lustre/obdclass/llog.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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/obdclass/llog.c
  32  *
  33  * OST<->MDS recovery logging infrastructure.
  34  * Invariants in implementation:
  35  * - we do not share logs among different OST<->MDS connections, so that
  36  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  40  * Author: Mikhail Pershin <tappro@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_LOG
  44
  45 #include <linux/pid_namespace.h>
  46 #include <linux/kthread.h>
  47 #include <llog_swab.h>
  48 #include <lustre_log.h>
  49 #include <obd_support.h>
  50 #include <obd_class.h>
  51 #include "llog_internal.h"
  52
  53 /*
  54  * Allocate a new log or catalog handle
  55  * Used inside llog_open().
  56  */
  57 static struct llog_handle *llog_alloc_handle(void)
  58 {
  59         struct llog_handle *loghandle;
  60
  61         OBD_ALLOC_PTR(loghandle);
  62         if (loghandle == NULL)
  63                 return NULL;
  64
  65         init_rwsem(&loghandle->lgh_lock);
  66         mutex_init(&loghandle->lgh_hdr_mutex);
  67         init_rwsem(&loghandle->lgh_last_sem);
  68         INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
  69         refcount_set(&loghandle->lgh_refcount, 1);
  70
  71         return loghandle;
  72 }
  73
  74 /*
  75  * Free llog handle and header data if exists. Used in llog_close() only
  76  */
  77 static void llog_free_handle(struct llog_handle *loghandle)
  78 {
  79         LASSERT(loghandle != NULL);
  80
  81         /* failed llog_init_handle */
  82         if (loghandle->lgh_hdr == NULL)
  83                 goto out;
  84
  85         if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
  86                 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
  87         else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
  88                 LASSERT(list_empty(&loghandle->u.chd.chd_head));
  89         OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
  90 out:
  91         OBD_FREE_PTR(loghandle);
  92 }
  93
  94 struct llog_handle *llog_handle_get(struct llog_handle *loghandle)
  95 {
  96         if (refcount_inc_not_zero(&loghandle->lgh_refcount))
  97                 return loghandle;
  98         return NULL;
  99 }
 100
 101 int llog_handle_put(const struct lu_env *env, struct llog_handle *loghandle)
 102 {
 103         int rc = 0;
 104
 105         if (refcount_dec_and_test(&loghandle->lgh_refcount)) {
 106                 const struct llog_operations *lop;
 107
 108                 rc = llog_handle2ops(loghandle, &lop);
 109                 if (!rc) {
 110                         if (lop->lop_close)
 111                                 rc = lop->lop_close(env, loghandle);
 112                         else
 113                                 rc = -EOPNOTSUPP;
 114                 }
 115                 llog_free_handle(loghandle);
 116         }
 117         return rc;
 118 }
 119
 120 static int llog_declare_destroy(const struct lu_env *env,
 121                                 struct llog_handle *handle,
 122                                 struct thandle *th)
 123 {
 124         const struct llog_operations *lop;
 125         int rc;
 126
 127         ENTRY;
 128
 129         rc = llog_handle2ops(handle, &lop);
 130         if (rc)
 131                 RETURN(rc);
 132         if (lop->lop_declare_destroy == NULL)
 133                 RETURN(-EOPNOTSUPP);
 134
 135         rc = lop->lop_declare_destroy(env, handle, th);
 136
 137         RETURN(rc);
 138 }
 139
 140 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
 141                        struct thandle *th)
 142 {
 143         const struct llog_operations *lop;
 144         int rc;
 145         ENTRY;
 146
 147         rc = llog_handle2ops(handle, &lop);
 148         if (rc < 0)
 149                 RETURN(rc);
 150         if (lop->lop_destroy == NULL)
 151                 RETURN(-EOPNOTSUPP);
 152
 153         LASSERT(handle->lgh_obj != NULL);
 154         if (!llog_exist(handle))
 155                 RETURN(0);
 156
 157         rc = lop->lop_destroy(env, handle, th);
 158
 159         RETURN(rc);
 160 }
 161
 162 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
 163 {
 164         const struct llog_operations *lop;
 165         struct dt_device *dt;
 166         struct thandle *th;
 167         int rc;
 168
 169         ENTRY;
 170
 171         rc = llog_handle2ops(handle, &lop);
 172         if (rc < 0)
 173                 RETURN(rc);
 174         if (lop->lop_destroy == NULL)
 175                 RETURN(-EOPNOTSUPP);
 176
 177         if (handle->lgh_obj == NULL) {
 178                 /* if lgh_obj == NULL, then it is from client side destroy */
 179                 rc = lop->lop_destroy(env, handle, NULL);
 180                 RETURN(rc);
 181         }
 182
 183         if (!llog_exist(handle))
 184                 RETURN(0);
 185
 186         dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
 187
 188         if (unlikely(unlikely(dt->dd_rdonly)))
 189                 RETURN(-EROFS);
 190
 191         th = dt_trans_create(env, dt);
 192         if (IS_ERR(th))
 193                 RETURN(PTR_ERR(th));
 194
 195         rc = llog_declare_destroy(env, handle, th);
 196         if (rc != 0)
 197                 GOTO(out_trans, rc);
 198
 199         rc = dt_trans_start_local(env, dt, th);
 200         if (rc < 0)
 201                 GOTO(out_trans, rc);
 202
 203         rc = lop->lop_destroy(env, handle, th);
 204
 205 out_trans:
 206         dt_trans_stop(env, dt, th);
 207
 208         RETURN(rc);
 209 }
 210 EXPORT_SYMBOL(llog_destroy);
 211
 212 /* returns negative on error; 0 if success; 1 if success & log destroyed */
 213 int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
 214                         int num, int *index)
 215 {
 216         struct llog_thread_info *lgi = llog_info(env);
 217         struct dt_device        *dt;
 218         struct llog_log_hdr     *llh;
 219         struct thandle          *th;
 220         __u32                    tmp_lgc_index;
 221         int                      rc, i = 0;
 222         int rc1;
 223         bool subtract_count = false;
 224
 225         ENTRY;
 226
 227         LASSERT(loghandle != NULL);
 228         LASSERT(loghandle->lgh_ctxt != NULL);
 229         LASSERT(loghandle->lgh_obj != NULL);
 230
 231         llh = loghandle->lgh_hdr;
 232
 233         CDEBUG(D_RPCTRACE, "Canceling %d records, first %d in log "DFID"\n",
 234                num, index[0], PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
 235
 236         dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
 237
 238         if (unlikely(unlikely(dt->dd_rdonly)))
 239                 RETURN(0);
 240
 241         th = dt_trans_create(env, dt);
