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/
31 * lustre/mgs/mgs_handler.c
33 * Author: Nathan Rutman <nathan@clusterfs.com>
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mikhail Pershin <tappro@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_MGS
39 #define D_MGS D_CONFIG
41 #include <obd_class.h>
42 #include <lprocfs_status.h>
43 #include <uapi/linux/lustre/lustre_ioctl.h>
44 #include <uapi/linux/lustre/lustre_param.h>
45 #include <uapi/linux/lustre/lustre_barrier_user.h>
47 #include "mgs_internal.h"
50 * Regular MGS handlers
52 static int mgs_connect(struct tgt_session_info *tsi)
54 struct ptlrpc_request *req = tgt_ses_req(tsi);
59 CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_CONNECT_NET, cfs_fail_val);
60 rc = tgt_connect(tsi);
64 if (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1)
65 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT);
70 static int mgs_disconnect(struct tgt_session_info *tsi)
76 LASSERT(tsi->tsi_exp);
78 rc = tgt_disconnect(tsi);
80 RETURN(err_serious(rc));
84 static int mgs_exception(struct tgt_session_info *tsi)
88 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_EXCEPTION);
93 static inline bool str_starts_with(const char *str, const char *prefix)
95 return strncmp(str, prefix, strlen(prefix)) == 0;
98 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 18, 53, 0)
99 static int mgs_set_info(struct tgt_session_info *tsi)
101 struct mgs_thread_info *mgi;
102 struct mgs_send_param *msp;
108 mgi = mgs_env_info(tsi->tsi_env);
110 RETURN(err_serious(PTR_ERR(mgi)));
112 msp = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_SEND_PARAM);
114 RETURN(err_serious(-EFAULT));
116 param_len = strnlen(msp->mgs_param, sizeof(msp->mgs_param));
117 if (param_len == 0 || param_len == sizeof(msp->mgs_param))
120 /* We only allow '*.lov.stripe{size,count,offset}=*' from an RPC. */
121 s = strchr(msp->mgs_param, '.');
125 if (!str_starts_with(s + 1, "lov.stripesize=") &&
126 !str_starts_with(s + 1, "lov.stripecount=") &&
127 !str_starts_with(s + 1, "lov.stripeoffset="))
131 CDEBUG(D_MGS, "%s: ignoring set info '%s'\n",
132 tgt_name(tsi->tsi_tgt), msp->mgs_param);
144 static int mgs_completion_ast_generic(struct ldlm_lock *lock, __u64 flags,
145 void *cbdata, enum ast_type type)
149 if (!(flags & LDLM_FL_BLOCKED_MASK)) {
152 /* l_ast_data is used as a marker to avoid cancel ldlm lock
153 * twice. See LU-2317. */
154 lock_res_and_lock(lock);
155 fsdb = (struct fs_db *)lock->l_ast_data;
156 lock->l_ast_data = NULL;
157 unlock_res_and_lock(lock);
160 struct lustre_handle lockh;
164 /* clear the bit before lock put */
165 clear_bit(FSDB_REVOKING_LOCK,
169 clear_bit(FSDB_REVOKING_PARAMS,
173 mgs_ir_notify_complete(fsdb);
181 ldlm_lock2handle(lock, &lockh);
182 ldlm_lock_decref_and_cancel(&lockh, LCK_EX);
186 RETURN(ldlm_completion_ast(lock, flags, cbdata));
189 static int mgs_completion_ast_config(struct ldlm_lock *lock, __u64 flags,
192 return mgs_completion_ast_generic(lock, flags, cbdata, AST_CONFIG);
195 static int mgs_completion_ast_params(struct ldlm_lock *lock, __u64 flags,
198 return mgs_completion_ast_generic(lock, flags, cbdata, AST_PARAMS);
201 static int mgs_completion_ast_ir(struct ldlm_lock *lock, __u64 flags,
204 return mgs_completion_ast_generic(lock, flags, cbdata, AST_IR);
207 static int mgs_completion_ast_barrier(struct ldlm_lock *lock, __u64 flags,
210 return mgs_completion_ast_generic(lock, flags, cbdata, AST_BARRIER);
213 void mgs_revoke_lock(struct mgs_device *mgs, struct fs_db *fsdb,
214 enum mgs_cfg_type type)
216 ldlm_completion_callback cp = NULL;
217 struct lustre_handle lockh = {
220 struct ldlm_res_id res_id;
221 __u64 flags = LDLM_FL_ATOMIC_CB;
225 LASSERT(fsdb->fsdb_name[0] != '\0');
226 rc = mgc_fsname2resid(fsdb->fsdb_name, &res_id, type);
229 case MGS_CFG_T_CONFIG:
230 case MGS_CFG_T_NODEMAP:
231 cp = mgs_completion_ast_config;
232 if (test_and_set_bit(FSDB_REVOKING_LOCK, &fsdb->fsdb_flags))
235 case MGS_CFG_T_PARAMS:
236 cp = mgs_completion_ast_params;
237 if (test_and_set_bit(FSDB_REVOKING_PARAMS, &fsdb->fsdb_flags))
240 case MGS_CFG_T_RECOVER:
241 cp = mgs_completion_ast_ir;
243 case MGS_CFG_T_BARRIER:
244 cp = mgs_completion_ast_barrier;
252 rc = ldlm_cli_enqueue_local(NULL, mgs->mgs_obd->obd_namespace,
253 &res_id, LDLM_PLAIN, NULL, LCK_EX,
254 &flags, ldlm_blocking_ast, cp,
255 NULL, fsdb, 0, LVB_T_NONE, NULL,
257 if (rc != ELDLM_OK) {
258 CERROR("%s: can't take cfg lock for %#llx/%#llx : rc = %d\n",
259 mgs->mgs_obd->obd_name,
260 le64_to_cpu(res_id.name[0]),
261 le64_to_cpu(res_id.name[1]), rc);
263 if (type == MGS_CFG_T_CONFIG)
264 clear_bit(FSDB_REVOKING_LOCK,
267 if (type == MGS_CFG_T_PARAMS)
268 clear_bit(FSDB_REVOKING_PARAMS,
271 /* lock has been cancelled in completion_ast. */
280 static int mgs_check_target(const struct lu_env *env,
281 struct mgs_device *mgs,
282 struct mgs_target_info *mti)
287 rc = mgs_check_index(env, mgs, mti);
289 LCONSOLE_ERROR_MSG(0x13b, "%s claims to have registered, but "
290 "this MGS does not know about it, preventing "
291 "registration.\n", mti->mti_svname);
293 } else if (rc == -1) {
294 LCONSOLE_ERROR_MSG(0x13c, "Client log %s-client has "
295 "disappeared! Regenerating all logs.\n",
297 mti->mti_flags |= LDD_F_WRITECONF;
