lustre/target/update_recovery.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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2014, Intel Corporation.
  24  */
  25
  26 /*
  27  * lustre/target/update_recovery.c
  28  *
  29  * This file implement the methods to handle the update recovery.
  30  *
  31  * During DNE recovery, the recovery thread will redo the operation according
  32  * to the transaction no, and these replay are either from client replay req
  33  * or update replay records(for distribute transaction) in the update log.
  34  * For distribute transaction replay, the replay thread will call
  35  * distribute_txn_replay_handle() to handle the updates.
  36  *
  37  * After the Master MDT restarts, it will retrieve the update records from all
  38  * of MDTs, for each distributed operation, it will check updates on all MDTs,
  39  * if some updates records are missing on some MDTs, the replay thread will redo
  40  * updates on these MDTs.
  41  *
  42  * Author: Di Wang <di.wang@intel.com>
  43  */
  44 #define DEBUG_SUBSYSTEM S_CLASS
  45
  46 #include <lu_target.h>
  47 #include <md_object.h>
  48 #include <lustre_update.h>
  49 #include <obd.h>
  50 #include <obd_class.h>
  51 #include "tgt_internal.h"
  52
  53 /**
  54  * Lookup distribute_txn_replay req
  55  *
  56  * Lookup distribute_txn_replay in the replay list by batchid.
  57  * It is assumed the list has been locked before calling this function.
  58  *
  59  * \param[in] tdtd      distribute_txn_data, which holds the replay
  60  *                      list.
  61  * \param[in] batchid   batchid used by lookup.
  62  *
  63  * \retval              pointer of the replay if succeeds.
  64  * \retval              NULL if can not find it.
  65  */
  66 static struct distribute_txn_replay_req *
  67 dtrq_lookup(struct target_distribute_txn_data *tdtd, __u64 batchid)
  68 {
  69         struct distribute_txn_replay_req        *tmp;
  70         struct distribute_txn_replay_req        *dtrq = NULL;
  71
  72         list_for_each_entry(tmp, &tdtd->tdtd_replay_list, dtrq_list) {
  73                 if (tmp->dtrq_lur->lur_update_rec.ur_batchid == batchid) {
  74                         dtrq = tmp;
  75                         break;
  76                 }
  77         }
  78         return dtrq;
  79 }
  80
  81 /**
  82  * insert distribute txn replay req
  83  *
  84  * Insert distribute txn replay to the replay list, and it assumes the
  85  * list has been looked. Note: the replay list is a sorted list, which
  86  * is sorted by master transno. It is assumed the replay list has been
  87  * locked before calling this function.
  88  *
  89  * \param[in] tdtd      target distribute txn data where replay list is
  90  * \param[in] new       distribute txn replay to be inserted
  91  *
  92  * \retval              0 if insertion succeeds
  93  * \retval              EEXIST if the dtrq already exists
  94  */
  95 static int dtrq_insert(struct target_distribute_txn_data *tdtd,
  96                         struct distribute_txn_replay_req *new)
  97 {
  98         struct distribute_txn_replay_req *iter;
  99
 100         list_for_each_entry_reverse(iter, &tdtd->tdtd_replay_list, dtrq_list) {
 101                 if (iter->dtrq_lur->lur_update_rec.ur_master_transno >
 102                     new->dtrq_lur->lur_update_rec.ur_master_transno)
 103                         continue;
 104
 105                 /* If there are mulitple replay req with same transno, then
 106                  * sort them with batchid */
 107                 if (iter->dtrq_lur->lur_update_rec.ur_master_transno ==
 108                     new->dtrq_lur->lur_update_rec.ur_master_transno &&
 109                     iter->dtrq_lur->lur_update_rec.ur_batchid ==
 110                     new->dtrq_lur->lur_update_rec.ur_batchid)
 111                         return -EEXIST;
 112
 113                 /* If there are mulitple replay req with same transno, then
 114                  * sort them with batchid */
 115                 if (iter->dtrq_lur->lur_update_rec.ur_master_transno ==
 116                     new->dtrq_lur->lur_update_rec.ur_master_transno &&
 117                     iter->dtrq_lur->lur_update_rec.ur_batchid >
 118                     new->dtrq_lur->lur_update_rec.ur_batchid)
 119                         continue;
 120
 121                 list_add(&new->dtrq_list, &iter->dtrq_list);
 122                 break;
 123         }
 124
 125         if (list_empty(&new->dtrq_list))
 126                 list_add(&new->dtrq_list, &tdtd->tdtd_replay_list);
 127
 128         return 0;
 129 }
 130
 131 /**
 132  * create distribute txn replay req
 133  *
 134  * Allocate distribute txn replay req according to the update records.
 135  *
 136  * \param[in] tdtd      target distribute txn data where replay list is.
 137  * \param[in] record    update records from the update log.
 138  *
 139  * \retval              the pointer of distribute txn replay req if
 140  *                      the creation succeeds.
 141  * \retval              NULL if the creation fails.
 142  */
 143 static struct distribute_txn_replay_req *
 144 dtrq_create(struct llog_update_record *lur)
 145 {
 146         struct distribute_txn_replay_req *new;
 147
 148         OBD_ALLOC_PTR(new);
 149         if (new == NULL)
 150                 RETURN(ERR_PTR(-ENOMEM));
 151
 152         new->dtrq_lur_size = llog_update_record_size(lur);
 153         OBD_ALLOC_LARGE(new->dtrq_lur, new->dtrq_lur_size);
 154         if (new->dtrq_lur == NULL) {
 155                 OBD_FREE_PTR(new);
 156                 RETURN(ERR_PTR(-ENOMEM));
 157         }
 158
 159         memcpy(new->dtrq_lur, lur, new->dtrq_lur_size);
 160
 161         spin_lock_init(&new->dtrq_sub_list_lock);
 162         INIT_LIST_HEAD(&new->dtrq_sub_list);
 163         INIT_LIST_HEAD(&new->dtrq_list);
 164
 165         RETURN(new);
 166 }
 167
 168 /**
 169  * Lookup distribute sub replay
 170  *
 171  * Lookup distribute sub replay in the sub list of distribute_txn_replay by
 172  * mdt_index.
 173  *
 174  * \param[in] distribute_txn_replay_req the distribute txn replay req to lookup
 175  * \param[in] mdt_index                 the mdt_index as the key of lookup
 176  *
 177  * \retval              the pointer of sub replay if it can be found.
