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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mgs/mgs_handler.c
34 * Author: Nathan Rutman <nathan@clusterfs.com>
35 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
36 * Author: Mikhail Pershin <tappro@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_MGS
40 #define D_MGS D_CONFIG
42 #include <obd_class.h>
43 #include <lprocfs_status.h>
44 #include <uapi/linux/lustre/lustre_ioctl.h>
45 #include <uapi/linux/lustre/lustre_param.h>
46 #include <uapi/linux/lustre/lustre_barrier_user.h>
48 #include "mgs_internal.h"
51 * Regular MGS handlers
53 static int mgs_connect(struct tgt_session_info *tsi)
55 struct ptlrpc_request *req = tgt_ses_req(tsi);
60 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_CONNECT_NET, cfs_fail_val);
61 rc = tgt_connect(tsi);
65 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
66 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
71 static int mgs_disconnect(struct tgt_session_info *tsi)
77 LASSERT(tsi->tsi_exp);
79 rc = tgt_disconnect(tsi);
81 RETURN(err_serious(rc));
85 static int mgs_exception(struct tgt_session_info *tsi)
89 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
94 static inline bool str_starts_with(const char *str, const char *prefix)
96 return strncmp(str, prefix, strlen(prefix)) == 0;
99 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
100 static int mgs_set_info(struct tgt_session_info *tsi)
102 struct mgs_thread_info *mgi;
103 struct mgs_send_param *msp, *rep_msp;
104 struct lustre_cfg *lcfg;
111 mgi = mgs_env_info(tsi->tsi_env);
113 RETURN(err_serious(PTR_ERR(mgi)));
115 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
117 RETURN(err_serious(-EFAULT));
119 param_len = strnlen(msp->mgs_param, sizeof(msp->mgs_param));
120 if (param_len == 0 || param_len == sizeof(msp->mgs_param))
123 /* We only allow '*.lov.stripe{size,count,offset}=*' from an RPC. */
124 s = strchr(msp->mgs_param, '.');
128 if (!str_starts_with(s + 1, "lov.stripesize=") &&
129 !str_starts_with(s + 1, "lov.stripecount=") &&
130 !str_starts_with(s + 1, "lov.stripeoffset="))
133 /* Construct lustre_cfg structure to pass to function mgs_set_param */
134 lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL);
135 lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, msp->mgs_param);
136 OBD_ALLOC(lcfg, lustre_cfg_len(mgi->mgi_bufs.lcfg_bufcount,
137 mgi->mgi_bufs.lcfg_buflen));
140 lustre_cfg_init(lcfg, LCFG_PARAM, &mgi->mgi_bufs);
142 rc = mgs_set_param(tsi->tsi_env, exp2mgs_dev(tsi->tsi_exp), lcfg);
144 LCONSOLE_WARN("%s: Unable to set parameter %s: %d\n",
145 tgt_name(tsi->tsi_tgt), msp->mgs_param, rc);
149 /* send back the whole msp in the reply */
150 rep_msp = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
154 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
166 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
167 void *cbdata, enum ast_type type)
171 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
174 /* l_ast_data is used as a marker to avoid cancel ldlm lock
175 * twice. See LU-2317. */
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;
186 /* clear the bit before lock put */
187 clear_bit(FSDB_REVOKING_LOCK,
191 clear_bit(FSDB_REVOKING_PARAMS,
195 mgs_ir_notify_complete(fsdb);
203 ldlm_lock2handle(lock, &lockh);
204 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
208 RETURN(ldlm_completion_ast(lock, flags, cbdata));
211 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
214 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
217 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
220 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
223 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
226 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
229 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
232 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
235 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb, int type)
237 ldlm_completion_callback cp = NULL;
238 struct lustre_handle lockh = {
241 struct ldlm_res_id res_id;
242 __u64 flags = LDLM_FL_ATOMIC_CB;
246 LASSERT(fsdb->fsdb_name[0] != '\0');
247 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
250 case CONFIG_T_CONFIG:
251 case CONFIG_T_NODEMAP:
252 cp = mgs_completion_ast_config;
253 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
256 case CONFIG_T_PARAMS:
257 cp = mgs_completion_ast_params;
258 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
261 case CONFIG_T_RECOVER:
262 cp = mgs_completion_ast_ir;
264 case CONFIG_T_BARRIER:
265 cp = mgs_completion_ast_barrier;
273 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
274 &res_id, LDLM_PLAIN, NULL, LCK_EX,
275 &flags, ldlm_blocking_ast, cp,
276 NULL, fsdb, 0, LVB_T_NONE, NULL,
278 if (rc != ELDLM_OK) {
279 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
280 mgs->mgs_obd->obd_name,
281 le64_to_cpu(res_id.name[0]),
282 le64_to_cpu(res_id.name[1]), rc);
284 if (type == CONFIG_T_CONFIG)
285 clear_bit(FSDB_REVOKING_LOCK,
288 if (type == CONFIG_T_PARAMS)
289 clear_bit(FSDB_REVOKING_PARAMS,
292 /* lock has been cancelled in completion_ast. */
301 static int mgs_check_target(const struct lu_env *env,
302 struct mgs_device *mgs,
303 struct mgs_target_info *mti)
308 rc = mgs_check_index(env, mgs, mti);
310 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
