lustre/ptlrpc/recov_thread.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2003 Cluster File Systems, Inc.
   5  *   Author: Andreas Dilger <adilger@clusterfs.com>
   6  *
   7  *   This file is part of the Lustre file system, http://www.lustre.org
   8  *   Lustre is a trademark of Cluster File Systems, Inc.
   9  *
  10  *   You may have signed or agreed to another license before downloading
  11  *   this software.  If so, you are bound by the terms and conditions
  12  *   of that agreement, and the following does not apply to you.  See the
  13  *   LICENSE file included with this distribution for more information.
  14  *
  15  *   If you did not agree to a different license, then this copy of Lustre
  16  *   is open source software; you can redistribute it and/or modify it
  17  *   under the terms of version 2 of the GNU General Public License as
  18  *   published by the Free Software Foundation.
  19  *
  20  *   In either case, Lustre is distributed in the hope that it will be
  21  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  22  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23  *   license text for more details.
  24  *
  25  * OST<->MDS recovery logging thread.
  26  *
  27  * Invariants in implementation:
  28  * - we do not share logs among different OST<->MDS connections, so that
  29  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LOG
  33
  34 #ifndef EXPORT_SYMTAB
  35 # define EXPORT_SYMTAB
  36 #endif
  37
  38 #ifdef __KERNEL__
  39 # include <libcfs/libcfs.h>
  40 #else
  41 # include <libcfs/list.h>
  42 # include <liblustre.h>
  43 #endif
  44
  45 #include <libcfs/kp30.h>
  46 #include <obd_class.h>
  47 #include <lustre_commit_confd.h>
  48 #include <obd_support.h>
  49 #include <obd_class.h>
  50 #include <lustre_net.h>
  51 #include <lnet/types.h>
  52 #include <libcfs/list.h>
  53 #include <lustre_log.h>
  54 #include "ptlrpc_internal.h"
  55
  56 #ifdef __KERNEL__
  57
  58 static struct llog_commit_master lustre_lcm;
  59 static struct llog_commit_master *lcm = &lustre_lcm;
  60
  61 /* Allocate new commit structs in case we do not have enough.
  62  * Make the llcd size small enough that it fits into a single page when we
  63  * are sending/receiving it. */
  64 static int llcd_alloc(void)
  65 {
  66         struct llog_canceld_ctxt *llcd;
  67         int llcd_size;
  68
  69         /* payload of lustre_msg V2 is bigger */
  70         llcd_size = 4096 - lustre_msg_size(LUSTRE_MSG_MAGIC_V2, 1, NULL);
  71         OBD_ALLOC(llcd,
  72                   llcd_size + offsetof(struct llog_canceld_ctxt, llcd_cookies));
  73         if (llcd == NULL)
  74                 return -ENOMEM;
  75
  76         llcd->llcd_size = llcd_size;
  77         llcd->llcd_lcm = lcm;
  78
  79         spin_lock(&lcm->lcm_llcd_lock);
  80         list_add(&llcd->llcd_list, &lcm->lcm_llcd_free);
  81         atomic_inc(&lcm->lcm_llcd_numfree);
  82         spin_unlock(&lcm->lcm_llcd_lock);
  83
  84         return 0;
  85 }
  86
  87 /* Get a free cookie struct from the list */
  88 struct llog_canceld_ctxt *llcd_grab(void)
  89 {
  90         struct llog_canceld_ctxt *llcd;
  91
  92 repeat:
  93         spin_lock(&lcm->lcm_llcd_lock);
  94         if (list_empty(&lcm->lcm_llcd_free)) {
  95                 spin_unlock(&lcm->lcm_llcd_lock);
  96                 if (llcd_alloc() < 0) {
  97                         CERROR("unable to allocate log commit data!\n");
  98                         return NULL;
  99                 }
 100                 /* check new llcd wasn't grabbed while lock dropped, b=7407 */
 101                 goto repeat;
 102         }
 103
 104         llcd = list_entry(lcm->lcm_llcd_free.next, typeof(*llcd), llcd_list);
 105         list_del(&llcd->llcd_list);
 106         atomic_dec(&lcm->lcm_llcd_numfree);
 107         spin_unlock(&lcm->lcm_llcd_lock);
 108
 109         llcd->llcd_cookiebytes = 0;
 110
 111         return llcd;
 112 }
 113 EXPORT_SYMBOL(llcd_grab);
 114
 115 static void llcd_put(struct llog_canceld_ctxt *llcd)
 116 {
 117         if (atomic_read(&lcm->lcm_llcd_numfree) >= lcm->lcm_llcd_maxfree) {
 118                 int llcd_size = llcd->llcd_size +