 178  * \retval              NULL if it can not find.
 179  */
 180 struct distribute_txn_replay_req_sub *
 181 dtrq_sub_lookup(struct distribute_txn_replay_req *dtrq, __u32 mdt_index)
 182 {
 183         struct distribute_txn_replay_req_sub *dtrqs = NULL;
 184         struct distribute_txn_replay_req_sub *tmp;
 185
 186         list_for_each_entry(tmp, &dtrq->dtrq_sub_list, dtrqs_list) {
 187                 if (tmp->dtrqs_mdt_index == mdt_index) {
 188                         dtrqs = tmp;
 189                         break;
 190                 }
 191         }
 192         return dtrqs;
 193 }
 194
 195 /**
 196  * Try to add cookie to sub distribute txn request
 197  *
 198  * Check if the update log cookie has been added to the request, if not,
 199  * add it to the dtrqs_cookie_list.
 200  *
 201  * \param[in] dtrqs     sub replay req where cookies to be added.
 202  * \param[in] cookie    cookie to be added.
 203  *
 204  * \retval              0 if the cookie is adding succeeds.
 205  * \retval              negative errno if adding fails.
 206  */
 207 static int dtrq_sub_add_cookie(struct distribute_txn_replay_req_sub *dtrqs,
 208                                struct llog_cookie *cookie)
 209 {
 210         struct sub_thandle_cookie *new;
 211
 212         OBD_ALLOC_PTR(new);
 213         if (new == NULL)
 214                 return -ENOMEM;
 215
 216         INIT_LIST_HEAD(&new->stc_list);
 217         new->stc_cookie = *cookie;
 218         /* Note: only single thread will access one sub_request each time,
 219          * so no need lock here */
 220         list_add(&new->stc_list, &dtrqs->dtrqs_cookie_list);
 221
 222         return 0;
 223 }
 224
 225 /**
 226  * Insert distribute txn sub req replay
 227  *
 228  * Allocate sub replay req and insert distribute txn replay list.
 229  *
 230  * \param[in] dtrq      d to be added
 231  * \param[in] cookie    the cookie of the update record
 232  * \param[in] mdt_index the mdt_index of the update record
 233  *
 234  * \retval              0 if the adding succeeds.
 235  * \retval              negative errno if the adding fails.
 236  */
 237 static int
 238 dtrq_sub_create_and_insert(struct distribute_txn_replay_req *dtrq,
 239                            struct llog_cookie *cookie,
 240                            __u32 mdt_index)
 241 {
 242         struct distribute_txn_replay_req_sub    *dtrqs = NULL;
 243         struct distribute_txn_replay_req_sub    *new;
 244         int                                     rc;
 245         ENTRY;
 246
 247         spin_lock(&dtrq->dtrq_sub_list_lock);
 248         dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
 249         spin_unlock(&dtrq->dtrq_sub_list_lock);
 250         if (dtrqs != NULL) {
 251                 rc = dtrq_sub_add_cookie(dtrqs, cookie);
 252                 RETURN(0);
 253         }
 254
 255         OBD_ALLOC_PTR(new);
 256         if (new == NULL)
 257                 RETURN(-ENOMEM);
 258
 259         INIT_LIST_HEAD(&new->dtrqs_list);
 260         INIT_LIST_HEAD(&new->dtrqs_cookie_list);
 261         new->dtrqs_mdt_index = mdt_index;
 262         spin_lock(&dtrq->dtrq_sub_list_lock);
 263         dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
 264         if (dtrqs == NULL) {
 265                 list_add(&new->dtrqs_list, &dtrq->dtrq_sub_list);
 266                 dtrqs = new;
 267         } else {
 268                 OBD_FREE_PTR(new);
 269         }
 270         spin_unlock(&dtrq->dtrq_sub_list_lock);
 271
 272         rc = dtrq_sub_add_cookie(dtrqs, cookie);
 273
 274         RETURN(rc);
 275 }
 276
 277 /**
 278  * append updates to the current replay updates
 279  *
 280  * Append more updates to the existent replay update. And this is only
 281  * used when combining mulitple updates into one large updates during
 282  * replay.
 283  *
 284  * \param[in] dtrq      the update replay request where the new update
 285  *                      records will be added.
 286  * \param[in] lur       the new update record.
 287  *
 288  * \retval              0 if appending succeeds.
 289  * \retval              negative errno if appending fails.
 290  */
 291 static int dtrq_append_updates(struct distribute_txn_replay_req *dtrq,
 292                                struct update_records *record)
 293 {
 294         struct llog_update_record *new_lur;
 295         size_t lur_size = dtrq->dtrq_lur_size;
 296         void *ptr;
 297         ENTRY;
 298
 299         /* Because several threads might retrieve the same records from
 300          * different targets, and we only need one copy of records. So
 301          * we will check if the records is in the next one, if not, just
 302          * skip it */
 303         spin_lock(&dtrq->dtrq_sub_list_lock);
 304         if (dtrq->dtrq_lur->lur_update_rec.ur_index + 1 != record->ur_index) {
 305                 spin_unlock(&dtrq->dtrq_sub_list_lock);
 306                 RETURN(0);
 307         }
 308         dtrq->dtrq_lur->lur_update_rec.ur_index++;
 309         spin_unlock(&dtrq->dtrq_sub_list_lock);
 310
 311         lur_size += update_records_size(record);
 312         OBD_ALLOC_LARGE(new_lur, lur_size);
 313         if (new_lur == NULL) {
 314                 spin_lock(&dtrq->dtrq_sub_list_lock);
 315                 dtrq->dtrq_lur->lur_update_rec.ur_index--;
 316                 spin_unlock(&dtrq->dtrq_sub_list_lock);
 317                 RETURN(-ENOMEM);
 318         }
 319
 320         /* Copy the old and new records to the new allocated buffer */
 321         memcpy(new_lur, dtrq->dtrq_lur, dtrq->dtrq_lur_size);
 322         ptr = (char *)&new_lur->lur_update_rec +
 323                 update_records_size(&new_lur->lur_update_rec);
 324         memcpy(ptr, &record->ur_ops,
 325                update_records_size(record) -
 326                offsetof(struct update_records, ur_ops));
 327
 328         new_lur->lur_update_rec.ur_update_count += record->ur_update_count;
 329         new_lur->lur_update_rec.ur_param_count += record->ur_param_count;
 330         new_lur->lur_hdr.lrh_len = llog_update_record_size(new_lur);
 331
 332         /* Replace the records */
 333         OBD_FREE_LARGE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
 334         dtrq->dtrq_lur = new_lur;
 335         dtrq->dtrq_lur_size = lur_size;
 336         dtrq->dtrq_lur->lur_update_rec.ur_flags = record->ur_flags;
 337         update_records_dump(&new_lur->lur_update_rec, D_INFO, true);
 338         RETURN(0);
 339 }
 340
 341 /**
 342  * Insert update records to the replay list.
