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) 2010, 2013, Intel Corporation.
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>
39 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40 * Author: Mikhail Pershin <tappro@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_MGS
44 #define D_MGS D_CONFIG
46 #include <obd_class.h>
47 #include <lprocfs_status.h>
48 #include <lustre_param.h>
50 #include "mgs_internal.h"
53 * Regular MGS handlers
55 static int mgs_connect(struct tgt_session_info *tsi)
57 struct ptlrpc_request *req = tgt_ses_req(tsi);
62 rc = tgt_connect(tsi);
66 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
67 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
72 static int mgs_disconnect(struct tgt_session_info *tsi)
78 LASSERT(tsi->tsi_exp);
80 rc = tgt_disconnect(tsi);
82 RETURN(err_serious(rc));
86 static int mgs_exception(struct tgt_session_info *tsi)
90 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
95 static int mgs_set_info(struct tgt_session_info *tsi)
97 struct mgs_thread_info *mgi;
98 struct mgs_send_param *msp, *rep_msp;
99 struct lustre_cfg *lcfg;
104 mgi = mgs_env_info(tsi->tsi_env);
106 RETURN(err_serious(PTR_ERR(mgi)));
108 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
110 RETURN(err_serious(-EFAULT));
112 /* Construct lustre_cfg structure to pass to function mgs_setparam */
113 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
114 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
115 lcfg = lustre_cfg_new(LCFG_PARAM, &mgi->mgi_bufs);
117 rc = mgs_setparam(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg,
120 LCONSOLE_WARN("%s: Unable to set parameter %s for %s: %d\n",
121 tgt_name(tsi->tsi_tgt), msp->mgs_param,
122 mgi->mgi_fsname, rc);
126 /* send back the whole msp in the reply */
127 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
131 lustre_cfg_free(lcfg);
135 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
140 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
141 LDLM_FL_BLOCK_CONV))) {
144 /* l_ast_data is used as a marker to avoid cancel ldlm lock
145 * twice. See LU-2317. */
146 lock_res_and_lock(lock);
147 fsdb = (struct fs_db *)lock->l_ast_data;
148 lock->l_ast_data = NULL;
149 unlock_res_and_lock(lock);
152 struct lustre_handle lockh;
154 /* clear the bit before lock put */
155 clear_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags);
157 ldlm_lock2handle(lock, &lockh);
158 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
162 RETURN(ldlm_completion_ast(lock, flags, cbdata));
165 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
170 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
171 LDLM_FL_BLOCK_CONV))) {
174 /* l_ast_data is used as a marker to avoid cancel ldlm lock
175 * twice. See LU-1259. */
176 lock_res_and_lock(lock);
177 fsdb = (struct fs_db *)lock->l_ast_data;
178 lock->l_ast_data = NULL;
179 unlock_res_and_lock(lock);
182 struct lustre_handle lockh;
184 mgs_ir_notify_complete(fsdb);
186 ldlm_lock2handle(lock, &lockh);
187 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
191 RETURN(ldlm_completion_ast(lock, flags, cbdata));
194 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
196 ldlm_completion_callback cp = NULL;
197 struct lustre_handle lockh = { 0 };
198 struct ldlm_res_id res_id;
199 __u64 flags = LDLM_FL_ATOMIC_CB;
203 LASSERT(fsdb->fsdb_name[0] != '\0');
204 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
208 case CONFIG_T_CONFIG:
209 cp = mgs_completion_ast_config;
210 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
213 case CONFIG_T_RECOVER:
214 cp = mgs_completion_ast_ir;
221 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
222 &res_id, LDLM_PLAIN, NULL, LCK_EX,
223 &flags, ldlm_blocking_ast, cp,
224 NULL, fsdb, 0, LVB_T_NONE, NULL,
226 if (rc != ELDLM_OK) {
227 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
228 le64_to_cpu(res_id.name[0]),
229 le64_to_cpu(res_id.name[1]), rc);
231 if (type == CONFIG_T_CONFIG)
232 clear_bit(FSDB_REVOKING_LOCK,
235 /* lock has been cancelled in completion_ast. */
244 static int mgs_check_target(const struct lu_env *env,
