lvars.module_vars, LUSTRE_MGS_NAME, NULL);
1204 static void /*__exit*/ mgs_exit(void)
1206 class_unregister_type(LUSTRE_MGS_NAME);
1209 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1210 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1211 MODULE_LICENSE("GPL");
1213 module_init(mgs_init);
1214 module_exit(mgs_exit);