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 /* Allocate new commit structs in case we do not have enough.
  59  * Make the llcd size small enough that it fits into a single page when we
  60  * are sending/receiving it. */
  61 static int llcd_alloc(struct llog_commit_master *lcm)
  62 {
  63         struct llog_canceld_ctxt *llcd;
  64         int llcd_size;
  65
  66         /* payload of lustre_msg V2 is bigger */
  67         llcd_size = 4096 - lustre_msg_size(LUSTRE_MSG_MAGIC_V2, 1, NULL);
  68         OBD_ALLOC(llcd,
  69                   llcd_size + offsetof(struct llog_canceld_ctxt, llcd_cookies));
  70         if (llcd == NULL)
  71                 return -ENOMEM;
  72
  73         llcd->llcd_size = llcd_size;
  74         llcd->llcd_lcm = lcm;
  75
  76         spin_lock(&lcm->lcm_llcd_lock);
  77         list_add(&llcd->llcd_list, &lcm->lcm_llcd_free);
  78         atomic_inc(&lcm->lcm_llcd_numfree);
  79         spin_unlock(&lcm->lcm_llcd_lock);
  80
  81         return 0;
  82 }
  83
  84 /* Get a free cookie struct from the list */
  85 static struct llog_canceld_ctxt *llcd_grab(struct llog_commit_master *lcm)
  86 {
  87         struct llog_canceld_ctxt *llcd;
  88
  89 repeat:
  90         spin_lock(&lcm->lcm_llcd_lock);
  91         if (list_empty(&lcm->lcm_llcd_free)) {
  92                 spin_unlock(&lcm->lcm_llcd_lock);
  93                 if (llcd_alloc(lcm) < 0) {
  94                         CERROR("unable to allocate log commit data!\n");
  95                         return NULL;
  96                 }
  97                 /* check new llcd wasn't grabbed while lock dropped, b=7407 */
  98                 goto repeat;
  99         }
 100
 101         llcd = list_entry(lcm->lcm_llcd_free.next, typeof(*llcd), llcd_list);
 102         list_del(&llcd->llcd_list);
 103         atomic_dec(&lcm->lcm_llcd_numfree);
 104         spin_unlock(&lcm->lcm_llcd_lock);
 105
 106         llcd->llcd_cookiebytes = 0;
 107
 108         return llcd;
 109 }
 110
 111 static void llcd_put(struct llog_canceld_ctxt *llcd)
 112 {
 113         struct llog_commit_master *lcm = llcd->llcd_lcm;
 114
 115         llog_ctxt_put(llcd->llcd_ctxt);
 116         if (atomic_read(&lcm->lcm_llcd_numfree) >= lcm->lcm_llcd_maxfree) {
 117                 int llcd_size = llcd->llcd_size +
 118                          offsetof(struct llog_canceld_ctxt, llcd_cookies);
 119                 OBD_FREE(llcd, llcd_size);
 120         } else {
 121                 spin_lock(&lcm->lcm_llcd_lock);
 122                 list_add(&llcd->llcd_list, &lcm->lcm_llcd_free);
 123                 atomic_inc(&lcm->lcm_llcd_numfree);
 124                 spin_unlock(&lcm->lcm_llcd_lock);
 125         }
 126 }
 127
 128 /* Send some cookies to the appropriate target */
 129 static void llcd_send(struct llog_canceld_ctxt *llcd)
 130 {
 131         if (!(llcd->llcd_lcm->lcm_flags & LLOG_LCM_FL_EXIT)) {
 132         spin_lock(&llcd->llcd_lcm->lcm_llcd_lock);
 133                 list_add_tail(&llcd->llcd_list,
 134                               &llcd->llcd_lcm->lcm_llcd_pending);
 135         spin_unlock(&llcd->llcd_lcm->lcm_llcd_lock);
 136         }
 137         cfs_waitq_signal_nr(&llcd->llcd_lcm->lcm_waitq, 1);
 138 }
 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(ctxt->loc_lcm);
 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 = llog_ctxt_get(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_init(&lcd->lcd_llcd_list, &lcm->lcm_llcd_resend);
 228         spin_unlock(&lcm->lcm_llcd_lock);
 229 }
 230
 231 static int log_commit_thread(void *arg)
 232 {
 233         struct llog_commit_master *lcm = arg;
 234         struct llog_commit_daemon *lcd;
 235         struct llog_canceld_ctxt *llcd, *n;
 236         struct obd_import *import = NULL;
 237         ENTRY;
 238
 239         OBD_ALLOC(lcd, sizeof(*lcd));
 240         if (lcd == NULL)
 241                 RETURN(-ENOMEM);
 242
 243         spin_lock(&lcm->lcm_thread_lock);
 244         THREAD_NAME(cfs_curproc_comm(), CFS_CURPROC_COMM_MAX - 1,
 245                     "ll_log_comt_%02d", atomic_read(&lcm->lcm_thread_total));
 246         atomic_inc(&lcm->lcm_thread_total);
 247         spin_unlock(&lcm->lcm_thread_lock);
 248
 249         ptlrpc_daemonize(cfs_curproc_comm()); /* thread never needs to do IO */