245 struct mgs_device *mgs,
246 struct mgs_target_info *mti)
251 rc = mgs_check_index(env, mgs, mti);
253 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
254 "this MGS does not know about it, preventing "
255 "registration.\n", mti->mti_svname);
257 } else if (rc == -1) {
258 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
259 "disappeared! Regenerating all logs.\n",
261 mti->mti_flags |= LDD_F_WRITECONF;
264 /* Index is correctly marked as used */
266 /* If the logs don't contain the mti_nids then add
267 them as failover nids */
268 rc = mgs_check_failnid(env, mgs, mti);
274 /* Ensure this is not a failover node that is connecting first*/
275 static int mgs_check_failover_reg(struct mgs_target_info *mti)
281 ptr = mti->mti_params;
282 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
283 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
284 for (i = 0; i < mti->mti_nid_count; i++) {
285 if (nid == mti->mti_nids[i]) {
286 LCONSOLE_WARN("Denying initial registra"
287 "tion attempt from nid %s"
288 ", specified as failover"
289 "\n",libcfs_nid2str(nid));
290 return -EADDRNOTAVAIL;
298 /* Called whenever a target starts up. Flags indicate first connect, etc. */
299 static int mgs_target_reg(struct tgt_session_info *tsi)
301 struct obd_device *obd = tsi->tsi_exp->exp_obd;
302 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
303 struct mgs_target_info *mti, *rep_mti;
310 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
312 return err_serious(rc);
314 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
316 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
318 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
319 RETURN(err_serious(-EFAULT));
322 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
324 opc = mti->mti_flags & LDD_F_OPC_MASK;
328 if (opc == LDD_F_OPC_READY) {
329 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
330 mti->mti_fsname, mti->mti_stripe_index);
331 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
333 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
334 CERROR("Update IR return with %d(ignore and IR "
337 GOTO(out_nolock, rc);
340 /* Do not support unregistering right now. */
341 if (opc != LDD_F_OPC_REG)
342 GOTO(out_nolock, rc = -EINVAL);
344 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
345 mti->mti_fsname, mti->mti_stripe_index);
347 if (mti->mti_flags & LDD_F_NEED_INDEX)
348 mti->mti_flags |= LDD_F_WRITECONF;
350 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
352 /* We're just here as a startup ping. */
353 CDEBUG(D_MGS, "Server %s is running on %s\n",
354 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
355 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
356 /* above will set appropriate mti flags */
358 /* Nothing wrong, or fatal error */
359 GOTO(out_nolock, rc);
360 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
361 rc = mgs_check_failover_reg(mti);
363 GOTO(out_nolock, rc);
366 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
368 if (mti->mti_flags & LDD_F_WRITECONF) {
369 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
370 mti->mti_stripe_index == 0) {
371 rc = mgs_erase_logs(tsi->tsi_env, mgs,
373 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
374 "request. All servers must be restarted "
375 "in order to regenerate the logs."
376 "\n", obd->obd_name, mti->mti_fsname);
377 } else if (mti->mti_flags &
378 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
379 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
380 LCONSOLE_WARN("%s: Regenerating %s log by user "
382 obd->obd_name, mti->mti_svname);
384 mti->mti_flags |= LDD_F_UPDATE;
385 /* Erased logs means start from scratch. */
386 mti->mti_flags &= ~LDD_F_UPGRADE14;
388 GOTO(out_nolock, rc);
391 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &fsdb);
393 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
394 GOTO(out_nolock, rc);
398 * Log writing contention is handled by the fsdb_mutex.
400 * It should be alright if someone was reading while we were
401 * updating the logs - if we revoke at the end they will just update
402 * from where they left off.