 242         if (IS_ERR(th))
 243                 RETURN(PTR_ERR(th));
 244
 245         rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, 0, th);
 246         if (rc < 0)
 247                 GOTO(out_trans, rc);
 248
 249         if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
 250                 rc = llog_declare_destroy(env, loghandle, th);
 251                 if (rc < 0)
 252                         GOTO(out_trans, rc);
 253         }
 254
 255         th->th_wait_submit = 1;
 256         rc = dt_trans_start_local(env, dt, th);
 257         if (rc < 0)
 258                 GOTO(out_trans, rc);
 259
 260         down_write(&loghandle->lgh_lock);
 261         /* clear bitmap */
 262         mutex_lock(&loghandle->lgh_hdr_mutex);
 263         for (i = 0; i < num; ++i) {
 264                 if (index[i] == 0) {
 265                         CERROR("Can't cancel index 0 which is header\n");
 266                         GOTO(out_unlock, rc = -EINVAL);
 267                 }
 268                 if (!__test_and_clear_bit_le(index[i], LLOG_HDR_BITMAP(llh))) {
 269                         CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n",
 270                                index[i]);
 271                         GOTO(out_unlock, rc = -ENOENT);
 272                 }
 273         }
 274         loghandle->lgh_hdr->llh_count -= num;
 275         subtract_count = true;
 276
 277         /* Since llog_process_thread use lgi_cookie, it`s better to save them
 278          * and restore after using
 279          */
 280         tmp_lgc_index = lgi->lgi_cookie.lgc_index;
 281         /* Pass this index to llog_osd_write_rec(), which will use the index
 282          * to only update the necesary bitmap. */
 283         lgi->lgi_cookie.lgc_index = index[0];
 284         /* update header */
 285         rc = llog_write_rec(env, loghandle, &llh->llh_hdr, (num != 1 ? NULL :
 286                             &lgi->lgi_cookie), LLOG_HEADER_IDX, th);
 287         lgi->lgi_cookie.lgc_index = tmp_lgc_index;
 288
 289         if (rc != 0)
 290                 GOTO(out_unlock, rc);
 291
 292         if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
 293             (llh->llh_count == 1) &&
 294             ((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
 295              (loghandle->u.phd.phd_cat_handle != NULL &&
 296               loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
 297                 loghandle))) {
 298                 /* never try to destroy it again */
 299                 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
 300                 rc = llog_trans_destroy(env, loghandle, th);
 301                 if (rc < 0) {
 302                         /* Sigh, can not destroy the final plain llog, but
 303                          * the bitmap has been clearly, so the record can not
 304                          * be accessed anymore, let's return 0 for now, and
 305                          * the orphan will be handled by LFSCK. */
 306                         CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
 307                                loghandle2name(loghandle),
 308                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
 309                         GOTO(out_unlock, rc = 0);
 310                 }
 311                 rc = LLOG_DEL_PLAIN;
 312         }
 313
 314 out_unlock:
 315         if (rc < 0) {
 316                 /* restore bitmap while holding a mutex */
 317                 if (subtract_count) {
 318                         loghandle->lgh_hdr->llh_count += num;
 319                         subtract_count = false;
 320                 }
 321                 for (i = i - 1; i >= 0; i--)
 322                         set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
 323         }
 324         mutex_unlock(&loghandle->lgh_hdr_mutex);
 325         up_write(&loghandle->lgh_lock);
 326 out_trans:
 327         rc1 = dt_trans_stop(env, dt, th);
 328         if (rc == 0)
 329                 rc = rc1;
 330         if (rc1 < 0) {
 331                 mutex_lock(&loghandle->lgh_hdr_mutex);
 332                 if (subtract_count)
 333                         loghandle->lgh_hdr->llh_count += num;
 334                 for (i = i - 1; i >= 0; i--)
 335                         set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
 336                 mutex_unlock(&loghandle->lgh_hdr_mutex);
 337         }
 338         RETURN(rc);
 339 }
 340
 341 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
 342                     int index)
 343 {
 344         return llog_cancel_arr_rec(env, loghandle, 1, &index);
 345 }
 346
 347 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
 348                      const struct obd_uuid *uuid)
 349 {
 350         const struct llog_operations *lop;
 351         int rc;
 352         ENTRY;
 353
 354         rc = llog_handle2ops(handle, &lop);
 355         if (rc)
 356                 RETURN(rc);
 357
 358         if (lop->lop_read_header == NULL)
 359                 RETURN(-EOPNOTSUPP);
 360
 361         rc = lop->lop_read_header(env, handle);
 362         if (rc == LLOG_EEMPTY) {
 363                 struct llog_log_hdr *llh = handle->lgh_hdr;
 364
 365                 /* lrh_len should be initialized in llog_init_handle */
 366                 handle->lgh_last_idx = 0; /* header is record with index 0 */
 367                 llh->llh_count = 1;         /* for the header record */
 368                 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
 369                 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
 370                                                 LLOG_MIN_CHUNK_SIZE);
 371                 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
 372                 llh->llh_hdr.lrh_index = 0;
 373                 llh->llh_timestamp = ktime_get_real_seconds();
 374                 if (uuid)
 375                         memcpy(&llh->llh_tgtuuid, uuid,
 376                                sizeof(llh->llh_tgtuuid));
 377                 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
 378                 /* Since update llog header might also call this function,
 379                  * let's reset the bitmap to 0 here */
 380                 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
 381                                                 llh->llh_bitmap_offset -
 382                                                 sizeof(llh->llh_tail));
 383                 set_bit_le(0, LLOG_HDR_BITMAP(llh));
 384                 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
 385                 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
 386                 rc = 0;
 387         }
 388         RETURN(rc);
 389 }
 390 EXPORT_SYMBOL(llog_read_header);
 391
