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);
141 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
142 void *cbdata, enum ast_type type)
146 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
147 LDLM_FL_BLOCK_CONV))) {
150 /* l_ast_data is used as a marker to avoid cancel ldlm lock
151 * twice. See LU-2317. */
152 lock_res_and_lock(lock);
153 fsdb = (struct fs_db *)lock->l_ast_data;
154 lock->l_ast_data = NULL;
155 unlock_res_and_lock(lock);
158 struct lustre_handle lockh;
162 /* clear the bit before lock put */
163 clear_bit(FSDB_REVOKING_LOCK,
167 clear_bit(FSDB_REVOKING_PARAMS,
171 mgs_ir_notify_complete(fsdb);
177 ldlm_lock2handle(lock, &lockh);
178 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
182 RETURN(ldlm_completion_ast(lock, flags, cbdata));
185 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
188 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
191 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
194 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
197 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
200 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
203 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
205 ldlm_completion_callback cp = NULL;
206 struct lustre_handle lockh = { 0 };
207 struct ldlm_res_id res_id;
208 __u64 flags = LDLM_FL_ATOMIC_CB;
212 LASSERT(fsdb->fsdb_name[0] != '\0');
213 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
216 case CONFIG_T_CONFIG:
217 cp = mgs_completion_ast_config;
218 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
221 case CONFIG_T_PARAMS:
222 cp = mgs_completion_ast_params;
223 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
226 case CONFIG_T_RECOVER:
227 cp = mgs_completion_ast_ir;
234 rc = ldlm_cli_enqueue_local(mgs->mgs_obd->obd_namespace,
235 &res_id, LDLM_PLAIN, NULL, LCK_EX,
236 &flags, ldlm_blocking_ast, cp,
237 NULL, fsdb, 0, LVB_T_NONE, NULL,
239 if (rc != ELDLM_OK) {
240 CERROR("can't take cfg lock for "LPX64"/"LPX64"(%d)\n",
241 le64_to_cpu(res_id.name[0]),
242 le64_to_cpu(res_id.name[1]), rc);
244 if (type == CONFIG_T_CONFIG)
245 clear_bit(FSDB_REVOKING_LOCK,
248 if (type == CONFIG_T_PARAMS)
249 clear_bit(FSDB_REVOKING_PARAMS,
252 /* lock has been cancelled in completion_ast. */
261 static int mgs_check_target(const struct lu_env *env,
262 struct mgs_device *mgs,
263 struct mgs_target_info *mti)
268 rc = mgs_check_index(env, mgs, mti);
270 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
271 "this MGS does not know about it, preventing "
272 "registration.\n", mti->mti_svname);
274 } else if (rc == -1) {
275 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
276 "disappeared! Regenerating all logs.\n",
278 mti->mti_flags |= LDD_F_WRITECONF;
281 /* Index is correctly marked as used */
283 /* If the logs don't contain the mti_nids then add
284 them as failover nids */
285 rc = mgs_check_failnid(env, mgs, mti);
291 /* Ensure this is not a failover node that is connecting first*/
292 static int mgs_check_failover_reg(struct mgs_target_info *mti)
298 ptr = mti->mti_params;
299 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
300 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
301 for (i = 0; i < mti->mti_nid_count; i++) {
302 if (nid == mti->mti_nids[i]) {
303 LCONSOLE_WARN("Denying initial registra"
304 "tion attempt from nid %s"
305 ", specified as failover"
306 "\n",libcfs_nid2str(nid));
307 return -EADDRNOTAVAIL;
315 /* Called whenever a target starts up. Flags indicate first connect, etc. */
316 static int mgs_target_reg(struct tgt_session_info *tsi)
318 struct obd_device *obd = tsi->tsi_exp->exp_obd;
319 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
320 struct mgs_target_info *mti, *rep_mti;
327 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
329 return err_serious(rc);
331 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
333 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
335 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
336 RETURN(err_serious(-EFAULT));
339 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
341 opc = mti->mti_flags & LDD_F_OPC_MASK;
345 if (opc == LDD_F_OPC_READY) {
346 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
347 mti->mti_fsname, mti->mti_stripe_index);
348 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
350 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
351 CERROR("Update IR return with %d(ignore and IR "
354 GOTO(out_nolock, rc);
357 /* Do not support unregistering right now. */
358 if (opc != LDD_F_OPC_REG)
359 GOTO(out_nolock, rc = -EINVAL);
361 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
362 mti->mti_fsname, mti->mti_stripe_index);
364 if (mti->mti_flags & LDD_F_NEED_INDEX)
365 mti->mti_flags |= LDD_F_WRITECONF;
367 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
369 /* We're just here as a startup ping. */
370 CDEBUG(D_MGS, "Server %s is running on %s\n",
371 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
372 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
373 /* above will set appropriate mti flags */
375 /* Nothing wrong, or fatal error */
376 GOTO(out_nolock, rc);
377 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
378 rc = mgs_check_failover_reg(mti);
380 GOTO(out_nolock, rc);
383 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
385 if (mti->mti_flags & LDD_F_WRITECONF) {
386 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
387 mti->mti_stripe_index == 0) {
388 rc = mgs_erase_logs(tsi->tsi_env, mgs,
390 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
391 "request. All servers must be restarted "
392 "in order to regenerate the logs."