311 "this MGS does not know about it, preventing "
312 "registration.\n", mti->mti_svname);
314 } else if (rc == -1) {
315 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
316 "disappeared! Regenerating all logs.\n",
318 mti->mti_flags |= LDD_F_WRITECONF;
321 /* Index is correctly marked as used */
328 /* Ensure this is not a failover node that is connecting first*/
329 static int mgs_check_failover_reg(struct mgs_target_info *mti)
335 ptr = mti->mti_params;
336 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
337 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
338 for (i = 0; i < mti->mti_nid_count; i++) {
339 if (nid == mti->mti_nids[i]) {
340 LCONSOLE_WARN("Denying initial registra"
341 "tion attempt from nid %s"
342 ", specified as failover"
343 "\n",libcfs_nid2str(nid));
344 return -EADDRNOTAVAIL;
352 /* Called whenever a target starts up. Flags indicate first connect, etc. */
353 static int mgs_target_reg(struct tgt_session_info *tsi)
355 struct obd_device *obd = tsi->tsi_exp->exp_obd;
356 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
357 struct mgs_target_info *mti, *rep_mti;
358 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
359 struct fs_db *c_fsdb = NULL; /* config fsdb */
360 char barrier_name[20];
366 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
368 return err_serious(rc);
370 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
372 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
374 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
375 RETURN(err_serious(-EFAULT));
378 down_read(&mgs->mgs_barrier_rwsem);
380 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
382 opc = mti->mti_flags & LDD_F_OPC_MASK;
386 if (opc == LDD_F_OPC_READY) {
387 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
388 mti->mti_fsname, mti->mti_stripe_index);
389 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
391 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
392 CERROR("%s: Update IR return failure: rc = %d\n",
393 mti->mti_fsname, rc);
396 GOTO(out_norevoke, rc);
399 /* Do not support unregistering right now. */
400 if (opc != LDD_F_OPC_REG)
401 GOTO(out_norevoke, rc = -EINVAL);
403 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
404 mti->mti_fsname, BARRIER_FILENAME);
405 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
407 CERROR("%s: Can't get db for %s: rc = %d\n",
408 mti->mti_fsname, barrier_name, rc);
410 GOTO(out_norevoke, rc);
413 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
414 mti->mti_fsname, mti->mti_stripe_index);
416 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
417 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
418 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
419 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
420 LCONSOLE_WARN("%s: the system is in barrier, refuse "
421 "the connection from MDT %s temporary\n",
422 obd->obd_name, mti->mti_svname);
424 GOTO(out_norevoke, rc = -EBUSY);
427 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
428 !b_fsdb->fsdb_barrier_disabled) {
429 LCONSOLE_WARN("%s: the MDT %s does not support write "
430 "barrier, so disable barrier on the "
432 obd->obd_name, mti->mti_svname);
434 b_fsdb->fsdb_barrier_disabled = 1;
438 if (mti->mti_flags & LDD_F_NEED_INDEX)
439 mti->mti_flags |= LDD_F_WRITECONF;
441 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
442 /* We're just here as a startup ping. */
443 CDEBUG(D_MGS, "Server %s is running on %s\n",
444 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
445 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
446 /* above will set appropriate mti flags */
448 /* Nothing wrong, or fatal error */
449 GOTO(out_norevoke, rc);
450 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
451 rc = mgs_check_failover_reg(mti);
453 GOTO(out_norevoke, rc);
456 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
458 if (mti->mti_flags & LDD_F_WRITECONF) {
459 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
460 mti->mti_stripe_index == 0) {
461 mgs_put_fsdb(mgs, b_fsdb);
463 rc = mgs_erase_logs(tsi->tsi_env, mgs,
465 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
466 "request. All servers must be restarted "
467 "in order to regenerate the logs: rc = %d"
468 "\n", obd->obd_name, mti->mti_fsname, rc);
469 if (rc && rc != -ENOENT)
470 GOTO(out_norevoke, rc);
472 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
473 barrier_name, &b_fsdb);
475 CERROR("Can't get db for %s: %d\n",
478 GOTO(out_norevoke, rc);
481 if (!(exp_connect_flags(tsi->tsi_exp) &
482 OBD_CONNECT_BARRIER)) {
483 LCONSOLE_WARN("%s: the MDT %s does not support "
484 "write barrier, disable barrier "
485 "on the whole system.\n",
486 obd->obd_name, mti->mti_svname);
488 b_fsdb->fsdb_barrier_disabled = 1;
490 } else if (mti->mti_flags &
491 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
492 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
493 LCONSOLE_WARN("%s: Regenerating %s log by user "
494 "request: rc = %d\n",
495 obd->obd_name, mti->mti_svname, rc);
497 GOTO(out_norevoke, rc);
500 mti->mti_flags |= LDD_F_UPDATE;
503 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
505 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
507 GOTO(out_norevoke, rc);
511 * Log writing contention is handled by the fsdb_mutex.