 250
 251         CFS_INIT_LIST_HEAD(&lcd->lcd_lcm_list);
 252         CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
 253         lcd->lcd_lcm = lcm;
 254
 255         CDEBUG(D_HA, "%s started\n", cfs_curproc_comm());
 256         do {
 257                 struct ptlrpc_request *request;
 258                 struct list_head *sending_list;
 259                 int rc = 0;
 260
 261                 if (import)
 262                         class_import_put(import);
 263                 import = NULL;
 264
 265                 /* If we do not have enough pages available, allocate some */
 266                 while (atomic_read(&lcm->lcm_llcd_numfree) <
 267                        lcm->lcm_llcd_minfree) {
 268                         if (llcd_alloc(lcm) < 0)
 269                                 break;
 270                 }
 271
 272                 spin_lock(&lcm->lcm_thread_lock);
 273                 atomic_inc(&lcm->lcm_thread_numidle);
 274                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_idle);
 275                 spin_unlock(&lcm->lcm_thread_lock);
 276
 277                 wait_event_interruptible(lcm->lcm_waitq,
 278                                          !list_empty(&lcm->lcm_llcd_pending) ||
 279                                          lcm->lcm_flags & LLOG_LCM_FL_EXIT);
 280
 281                 /* If we are the last available thread, start a new one in case
 282                  * we get blocked on an RPC (nobody else will start a new one)*/
 283                 spin_lock(&lcm->lcm_thread_lock);
 284                 atomic_dec(&lcm->lcm_thread_numidle);
 285                 list_move(&lcd->lcd_lcm_list, &lcm->lcm_thread_busy);
 286                 spin_unlock(&lcm->lcm_thread_lock);
 287
 288                 sending_list = &lcm->lcm_llcd_pending;
 289         resend:
 290                 if (import)
 291                         class_import_put(import);
 292                 import = NULL;
 293                 if (lcm->lcm_flags & LLOG_LCM_FL_EXIT) {
 294                         lcm->lcm_llcd_maxfree = 0;
 295                         lcm->lcm_llcd_minfree = 0;
 296                         lcm->lcm_thread_max = 0;
 297
 298                         if (list_empty(&lcm->lcm_llcd_pending) ||
 299                             lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE)
 300                                 break;
 301                 }
 302
 303                 if (atomic_read(&lcm->lcm_thread_numidle) <= 1 &&
 304                     atomic_read(&lcm->lcm_thread_total) < lcm->lcm_thread_max) {
 305                         rc = llog_start_commit_thread(lcm);
 306                         if (rc < 0)
 307                                 CERROR("error starting thread: rc %d\n", rc);
 308                 }
 309
 310                 /* Move all of the pending cancels from the same OST off of
 311                  * the list, so we don't get multiple threads blocked and/or
 312                  * doing upcalls on the same OST in case of failure. */
 313                 spin_lock(&lcm->lcm_llcd_lock);
 314                 if (!list_empty(sending_list)) {
 315                         list_move_tail(sending_list->next,
 316                                        &lcd->lcd_llcd_list);
 317                         llcd = list_entry(lcd->lcd_llcd_list.next,
 318                                           typeof(*llcd), llcd_list);
 319                         LASSERT(llcd->llcd_lcm == lcm);
 320                         import = llcd->llcd_ctxt->loc_imp;
 321                         if (import)
 322                                 class_import_get(import);
 323                 }
 324                 list_for_each_entry_safe(llcd, n, sending_list, llcd_list) {
 325                         LASSERT(llcd->llcd_lcm == lcm);
 326                         if (import == llcd->llcd_ctxt->loc_imp)
 327                                 list_move_tail(&llcd->llcd_list,
 328                                                &lcd->lcd_llcd_list);
 329                 }
 330                 if (sending_list != &lcm->lcm_llcd_resend) {
 331                         list_for_each_entry_safe(llcd, n, &lcm->lcm_llcd_resend,
 332                                                  llcd_list) {