 392 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
 393                      int flags, struct obd_uuid *uuid)
 394 {
 395         struct llog_log_hdr     *llh;
 396         enum llog_flag           fmt = flags & LLOG_F_EXT_MASK;
 397         int                      rc;
 398         int                     chunk_size = handle->lgh_ctxt->loc_chunk_size;
 399         ENTRY;
 400
 401         LASSERT(handle->lgh_hdr == NULL);
 402
 403         LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
 404         OBD_ALLOC_LARGE(llh, chunk_size);
 405         if (llh == NULL)
 406                 RETURN(-ENOMEM);
 407
 408         handle->lgh_hdr = llh;
 409         handle->lgh_hdr_size = chunk_size;
 410         /* first assign flags to use llog_client_ops */
 411         llh->llh_flags = flags;
 412         rc = llog_read_header(env, handle, uuid);
 413         if (rc == 0) {
 414                 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
 415                               flags & LLOG_F_IS_CAT) ||
 416                              (llh->llh_flags & LLOG_F_IS_CAT &&
 417                               flags & LLOG_F_IS_PLAIN))) {
 418                         CERROR("%s: llog type is %s but initializing %s\n",
 419                                loghandle2name(handle),
 420                                llh->llh_flags & LLOG_F_IS_CAT ?
 421                                "catalog" : "plain",
 422                                flags & LLOG_F_IS_CAT ? "catalog" : "plain");
 423                         GOTO(out, rc = -EINVAL);
 424                 } else if (llh->llh_flags &
 425                            (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
 426                         /*
 427                          * it is possible to open llog without specifying llog
 428                          * type so it is taken from llh_flags
 429                          */
 430                         flags = llh->llh_flags;
 431                 } else {
 432                         /* for some reason the llh_flags has no type set */
 433                         CERROR("llog type is not specified!\n");
 434                         GOTO(out, rc = -EINVAL);
 435                 }
 436                 if (unlikely(uuid &&
 437                              !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
 438                         CERROR("%s: llog uuid mismatch: %s/%s\n",
 439                                loghandle2name(handle),
 440                                (char *)uuid->uuid,
 441                                (char *)llh->llh_tgtuuid.uuid);
 442                         GOTO(out, rc = -EEXIST);
 443                 }
 444         }
 445         if (flags & LLOG_F_IS_CAT) {
 446                 LASSERT(list_empty(&handle->u.chd.chd_head));
 447                 INIT_LIST_HEAD(&handle->u.chd.chd_head);
 448                 llh->llh_size = sizeof(struct llog_logid_rec);
 449                 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
 450         } else if (!(flags & LLOG_F_IS_PLAIN)) {
 451                 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
 452                        loghandle2name(handle), flags, LLOG_F_IS_CAT,
 453                        LLOG_F_IS_PLAIN);
 454                 rc = -EINVAL;
 455         }
 456         llh->llh_flags |= fmt;
 457 out:
 458         if (rc) {
 459                 OBD_FREE_LARGE(llh, chunk_size);
 460                 handle->lgh_hdr = NULL;
 461         }
 462         RETURN(rc);
 463 }
 464 EXPORT_SYMBOL(llog_init_handle);
 465
 466 int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
 467 {
 468         int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
 469
 470         if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
 471                 CERROR("%s: record is too large: %d > %d\n",
 472                        loghandle2name(llh), rec->lrh_len, chunk_size);
 473                 return -EINVAL;
 474         }
 475         if (rec->lrh_index >= LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr)) {
 476                 CERROR("%s: index is too high: %d\n",
 477                        loghandle2name(llh), rec->lrh_index);
 478                 return -EINVAL;
 479         }
 480         if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC) {
 481                 CERROR("%s: magic %x is bad\n",
 482                        loghandle2name(llh), rec->lrh_type);
 483                 return -EINVAL;
 484         }
 485
 486         return 0;
 487 }
 488 EXPORT_SYMBOL(llog_verify_record);
 489
 490 static int llog_process_thread(void *arg)
 491 {
 492         struct llog_process_info        *lpi = arg;
 493         struct llog_handle              *loghandle = lpi->lpi_loghandle;
 494         struct llog_log_hdr             *llh = loghandle->lgh_hdr;
 495         struct llog_process_cat_data    *cd  = lpi->lpi_catdata;
 496         struct llog_thread_info         *lti;
 497         char                            *buf;
 498         size_t                           chunk_size;
 499         __u64                            cur_offset;
 500         int                              rc = 0, index = 1, last_index;
 501         int                              saved_index = 0;
 502         int                              last_called_index = 0;
 503         bool                             repeated = false;
 504         bool                            refresh_idx = false;
 505
 506         ENTRY;
 507
 508         if (llh == NULL)
 509                 RETURN(-EINVAL);
 510
 511         lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
 512
 513         cur_offset = chunk_size = llh->llh_hdr.lrh_len;
 514         /* expect chunk_size to be power of two */
 515         LASSERT(is_power_of_2(chunk_size));
 516
 517         OBD_ALLOC_LARGE(buf, chunk_size);
 518         if (buf == NULL) {
 519                 lpi->lpi_rc = -ENOMEM;
 520                 RETURN(0);
 521         }
 522
 523         if (cd != NULL) {
 524                 last_called_index = cd->lpcd_first_idx;
 525                 index = cd->lpcd_first_idx + 1;
 526         }
 527         if (cd != NULL && cd->lpcd_last_idx)
 528                 last_index = cd->lpcd_last_idx;
 529         else
 530                 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 531
 532         while (rc == 0) {
 533                 struct llog_rec_hdr *rec;
 534                 off_t chunk_offset = 0;
 535                 unsigned int buf_offset = 0;
 536                 int     lh_last_idx;
 537                 int     synced_idx = 0;
 538
 539                 /* skip records not set in bitmap */
 540                 while (index <= last_index &&
 541                        !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
 542                         ++index;
 543