300 /* Index is correctly marked as used */
307 /* Ensure this is not a failover node that is connecting first*/
308 static int mgs_check_failover_reg(struct mgs_target_info *mti)
314 ptr = mti->mti_params;
315 while (class_find_param(ptr, PARAM_FAILNODE, &ptr) == 0) {
316 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
317 for (i = 0; i < mti->mti_nid_count; i++) {
318 if (nid == mti->mti_nids[i]) {
319 LCONSOLE_WARN("Denying initial registration attempt from nid %s, specified as failover\n",
320 libcfs_nid2str(nid));
321 return -EADDRNOTAVAIL;
329 /* Called whenever a target starts up. Flags indicate first connect, etc. */
330 static int mgs_target_reg(struct tgt_session_info *tsi)
332 struct obd_device *obd = tsi->tsi_exp->exp_obd;
333 struct mgs_device *mgs = exp2mgs_dev(tsi->tsi_exp);
334 struct mgs_target_info *mti, *rep_mti;
335 struct fs_db *b_fsdb = NULL; /* barrier fsdb */
336 struct fs_db *c_fsdb = NULL; /* config fsdb */
337 char barrier_name[20];
343 rc = lu_env_refill((struct lu_env *)tsi->tsi_env);
345 return err_serious(rc);
347 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_REG);
349 mti = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
351 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no mgs_send_param");
352 RETURN(err_serious(-EFAULT));
355 down_read(&mgs->mgs_barrier_rwsem);
357 if (OCD_HAS_FLAG(&tgt_ses_req(tsi)->rq_export->exp_connect_data,
359 opc = mti->mti_flags & LDD_F_OPC_MASK;
363 if (opc == LDD_F_OPC_READY) {
364 CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
365 mti->mti_fsname, mti->mti_stripe_index);
366 rc = mgs_ir_update(tsi->tsi_env, mgs, mti);
368 LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
369 CERROR("%s: Update IR return failure: rc = %d\n",
370 mti->mti_fsname, rc);
373 GOTO(out_norevoke, rc);
376 /* Do not support unregistering right now. */
377 if (opc != LDD_F_OPC_REG)
378 GOTO(out_norevoke, rc = -EINVAL);
380 snprintf(barrier_name, sizeof(barrier_name) - 1, "%s-%s",
381 mti->mti_fsname, BARRIER_FILENAME);
382 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, barrier_name, &b_fsdb);
384 CERROR("%s: Can't get db for %s: rc = %d\n",
385 mti->mti_fsname, barrier_name, rc);
387 GOTO(out_norevoke, rc);
390 CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
391 mti->mti_fsname, mti->mti_stripe_index);
393 if (mti->mti_flags & LDD_F_SV_TYPE_MDT) {
394 if (b_fsdb->fsdb_barrier_status == BS_FREEZING_P1 ||
395 b_fsdb->fsdb_barrier_status == BS_FREEZING_P2 ||
396 b_fsdb->fsdb_barrier_status == BS_FROZEN) {
397 LCONSOLE_WARN("%s: the system is in barrier, refuse "
398 "the connection from MDT %s temporary\n",
399 obd->obd_name, mti->mti_svname);
401 GOTO(out_norevoke, rc = -EBUSY);
404 if (!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_BARRIER) &&
405 !b_fsdb->fsdb_barrier_disabled) {
406 LCONSOLE_WARN("%s: the MDT %s does not support write "
407 "barrier, so disable barrier on the "
409 obd->obd_name, mti->mti_svname);
411 b_fsdb->fsdb_barrier_disabled = 1;
415 if (mti->mti_flags & LDD_F_NEED_INDEX)
416 mti->mti_flags |= LDD_F_WRITECONF;
418 if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPDATE))) {
419 /* We're just here as a startup ping. */
420 CDEBUG(D_MGS, "Server %s is running on %s\n",
421 mti->mti_svname, obd_export_nid2str(tsi->tsi_exp));
422 rc = mgs_check_target(tsi->tsi_env, mgs, mti);
423 /* above will set appropriate mti flags */
425 /* Nothing wrong, or fatal error */
426 GOTO(out_norevoke, rc);
427 } else if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)) {
428 rc = mgs_check_failover_reg(mti);
430 GOTO(out_norevoke, rc);
433 OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
435 if (mti->mti_flags & LDD_F_WRITECONF) {
436 if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
437 mti->mti_stripe_index == 0) {
438 mgs_put_fsdb(mgs, b_fsdb);
440 rc = mgs_erase_logs(tsi->tsi_env, mgs,
442 LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
443 "request. All servers must be restarted "
444 "in order to regenerate the logs: rc = %d"
445 "\n", obd->obd_name, mti->mti_fsname, rc);
446 if (rc && rc != -ENOENT)
447 GOTO(out_norevoke, rc);
449 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
450 barrier_name, &b_fsdb);
452 CERROR("Can't get db for %s: %d\n",
455 GOTO(out_norevoke, rc);
458 if (!(exp_connect_flags(tsi->tsi_exp) &
459 OBD_CONNECT_BARRIER)) {
460 LCONSOLE_WARN("%s: the MDT %s does not support "
461 "write barrier, disable barrier "
462 "on the whole system.\n",
463 obd->obd_name, mti->mti_svname);
465 b_fsdb->fsdb_barrier_disabled = 1;
467 } else if (mti->mti_flags &
468 (LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
469 rc = mgs_erase_log(tsi->tsi_env, mgs, mti->mti_svname);
470 LCONSOLE_WARN("%s: Regenerating %s log by user "
471 "request: rc = %d\n",
472 obd->obd_name, mti->mti_svname, rc);
474 GOTO(out_norevoke, rc);
477 mti->mti_flags |= LDD_F_UPDATE;
480 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs, mti->mti_fsname, &c_fsdb);
482 CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
484 GOTO(out_norevoke, rc);
488 * Log writing contention is handled by the fsdb_mutex.
490 * It should be alright if someone was reading while we were
491 * updating the logs - if we revoke at the end they will just update
492 * from where they left off.