513 * It should be alright if someone was reading while we were
514 * updating the logs - if we revoke at the end they will just update
515 * from where they left off.
517 if (mti->mti_flags & LDD_F_UPDATE) {
518 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
519 mti->mti_stripe_index);
521 /* create or update the target log
522 and update the client/mdt logs */
523 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
525 CERROR("Failed to write %s log (%d)\n",
526 mti->mti_svname, rc);
530 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
531 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
532 mti->mti_flags |= LDD_F_REWRITE_LDD;
536 mgs_revoke_lock(mgs, c_fsdb, CONFIG_T_CONFIG);
539 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
541 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
542 mti->mti_fsname, &c_fsdb);
544 CERROR("Fail to get db for %s: %d\n",
545 mti->mti_fsname, rc);
549 memcpy(b_fsdb->fsdb_mdt_index_map,
550 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
551 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
555 up_read(&mgs->mgs_barrier_rwsem);
557 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
558 mti->mti_stripe_index, rc);
559 /* An error flag is set in the mti reply rather than an error code */
561 mti->mti_flags |= LDD_F_ERROR;
563 /* send back the whole mti in the reply */
564 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
567 /* Flush logs to disk */
568 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
570 mgs_put_fsdb(mgs, b_fsdb);
572 mgs_put_fsdb(mgs, c_fsdb);
576 /* Called whenever a target cleans up. */
577 static int mgs_target_del(struct tgt_session_info *tsi)
581 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
586 static int mgs_config_read(struct tgt_session_info *tsi)
588 struct ptlrpc_request *req = tgt_ses_req(tsi);
589 struct mgs_config_body *body;
594 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
596 DEBUG_REQ(D_HA, req, "no mgs_config_body");
597 RETURN(err_serious(-EFAULT));
600 switch (body->mcb_type) {
601 case CONFIG_T_RECOVER:
602 rc = mgs_get_ir_logs(req);
604 case CONFIG_T_NODEMAP:
605 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
607 case CONFIG_T_CONFIG:
618 static int mgs_llog_open(struct tgt_session_info *tsi)
620 struct mgs_thread_info *mgi;
621 struct ptlrpc_request *req = tgt_ses_req(tsi);
627 rc = tgt_llog_open(tsi);
632 * For old clients there is no direct way of knowing which file system
633 * a client is operating at the MGS side. But we need to pick up those
634 * clients so that the MGS can mark the corresponding file system as
635 * non-IR capable because old clients are not ready to be notified.
637 * Therefore we attempt to detect the file systems name by hacking the
638 * llog operation which is currently used by the clients to fetch
639 * configuration logs. At present this is fine because this is the
640 * ONLY llog operation between mgc and the MGS.
642 * If extra llog operation are going to be added, this function needs
645 * When releases prior than 2.0 are not supported, the following code
648 mgi = mgs_env_info(tsi->tsi_env);
650 RETURN(PTR_ERR(mgi));
652 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
654 char *ptr = strrchr(logname, '-');
655 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
657 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
658 if (strcmp(logname, PARAMS_FILENAME) != 0)
659 LCONSOLE_WARN("%s: non-config logname "
661 tgt_name(tsi->tsi_tgt),
663 /* not error, this can be llog test name */
665 strncpy(mgi->mgi_fsname, logname, len);
666 mgi->mgi_fsname[len] = 0;
668 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
670 if (rc && rc != -EEXIST) {
671 LCONSOLE_WARN("%s: Unable to add client %s "
672 "to file system %s: %d\n",
673 tgt_name(tsi->tsi_tgt),
674 libcfs_nid2str(req->rq_peer.nid),
675 mgi->mgi_fsname, rc);
681 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
687 static inline int mgs_init_export(struct obd_export *exp)
689 struct mgs_export_data *data = &exp->u.eu_mgs_data;
691 /* init mgs_export_data for fsc */
692 spin_lock_init(&data->med_lock);
693 INIT_LIST_HEAD(&data->med_clients);
695 spin_lock(&exp->exp_lock);
696 exp->exp_connecting = 1;
697 spin_unlock(&exp->exp_lock);
699 /* self-export doesn't need client data and ldlm initialization */
700 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
701 &exp->exp_client_uuid)))
703 return ldlm_init_export(exp);
706 static inline int mgs_destroy_export(struct obd_export *exp)
710 target_destroy_export(exp);
711 mgs_client_free(exp);
713 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
714 &exp->exp_client_uuid)))
717 ldlm_destroy_export(exp);
722 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
729 for (ptr = arg; *ptr != '\0'; ptr++) {
730 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
734 /* Test for fsname.poolname format.