405 if (mti->mti_flags & LDD_F_UPGRADE14) {
406 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
410 if (mti->mti_flags & LDD_F_UPDATE) {
411 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
412 mti->mti_stripe_index);
414 /* create or update the target log
415 and update the client/mdt logs */
416 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, fsdb);
418 CERROR("Failed to write %s log (%d)\n",
419 mti->mti_svname, rc);
423 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
424 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
426 mti->mti_flags |= LDD_F_REWRITE_LDD;
430 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
433 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
434 mti->mti_stripe_index, rc);
435 /* An error flag is set in the mti reply rather than an error code */
437 mti->mti_flags |= LDD_F_ERROR;
439 /* send back the whole mti in the reply */
440 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
443 /* Flush logs to disk */
444 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
448 /* Called whenever a target cleans up. */
449 static int mgs_target_del(struct tgt_session_info *tsi)
453 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
458 static int mgs_config_read(struct tgt_session_info *tsi)
460 struct ptlrpc_request *req = tgt_ses_req(tsi);
461 struct mgs_config_body *body;
466 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
468 DEBUG_REQ(D_HA, req, "no mgs_config_body");
469 RETURN(err_serious(-EFAULT));
472 switch (body->mcb_type) {
473 case CONFIG_T_RECOVER:
474 rc = mgs_get_ir_logs(req);
476 case CONFIG_T_CONFIG:
487 static int mgs_llog_open(struct tgt_session_info *tsi)
489 struct mgs_thread_info *mgi;
490 struct ptlrpc_request *req = tgt_ses_req(tsi);
496 rc = tgt_llog_open(tsi);
501 * For old clients there is no direct way of knowing which file system
502 * a client is operating at the MGS side. But we need to pick up those
503 * clients so that the MGS can mark the corresponding file system as
504 * non-IR capable because old clients are not ready to be notified.
506 * Therefore we attempt to detect the file systems name by hacking the
507 * llog operation which is currently used by the clients to fetch
508 * configuration logs. At present this is fine because this is the
509 * ONLY llog operation between mgc and the MGS.
511 * If extra llog operation are going to be added, this function needs
514 * When releases prior than 2.0 are not supported, the following code
517 mgi = mgs_env_info(tsi->tsi_env);
519 RETURN(PTR_ERR(mgi));
521 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
523 char *ptr = strchr(logname, '-');
524 int len = (int)(ptr - logname);
526 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
527 LCONSOLE_WARN("%s: non-config logname received: %s\n",
528 tgt_name(tsi->tsi_tgt), logname);
529 /* not error, this can be llog test name */
531 strncpy(mgi->mgi_fsname, logname, len);
532 mgi->mgi_fsname[len] = 0;
534 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
536 if (rc && rc != -EEXIST) {
537 LCONSOLE_WARN("%s: Unable to add client %s "
538 "to file system %s: %d\n",
539 tgt_name(tsi->tsi_tgt),
540 libcfs_nid2str(req->rq_peer.nid),
541 mgi->mgi_fsname, rc);
547 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
553 static inline int mgs_init_export(struct obd_export *exp)
555 struct mgs_export_data *data = &exp->u.eu_mgs_data;
557 /* init mgs_export_data for fsc */
558 spin_lock_init(&data->med_lock);
559 CFS_INIT_LIST_HEAD(&data->med_clients);
561 spin_lock(&exp->exp_lock);
562 exp->exp_connecting = 1;
563 spin_unlock(&exp->exp_lock);
565 /* self-export doesn't need client data and ldlm initialization */
566 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
567 &exp->exp_client_uuid)))
569 return ldlm_init_export(exp);
572 static inline int mgs_destroy_export(struct obd_export *exp)
576 target_destroy_export(exp);
577 mgs_client_free(exp);
579 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
580 &exp->exp_client_uuid)))
583 ldlm_destroy_export(exp);
588 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
