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 inline void stop_log_commit(struct llog_commit_master *lcm,
 221                                    struct llog_commit_daemon *lcd,
 222                                    int rc)
 223 {
 224         CERROR("error preparing commit: rc %d\n", rc);
 225
 226         spin_lock(&lcm->lcm_llcd_lock);
 227         list_splice(&lcd->lcd_llcd_list, &lcm->lcm_llcd_resend);
 228         CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
 229         spin_unlock(&lcm->lcm_llcd_lock);
 230 }
 231
 232 static int log_commit_thread(void *arg)
 233 {
 234         struct llog_commit_master *lcm = arg;
 235         struct llog_commit_daemon *lcd;
 236         struct llog_canceld_ctxt *llcd, *n;
 237         struct obd_import *import = NULL;
 238         ENTRY;
 239
 240         OBD_ALLOC(lcd, sizeof(*lcd));
 241         if (lcd == NULL)
 242                 RETURN(-ENOMEM);
 243
 244         spin_lock(&lcm->lcm_thread_lock);
 245         THREAD_NAME(cfs_curproc_comm(), CFS_CURPROC_COMM_MAX - 1,
 246                     "ll_log_comt_%02d", atomic_read(&lcm->lcm_thread_total));
 247         atomic_inc(&lcm->lcm_thread_total);
 248         spin_unlock(&lcm->lcm_thread_lock);
 249
 250         ptlrpc_daemonize(cfs_curproc_comm()); /* thread never needs to do IO */
 251
 252         CFS_INIT_LIST_HEAD(&lcd->lcd_lcm_list);
 253         CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
 254         lcd->lcd_lcm = lcm;
 255
 256         CDEBUG(D_HA, "%s started\n", cfs_curproc_comm());
 257         do {
 258                 struct ptlrpc_request *request;
 259                 struct list_head *sending_list;
 260                 int rc = 0;
 261
 262                 if (import)
 263                         class_import_put(import);
 264                 import = NULL;
 265
 266                 /* If we do not have enough pages available, allocate some */
 267                 while (atomic_read(&lcm->lcm_llcd_numfree) <
 268                        lcm->lcm_llcd_minfree) {
 269                         if (llcd_alloc() < 0)
 270                                 break;
 271                 }
 272
 273                 spin_lock(&lcm->lcm_thread_lock);
 274                 atomic_inc(&lcm->lcm_thread_numidle);
 275                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_idle);
 276                 spin_unlock(&lcm->lcm_thread_lock);
 277
 278                 wait_event_interruptible(lcm->lcm_waitq,
 279                                          !list_empty(&lcm->lcm_llcd_pending) ||
 280                                          lcm->lcm_flags & LLOG_LCM_FL_EXIT);
 281
 282                 /* If we are the last available thread, start a new one in case
 283                  * we get blocked on an RPC (nobody else will start a new one)*/
 284                 spin_lock(&lcm->lcm_thread_lock);
 285                 atomic_dec(&lcm->lcm_thread_numidle);
 286                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_busy);
 287                 spin_unlock(&lcm->lcm_thread_lock);
 288
 289                 sending_list = &lcm->lcm_llcd_pending;
 290         resend:
 291                 if (import)
 292                         class_import_put(import);
 293                 import = NULL;
 294                 if (lcm->lcm_flags & LLOG_LCM_FL_EXIT) {
 295                         lcm->lcm_llcd_maxfree = 0;
 296                         lcm->lcm_llcd_minfree = 0;
 297                         lcm->lcm_thread_max = 0;
 298
 299                         if (list_empty(&lcm->lcm_llcd_pending) ||
 300                             lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE)
 301                                 break;
 302                 }
 303
 304                 if (atomic_read(&lcm->lcm_thread_numidle) <= 1 &&
 305                     atomic_read(&lcm->lcm_thread_total) < lcm->lcm_thread_max) {
 306                         rc = llog_start_commit_thread();
 307                         if (rc < 0)
 308                                 CERROR("error starting thread: rc %d\n", rc);
 309                 }
 310
 311                 /* Move all of the pending cancels from the same OST off of
 312                  * the list, so we don't get multiple threads blocked and/or
 313                  * doing upcalls on the same OST in case of failure. */
 314                 spin_lock(&lcm->lcm_llcd_lock);
 315                 if (!list_empty(sending_list)) {