 343  *
 344  * Allocate distribute txn replay req and insert it into the replay
 345  * list, then insert the update records into the replay req.
 346  *
 347  * \param[in] tdtd      distribute txn replay data where the replay list
 348  *                      is.
 349  * \param[in] record    the update record
 350  * \param[in] cookie    cookie of the record
 351  * \param[in] index     mdt index of the record
 352  *
 353  * \retval              0 if the adding succeeds.
 354  * \retval              negative errno if the adding fails.
 355  */
 356 int
 357 insert_update_records_to_replay_list(struct target_distribute_txn_data *tdtd,
 358                                      struct llog_update_record *lur,
 359                                      struct llog_cookie *cookie,
 360                                      __u32 mdt_index)
 361 {
 362         struct distribute_txn_replay_req *dtrq;
 363         struct update_records            *record = &lur->lur_update_rec;
 364         int rc = 0;
 365         ENTRY;
 366
 367         CDEBUG(D_HA, "%s: insert record batchid = "LPU64" transno = "LPU64
 368                " mdt_index %u\n", tdtd->tdtd_lut->lut_obd->obd_name,
 369                record->ur_batchid, record->ur_master_transno, mdt_index);
 370
 371         /* First try to build the replay update request with the records */
 372         spin_lock(&tdtd->tdtd_replay_list_lock);
 373         dtrq = dtrq_lookup(tdtd, record->ur_batchid);
 374         spin_unlock(&tdtd->tdtd_replay_list_lock);
 375         if (dtrq == NULL) {
 376                 /* If the transno in the update record is 0, it means the
 377                  * update are from master MDT, and we will use the master
 378                  * last committed transno as its batchid. Note: if it got
 379                  * the records from the slave later, it needs to update
 380                  * the batchid by the transno in slave update log (see below) */
 381                 dtrq = dtrq_create(lur);
 382                 if (IS_ERR(dtrq))
 383                         RETURN(PTR_ERR(dtrq));
 384
 385                 if (record->ur_master_transno == 0)
 386                         dtrq->dtrq_lur->lur_update_rec.ur_master_transno =
 387                                 tdtd->tdtd_lut->lut_last_transno;
 388                 spin_lock(&tdtd->tdtd_replay_list_lock);
 389                 rc = dtrq_insert(tdtd, dtrq);
 390                 spin_unlock(&tdtd->tdtd_replay_list_lock);
 391                 if (rc == -EEXIST) {
 392                         /* Some one else already add the record */
 393                         dtrq_destroy(dtrq);
 394                         rc = 0;
 395                 }
 396         } else {
 397                 struct update_records *dtrq_rec;
 398
 399                 /* If the master transno in update header is not
 400                  * matched with the one in the record, then it means
 401                  * the dtrq is originally created by master record,
 402                  * and we need update master transno and reposition
 403                  * the dtrq(by master transno). */
 404                 dtrq_rec = &dtrq->dtrq_lur->lur_update_rec;
 405                 if (record->ur_master_transno != 0 &&
 406                     dtrq_rec->ur_master_transno != record->ur_master_transno) {
 407                         dtrq_rec->ur_master_transno = record->ur_master_transno;
 408                         spin_lock(&tdtd->tdtd_replay_list_lock);
 409                         list_del_init(&dtrq->dtrq_list);
 410                         rc = dtrq_insert(tdtd, dtrq);
 411                         spin_unlock(&tdtd->tdtd_replay_list_lock);
 412                         if (rc < 0)
 413                                 return rc;
 414                 }
 415
 416                 /* This is a partial update records, let's try to append
 417                  * the record to the current replay request */
 418                 if (record->ur_flags & UPDATE_RECORD_CONTINUE)
 419                         rc = dtrq_append_updates(dtrq, record);
 420         }
 421
 422         /* Then create and add sub update request */
 423         rc = dtrq_sub_create_and_insert(dtrq, cookie, mdt_index);
 424
 425         RETURN(rc);
 426 }
 427 EXPORT_SYMBOL(insert_update_records_to_replay_list);
 428
 429 /**
 430  * Dump updates of distribute txns.
 431  *
 432  * Output all of recovery updates in the distribute txn list to the
 433  * debug log.
 434  *
 435  * \param[in] tdtd      distribute txn data where all of distribute txn
 436  *                      are listed.
 437  * \param[in] mask      debug mask
 438  */
 439 void dtrq_list_dump(struct target_distribute_txn_data *tdtd, unsigned int mask)
 440 {
 441         struct distribute_txn_replay_req *dtrq;
 442
 443         spin_lock(&tdtd->tdtd_replay_list_lock);
 444         list_for_each_entry(dtrq, &tdtd->tdtd_replay_list, dtrq_list)
 445                 update_records_dump(&dtrq->dtrq_lur->lur_update_rec, mask,
 446                                     false);
 447         spin_unlock(&tdtd->tdtd_replay_list_lock);
 448 }
 449 EXPORT_SYMBOL(dtrq_list_dump);
 450
 451 /**
 452  * Destroy distribute txn replay req
 453  *
 454  * Destroy distribute txn replay req and all of subs.
 455  *
 456  * \param[in] dtrq      distribute txn replqy req to be destroyed.
 457  */
 458 void dtrq_destroy(struct distribute_txn_replay_req *dtrq)
 459 {
 460         struct distribute_txn_replay_req_sub    *dtrqs;
 461         struct distribute_txn_replay_req_sub    *tmp;
 462
 463         LASSERT(list_empty(&dtrq->dtrq_list));
 464         spin_lock(&dtrq->dtrq_sub_list_lock);
 465         list_for_each_entry_safe(dtrqs, tmp, &dtrq->dtrq_sub_list, dtrqs_list) {
 466                 struct sub_thandle_cookie *stc;
 467                 struct sub_thandle_cookie *tmp;
 468
 469                 list_del(&dtrqs->dtrqs_list);
 470                 list_for_each_entry_safe(stc, tmp, &dtrqs->dtrqs_cookie_list,
 471                                          stc_list) {
 472                         list_del(&stc->stc_list);
 473                         OBD_FREE_PTR(stc);
 474                 }
 475                 OBD_FREE_PTR(dtrqs);
 476         }
 477         spin_unlock(&dtrq->dtrq_sub_list_lock);
 478
 479         if (dtrq->dtrq_lur != NULL)
 480                 OBD_FREE_LARGE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
 481
 482         OBD_FREE_PTR(dtrq);
 483 }
 484 EXPORT_SYMBOL(dtrq_destroy);
 485
 486 /**
 487  * Destroy all of replay req.