593 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
601 strcpy(poolname, ptr);
606 static int mgs_iocontrol_pool(const struct lu_env *env,
607 struct mgs_device *mgs,
608 struct obd_ioctl_data *data)
610 struct mgs_thread_info *mgi = mgs_env_info(env);
612 struct lustre_cfg *lcfg = NULL;
613 char *poolname = NULL;
616 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
617 if (poolname == NULL)
620 if (data->ioc_type != LUSTRE_CFG_TYPE) {
621 CERROR("%s: unknown cfg record type: %d\n",
622 mgs->mgs_obd->obd_name, data->ioc_type);
623 GOTO(out_pool, rc = -EINVAL);
626 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
627 GOTO(out_pool, rc = -E2BIG);
629 OBD_ALLOC(lcfg, data->ioc_plen1);
631 GOTO(out_pool, rc = -ENOMEM);
633 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
634 GOTO(out_lcfg, rc = -EFAULT);
636 if (lcfg->lcfg_bufcount < 2)
637 GOTO(out_lcfg, rc = -EFAULT);
639 /* first arg is always <fsname>.<poolname> */
640 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
645 switch (lcfg->lcfg_command) {
647 if (lcfg->lcfg_bufcount != 2)
648 GOTO(out_lcfg, rc = -EINVAL);
649 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
653 if (lcfg->lcfg_bufcount != 3)
654 GOTO(out_lcfg, rc = -EINVAL);
655 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
656 poolname, lustre_cfg_string(lcfg, 2));
659 if (lcfg->lcfg_bufcount != 3)
660 GOTO(out_lcfg, rc = -EINVAL);
661 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
662 poolname, lustre_cfg_string(lcfg, 2));
665 if (lcfg->lcfg_bufcount != 2)
666 GOTO(out_lcfg, rc = -EINVAL);
667 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
675 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
676 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
681 OBD_FREE(lcfg, data->ioc_plen1);
683 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
687 /* from mdt_iocontrol */
688 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
689 void *karg, void *uarg)
691 struct mgs_device *mgs = exp2mgs_dev(exp);
692 struct obd_ioctl_data *data = karg;
697 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
699 rc = lu_env_init(&env, LCT_MG_THREAD);
705 case OBD_IOC_PARAM: {
706 struct mgs_thread_info *mgi = mgs_env_info(&env);
707 struct lustre_cfg *lcfg;
709 if (data->ioc_type != LUSTRE_CFG_TYPE) {
710 CERROR("%s: unknown cfg record type: %d\n",
711 mgs->mgs_obd->obd_name, data->ioc_type);
712 GOTO(out, rc = -EINVAL);
715 OBD_ALLOC(lcfg, data->ioc_plen1);
717 GOTO(out, rc = -ENOMEM);
718 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
719 GOTO(out_free, rc = -EFAULT);
721 if (lcfg->lcfg_bufcount < 1)
722 GOTO(out_free, rc = -EINVAL);
724 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
726 CERROR("%s: setparam err: rc = %d\n",
727 exp->exp_obd->obd_name, rc);
729 OBD_FREE(lcfg, data->ioc_plen1);
733 case OBD_IOC_REPLACE_NIDS: {
734 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
735 CERROR("No device name specified!\n");
740 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
741 CERROR("Device name is not NUL terminated!\n");
746 if (data->ioc_plen1 > MTI_NAME_MAXLEN) {
747 CERROR("Device name is too long\n");
752 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
753 CERROR("No NIDs were specified!\n");
758 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
759 CERROR("NID list is not NUL terminated!\n");
764 /* replace nids in llog */
765 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
768 CERROR("%s: error replacing nids: rc = %d\n",
769 exp->exp_obd->obd_name, rc);
775 rc = mgs_iocontrol_pool(&env, mgs, data);
778 case OBD_IOC_DUMP_LOG: {
779 struct llog_ctxt *ctxt;
781 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
782 rc = class_config_dump_llog(&env, ctxt, data->ioc_inlbuf1,
789 case OBD_IOC_LLOG_CANCEL:
790 case OBD_IOC_LLOG_REMOVE:
791 case OBD_IOC_LLOG_CHECK:
792 case OBD_IOC_LLOG_INFO:
793 case OBD_IOC_LLOG_PRINT: {
794 struct llog_ctxt *ctxt;
796 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
797 rc = llog_ioctl(&env, ctxt, cmd, data);
803 CERROR("%s: unknown command %#x\n",
804 mgs->mgs_obd->obd_name, cmd);
