drivers/staging/lustre/lustre/mgc/mgc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  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
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mgc/mgc_request.c
  37  *
  38  * Author: Nathan Rutman <nathan@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_MGC
  42 #define D_MGC D_CONFIG /*|D_WARNING*/
  43
  44 #include <linux/module.h>
  45 #include "../include/obd_class.h"
  46 #include "../include/lustre_dlm.h"
  47 #include "../include/lprocfs_status.h"
  48 #include "../include/lustre_log.h"
  49 #include "../include/lustre_disk.h"
  50
  51 #include "mgc_internal.h"
  52
  53 static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id,
  54                           int type)
  55 {
  56         __u64 resname = 0;
  57
  58         if (len > sizeof(resname)) {
  59                 CERROR("name too long: %s\n", name);
  60                 return -EINVAL;
  61         }
  62         if (len <= 0) {
  63                 CERROR("missing name: %s\n", name);
  64                 return -EINVAL;
  65         }
  66         memcpy(&resname, name, len);
  67
  68         /* Always use the same endianness for the resid */
  69         memset(res_id, 0, sizeof(*res_id));
  70         res_id->name[0] = cpu_to_le64(resname);
  71         /* XXX: unfortunately, sptlprc and config llog share one lock */
  72         switch (type) {
  73         case CONFIG_T_CONFIG:
  74         case CONFIG_T_SPTLRPC:
  75                 resname = 0;
  76                 break;
  77         case CONFIG_T_RECOVER:
  78         case CONFIG_T_PARAMS:
  79                 resname = type;
  80                 break;
  81         default:
  82                 LBUG();
  83         }
  84         res_id->name[1] = cpu_to_le64(resname);
  85         CDEBUG(D_MGC, "log %s to resid %#llx/%#llx (%.8s)\n", name,
  86                res_id->name[0], res_id->name[1], (char *)&res_id->name[0]);
  87         return 0;
  88 }
  89
  90 int mgc_fsname2resid(char *fsname, struct ldlm_res_id *res_id, int type)
  91 {
  92         /* fsname is at most 8 chars long, maybe contain "-".
  93          * e.g. "lustre", "SUN-000"
  94          */
  95         return mgc_name2resid(fsname, strlen(fsname), res_id, type);
  96 }
  97 EXPORT_SYMBOL(mgc_fsname2resid);
  98
  99 static int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type)
 100 {
 101         char *name_end;
 102         int len;
 103
 104         /* logname consists of "fsname-nodetype".
 105          * e.g. "lustre-MDT0001", "SUN-000-client"
 106          * there is an exception: llog "params"
 107          */
 108         name_end = strrchr(logname, '-');
 109         if (!name_end)
 110                 len = strlen(logname);
 111         else
 112                 len = name_end - logname;
 113         return mgc_name2resid(logname, len, res_id, type);
 114 }
 115
 116 /********************** config llog list **********************/
 117 static LIST_HEAD(config_llog_list);
 118 static DEFINE_SPINLOCK(config_list_lock);
 119
 120 /* Take a reference to a config log */
 121 static int config_log_get(struct config_llog_data *cld)
 122 {
 123         atomic_inc(&cld->cld_refcount);
 124         CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
 125                atomic_read(&cld->cld_refcount));
 126         return 0;
 127 }
 128
 129 /* Drop a reference to a config log.  When no longer referenced,
 130  * we can free the config log data
 131  */
 132 static void config_log_put(struct config_llog_data *cld)
 133 {
 134         CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
 135                atomic_read(&cld->cld_refcount));
 136         LASSERT(atomic_read(&cld->cld_refcount) > 0);
 137
 138         /* spinlock to make sure no item with 0 refcount in the list */
 139         if (atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) {
 140                 list_del(&cld->cld_list_chain);
 141                 spin_unlock(&config_list_lock);
 142
 143                 CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname);
 144
 145                 if (cld->cld_recover)
 146                         config_log_put(cld->cld_recover);
 147                 if (cld->cld_sptlrpc)
 148                         config_log_put(cld->cld_sptlrpc);
 149                 if (cld->cld_params)
 150                         config_log_put(cld->cld_params);
 151                 if (cld_is_sptlrpc(cld))
 152                         sptlrpc_conf_log_stop(cld->cld_logname);
 153
 154                 class_export_put(cld->cld_mgcexp);
 155                 kfree(cld);
 156         }
 157 }
 158
 159 /* Find a config log by name */
 160 static
 161 struct config_llog_data *config_log_find(char *logname,
 162                                          struct config_llog_instance *cfg)
 163 {
 164         struct config_llog_data *cld;
 165         struct config_llog_data *found = NULL;
 166         void *instance;
 167
 168         LASSERT(logname);
 169
 170         instance = cfg ? cfg->cfg_instance : NULL;
 171         spin_lock(&config_list_lock);
 172         list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
 173                 /* check if instance equals */
 174                 if (instance != cld->cld_cfg.cfg_instance)
 175                         continue;
 176
 177                 /* instance may be NULL, should check name */
 178                 if (strcmp(logname, cld->cld_logname) == 0) {
 179                         found = cld;
 180                         break;
 181                 }
 182         }
 183         if (found) {
 184                 atomic_inc(&found->cld_refcount);
 185                 LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0);
 186         }
 187         spin_unlock(&config_list_lock);
 188         return found;
 189 }
 190
 191 static
 192 struct config_llog_data *do_config_log_add(struct obd_device *obd,
 193                                            char *logname,
 194                                            int type,
 195                                            struct config_llog_instance *cfg,
 196                                            struct super_block *sb)
 197 {
 198         struct config_llog_data *cld;
 199         int                   rc;
 200
 201         CDEBUG(D_MGC, "do adding config log %s:%p\n", logname,
 202                cfg ? cfg->cfg_instance : NULL);
 203
 204         cld = kzalloc(sizeof(*cld) + strlen(logname) + 1, GFP_NOFS);
 205         if (!cld)
 206                 return ERR_PTR(-ENOMEM);
 207
 208         strcpy(cld->cld_logname, logname);
 209         if (cfg)
 210                 cld->cld_cfg = *cfg;
 211         else
 212                 cld->cld_cfg.cfg_callback = class_config_llog_handler;
 213         mutex_init(&cld->cld_lock);
 214         cld->cld_cfg.cfg_last_idx = 0;
 215         cld->cld_cfg.cfg_flags = 0;
 216         cld->cld_cfg.cfg_sb = sb;
 217         cld->cld_type = type;
 218         atomic_set(&cld->cld_refcount, 1);
 219
 220         /* Keep the mgc around until we are done */
 221         cld->cld_mgcexp = class_export_get(obd->obd_self_export);
 222
 223         if (cld_is_sptlrpc(cld)) {
 224                 sptlrpc_conf_log_start(logname);
 225                 cld->cld_cfg.cfg_obdname = obd->obd_name;
 226         }
 227
 228         rc = mgc_logname2resid(logname, &cld->cld_resid, type);
 229
 230         spin_lock(&config_list_lock);
 231         list_add(&cld->cld_list_chain, &config_llog_list);
 232         spin_unlock(&config_list_lock);
 233
 234         if (rc) {
 235                 config_log_put(cld);
 236                 return ERR_PTR(rc);
 237         }
 238
 239         if (cld_is_sptlrpc(cld)) {
 240                 rc = mgc_process_log(obd, cld);
 241                 if (rc && rc != -ENOENT)
 242                         CERROR("failed processing sptlrpc log: %d\n", rc);
 243         }
 244
 245         return cld;
 246 }
 247
 248 static struct config_llog_data *
 249 config_recover_log_add(struct obd_device *obd, char *fsname,
 250                        struct config_llog_instance *cfg,
 251                        struct super_block *sb)
 252 {
 253         struct config_llog_instance lcfg = *cfg;
 254         struct config_llog_data *cld;
 255         char logname[32];
 256
 257         /* we have to use different llog for clients and mdts for cmd
 258          * where only clients are notified if one of cmd server restarts
 259          */
 260         LASSERT(strlen(fsname) < sizeof(logname) / 2);
 261         strcpy(logname, fsname);
 262         LASSERT(lcfg.cfg_instance);
 263         strcat(logname, "-cliir");
 264
 265         cld = do_config_log_add(obd, logname, CONFIG_T_RECOVER, &lcfg, sb);
 266         return cld;
 267 }
 268
 269 static struct config_llog_data *
 270 config_params_log_add(struct obd_device *obd,
 271                       struct config_llog_instance *cfg, struct super_block *sb)
 272 {
 273         struct config_llog_instance     lcfg = *cfg;
 274         struct config_llog_data         *cld;
 275
 276         lcfg.cfg_instance = sb;
 277
 278         cld = do_config_log_add(obd, PARAMS_FILENAME, CONFIG_T_PARAMS,
 279                                 &lcfg, sb);
 280
 281         return cld;
 282 }
 283
 284 /** Add this log to the list of active logs watched by an MGC.