 544                 /* There are no indices prior the last_index */
 545                 if (index > last_index)
 546                         break;
 547
 548                 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
 549                        last_index);
 550
 551 repeat:
 552                 /* get the buf with our target record; avoid old garbage */
 553                 memset(buf, 0, chunk_size);
 554                 /* the record index for outdated chunk data */
 555                 /* it is safe to process buffer until saved lgh_last_idx */
 556                 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 557                 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
 558                                      index, &cur_offset, buf, chunk_size);
 559                 if (repeated && rc)
 560                         CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
 561                                " buf_offset %u, rc = %d\n", cur_offset,
 562                                (__u64)chunk_offset, buf_offset, rc);
 563                 if (rc == -ESTALE)
 564                         GOTO(out, rc = 0);
 565                 /* we`ve tried to reread the chunk, but there is no
 566                  * new records */
 567                 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
 568                     cur_offset)
 569                         GOTO(out, rc = 0);
 570                 if (rc != 0)
 571                         GOTO(out, rc);
 572
 573                 /* NB: after llog_next_block() call the cur_offset is the
 574                  * offset of the next block after read one.
 575                  * The absolute offset of the current chunk is calculated
 576                  * from cur_offset value and stored in chunk_offset variable.
 577                  */
 578                 if ((cur_offset & (chunk_size - 1)) != 0)
 579                         chunk_offset = cur_offset & ~(chunk_size - 1);
 580                 else
 581                         chunk_offset = cur_offset - chunk_size;
 582
 583                 /* NB: when rec->lrh_len is accessed it is already swabbed
 584                  * since it is used at the "end" of the loop and the rec
 585                  * swabbing is done at the beginning of the loop. */
 586                 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
 587                      (char *)rec < buf + chunk_size;
 588                      rec = llog_rec_hdr_next(rec)) {
 589
 590                         CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
 591                                rec, rec->lrh_type);
 592
 593                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 594                                 lustre_swab_llog_rec(rec);
 595
 596                         CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
 597                                rec->lrh_type, rec->lrh_index);
 598
 599                         if (index == (synced_idx + 1) &&
 600                             synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
 601                                 GOTO(out, rc = 0);
 602
 603                         if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
 604                                 cfs_fail_val == (unsigned int)
 605                                         (loghandle->lgh_id.lgl_oi.oi.oi_id &
 606                                          0xFFFFFFFF)) {
 607                                 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
 608                         }
 609
 610                         /* the bitmap could be changed during processing
 611                          * records from the chunk. For wrapped catalog
 612                          * it means we can read deleted record and try to
 613                          * process it. Check this case and reread the chunk.
 614                          * It is safe to process to lh_last_idx, including
 615                          * lh_last_idx if it was synced. We can not do <=
 616                          * comparison, cause for wrapped catalog lgh_last_idx
 617                          * could be less than index. So we detect last index
 618                          * for processing as index == lh_last_idx+1. But when
 619                          * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
 620                          * the first processing index is llh_cat_idx+1.The
 621                          * exception is !(lgh_last_idx == llh_cat_idx &&
 622                          * index == llh_cat_idx + 1), and after simplification
 623                          * it turns to
 624                          * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
 625                          * This exception is working for catalog only.
 626                          */
 627
 628                         if ((index == lh_last_idx && synced_idx != index) ||
 629                             (index == (lh_last_idx + 1) &&
 630                              lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
 631                             (rec->lrh_index == 0 && !repeated)) {
 632
 633                                 /* save offset inside buffer for the re-read */
 634                                 buf_offset = (char *)rec - (char *)buf;
 635                                 cur_offset = chunk_offset;
 636                                 repeated = true;
 637                                 /* We need to be sure lgh_last_idx
 638                                  * record was saved to disk
 639                                  */
 640                                 down_read(&loghandle->lgh_last_sem);
 641                                 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 642                                 up_read(&loghandle->lgh_last_sem);
 643                                 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
 644                                 goto repeat;
 645
 646                         }
 647
 648                         repeated = false;
 649
 650                         rc = llog_verify_record(loghandle, rec);
 651                         if (rc) {
 652                                 CERROR("%s: invalid record in llog "DFID
 653                                        " record for index %d/%d: rc = %d\n",
 654                                        loghandle2name(loghandle),
 655                                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 656                                        rec->lrh_index, index, rc);
 657                                 /*
 658                                  * the block seem to be corrupted, let's try
 659                                  * with the next one. reset rc to go to the
 660                                  * next chunk.