494 if (mti->mti_flags & LDD_F_UPDATE) {
495 CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
496 mti->mti_stripe_index);
498 /* create or update the target log
499 and update the client/mdt logs */
500 rc = mgs_write_log_target(tsi->tsi_env, mgs, mti, c_fsdb);
502 CERROR("Failed to write %s log (%d)\n",
503 mti->mti_svname, rc);
507 mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
508 LDD_F_NEED_INDEX | LDD_F_WRITECONF);
509 mti->mti_flags |= LDD_F_REWRITE_LDD;
513 mgs_revoke_lock(mgs, c_fsdb, MGS_CFG_T_CONFIG);
516 if (!rc && mti->mti_flags & LDD_F_SV_TYPE_MDT && b_fsdb) {
518 rc = mgs_find_or_make_fsdb(tsi->tsi_env, mgs,
519 mti->mti_fsname, &c_fsdb);
521 CERROR("Fail to get db for %s: %d\n",
522 mti->mti_fsname, rc);
526 memcpy(b_fsdb->fsdb_mdt_index_map,
527 c_fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE);
528 b_fsdb->fsdb_mdt_count = c_fsdb->fsdb_mdt_count;
532 up_read(&mgs->mgs_barrier_rwsem);
534 CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
535 mti->mti_stripe_index, rc);
536 /* An error flag is set in the mti reply rather than an error code */
538 mti->mti_flags |= LDD_F_ERROR;
540 /* send back the whole mti in the reply */
541 rep_mti = req_capsule_server_get(tsi->tsi_pill, &RMF_MGS_TARGET_INFO);
544 /* Flush logs to disk */
545 dt_sync(tsi->tsi_env, mgs->mgs_bottom);
547 mgs_put_fsdb(mgs, b_fsdb);
549 mgs_put_fsdb(mgs, c_fsdb);
553 /* Called whenever a target cleans up. */
554 static int mgs_target_del(struct tgt_session_info *tsi)
558 tgt_counter_incr(tsi->tsi_exp, LPROC_MGS_TARGET_DEL);
563 static int mgs_config_read(struct tgt_session_info *tsi)
565 struct ptlrpc_request *req = tgt_ses_req(tsi);
566 struct mgs_config_body *body;
571 body = req_capsule_client_get(tsi->tsi_pill, &RMF_MGS_CONFIG_BODY);
573 DEBUG_REQ(D_HA, req, "no mgs_config_body");
574 RETURN(err_serious(-EFAULT));
577 switch (body->mcb_type) {
578 case MGS_CFG_T_RECOVER:
579 rc = mgs_get_ir_logs(req);
581 case MGS_CFG_T_NODEMAP:
582 rc = nodemap_get_config_req(req->rq_export->exp_obd, req);
584 case MGS_CFG_T_CONFIG:
595 static int mgs_llog_open(struct tgt_session_info *tsi)
597 struct mgs_thread_info *mgi;
598 struct ptlrpc_request *req = tgt_ses_req(tsi);
604 rc = tgt_llog_open(tsi);
609 * For old clients there is no direct way of knowing which file system
610 * a client is operating at the MGS side. But we need to pick up those
611 * clients so that the MGS can mark the corresponding file system as
612 * non-IR capable because old clients are not ready to be notified.
614 * Therefore we attempt to detect the file systems name by hacking the
615 * llog operation which is currently used by the clients to fetch
616 * configuration logs. At present this is fine because this is the
617 * ONLY llog operation between mgc and the MGS.
619 * If extra llog operation are going to be added, this function needs
622 * When releases prior than 2.0 are not supported, the following code
625 mgi = mgs_env_info(tsi->tsi_env);
627 RETURN(PTR_ERR(mgi));
629 logname = req_capsule_client_get(tsi->tsi_pill, &RMF_NAME);
631 char *ptr = strrchr(logname, '-');
632 int len = (ptr != NULL) ? (int)(ptr - logname) : 0;
634 if (ptr == NULL || len >= sizeof(mgi->mgi_fsname)) {
635 if (strcmp(logname, PARAMS_FILENAME) != 0)
636 LCONSOLE_WARN("%s: non-config logname "
638 tgt_name(tsi->tsi_tgt),
640 /* not error, this can be llog test name */
642 strncpy(mgi->mgi_fsname, logname, len);
643 mgi->mgi_fsname[len] = 0;
645 rc = mgs_fsc_attach(tsi->tsi_env, tsi->tsi_exp,
647 if (rc && rc != -EEXIST) {
648 LCONSOLE_WARN("%s: Unable to add client %s "
649 "to file system %s: %d\n",
650 tgt_name(tsi->tsi_tgt),
651 libcfs_nidstr(&req->rq_peer.nid),
652 mgi->mgi_fsname, rc);
658 CERROR("%s: no logname in request\n", tgt_name(tsi->tsi_tgt));
664 static inline int mgs_init_export(struct obd_export *exp)
666 struct mgs_export_data *data = &exp->u.eu_mgs_data;
668 /* init mgs_export_data for fsc */
669 spin_lock_init(&data->med_lock);
670 INIT_LIST_HEAD(&data->med_clients);
672 spin_lock(&exp->exp_lock);
673 exp->exp_connecting = 1;
674 spin_unlock(&exp->exp_lock);
676 /* self-export doesn't need client data and ldlm initialization */
677 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
678 &exp->exp_client_uuid)))
680 return ldlm_init_export(exp);
683 static inline int mgs_destroy_export(struct obd_export *exp)
687 target_destroy_export(exp);
688 mgs_client_free(exp);
690 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
691 &exp->exp_client_uuid)))
694 ldlm_destroy_export(exp);
699 static int mgs_extract_fs_pool(char *arg, char *fsname, char *poolname)
706 for (ptr = arg; *ptr != '\0'; ptr++) {
707 if (!isalnum(*ptr) && *ptr != '_' && *ptr != '-' && *ptr != '.')
711 /* Test for fsname.poolname format.