 333                                 LASSERT(llcd->llcd_lcm == lcm);
 334                                 if (import == llcd->llcd_ctxt->loc_imp)
 335                                         list_move_tail(&llcd->llcd_list,
 336                                                        &lcd->lcd_llcd_list);
 337                         }
 338                 }
 339                 spin_unlock(&lcm->lcm_llcd_lock);
 340
 341                 /* We are the only one manipulating our local list - no lock */
 342                 list_for_each_entry_safe(llcd,n, &lcd->lcd_llcd_list,llcd_list){
 343                         char *bufs[2] = { NULL, (char *)llcd->llcd_cookies };
 344
 345                         list_del(&llcd->llcd_list);
 346                         if (llcd->llcd_cookiebytes == 0) {
 347                                 CDEBUG(D_RPCTRACE, "put empty llcd %p:%p\n",
 348                                        llcd, llcd->llcd_ctxt);
 349                                 llcd_put(llcd);
 350                                 continue;
 351                         }
 352
 353                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 354                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 355                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 356                                 CWARN("import will be destroyed, put "
 357                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 358                                 llcd_put(llcd);
 359                                 continue;
 360                         }
 361                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 362
 363                         if (!import || (import == LP_POISON) ||
 364                             (import->imp_client == LP_POISON)) {
 365                                 CERROR("No import %p (llcd=%p, ctxt=%p)\n",
 366                                        import, llcd, llcd->llcd_ctxt);
 367                                 llcd_put(llcd);
 368                                 continue;
 369                         }
 370
 371                         OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_RECOV, 10);
 372
 373                         request = ptlrpc_request_alloc(import, &RQF_LOG_CANCEL);
 374                         if (request == NULL) {
 375                                 rc = -ENOMEM;
 376                                 stop_log_commit(lcm, lcd, rc);
 377                                 break;
 378                         }
 379
 380                         req_capsule_set_size(&request->rq_pill, &RMF_LOGCOOKIES,
 381                                              RCL_CLIENT,llcd->llcd_cookiebytes);
 382
 383                         rc = ptlrpc_request_bufs_pack(request,
 384                                                       LUSTRE_LOG_VERSION,
 385                                                       OBD_LOG_CANCEL, bufs,
 386                                                       NULL);
 387                         if (rc) {
 388                                 ptlrpc_request_free(request);
 389                                 stop_log_commit(lcm, lcd, rc);
 390                                 break;
 391                         }
 392
 393                         /* XXX FIXME bug 249, 5515 */
 394                         request->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
 395                         request->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
 396
 397                         ptlrpc_request_set_replen(request);
 398                         mutex_down(&llcd->llcd_ctxt->loc_sem);
 399                         if (llcd->llcd_ctxt->loc_imp == NULL) {
 400                                 mutex_up(&llcd->llcd_ctxt->loc_sem);
 401                                 CWARN("import will be destroyed, put "
 402                                       "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
 403                                 llcd_put(llcd);
 404                                 ptlrpc_req_finished(request);
 405                                 continue;
 406                         }
 407                         mutex_up(&llcd->llcd_ctxt->loc_sem);
 408                         rc = ptlrpc_queue_wait(request);
 409                         ptlrpc_req_finished(request);