 285  * Active means we're watching for updates.
 286  * We have one active log per "mount" - client instance or servername.
 287  * Each instance may be at a different point in the log.
 288  */
 289 static int config_log_add(struct obd_device *obd, char *logname,
 290                           struct config_llog_instance *cfg,
 291                           struct super_block *sb)
 292 {
 293         struct lustre_sb_info *lsi = s2lsi(sb);
 294         struct config_llog_data *cld;
 295         struct config_llog_data *sptlrpc_cld;
 296         struct config_llog_data *params_cld;
 297         char                    seclogname[32];
 298         char                    *ptr;
 299         int                     rc;
 300
 301         CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance);
 302
 303         /*
 304          * for each regular log, the depended sptlrpc log name is
 305          * <fsname>-sptlrpc. multiple regular logs may share one sptlrpc log.
 306          */
 307         ptr = strrchr(logname, '-');
 308         if (!ptr || ptr - logname > 8) {
 309                 CERROR("logname %s is too long\n", logname);
 310                 return -EINVAL;
 311         }
 312
 313         memcpy(seclogname, logname, ptr - logname);
 314         strcpy(seclogname + (ptr - logname), "-sptlrpc");
 315
 316         sptlrpc_cld = config_log_find(seclogname, NULL);
 317         if (!sptlrpc_cld) {
 318                 sptlrpc_cld = do_config_log_add(obd, seclogname,
 319                                                 CONFIG_T_SPTLRPC, NULL, NULL);
 320                 if (IS_ERR(sptlrpc_cld)) {
 321                         CERROR("can't create sptlrpc log: %s\n", seclogname);
 322                         rc = PTR_ERR(sptlrpc_cld);
 323                         goto out_err;
 324                 }
 325         }
 326         params_cld = config_params_log_add(obd, cfg, sb);
 327         if (IS_ERR(params_cld)) {
 328                 rc = PTR_ERR(params_cld);
 329                 CERROR("%s: can't create params log: rc = %d\n",
 330                        obd->obd_name, rc);
 331                 goto out_err1;
 332         }
 333
 334         cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb);
 335         if (IS_ERR(cld)) {
 336                 CERROR("can't create log: %s\n", logname);
 337                 rc = PTR_ERR(cld);
 338                 goto out_err2;
 339         }
 340
 341         cld->cld_sptlrpc = sptlrpc_cld;
 342         cld->cld_params = params_cld;
 343
 344         LASSERT(lsi->lsi_lmd);
 345         if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) {
 346                 struct config_llog_data *recover_cld;
 347
 348                 ptr = strrchr(seclogname, '-');
 349                 if (ptr) {
 350                         *ptr = 0;
 351                 } else {
 352                         CERROR("%s: sptlrpc log name not correct, %s: rc = %d\n",
 353                                obd->obd_name, seclogname, -EINVAL);
 354                         config_log_put(cld);
 355                         return -EINVAL;
 356                 }
 357                 recover_cld = config_recover_log_add(obd, seclogname, cfg, sb);
 358                 if (IS_ERR(recover_cld)) {
 359                         rc = PTR_ERR(recover_cld);
 360                         goto out_err3;
 361                 }
 362                 cld->cld_recover = recover_cld;
 363         }
 364
 365         return 0;
 366
 367 out_err3:
 368         config_log_put(cld);
 369
 370 out_err2:
 371         config_log_put(params_cld);
 372
 373 out_err1:
 374         config_log_put(sptlrpc_cld);
 375
 376 out_err:
 377         return rc;
 378 }
 379
 380 DEFINE_MUTEX(llog_process_lock);
 381
 382 /** Stop watching for updates on this log.
 383  */
 384 static int config_log_end(char *logname, struct config_llog_instance *cfg)
 385 {
 386         struct config_llog_data *cld;
 387         struct config_llog_data *cld_sptlrpc = NULL;
 388         struct config_llog_data *cld_params = NULL;
 389         struct config_llog_data *cld_recover = NULL;
 390         int rc = 0;
 391
 392         cld = config_log_find(logname, cfg);
 393         if (!cld)
 394                 return -ENOENT;
 395
 396         mutex_lock(&cld->cld_lock);
 397         /*
 398          * if cld_stopping is set, it means we didn't start the log thus
 399          * not owning the start ref. this can happen after previous umount:
 400          * the cld still hanging there waiting for lock cancel, and we
 401          * remount again but failed in the middle and call log_end without
 402          * calling start_log.
 403          */
 404         if (unlikely(cld->cld_stopping)) {
 405                 mutex_unlock(&cld->cld_lock);
 406                 /* drop the ref from the find */
 407                 config_log_put(cld);
 408                 return rc;
 409         }
 410
 411         cld->cld_stopping = 1;
 412
 413         cld_recover = cld->cld_recover;
 414         cld->cld_recover = NULL;
 415         mutex_unlock(&cld->cld_lock);
 416
 417         if (cld_recover) {
 418                 mutex_lock(&cld_recover->cld_lock);
 419                 cld_recover->cld_stopping = 1;
 420                 mutex_unlock(&cld_recover->cld_lock);
 421                 config_log_put(cld_recover);
 422         }
 423
 424         spin_lock(&config_list_lock);
 425         cld_sptlrpc = cld->cld_sptlrpc;
 426         cld->cld_sptlrpc = NULL;
 427         cld_params = cld->cld_params;
 428         cld->cld_params = NULL;
 429         spin_unlock(&config_list_lock);
 430
 431         if (cld_sptlrpc)
 432                 config_log_put(cld_sptlrpc);
 433
 434         if (cld_params) {
 435                 mutex_lock(&cld_params->cld_lock);
 436                 cld_params->cld_stopping = 1;
 437                 mutex_unlock(&cld_params->cld_lock);
 438                 config_log_put(cld_params);
 439         }
 440
 441         /* drop the ref from the find */
 442         config_log_put(cld);
 443         /* drop the start ref */
 444         config_log_put(cld);
 445
 446         CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client",
 447                rc);
 448         return rc;
 449 }
 450
 451 int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data)
 452 {
 453         struct obd_device       *obd = data;
 454         struct obd_import       *imp;
 455         struct obd_connect_data *ocd;
 456         struct config_llog_data *cld;
 457         int rc;
 458
 459         rc = lprocfs_climp_check(obd);
 460         if (rc)
 461                 return rc;
 462
 463         imp = obd->u.cli.cl_import;
 464         ocd = &imp->imp_connect_data;
 465
 466         seq_printf(m, "imperative_recovery: %s\n",
 467                    OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED");
 468         seq_printf(m, "client_state:\n");
 469
 470         spin_lock(&config_list_lock);
 471         list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
 472                 if (!cld->cld_recover)
 473                         continue;
 474                 seq_printf(m, "    - { client: %s, nidtbl_version: %u }\n",
 475                            cld->cld_logname,
 476                            cld->cld_recover->cld_cfg.cfg_last_idx);
 477         }
 478         spin_unlock(&config_list_lock);
 479
 480         up_read(&obd->u.cli.cl_sem);
 481         return 0;
 482 }
 483
 484 /* reenqueue any lost locks */
 485 #define RQ_RUNNING 0x1
 486 #define RQ_NOW     0x2
 487 #define RQ_LATER   0x4
 488 #define RQ_STOP    0x8
 489 #define RQ_PRECLEANUP  0x10
 490 static int rq_state;
 491 static wait_queue_head_t            rq_waitq;
 492 static DECLARE_COMPLETION(rq_exit);
 493 static DECLARE_COMPLETION(rq_start);
 494
 495 static void do_requeue(struct config_llog_data *cld)
 496 {
 497         LASSERT(atomic_read(&cld->cld_refcount) > 0);
 498
 499         /* Do not run mgc_process_log on a disconnected export or an
 500          * export which is being disconnected. Take the client
 501          * semaphore to make the check non-racy.