712 * strlen() test if poolname is empty
714 ptr = strchr(arg, '.');
715 if (!ptr || !strlen(ptr))
719 /* Check pool name validity. */
720 if (ptr[0] == '\0' || lov_pool_is_reserved(ptr))
722 /* Also make sure poolname is not to long. */
723 if (strlen(ptr) > LOV_MAXPOOLNAME)
724 return -ENAMETOOLONG;
725 strlcpy(poolname, ptr, LOV_MAXPOOLNAME + 1);
727 /* Test if fsname is empty */
728 len = strlen(arg) - strlen(ptr) - 1;
733 if (len > LUSTRE_MAXFSNAME)
734 return -ENAMETOOLONG;
736 strncpy(fsname, arg, len);
741 static int mgs_iocontrol_nodemap(const struct lu_env *env,
742 struct mgs_device *mgs,
743 struct obd_ioctl_data *data)
745 struct lustre_cfg *lcfg = NULL;
748 const char *nodemap_name = NULL;
749 const char *nidstr = NULL;
750 const char *client_idstr = NULL;
751 const char *idtype_str = NULL;
754 char name_buf[LUSTRE_NODEMAP_NAME_LENGTH + 1];
756 unsigned long client_id;
763 if (data->ioc_type != LUSTRE_CFG_TYPE) {
764 CERROR("%s: unknown cfg record type: %d\n",
765 mgs->mgs_obd->obd_name, data->ioc_type);
766 GOTO(out, rc = -EINVAL);
769 if (data->ioc_plen1 > PAGE_SIZE)
770 GOTO(out, rc = -E2BIG);
772 OBD_ALLOC(lcfg, data->ioc_plen1);
774 GOTO(out, rc = -ENOMEM);
776 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
777 GOTO(out_lcfg, rc = -EFAULT);
779 cmd = lcfg->lcfg_command;
782 case LCFG_NODEMAP_ACTIVATE:
783 if (lcfg->lcfg_bufcount != 2)
784 GOTO(out_lcfg, rc = -EINVAL);
785 param = lustre_cfg_string(lcfg, 1);
786 if (strcmp(param, "1") == 0)
791 case LCFG_NODEMAP_ADD:
792 case LCFG_NODEMAP_DEL:
793 if (lcfg->lcfg_bufcount != 2)
794 GOTO(out_lcfg, rc = -EINVAL);
795 nodemap_name = lustre_cfg_string(lcfg, 1);
796 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
798 case LCFG_NODEMAP_TEST_NID:
799 if (lcfg->lcfg_bufcount != 2)
800 GOTO(out_lcfg, rc = -EINVAL);
801 nidstr = lustre_cfg_string(lcfg, 1);
802 nid = libcfs_str2nid(nidstr);
803 nodemap_test_nid(nid, name_buf, sizeof(name_buf));
804 rc = copy_to_user(data->ioc_pbuf1, name_buf,
805 min_t(size_t, data->ioc_plen1,
808 GOTO(out_lcfg, rc = -EFAULT);
810 case LCFG_NODEMAP_TEST_ID:
811 if (lcfg->lcfg_bufcount != 4)
812 GOTO(out_lcfg, rc = -EINVAL);
813 nidstr = lustre_cfg_string(lcfg, 1);
814 idtype_str = lustre_cfg_string(lcfg, 2);
815 client_idstr = lustre_cfg_string(lcfg, 3);
817 nid = libcfs_str2nid(nidstr);
818 if (strcmp(idtype_str, "uid") == 0)
819 idtype = NODEMAP_UID;
820 else if (strcmp(idtype_str, "gid") == 0)
821 idtype = NODEMAP_GID;
822 else if (strcmp(idtype_str, "projid") == 0)
823 idtype = NODEMAP_PROJID;
825 GOTO(out_lcfg, rc = -EINVAL);
827 rc = kstrtoul(client_idstr, 10, &client_id);
829 GOTO(out_lcfg, rc = -EINVAL);
831 rc = nodemap_test_id(nid, idtype, client_id, &fs_id);
833 GOTO(out_lcfg, rc = -EINVAL);
835 if (data->ioc_plen1 < sizeof(fs_idstr))
836 GOTO(out_lcfg, rc = -EINVAL);
838 snprintf(fs_idstr, sizeof(fs_idstr), "%u", fs_id);
839 if (copy_to_user(data->ioc_pbuf1, fs_idstr,
840 sizeof(fs_idstr)) != 0)
841 GOTO(out_lcfg, rc = -EINVAL);
843 case LCFG_NODEMAP_ADD_RANGE:
844 case LCFG_NODEMAP_DEL_RANGE:
845 case LCFG_NODEMAP_ADD_UIDMAP:
846 case LCFG_NODEMAP_DEL_UIDMAP:
847 case LCFG_NODEMAP_ADD_GIDMAP:
848 case LCFG_NODEMAP_DEL_GIDMAP:
849 case LCFG_NODEMAP_ADD_PROJIDMAP:
850 case LCFG_NODEMAP_DEL_PROJIDMAP:
851 case LCFG_NODEMAP_SET_FILESET:
852 case LCFG_NODEMAP_SET_SEPOL:
853 if (lcfg->lcfg_bufcount != 3)
854 GOTO(out_lcfg, rc = -EINVAL);
855 nodemap_name = lustre_cfg_string(lcfg, 1);
856 param = lustre_cfg_string(lcfg, 2);
857 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
859 case LCFG_NODEMAP_ADMIN:
860 case LCFG_NODEMAP_TRUSTED:
861 case LCFG_NODEMAP_DENY_UNKNOWN:
862 case LCFG_NODEMAP_SQUASH_UID:
863 case LCFG_NODEMAP_SQUASH_GID:
864 case LCFG_NODEMAP_SQUASH_PROJID:
865 case LCFG_NODEMAP_MAP_MODE:
866 case LCFG_NODEMAP_AUDIT_MODE:
867 case LCFG_NODEMAP_FORBID_ENCRYPT:
868 case LCFG_NODEMAP_READONLY_MOUNT:
869 if (lcfg->lcfg_bufcount != 4)
870 GOTO(out_lcfg, rc = -EINVAL);
871 nodemap_name = lustre_cfg_string(lcfg, 1);
872 param = lustre_cfg_string(lcfg, 3);
873 rc = mgs_nodemap_cmd(env, mgs, cmd, nodemap_name, param);
880 CDEBUG_LIMIT(rc == -EEXIST ? D_INFO : D_ERROR,
881 "%s: OBD_IOC_NODEMAP command %X for %s: rc = %d\n",
882 mgs->mgs_obd->obd_name, lcfg->lcfg_command,
887 /* revoke nodemap lock */
888 rc = mgs_find_or_make_fsdb(env, mgs, LUSTRE_NODEMAP_NAME, &fsdb);
890 CWARN("%s: cannot make nodemap fsdb: rc = %d\n",
891 mgs->mgs_obd->obd_name, rc);
893 mgs_revoke_lock(mgs, fsdb, MGS_CFG_T_NODEMAP);
894 mgs_put_fsdb(mgs, fsdb);
898 OBD_FREE(lcfg, data->ioc_plen1);
903 static int mgs_iocontrol_pool(const struct lu_env *env,
904 struct mgs_device *mgs,
905 struct obd_ioctl_data *data)
907 struct mgs_thread_info *mgi = mgs_env_info(env);
909 struct lustre_cfg *lcfg = NULL;
910 char *poolname = NULL;