735 * strlen() test if poolname is empty
737 ptr = strchr(arg, '.');
738 if (!ptr || !strlen(ptr))
742 /* Also make sure poolname is not to long. */
743 if (strlen(ptr) > LOV_MAXPOOLNAME)
744 return -ENAMETOOLONG;
745 strncpy(poolname, ptr, LOV_MAXPOOLNAME);
747 /* Test if fsname is empty */
748 len = strlen(arg) - strlen(ptr) - 1;
753 if (len > LUSTRE_MAXFSNAME)
754 return -ENAMETOOLONG;
756 strncpy(fsname, arg, len);
761 static int mgs_iocontrol_nodemap(const struct lu_env *env,
762 struct mgs_device *mgs,
763 struct obd_ioctl_data *data)
765 struct lustre_cfg *lcfg = NULL;
768 const char *nodemap_name = NULL;
769 const char *nidstr = NULL;
770 const char *client_idstr = NULL;
771 const char *idtype_str = NULL;
774 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
776 unsigned long client_id;
783 if (data->ioc_type != LUSTRE_CFG_TYPE) {
784 CERROR("%s: unknown cfg record type: %d\n",
785 mgs->mgs_obd->obd_name, data->ioc_type);
786 GOTO(out, rc = -EINVAL);
789 if (data->ioc_plen1 > PAGE_SIZE)
790 GOTO(out, rc = -E2BIG);
792 OBD_ALLOC(lcfg, data->ioc_plen1);
794 GOTO(out, rc = -ENOMEM);
796 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
797 GOTO(out_lcfg, rc = -EFAULT);
799 cmd = lcfg->lcfg_command;
802 case LCFG_NODEMAP_ACTIVATE:
803 if (lcfg->lcfg_bufcount != 2)
804 GOTO(out_lcfg, rc = -EINVAL);
805 param = lustre_cfg_string(lcfg, 1);
806 if (strcmp(param, "1") == 0)
811 case LCFG_NODEMAP_ADD:
812 case LCFG_NODEMAP_DEL:
813 if (lcfg->lcfg_bufcount != 2)
814 GOTO(out_lcfg, rc = -EINVAL);
815 nodemap_name = lustre_cfg_string(lcfg, 1);
816 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
818 case LCFG_NODEMAP_TEST_NID:
819 if (lcfg->lcfg_bufcount != 2)
820 GOTO(out_lcfg, rc = -EINVAL);
821 nidstr = lustre_cfg_string(lcfg, 1);
822 nid = libcfs_str2nid(nidstr);
823 nodemap_test_nid(nid, name_buf, sizeof(name_buf));
824 rc = copy_to_user(data->ioc_pbuf1, name_buf,
825 MIN(data->ioc_plen1, sizeof(name_buf)));
827 GOTO(out_lcfg, rc = -EFAULT);
829 case LCFG_NODEMAP_TEST_ID:
830 if (lcfg->lcfg_bufcount != 4)
831 GOTO(out_lcfg, rc = -EINVAL);
832 nidstr = lustre_cfg_string(lcfg, 1);
833 idtype_str = lustre_cfg_string(lcfg, 2);
834 client_idstr = lustre_cfg_string(lcfg, 3);
836 nid = libcfs_str2nid(nidstr);
837 if (strcmp(idtype_str, "uid") == 0)
838 idtype = NODEMAP_UID;
840 idtype = NODEMAP_GID;
842 rc = kstrtoul(client_idstr, 10, &client_id);
844 GOTO(out_lcfg, rc = -EINVAL);
846 rc = nodemap_test_id(nid, idtype, client_id, &fs_id);
848 GOTO(out_lcfg, rc = -EINVAL);
850 if (data->ioc_plen1 < sizeof(fs_idstr))
851 GOTO(out_lcfg, rc = -EINVAL);
853 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
854 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
855 sizeof(fs_idstr)) != 0)
856 GOTO(out_lcfg, rc = -EINVAL);
858 case LCFG_NODEMAP_ADD_RANGE:
859 case LCFG_NODEMAP_DEL_RANGE:
860 case LCFG_NODEMAP_ADD_UIDMAP:
861 case LCFG_NODEMAP_DEL_UIDMAP:
862 case LCFG_NODEMAP_ADD_GIDMAP:
863 case LCFG_NODEMAP_DEL_GIDMAP:
864 case LCFG_NODEMAP_SET_FILESET:
865 case LCFG_NODEMAP_SET_SEPOL:
866 if (lcfg->lcfg_bufcount != 3)
867 GOTO(out_lcfg, rc = -EINVAL);
868 nodemap_name = lustre_cfg_string(lcfg, 1);
869 param = lustre_cfg_string(lcfg, 2);
870 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
872 case LCFG_NODEMAP_ADMIN:
873 case LCFG_NODEMAP_TRUSTED:
874 case LCFG_NODEMAP_DENY_UNKNOWN:
875 case LCFG_NODEMAP_SQUASH_UID:
876 case LCFG_NODEMAP_SQUASH_GID:
877 case LCFG_NODEMAP_MAP_MODE:
878 case LCFG_NODEMAP_AUDIT_MODE:
879 if (lcfg->lcfg_bufcount != 4)
880 GOTO(out_lcfg, rc = -EINVAL);
881 nodemap_name = lustre_cfg_string(lcfg, 1);
882 param = lustre_cfg_string(lcfg, 3);
883 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
890 CERROR("%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
891 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
896 /* revoke nodemap lock */
897 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
899 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
900 mgs->mgs_obd->obd_name, rc);
902 mgs_revoke_lock(mgs, fsdb, CONFIG_T_NODEMAP);
903 mgs_put_fsdb(mgs, fsdb);
907 OBD_FREE(lcfg, data->ioc_plen1);
912 static int mgs_iocontrol_pool(const struct lu_env *env,
913 struct mgs_device *mgs,
914 struct obd_ioctl_data *data)
916 struct mgs_thread_info *mgi = mgs_env_info(env);
918 struct lustre_cfg *lcfg = NULL;
919 char *poolname = NULL;
922 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
923 if (poolname == NULL)
926 if (data->ioc_type != LUSTRE_CFG_TYPE) {
927 CERROR("%s: unknown cfg record type: %d\n",
928 mgs->mgs_obd->obd_name, data->ioc_type);
929 GOTO(out_pool, rc = -EINVAL);
932 if (data->ioc_plen1 > PAGE_SIZE)
933 GOTO(out_pool, rc = -E2BIG);