393 "\n", obd->obd_name, mti->mti_fsname);
394 } else if (mti->mti_flags &
395 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
396 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
397 LCONSOLE_WARN("%s: Regenerating %s log by user "
399 obd->obd_name, mti->mti_svname);
401 mti->mti_flags |= LDD_F_UPDATE;
402 /* Erased logs means start from scratch. */
403 mti->mti_flags &= ~LDD_F_UPGRADE14;
405 GOTO(out_nolock, rc);
408 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &fsdb);
410 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
411 GOTO(out_nolock, rc);
415 * Log writing contention is handled by the fsdb_mutex.
417 * It should be alright if someone was reading while we were
418 * updating the logs - if we revoke at the end they will just update
419 * from where they left off.
422 if (mti->mti_flags & LDD_F_UPGRADE14) {
423 CERROR("Can't upgrade from 1.4 (%d)\n", rc);
427 if (mti->mti_flags & LDD_F_UPDATE) {
428 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
429 mti->mti_stripe_index);
431 /* create or update the target log
432 and update the client/mdt logs */
433 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, fsdb);
435 CERROR("Failed to write %s log (%d)\n",
436 mti->mti_svname, rc);
440 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
441 LDD_F_NEED_INDEX | LDD_F_WRITECONF |
443 mti->mti_flags |= LDD_F_REWRITE_LDD;
447 mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG);
450 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
451 mti->mti_stripe_index, rc);
452 /* An error flag is set in the mti reply rather than an error code */
454 mti->mti_flags |= LDD_F_ERROR;
456 /* send back the whole mti in the reply */
457 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
460 /* Flush logs to disk */
461 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
465 /* Called whenever a target cleans up. */
466 static int mgs_target_del(struct tgt_session_info *tsi)
470 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
475 static int mgs_config_read(struct tgt_session_info *tsi)
477 struct ptlrpc_request *req = tgt_ses_req(tsi);
478 struct mgs_config_body *body;
483 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
485 DEBUG_REQ(D_HA, req, "no mgs_config_body");
486 RETURN(err_serious(-EFAULT));
489 switch (body->mcb_type) {
490 case CONFIG_T_RECOVER:
491 rc = mgs_get_ir_logs(req);
493 case CONFIG_T_CONFIG:
504 static int mgs_llog_open(struct tgt_session_info *tsi)
506 struct mgs_thread_info *mgi;
507 struct ptlrpc_request *req = tgt_ses_req(tsi);
513 rc = tgt_llog_open(tsi);
518 * For old clients there is no direct way of knowing which file system
519 * a client is operating at the MGS side. But we need to pick up those
520 * clients so that the MGS can mark the corresponding file system as
521 * non-IR capable because old clients are not ready to be notified.
523 * Therefore we attempt to detect the file systems name by hacking the
524 * llog operation which is currently used by the clients to fetch
525 * configuration logs. At present this is fine because this is the
526 * ONLY llog operation between mgc and the MGS.