913 OBD_ALLOC(poolname, LOV_MAXPOOLNAME + 1);
914 if (poolname == NULL)
917 if (data->ioc_type != LUSTRE_CFG_TYPE) {
918 CERROR("%s: unknown cfg record type: %d\n",
919 mgs->mgs_obd->obd_name, data->ioc_type);
920 GOTO(out_pool, rc = -EINVAL);
923 if (data->ioc_plen1 > PAGE_SIZE)
924 GOTO(out_pool, rc = -E2BIG);
926 OBD_ALLOC(lcfg, data->ioc_plen1);
928 GOTO(out_pool, rc = -ENOMEM);
930 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
931 GOTO(out_lcfg, rc = -EFAULT);
933 if (lcfg->lcfg_bufcount < 2)
934 GOTO(out_lcfg, rc = -EFAULT);
936 /* first arg is always <fsname>.<poolname> */
937 rc = mgs_extract_fs_pool(lustre_cfg_string(lcfg, 1), mgi->mgi_fsname,
942 switch (lcfg->lcfg_command) {
944 if (lcfg->lcfg_bufcount != 2)
945 GOTO(out_lcfg, rc = -EINVAL);
946 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_NEW, mgi->mgi_fsname,
950 if (lcfg->lcfg_bufcount != 3)
951 GOTO(out_lcfg, rc = -EINVAL);
952 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_ADD, mgi->mgi_fsname,
953 poolname, lustre_cfg_string(lcfg, 2));
956 if (lcfg->lcfg_bufcount != 3)
957 GOTO(out_lcfg, rc = -EINVAL);
958 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_REM, mgi->mgi_fsname,
959 poolname, lustre_cfg_string(lcfg, 2));
962 if (lcfg->lcfg_bufcount != 2)
963 GOTO(out_lcfg, rc = -EINVAL);
964 rc = mgs_pool_cmd(env, mgs, LCFG_POOL_DEL, mgi->mgi_fsname,
972 CERROR("OBD_IOC_POOL err %d, cmd %X for pool %s.%s\n",
973 rc, lcfg->lcfg_command, mgi->mgi_fsname, poolname);
978 OBD_FREE(lcfg, data->ioc_plen1);
980 OBD_FREE(poolname, LOV_MAXPOOLNAME + 1);
984 /* from mdt_iocontrol */
985 static int mgs_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
986 void *karg, void __user *uarg)
988 struct mgs_device *mgs = exp2mgs_dev(exp);
989 struct obd_ioctl_data *data = karg;
994 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
996 rc = lu_env_init(&env, LCT_MG_THREAD);
1002 case OBD_IOC_PARAM: {
1003 struct lustre_cfg *lcfg;
1005 if (data->ioc_type != LUSTRE_CFG_TYPE) {
1006 CERROR("%s: unknown cfg record type: %d\n",
1007 mgs->mgs_obd->obd_name, data->ioc_type);
1008 GOTO(out, rc = -EINVAL);
1011 OBD_ALLOC(lcfg, data->ioc_plen1);
1013 GOTO(out, rc = -ENOMEM);
1014 if (copy_from_user(lcfg, data->ioc_pbuf1, data->ioc_plen1))
1015 GOTO(out_free, rc = -EFAULT);
1017 if (lcfg->lcfg_bufcount < 1)
1018 GOTO(out_free, rc = -EINVAL);
1020 rc = mgs_set_param(&env, mgs, lcfg);
1022 CERROR("%s: setparam err: rc = %d\n",
1023 exp->exp_obd->obd_name, rc);
1025 OBD_FREE(lcfg, data->ioc_plen1);
1029 case OBD_IOC_REPLACE_NIDS: {
1030 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1032 CERROR("%s: no device or fsname specified: rc = %d\n",
1033 exp->exp_obd->obd_name, rc);
1037 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1039 CERROR("%s: device or fsname is too long: rc = %d\n",
1040 exp->exp_obd->obd_name, rc);
1044 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1046 CERROR("%s: device or fsname is not NUL terminated: "
1047 "rc = %d\n", exp->exp_obd->obd_name, rc);
1051 if (!data->ioc_inllen2 || !data->ioc_inlbuf2) {
1053 CERROR("%s: no NIDs specified: rc = %d\n",
1054 exp->exp_obd->obd_name, rc);
1058 if (data->ioc_inlbuf2[data->ioc_inllen2 - 1] != 0) {
1060 CERROR("%s: NID list is not NUL terminated: "
1061 "rc = %d\n", exp->exp_obd->obd_name, rc);
1065 /* replace nids in llog */
1066 rc = mgs_replace_nids(&env, mgs, data->ioc_inlbuf1,
1069 CERROR("%s: error replacing nids: rc = %d\n",
1070 exp->exp_obd->obd_name, rc);
1075 case OBD_IOC_CLEAR_CONFIGS: {
1076 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
1078 CERROR("%s: no device or fsname specified: rc = %d\n",
1079 exp->exp_obd->obd_name, rc);
1083 if (data->ioc_inllen1 > MTI_NAME_MAXLEN) {
1085 CERROR("%s: device or fsname is too long: rc = %d\n",
1086 exp->exp_obd->obd_name, rc);
1090 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
1092 CERROR("%s: device or fsname is not NUL terminated: "
1093 "rc = %d\n", exp->exp_obd->obd_name, rc);
1097 /* remove records marked SKIP from config logs */
1098 rc = mgs_clear_configs(&env, mgs, data->ioc_inlbuf1);
1100 CERROR("%s: error clearing config log: rc = %d\n",
1101 exp->exp_obd->obd_name, rc);
1107 rc = mgs_iocontrol_pool(&env, mgs, data);
1110 case OBD_IOC_BARRIER:
1111 rc = mgs_iocontrol_barrier(&env, mgs, data);
1114 case OBD_IOC_NODEMAP:
1115 rc = mgs_iocontrol_nodemap(&env, mgs, data);
1118 case OBD_IOC_LCFG_FORK:
1119 rc = mgs_lcfg_fork(&env, mgs, data->ioc_inlbuf1,
1123 case OBD_IOC_LCFG_ERASE:
1124 rc = mgs_lcfg_erase(&env, mgs, data->ioc_inlbuf1);
1127 case OBD_IOC_CATLOGLIST:
1128 rc = mgs_list_logs(&env, mgs, data);
1130 case OBD_IOC_LLOG_CANCEL:
1131 case OBD_IOC_LLOG_REMOVE:
1132 case OBD_IOC_LLOG_CHECK:
1133 case OBD_IOC_LLOG_INFO:
1134 case OBD_IOC_LLOG_PRINT: {
1135 struct llog_ctxt *ctxt;