 488  *
 489  * Destroy all of replay req in the replay list.
 490  *
 491  * \param[in] tdtd      target distribute txn data where the replay list is.
 492  */
 493 void dtrq_list_destroy(struct target_distribute_txn_data *tdtd)
 494 {
 495         struct distribute_txn_replay_req *dtrq;
 496         struct distribute_txn_replay_req *tmp;
 497
 498         spin_lock(&tdtd->tdtd_replay_list_lock);
 499         list_for_each_entry_safe(dtrq, tmp, &tdtd->tdtd_replay_list,
 500                                  dtrq_list) {
 501                 list_del_init(&dtrq->dtrq_list);
 502                 dtrq_destroy(dtrq);
 503         }
 504         spin_unlock(&tdtd->tdtd_replay_list_lock);
 505 }
 506 EXPORT_SYMBOL(dtrq_list_destroy);
 507
 508 /**
 509  * Get next req in the replay list
 510  *
 511  * Get next req needs to be replayed, since it is a sorted list
 512  * (by master MDT transno)
 513  *
 514  * \param[in] tdtd      distribute txn data where the replay list is
 515  *
 516  * \retval              the pointer of update recovery header
 517  */
 518 struct distribute_txn_replay_req *
 519 distribute_txn_get_next_req(struct target_distribute_txn_data *tdtd)
 520 {
 521         struct distribute_txn_replay_req *dtrq = NULL;
 522
 523         spin_lock(&tdtd->tdtd_replay_list_lock);
 524         if (!list_empty(&tdtd->tdtd_replay_list)) {
 525                 dtrq = list_entry(tdtd->tdtd_replay_list.next,
 526                                  struct distribute_txn_replay_req, dtrq_list);
 527                 list_del_init(&dtrq->dtrq_list);
 528         }
 529         spin_unlock(&tdtd->tdtd_replay_list_lock);
 530
 531         return dtrq;
 532 }
 533 EXPORT_SYMBOL(distribute_txn_get_next_req);
 534
 535 /**
 536  * Get next transno in the replay list, because this is the sorted
 537  * list, so it will return the transno of next req in the list.
 538  *
 539  * \param[in] tdtd      distribute txn data where the replay list is
 540  *
 541  * \retval              the transno of next update in the list
 542  */
 543 __u64 distribute_txn_get_next_transno(struct target_distribute_txn_data *tdtd)
 544 {
 545         struct distribute_txn_replay_req        *dtrq = NULL;
 546         __u64                                   transno = 0;
 547
 548         spin_lock(&tdtd->tdtd_replay_list_lock);
 549         if (!list_empty(&tdtd->tdtd_replay_list)) {
 550                 dtrq = list_entry(tdtd->tdtd_replay_list.next,
 551                                  struct distribute_txn_replay_req, dtrq_list);
 552                 transno = dtrq->dtrq_lur->lur_update_rec.ur_master_transno;
 553         }
 554         spin_unlock(&tdtd->tdtd_replay_list_lock);
 555
 556         CDEBUG(D_HA, "%s: Next update transno "LPU64"\n",
 557                tdtd->tdtd_lut->lut_obd->obd_name, transno);
 558         return transno;
 559 }
 560 EXPORT_SYMBOL(distribute_txn_get_next_transno);
 561
 562 /**
 563  * Check if the update of one object is committed
 564  *
 565  * Check whether the update for the object is committed by checking whether
 566  * the correspondent sub exists in the replay req. If it is committed, mark
 567  * the committed flag in correspondent the sub thandle.
 568  *
 569  * \param[in] env       execution environment
 570  * \param[in] dtrq      replay request
 571  * \param[in] dt_obj    object for the update
 572  * \param[in] top_th    top thandle
 573  * \param[in] sub_th    sub thandle which the update belongs to
 574  *
 575  * \retval              1 if the update is not committed.
 576  * \retval              0 if the update is committed.
 577  * \retval              negative errno if some other failures happen.
 578  */
 579 static int update_is_committed(const struct lu_env *env,
 580                                struct distribute_txn_replay_req *dtrq,
 581                                struct dt_object *dt_obj,
 582                                struct top_thandle *top_th,
 583                                struct sub_thandle *st)
 584 {
 585         struct seq_server_site  *seq_site;
 586         const struct lu_fid     *fid = lu_object_fid(&dt_obj->do_lu);
 587         struct distribute_txn_replay_req_sub    *dtrqs;
 588         __u32                   mdt_index;
 589         ENTRY;
 590
 591         if (st->st_sub_th != NULL)
 592                 RETURN(1);
 593
 594         if (st->st_committed)
 595                 RETURN(0);
 596
 597         seq_site = lu_site2seq(dt_obj->do_lu.lo_dev->ld_site);
 598         if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
 599                 mdt_index = fid_oid(fid);
 600         } else if (!fid_seq_in_fldb(fid_seq(fid))) {
 601                 mdt_index = seq_site->ss_node_id;
 602         } else {
 603                 struct lu_server_fld *fld;
 604                 struct lu_seq_range range = {0};
 605                 int rc;
 606
 607                 fld = seq_site->ss_server_fld;
 608                 fld_range_set_type(&range, LU_SEQ_RANGE_MDT);
 609                 LASSERT(fld->lsf_seq_lookup != NULL);
 610                 rc = fld->lsf_seq_lookup(env, fld, fid_seq(fid),
 611                                          &range);
 612                 if (rc < 0)
 613                         RETURN(rc);
 614                 mdt_index = range.lsr_index;
 615         }
 616
 617         dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
 618         if (dtrqs != NULL || top_th->tt_multiple_thandle->tmt_committed) {
 619                 st->st_committed = 1;
 620                 if (dtrqs != NULL) {
 621                         struct sub_thandle_cookie *stc;
 622                         struct sub_thandle_cookie *tmp;
 623
 624                         list_for_each_entry_safe(stc, tmp,
 625                                                  &dtrqs->dtrqs_cookie_list,
 626                                                  stc_list)
 627                                 list_move(&stc->stc_list, &st->st_cookie_list);
 628                 }
 629                 RETURN(0);
 630         }
 631
 632         CDEBUG(D_HA, "Update of "DFID "on MDT%u is not committed\n", PFID(fid),
 633                mdt_index);
 634
 635         RETURN(1);
 636 }
 637
 638 /**
 639  * Implementation of different update methods for update recovery.