 119                          offsetof(struct llog_canceld_ctxt, llcd_cookies);
 120                 OBD_FREE(llcd, llcd_size);
 121         } else {
 122                 spin_lock(&lcm->lcm_llcd_lock);
 123                 list_add(&llcd->llcd_list, &lcm->lcm_llcd_free);
 124                 atomic_inc(&lcm->lcm_llcd_numfree);
 125                 spin_unlock(&lcm->lcm_llcd_lock);
 126         }
 127 }
 128
 129 /* Send some cookies to the appropriate target */
 130 void llcd_send(struct llog_canceld_ctxt *llcd)
 131 {
 132         spin_lock(&llcd->llcd_lcm->lcm_llcd_lock);
 133         list_add_tail(&llcd->llcd_list, &llcd->llcd_lcm->lcm_llcd_pending);
 134         spin_unlock(&llcd->llcd_lcm->lcm_llcd_lock);
 135
 136         cfs_waitq_signal_nr(&llcd->llcd_lcm->lcm_waitq, 1);
 137 }
 138 EXPORT_SYMBOL(llcd_send);
 139
 140 /* deleted objects have a commit callback that cancels the MDS
 141  * log record for the deletion.  The commit callback calls this
 142  * function
 143  */
 144 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
 145                          struct lov_stripe_md *lsm, int count,
 146                          struct llog_cookie *cookies, int flags)
 147 {
 148         struct llog_canceld_ctxt *llcd;
 149         int rc = 0;
 150         ENTRY;
 151
 152         LASSERT(ctxt);
 153
 154         mutex_down(&ctxt->loc_sem);
 155         if (ctxt->loc_imp == NULL) {
 156                 CDEBUG(D_RPCTRACE, "no import for ctxt %p\n", ctxt);
 157                 GOTO(out, rc = 0);
 158         }
 159
 160         llcd = ctxt->loc_llcd;
 161
 162         if (count > 0 && cookies != NULL) {
 163                 if (llcd == NULL) {
 164                         llcd = llcd_grab();
 165                         if (llcd == NULL) {
 166                                 CERROR("couldn't get an llcd - dropped "LPX64
 167                                        ":%x+%u\n",
 168                                        cookies->lgc_lgl.lgl_oid,
 169                                        cookies->lgc_lgl.lgl_ogen,
 170                                        cookies->lgc_index);
 171                                 GOTO(out, rc = -ENOMEM);
 172                         }
 173                         llcd->llcd_ctxt = ctxt;
 174                         ctxt->loc_llcd = llcd;
 175                 }
 176
 177                 memcpy((char *)llcd->llcd_cookies + llcd->llcd_cookiebytes,
 178                        cookies, sizeof(*cookies));
 179                 llcd->llcd_cookiebytes += sizeof(*cookies);
 180         } else {
 181                 if (llcd == NULL || !(flags & OBD_LLOG_FL_SENDNOW))
 182                         GOTO(out, rc);
 183         }
 184
 185         if ((llcd->llcd_size - llcd->llcd_cookiebytes) < sizeof(*cookies) ||
 186             (flags & OBD_LLOG_FL_SENDNOW)) {
 187                 CDEBUG(D_RPCTRACE, "send llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 188                 ctxt->loc_llcd = NULL;
 189                 llcd_send(llcd);
 190         }
 191 out:
 192         mutex_up(&ctxt->loc_sem);
 193         return rc;
 194 }
 195 EXPORT_SYMBOL(llog_obd_repl_cancel);
 196
 197 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp)
 198 {
 199         int rc = 0;
 200         ENTRY;
 201
 202         if (exp && (ctxt->loc_imp == exp->exp_imp_reverse)) {
 203                 CDEBUG(D_RPCTRACE,"reverse import disconnect, put llcd %p:%p\n",
 204                        ctxt->loc_llcd, ctxt);
 205                 mutex_down(&ctxt->loc_sem);
 206                 if (ctxt->loc_llcd != NULL) {
 207                         llcd_put(ctxt->loc_llcd);
 208                         ctxt->loc_llcd = NULL;
 209                 }
 210                 ctxt->loc_imp = NULL;
 211                 mutex_up(&ctxt->loc_sem);
 212         } else {
 213                 rc = llog_cancel(ctxt, NULL, 0, NULL, OBD_LLOG_FL_SENDNOW);
 214         }
 215
 216         RETURN(rc);
 217 }
 218 EXPORT_SYMBOL(llog_obd_repl_sync);
 219
 220 static int log_commit_thread(void *arg)
 221 {
 222         struct llog_commit_master *lcm = arg;
 223         struct llog_commit_daemon *lcd;
 224         struct llog_canceld_ctxt *llcd, *n;
 225         struct obd_import *import = NULL;
 226         ENTRY;
 227