1137 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1138 rc = llog_ioctl(&env, ctxt, cmd, data);
1139 llog_ctxt_put(ctxt);
1144 CERROR("%s: unknown command %#x\n",
1145 mgs->mgs_obd->obd_name, cmd);
1154 static int mgs_connect_to_osd(struct mgs_device *m, const char *nextdev)
1156 struct obd_connect_data *data = NULL;
1157 struct obd_device *obd;
1161 OBD_ALLOC_PTR(data);
1165 obd = class_name2obd(nextdev);
1167 CERROR("can't locate next device: %s\n", nextdev);
1168 GOTO(out, rc = -ENOTCONN);
1171 data->ocd_version = LUSTRE_VERSION_CODE;
1173 rc = obd_connect(NULL, &m->mgs_bottom_exp, obd,
1174 &obd->obd_uuid, data, NULL);
1176 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1180 m->mgs_bottom = lu2dt_dev(m->mgs_bottom_exp->exp_obd->obd_lu_dev);
1181 m->mgs_dt_dev.dd_lu_dev.ld_site = m->mgs_bottom->dd_lu_dev.ld_site;
1182 LASSERT(m->mgs_dt_dev.dd_lu_dev.ld_site);
1188 static struct tgt_handler mgs_mgs_handlers[] = {
1189 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1190 0, MGS_CONNECT, mgs_connect,
1191 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1192 TGT_RPC_HANDLER(MGS_FIRST_OPC,
1193 0, MGS_DISCONNECT, mgs_disconnect,
1194 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
1195 TGT_MGS_HDL_VAR(0, MGS_EXCEPTION, mgs_exception),
1196 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 18, 53, 0)
1197 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_SET_INFO, mgs_set_info),
1199 TGT_MGS_HDL(HAS_REPLY | IS_MUTABLE, MGS_TARGET_REG, mgs_target_reg),
1200 TGT_MGS_HDL_VAR(0, MGS_TARGET_DEL, mgs_target_del),
1201 TGT_MGS_HDL(HAS_REPLY, MGS_CONFIG_READ, mgs_config_read),
1204 static struct tgt_handler mgs_obd_handlers[] = {
1205 TGT_OBD_HDL(0, OBD_PING, tgt_obd_ping),
1208 static struct tgt_handler mgs_dlm_handlers[] = {
1209 [LDLM_ENQUEUE - LDLM_FIRST_OPC] = {
1210 .th_name = "LDLM_ENQUEUE",
1211 /* don't use th_fail_id for MGS to don't interfere with MDS tests.
1212 * * There are no tests for MGS with OBD_FAIL_LDLM_ENQUEUE_NET so it
1213 * * is safe. If such tests will be needed we have to distinguish
1214 * * MDS and MGS fail ids, e.g use OBD_FAIL_MGS_ENQUEUE_NET for MGS
1215 * * instead of common OBD_FAIL_LDLM_ENQUEUE_NET */
1217 .th_opc = LDLM_ENQUEUE,
1218 .th_flags = HAS_KEY,
1219 .th_act = tgt_enqueue,
1220 .th_fmt = &RQF_LDLM_ENQUEUE,
1221 .th_version = LUSTRE_DLM_VERSION,
1225 static struct tgt_handler mgs_llog_handlers[] = {
1226 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, mgs_llog_open),
1227 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1228 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1229 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1232 static struct tgt_opc_slice mgs_common_slice[] = {
1234 .tos_opc_start = MGS_FIRST_OPC,
1235 .tos_opc_end = MGS_LAST_OPC,
1236 .tos_hs = mgs_mgs_handlers
1239 .tos_opc_start = OBD_FIRST_OPC,
1240 .tos_opc_end = OBD_LAST_OPC,
1241 .tos_hs = mgs_obd_handlers
1244 .tos_opc_start = LDLM_FIRST_OPC,
1245 .tos_opc_end = LDLM_LAST_OPC,
1246 .tos_hs = mgs_dlm_handlers
1249 .tos_opc_start = LLOG_FIRST_OPC,
1250 .tos_opc_end = LLOG_LAST_OPC,
1251 .tos_hs = mgs_llog_handlers
1254 .tos_opc_start = SEC_FIRST_OPC,
1255 .tos_opc_end = SEC_LAST_OPC,
1256 .tos_hs = tgt_sec_ctx_handlers
1263 static int mgs_init0(const struct lu_env *env, struct mgs_device *mgs,
1264 struct lu_device_type *ldt, struct lustre_cfg *lcfg)
1266 struct ptlrpc_service_conf conf;
1267 struct obd_device *obd;
1268 struct lustre_mount_info *lmi;
1269 struct llog_ctxt *ctxt;
1274 lmi = server_get_mount(lustre_cfg_string(lcfg, 0));
1278 mgs->mgs_dt_dev.dd_lu_dev.ld_ops = &mgs_lu_ops;
1280 rc = mgs_connect_to_osd(mgs, lustre_cfg_string(lcfg, 3));
1284 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1287 mgs->mgs_obd->obd_lu_dev = &mgs->mgs_dt_dev.dd_lu_dev;
1291 /* namespace for mgs llog */
1292 obd->obd_namespace = ldlm_namespace_new(obd ,"MGS",
1293 LDLM_NAMESPACE_SERVER,
1294 LDLM_NAMESPACE_MODEST,
1296 if (IS_ERR(obd->obd_namespace)) {
1297 rc = PTR_ERR(obd->obd_namespace);
1298 CERROR("%s: unable to create server namespace: rc = %d\n",
1300 obd->obd_namespace = NULL;
1304 /* No recovery for MGCs */
1305 obd->obd_replayable = 0;
1307 rc = tgt_init(env, &mgs->mgs_lut, obd, mgs->mgs_bottom,
1308 mgs_common_slice, OBD_FAIL_MGS_ALL_REQUEST_NET,
1309 OBD_FAIL_MGS_ALL_REPLY_NET);
1313 rc = mgs_fs_setup(env, mgs);
1315 CERROR("%s: MGS filesystem method init failed: rc = %d\n",
1320 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT,
1321 obd, &llog_osd_ops);
1325 /* XXX: we need this trick till N:1 stack is supported
1326 * set "current" directory for named llogs */
1327 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1329 ctxt->loc_dir = mgs->mgs_configs_dir;
1330 llog_ctxt_put(ctxt);
1332 /* Internal mgs setup */
1333 mgs_init_fsdb_list(mgs);
1334 mutex_init(&mgs->mgs_mutex);
1335 mgs->mgs_start_time = ktime_get_real_seconds();