935 OBD_ALLOC(lcfg, data->ioc_plen1);
937 GOTO(out_pool, rc = -ENOMEM);
939 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
940 GOTO(out_lcfg, rc = -EFAULT);
942 if (lcfg->lcfg_bufcount < 2)
943 GOTO(out_lcfg, rc = -EFAULT);
945 /* first arg is always <fsname>.<poolname> */
946 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
951 switch (lcfg->lcfg_command) {
953 if (lcfg->lcfg_bufcount != 2)
954 GOTO(out_lcfg, rc = -EINVAL);
955 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
959 if (lcfg->lcfg_bufcount != 3)
960 GOTO(out_lcfg, rc = -EINVAL);
961 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
962 poolname, lustre_cfg_string(lcfg, 2));
965 if (lcfg->lcfg_bufcount != 3)
966 GOTO(out_lcfg, rc = -EINVAL);
967 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
968 poolname, lustre_cfg_string(lcfg, 2));
971 if (lcfg->lcfg_bufcount != 2)
972 GOTO(out_lcfg, rc = -EINVAL);
973 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
981 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
982 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
987 OBD_FREE(lcfg, data->ioc_plen1);
989 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
993 /* from mdt_iocontrol */
994 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
995 void *karg, void __user *uarg)
997 struct mgs_device *mgs = exp2mgs_dev(exp);
998 struct obd_ioctl_data *data = karg;
1003 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1005 rc = lu_env_init(&env, LCT_MG_THREAD);
1011 case OBD_IOC_PARAM: {
1012 struct lustre_cfg *lcfg;
1014 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1015 CERROR("%s: unknown cfg record type: %d\n",
1016 mgs->mgs_obd->obd_name, data->ioc_type);
1017 GOTO(out, rc = -EINVAL);
1020 OBD_ALLOC(lcfg, data->ioc_plen1);
1022 GOTO(out, rc = -ENOMEM);
1023 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1024 GOTO(out_free, rc = -EFAULT);
1026 if (lcfg->lcfg_bufcount < 1)
1027 GOTO(out_free, rc = -EINVAL);
1029 rc = mgs_set_param(&env, mgs, lcfg);
1031 CERROR("%s: setparam err: rc = %d\n",
1032 exp->exp_obd->obd_name, rc);
1034 OBD_FREE(lcfg, data->ioc_plen1);
1038 case OBD_IOC_REPLACE_NIDS: {
1039 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1041 CERROR("%s: no device or fsname specified: rc = %d\n",
1042 exp->exp_obd->obd_name, rc);
1046 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1048 CERROR("%s: device or fsname is too long: rc = %d\n",
1049 exp->exp_obd->obd_name, rc);
1053 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1055 CERROR("%s: device or fsname is not NUL terminated: "
1056 "rc = %d\n", exp->exp_obd->obd_name, rc);
1060 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1062 CERROR("%s: no NIDs specified: rc = %d\n",
1063 exp->exp_obd->obd_name, rc);
1067 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1069 CERROR("%s: NID list is not NUL terminated: "
1070 "rc = %d\n", exp->exp_obd->obd_name, rc);
1074 /* replace nids in llog */
1075 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1078 CERROR("%s: error replacing nids: rc = %d\n",
1079 exp->exp_obd->obd_name, rc);
1084 case OBD_IOC_CLEAR_CONFIGS: {
1085 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1087 CERROR("%s: no device or fsname specified: rc = %d\n",
1088 exp->exp_obd->obd_name, rc);
1092 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1094 CERROR("%s: device or fsname is too long: rc = %d\n",
1095 exp->exp_obd->obd_name, rc);
1099 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1101 CERROR("%s: device or fsname is not NUL terminated: "
1102 "rc = %d\n", exp->exp_obd->obd_name, rc);
1106 /* remove records marked SKIP from config logs */
1107 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1109 CERROR("%s: error clearing config log: rc = %d\n",
1110 exp->exp_obd->obd_name, rc);
1116 rc = mgs_iocontrol_pool(&env, mgs, data);
1119 case OBD_IOC_BARRIER:
1120 rc = mgs_iocontrol_barrier(&env, mgs, data);
1123 case OBD_IOC_NODEMAP:
1124 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1127 case OBD_IOC_LCFG_FORK:
1128 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1132 case OBD_IOC_LCFG_ERASE:
1133 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1136 case OBD_IOC_CATLOGLIST:
1137 rc = mgs_list_logs(&env, mgs, data);
1139 case OBD_IOC_LLOG_CANCEL:
1140 case OBD_IOC_LLOG_REMOVE:
1141 case OBD_IOC_LLOG_CHECK:
1142 case OBD_IOC_LLOG_INFO:
1143 case OBD_IOC_LLOG_PRINT: {
1144 struct llog_ctxt *ctxt;
1146 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1147 rc = llog_ioctl(&env, ctxt, cmd, data);
1148 llog_ctxt_put(ctxt);
1153 CERROR("%s: unknown command %#x\n",
1154 mgs->mgs_obd->obd_name, cmd);