 228         OBD_ALLOC(lcd, sizeof(*lcd));
 229         if (lcd == NULL)
 230                 RETURN(-ENOMEM);
 231
 232         spin_lock(&lcm->lcm_thread_lock);
 233         THREAD_NAME(cfs_curproc_comm(), CFS_CURPROC_COMM_MAX - 1,
 234                     "ll_log_comt_%02d", atomic_read(&lcm->lcm_thread_total));
 235         atomic_inc(&lcm->lcm_thread_total);
 236         spin_unlock(&lcm->lcm_thread_lock);
 237
 238         ptlrpc_daemonize(cfs_curproc_comm()); /* thread never needs to do IO */
 239
 240         CFS_INIT_LIST_HEAD(&lcd->lcd_lcm_list);
 241         CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
 242         lcd->lcd_lcm = lcm;
 243
 244         CDEBUG(D_HA, "%s started\n", cfs_curproc_comm());
 245         do {
 246                 struct ptlrpc_request *request;
 247                 struct list_head *sending_list;
 248                 int rc = 0;
 249
 250                 if (import)
 251                         class_import_put(import);
 252                 import = NULL;
 253
 254                 /* If we do not have enough pages available, allocate some */
 255                 while (atomic_read(&lcm->lcm_llcd_numfree) <
 256                        lcm->lcm_llcd_minfree) {
 257                         if (llcd_alloc() < 0)
 258                                 break;
 259                 }
 260
 261                 spin_lock(&lcm->lcm_thread_lock);
 262                 atomic_inc(&lcm->lcm_thread_numidle);
 263                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_idle);
 264                 spin_unlock(&lcm->lcm_thread_lock);
 265
 266                 wait_event_interruptible(lcm->lcm_waitq,
 267                                          !list_empty(&lcm->lcm_llcd_pending) ||
 268                                          lcm->lcm_flags & LLOG_LCM_FL_EXIT);
 269
 270                 /* If we are the last available thread, start a new one in case
 271                  * we get blocked on an RPC (nobody else will start a new one)*/
 272                 spin_lock(&lcm->lcm_thread_lock);
 273                 atomic_dec(&lcm->lcm_thread_numidle);
 274                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_busy);
 275                 spin_unlock(&lcm->lcm_thread_lock);
 276
 277                 sending_list = &lcm->lcm_llcd_pending;
 278         resend:
 279                 if (import)
 280                         class_import_put(import);
 281                 import = NULL;
 282                 if (lcm->lcm_flags & LLOG_LCM_FL_EXIT) {
 283                         lcm->lcm_llcd_maxfree = 0;
 284                         lcm->lcm_llcd_minfree = 0;
 285                         lcm->lcm_thread_max = 0;
 286
 287                         if (list_empty(&lcm->lcm_llcd_pending) ||
 288                             lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE)
 289                                 break;
 290                 }
 291
 292                 if (atomic_read(&lcm->lcm_thread_numidle) <= 1 &&
 293                     atomic_read(&lcm->lcm_thread_total) < lcm->lcm_thread_max) {
 294                         rc = llog_start_commit_thread();
 295                         if (rc < 0)
 296                                 CERROR("error starting thread: rc %d\n", rc);
 297                 }
 298
 299                 /* Move all of the pending cancels from the same OST off of
 300                  * the list, so we don't get multiple threads blocked and/or
 301                  * doing upcalls on the same OST in case of failure. */
 302                 spin_lock(&lcm->lcm_llcd_lock);
 303                 if (!list_empty(sending_list)) {
 304                         list_move_tail(sending_list->next,
 305                                        &lcd->lcd_llcd_list);
 306                         llcd = list_entry(lcd->lcd_llcd_list.next,
 307                                           typeof(*llcd), llcd_list);
 308                         LASSERT(llcd->llcd_lcm == lcm);
 309                         import = llcd->llcd_ctxt->loc_imp;
 310                         if (import)
 311                                 class_import_get(import);
 312                 }
 313                 list_for_each_entry_safe(llcd, n, sending_list, llcd_list) {