1336 spin_lock_init(&mgs->mgs_lock);
1337 mutex_init(&mgs->mgs_health_mutex);
1338 init_rwsem(&mgs->mgs_barrier_rwsem);
1340 rc = mgs_lcfg_rename(env, mgs);
1344 rc = lproc_mgs_setup(mgs, lustre_cfg_string(lcfg, 3));
1346 CERROR("%s: cannot initialize proc entry: rc = %d\n",
1351 /* Setup params fsdb and log, so that other servers can make a local
1352 * copy successfully when they are mounted. See LU-4783 */
1353 rc = mgs_params_fsdb_setup(env, mgs);
1355 /* params fsdb and log can be setup later */
1356 CERROR("%s: %s fsdb and log setup failed: rc = %d\n",
1357 obd->obd_name, PARAMS_FILENAME, rc);
1359 /* Setup _mgs fsdb, useful for srpc */
1360 mgs__mgs_fsdb_setup(env, mgs);
1362 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1363 "mgs_ldlm_client", &obd->obd_ldlm_client);
1365 conf = (typeof(conf)) {
1366 .psc_name = LUSTRE_MGS_NAME,
1367 .psc_watchdog_factor = MGS_SERVICE_WATCHDOG_FACTOR,
1369 .bc_nbufs = MGS_NBUFS,
1370 .bc_buf_size = MGS_BUFSIZE,
1371 .bc_req_max_size = MGS_MAXREQSIZE,
1372 .bc_rep_max_size = MGS_MAXREPSIZE,
1373 .bc_req_portal = MGS_REQUEST_PORTAL,
1374 .bc_rep_portal = MGC_REPLY_PORTAL,
1377 .tc_thr_name = "ll_mgs",
1378 .tc_nthrs_init = MGS_NTHRS_INIT,
1379 .tc_nthrs_max = MGS_NTHRS_MAX,
1380 .tc_ctx_tags = LCT_MG_THREAD,
1383 .so_req_handler = tgt_request_handle,
1384 .so_req_printer = target_print_req,
1388 /* Start the service threads */
1389 mgs->mgs_service = ptlrpc_register_service(&conf, &obd->obd_kset,
1390 obd->obd_debugfs_entry);
1391 if (IS_ERR(mgs->mgs_service)) {
1392 rc = PTR_ERR(mgs->mgs_service);
1393 CERROR("failed to start mgs service: %d\n", rc);
1394 mgs->mgs_service = NULL;
1395 GOTO(err_lproc, rc);
1398 ping_evictor_start();
1400 CDEBUG(D_INFO, "MGS %s started\n", obd->obd_name);
1402 /* device stack is not yet fully setup to keep no objects behind */
1403 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1406 mgs_params_fsdb_cleanup(env, mgs);
1407 lproc_mgs_cleanup(mgs);
1409 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1411 ctxt->loc_dir = NULL;
1412 llog_cleanup(env, ctxt);
1415 tgt_fini(env, &mgs->mgs_lut);
1417 /* No extra cleanup needed for llog_init_commit_thread() */
1418 mgs_fs_cleanup(env, mgs);
1420 ldlm_namespace_free(obd->obd_namespace, NULL, 0);
1421 obd->obd_namespace = NULL;
1423 lu_site_purge(env, mgs2lu_dev(mgs)->ld_site, ~0);
1424 lu_site_print(env, mgs2lu_dev(mgs)->ld_site,
1425 &mgs2lu_dev(mgs)->ld_site->ls_obj_hash.nelems,
1426 D_OTHER, lu_cdebug_printer);
1427 obd_disconnect(mgs->mgs_bottom_exp);
1430 server_put_mount(lustre_cfg_string(lcfg, 0), true);
1434 static struct lu_device *mgs_device_free(const struct lu_env *env,
1435 struct lu_device *lu)
1437 struct mgs_device *mgs = lu2mgs_dev(lu);
1440 dt_device_fini(&mgs->mgs_dt_dev);
1445 static int mgs_process_config(const struct lu_env *env,
1446 struct lu_device *dev,
1447 struct lustre_cfg *lcfg)
1453 static int mgs_object_init(const struct lu_env *env, struct lu_object *o,
1454 const struct lu_object_conf *unused)
1456 struct mgs_device *d = lu2mgs_dev(o->lo_dev);
1457 struct lu_device *under;
1458 struct lu_object *below;
1462 /* do no set .do_ops as mgs calls to bottom osd directly */
1464 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
1465 PFID(lu_object_fid(o)));
1467 under = &d->mgs_bottom->dd_lu_dev;
1468 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1470 lu_object_add(o, below);
1477 static void mgs_object_free(const struct lu_env *env, struct lu_object *o)
1479 struct mgs_object *obj = lu2mgs_obj(o);
1480 struct lu_object_header *h = o->lo_header;
1482 dt_object_fini(&obj->mgo_obj);
1483 lu_object_header_fini(h);
1484 OBD_FREE_PRE(obj, sizeof(*obj), "kfreed");
1485 kfree_rcu(obj, mgo_header.loh_rcu);
1488 static int mgs_object_print(const struct lu_env *env, void *cookie,
1489 lu_printer_t p, const struct lu_object *l)
1491 const struct mgs_object *o = lu2mgs_obj((struct lu_object *) l);
1493 return (*p)(env, cookie, LUSTRE_MGS_NAME"-object@%p", o);
1496 static const struct lu_object_operations mgs_lu_obj_ops = {
1497 .loo_object_init = mgs_object_init,
1498 .loo_object_free = mgs_object_free,
1499 .loo_object_print = mgs_object_print,
1502 static struct lu_object *mgs_object_alloc(const struct lu_env *env,
1503 const struct lu_object_header *hdr,
1504 struct lu_device *d)
1506 struct lu_object_header *h;
1507 struct mgs_object *o;
1508 struct lu_object *l;
1510 LASSERT(hdr == NULL);
1514 l = &o->mgo_obj.do_lu;
1517 lu_object_header_init(h);
1518 dt_object_init(&o->mgo_obj, h, d);
1519 lu_object_add_top(h, l);
1521 l->lo_ops = &mgs_lu_obj_ops;
1529 const struct lu_device_operations mgs_lu_ops = {
1530 .ldo_object_alloc = mgs_object_alloc,
1531 .ldo_process_config = mgs_process_config,
1534 static struct lu_device *mgs_device_alloc(const struct lu_env *env,