1163 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1165 struct obd_connect_data *data = NULL;
1166 struct obd_device *obd;
1170 OBD_ALLOC_PTR(data);
1174 obd = class_name2obd(nextdev);
1176 CERROR("can't locate next device: %s\n", nextdev);
1177 GOTO(out, rc = -ENOTCONN);
1180 data->ocd_version = LUSTRE_VERSION_CODE;
1182 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1183 &obd->obd_uuid, data, NULL);
1185 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1189 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1190 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1191 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1197 static struct tgt_handler mgs_mgs_handlers[] = {
1198 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1199 0, MGS_CONNECT, mgs_connect,
1200 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1201 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1202 0, MGS_DISCONNECT, mgs_disconnect,
1203 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1204 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1205 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
1206 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_SET_INFO, mgs_set_info),
1208 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1209 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1210 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1213 static struct tgt_handler mgs_obd_handlers[] = {
1214 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1217 static struct tgt_handler mgs_dlm_handlers[] = {
1218 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1219 .th_name = "LDLM_ENQUEUE",
1220 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1221 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1222 * * is safe. If such tests will be needed we have to distinguish
1223 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1224 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1226 .th_opc = LDLM_ENQUEUE,
1227 .th_flags = HAS_KEY,
1228 .th_act = tgt_enqueue,
1229 .th_fmt = &RQF_LDLM_ENQUEUE,
1230 .th_version = LUSTRE_DLM_VERSION,
1234 static struct tgt_handler mgs_llog_handlers[] = {
1235 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1236 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1237 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1238 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1241 static struct tgt_opc_slice mgs_common_slice[] = {
1243 .tos_opc_start = MGS_FIRST_OPC,
1244 .tos_opc_end = MGS_LAST_OPC,
1245 .tos_hs = mgs_mgs_handlers
1248 .tos_opc_start = OBD_FIRST_OPC,
1249 .tos_opc_end = OBD_LAST_OPC,
1250 .tos_hs = mgs_obd_handlers
1253 .tos_opc_start = LDLM_FIRST_OPC,
1254 .tos_opc_end = LDLM_LAST_OPC,
1255 .tos_hs = mgs_dlm_handlers
1258 .tos_opc_start = LLOG_FIRST_OPC,
1259 .tos_opc_end = LLOG_LAST_OPC,
1260 .tos_hs = mgs_llog_handlers
1263 .tos_opc_start = SEC_FIRST_OPC,
1264 .tos_opc_end = SEC_LAST_OPC,
1265 .tos_hs = tgt_sec_ctx_handlers
1272 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1273 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1275 struct ptlrpc_service_conf conf;
1276 struct obd_device *obd;
1277 struct lustre_mount_info *lmi;
1278 struct llog_ctxt *ctxt;
1283 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1287 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1289 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1293 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1296 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1298 obd->u.obt.obt_magic = OBT_MAGIC;
1299 obd->u.obt.obt_instance = 0;
1301 /* namespace for mgs llog */
1302 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1303 LDLM_NAMESPACE_SERVER,
1304 LDLM_NAMESPACE_MODEST,
1306 if (obd->obd_namespace == NULL)
1307 GOTO(err_ops, rc = -ENOMEM);
1309 /* No recovery for MGCs */
1310 obd->obd_replayable = 0;
1312 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1313 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1314 OBD_FAIL_MGS_ALL_REPLY_NET);
1318 rc = mgs_fs_setup(env, mgs);
1320 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1325 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1326 obd, &llog_osd_ops);
1330 /* XXX: we need this trick till N:1 stack is supported
1331 * set "current" directory for named llogs */
1332 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1334 ctxt->loc_dir = mgs->mgs_configs_dir;
1335 llog_ctxt_put(ctxt);
1337 /* Internal mgs setup */
1338 mgs_init_fsdb_list(mgs);
1339 mutex_init(&mgs->mgs_mutex);
1340 mgs->mgs_start_time = ktime_get_real_seconds();
1341 spin_lock_init(&mgs->mgs_lock);
1342 mutex_init(&mgs->mgs_health_mutex);
1343 init_rwsem(&mgs->mgs_barrier_rwsem);
1345 rc = mgs_lcfg_rename(env, mgs);