 640  *
 641  * These following functions update_recovery_$(update_name) implement
 642  * different updates recovery methods. They will extract the parameters
 643  * from the common parameters area and call correspondent dt API to redo
 644  * the update.
 645  *
 646  * \param[in] env       execution environment
 647  * \param[in] op        update operation to be replayed
 648  * \param[in] params    common update parameters which holds all parameters
 649  *                      of the operation
 650  * \param[in] th        transaction handle
 651  * \param[in] declare   indicate it will do declare or real execution, true
 652  *                      means declare, false means real execution
 653  *
 654  * \retval              0 if it succeeds.
 655  * \retval              negative errno if it fails.
 656  */
 657 static int update_recovery_create(const struct lu_env *env,
 658                                   struct dt_object *dt_obj,
 659                                   const struct update_op *op,
 660                                   const struct update_params *params,
 661                                   struct thandle_exec_args *ta,
 662                                   struct thandle *th)
 663 {
 664         struct update_thread_info *uti = update_env_info(env);
 665         struct llog_update_record *lur = uti->uti_dtrq->dtrq_lur;
 666         struct lu_attr          *attr = &uti->uti_attr;
 667         struct obdo             *wobdo;
 668         struct obdo             *lobdo = &uti->uti_obdo;
 669         struct dt_object_format dof;
 670         __u16                   size;
 671         unsigned int            param_count;
 672         int rc;
 673         ENTRY;
 674
 675         if (dt_object_exists(dt_obj))
 676                 RETURN(-EEXIST);
 677
 678         param_count = lur->lur_update_rec.ur_param_count;
 679         wobdo = update_params_get_param_buf(params, op->uop_params_off[0],
 680                                             param_count, &size);
 681         if (wobdo == NULL)
 682                 RETURN(-EIO);
 683         if (size != sizeof(*wobdo))
 684                 RETURN(-EIO);
 685
 686         if (LLOG_REC_HDR_NEEDS_SWABBING(&lur->lur_hdr))
 687                 lustre_swab_obdo(wobdo);
 688
 689         lustre_get_wire_obdo(NULL, lobdo, wobdo);
 690         la_from_obdo(attr, lobdo, lobdo->o_valid);
 691
 692         dof.dof_type = dt_mode_to_dft(attr->la_mode);
 693
 694         rc = out_tx_create(env, dt_obj, attr, NULL, &dof,
 695                            ta, th, NULL, 0);
 696
 697         RETURN(rc);
 698 }
 699
 700 static int update_recovery_destroy(const struct lu_env *env,
 701                                    struct dt_object *dt_obj,
 702                                    const struct update_op *op,
 703                                    const struct update_params *params,
 704                                    struct thandle_exec_args *ta,
 705                                    struct thandle *th)
 706 {
 707         int rc;
 708         ENTRY;
 709
 710         rc = out_tx_destroy(env, dt_obj, ta, th, NULL, 0);
 711
 712         RETURN(rc);
 713 }
 714
 715 static int update_recovery_ref_add(const struct lu_env *env,
 716                                    struct dt_object *dt_obj,
 717                                    const struct update_op *op,
 718                                    const struct update_params *params,
 719                                    struct thandle_exec_args *ta,
 720                                    struct thandle *th)
 721 {
 722         int rc;
 723         ENTRY;
 724
 725         rc = out_tx_ref_add(env, dt_obj, ta, th, NULL, 0);
 726
 727         RETURN(rc);
 728 }
 729
 730 static int update_recovery_ref_del(const struct lu_env *env,
 731                                    struct dt_object *dt_obj,
 732                                    const struct update_op *op,
 733                                    const struct update_params *params,
 734                                    struct thandle_exec_args *ta,
 735                                    struct thandle *th)
 736 {
 737         int rc;
 738         ENTRY;
 739
 740         rc = out_tx_ref_del(env, dt_obj, ta, th, NULL, 0);
 741
 742         RETURN(rc);
 743 }
 744
 745 static int update_recovery_attr_set(const struct lu_env *env,
 746                                     struct dt_object *dt_obj,
 747                                     const struct update_op *op,
 748                                     const struct update_params *params,
 749                                     struct thandle_exec_args *ta,
 750                                     struct thandle *th)
 751 {
 752         struct update_thread_info *uti = update_env_info(env);
 753         struct llog_update_record *lur = uti->uti_dtrq->dtrq_lur;
 754         struct obdo     *wobdo;
 755         struct obdo     *lobdo = &uti->uti_obdo;
 756         struct lu_attr  *attr = &uti->uti_attr;
 757         __u16           size;
 758         unsigned int    param_count;
 759         int             rc;
 760         ENTRY;
 761
 762         param_count = lur->lur_update_rec.ur_param_count;
 763         wobdo = update_params_get_param_buf(params, op->uop_params_off[0],
 764                                             param_count, &size);
 765         if (wobdo == NULL)
 766                 RETURN(-EIO);
 767         if (size != sizeof(*wobdo))
 768                 RETURN(-EIO);
 769
 770         if (LLOG_REC_HDR_NEEDS_SWABBING(&lur->lur_hdr))
 771                 lustre_swab_obdo(wobdo);
 772
 773         lustre_get_wire_obdo(NULL, lobdo, wobdo);
 774         la_from_obdo(attr, lobdo, lobdo->o_valid);
 775
 776         rc = out_tx_attr_set(env, dt_obj, attr, ta, th, NULL, 0);
 777
 778         RETURN(rc);
 779 }
 780
 781 static int update_recovery_xattr_set(const struct lu_env *env,
 782                                      struct dt_object *dt_obj,
 783                                      const struct update_op *op,
 784                                      const struct update_params *params,
 785                                      struct thandle_exec_args *ta,
 786                                      struct thandle *th)
 787 {
 788         struct update_thread_info *uti = update_env_info(env);
 789         char            *buf;
 790         char            *name;
 791         int             fl;
 792         __u16           size;
 793         __u32           param_count;
 794         int             rc;
 795         ENTRY;
 796
 797         param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
 798         name = update_params_get_param_buf(params,
 799                                            op->uop_params_off[0],
 800                                            param_count, &size);
 801         if (name == NULL)
 802                 RETURN(-EIO);
 803
 804         buf = update_params_get_param_buf(params,
 805                                           op->uop_params_off[1],