1535 struct lu_device_type *type,
1536 struct lustre_cfg *lcfg)
1538 struct mgs_device *mgs;
1539 struct lu_device *ludev;
1543 ludev = ERR_PTR(-ENOMEM);
1547 ludev = mgs2lu_dev(mgs);
1548 dt_device_init(&mgs->mgs_dt_dev, type);
1549 rc = mgs_init0(env, mgs, type, lcfg);
1551 mgs_device_free(env, ludev);
1552 ludev = ERR_PTR(rc);
1558 static struct lu_device *mgs_device_fini(const struct lu_env *env,
1559 struct lu_device *d)
1561 struct mgs_device *mgs = lu2mgs_dev(d);
1562 struct obd_device *obd = mgs->mgs_obd;
1563 struct llog_ctxt *ctxt;
1567 LASSERT(mgs->mgs_bottom);
1569 class_disconnect_exports(obd);
1571 ping_evictor_stop();
1573 mutex_lock(&mgs->mgs_health_mutex);
1574 ptlrpc_unregister_service(mgs->mgs_service);
1575 mutex_unlock(&mgs->mgs_health_mutex);
1577 mgs_params_fsdb_cleanup(env, mgs);
1578 mgs_cleanup_fsdb_list(mgs);
1580 ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1);
1581 obd_exports_barrier(obd);
1582 obd_zombie_barrier();
1584 tgt_fini(env, &mgs->mgs_lut);
1585 lproc_mgs_cleanup(mgs);
1587 ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT);
1589 ctxt->loc_dir = NULL;
1590 llog_cleanup(env, ctxt);
1593 mgs_fs_cleanup(env, mgs);
1595 ldlm_namespace_free_post(obd->obd_namespace);
1596 obd->obd_namespace = NULL;
1598 lu_site_purge(env, d->ld_site, ~0);
1599 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
1600 D_OTHER, lu_cdebug_printer);
1601 LASSERT(mgs->mgs_bottom_exp);
1602 obd_disconnect(mgs->mgs_bottom_exp);
1604 server_put_mount(obd->obd_name, true);
1609 /* context key constructor/destructor: mgs_key_init, mgs_key_fini */
1610 LU_KEY_INIT_FINI(mgs, struct mgs_thread_info);
1612 LU_TYPE_INIT_FINI(mgs, &mgs_thread_key);
1614 LU_CONTEXT_KEY_DEFINE(mgs, LCT_MG_THREAD);
1616 static const struct lu_device_type_operations mgs_device_type_ops = {
1617 .ldto_init = mgs_type_init,
1618 .ldto_fini = mgs_type_fini,
1620 .ldto_start = mgs_type_start,
1621 .ldto_stop = mgs_type_stop,
1623 .ldto_device_alloc = mgs_device_alloc,
1624 .ldto_device_free = mgs_device_free,
1626 .ldto_device_fini = mgs_device_fini
1629 static struct lu_device_type mgs_device_type = {
1630 .ldt_tags = LU_DEVICE_DT,
1631 .ldt_name = LUSTRE_MGS_NAME,
1632 .ldt_ops = &mgs_device_type_ops,
1633 .ldt_ctx_tags = LCT_MG_THREAD
1636 static int mgs_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
1637 struct obd_device *obd, struct obd_uuid *cluuid,
1638 struct obd_connect_data *data, void *localdata)
1642 if (exp == NULL || obd == NULL || cluuid == NULL)
1645 tgt_counter_incr(exp, LPROC_MGS_CONNECT);
1648 data->ocd_connect_flags &= MGS_CONNECT_SUPPORTED;
1650 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1651 data->ocd_connect_flags2 &= MGS_CONNECT_SUPPORTED2;
1653 exp->exp_connect_data = *data;
1654 data->ocd_version = LUSTRE_VERSION_CODE;
1657 RETURN(mgs_export_stats_init(obd, exp, localdata));
1660 static int mgs_obd_connect(const struct lu_env *env, struct obd_export **exp,
1661 struct obd_device *obd, struct obd_uuid *cluuid,
1662 struct obd_connect_data *data, void *localdata)
1664 struct obd_export *lexp;
1665 struct lustre_handle conn = {
1672 if (exp == NULL || obd == NULL || cluuid == NULL)
1675 rc = class_connect(&conn, obd, cluuid);
1679 lexp = class_conn2export(&conn);
1683 rc = mgs_obd_reconnect(env, lexp, obd, cluuid, data, localdata);
1685 GOTO(out_disconnect, rc);
1692 class_disconnect(lexp);
1697 static int mgs_obd_disconnect(struct obd_export *exp)
1705 mgs_fsc_cleanup(exp);
1707 class_export_get(exp);
1708 tgt_counter_incr(exp, LPROC_MGS_DISCONNECT);
1710 rc = server_disconnect_export(exp);
1711 class_export_put(exp);
1715 static int mgs_health_check(const struct lu_env *env, struct obd_device *obd)
1717 struct mgs_device *mgs = lu2mgs_dev(obd->obd_lu_dev);
1720 mutex_lock(&mgs->mgs_health_mutex);
1721 rc |= ptlrpc_service_health_check(mgs->mgs_service);
1722 mutex_unlock(&mgs->mgs_health_mutex);
1724 return rc != 0 ? 1 : 0;
1727 /* use obd ops to offer management infrastructure */
1728 static const struct obd_ops mgs_obd_device_ops = {
1729 .o_owner = THIS_MODULE,
1730 .o_connect = mgs_obd_connect,
1731 .o_reconnect = mgs_obd_reconnect,
1732 .o_disconnect = mgs_obd_disconnect,
1733 .o_init_export = mgs_init_export,
1734 .o_destroy_export = mgs_destroy_export,
1735 .o_iocontrol = mgs_iocontrol,
1736 .o_health_check = mgs_health_check,
1739 static int __init mgs_init(void)
1741 return class_register_type(&mgs_obd_device_ops, NULL, true,
1742 LUSTRE_MGS_NAME, &mgs_device_type);
1745 static void __exit mgs_exit(void)
1747 class_unregister_type(LUSTRE_MGS_NAME);
1750 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1751 MODULE_DESCRIPTION("Lustre Management Server (MGS)");
1752 MODULE_VERSION(LUSTRE_VERSION_STRING);
1753 MODULE_LICENSE("GPL");
1755 module_init(mgs_init);
1756 module_exit(mgs_exit);