528 * If extra llog operation are going to be added, this function needs
531 * When releases prior than 2.0 are not supported, the following code
534 mgi = mgs_env_info(tsi->tsi_env);
536 RETURN(PTR_ERR(mgi));
538 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
540 char *ptr = strchr(logname, '-');
541 int len = (int)(ptr - logname);
543 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
544 LCONSOLE_WARN("%s: non-config logname received: %s\n",
545 tgt_name(tsi->tsi_tgt), logname);
546 /* not error, this can be llog test name */
548 strncpy(mgi->mgi_fsname, logname, len);
549 mgi->mgi_fsname[len] = 0;
551 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
553 if (rc && rc != -EEXIST) {
554 LCONSOLE_WARN("%s: Unable to add client %s "
555 "to file system %s: %d\n",
556 tgt_name(tsi->tsi_tgt),
557 libcfs_nid2str(req->rq_peer.nid),
558 mgi->mgi_fsname, rc);
564 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
570 static inline int mgs_init_export(struct obd_export *exp)
572 struct mgs_export_data *data = &exp->u.eu_mgs_data;
574 /* init mgs_export_data for fsc */
575 spin_lock_init(&data->med_lock);
576 CFS_INIT_LIST_HEAD(&data->med_clients);
578 spin_lock(&exp->exp_lock);
579 exp->exp_connecting = 1;
580 spin_unlock(&exp->exp_lock);
582 /* self-export doesn't need client data and ldlm initialization */
583 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
584 &exp->exp_client_uuid)))
586 return ldlm_init_export(exp);
589 static inline int mgs_destroy_export(struct obd_export *exp)
593 target_destroy_export(exp);
594 mgs_client_free(exp);
596 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
597 &exp->exp_client_uuid)))
600 ldlm_destroy_export(exp);
605 static int mgs_extract_fs_pool(char * arg, char *fsname, char *poolname)
610 for (ptr = arg; (*ptr != '\0') && (*ptr != '.'); ptr++ ) {
618 strcpy(poolname, ptr);
623 static int mgs_iocontrol_nodemap(const struct lu_env *env,
624 struct mgs_device *mgs,
625 struct obd_ioctl_data *data)
627 struct lustre_cfg *lcfg = NULL;
628 struct lu_nodemap *nodemap;
630 const char *nodemap_name = NULL;
631 const char *nidstr = NULL;
632 const char *client_idstr = NULL;
633 const char *idtype_str = NULL;
644 if (data->ioc_type != LUSTRE_CFG_TYPE) {
645 CERROR("%s: unknown cfg record type: %d\n",
646 mgs->mgs_obd->obd_name, data->ioc_type);
647 GOTO(out, rc = -EINVAL);
650 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
651 GOTO(out, rc = -E2BIG);
653 OBD_ALLOC(lcfg, data->ioc_plen1);
655 GOTO(out, rc = -ENOMEM);
657 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
658 GOTO(out_lcfg, rc = -EFAULT);
660 cmd = lcfg->lcfg_command;
663 case LCFG_NODEMAP_ACTIVATE:
664 if (lcfg->lcfg_bufcount != 2)
665 GOTO(out_lcfg, rc = -EINVAL);
666 param = lustre_cfg_string(lcfg, 1);
667 if (strcmp(param, "1") == 0)
672 case LCFG_NODEMAP_ADD:
673 case LCFG_NODEMAP_DEL:
674 if (lcfg->lcfg_bufcount != 2)
675 GOTO(out_lcfg, rc = -EINVAL);
676 nodemap_name = lustre_cfg_string(lcfg, 1);
677 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
679 case LCFG_NODEMAP_TEST_NID:
680 if (lcfg->lcfg_bufcount != 2)
681 GOTO(out_lcfg, rc = -EINVAL);
682 nidstr = lustre_cfg_string(lcfg, 1);
683 nid = libcfs_str2nid(nidstr);
684 nodemap = nodemap_classify_nid(nid);
685 memset(data->ioc_pbuf1, 0, data->ioc_plen1);
686 if (copy_to_user(data->ioc_pbuf1, nodemap->nm_name,
687 strlen(nodemap->nm_name)) != 0)
688 GOTO(out_lcfg, rc = -EFAULT);
690 case LCFG_NODEMAP_TEST_ID:
691 if (lcfg->lcfg_bufcount != 4)
692 GOTO(out_lcfg, rc = -EINVAL);
693 nidstr = lustre_cfg_string(lcfg, 1);
694 idtype_str = lustre_cfg_string(lcfg, 2);
695 client_idstr = lustre_cfg_string(lcfg, 3);
697 nid = libcfs_str2nid(nidstr);
698 nodemap = nodemap_classify_nid(nid);