 806                                           param_count, &size);
 807         if (buf == NULL)
 808                 RETURN(-EIO);
 809
 810         uti->uti_buf.lb_buf = buf;
 811         uti->uti_buf.lb_len = (size_t)size;
 812
 813         buf = update_params_get_param_buf(params, op->uop_params_off[2],
 814                                           param_count, &size);
 815         if (buf == NULL)
 816                 RETURN(-EIO);
 817         if (size != sizeof(fl))
 818                 RETURN(-EIO);
 819
 820         fl = le32_to_cpu(*(int *)buf);
 821
 822         rc = out_tx_xattr_set(env, dt_obj, &uti->uti_buf, name, fl, ta, th,
 823                               NULL, 0);
 824
 825         RETURN(rc);
 826 }
 827
 828 static int update_recovery_index_insert(const struct lu_env *env,
 829                                         struct dt_object *dt_obj,
 830                                         const struct update_op *op,
 831                                         const struct update_params *params,
 832                                         struct thandle_exec_args *ta,
 833                                         struct thandle *th)
 834 {
 835         struct update_thread_info *uti = update_env_info(env);
 836         struct lu_fid           *fid;
 837         char                    *name;
 838         __u32                   param_count;
 839         __u32                   *ptype;
 840         __u32                   type;
 841         __u16                   size;
 842         int rc;
 843         ENTRY;
 844
 845         param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
 846         name = update_params_get_param_buf(params, op->uop_params_off[0],
 847                                            param_count, &size);
 848         if (name == NULL)
 849                 RETURN(-EIO);
 850
 851         fid = update_params_get_param_buf(params, op->uop_params_off[1],
 852                                           param_count, &size);
 853         if (fid == NULL)
 854                 RETURN(-EIO);
 855         if (size != sizeof(*fid))
 856                 RETURN(-EIO);
 857
 858         fid_le_to_cpu(fid, fid);
 859
 860         ptype = update_params_get_param_buf(params, op->uop_params_off[2],
 861                                             param_count, &size);
 862         if (ptype == NULL)
 863                 RETURN(-EIO);
 864         if (size != sizeof(*ptype))
 865                 RETURN(-EIO);
 866         type = le32_to_cpu(*ptype);
 867
 868         if (dt_try_as_dir(env, dt_obj) == 0)
 869                 RETURN(-ENOTDIR);
 870
 871         uti->uti_rec.rec_fid = fid;
 872         uti->uti_rec.rec_type = type;
 873
 874         rc = out_tx_index_insert(env, dt_obj,
 875                                  (const struct dt_rec *)&uti->uti_rec,
 876                                  (const struct dt_key *)name, ta, th,
 877                                  NULL, 0);
 878
 879         RETURN(rc);
 880 }
 881
 882 static int update_recovery_index_delete(const struct lu_env *env,
 883                                         struct dt_object *dt_obj,
 884                                         const struct update_op *op,
 885                                         const struct update_params *params,
 886                                         struct thandle_exec_args *ta,
 887                                         struct thandle *th)
 888 {
 889         struct update_thread_info *uti = update_env_info(env);
 890         __u32   param_count;
 891         char    *name;
 892         __u16   size;
 893         int     rc;
 894         ENTRY;
 895
 896         param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
 897         name = update_params_get_param_buf(params, op->uop_params_off[0],
 898                                            param_count, &size);
 899         if (name == NULL)
 900                 RETURN(-EIO);
 901
 902         if (dt_try_as_dir(env, dt_obj) == 0)
 903                 RETURN(-ENOTDIR);
 904
 905         rc = out_tx_index_delete(env, dt_obj,
 906                                  (const struct dt_key *)name, ta, th, NULL, 0);
 907
 908         RETURN(rc);
 909 }
 910
 911 static int update_recovery_write(const struct lu_env *env,
 912                                  struct dt_object *dt_obj,
 913                                  const struct update_op *op,
 914                                  const struct update_params *params,
 915                                  struct thandle_exec_args *ta,
 916                                  struct thandle *th)
 917 {
 918         struct update_thread_info *uti = update_env_info(env);
 919         char            *buf;
 920         __u32           param_count;
 921         __u64           pos;
 922         __u16           size;
 923         int rc;
 924         ENTRY;
 925
 926         param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
 927         buf = update_params_get_param_buf(params, op->uop_params_off[0],
 928                                           param_count, &size);
 929         if (buf == NULL)
 930                 RETURN(-EIO);
 931
 932         uti->uti_buf.lb_buf = buf;
 933         uti->uti_buf.lb_len = size;
 934
 935         buf = update_params_get_param_buf(params, op->uop_params_off[1],
 936                                           param_count, &size);
 937         if (buf == NULL)
 938                 RETURN(-EIO);
 939
 940         pos = le64_to_cpu(*(__u64 *)buf);
 941
 942         rc = out_tx_write(env, dt_obj, &uti->uti_buf, pos,
 943                           ta, th, NULL, 0);
 944
 945         RETURN(rc);
 946 }
 947
 948 static int update_recovery_xattr_del(const struct lu_env *env,
 949                                      struct dt_object *dt_obj,
 950                                      const struct update_op *op,
 951                                      const struct update_params *params,
 952                                      struct thandle_exec_args *ta,
 953                                      struct thandle *th)
 954 {
 955         struct update_thread_info *uti = update_env_info(env);
 956         __u32   param_count;
 957         char    *name;
 958         __u16   size;
 959         int     rc;
 960         ENTRY;
 961
 962         param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
 963         name = update_params_get_param_buf(params, op->uop_params_off[0],
 964                                            param_count, &size);
 965         if (name == NULL)
 966                 RETURN(-EIO);
 967
 968         rc = out_tx_xattr_del(env, dt_obj, name, ta, th, NULL, 0);
 969
 970         RETURN(rc);
 971 }
 972
 973 /**
 974  * Execute updates in the update replay records
 975  *
 976  * Declare distribute txn replay by update records and add the updates
 977  * to the execution list. Note: it will check if the update has been
 978  * committed, and only execute the updates if it is not committed to
 979  * disk.