 502          */
 503         down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
 504         if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) {
 505                 CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname);
 506                 mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
 507         } else {
 508                 CDEBUG(D_MGC, "disconnecting, won't update log %s\n",
 509                        cld->cld_logname);
 510         }
 511         up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
 512 }
 513
 514 /* this timeout represents how many seconds MGC should wait before
 515  * requeue config and recover lock to the MGS. We need to randomize this
 516  * in order to not flood the MGS.
 517  */
 518 #define MGC_TIMEOUT_MIN_SECONDS   5
 519 #define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 */
 520
 521 static int mgc_requeue_thread(void *data)
 522 {
 523         bool first = true;
 524
 525         CDEBUG(D_MGC, "Starting requeue thread\n");
 526
 527         /* Keep trying failed locks periodically */
 528         spin_lock(&config_list_lock);
 529         rq_state |= RQ_RUNNING;
 530         while (1) {
 531                 struct l_wait_info lwi;
 532                 struct config_llog_data *cld, *cld_prev;
 533                 int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC;
 534                 int stopped = !!(rq_state & RQ_STOP);
 535                 int to;
 536
 537                 /* Any new or requeued lostlocks will change the state */
 538                 rq_state &= ~(RQ_NOW | RQ_LATER);
 539                 spin_unlock(&config_list_lock);
 540
 541                 if (first) {
 542                         first = false;
 543                         complete(&rq_start);
 544                 }
 545
 546                 /* Always wait a few seconds to allow the server who
 547                  * caused the lock revocation to finish its setup, plus some
 548                  * random so everyone doesn't try to reconnect at once.
 549                  */
 550                 to = MGC_TIMEOUT_MIN_SECONDS * HZ;
 551                 to += rand * HZ / 100; /* rand is centi-seconds */
 552                 lwi = LWI_TIMEOUT(to, NULL, NULL);
 553                 l_wait_event(rq_waitq, rq_state & (RQ_STOP | RQ_PRECLEANUP),
 554                              &lwi);
 555
 556                 /*
 557                  * iterate & processing through the list. for each cld, process
 558                  * its depending sptlrpc cld firstly (if any) and then itself.
 559                  *
 560                  * it's guaranteed any item in the list must have
 561                  * reference > 0; and if cld_lostlock is set, at
 562                  * least one reference is taken by the previous enqueue.
 563                  */
 564                 cld_prev = NULL;
 565
 566                 spin_lock(&config_list_lock);
 567                 rq_state &= ~RQ_PRECLEANUP;
 568                 list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
 569                         if (!cld->cld_lostlock)
 570                                 continue;
 571
 572                         spin_unlock(&config_list_lock);
 573
 574                         LASSERT(atomic_read(&cld->cld_refcount) > 0);
 575
 576                         /* Whether we enqueued again or not in mgc_process_log,
 577                          * we're done with the ref from the old enqueue
 578                          */
 579                         if (cld_prev)
 580                                 config_log_put(cld_prev);
 581                         cld_prev = cld;
 582
 583                         cld->cld_lostlock = 0;
 584                         if (likely(!stopped))
 585                                 do_requeue(cld);
 586
 587                         spin_lock(&config_list_lock);
 588                 }
 589                 spin_unlock(&config_list_lock);
 590                 if (cld_prev)
 591                         config_log_put(cld_prev);
 592
 593                 /* break after scanning the list so that we can drop
 594                  * refcount to losing lock clds
 595                  */
 596                 if (unlikely(stopped)) {
 597                         spin_lock(&config_list_lock);
 598                         break;
 599                 }
 600
 601                 /* Wait a bit to see if anyone else needs a requeue */
 602                 lwi = (struct l_wait_info) { 0 };
 603                 l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
 604                              &lwi);
 605                 spin_lock(&config_list_lock);
 606         }
 607         /* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */
 608         rq_state &= ~RQ_RUNNING;
 609         spin_unlock(&config_list_lock);
 610
 611         complete(&rq_exit);
 612
 613         CDEBUG(D_MGC, "Ending requeue thread\n");
 614         return 0;
 615 }
 616
 617 /* Add a cld to the list to requeue.  Start the requeue thread if needed.
 618  * We are responsible for dropping the config log reference from here on out.
 619  */
 620 static void mgc_requeue_add(struct config_llog_data *cld)
 621 {
 622         CDEBUG(D_INFO, "log %s: requeue (r=%d sp=%d st=%x)\n",
 623                cld->cld_logname, atomic_read(&cld->cld_refcount),
 624                cld->cld_stopping, rq_state);
 625         LASSERT(atomic_read(&cld->cld_refcount) > 0);
 626
 627         mutex_lock(&cld->cld_lock);
 628         if (cld->cld_stopping || cld->cld_lostlock) {
 629                 mutex_unlock(&cld->cld_lock);
 630                 return;
 631         }
 632         /* this refcount will be released in mgc_requeue_thread. */
 633         config_log_get(cld);
 634         cld->cld_lostlock = 1;
 635         mutex_unlock(&cld->cld_lock);
 636
 637         /* Hold lock for rq_state */
 638         spin_lock(&config_list_lock);
 639         if (rq_state & RQ_STOP) {
 640                 spin_unlock(&config_list_lock);
 641                 cld->cld_lostlock = 0;
 642                 config_log_put(cld);
 643         } else {
 644                 rq_state |= RQ_NOW;
 645                 spin_unlock(&config_list_lock);
 646                 wake_up(&rq_waitq);
 647         }
 648 }
 649
 650 static int mgc_llog_init(const struct lu_env *env, struct obd_device *obd)
 651 {
 652         struct llog_ctxt        *ctxt;
 653         int                      rc;
 654
 655         /* setup only remote ctxt, the local disk context is switched per each
 656          * filesystem during mgc_fs_setup()
 657          */
 658         rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, obd,
 659                         &llog_client_ops);
 660         if (rc)
 661                 return rc;
 662
 663         ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 664         LASSERT(ctxt);
 665
 666         llog_initiator_connect(ctxt);
 667         llog_ctxt_put(ctxt);
 668
 669         return 0;
 670 }
 671
 672 static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd)
 673 {
 674         struct llog_ctxt *ctxt;
 675
 676         ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 677         if (ctxt)
 678                 llog_cleanup(env, ctxt);
 679
 680         return 0;
 681 }
 682
 683 static atomic_t mgc_count = ATOMIC_INIT(0);
 684 static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 685 {
 686         int rc = 0;
 687         int temp;
 688
 689         switch (stage) {
 690         case OBD_CLEANUP_EARLY:
 691                 break;
 692         case OBD_CLEANUP_EXPORTS:
 693                 if (atomic_dec_and_test(&mgc_count)) {
 694                         LASSERT(rq_state & RQ_RUNNING);
 695                         /* stop requeue thread */
 696                         temp = RQ_STOP;
 697                 } else {
 698                         /* wakeup requeue thread to clean our cld */
 699                         temp = RQ_NOW | RQ_PRECLEANUP;
 700                 }
 701                 spin_lock(&config_list_lock);
 702                 rq_state |= temp;
 703                 spin_unlock(&config_list_lock);
 704                 wake_up(&rq_waitq);
 705                 if (temp & RQ_STOP)
 706                         wait_for_completion(&rq_exit);
 707                 obd_cleanup_client_import(obd);
 708                 rc = mgc_llog_fini(NULL, obd);
 709                 if (rc != 0)