813 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
815 struct obd_connect_data *data = NULL;
816 struct obd_device *obd;
824 obd = class_name2obd(nextdev);
826 CERROR("can't locate next device: %s\n", nextdev);
827 GOTO(out, rc = -ENOTCONN);
830 data->ocd_version = LUSTRE_VERSION_CODE;
832 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
833 &obd->obd_uuid, data, NULL);
835 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
839 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
840 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
841 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
847 static struct tgt_handler mgs_mgs_handlers[] = {
848 TGT_RPC_HANDLER(MGS_FIRST_OPC,
849 0, MGS_CONNECT, mgs_connect,
850 &RQF_CONNECT, LUSTRE_OBD_VERSION),
851 TGT_RPC_HANDLER(MGS_FIRST_OPC,
852 0, MGS_DISCONNECT, mgs_disconnect,
853 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
854 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
855 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
856 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
857 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
858 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
861 static struct tgt_handler mgs_obd_handlers[] = {
862 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
865 static struct tgt_handler mgs_dlm_handlers[] = {
866 TGT_DLM_HDL(HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
869 static struct tgt_handler mgs_llog_handlers[] = {
870 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
871 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
872 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
873 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
874 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
877 static struct tgt_opc_slice mgs_common_slice[] = {
879 .tos_opc_start = MGS_FIRST_OPC,
880 .tos_opc_end = MGS_LAST_OPC,
881 .tos_hs = mgs_mgs_handlers
884 .tos_opc_start = OBD_FIRST_OPC,
885 .tos_opc_end = OBD_LAST_OPC,
886 .tos_hs = mgs_obd_handlers
889 .tos_opc_start = LDLM_FIRST_OPC,
890 .tos_opc_end = LDLM_LAST_OPC,
891 .tos_hs = mgs_dlm_handlers
894 .tos_opc_start = LLOG_FIRST_OPC,
895 .tos_opc_end = LLOG_LAST_OPC,
896 .tos_hs = mgs_llog_handlers
899 .tos_opc_start = SEC_FIRST_OPC,
900 .tos_opc_end = SEC_LAST_OPC,
901 .tos_hs = tgt_sec_ctx_handlers
908 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
909 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
911 struct ptlrpc_service_conf conf;
912 struct obd_device *obd;
913 struct lustre_mount_info *lmi;
914 struct llog_ctxt *ctxt;
919 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
923 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
925 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
929 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
932 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
934 obd->u.obt.obt_magic = OBT_MAGIC;
935 obd->u.obt.obt_instance = 0;
937 /* namespace for mgs llog */
938 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
939 LDLM_NAMESPACE_SERVER,
940 LDLM_NAMESPACE_MODEST,
942 if (obd->obd_namespace == NULL)
943 GOTO(err_ops, rc = -ENOMEM);
945 /* No recovery for MGCs */
946 obd->obd_replayable = 0;
948 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
949 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
950 OBD_FAIL_MGS_ALL_REPLY_NET);
954 rc = mgs_fs_setup(env, mgs);
956 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
961 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
966 /* XXX: we need this trick till N:1 stack is supported
967 * set "current" directory for named llogs */
968 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
970 ctxt->loc_dir = mgs->mgs_configs_dir;
973 /* Internal mgs setup */
974 mgs_init_fsdb_list(mgs);
975 mutex_init(&mgs->mgs_mutex);
976 mgs->mgs_start_time = cfs_time_current_sec();
977 spin_lock_init(&mgs->mgs_lock);
979 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
981 CERROR("%s: cannot initialize proc entry: rc = %d\n",
986 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
987 "mgs_ldlm_client", &obd->obd_ldlm_client);