699 client_id = simple_strtoul(client_idstr, NULL, 10);
701 if (strcmp(idtype_str, "uid") == 0)
702 idtype = NODEMAP_UID;
704 idtype = NODEMAP_GID;
706 fs_id = nodemap_map_id(nodemap, idtype, NODEMAP_CLIENT_TO_FS,
709 if (data->ioc_plen1 < sizeof(fs_idstr))
710 GOTO(out_lcfg, rc = -EINVAL);
712 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
713 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
714 sizeof(fs_idstr)) != 0)
715 GOTO(out_lcfg, rc = -EINVAL);
717 case LCFG_NODEMAP_ADD_RANGE:
718 case LCFG_NODEMAP_DEL_RANGE:
719 case LCFG_NODEMAP_ADD_UIDMAP:
720 case LCFG_NODEMAP_DEL_UIDMAP:
721 case LCFG_NODEMAP_ADD_GIDMAP:
722 case LCFG_NODEMAP_DEL_GIDMAP:
723 if (lcfg->lcfg_bufcount != 3)
724 GOTO(out_lcfg, rc = -EINVAL);
725 nodemap_name = lustre_cfg_string(lcfg, 1);
726 param = lustre_cfg_string(lcfg, 2);
727 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
729 case LCFG_NODEMAP_ADMIN:
730 case LCFG_NODEMAP_TRUSTED:
731 case LCFG_NODEMAP_SQUASH_UID:
732 case LCFG_NODEMAP_SQUASH_GID:
733 if (lcfg->lcfg_bufcount != 4)
734 GOTO(out_lcfg, rc = -EINVAL);
735 nodemap_name = lustre_cfg_string(lcfg, 1);
736 param = lustre_cfg_string(lcfg, 3);
737 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
744 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
745 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
751 OBD_FREE(lcfg, data->ioc_plen1);
756 static int mgs_iocontrol_pool(const struct lu_env *env,
757 struct mgs_device *mgs,
758 struct obd_ioctl_data *data)
760 struct mgs_thread_info *mgi = mgs_env_info(env);
762 struct lustre_cfg *lcfg = NULL;
763 char *poolname = NULL;
766 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
767 if (poolname == NULL)
770 if (data->ioc_type != LUSTRE_CFG_TYPE) {
771 CERROR("%s: unknown cfg record type: %d\n",
772 mgs->mgs_obd->obd_name, data->ioc_type);
773 GOTO(out_pool, rc = -EINVAL);
776 if (data->ioc_plen1 > PAGE_CACHE_SIZE)
777 GOTO(out_pool, rc = -E2BIG);
779 OBD_ALLOC(lcfg, data->ioc_plen1);
781 GOTO(out_pool, rc = -ENOMEM);
783 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
784 GOTO(out_lcfg, rc = -EFAULT);
786 if (lcfg->lcfg_bufcount < 2)
787 GOTO(out_lcfg, rc = -EFAULT);
789 /* first arg is always <fsname>.<poolname> */
790 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
795 switch (lcfg->lcfg_command) {
797 if (lcfg->lcfg_bufcount != 2)
798 GOTO(out_lcfg, rc = -EINVAL);
799 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
803 if (lcfg->lcfg_bufcount != 3)
804 GOTO(out_lcfg, rc = -EINVAL);
805 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
806 poolname, lustre_cfg_string(lcfg, 2));
809 if (lcfg->lcfg_bufcount != 3)
810 GOTO(out_lcfg, rc = -EINVAL);
811 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
812 poolname, lustre_cfg_string(lcfg, 2));
815 if (lcfg->lcfg_bufcount != 2)
816 GOTO(out_lcfg, rc = -EINVAL);
817 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
825 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
826 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
831 OBD_FREE(lcfg, data->ioc_plen1);
833 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
837 /* from mdt_iocontrol */
838 int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
839 void *karg, void *uarg)
841 struct mgs_device *mgs = exp2mgs_dev(exp);
842 struct obd_ioctl_data *data = karg;
847 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
849 rc = lu_env_init(&env, LCT_MG_THREAD);
855 case OBD_IOC_PARAM: {
856 struct mgs_thread_info *mgi = mgs_env_info(&env);
857 struct lustre_cfg *lcfg;
859 if (data->ioc_type != LUSTRE_CFG_TYPE) {
860 CERROR("%s: unknown cfg record type: %d\n",
861 mgs->mgs_obd->obd_name, data->ioc_type);
862 GOTO(out, rc = -EINVAL);