 410
 411                         /* If the RPC failed, we put this and the remaining
 412                          * messages onto the resend list for another time. */
 413                         if (rc == 0) {
 414                                 llcd_put(llcd);
 415                                 continue;
 416                         }
 417
 418                         CERROR("commit %p:%p drop %d cookies: rc %d\n",
 419                                llcd, llcd->llcd_ctxt,
 420                                (int)(llcd->llcd_cookiebytes /
 421                                      sizeof(*llcd->llcd_cookies)), rc);
 422                         llcd_put(llcd);
 423                 }
 424
 425                 if (rc == 0) {
 426                         sending_list = &lcm->lcm_llcd_resend;
 427                         if (!list_empty(sending_list))
 428                                 goto resend;
 429                 }
 430         } while(1);
 431
 432         if (import)
 433                 class_import_put(import);
 434
 435         /* If we are force exiting, just drop all of the cookies. */
 436         if (lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE) {
 437                 spin_lock(&lcm->lcm_llcd_lock);
 438                 list_splice_init(&lcm->lcm_llcd_pending, &lcd->lcd_llcd_list);
 439                 list_splice_init(&lcm->lcm_llcd_resend, &lcd->lcd_llcd_list);
 440                 list_splice_init(&lcm->lcm_llcd_free, &lcd->lcd_llcd_list);
 441                 spin_unlock(&lcm->lcm_llcd_lock);
 442
 443                 list_for_each_entry_safe(llcd, n, &lcd->lcd_llcd_list,llcd_list)
 444                         llcd_put(llcd);
 445         }
 446
 447         spin_lock(&lcm->lcm_thread_lock);
 448         list_del(&lcd->lcd_lcm_list);
 449         spin_unlock(&lcm->lcm_thread_lock);
 450         OBD_FREE(lcd, sizeof(*lcd));
 451
 452         CDEBUG(D_HA, "%s exiting\n", cfs_curproc_comm());
 453
 454         spin_lock(&lcm->lcm_thread_lock);
 455         atomic_dec(&lcm->lcm_thread_total);
 456         spin_unlock(&lcm->lcm_thread_lock);
 457         cfs_waitq_signal(&lcm->lcm_waitq);
 458
 459         return 0;
 460 }
 461
 462 int llog_start_commit_thread(struct llog_commit_master *lcm)
 463 {
 464         int rc;
 465         ENTRY;
 466
 467         if (atomic_read(&lcm->lcm_thread_total) >= lcm->lcm_thread_max)
 468                 RETURN(0);
 469
 470         rc = cfs_kernel_thread(log_commit_thread, lcm, CLONE_VM | CLONE_FILES);
 471         if (rc < 0) {
 472                 CERROR("error starting thread #%d: %d\n",
 473                        atomic_read(&lcm->lcm_thread_total), rc);
 474                 RETURN(rc);
 475         }
 476
 477         RETURN(0);
 478 }
 479 EXPORT_SYMBOL(llog_start_commit_thread);
 480
 481 static struct llog_process_args {
 482         struct semaphore         llpa_sem;
 483         struct llog_ctxt        *llpa_ctxt;
 484         void                    *llpa_cb;
 485         void                    *llpa_arg;
 486 } llpa;
 487
 488 int llog_init_commit_master(struct llog_commit_master *lcm)
 489 {
 490         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_busy);
 491         CFS_INIT_LIST_HEAD(&lcm->lcm_thread_idle);
 492         spin_lock_init(&lcm->lcm_thread_lock);
 493         atomic_set(&lcm->lcm_thread_numidle, 0);
 494         cfs_waitq_init(&lcm->lcm_waitq);
 495         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_pending);
 496         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_resend);
 497         CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_free);
 498         spin_lock_init(&lcm->lcm_llcd_lock);
 499         atomic_set(&lcm->lcm_llcd_numfree, 0);
 500         lcm->lcm_llcd_minfree = 0;
 501         lcm->lcm_thread_max = 5;
 502         /* FIXME initialize semaphore for llog_process_args */
 503         sema_init(&llpa.llpa_sem, 1);
 504         return 0;
 505 }
 506 EXPORT_SYMBOL(llog_init_commit_master);
 507
 508 int llog_cleanup_commit_master(struct llog_commit_master *lcm,
 509                                int force)
 510 {
 511         lcm->lcm_flags |= LLOG_LCM_FL_EXIT;
 512         if (force)
 513                 lcm->lcm_flags |= LLOG_LCM_FL_EXIT_FORCE;
 514         cfs_waitq_signal(&lcm->lcm_waitq);