 316                         list_move_tail(sending_list->next,
 317                                        &lcd->lcd_llcd_list);
 318                         llcd = list_entry(lcd->lcd_llcd_list.next,
 319                                           typeof(*llcd), llcd_list);
 320                         LASSERT(llcd->llcd_lcm == lcm);
 321                         import = llcd->llcd_ctxt->loc_imp;
 322                         if (import)
 323                                 class_import_get(import);
 324                 }
 325                 list_for_each_entry_safe(llcd, n, sending_list, llcd_list) {
 326                         LASSERT(llcd->llcd_lcm == lcm);
 327                         if (import == llcd->llcd_ctxt->loc_imp)
 328                                 list_move_tail(&llcd->llcd_list,
 329                                                &lcd->lcd_llcd_list);
 330                 }
 331                 if (sending_list != &lcm->lcm_llcd_resend) {
 332                         list_for_each_entry_safe(llcd, n, &lcm->lcm_llcd_resend,
 333                                                  llcd_list) {
 334                                 LASSERT(llcd->llcd_lcm == lcm);
 335                                 if (import == llcd->llcd_ctxt->loc_imp)
 336                                         list_move_tail(&llcd->llcd_list,
 337                                                        &lcd->lcd_llcd_list);
 338                         }
 339                 }
 340                 spin_unlock(&lcm->lcm_llcd_lock);
 341
 342                 /* We are the only one manipulating our local list - no lock */
 343                 list_for_each_entry_safe(llcd,n, &lcd->lcd_llcd_list,llcd_list){
 344                         char *bufs[2] = { NULL, (char *)llcd->llcd_cookies };
 345
 346                         list_del(&llcd->llcd_list);
 347                         if (llcd->llcd_cookiebytes == 0) {
 348                                 CDEBUG(D_RPCTRACE, "put empty llcd %p:%p\n",
 349                                        llcd, llcd->llcd_ctxt);
 350                                 llcd_put(llcd);
 351                                 continue;
 352                         }
 353
 354                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 355                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 356                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 357                                 CWARN("import will be destroyed, put "
 358                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 359                                 llcd_put(llcd);
 360                                 continue;
 361                         }
 362                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 363
 364                         if (!import || (import == LP_POISON) ||
 365                             (import->imp_client == LP_POISON)) {
 366                                 CERROR("No import %p (llcd=%p, ctxt=%p)\n",
 367                                        import, llcd, llcd->llcd_ctxt);
 368                                 llcd_put(llcd);
 369                                 continue;
 370                         }
 371
 372                         OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_RECOV, 10);
 373
 374                         request = ptlrpc_request_alloc(import, &RQF_LOG_CANCEL);
 375                         if (request == NULL) {
 376                                 rc = -ENOMEM;
 377                                 stop_log_commit(lcm, lcd, rc);
 378                                 break;
 379                         }
 380
 381                         req_capsule_set_size(&request->rq_pill, &RMF_LOGCOOKIES,
 382                                              RCL_CLIENT,llcd->llcd_cookiebytes);
 383
 384                         rc = ptlrpc_request_bufs_pack(request,
 385                                                       LUSTRE_LOG_VERSION,
 386                                                       OBD_LOG_CANCEL, bufs,
 387                                                       NULL);
 388                         if (rc) {
 389                                 ptlrpc_request_free(request);
 390                                 stop_log_commit(lcm, lcd, rc);
 391                                 break;
 392                         }
 393
 394                         /* XXX FIXME bug 249, 5515 */