 661                                  */
 662                                 refresh_idx = true;
 663                                 index = 0;
 664                                 GOTO(repeat, rc = 0);
 665                         }
 666
 667                         if (rec->lrh_index < index) {
 668                                 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
 669                                        rec->lrh_index);
 670                                 continue;
 671                         }
 672
 673                         if (rec->lrh_index > index) {
 674                                 /* the record itself looks good, but we met a
 675                                  * gap which can be result of old bugs, just
 676                                  * keep going */
 677                                 CERROR("%s: "DFID" index %u, expected %u\n",
 678                                        loghandle2name(loghandle),
 679                                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 680                                        rec->lrh_index, index);
 681                                 index = rec->lrh_index;
 682                         }
 683
 684                         CDEBUG(D_OTHER,
 685                                "lrh_index: %d lrh_len: %d (%d remains)\n",
 686                                rec->lrh_index, rec->lrh_len,
 687                                (int)(buf + chunk_size - (char *)rec));
 688
 689                         /* lgh_cur_offset is used only at llog_test_3 */
 690                         loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
 691                                                     chunk_offset;
 692
 693                         /* if set, process the callback on this record */
 694                         if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
 695                                 struct llog_cookie *lgc;
 696                                 __u64   tmp_off;
 697                                 int     tmp_idx;
 698
 699                                 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
 700                                         D_HA : D_OTHER),
 701                                        "index: %d, lh_last_idx: %d "
 702                                        "synced_idx: %d lgh_last_idx: %d\n",
 703                                        index, lh_last_idx, synced_idx,
 704                                        loghandle->lgh_last_idx);
 705
 706                                 if (lti != NULL) {
 707                                         lgc = &lti->lgi_cookie;
 708                                         /* store lu_env for recursive calls */
 709                                         tmp_off = lgc->lgc_offset;
 710                                         tmp_idx = lgc->lgc_index;
 711
 712                                         lgc->lgc_offset = (char *)rec -
 713                                                 (char *)buf + chunk_offset;
 714                                         lgc->lgc_index = rec->lrh_index;
 715                                 }
 716                                 /* using lu_env for passing record offset to
 717                                  * llog_write through various callbacks */
 718                                 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
 719                                                  lpi->lpi_cbdata);
 720                                 last_called_index = index;
 721
 722                                 if (lti != NULL) {
 723                                         lgc->lgc_offset = tmp_off;
 724                                         lgc->lgc_index = tmp_idx;
 725                                 }
 726
 727                                 if (rc == LLOG_PROC_BREAK) {
 728                                         GOTO(out, rc);
 729                                 } else if (rc == LLOG_DEL_RECORD) {
 730                                         rc = llog_cancel_rec(lpi->lpi_env,
 731                                                              loghandle,
 732                                                              rec->lrh_index);
 733                                         /* Allow parallel cancelling, ENOENT
 734                                          * means record was canceled at another
 735                                          * processing thread or callback
 736                                          */
 737                                         if (rc == -ENOENT)
 738                                                 rc = 0;
 739                                 }
 740                                 if (rc)
 741                                         GOTO(out, rc);
 742                                 /* some stupid callbacks directly cancel records
 743                                  * and delete llog. Check it and stop
 744                                  * processing. */
 745                                 if (loghandle->lgh_hdr == NULL ||
 746                                     loghandle->lgh_hdr->llh_count == 1)
 747                                         GOTO(out, rc = 0);
 748                         }
 749                         /* exit if the last index is reached */
 750                         if (index >= last_index)
 751                                 GOTO(out, rc = 0);
 752                         ++index;
 753                 }
 754         }
 755
 756 out:
 757         CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
 758                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
 759                POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
 760                index, llh->llh_count);
 761
 762         if (cd != NULL)
 763                 cd->lpcd_last_idx = last_called_index;
 764
 765         if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
 766                 if (dt_object_remote(loghandle->lgh_obj)) {
 767                         /* If it is remote object, then -EIO might means
 768                          * disconnection or eviction, let's return -EAGAIN,
 769                          * so for update recovery log processing, it will
 770                          * retry until the umount or abort recovery, see
 771                          * lod_sub_recovery_thread() */
 772                         CERROR("%s retry remote llog process\n",
 773                                loghandle2name(loghandle));
 774                         rc = -EAGAIN;
 775                 } else {
 776                         /* something bad happened to the processing of a local