 710                         CERROR("failed to cleanup llogging subsystems\n");
 711                 break;
 712         }
 713         return rc;
 714 }
 715
 716 static int mgc_cleanup(struct obd_device *obd)
 717 {
 718         /* COMPAT_146 - old config logs may have added profiles we don't
 719          * know about
 720          */
 721         if (obd->obd_type->typ_refcnt <= 1)
 722                 /* Only for the last mgc */
 723                 class_del_profiles();
 724
 725         lprocfs_obd_cleanup(obd);
 726         ptlrpcd_decref();
 727
 728         return client_obd_cleanup(obd);
 729 }
 730
 731 static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 732 {
 733         struct lprocfs_static_vars lvars = { NULL };
 734         struct task_struct *task;
 735         int rc;
 736
 737         ptlrpcd_addref();
 738
 739         rc = client_obd_setup(obd, lcfg);
 740         if (rc)
 741                 goto err_decref;
 742
 743         rc = mgc_llog_init(NULL, obd);
 744         if (rc) {
 745                 CERROR("failed to setup llogging subsystems\n");
 746                 goto err_cleanup;
 747         }
 748
 749         lprocfs_mgc_init_vars(&lvars);
 750         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
 751         sptlrpc_lprocfs_cliobd_attach(obd);
 752
 753         if (atomic_inc_return(&mgc_count) == 1) {
 754                 rq_state = 0;
 755                 init_waitqueue_head(&rq_waitq);
 756
 757                 /* start requeue thread */
 758                 task = kthread_run(mgc_requeue_thread, NULL, "ll_cfg_requeue");
 759                 if (IS_ERR(task)) {
 760                         rc = PTR_ERR(task);
 761                         CERROR("%s: cannot start requeue thread: rc = %d; no more log updates\n",
 762                                obd->obd_name, rc);
 763                         goto err_cleanup;
 764                 }
 765                 /* rc is the task_struct pointer of mgc_requeue_thread. */
 766                 rc = 0;
 767                 wait_for_completion(&rq_start);
 768         }
 769
 770         return rc;
 771
 772 err_cleanup:
 773         client_obd_cleanup(obd);
 774 err_decref:
 775         ptlrpcd_decref();
 776         return rc;
 777 }
 778
 779 /* based on ll_mdc_blocking_ast */
 780 static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 781                             void *data, int flag)
 782 {
 783         struct lustre_handle lockh;
 784         struct config_llog_data *cld = data;
 785         int rc = 0;
 786
 787         switch (flag) {
 788         case LDLM_CB_BLOCKING:
 789                 /* mgs wants the lock, give it up... */
 790                 LDLM_DEBUG(lock, "MGC blocking CB");
 791                 ldlm_lock2handle(lock, &lockh);
 792                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 793                 break;
 794         case LDLM_CB_CANCELING:
 795                 /* We've given up the lock, prepare ourselves to update. */
 796                 LDLM_DEBUG(lock, "MGC cancel CB");
 797
 798                 CDEBUG(D_MGC, "Lock res "DLDLMRES" (%.8s)\n",
 799                        PLDLMRES(lock->l_resource),
 800                        (char *)&lock->l_resource->lr_name.name[0]);
 801
 802                 if (!cld) {
 803                         CDEBUG(D_INFO, "missing data, won't requeue\n");
 804                         break;
 805                 }
 806
 807                 /* held at mgc_process_log(). */
 808                 LASSERT(atomic_read(&cld->cld_refcount) > 0);
 809                 /* Are we done with this log? */
 810                 if (cld->cld_stopping) {
 811                         CDEBUG(D_MGC, "log %s: stopping, won't requeue\n",
 812                                cld->cld_logname);
 813                         config_log_put(cld);
 814                         break;
 815                 }
 816                 /* Make sure not to re-enqueue when the mgc is stopping
 817                  * (we get called from client_disconnect_export)
 818                  */
 819                 if (!lock->l_conn_export ||
 820                     !lock->l_conn_export->exp_obd->u.cli.cl_conn_count) {
 821                         CDEBUG(D_MGC, "log %.8s: disconnecting, won't requeue\n",
 822                                cld->cld_logname);
 823                         config_log_put(cld);
 824                         break;
 825                 }
 826
 827                 /* Re-enqueue now */
 828                 mgc_requeue_add(cld);
 829                 config_log_put(cld);
 830                 break;
 831         default:
 832                 LBUG();
 833         }
 834
 835         return rc;
 836 }
 837
 838 /* Not sure where this should go... */
 839 /* This is the timeout value for MGS_CONNECT request plus a ping interval, such
 840  * that we can have a chance to try the secondary MGS if any.
 841  */
 842 #define  MGC_ENQUEUE_LIMIT (INITIAL_CONNECT_TIMEOUT + (AT_OFF ? 0 : at_min) \
 843                                 + PING_INTERVAL)
 844 #define  MGC_TARGET_REG_LIMIT 10
 845 #define  MGC_SEND_PARAM_LIMIT 10
 846
 847 /* Send parameter to MGS*/
 848 static int mgc_set_mgs_param(struct obd_export *exp,
 849                              struct mgs_send_param *msp)
 850 {
 851         struct ptlrpc_request *req;
 852         struct mgs_send_param *req_msp, *rep_msp;
 853         int rc;
 854
 855         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 856                                         &RQF_MGS_SET_INFO, LUSTRE_MGS_VERSION,
 857                                         MGS_SET_INFO);
 858         if (!req)
 859                 return -ENOMEM;
 860
 861         req_msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
 862         if (!req_msp) {
 863                 ptlrpc_req_finished(req);
 864                 return -ENOMEM;
 865         }
 866
 867         memcpy(req_msp, msp, sizeof(*req_msp));
 868         ptlrpc_request_set_replen(req);
 869
 870         /* Limit how long we will wait for the enqueue to complete */
 871         req->rq_delay_limit = MGC_SEND_PARAM_LIMIT;
 872         rc = ptlrpc_queue_wait(req);
 873         if (!rc) {
 874                 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
 875                 memcpy(msp, rep_msp, sizeof(*rep_msp));
 876         }
 877
 878         ptlrpc_req_finished(req);
 879
 880         return rc;
 881 }
 882
 883 /* Take a config lock so we can get cancel notifications */
 884 static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
 885                        __u32 type, ldlm_policy_data_t *policy, __u32 mode,
 886                        __u64 *flags, void *bl_cb, void *cp_cb, void *gl_cb,
 887                        void *data, __u32 lvb_len, void *lvb_swabber,
 888                        struct lustre_handle *lockh)
 889 {
 890         struct config_llog_data *cld = data;
 891         struct ldlm_enqueue_info einfo = {
 892                 .ei_type        = type,
 893                 .ei_mode        = mode,
 894                 .ei_cb_bl       = mgc_blocking_ast,
 895                 .ei_cb_cp       = ldlm_completion_ast,
 896         };
 897         struct ptlrpc_request *req;
 898         int short_limit = cld_is_sptlrpc(cld);
 899         int rc;
 900
 901         CDEBUG(D_MGC, "Enqueue for %s (res %#llx)\n", cld->cld_logname,
 902                cld->cld_resid.name[0]);
 903
 904         /* We need a callback for every lockholder, so don't try to
 905          * ldlm_lock_match (see rev 1.1.2.11.2.47)
 906          */
 907         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 908                                         &RQF_LDLM_ENQUEUE, LUSTRE_DLM_VERSION,
 909                                         LDLM_ENQUEUE);
 910         if (!req)
 911                 return -ENOMEM;
 912
 913         req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, 0);
 914         ptlrpc_request_set_replen(req);
 915
 916         /* Limit how long we will wait for the enqueue to complete */
 917         req->rq_delay_limit = short_limit ? 5 : MGC_ENQUEUE_LIMIT;
 918         rc = ldlm_cli_enqueue(exp, &req, &einfo, &cld->cld_resid, NULL, flags,
 919                               NULL, 0, LVB_T_NONE, lockh, 0);
 920         /* A failed enqueue should still call the mgc_blocking_ast,
 921          * where it will be requeued if needed ("grant failed").