865 OBD_ALLOC(lcfg, data->ioc_plen1);
867 GOTO(out, rc = -ENOMEM);
868 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
869 GOTO(out_free, rc = -EFAULT);
871 if (lcfg->lcfg_bufcount < 1)
872 GOTO(out_free, rc = -EINVAL);
874 rc = mgs_setparam(&env, mgs, lcfg, mgi->mgi_fsname);
876 CERROR("%s: setparam err: rc = %d\n",
877 exp->exp_obd->obd_name, rc);
879 OBD_FREE(lcfg, data->ioc_plen1);
883 case OBD_IOC_REPLACE_NIDS: {
884 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
885 CERROR("No device name specified!\n");
890 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
891 CERROR("Device name is not NUL terminated!\n");
896 if (data->ioc_plen1 > MTI_NAME_MAXLEN) {
897 CERROR("Device name is too long\n");
902 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
903 CERROR("No NIDs were specified!\n");
908 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
909 CERROR("NID list is not NUL terminated!\n");
914 /* replace nids in llog */
915 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
918 CERROR("%s: error replacing nids: rc = %d\n",
919 exp->exp_obd->obd_name, rc);
925 rc = mgs_iocontrol_pool(&env, mgs, data);
928 case OBD_IOC_NODEMAP:
929 rc = mgs_iocontrol_nodemap(&env, mgs, data);
932 case OBD_IOC_DUMP_LOG: {
933 struct llog_ctxt *ctxt;
935 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
936 rc = class_config_dump_llog(&env, ctxt, data->ioc_inlbuf1,
943 case OBD_IOC_CATLOGLIST:
944 rc = mgs_list_logs(&env, mgs, data);
946 case OBD_IOC_LLOG_CANCEL:
947 case OBD_IOC_LLOG_REMOVE:
948 case OBD_IOC_LLOG_CHECK:
949 case OBD_IOC_LLOG_INFO:
950 case OBD_IOC_LLOG_PRINT: {
951 struct llog_ctxt *ctxt;
953 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
954 rc = llog_ioctl(&env, ctxt, cmd, data);
960 CERROR("%s: unknown command %#x\n",
961 mgs->mgs_obd->obd_name, cmd);
970 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
972 struct obd_connect_data *data = NULL;
973 struct obd_device *obd;
981 obd = class_name2obd(nextdev);
983 CERROR("can't locate next device: %s\n", nextdev);
984 GOTO(out, rc = -ENOTCONN);
987 data->ocd_version = LUSTRE_VERSION_CODE;
989 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
990 &obd->obd_uuid, data, NULL);
992 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
996 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
997 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
998 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1004 static struct tgt_handler mgs_mgs_handlers[] = {
1005 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1006 0, MGS_CONNECT, mgs_connect,
1007 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1008 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1009 0, MGS_DISCONNECT, mgs_disconnect,
1010 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1011 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1012 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_SET_INFO, mgs_set_info),
1013 TGT_MGS_HDL (HABEO_REFERO | MUTABOR, MGS_TARGET_REG, mgs_target_reg),
1014 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1015 TGT_MGS_HDL (HABEO_REFERO, MGS_CONFIG_READ, mgs_config_read),
1018 static struct tgt_handler mgs_obd_handlers[] = {
1019 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1022 static struct tgt_handler mgs_dlm_handlers[] = {
1023 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1024 .th_name = "LDLM_ENQUEUE",
1025 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1026 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1027 * * is safe. If such tests will be needed we have to distinguish
1028 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1029 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1031 .th_opc = LDLM_ENQUEUE,