 980  *
 981  * \param[in] env       execution environment
 982  * \param[in] tdtd      distribute txn replay data which hold all of replay
 983  *                      reqs and all replay parameters.
 984  * \param[in] dtrq      distribute transaction replay req.
 985  * \param[in] ta        thandle execute args.
 986  *
 987  * \retval              0 if declare succeeds.
 988  * \retval              negative errno if declare fails.
 989  */
 990 static int update_recovery_exec(const struct lu_env *env,
 991                                 struct target_distribute_txn_data *tdtd,
 992                                 struct distribute_txn_replay_req *dtrq,
 993                                 struct thandle_exec_args *ta)
 994 {
 995         struct llog_update_record *lur = dtrq->dtrq_lur;
 996         struct update_records   *records = &lur->lur_update_rec;
 997         struct update_ops       *ops = &records->ur_ops;
 998         struct update_params    *params = update_records_get_params(records);
 999         struct top_thandle      *top_th = container_of(ta->ta_handle,
1000                                                        struct top_thandle,
1001                                                        tt_super);
1002         struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
1003         struct update_op        *op;
1004         unsigned int            i;
1005         int                     rc = 0;
1006         ENTRY;
1007
1008         /* These records have been swabbed in llog_cat_process() */
1009         for (i = 0, op = &ops->uops_op[0]; i < records->ur_update_count;
1010              i++, op = update_op_next_op(op)) {
1011                 struct lu_fid           *fid = &op->uop_fid;
1012                 struct dt_object        *dt_obj;
1013                 struct dt_object        *sub_dt_obj;
1014                 struct dt_device        *sub_dt;
1015                 struct sub_thandle      *st;
1016
1017                 if (op->uop_type == OUT_NOOP)
1018                         continue;
1019
1020                 dt_obj = dt_locate(env, tdtd->tdtd_dt, fid);
1021                 if (IS_ERR(dt_obj)) {
1022                         rc = PTR_ERR(dt_obj);
1023                         break;
1024                 }
1025                 sub_dt_obj = dt_object_child(dt_obj);
1026
1027                 /* Create sub thandle if not */
1028                 sub_dt = lu2dt_dev(sub_dt_obj->do_lu.lo_dev);
1029                 st = lookup_sub_thandle(tmt, sub_dt);
1030                 if (st == NULL) {
1031                         st = create_sub_thandle(tmt, sub_dt);
1032                         if (IS_ERR(st))
1033                                 GOTO(next, rc = PTR_ERR(st));
1034                 }
1035
1036                 /* check if updates on the OSD/OSP are committed */
1037                 rc = update_is_committed(env, dtrq, dt_obj, top_th, st);
1038                 if (rc == 0)
1039                         /* If this is committed, goto next */
1040                         goto next;
1041
1042                 if (rc < 0)
1043                         GOTO(next, rc);
1044
1045                 /* Create thandle for sub thandle if needed */
1046                 if (st->st_sub_th == NULL) {
1047                         rc = sub_thandle_trans_create(env, top_th, st);
1048                         if (rc != 0)
1049                                 GOTO(next, rc);
1050                 }
1051
1052                 CDEBUG(D_HA, "replay %uth update\n", i);
1053                 switch (op->uop_type) {
1054                 case OUT_CREATE:
1055                         rc = update_recovery_create(env, sub_dt_obj,
1056                                                     op, params, ta,
1057                                                     st->st_sub_th);
1058                         break;
1059                 case OUT_DESTROY:
1060                         rc = update_recovery_destroy(env, sub_dt_obj,
1061                                                      op, params, ta,
1062                                                      st->st_sub_th);
1063                         break;
1064                 case OUT_REF_ADD:
1065                         rc = update_recovery_ref_add(env, sub_dt_obj,
1066                                                      op, params, ta,
1067                                                      st->st_sub_th);
1068                         break;
1069                 case OUT_REF_DEL:
1070                         rc = update_recovery_ref_del(env, sub_dt_obj,
1071                                                      op, params, ta,
1072                                                      st->st_sub_th);
1073                         break;
1074                 case OUT_ATTR_SET:
1075                         rc = update_recovery_attr_set(env, sub_dt_obj,
1076                                                       op, params, ta,
1077                                                       st->st_sub_th);
1078                         break;
1079                 case OUT_XATTR_SET:
1080                         rc = update_recovery_xattr_set(env, sub_dt_obj,
1081                                                        op, params, ta,
1082                                                        st->st_sub_th);
1083                         break;
1084                 case OUT_INDEX_INSERT:
1085                         rc = update_recovery_index_insert(env, sub_dt_obj,
1086                                                           op, params, ta,
1087                                                           st->st_sub_th);
1088                         break;
1089                 case OUT_INDEX_DELETE:
1090                         rc = update_recovery_index_delete(env, sub_dt_obj,
1091                                                           op, params, ta,
1092                                                           st->st_sub_th);
1093                         break;
1094                 case OUT_WRITE:
1095                         rc = update_recovery_write(env, sub_dt_obj,
1096                                                    op, params, ta,
1097                                                    st->st_sub_th);
1098                         break;
1099                 case OUT_XATTR_DEL:
1100                         rc = update_recovery_xattr_del(env, sub_dt_obj,
1101                                                        op, params, ta,
1102                                                        st->st_sub_th);
1103                         break;
1104                 default:
1105                         CERROR("Unknown update type %u\n", (__u32)op->uop_type);
1106                         rc = -EINVAL;
1107                         break;
1108                 }
1109 next:
1110                 lu_object_put(env, &dt_obj->do_lu);
1111                 if (rc < 0)
1112                         break;
1113         }
1114
1115         ta->ta_handle->th_result = rc;
1116         RETURN(rc);
1117 }
1118
1119 /**
1120  * redo updates on MDT if needed.
1121  *
1122  * During DNE recovery, the recovery thread (target_recovery_thread) will call
1123  * this function to replay distribute txn updates on all MDTs. It only replay
1124  * updates on the MDT where the update record is missing.
1125  *
1126  * If the update already exists on the MDT, then it does not need replay the
1127  * updates on that MDT, and only mark the sub transaction has been committed
1128  * there.
1129  *
1130  * \param[in] env       execution environment
1131  * \param[in] tdtd      target distribute txn data, which holds the replay list
1132  *                      and all parameters needed by replay process.
1133  * \param[in] dtrq      distribute txn replay req.