 314                         LASSERT(llcd->llcd_lcm == lcm);
 315                         if (import == llcd->llcd_ctxt->loc_imp)
 316                                 list_move_tail(&llcd->llcd_list,
 317                                                &lcd->lcd_llcd_list);
 318                 }
 319                 if (sending_list != &lcm->lcm_llcd_resend) {
 320                         list_for_each_entry_safe(llcd, n, &lcm->lcm_llcd_resend,
 321                                                  llcd_list) {
 322                                 LASSERT(llcd->llcd_lcm == lcm);
 323                                 if (import == llcd->llcd_ctxt->loc_imp)
 324                                         list_move_tail(&llcd->llcd_list,
 325                                                        &lcd->lcd_llcd_list);
 326                         }
 327                 }
 328                 spin_unlock(&lcm->lcm_llcd_lock);
 329
 330                 /* We are the only one manipulating our local list - no lock */
 331                 list_for_each_entry_safe(llcd,n, &lcd->lcd_llcd_list,llcd_list){
 332                         int size[2] = { sizeof(struct ptlrpc_body),
 333                                         llcd->llcd_cookiebytes };
 334                         char *bufs[2] = { NULL, (char *)llcd->llcd_cookies };
 335
 336                         list_del(&llcd->llcd_list);
 337                         if (llcd->llcd_cookiebytes == 0) {
 338                                 CDEBUG(D_RPCTRACE, "put empty llcd %p:%p\n",
 339                                        llcd, llcd->llcd_ctxt);
 340                                 llcd_put(llcd);
 341                                 continue;
 342                         }
 343
 344                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 345                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 346                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 347                                 CWARN("import will be destroyed, put "
 348                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 349                                 llcd_put(llcd);
 350                                 continue;
 351                         }
 352                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 353
 354                         if (!import || (import == LP_POISON) ||
 355                             (import->imp_client == LP_POISON)) {
 356                                 CERROR("No import %p (llcd=%p, ctxt=%p)\n",
 357                                        import, llcd, llcd->llcd_ctxt);
 358                                 llcd_put(llcd);
 359                                 continue;
 360                         }
 361
 362                         OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_RECOV, 10);
 363
 364                         request = ptlrpc_prep_req(import, LUSTRE_LOG_VERSION,
 365                                                   OBD_LOG_CANCEL, 2, size,bufs);
 366                         if (request == NULL) {
 367                                 rc = -ENOMEM;
 368                                 CERROR("error preparing commit: rc %d\n", rc);
 369
 370                                 spin_lock(&lcm->lcm_llcd_lock);
 371                                 list_splice(&lcd->lcd_llcd_list,
 372                                             &lcm->lcm_llcd_resend);
 373                                 CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
 374                                 spin_unlock(&lcm->lcm_llcd_lock);
 375                                 break;
 376                         }
 377
 378                         /* XXX FIXME bug 249, 5515 */
 379                         request->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
 380                         request->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
 381
 382                         ptlrpc_req_set_repsize(request, 1, NULL);
 383                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 384                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 385                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 386                                 CWARN("import will be destroyed, put "