1032 .th_flags = HABEO_CLAVIS,
1033 .th_act = tgt_enqueue,
1034 .th_fmt = &RQF_LDLM_ENQUEUE,
1035 .th_version = LUSTRE_DLM_VERSION,
1039 static struct tgt_handler mgs_llog_handlers[] = {
1040 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1041 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1042 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1043 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1044 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1047 static struct tgt_opc_slice mgs_common_slice[] = {
1049 .tos_opc_start = MGS_FIRST_OPC,
1050 .tos_opc_end = MGS_LAST_OPC,
1051 .tos_hs = mgs_mgs_handlers
1054 .tos_opc_start = OBD_FIRST_OPC,
1055 .tos_opc_end = OBD_LAST_OPC,
1056 .tos_hs = mgs_obd_handlers
1059 .tos_opc_start = LDLM_FIRST_OPC,
1060 .tos_opc_end = LDLM_LAST_OPC,
1061 .tos_hs = mgs_dlm_handlers
1064 .tos_opc_start = LLOG_FIRST_OPC,
1065 .tos_opc_end = LLOG_LAST_OPC,
1066 .tos_hs = mgs_llog_handlers
1069 .tos_opc_start = SEC_FIRST_OPC,
1070 .tos_opc_end = SEC_LAST_OPC,
1071 .tos_hs = tgt_sec_ctx_handlers
1078 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1079 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1081 struct ptlrpc_service_conf conf;
1082 struct obd_device *obd;
1083 struct lustre_mount_info *lmi;
1084 struct llog_ctxt *ctxt;
1089 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1093 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1095 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1099 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1102 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1104 obd->u.obt.obt_magic = OBT_MAGIC;
1105 obd->u.obt.obt_instance = 0;
1107 /* namespace for mgs llog */
1108 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1109 LDLM_NAMESPACE_SERVER,
1110 LDLM_NAMESPACE_MODEST,
1112 if (obd->obd_namespace == NULL)
1113 GOTO(err_ops, rc = -ENOMEM);
1115 /* No recovery for MGCs */
1116 obd->obd_replayable = 0;
1118 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1119 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1120 OBD_FAIL_MGS_ALL_REPLY_NET);
1124 rc = mgs_fs_setup(env, mgs);
1126 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1131 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1132 obd, &llog_osd_ops);
1136 /* XXX: we need this trick till N:1 stack is supported
1137 * set "current" directory for named llogs */
1138 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1140 ctxt->loc_dir = mgs->mgs_configs_dir;
1141 llog_ctxt_put(ctxt);
1143 /* Internal mgs setup */
1144 mgs_init_fsdb_list(mgs);
1145 mutex_init(&mgs->mgs_mutex);
1146 mgs->mgs_start_time = cfs_time_current_sec();
1147 spin_lock_init(&mgs->mgs_lock);
1149 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1151 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1156 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1157 "mgs_ldlm_client", &obd->obd_ldlm_client);
1159 conf = (typeof(conf)) {
1160 .psc_name = LUSTRE_MGS_NAME,
1161 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1163 .bc_nbufs = MGS_NBUFS,
1164 .bc_buf_size = MGS_BUFSIZE,
1165 .bc_req_max_size = MGS_MAXREQSIZE,
1166 .bc_rep_max_size = MGS_MAXREPSIZE,
1167 .bc_req_portal = MGS_REQUEST_PORTAL,
1168 .bc_rep_portal = MGC_REPLY_PORTAL,
1171 .tc_thr_name = "ll_mgs",
1172 .tc_nthrs_init = MGS_NTHRS_INIT,
1173 .tc_nthrs_max = MGS_NTHRS_MAX,
1174 .tc_ctx_tags = LCT_MG_THREAD,
1177 .so_req_handler = tgt_request_handle,
1178 .so_req_printer = target_print_req,
1182 /* Start the service threads */
1183 mgs->mgs_service = ptlrpc_register_service(&conf, obd->obd_proc_entry);
1184 if (IS_ERR(mgs->mgs_service)) {
1185 rc = PTR_ERR(mgs->mgs_service);
1186 CERROR("failed to start mgs service: %d\n", rc);