 922          */
 923         ptlrpc_req_finished(req);
 924         return rc;
 925 }
 926
 927 static void mgc_notify_active(struct obd_device *unused)
 928 {
 929         /* wakeup mgc_requeue_thread to requeue mgc lock */
 930         spin_lock(&config_list_lock);
 931         rq_state |= RQ_NOW;
 932         spin_unlock(&config_list_lock);
 933         wake_up(&rq_waitq);
 934
 935         /* TODO: Help the MGS rebuild nidtbl. -jay */
 936 }
 937
 938 /* Send target_reg message to MGS */
 939 static int mgc_target_register(struct obd_export *exp,
 940                                struct mgs_target_info *mti)
 941 {
 942         struct ptlrpc_request  *req;
 943         struct mgs_target_info *req_mti, *rep_mti;
 944         int                  rc;
 945
 946         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 947                                         &RQF_MGS_TARGET_REG, LUSTRE_MGS_VERSION,
 948                                         MGS_TARGET_REG);
 949         if (!req)
 950                 return -ENOMEM;
 951
 952         req_mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
 953         if (!req_mti) {
 954                 ptlrpc_req_finished(req);
 955                 return -ENOMEM;
 956         }
 957
 958         memcpy(req_mti, mti, sizeof(*req_mti));
 959         ptlrpc_request_set_replen(req);
 960         CDEBUG(D_MGC, "register %s\n", mti->mti_svname);
 961         /* Limit how long we will wait for the enqueue to complete */
 962         req->rq_delay_limit = MGC_TARGET_REG_LIMIT;
 963
 964         rc = ptlrpc_queue_wait(req);
 965         if (!rc) {
 966                 rep_mti = req_capsule_server_get(&req->rq_pill,
 967                                                  &RMF_MGS_TARGET_INFO);
 968                 memcpy(mti, rep_mti, sizeof(*rep_mti));
 969                 CDEBUG(D_MGC, "register %s got index = %d\n",
 970                        mti->mti_svname, mti->mti_stripe_index);
 971         }
 972         ptlrpc_req_finished(req);
 973
 974         return rc;
 975 }
 976
 977 static int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp,
 978                               u32 keylen, void *key, u32 vallen,
 979                               void *val, struct ptlrpc_request_set *set)
 980 {
 981         int rc = -EINVAL;
 982
 983         /* Turn off initial_recov after we try all backup servers once */
 984         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
 985                 struct obd_import *imp = class_exp2cliimp(exp);
 986                 int value;
 987
 988                 if (vallen != sizeof(int))
 989                         return -EINVAL;
 990                 value = *(int *)val;
 991                 CDEBUG(D_MGC, "InitRecov %s %d/d%d:i%d:r%d:or%d:%s\n",
 992                        imp->imp_obd->obd_name, value,
 993                        imp->imp_deactive, imp->imp_invalid,
 994                        imp->imp_replayable, imp->imp_obd->obd_replayable,
 995                        ptlrpc_import_state_name(imp->imp_state));
 996                 /* Resurrect if we previously died */
 997                 if ((imp->imp_state != LUSTRE_IMP_FULL &&
 998                      imp->imp_state != LUSTRE_IMP_NEW) || value > 1)
 999                         ptlrpc_reconnect_import(imp);
1000                 return 0;
1001         }
1002         if (KEY_IS(KEY_SET_INFO)) {
1003                 struct mgs_send_param *msp;
1004
1005                 msp = val;
1006                 rc =  mgc_set_mgs_param(exp, msp);
1007                 return rc;
1008         }
1009         if (KEY_IS(KEY_MGSSEC)) {
1010                 struct client_obd     *cli = &exp->exp_obd->u.cli;
1011                 struct sptlrpc_flavor  flvr;
1012
1013                 /*
1014                  * empty string means using current flavor, if which haven't
1015                  * been set yet, set it as null.
1016                  *
1017                  * if flavor has been set previously, check the asking flavor
1018                  * must match the existing one.
1019                  */
1020                 if (vallen == 0) {
1021                         if (cli->cl_flvr_mgc.sf_rpc != SPTLRPC_FLVR_INVALID)
1022                                 return 0;
1023                         val = "null";
1024                         vallen = 4;
1025                 }
1026
1027                 rc = sptlrpc_parse_flavor(val, &flvr);
1028                 if (rc) {
1029                         CERROR("invalid sptlrpc flavor %s to MGS\n",
1030                                (char *) val);
1031                         return rc;
1032                 }
1033
1034                 /*
1035                  * caller already hold a mutex
1036                  */
1037                 if (cli->cl_flvr_mgc.sf_rpc == SPTLRPC_FLVR_INVALID) {
1038                         cli->cl_flvr_mgc = flvr;
1039                 } else if (memcmp(&cli->cl_flvr_mgc, &flvr,
1040                                   sizeof(flvr)) != 0) {
1041                         char    str[20];
1042
1043                         sptlrpc_flavor2name(&cli->cl_flvr_mgc,
1044                                             str, sizeof(str));
1045                         LCONSOLE_ERROR("asking sptlrpc flavor %s to MGS but currently %s is in use\n",
1046                                        (char *) val, str);
1047                         rc = -EPERM;
1048                 }
1049                 return rc;
1050         }
1051
1052         return rc;
1053 }
1054
1055 static int mgc_get_info(const struct lu_env *env, struct obd_export *exp,
1056                         __u32 keylen, void *key, __u32 *vallen, void *val,
1057                         struct lov_stripe_md *unused)
1058 {
1059         int rc = -EINVAL;
1060
1061         if (KEY_IS(KEY_CONN_DATA)) {
1062                 struct obd_import *imp = class_exp2cliimp(exp);
1063                 struct obd_connect_data *data = val;
1064
1065                 if (*vallen == sizeof(*data)) {
1066                         *data = imp->imp_connect_data;
1067                         rc = 0;
1068                 }
1069         }
1070
1071         return rc;
1072 }
1073
1074 static int mgc_import_event(struct obd_device *obd,
1075                             struct obd_import *imp,
1076                             enum obd_import_event event)
1077 {
1078         LASSERT(imp->imp_obd == obd);
1079         CDEBUG(D_MGC, "import event %#x\n", event);
1080
1081         switch (event) {
1082         case IMP_EVENT_DISCON:
1083                 /* MGC imports should not wait for recovery */
1084                 if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1085                         ptlrpc_pinger_ir_down();
1086                 break;
1087         case IMP_EVENT_INACTIVE:
1088                 break;
1089         case IMP_EVENT_INVALIDATE: {
1090                 struct ldlm_namespace *ns = obd->obd_namespace;
1091
1092                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1093                 break;
1094         }
1095         case IMP_EVENT_ACTIVE:
1096                 CDEBUG(D_INFO, "%s: Reactivating import\n", obd->obd_name);
1097                 /* Clearing obd_no_recov allows us to continue pinging */
1098                 obd->obd_no_recov = 0;
1099                 mgc_notify_active(obd);
1100                 if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1101                         ptlrpc_pinger_ir_up();
1102                 break;
1103         case IMP_EVENT_OCD:
1104                 break;
1105         case IMP_EVENT_DEACTIVATE:
1106         case IMP_EVENT_ACTIVATE:
1107                 break;
1108         default:
1109                 CERROR("Unknown import event %#x\n", event);
1110                 LBUG();
1111         }
1112         return 0;
1113 }
1114
1115 enum {
1116         CONFIG_READ_NRPAGES_INIT = 1 << (20 - PAGE_SHIFT),
1117         CONFIG_READ_NRPAGES      = 4
1118 };
1119
1120 static int mgc_apply_recover_logs(struct obd_device *mgc,
1121                                   struct config_llog_data *cld,
1122                                   __u64 max_version,
1123                                   void *data, int datalen, bool mne_swab)