 395                         request->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
 396                         request->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
 397
 398                         ptlrpc_request_set_replen(request);
 399                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 400                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 401                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 402                                 CWARN("import will be destroyed, put "
 403                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 404                                 llcd_put(llcd);
 405                                 ptlrpc_req_finished(request);
 406                                 continue;
 407                         }
 408                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 409                         rc = ptlrpc_queue_wait(request);
 410                         ptlrpc_req_finished(request);
 411
 412                         /* If the RPC failed, we put this and the remaining
 413                          * messages onto the resend list for another time. */
 414                         if (rc == 0) {
 415                                 llcd_put(llcd);
 416                                 continue;
 417                         }
 418
 419                         CERROR("commit %p:%p drop %d cookies: rc %d\n",
 420                                llcd, llcd->llcd_ctxt,
 421                                (int)(llcd->llcd_cookiebytes /
 422                                      sizeof(*llcd->llcd_cookies)), rc);
 423                         llcd_put(llcd);
 424                 }
 425
 426                 if (rc == 0) {
 427                         sending_list = &lcm->lcm_llcd_resend;
 428                         if (!list_empty(sending_list))
 429                                 goto resend;
 430                 }
 431         } while(1);
 432
 433         if (import)
 434                 class_import_put(import);
 435
 436         /* If we are force exiting, just drop all of the cookies. */
 437         if (lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE) {
 438                 spin_lock(&lcm->lcm_llcd_lock);
 439                 list_splice(&lcm->lcm_llcd_pending, &lcd->lcd_llcd_list);
 440                 list_splice(&lcm->lcm_llcd_resend, &lcd->lcd_llcd_list);
 441                 list_splice(&lcm->lcm_llcd_free, &lcd->lcd_llcd_list);
 442                 spin_unlock(&lcm->lcm_llcd_lock);
 443
 444                 list_for_each_entry_safe(llcd, n, &lcd->lcd_llcd_list,llcd_list)
 445                         llcd_put(llcd);
 446         }
 447
 448         spin_lock(&lcm->lcm_thread_lock);
 449         list_del(&lcd->lcd_lcm_list);
 450         spin_unlock(&lcm->lcm_thread_lock);
 451         OBD_FREE(lcd, sizeof(*lcd));
 452
 453         CDEBUG(D_HA, "%s exiting\n", cfs_curproc_comm());
 454
 455         spin_lock(&lcm->lcm_thread_lock);
 456         atomic_dec(&lcm->lcm_thread_total);
 457         spin_unlock(&lcm->lcm_thread_lock);
 458         cfs_waitq_signal(&lcm->lcm_waitq);
 459
 460         return 0;
 461 }
 462
 463 int llog_start_commit_thread(void)
 464 {
 465         int rc;
 466         ENTRY;
 467
 468         if (atomic_read(&lcm->lcm_thread_total) >= lcm->lcm_thread_max)
 469                 RETURN(0);
 470
 471         rc = cfs_kernel_thread(log_commit_thread, lcm, CLONE_VM | CLONE_FILES);
 472         if (rc < 0) {
 473                 CERROR("error starting thread #%d: %d\n",
 474                        atomic_read(&lcm->lcm_thread_total), rc);
 475                 RETURN(rc);
 476         }
 477
 478         RETURN(0);
 479 }
 480 EXPORT_SYMBOL(llog_start_commit_thread);
 481
 482 static struct llog_process_args {
 483         struct semaphore         llpa_sem;
 484         struct llog_ctxt        *llpa_ctxt;
 485         void                    *llpa_cb;
 486         void                    *llpa_arg;
 487 } llpa;
 488
 489 int llog_init_commit_master(void)
 490 {
 491         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_busy);
 492         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_idle);
 493         spin_lock_init(&lcm->lcm_thread_lock);
 494         atomic_set(&lcm->lcm_thread_numidle, 0);
 495         cfs_waitq_init(&lcm->lcm_waitq);
 496         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_pending);
 497         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_resend);