 387                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 388                                 llcd_put(llcd);
 389                                 ptlrpc_req_finished(request);
 390                                 continue;
 391                         }
 392                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 393                         rc = ptlrpc_queue_wait(request);
 394                         ptlrpc_req_finished(request);
 395
 396                         /* If the RPC failed, we put this and the remaining
 397                          * messages onto the resend list for another time. */
 398                         if (rc == 0) {
 399                                 llcd_put(llcd);
 400                                 continue;
 401                         }
 402
 403                         CERROR("commit %p:%p drop %d cookies: rc %d\n",
 404                                llcd, llcd->llcd_ctxt,
 405                                (int)(llcd->llcd_cookiebytes /
 406                                      sizeof(*llcd->llcd_cookies)), rc);
 407                         llcd_put(llcd);
 408                 }
 409
 410                 if (rc == 0) {
 411                         sending_list = &lcm->lcm_llcd_resend;
 412                         if (!list_empty(sending_list))
 413                                 goto resend;
 414                 }
 415         } while(1);
 416
 417         if (import)
 418                 class_import_put(import);
 419
 420         /* If we are force exiting, just drop all of the cookies. */
 421         if (lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE) {
 422                 spin_lock(&lcm->lcm_llcd_lock);
 423                 list_splice(&lcm->lcm_llcd_pending, &lcd->lcd_llcd_list);
 424                 list_splice(&lcm->lcm_llcd_resend, &lcd->lcd_llcd_list);
 425                 list_splice(&lcm->lcm_llcd_free, &lcd->lcd_llcd_list);
 426                 spin_unlock(&lcm->lcm_llcd_lock);
 427
 428                 list_for_each_entry_safe(llcd, n, &lcd->lcd_llcd_list,llcd_list)
 429                         llcd_put(llcd);
 430         }
 431
 432         spin_lock(&lcm->lcm_thread_lock);
 433         list_del(&lcd->lcd_lcm_list);
 434         spin_unlock(&lcm->lcm_thread_lock);
 435         OBD_FREE(lcd, sizeof(*lcd));
 436
 437         CDEBUG(D_HA, "%s exiting\n", cfs_curproc_comm());
 438
 439         spin_lock(&lcm->lcm_thread_lock);
 440         atomic_dec(&lcm->lcm_thread_total);
 441         spin_unlock(&lcm->lcm_thread_lock);
 442         cfs_waitq_signal(&lcm->lcm_waitq);
 443
 444         return 0;
 445 }
 446
 447 int llog_start_commit_thread(void)
 448 {
 449         int rc;
 450         ENTRY;
 451
 452         if (atomic_read(&lcm->lcm_thread_total) >= lcm->lcm_thread_max)
 453                 RETURN(0);
 454
 455         rc = cfs_kernel_thread(log_commit_thread, lcm, CLONE_VM | CLONE_FILES);
 456         if (rc < 0) {
 457                 CERROR("error starting thread #%d: %d\n",
 458                        atomic_read(&lcm->lcm_thread_total), rc);
 459                 RETURN(rc);
 460         }
 461
 462         RETURN(0);
 463 }
 464 EXPORT_SYMBOL(llog_start_commit_thread);
 465
 466 static struct llog_process_args {
 467         struct semaphore         llpa_sem;
 468         struct llog_ctxt        *llpa_ctxt;
 469         void                    *llpa_cb;
 470         void                    *llpa_arg;
 471 } llpa;
 472
 473 int llog_init_commit_master(void)
 474 {
 475         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_busy);
 476         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_idle);
 477         spin_lock_init(&lcm->lcm_thread_lock);
 478         atomic_set(&lcm->lcm_thread_numidle, 0);
 479         cfs_waitq_init(&lcm->lcm_waitq);
 480         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_pending);
 481         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_resend);
 482         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_free);
 483         spin_lock_init(&lcm->lcm_llcd_lock);
 484         atomic_set(&lcm->lcm_llcd_numfree, 0);
 485         lcm->lcm_llcd_minfree = 0;