 515
 516         wait_event_interruptible(lcm->lcm_waitq,
 517                                  atomic_read(&lcm->lcm_thread_total) == 0);
 518         return 0;
 519 }
 520 EXPORT_SYMBOL(llog_cleanup_commit_master);
 521
 522 static int log_process_thread(void *args)
 523 {
 524         struct llog_process_args *data = args;
 525         struct llog_ctxt *ctxt = data->llpa_ctxt;
 526         void   *cb = data->llpa_cb;
 527         struct llog_logid logid = *(struct llog_logid *)(data->llpa_arg);
 528         struct llog_handle *llh = NULL;
 529         int rc;
 530         ENTRY;
 531
 532         mutex_up(&data->llpa_sem);
 533         ptlrpc_daemonize("llog_process");     /* thread does IO to log files */
 534
 535         rc = llog_create(ctxt, &llh, &logid, NULL);
 536         if (rc) {
 537                 CERROR("llog_create failed %d\n", rc);
 538                 GOTO(out, rc);
 539         }
 540         rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
 541         if (rc) {
 542                 CERROR("llog_init_handle failed %d\n", rc);
 543                 GOTO(release_llh, rc);
 544         }
 545
 546         if (cb) {
 547                 rc = llog_cat_process(llh, (llog_cb_t)cb, NULL);
 548                 if (rc != LLOG_PROC_BREAK)
 549                         CERROR("llog_cat_process failed %d\n", rc);
 550         } else {
 551                 CWARN("no callback function for recovery\n");
 552         }
 553
 554         CDEBUG(D_HA, "send llcd %p:%p forcibly after recovery\n",
 555                ctxt->loc_llcd, ctxt);
 556         llog_sync(ctxt, NULL);
 557
 558 release_llh:
 559         rc = llog_cat_put(llh);
 560         if (rc)
 561                 CERROR("llog_cat_put failed %d\n", rc);
 562 out:
 563         llog_ctxt_put(ctxt);
 564         RETURN(rc);
 565 }
 566
 567 static int llog_recovery_generic(struct llog_ctxt *ctxt, void *handle,void *arg)
 568 {
 569         struct obd_device *obd = ctxt->loc_obd;
 570         int rc;
 571         ENTRY;
 572
 573         if (obd->obd_stopping)
 574                 RETURN(-ENODEV);
 575
 576         mutex_down(&llpa.llpa_sem);
 577         llpa.llpa_cb = handle;
 578         llpa.llpa_arg = arg;
 579         llpa.llpa_ctxt = llog_ctxt_get(ctxt); //llog_group_get_ctxt(ctxt->loc_olg, ctxt->loc_idx);
 580         if (!llpa.llpa_ctxt) {
 581                 up(&llpa.llpa_sem);
 582                 RETURN(-ENODEV);
 583         }
 584         rc = cfs_kernel_thread(log_process_thread, &llpa, CLONE_VM | CLONE_FILES);
 585         if (rc < 0)
 586                 CERROR("error starting log_process_thread: %d\n", rc);
 587         else {
 588                 CDEBUG(D_HA, "log_process_thread: %d\n", rc);
 589                 rc = 0;
 590         }
 591
 592         RETURN(rc);
 593 }
 594
 595 int llog_repl_connect(struct llog_ctxt *ctxt, int count,
 596                       struct llog_logid *logid, struct llog_gen *gen,
 597                       struct obd_uuid *uuid)
 598 {
 599         struct llog_canceld_ctxt *llcd;
 600         int rc;
 601         ENTRY;
 602
 603         /* send back llcd before recovery from llog */
 604         if (ctxt->loc_llcd != NULL) {
 605                 CWARN("llcd %p:%p not empty\n", ctxt->loc_llcd, ctxt);
 606                 llog_sync(ctxt, NULL);
 607         }
 608
 609         mutex_down(&ctxt->loc_sem);
 610         ctxt->loc_gen = *gen;
 611         llcd = llcd_grab(ctxt->loc_lcm);
 612         if (llcd == NULL) {
 613                 CERROR("couldn't get an llcd\n");
 614                 mutex_up(&ctxt->loc_sem);
 615                 RETURN(-ENOMEM);
 616         }
 617         llcd->llcd_ctxt = llog_ctxt_get(ctxt);
 618         ctxt->loc_llcd = llcd;
 619         mutex_up(&ctxt->loc_sem);
 620
 621         rc = llog_recovery_generic(ctxt, ctxt->llog_proc_cb, logid);
 622         if (rc != 0)
 623                 CERROR("error recovery process: %d\n", rc);
 624
 625         RETURN(rc);
 626 }
 627 EXPORT_SYMBOL(llog_repl_connect);
 628
 629 #else /* !__KERNEL__ */
 630
 631 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
 632                          struct lov_stripe_md *lsm, int count,
 633                          struct llog_cookie *cookies, int flags)
 634 {
 635         return 0;
 636 }
 637 #endif