 777                          * llog file, probably I/O error or the log got
 778                          * corrupted to be able to finally release the log we
 779                          * discard any remaining bits in the header */
 780                         CERROR("%s: Local llog found corrupted #"DOSTID":%x"
 781                                " %s index %d count %d\n",
 782                                loghandle2name(loghandle),
 783                                POSTID(&loghandle->lgh_id.lgl_oi),
 784                                loghandle->lgh_id.lgl_ogen,
 785                                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
 786                                 "plain"), index, llh->llh_count);
 787
 788                         while (index <= last_index) {
 789                                 if (test_bit_le(index,
 790                                                   LLOG_HDR_BITMAP(llh)) != 0)
 791                                         llog_cancel_rec(lpi->lpi_env, loghandle,
 792                                                         index);
 793                                 index++;
 794                         }
 795                         rc = 0;
 796                 }
 797         }
 798
 799         OBD_FREE_LARGE(buf, chunk_size);
 800         lpi->lpi_rc = rc;
 801         return 0;
 802 }
 803
 804 static int llog_process_thread_daemonize(void *arg)
 805 {
 806         struct llog_process_info        *lpi = arg;
 807         struct lu_env                    env;
 808         int                              rc;
 809         struct nsproxy                  *new_ns, *curr_ns = current->nsproxy;
 810
 811         task_lock(lpi->lpi_reftask);
 812         new_ns = lpi->lpi_reftask->nsproxy;
 813         if (curr_ns != new_ns) {
 814                 get_nsproxy(new_ns);
 815
 816                 current->nsproxy = new_ns;
 817                 /* XXX: we should call put_nsproxy() instead of
 818                  * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
 819                  * used outside of the kernel itself, because it calls
 820                  * free_nsproxy() which is not exported by the kernel
 821                  * (defined in kernel/nsproxy.c) */
 822                 if (curr_ns)
 823                         atomic_dec(&curr_ns->count);
 824         }
 825         task_unlock(lpi->lpi_reftask);
 826
 827         /* client env has no keys, tags is just 0 */
 828         rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
 829         if (rc)
 830                 goto out;
 831         lpi->lpi_env = &env;
 832
 833         rc = llog_process_thread(arg);
 834
 835         lu_env_fini(&env);
 836 out:
 837         complete(&lpi->lpi_completion);
 838         return rc;
 839 }
 840
 841 int llog_process_or_fork(const struct lu_env *env,
 842                          struct llog_handle *loghandle,
 843                          llog_cb_t cb, void *data, void *catdata, bool fork)
 844 {
 845         struct llog_process_info *lpi;
 846         struct llog_process_data *d = data;
 847         struct llog_process_cat_data *cd = catdata;
 848         __u32 flags = loghandle->lgh_hdr->llh_flags;
 849         int rc;
 850
 851         ENTRY;
 852
 853         OBD_ALLOC_PTR(lpi);
 854         if (lpi == NULL) {
 855                 CERROR("cannot alloc pointer\n");
 856                 RETURN(-ENOMEM);
 857         }
 858         lpi->lpi_loghandle = loghandle;
 859         lpi->lpi_cb        = cb;
 860         lpi->lpi_cbdata    = data;
 861         lpi->lpi_catdata   = catdata;
 862
 863         CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d\n",
 864                PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
 865                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
 866                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
 867                cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1);
 868         if (fork) {
 869                 struct task_struct *task;
 870
 871                 /* The new thread can't use parent env,
 872                  * init the new one in llog_process_thread_daemonize. */
 873                 lpi->lpi_env = NULL;
 874                 init_completion(&lpi->lpi_completion);
 875                 /* take reference to current, so that
 876                  * llog_process_thread_daemonize() can use it to switch to
 877                  * namespace associated with current  */
 878                 lpi->lpi_reftask = current;
 879                 task = kthread_run(llog_process_thread_daemonize, lpi,
 880                                    "llog_process_thread");
 881                 if (IS_ERR(task)) {
 882                         rc = PTR_ERR(task);
 883                         CERROR("%s: cannot start thread: rc = %d\n",
 884                                loghandle2name(loghandle), rc);
 885                         GOTO(out_lpi, rc);
 886                 }
 887                 wait_for_completion(&lpi->lpi_completion);
 888         } else {
 889                 lpi->lpi_env = env;
 890                 llog_process_thread(lpi);
 891         }
 892         rc = lpi->lpi_rc;
 893
 894 out_lpi:
 895         OBD_FREE_PTR(lpi);
 896         RETURN(rc);
 897 }
 898 EXPORT_SYMBOL(llog_process_or_fork);
 899
 900 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
 901                  llog_cb_t cb, void *data, void *catdata)
 902 {
 903         int rc;
 904         rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
 905         return rc == LLOG_DEL_PLAIN ? 0 : rc;
 906 }
 907 EXPORT_SYMBOL(llog_process);
 908
 909 static inline const struct cred *llog_raise_resource(void)
 910 {
 911         struct cred *cred = NULL;
 912
 913         if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
 914                 return cred;
 915
 916         cred = prepare_creds();
 917         if (!cred)
 918                 return cred;
 919
 920         cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
 921         return override_creds(cred);
 922 }
 923
 924 static inline void llog_restore_resource(const struct cred *old_cred)
 925 {
 926         if (old_cred)
 927                 revert_creds(old_cred);
 928 }
 929
 930 int llog_reverse_process(const struct lu_env *env,
 931                          struct llog_handle *loghandle, llog_cb_t cb,
 932                          void *data, void *catdata)
 933 {
 934         struct llog_log_hdr *llh = loghandle->lgh_hdr;
 935         struct llog_process_cat_data *cd = catdata;
 936         void *buf;