1124 {
1125         struct config_llog_instance *cfg = &cld->cld_cfg;
1126         struct mgs_nidtbl_entry *entry;
1127         struct lustre_cfg       *lcfg;
1128         struct lustre_cfg_bufs   bufs;
1129         u64   prev_version = 0;
1130         char *inst;
1131         char *buf;
1132         int   bufsz;
1133         int   pos;
1134         int   rc  = 0;
1135         int   off = 0;
1136
1137         LASSERT(cfg->cfg_instance);
1138         LASSERT(cfg->cfg_sb == cfg->cfg_instance);
1139
1140         inst = kzalloc(PAGE_SIZE, GFP_KERNEL);
1141         if (!inst)
1142                 return -ENOMEM;
1143
1144         pos = snprintf(inst, PAGE_SIZE, "%p", cfg->cfg_instance);
1145         if (pos >= PAGE_SIZE) {
1146                 kfree(inst);
1147                 return -E2BIG;
1148         }
1149
1150         ++pos;
1151         buf   = inst + pos;
1152         bufsz = PAGE_SIZE - pos;
1153
1154         while (datalen > 0) {
1155                 int   entry_len = sizeof(*entry);
1156                 int   is_ost;
1157                 struct obd_device *obd;
1158                 char *obdname;
1159                 char *cname;
1160                 char *params;
1161                 char *uuid;
1162
1163                 rc = -EINVAL;
1164                 if (datalen < sizeof(*entry))
1165                         break;
1166
1167                 entry = (typeof(entry))(data + off);
1168
1169                 /* sanity check */
1170                 if (entry->mne_nid_type != 0) /* only support type 0 for ipv4 */
1171                         break;
1172                 if (entry->mne_nid_count == 0) /* at least one nid entry */
1173                         break;
1174                 if (entry->mne_nid_size != sizeof(lnet_nid_t))
1175                         break;
1176
1177                 entry_len += entry->mne_nid_count * entry->mne_nid_size;
1178                 if (datalen < entry_len) /* must have entry_len at least */
1179                         break;
1180
1181                 /* Keep this swab for normal mixed endian handling. LU-1644 */
1182                 if (mne_swab)
1183                         lustre_swab_mgs_nidtbl_entry(entry);
1184                 if (entry->mne_length > PAGE_SIZE) {
1185                         CERROR("MNE too large (%u)\n", entry->mne_length);
1186                         break;
1187                 }
1188
1189                 if (entry->mne_length < entry_len)
1190                         break;
1191
1192                 off     += entry->mne_length;
1193                 datalen -= entry->mne_length;
1194                 if (datalen < 0)
1195                         break;
1196
1197                 if (entry->mne_version > max_version) {
1198                         CERROR("entry index(%lld) is over max_index(%lld)\n",
1199                                entry->mne_version, max_version);
1200                         break;
1201                 }
1202
1203                 if (prev_version >= entry->mne_version) {
1204                         CERROR("index unsorted, prev %lld, now %lld\n",
1205                                prev_version, entry->mne_version);
1206                         break;
1207                 }
1208                 prev_version = entry->mne_version;
1209
1210                 /*
1211                  * Write a string with format "nid::instance" to
1212                  * lustre/<osc|mdc>/<target>-<osc|mdc>-<instance>/import.
1213                  */
1214
1215                 is_ost = entry->mne_type == LDD_F_SV_TYPE_OST;
1216                 memset(buf, 0, bufsz);
1217                 obdname = buf;
1218                 pos = 0;
1219
1220                 /* lustre-OST0001-osc-<instance #> */
1221                 strcpy(obdname, cld->cld_logname);
1222                 cname = strrchr(obdname, '-');
1223                 if (!cname) {
1224                         CERROR("mgc %s: invalid logname %s\n",
1225                                mgc->obd_name, obdname);
1226                         break;
1227                 }
1228
1229                 pos = cname - obdname;
1230                 obdname[pos] = 0;
1231                 pos += sprintf(obdname + pos, "-%s%04x",
1232                                   is_ost ? "OST" : "MDT", entry->mne_index);
1233
1234                 cname = is_ost ? "osc" : "mdc";
1235                 pos += sprintf(obdname + pos, "-%s-%s", cname, inst);
1236                 lustre_cfg_bufs_reset(&bufs, obdname);
1237
1238                 /* find the obd by obdname */
1239                 obd = class_name2obd(obdname);
1240                 if (!obd) {
1241                         CDEBUG(D_INFO, "mgc %s: cannot find obdname %s\n",
1242                                mgc->obd_name, obdname);
1243                         rc = 0;
1244                         /* this is a safe race, when the ost is starting up...*/
1245                         continue;
1246                 }
1247
1248                 /* osc.import = "connection=<Conn UUID>::<target instance>" */
1249                 ++pos;
1250                 params = buf + pos;
1251                 pos += sprintf(params, "%s.import=%s", cname, "connection=");
1252                 uuid = buf + pos;
1253
1254                 down_read(&obd->u.cli.cl_sem);
1255                 if (!obd->u.cli.cl_import) {
1256                         /* client does not connect to the OST yet */
1257                         up_read(&obd->u.cli.cl_sem);
1258                         rc = 0;
1259                         continue;
1260                 }
1261
1262                 /* TODO: iterate all nids to find one */
1263                 /* find uuid by nid */
1264                 rc = client_import_find_conn(obd->u.cli.cl_import,
1265                                              entry->u.nids[0],
1266                                              (struct obd_uuid *)uuid);
1267                 up_read(&obd->u.cli.cl_sem);
1268                 if (rc < 0) {
1269                         CERROR("mgc: cannot find uuid by nid %s\n",
1270                                libcfs_nid2str(entry->u.nids[0]));
1271                         break;
1272                 }
1273
1274                 CDEBUG(D_INFO, "Find uuid %s by nid %s\n",
1275                        uuid, libcfs_nid2str(entry->u.nids[0]));
1276
1277                 pos += strlen(uuid);
1278                 pos += sprintf(buf + pos, "::%u", entry->mne_instance);
1279                 LASSERT(pos < bufsz);
1280
1281                 lustre_cfg_bufs_set_string(&bufs, 1, params);
1282
1283                 rc = -ENOMEM;
1284                 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
1285                 if (IS_ERR(lcfg)) {
1286                         CERROR("mgc: cannot allocate memory\n");
1287                         break;
1288                 }
1289
1290                 CDEBUG(D_INFO, "ir apply logs %lld/%lld for %s -> %s\n",
1291                        prev_version, max_version, obdname, params);
1292
1293                 rc = class_process_config(lcfg);
1294                 lustre_cfg_free(lcfg);
1295                 if (rc)
1296                         CDEBUG(D_INFO, "process config for %s error %d\n",
1297                                obdname, rc);
1298
1299                 /* continue, even one with error */
1300         }
1301
1302         kfree(inst);
1303         return rc;
1304 }
1305
1306 /**
1307  * This function is called if this client was notified for target restarting
1308  * by the MGS. A CONFIG_READ RPC is going to send to fetch recovery logs.
1309  */
1310 static int mgc_process_recover_log(struct obd_device *obd,
1311                                    struct config_llog_data *cld)
1312 {
1313         struct ptlrpc_request *req = NULL;
1314         struct config_llog_instance *cfg = &cld->cld_cfg;
1315         struct mgs_config_body *body;
1316         struct mgs_config_res  *res;
1317         struct ptlrpc_bulk_desc *desc;
1318         struct page **pages;
1319         int nrpages;
1320         bool eof = true;
1321         bool mne_swab;
1322         int i;
1323         int ealen;
1324         int rc;
1325
1326         /* allocate buffer for bulk transfer.