1187 mgs->mgs_service = NULL;
1188 GOTO(err_lproc, rc);
1191 ping_evictor_start();
1193 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1195 /* device stack is not yet fully setup to keep no objects behind */
1196 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1199 lproc_mgs_cleanup(mgs);
1201 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1203 ctxt->loc_dir = NULL;
1204 llog_cleanup(env, ctxt);
1207 tgt_fini(env, &mgs->mgs_lut);
1209 /* No extra cleanup needed for llog_init_commit_thread() */
1210 mgs_fs_cleanup(env, mgs);
1212 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1213 obd->obd_namespace = NULL;
1215 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1216 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1217 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1218 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1221 obd_disconnect(mgs->mgs_bottom_exp);
1224 server_put_mount(lustre_cfg_string(lcfg, 0));
1228 static struct lu_device *mgs_device_free(const struct lu_env *env,
1229 struct lu_device *lu)
1231 struct mgs_device *mgs = lu2mgs_dev(lu);
1234 dt_device_fini(&mgs->mgs_dt_dev);
1239 static int mgs_process_config(const struct lu_env *env,
1240 struct lu_device *dev,
1241 struct lustre_cfg *lcfg)
1247 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1248 const struct lu_object_conf *unused)
1250 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1251 struct lu_device *under;
1252 struct lu_object *below;
1256 /* do no set .do_ops as mgs calls to bottom osd directly */
1258 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1259 PFID(lu_object_fid(o)));
1261 under = &d->mgs_bottom->dd_lu_dev;
1262 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1264 lu_object_add(o, below);
1271 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1273 struct mgs_object *obj = lu2mgs_obj(o);
1274 struct lu_object_header *h = o->lo_header;
1276 dt_object_fini(&obj->mgo_obj);
1277 lu_object_header_fini(h);
1281 static int mgs_object_print(const struct lu_env *env, void *cookie,
1282 lu_printer_t p, const struct lu_object *l)
1284 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1286 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1289 struct lu_object_operations mgs_lu_obj_ops = {
1290 .loo_object_init = mgs_object_init,
1291 .loo_object_free = mgs_object_free,
1292 .loo_object_print = mgs_object_print,
1295 struct lu_object *mgs_object_alloc(const struct lu_env *env,
1296 const struct lu_object_header *hdr,
1297 struct lu_device *d)
1299 struct lu_object_header *h;
1300 struct mgs_object *o;
1301 struct lu_object *l;
1303 LASSERT(hdr == NULL);
1307 l = &o->mgo_obj.do_lu;
1310 lu_object_header_init(h);
1311 dt_object_init(&o->mgo_obj, h, d);
1312 lu_object_add_top(h, l);
1314 l->lo_ops = &mgs_lu_obj_ops;
1322 const struct lu_device_operations mgs_lu_ops = {
1323 .ldo_object_alloc = mgs_object_alloc,
1324 .ldo_process_config = mgs_process_config,
1327 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1328 struct lu_device_type *type,
1329 struct lustre_cfg *lcfg)
1331 struct mgs_device *mgs;
1332 struct lu_device *ludev;
1336 ludev = ERR_PTR(-ENOMEM);
1340 ludev = mgs2lu_dev(mgs);
1341 dt_device_init(&mgs->mgs_dt_dev, type);
1342 rc = mgs_init0(env, mgs, type, lcfg);
1344 mgs_device_free(env, ludev);
1345 ludev = ERR_PTR(rc);
1351 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1352 struct lu_device *d)
1354 struct mgs_device *mgs = lu2mgs_dev(d);
1355 struct obd_device *obd = mgs->mgs_obd;
1356 struct llog_ctxt *ctxt;
1360 LASSERT(mgs->mgs_bottom);
1362 class_disconnect_exports(obd);
1364 ping_evictor_stop();
1366 ptlrpc_unregister_service(mgs->mgs_service);
1368 obd_exports_barrier(obd);