 937         int rc = 0, first_index = 1, index, idx;
 938         __u32   chunk_size = llh->llh_hdr.lrh_len;
 939         ENTRY;
 940
 941         OBD_ALLOC_LARGE(buf, chunk_size);
 942         if (buf == NULL)
 943                 RETURN(-ENOMEM);
 944
 945         if (cd != NULL)
 946                 first_index = cd->lpcd_first_idx + 1;
 947         if (cd != NULL && cd->lpcd_last_idx)
 948                 index = cd->lpcd_last_idx;
 949         else
 950                 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 951
 952         while (rc == 0) {
 953                 struct llog_rec_hdr *rec;
 954                 struct llog_rec_tail *tail;
 955
 956                 /* skip records not set in bitmap */
 957                 while (index >= first_index &&
 958                        !test_bit_le(index, LLOG_HDR_BITMAP(llh)))
 959                         --index;
 960
 961                 LASSERT(index >= first_index - 1);
 962                 if (index == first_index - 1)
 963                         break;
 964
 965                 /* get the buf with our target record; avoid old garbage */
 966                 memset(buf, 0, chunk_size);
 967                 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
 968                 if (rc)
 969                         GOTO(out, rc);
 970
 971                 rec = buf;
 972                 idx = rec->lrh_index;
 973                 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
 974                 while (idx < index) {
 975                         rec = (void *)rec + rec->lrh_len;
 976                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 977                                 lustre_swab_llog_rec(rec);
 978                         idx ++;
 979                 }
 980                 LASSERT(idx == index);
 981                 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
 982
 983                 /* process records in buffer, starting where we found one */
 984                 while ((void *)tail > buf) {
 985                         if (tail->lrt_index == 0)
 986                                 GOTO(out, rc = 0); /* no more records */
 987
 988                         /* if set, process the callback on this record */
 989                         if (test_bit_le(index, LLOG_HDR_BITMAP(llh))) {
 990                                 rec = (void *)tail - tail->lrt_len +
 991                                       sizeof(*tail);
 992
 993                                 rc = cb(env, loghandle, rec, data);
 994                                 if (rc == LLOG_PROC_BREAK) {
 995                                         GOTO(out, rc);
 996                                 } else if (rc == LLOG_DEL_RECORD) {
 997                                         rc = llog_cancel_rec(env, loghandle,
 998                                                              tail->lrt_index);
 999                                 }
1000                                 if (rc)
1001                                         GOTO(out, rc);
1002                         }
1003
1004                         /* previous record, still in buffer? */
1005                         --index;
1006                         if (index < first_index)
1007                                 GOTO(out, rc = 0);
1008                         tail = (void *)tail - tail->lrt_len;
1009                 }
1010         }
1011
1012 out:
1013         if (buf != NULL)
1014                 OBD_FREE_LARGE(buf, chunk_size);
1015         RETURN(rc);
1016 }
1017 EXPORT_SYMBOL(llog_reverse_process);
1018
1019 /**
1020  * new llog API
1021  *
1022  * API functions:
1023  *      llog_open - open llog, may not exist
1024  *      llog_exist - check if llog exists
1025  *      llog_close - close opened llog, pair for open, frees llog_handle
1026  *      llog_declare_create - declare llog creation
1027  *      llog_create - create new llog on disk, need transaction handle
1028  *      llog_declare_write_rec - declaration of llog write
1029  *      llog_write_rec - write llog record on disk, need transaction handle
1030  *      llog_declare_add - declare llog catalog record addition
1031  *      llog_add - add llog record in catalog, need transaction handle
1032  */
1033 int llog_exist(struct llog_handle *loghandle)
1034 {
1035         const struct llog_operations *lop;
1036         int rc;
1037
1038         ENTRY;
1039
1040         rc = llog_handle2ops(loghandle, &lop);
1041         if (rc)
1042                 RETURN(rc);
1043         if (lop->lop_exist == NULL)
1044                 RETURN(-EOPNOTSUPP);
1045
1046         rc = lop->lop_exist(loghandle);
1047         RETURN(rc);
1048 }
1049 EXPORT_SYMBOL(llog_exist);
1050
1051 int llog_declare_create(const struct lu_env *env,
1052                         struct llog_handle *loghandle, struct thandle *th)
1053 {
1054         const struct cred *old_cred;
1055         const struct llog_operations *lop;
1056         int rc;
1057
1058         ENTRY;
1059
1060         rc = llog_handle2ops(loghandle, &lop);
1061         if (rc)
1062                 RETURN(rc);
1063         if (lop->lop_declare_create == NULL)
1064                 RETURN(-EOPNOTSUPP);
1065
1066         old_cred = llog_raise_resource();
1067         rc = lop->lop_declare_create(env, loghandle, th);
1068         llog_restore_resource(old_cred);
1069         RETURN(rc);
1070 }
1071
1072 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1073                 struct thandle *th)
1074 {
1075         const struct cred *old_cred;
1076         const struct llog_operations *lop;
1077         int rc;
1078
1079         ENTRY;
1080
1081         rc = llog_handle2ops(handle, &lop);
1082         if (rc)
1083                 RETURN(rc);
1084         if (lop->lop_create == NULL)
1085                 RETURN(-EOPNOTSUPP);
1086
1087         old_cred = llog_raise_resource();
1088         rc = lop->lop_create(env, handle, th);
1089         llog_restore_resource(old_cred);
1090         RETURN(rc);
1091 }
1092
1093 int llog_declare_write_rec(const struct lu_env *env,
1094                            struct llog_handle *handle,
1095                            struct llog_rec_hdr *rec, int idx,
1096                            struct thandle *th)
1097 {
1098         const struct cred *old_cred;
1099         const struct llog_operations *lop;
1100         int rc;
1101
1102         ENTRY;
1103
1104         rc = llog_handle2ops(handle, &lop);
1105         if (rc)
1106                 RETURN(rc);
1107         LASSERT(lop);
1108         if (lop->lop_declare_write_rec == NULL)
1109                 RETURN(-EOPNOTSUPP);
1110
1111         old_cred = llog_raise_resource();