 486         lcm->lcm_thread_max = 5;
 487         /* FIXME initialize semaphore for llog_process_args */
 488         sema_init(&llpa.llpa_sem, 1);
 489         return 0;
 490 }
 491
 492 int llog_cleanup_commit_master(int force)
 493 {
 494         lcm->lcm_flags |= LLOG_LCM_FL_EXIT;
 495         if (force)
 496                 lcm->lcm_flags |= LLOG_LCM_FL_EXIT_FORCE;
 497         cfs_waitq_signal(&lcm->lcm_waitq);
 498
 499         wait_event_interruptible(lcm->lcm_waitq,
 500                                  atomic_read(&lcm->lcm_thread_total) == 0);
 501         return 0;
 502 }
 503
 504 static int log_process_thread(void *args)
 505 {
 506         struct llog_process_args *data = args;
 507         struct llog_ctxt *ctxt = data->llpa_ctxt;
 508         void   *cb = data->llpa_cb;
 509         struct llog_logid logid = *(struct llog_logid *)(data->llpa_arg);
 510         struct llog_handle *llh = NULL;
 511         int rc;
 512         ENTRY;
 513
 514         mutex_up(&data->llpa_sem);
 515         ptlrpc_daemonize("llog_process");     /* thread does IO to log files */
 516
 517         rc = llog_create(ctxt, &llh, &logid, NULL);
 518         if (rc) {
 519                 CERROR("llog_create failed %d\n", rc);
 520                 RETURN(rc);
 521         }
 522         rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
 523         if (rc) {
 524                 CERROR("llog_init_handle failed %d\n", rc);
 525                 GOTO(out, rc);
 526         }
 527
 528         if (cb) {
 529                 rc = llog_cat_process(llh, (llog_cb_t)cb, NULL);
 530                 if (rc != LLOG_PROC_BREAK)
 531                         CERROR("llog_cat_process failed %d\n", rc);
 532         } else {
 533                 CWARN("no callback function for recovery\n");
 534         }
 535
 536         CDEBUG(D_HA, "send llcd %p:%p forcibly after recovery\n",
 537                ctxt->loc_llcd, ctxt);
 538         llog_sync(ctxt, NULL);
 539 out:
 540         rc = llog_cat_put(llh);
 541         if (rc)
 542                 CERROR("llog_cat_put failed %d\n", rc);
 543
 544         RETURN(rc);
 545 }
 546
 547 static int llog_recovery_generic(struct llog_ctxt *ctxt, void *handle,void *arg)
 548 {
 549         int rc;
 550         ENTRY;
 551
 552         mutex_down(&llpa.llpa_sem);
 553         llpa.llpa_ctxt = ctxt;
 554         llpa.llpa_cb = handle;
 555         llpa.llpa_arg = arg;
 556
 557         rc = cfs_kernel_thread(log_process_thread, &llpa, CLONE_VM | CLONE_FILES);
 558         if (rc < 0)
 559                 CERROR("error starting log_process_thread: %d\n", rc);
 560         else {
 561                 CDEBUG(D_HA, "log_process_thread: %d\n", rc);
 562                 rc = 0;
 563         }
 564
 565         RETURN(rc);
 566 }
 567
 568 int llog_repl_connect(struct llog_ctxt *ctxt, int count,
 569                       struct llog_logid *logid, struct llog_gen *gen,
 570                       struct obd_uuid *uuid)
 571 {
 572         struct llog_canceld_ctxt *llcd;
 573         int rc;
 574         ENTRY;
 575
 576         /* send back llcd before recovery from llog */
 577         if (ctxt->loc_llcd != NULL) {
 578                 CWARN("llcd %p:%p not empty\n", ctxt->loc_llcd, ctxt);
 579                 llog_sync(ctxt, NULL);
 580         }
 581
 582         mutex_down(&ctxt->loc_sem);
 583         ctxt->loc_gen = *gen;
 584         llcd = llcd_grab();
 585         if (llcd == NULL) {
 586                 CERROR("couldn't get an llcd\n");
 587                 mutex_up(&ctxt->loc_sem);
 588                 RETURN(-ENOMEM);
 589         }
 590         llcd->llcd_ctxt = ctxt;
 591         ctxt->loc_llcd = llcd;
 592         mutex_up(&ctxt->loc_sem);
 593
 594         rc = llog_recovery_generic(ctxt, ctxt->llog_proc_cb, logid);
 595         if (rc != 0)
 596                 CERROR("error recovery process: %d\n", rc);
 597
 598         RETURN(rc);
 599 }
 600 EXPORT_SYMBOL(llog_repl_connect);
 601
 602 #else /* !__KERNEL__ */
 603
 604 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
 605                          struct lov_stripe_md *lsm, int count,
 606                          struct llog_cookie *cookies, int flags)
 607 {
 608         return 0;
 609 }
 610 #endif