1327          * if this is the first time for this mgs to read logs,
1328          * CONFIG_READ_NRPAGES_INIT will be used since it will read all logs
1329          * once; otherwise, it only reads increment of logs, this should be
1330          * small and CONFIG_READ_NRPAGES will be used.
1331          */
1332         nrpages = CONFIG_READ_NRPAGES;
1333         if (cfg->cfg_last_idx == 0) /* the first time */
1334                 nrpages = CONFIG_READ_NRPAGES_INIT;
1335
1336         pages = kcalloc(nrpages, sizeof(*pages), GFP_KERNEL);
1337         if (!pages) {
1338                 rc = -ENOMEM;
1339                 goto out;
1340         }
1341
1342         for (i = 0; i < nrpages; i++) {
1343                 pages[i] = alloc_page(GFP_KERNEL);
1344                 if (!pages[i]) {
1345                         rc = -ENOMEM;
1346                         goto out;
1347                 }
1348         }
1349
1350 again:
1351         LASSERT(cld_is_recover(cld));
1352         LASSERT(mutex_is_locked(&cld->cld_lock));
1353         req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp),
1354                                    &RQF_MGS_CONFIG_READ);
1355         if (!req) {
1356                 rc = -ENOMEM;
1357                 goto out;
1358         }
1359
1360         rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ);
1361         if (rc)
1362                 goto out;
1363
1364         /* pack request */
1365         body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
1366         LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname));
1367         if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name))
1368             >= sizeof(body->mcb_name)) {
1369                 rc = -E2BIG;
1370                 goto out;
1371         }
1372         body->mcb_offset = cfg->cfg_last_idx + 1;
1373         body->mcb_type   = cld->cld_type;
1374         body->mcb_bits   = PAGE_SHIFT;
1375         body->mcb_units  = nrpages;
1376
1377         /* allocate bulk transfer descriptor */
1378         desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, BULK_PUT_SINK,
1379                                     MGS_BULK_PORTAL);
1380         if (!desc) {
1381                 rc = -ENOMEM;
1382                 goto out;
1383         }
1384
1385         for (i = 0; i < nrpages; i++)
1386                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE);
1387
1388         ptlrpc_request_set_replen(req);
1389         rc = ptlrpc_queue_wait(req);
1390         if (rc)
1391                 goto out;
1392
1393         res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
1394         if (res->mcr_size < res->mcr_offset) {
1395                 rc = -EINVAL;
1396                 goto out;
1397         }
1398
1399         /* always update the index even though it might have errors with
1400          * handling the recover logs
1401          */
1402         cfg->cfg_last_idx = res->mcr_offset;
1403         eof = res->mcr_offset == res->mcr_size;
1404
1405         CDEBUG(D_INFO, "Latest version %lld, more %d.\n",
1406                res->mcr_offset, eof == false);
1407
1408         ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0);
1409         if (ealen < 0) {
1410                 rc = ealen;
1411                 goto out;
1412         }
1413
1414         if (ealen > nrpages << PAGE_SHIFT) {
1415                 rc = -EINVAL;
1416                 goto out;
1417         }
1418
1419         if (ealen == 0) { /* no logs transferred */
1420                 if (!eof)
1421                         rc = -EINVAL;
1422                 goto out;
1423         }
1424
1425         mne_swab = !!ptlrpc_rep_need_swab(req);
1426 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
1427         /* This import flag means the server did an extra swab of IR MNE
1428          * records (fixed in LU-1252), reverse it here if needed. LU-1644
1429          */
1430         if (unlikely(req->rq_import->imp_need_mne_swab))
1431                 mne_swab = !mne_swab;
1432 #else
1433 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
1434 #endif
1435
1436         for (i = 0; i < nrpages && ealen > 0; i++) {
1437                 int rc2;
1438                 void *ptr;
1439
1440                 ptr = kmap(pages[i]);
1441                 rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr,
1442                                              min_t(int, ealen, PAGE_SIZE),
1443                                              mne_swab);
1444                 kunmap(pages[i]);
1445                 if (rc2 < 0) {
1446                         CWARN("Process recover log %s error %d\n",
1447                               cld->cld_logname, rc2);
1448                         break;
1449                 }
1450
1451                 ealen -= PAGE_SIZE;
1452         }
1453
1454 out:
1455         if (req)
1456                 ptlrpc_req_finished(req);
1457
1458         if (rc == 0 && !eof)
1459                 goto again;
1460
1461         if (pages) {
1462                 for (i = 0; i < nrpages; i++) {
1463                         if (!pages[i])
1464                                 break;
1465                         __free_page(pages[i]);
1466                 }
1467                 kfree(pages);
1468         }
1469         return rc;
1470 }
1471
1472 /* local_only means it cannot get remote llogs */
1473 static int mgc_process_cfg_log(struct obd_device *mgc,
1474                                struct config_llog_data *cld, int local_only)
1475 {
1476         struct llog_ctxt        *ctxt;
1477         struct lustre_sb_info   *lsi = NULL;
1478         int                      rc = 0;
1479         bool                     sptlrpc_started = false;
1480         struct lu_env           *env;
1481
1482         LASSERT(cld);
1483         LASSERT(mutex_is_locked(&cld->cld_lock));
1484
1485         /*
1486          * local copy of sptlrpc log is controlled elsewhere, don't try to
1487          * read it up here.
1488          */
1489         if (cld_is_sptlrpc(cld) && local_only)
1490                 return 0;
1491
1492         if (cld->cld_cfg.cfg_sb)
1493                 lsi = s2lsi(cld->cld_cfg.cfg_sb);
1494
1495         env = kzalloc(sizeof(*env), GFP_KERNEL);
1496         if (!env)
1497                 return -ENOMEM;
1498
1499         rc = lu_env_init(env, LCT_MG_THREAD);
1500         if (rc)
1501                 goto out_free;
1502
1503         ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
1504         LASSERT(ctxt);
1505
1506         if (local_only) /* no local log at client side */ {
1507                 rc = -EIO;
1508                 goto out_pop;
1509         }
1510
1511         if (cld_is_sptlrpc(cld)) {
1512                 sptlrpc_conf_log_update_begin(cld->cld_logname);
1513                 sptlrpc_started = true;
1514         }
1515
1516         /* logname and instance info should be the same, so use our
1517          * copy of the instance for the update.  The cfg_last_idx will
1518          * be updated here.
1519          */
1520         rc = class_config_parse_llog(env, ctxt, cld->cld_logname,
1521                                      &cld->cld_cfg);
1522
1523 out_pop:
1524         __llog_ctxt_put(env, ctxt);
1525
1526         /*
1527          * update settings on existing OBDs. doing it inside
1528          * of llog_process_lock so no device is attaching/detaching
1529          * in parallel.
1530          * the logname must be <fsname>-sptlrpc
1531          */
1532         if (sptlrpc_started) {
1533                 LASSERT(cld_is_sptlrpc(cld));
1534                 sptlrpc_conf_log_update_end(cld->cld_logname);
1535                 class_notify_sptlrpc_conf(cld->cld_logname,
1536                                           strlen(cld->cld_logname) -
1537                                           strlen("-sptlrpc"));
1538         }
1539
1540         lu_env_fini(env);
1541 out_free:
1542         kfree(env);
1543         return rc;
1544 }
1545
1546 /** Get a config log from the MGS and process it.
1547  * This func is called for both clients and servers.
1548  * Copy the log locally before parsing it if appropriate (non-MGS server)
1549  */
1550 int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld)
1551 {
1552         struct lustre_handle lockh = { 0 };
1553         __u64 flags = LDLM_FL_NO_LRU;
1554         int rc = 0, rcl;
1555
1556         LASSERT(cld);
1557
1558         /* I don't want multiple processes running process_log at once --
1559          * sounds like badness.  It actually might be fine, as long as
1560          * we're not trying to update from the same log
1561          * simultaneously (in which case we should use a per-log sem.)