1349 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1351 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1356 /* Setup params fsdb and log, so that other servers can make a local
1357 * copy successfully when they are mounted. See LU-4783 */
1358 rc = mgs_params_fsdb_setup(env, mgs);
1360 /* params fsdb and log can be setup later */
1361 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1362 obd->obd_name, PARAMS_FILENAME, rc);
1364 /* Setup _mgs fsdb, useful for srpc */
1365 mgs__mgs_fsdb_setup(env, mgs);
1367 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1368 "mgs_ldlm_client", &obd->obd_ldlm_client);
1370 conf = (typeof(conf)) {
1371 .psc_name = LUSTRE_MGS_NAME,
1372 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1374 .bc_nbufs = MGS_NBUFS,
1375 .bc_buf_size = MGS_BUFSIZE,
1376 .bc_req_max_size = MGS_MAXREQSIZE,
1377 .bc_rep_max_size = MGS_MAXREPSIZE,
1378 .bc_req_portal = MGS_REQUEST_PORTAL,
1379 .bc_rep_portal = MGC_REPLY_PORTAL,
1382 .tc_thr_name = "ll_mgs",
1383 .tc_nthrs_init = MGS_NTHRS_INIT,
1384 .tc_nthrs_max = MGS_NTHRS_MAX,
1385 .tc_ctx_tags = LCT_MG_THREAD,
1388 .so_req_handler = tgt_request_handle,
1389 .so_req_printer = target_print_req,
1393 /* Start the service threads */
1394 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1395 obd->obd_debugfs_entry);
1396 if (IS_ERR(mgs->mgs_service)) {
1397 rc = PTR_ERR(mgs->mgs_service);
1398 CERROR("failed to start mgs service: %d\n", rc);
1399 mgs->mgs_service = NULL;
1400 GOTO(err_lproc, rc);
1403 ping_evictor_start();
1405 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1407 /* device stack is not yet fully setup to keep no objects behind */
1408 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1411 mgs_params_fsdb_cleanup(env, mgs);
1412 lproc_mgs_cleanup(mgs);
1414 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1416 ctxt->loc_dir = NULL;
1417 llog_cleanup(env, ctxt);
1420 tgt_fini(env, &mgs->mgs_lut);
1422 /* No extra cleanup needed for llog_init_commit_thread() */
1423 mgs_fs_cleanup(env, mgs);
1425 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1426 obd->obd_namespace = NULL;
1428 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1429 if (!cfs_hash_is_empty(mgs2lu_dev(mgs)->ld_site->ls_obj_hash)) {
1430 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1431 lu_site_print(env, mgs2lu_dev(mgs)->ld_site, &msgdata,
1434 obd_disconnect(mgs->mgs_bottom_exp);
1437 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1441 static struct lu_device *mgs_device_free(const struct lu_env *env,
1442 struct lu_device *lu)
1444 struct mgs_device *mgs = lu2mgs_dev(lu);
1447 dt_device_fini(&mgs->mgs_dt_dev);
1452 static int mgs_process_config(const struct lu_env *env,
1453 struct lu_device *dev,
1454 struct lustre_cfg *lcfg)
1460 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1461 const struct lu_object_conf *unused)
1463 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1464 struct lu_device *under;
1465 struct lu_object *below;
1469 /* do no set .do_ops as mgs calls to bottom osd directly */
1471 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1472 PFID(lu_object_fid(o)));
1474 under = &d->mgs_bottom->dd_lu_dev;
1475 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1477 lu_object_add(o, below);
1484 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1486 struct mgs_object *obj = lu2mgs_obj(o);
1487 struct lu_object_header *h = o->lo_header;
1489 dt_object_fini(&obj->mgo_obj);
1490 lu_object_header_fini(h);
1494 static int mgs_object_print(const struct lu_env *env, void *cookie,
1495 lu_printer_t p, const struct lu_object *l)
1497 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1499 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1502 static struct lu_object_operations mgs_lu_obj_ops = {
1503 .loo_object_init = mgs_object_init,
1504 .loo_object_free = mgs_object_free,
1505 .loo_object_print = mgs_object_print,
1508 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1509 const struct lu_object_header *hdr,
1510 struct lu_device *d)
1512 struct lu_object_header *h;
1513 struct mgs_object *o;
1514 struct lu_object *l;
1516 LASSERT(hdr == NULL);
1520 l = &o->mgo_obj.do_lu;
1523 lu_object_header_init(h);
1524 dt_object_init(&o->mgo_obj, h, d);
1525 lu_object_add_top(h, l);
1527 l->lo_ops = &mgs_lu_obj_ops;
1535 const struct lu_device_operations mgs_lu_ops = {
1536 .ldo_object_alloc = mgs_object_alloc,
1537 .ldo_process_config = mgs_process_config,
1540 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1541 struct lu_device_type *type,
1542 struct lustre_cfg *lcfg)
1544 struct mgs_device *mgs;
1545 struct lu_device *ludev;
1549 ludev = ERR_PTR(-ENOMEM);