989 conf = (typeof(conf)) {
990 .psc_name = LUSTRE_MGS_NAME,
991 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
993 .bc_nbufs = MGS_NBUFS,
994 .bc_buf_size = MGS_BUFSIZE,
995 .bc_req_max_size = MGS_MAXREQSIZE,
996 .bc_rep_max_size = MGS_MAXREPSIZE,
997 .bc_req_portal = MGS_REQUEST_PORTAL,
998 .bc_rep_portal = MGC_REPLY_PORTAL,
1001 .tc_thr_name = "ll_mgs",
1002 .tc_nthrs_init = MGS_NTHRS_INIT,
1003 .tc_nthrs_max = MGS_NTHRS_MAX,
1004 .tc_ctx_tags = LCT_MG_THREAD,
1007 .so_req_handler = tgt_request_handle,
1008 .so_req_printer = target_print_req,
1012 /* Start the service threads */
1013 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1014 if (IS_ERR(mgs->mgs_service)) {
1015 rc = PTR_ERR(mgs->mgs_service);
1016 CERROR("failed to start mgs service: %d\n", rc);
1017 mgs->mgs_service = NULL;
1018 GOTO(err_lproc, rc);
1021 ping_evictor_start();
1023 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1025 /* device stack is not yet fully setup to keep no objects behind */
1026 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1029 lproc_mgs_cleanup(mgs);
1031 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1033 ctxt->loc_dir = NULL;
1034 llog_cleanup(env, ctxt);
1037 tgt_fini(env, &mgs->mgs_lut);
1039 /* No extra cleanup needed for llog_init_commit_thread() */
1040 mgs_fs_cleanup(env, mgs);
1042 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1043 obd->obd_namespace = NULL;
1045 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1046 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1047 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1048 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1051 obd_disconnect(mgs->mgs_bottom_exp);
1054 server_put_mount(lustre_cfg_string(lcfg, 0), lmi->lmi_mnt);
1058 static struct lu_device *mgs_device_free(const struct lu_env *env,
1059 struct lu_device *lu)
1061 struct mgs_device *mgs = lu2mgs_dev(lu);
1064 dt_device_fini(&mgs->mgs_dt_dev);
1069 static int mgs_process_config(const struct lu_env *env,
1070 struct lu_device *dev,
1071 struct lustre_cfg *lcfg)
1077 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1078 const struct lu_object_conf *unused)
1080 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1081 struct lu_device *under;
1082 struct lu_object *below;
1086 /* do no set .do_ops as mgs calls to bottom osd directly */
1088 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1089 PFID(lu_object_fid(o)));
1091 under = &d->mgs_bottom->dd_lu_dev;
1092 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1094 lu_object_add(o, below);
1101 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1103 struct mgs_object *obj = lu2mgs_obj(o);
1104 struct lu_object_header *h = o->lo_header;
1106 dt_object_fini(&obj->mgo_obj);
1107 lu_object_header_fini(h);
1111 static int mgs_object_print(const struct lu_env *env, void *cookie,
1112 lu_printer_t p, const struct lu_object *l)
1114 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1116 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1119 struct lu_object_operations mgs_lu_obj_ops = {
1120 .loo_object_init = mgs_object_init,
1121 .loo_object_free = mgs_object_free,
1122 .loo_object_print = mgs_object_print,
1125 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1126 const struct lu_object_header *hdr,
1127 struct lu_device *d)
1129 struct lu_object_header *h;
1130 struct mgs_object *o;
1131 struct lu_object *l;
1133 LASSERT(hdr == NULL);
1137 l = &o->mgo_obj.do_lu;
1140 lu_object_header_init(h);
1141 dt_object_init(&o->mgo_obj, h, d);
1142 lu_object_add_top(h, l);
1144 l->lo_ops = &mgs_lu_obj_ops;
1152 const struct lu_device_operations mgs_lu_ops = {
1153 .ldo_object_alloc = mgs_object_alloc,
1154 .ldo_process_config = mgs_process_config,
1157 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1158 struct lu_device_type *type,
1159 struct lustre_cfg *lcfg)
1161 struct mgs_device *mgs;
1162 struct lu_device *ludev;
1166 ludev = ERR_PTR(-ENOMEM);
1170 ludev = mgs2lu_dev(mgs);
1171 dt_device_init(&mgs->mgs_dt_dev, type);
1172 rc = mgs_init0(env, mgs, type, lcfg);
1174 mgs_device_free(env, ludev);