1562          */
1563         mutex_lock(&cld->cld_lock);
1564         if (cld->cld_stopping) {
1565                 mutex_unlock(&cld->cld_lock);
1566                 return 0;
1567         }
1568
1569         OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PAUSE_PROCESS_LOG, 20);
1570
1571         CDEBUG(D_MGC, "Process log %s:%p from %d\n", cld->cld_logname,
1572                cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1);
1573
1574         /* Get the cfg lock on the llog */
1575         rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL,
1576                           LCK_CR, &flags, NULL, NULL, NULL,
1577                           cld, 0, NULL, &lockh);
1578         if (rcl == 0) {
1579                 /* Get the cld, it will be released in mgc_blocking_ast. */
1580                 config_log_get(cld);
1581                 rc = ldlm_lock_set_data(&lockh, (void *)cld);
1582                 LASSERT(rc == 0);
1583         } else {
1584                 CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl);
1585
1586                 /* mark cld_lostlock so that it will requeue
1587                  * after MGC becomes available.
1588                  */
1589                 cld->cld_lostlock = 1;
1590                 /* Get extra reference, it will be put in requeue thread */
1591                 config_log_get(cld);
1592         }
1593
1594         if (cld_is_recover(cld)) {
1595                 rc = 0; /* this is not a fatal error for recover log */
1596                 if (rcl == 0)
1597                         rc = mgc_process_recover_log(mgc, cld);
1598         } else {
1599                 rc = mgc_process_cfg_log(mgc, cld, rcl != 0);
1600         }
1601
1602         CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n",
1603                mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc);
1604
1605         mutex_unlock(&cld->cld_lock);
1606
1607         /* Now drop the lock so MGS can revoke it */
1608         if (!rcl)
1609                 ldlm_lock_decref(&lockh, LCK_CR);
1610
1611         return rc;
1612 }
1613
1614 /** Called from lustre_process_log.
1615  * LCFG_LOG_START gets the config log from the MGS, processes it to start
1616  * any services, and adds it to the list logs to watch (follow).
1617  */
1618 static int mgc_process_config(struct obd_device *obd, u32 len, void *buf)
1619 {
1620         struct lustre_cfg *lcfg = buf;
1621         struct config_llog_instance *cfg = NULL;
1622         char *logname;
1623         int rc = 0;
1624
1625         switch (lcfg->lcfg_command) {
1626         case LCFG_LOV_ADD_OBD: {
1627                 /* Overloading this cfg command: register a new target */
1628                 struct mgs_target_info *mti;
1629
1630                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) !=
1631                     sizeof(struct mgs_target_info)) {
1632                         rc = -EINVAL;
1633                         goto out;
1634                 }
1635
1636                 mti = (struct mgs_target_info *)lustre_cfg_buf(lcfg, 1);
1637                 CDEBUG(D_MGC, "add_target %s %#x\n",
1638                        mti->mti_svname, mti->mti_flags);
1639                 rc = mgc_target_register(obd->u.cli.cl_mgc_mgsexp, mti);
1640                 break;
1641         }
1642         case LCFG_LOV_DEL_OBD:
1643                 /* Unregister has no meaning at the moment. */
1644                 CERROR("lov_del_obd unimplemented\n");
1645                 rc = -ENOSYS;
1646                 break;
1647         case LCFG_SPTLRPC_CONF: {
1648                 rc = sptlrpc_process_config(lcfg);
1649                 break;
1650         }
1651         case LCFG_LOG_START: {
1652                 struct config_llog_data *cld;
1653                 struct super_block *sb;
1654
1655                 logname = lustre_cfg_string(lcfg, 1);
1656                 cfg = (struct config_llog_instance *)lustre_cfg_buf(lcfg, 2);
1657                 sb = *(struct super_block **)lustre_cfg_buf(lcfg, 3);
1658
1659                 CDEBUG(D_MGC, "parse_log %s from %d\n", logname,
1660                        cfg->cfg_last_idx);
1661
1662                 /* We're only called through here on the initial mount */
1663                 rc = config_log_add(obd, logname, cfg, sb);
1664                 if (rc)
1665                         break;
1666                 cld = config_log_find(logname, cfg);
1667                 if (!cld) {
1668                         rc = -ENOENT;
1669                         break;
1670                 }
1671
1672                 /* COMPAT_146 */
1673                 /* FIXME only set this for old logs!  Right now this forces
1674                  * us to always skip the "inside markers" check
1675                  */
1676                 cld->cld_cfg.cfg_flags |= CFG_F_COMPAT146;
1677
1678                 rc = mgc_process_log(obd, cld);
1679                 if (rc == 0 && cld->cld_recover) {
1680                         if (OCD_HAS_FLAG(&obd->u.cli.cl_import->
1681                                          imp_connect_data, IMP_RECOV)) {
1682                                 rc = mgc_process_log(obd, cld->cld_recover);
1683                         } else {
1684                                 struct config_llog_data *cir = cld->cld_recover;
1685
1686                                 cld->cld_recover = NULL;
1687                                 config_log_put(cir);
1688                         }
1689                         if (rc)
1690                                 CERROR("Cannot process recover llog %d\n", rc);
1691                 }
1692
1693                 if (rc == 0 && cld->cld_params) {
1694                         rc = mgc_process_log(obd, cld->cld_params);
1695                         if (rc == -ENOENT) {
1696                                 CDEBUG(D_MGC,
1697                                        "There is no params config file yet\n");
1698                                 rc = 0;
1699                         }
1700                         /* params log is optional */
1701                         if (rc)
1702                                 CERROR(
1703                                        "%s: can't process params llog: rc = %d\n",
1704                                        obd->obd_name, rc);
1705                 }
1706                 config_log_put(cld);
1707
1708                 break;
1709         }
1710         case LCFG_LOG_END: {
1711                 logname = lustre_cfg_string(lcfg, 1);
1712
1713                 if (lcfg->lcfg_bufcount >= 2)
1714                         cfg = (struct config_llog_instance *)lustre_cfg_buf(
1715                                 lcfg, 2);
1716                 rc = config_log_end(logname, cfg);
1717                 break;
1718         }
1719         default: {
1720                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1721                 rc = -EINVAL;
1722                 goto out;
1723
1724         }
1725         }
1726 out:
1727         return rc;
1728 }
1729
1730 static struct obd_ops mgc_obd_ops = {
1731         .owner          = THIS_MODULE,
1732         .setup          = mgc_setup,
1733         .precleanup     = mgc_precleanup,
1734         .cleanup        = mgc_cleanup,
1735         .add_conn       = client_import_add_conn,
1736         .del_conn       = client_import_del_conn,
1737         .connect        = client_connect_import,
1738         .disconnect     = client_disconnect_export,
1739         /* .enqueue     = mgc_enqueue, */
1740         /* .iocontrol   = mgc_iocontrol, */
1741         .set_info_async = mgc_set_info_async,
1742         .get_info       = mgc_get_info,
1743         .import_event   = mgc_import_event,
1744         .process_config = mgc_process_config,
1745 };
1746
1747 static int __init mgc_init(void)
1748 {
1749         return class_register_type(&mgc_obd_ops, NULL,
1750                                    LUSTRE_MGC_NAME, NULL);
1751 }
1752
1753 static void /*__exit*/ mgc_exit(void)
1754 {
1755         class_unregister_type(LUSTRE_MGC_NAME);
1756 }
1757
1758 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1759 MODULE_DESCRIPTION("Lustre Management Client");
1760 MODULE_VERSION(LUSTRE_VERSION_STRING);
1761 MODULE_LICENSE("GPL");
1762
1763 module_init(mgc_init);
1764 module_exit(mgc_exit);