1369 obd_zombie_barrier();
1371 tgt_fini(env, &mgs->mgs_lut);
1372 mgs_cleanup_fsdb_list(mgs);
1373 lproc_mgs_cleanup(mgs);
1375 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1377 ctxt->loc_dir = NULL;
1378 llog_cleanup(env, ctxt);
1381 mgs_fs_cleanup(env, mgs);
1383 ldlm_namespace_free(obd->obd_namespace, NULL, 1);
1384 obd->obd_namespace = NULL;
1386 lu_site_purge(env, d->ld_site, ~0);
1387 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1388 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1389 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1392 LASSERT(mgs->mgs_bottom_exp);
1393 obd_disconnect(mgs->mgs_bottom_exp);
1395 server_put_mount(obd->obd_name);
1400 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1401 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1403 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1405 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1407 static struct lu_device_type_operations mgs_device_type_ops = {
1408 .ldto_init = mgs_type_init,
1409 .ldto_fini = mgs_type_fini,
1411 .ldto_start = mgs_type_start,
1412 .ldto_stop = mgs_type_stop,
1414 .ldto_device_alloc = mgs_device_alloc,
1415 .ldto_device_free = mgs_device_free,
1417 .ldto_device_fini = mgs_device_fini
1420 static struct lu_device_type mgs_device_type = {
1421 .ldt_tags = LU_DEVICE_DT,
1422 .ldt_name = LUSTRE_MGS_NAME,
1423 .ldt_ops = &mgs_device_type_ops,
1424 .ldt_ctx_tags = LCT_MG_THREAD
1427 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1428 struct obd_device *obd, struct obd_uuid *cluuid,
1429 struct obd_connect_data *data, void *localdata)
1431 struct obd_export *lexp;
1432 struct lustre_handle conn = { 0 };
1437 if (exp == NULL || obd == NULL || cluuid == NULL)
1440 rc = class_connect(&conn, obd, cluuid);
1444 lexp = class_conn2export(&conn);
1449 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1450 data->ocd_version = LUSTRE_VERSION_CODE;
1451 lexp->exp_connect_data = *data;
1454 tgt_counter_incr(lexp, LPROC_MGS_CONNECT);
1456 rc = mgs_export_stats_init(obd, lexp, localdata);
1458 class_disconnect(lexp);
1465 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1466 struct obd_device *obd, struct obd_uuid *cluuid,
1467 struct obd_connect_data *data, void *localdata)
1471 if (exp == NULL || obd == NULL || cluuid == NULL)
1474 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1477 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1478 data->ocd_version = LUSTRE_VERSION_CODE;
1479 exp->exp_connect_data = *data;
1482 RETURN(mgs_export_stats_init(obd, exp, localdata));
1485 static int mgs_obd_disconnect(struct obd_export *exp)
1493 mgs_fsc_cleanup(exp);
1495 class_export_get(exp);
1496 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1498 rc = server_disconnect_export(exp);
1499 class_export_put(exp);
1503 /* use obd ops to offer management infrastructure */
1504 static struct obd_ops mgs_obd_device_ops = {
1505 .o_owner = THIS_MODULE,
1506 .o_connect = mgs_obd_connect,
1507 .o_reconnect = mgs_obd_reconnect,
1508 .o_disconnect = mgs_obd_disconnect,
1509 .o_init_export = mgs_init_export,
1510 .o_destroy_export = mgs_destroy_export,
1511 .o_iocontrol = mgs_iocontrol,
1514 static int __init mgs_init(void)
1516 struct lprocfs_static_vars lvars;
1518 lprocfs_mgs_init_vars(&lvars);
1519 return class_register_type(&mgs_obd_device_ops, NULL, NULL,
1520 #ifndef HAVE_ONLY_PROCFS_SEQ
1523 LUSTRE_MGS_NAME, &mgs_device_type);
1526 static void /*__exit*/ mgs_exit(void)
1528 class_unregister_type(LUSTRE_MGS_NAME);
1531 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1532 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1533 MODULE_LICENSE("GPL");
1535 module_init(mgs_init);
1536 module_exit(mgs_exit);