1175 ludev = ERR_PTR(rc);
1181 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1182 struct lu_device *d)
1184 struct mgs_device *mgs = lu2mgs_dev(d);
1185 struct obd_device *obd = mgs->mgs_obd;
1186 struct llog_ctxt *ctxt;
1190 LASSERT(mgs->mgs_bottom);
1192 class_disconnect_exports(obd);
1194 ping_evictor_stop();
1196 ptlrpc_unregister_service(mgs->mgs_service);
1198 obd_exports_barrier(obd);
1199 obd_zombie_barrier();
1201 tgt_fini(env, &mgs->mgs_lut);
1202 mgs_cleanup_fsdb_list(mgs);
1203 lproc_mgs_cleanup(mgs);
1205 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1207 ctxt->loc_dir = NULL;
1208 llog_cleanup(env, ctxt);
1211 mgs_fs_cleanup(env, mgs);
1213 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1214 obd->obd_namespace = NULL;
1216 lu_site_purge(env, d->ld_site, ~0);
1217 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1218 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1219 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1222 LASSERT(mgs->mgs_bottom_exp);
1223 obd_disconnect(mgs->mgs_bottom_exp);
1225 server_put_mount(obd->obd_name, NULL);
1230 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1231 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1233 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1235 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1237 static struct lu_device_type_operations mgs_device_type_ops = {
1238 .ldto_init = mgs_type_init,
1239 .ldto_fini = mgs_type_fini,
1241 .ldto_start = mgs_type_start,
1242 .ldto_stop = mgs_type_stop,
1244 .ldto_device_alloc = mgs_device_alloc,
1245 .ldto_device_free = mgs_device_free,
1247 .ldto_device_fini = mgs_device_fini
1250 static struct lu_device_type mgs_device_type = {
1251 .ldt_tags = LU_DEVICE_DT,
1252 .ldt_name = LUSTRE_MGS_NAME,
1253 .ldt_ops = &mgs_device_type_ops,
1254 .ldt_ctx_tags = LCT_MG_THREAD
1257 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1258 struct obd_device *obd, struct obd_uuid *cluuid,
1259 struct obd_connect_data *data, void *localdata)
1261 struct obd_export *lexp;
1262 struct lustre_handle conn = { 0 };
1267 if (exp == NULL || obd == NULL || cluuid == NULL)
1270 rc = class_connect(&conn, obd, cluuid);
1274 lexp = class_conn2export(&conn);
1279 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1280 data->ocd_version = LUSTRE_VERSION_CODE;
1281 lexp->exp_connect_data = *data;
1284 tgt_counter_incr(lexp, LPROC_MGS_CONNECT);
1286 rc = mgs_export_stats_init(obd, lexp, localdata);
1288 class_disconnect(lexp);
1295 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1296 struct obd_device *obd, struct obd_uuid *cluuid,
1297 struct obd_connect_data *data, void *localdata)
1301 if (exp == NULL || obd == NULL || cluuid == NULL)
1304 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1307 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1308 data->ocd_version = LUSTRE_VERSION_CODE;
1309 exp->exp_connect_data = *data;
1312 RETURN(mgs_export_stats_init(obd, exp, localdata));
1315 static int mgs_obd_disconnect(struct obd_export *exp)
1323 mgs_fsc_cleanup(exp);
1325 class_export_get(exp);
1326 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1328 rc = server_disconnect_export(exp);
1329 class_export_put(exp);
1333 /* use obd ops to offer management infrastructure */
1334 static struct obd_ops mgs_obd_device_ops = {
1335 .o_owner = THIS_MODULE,
1336 .o_connect = mgs_obd_connect,
1337 .o_reconnect = mgs_obd_reconnect,
1338 .o_disconnect = mgs_obd_disconnect,
1339 .o_init_export = mgs_init_export,
1340 .o_destroy_export = mgs_destroy_export,
1341 .o_iocontrol = mgs_iocontrol,
1344 static int __init mgs_init(void)
1346 struct lprocfs_static_vars lvars;
1348 lprocfs_mgs_init_vars(&lvars);
1349 class_register_type(&mgs_obd_device_ops, NULL, lvars.module_vars,
1350 LUSTRE_MGS_NAME, &mgs_device_type);
1355 static void /*__exit*/ mgs_exit(void)
1357 class_unregister_type(LUSTRE_MGS_NAME);
1360 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1361 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1362 MODULE_LICENSE("GPL");
1364 module_init(mgs_init);
1365 module_exit(mgs_exit);