1134  *
1135  * \retval              0 if replay succeeds.
1136  * \retval              negative errno if replay failes.
1137  */
1138 int distribute_txn_replay_handle(struct lu_env *env,
1139                                  struct target_distribute_txn_data *tdtd,
1140                                  struct distribute_txn_replay_req *dtrq)
1141 {
1142         struct update_records   *records = &dtrq->dtrq_lur->lur_update_rec;
1143         struct thandle_exec_args *ta;
1144         struct lu_context       session_env;
1145         struct thandle          *th = NULL;
1146         struct top_thandle      *top_th;
1147         struct top_multiple_thandle *tmt;
1148         struct thandle_update_records *tur = NULL;
1149         int                     i;
1150         int                     rc = 0;
1151         ENTRY;
1152
1153         /* initialize session, it is needed for the handler of target */
1154         rc = lu_context_init(&session_env, LCT_SERVER_SESSION | LCT_NOREF);
1155         if (rc) {
1156                 CERROR("%s: failure to initialize session: rc = %d\n",
1157                        tdtd->tdtd_lut->lut_obd->obd_name, rc);
1158                 RETURN(rc);
1159         }
1160         lu_context_enter(&session_env);
1161         env->le_ses = &session_env;
1162         lu_env_refill(env);
1163         update_records_dump(records, D_HA, true);
1164         th = top_trans_create(env, NULL);
1165         if (IS_ERR(th))
1166                 GOTO(exit_session, rc = PTR_ERR(th));
1167
1168         ta = &update_env_info(env)->uti_tea;
1169         ta->ta_argno = 0;
1170
1171         update_env_info(env)->uti_dtrq = dtrq;
1172         /* Create distribute transaction structure for this top thandle */
1173         top_th = container_of(th, struct top_thandle, tt_super);
1174         rc = top_trans_create_tmt(env, top_th);
1175         if (rc < 0)
1176                 GOTO(stop_trans, rc);
1177
1178         ta->ta_handle = th;
1179
1180         /* check if the distribute transaction has been committed */
1181         tmt = top_th->tt_multiple_thandle;
1182         tmt->tmt_master_sub_dt = tdtd->tdtd_lut->lut_bottom;
1183         tmt->tmt_batchid = records->ur_batchid;
1184         tgt_th_info(env)->tti_transno = records->ur_master_transno;
1185
1186         if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid)
1187                 tmt->tmt_committed = 1;
1188
1189         rc = update_recovery_exec(env, tdtd, dtrq, ta);
1190         if (rc < 0)
1191                 GOTO(stop_trans, rc);
1192
1193         /* If no updates are needed to be replayed, then
1194          * mark this records as committed, so commit thread
1195          * distribute_txn_commit_thread() will delete the
1196          * record */
1197         if (ta->ta_argno == 0)
1198                 tmt->tmt_committed = 1;
1199
1200         tur = &update_env_info(env)->uti_tur;
1201         tur->tur_update_records = dtrq->dtrq_lur;
1202         tur->tur_update_records_buf_size = dtrq->dtrq_lur_size;
1203         tur->tur_update_params = NULL;
1204         tur->tur_update_param_count = 0;
1205         tmt->tmt_update_records = tur;
1206
1207         distribute_txn_insert_by_batchid(tmt);
1208         rc = top_trans_start(env, NULL, th);
1209         if (rc < 0)
1210                 GOTO(stop_trans, rc);
1211
1212         for (i = 0; i < ta->ta_argno; i++) {
1213                 struct tx_arg           *ta_arg;
1214                 struct dt_object        *dt_obj;
1215                 struct dt_device        *sub_dt;
1216                 struct sub_thandle      *st;
1217
1218                 ta_arg = ta->ta_args[i];
1219                 dt_obj = ta_arg->object;
1220
1221                 LASSERT(tmt->tmt_committed == 0);
1222                 sub_dt = lu2dt_dev(dt_obj->do_lu.lo_dev);
1223                 st = lookup_sub_thandle(tmt, sub_dt);
1224                 LASSERT(st != NULL);
1225                 LASSERT(st->st_sub_th != NULL);
1226                 rc = ta->ta_args[i]->exec_fn(env, st->st_sub_th,
1227                                              ta->ta_args[i]);
1228                 if (unlikely(rc < 0)) {
1229                         CDEBUG(D_HA, "error during execution of #%u from"
1230                                " %s:%d: rc = %d\n", i, ta->ta_args[i]->file,
1231                                ta->ta_args[i]->line, rc);
1232                         while (--i > 0) {
1233                                 if (ta->ta_args[i]->undo_fn != NULL) {
1234                                         dt_obj = ta->ta_args[i]->object;
1235                                         sub_dt =
1236                                                 lu2dt_dev(dt_obj->do_lu.lo_dev);
1237                                         st = lookup_sub_thandle(tmt, sub_dt);
1238                                         LASSERT(st != NULL);
1239                                         LASSERT(st->st_sub_th != NULL);
1240
1241                                         ta->ta_args[i]->undo_fn(env,
1242                                                                st->st_sub_th,
1243                                                                ta->ta_args[i]);
1244                                 } else {
1245                                         CERROR("%s: undo for %s:%d: rc = %d\n",
1246                                              dt_obd_name(ta->ta_handle->th_dev),
1247                                                ta->ta_args[i]->file,
1248                                                ta->ta_args[i]->line, -ENOTSUPP);
1249                                 }
1250                         }
1251                         break;
1252                 }
1253                 CDEBUG(D_HA, "%s: executed %u/%u: rc = %d\n",
1254                        dt_obd_name(sub_dt), i, ta->ta_argno, rc);
1255         }
1256
1257 stop_trans:
1258         if (rc < 0)
1259                 th->th_result = rc;
1260         rc = top_trans_stop(env, tdtd->tdtd_dt, th);
1261         for (i = 0; i < ta->ta_argno; i++) {
1262                 if (ta->ta_args[i]->object != NULL) {
1263                         lu_object_put(env, &ta->ta_args[i]->object->do_lu);
1264                         ta->ta_args[i]->object = NULL;
1265                 }
1266         }
1267
1268         if (tur != NULL)
1269                 tur->tur_update_records = NULL;
1270 exit_session:
1271         lu_context_exit(&session_env);
1272         lu_context_fini(&session_env);
1273         RETURN(rc);
1274 }
1275 EXPORT_SYMBOL(distribute_txn_replay_handle);