 498         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_free);
 499         spin_lock_init(&lcm->lcm_llcd_lock);
 500         atomic_set(&lcm->lcm_llcd_numfree, 0);
 501         lcm->lcm_llcd_minfree = 0;
 502         lcm->lcm_thread_max = 5;
 503         /* FIXME initialize semaphore for llog_process_args */
 504         sema_init(&llpa.llpa_sem, 1);
 505         return 0;
 506 }
 507
 508 int llog_cleanup_commit_master(int force)
 509 {
 510         lcm->lcm_flags |= LLOG_LCM_FL_EXIT;
 511         if (force)
 512                 lcm->lcm_flags |= LLOG_LCM_FL_EXIT_FORCE;
 513         cfs_waitq_signal(&lcm->lcm_waitq);
 514
 515         wait_event_interruptible(lcm->lcm_waitq,
 516                                  atomic_read(&lcm->lcm_thread_total) == 0);
 517         return 0;
 518 }
 519
 520 static int log_process_thread(void *args)
 521 {
 522         struct llog_process_args *data = args;
 523         struct llog_ctxt *ctxt = data->llpa_ctxt;
 524         void   *cb = data->llpa_cb;
 525         struct llog_logid logid = *(struct llog_logid *)(data->llpa_arg);
 526         struct llog_handle *llh = NULL;
 527         int rc;
 528         ENTRY;
 529
 530         mutex_up(&data->llpa_sem);
 531         ptlrpc_daemonize("llog_process");     /* thread does IO to log files */
 532
 533         rc = llog_create(ctxt, &llh, &logid, NULL);
 534         if (rc) {
 535                 CERROR("llog_create failed %d\n", rc);
 536                 RETURN(rc);
 537         }
 538         rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
 539         if (rc) {
 540                 CERROR("llog_init_handle failed %d\n", rc);
 541                 GOTO(out, rc);
 542         }
 543
 544         if (cb) {
 545                 rc = llog_cat_process(llh, (llog_cb_t)cb, NULL);
 546                 if (rc != LLOG_PROC_BREAK)
 547                         CERROR("llog_cat_process failed %d\n", rc);
 548         } else {
 549                 CWARN("no callback function for recovery\n");
 550         }
 551
 552         CDEBUG(D_HA, "send llcd %p:%p forcibly after recovery\n",
 553                ctxt->loc_llcd, ctxt);
 554         llog_sync(ctxt, NULL);
 555 out:
 556         rc = llog_cat_put(llh);
 557         if (rc)
 558                 CERROR("llog_cat_put failed %d\n", rc);
 559
 560         RETURN(rc);
 561 }
 562
 563 static int llog_recovery_generic(struct llog_ctxt *ctxt, void *handle,void *arg)
 564 {
 565         int rc;
 566         ENTRY;
 567
 568         mutex_down(&llpa.llpa_sem);
 569         llpa.llpa_ctxt = ctxt;
 570         llpa.llpa_cb = handle;
 571         llpa.llpa_arg = arg;
 572
 573         rc = cfs_kernel_thread(log_process_thread, &llpa, CLONE_VM | CLONE_FILES);
 574         if (rc < 0)
 575                 CERROR("error starting log_process_thread: %d\n", rc);
 576         else {
 577                 CDEBUG(D_HA, "log_process_thread: %d\n", rc);
 578                 rc = 0;
 579         }
 580
 581         RETURN(rc);
 582 }
 583
 584 int llog_repl_connect(struct llog_ctxt *ctxt, int count,
 585                       struct llog_logid *logid, struct llog_gen *gen,
 586                       struct obd_uuid *uuid)
 587 {
 588         struct llog_canceld_ctxt *llcd;
 589         int rc;
 590         ENTRY;
 591
 592         /* send back llcd before recovery from llog */
 593         if (ctxt->loc_llcd != NULL) {
 594                 CWARN("llcd %p:%p not empty\n", ctxt->loc_llcd, ctxt);
 595                 llog_sync(ctxt, NULL);
 596         }
 597
 598         mutex_down(&ctxt->loc_sem);
 599         ctxt->loc_gen = *gen;
 600         llcd = llcd_grab();
 601         if (llcd == NULL) {
 602                 CERROR("couldn't get an llcd\n");
 603                 mutex_up(&ctxt->loc_sem);
 604                 RETURN(-ENOMEM);
 605         }
 606         llcd->llcd_ctxt = ctxt;
 607         ctxt->loc_llcd = llcd;
 608         mutex_up(&ctxt->loc_sem);
 609
 610         rc = llog_recovery_generic(ctxt, ctxt->llog_proc_cb, logid);
 611         if (rc != 0)
 612                 CERROR("error recovery process: %d\n", rc);
 613
 614         RETURN(rc);
 615 }
 616 EXPORT_SYMBOL(llog_repl_connect);
 617
 618 #else /* !__KERNEL__ */
 619
 620 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
 621                          struct lov_stripe_md *lsm, int count,
 622                          struct llog_cookie *cookies, int flags)
 623 {
 624         return 0;
 625 }
 626 #endif