1553 ludev = mgs2lu_dev(mgs);
1554 dt_device_init(&mgs->mgs_dt_dev, type);
1555 rc = mgs_init0(env, mgs, type, lcfg);
1557 mgs_device_free(env, ludev);
1558 ludev = ERR_PTR(rc);
1564 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1565 struct lu_device *d)
1567 struct mgs_device *mgs = lu2mgs_dev(d);
1568 struct obd_device *obd = mgs->mgs_obd;
1569 struct llog_ctxt *ctxt;
1573 LASSERT(mgs->mgs_bottom);
1575 class_disconnect_exports(obd);
1577 ping_evictor_stop();
1579 mutex_lock(&mgs->mgs_health_mutex);
1580 ptlrpc_unregister_service(mgs->mgs_service);
1581 mutex_unlock(&mgs->mgs_health_mutex);
1583 mgs_params_fsdb_cleanup(env, mgs);
1584 mgs_cleanup_fsdb_list(mgs);
1586 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1587 obd_exports_barrier(obd);
1588 obd_zombie_barrier();
1590 tgt_fini(env, &mgs->mgs_lut);
1591 lproc_mgs_cleanup(mgs);
1593 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1595 ctxt->loc_dir = NULL;
1596 llog_cleanup(env, ctxt);
1599 mgs_fs_cleanup(env, mgs);
1601 ldlm_namespace_free_post(obd->obd_namespace);
1602 obd->obd_namespace = NULL;
1604 lu_site_purge(env, d->ld_site, ~0);
1605 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
1606 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
1607 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
1610 LASSERT(mgs->mgs_bottom_exp);
1611 obd_disconnect(mgs->mgs_bottom_exp);
1613 server_put_mount(obd->obd_name, true);
1618 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1619 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1621 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1623 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1625 static struct lu_device_type_operations mgs_device_type_ops = {
1626 .ldto_init = mgs_type_init,
1627 .ldto_fini = mgs_type_fini,
1629 .ldto_start = mgs_type_start,
1630 .ldto_stop = mgs_type_stop,
1632 .ldto_device_alloc = mgs_device_alloc,
1633 .ldto_device_free = mgs_device_free,
1635 .ldto_device_fini = mgs_device_fini
1638 static struct lu_device_type mgs_device_type = {
1639 .ldt_tags = LU_DEVICE_DT,
1640 .ldt_name = LUSTRE_MGS_NAME,
1641 .ldt_ops = &mgs_device_type_ops,
1642 .ldt_ctx_tags = LCT_MG_THREAD
1645 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1646 struct obd_device *obd, struct obd_uuid *cluuid,
1647 struct obd_connect_data *data, void *localdata)
1651 if (exp == NULL || obd == NULL || cluuid == NULL)
1654 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1657 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1659 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1660 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1662 exp->exp_connect_data = *data;
1663 data->ocd_version = LUSTRE_VERSION_CODE;
1666 RETURN(mgs_export_stats_init(obd, exp, localdata));
1669 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1670 struct obd_device *obd, struct obd_uuid *cluuid,
1671 struct obd_connect_data *data, void *localdata)
1673 struct obd_export *lexp;
1674 struct lustre_handle conn = {
1681 if (exp == NULL || obd == NULL || cluuid == NULL)
1684 rc = class_connect(&conn, obd, cluuid);
1688 lexp = class_conn2export(&conn);
1692 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1694 GOTO(out_disconnect, rc);
1701 class_disconnect(lexp);
1706 static int mgs_obd_disconnect(struct obd_export *exp)
1714 mgs_fsc_cleanup(exp);
1716 class_export_get(exp);
1717 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1719 rc = server_disconnect_export(exp);
1720 class_export_put(exp);
1724 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1726 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1729 mutex_lock(&mgs->mgs_health_mutex);
1730 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1731 mutex_unlock(&mgs->mgs_health_mutex);
1733 return rc != 0 ? 1 : 0;
1736 /* use obd ops to offer management infrastructure */
1737 static struct obd_ops mgs_obd_device_ops = {
1738 .o_owner = THIS_MODULE,
1739 .o_connect = mgs_obd_connect,
1740 .o_reconnect = mgs_obd_reconnect,
1741 .o_disconnect = mgs_obd_disconnect,
1742 .o_init_export = mgs_init_export,
1743 .o_destroy_export = mgs_destroy_export,
1744 .o_iocontrol = mgs_iocontrol,
1745 .o_health_check = mgs_health_check,
1748 static int __init mgs_init(void)
1750 return class_register_type(&mgs_obd_device_ops, NULL, true, NULL,
1751 LUSTRE_MGS_NAME, &mgs_device_type);
1754 static void __exit mgs_exit(void)
1756 class_unregister_type(LUSTRE_MGS_NAME);
1759 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1760 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1761 MODULE_VERSION(LUSTRE_VERSION_STRING);
1762 MODULE_LICENSE("GPL");
1764 module_init(mgs_init);
1765 module_exit(mgs_exit);