1112         rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1113         llog_restore_resource(old_cred);
1114         RETURN(rc);
1115 }
1116
1117 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1118                    struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1119                    int idx, struct thandle *th)
1120 {
1121         const struct cred *old_cred;
1122         const struct llog_operations *lop;
1123         int rc, buflen;
1124
1125         ENTRY;
1126
1127         /* API sanity checks */
1128         if (handle == NULL) {
1129                 CERROR("loghandle is missed\n");
1130                 RETURN(-EPROTO);
1131         } else if (handle->lgh_obj == NULL) {
1132                 CERROR("loghandle %p with NULL object\n",
1133                         handle);
1134                 RETURN(-EPROTO);
1135         } else if (th == NULL) {
1136                 CERROR("%s: missed transaction handle\n",
1137                        loghandle2name(handle));
1138                 RETURN(-EPROTO);
1139         } else if (handle->lgh_hdr == NULL) {
1140                 CERROR("%s: loghandle %p with no header\n",
1141                        loghandle2name(handle), handle);
1142                 RETURN(-EPROTO);
1143         }
1144
1145         rc = llog_handle2ops(handle, &lop);
1146         if (rc)
1147                 RETURN(rc);
1148
1149         if (lop->lop_write_rec == NULL)
1150                 RETURN(-EOPNOTSUPP);
1151
1152         buflen = rec->lrh_len;
1153         LASSERT(cfs_size_round(buflen) == buflen);
1154
1155         old_cred = llog_raise_resource();
1156         rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1157         llog_restore_resource(old_cred);
1158         RETURN(rc);
1159 }
1160
1161 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1162              struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1163              struct thandle *th)
1164 {
1165         const struct cred *old_cred;
1166         int rc;
1167
1168         ENTRY;
1169
1170         if (lgh->lgh_logops->lop_add == NULL)
1171                 RETURN(-EOPNOTSUPP);
1172
1173         old_cred = llog_raise_resource();
1174         rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1175         llog_restore_resource(old_cred);
1176         RETURN(rc);
1177 }
1178 EXPORT_SYMBOL(llog_add);
1179
1180 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1181                      struct llog_rec_hdr *rec, struct thandle *th)
1182 {
1183         const struct cred *old_cred;
1184         int rc;
1185
1186         ENTRY;
1187
1188         if (lgh->lgh_logops->lop_declare_add == NULL)
1189                 RETURN(-EOPNOTSUPP);
1190
1191         old_cred = llog_raise_resource();
1192         rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1193         llog_restore_resource(old_cred);
1194         RETURN(rc);
1195 }
1196 EXPORT_SYMBOL(llog_declare_add);
1197
1198 /**
1199  * Helper function to open llog or create it if doesn't exist.
1200  * It hides all transaction handling from caller.
1201  */
1202 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1203                      struct llog_handle **res, struct llog_logid *logid,
1204                      char *name)
1205 {
1206         struct dt_device        *d;
1207         struct thandle          *th;
1208         int                      rc;
1209
1210         ENTRY;
1211
1212         rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1213         if (rc)
1214                 RETURN(rc);
1215
1216         if (llog_exist(*res))
1217                 RETURN(0);
1218
1219         LASSERT((*res)->lgh_obj != NULL);
1220
1221         d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1222
1223         if (unlikely(unlikely(d->dd_rdonly)))
1224                 RETURN(-EROFS);
1225
1226         th = dt_trans_create(env, d);
1227         if (IS_ERR(th))
1228                 GOTO(out, rc = PTR_ERR(th));
1229
1230         /* Create update llog object synchronously, which
1231          * happens during inialization process see
1232          * lod_sub_prep_llog(), to make sure the update
1233          * llog object is created before corss-MDT writing
1234          * updates into the llog object */
1235         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1236                 th->th_sync = 1;
1237
1238         th->th_wait_submit = 1;
1239         rc = llog_declare_create(env, *res, th);
1240         if (rc == 0) {
1241                 rc = dt_trans_start_local(env, d, th);
1242                 if (rc == 0)
1243                         rc = llog_create(env, *res, th);
1244         }
1245         dt_trans_stop(env, d, th);
1246 out:
1247         if (rc)
1248                 llog_close(env, *res);
1249         RETURN(rc);
1250 }
1251 EXPORT_SYMBOL(llog_open_create);
1252
1253 /**
1254  * Helper function to delete existent llog.
1255  */
1256 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1257                struct llog_logid *logid, char *name)
1258 {
1259         struct llog_handle      *handle;
1260         int                      rc = 0, rc2;
1261
1262         ENTRY;
1263
1264         /* nothing to erase */
1265         if (name == NULL && logid == NULL)
1266                 RETURN(0);
1267
1268         rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1269         if (rc < 0)
1270                 RETURN(rc);
1271
1272         rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1273         if (rc == 0)
1274                 rc = llog_destroy(env, handle);
1275
1276         rc2 = llog_close(env, handle);
1277         if (rc == 0)
1278                 rc = rc2;
1279         RETURN(rc);
1280 }
1281 EXPORT_SYMBOL(llog_erase);
1282
1283 /*
1284  * Helper function for write record in llog.
1285  * It hides all transaction handling from caller.
1286  * Valid only with local llog.
1287  */
1288 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1289                struct llog_rec_hdr *rec, int idx)
1290 {
1291         struct dt_device        *dt;
1292         struct thandle          *th;
1293         bool                    need_cookie;
1294         int                     rc;
1295
1296         ENTRY;
1297
1298         LASSERT(loghandle);
1299         LASSERT(loghandle->lgh_ctxt);
1300         LASSERT(loghandle->lgh_obj != NULL);
1301
1302         dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1303
1304         if (unlikely(unlikely(dt->dd_rdonly)))
1305                 RETURN(-EROFS);
1306
1307         th = dt_trans_create(env, dt);
1308         if (IS_ERR(th))
1309                 RETURN(PTR_ERR(th));
1310
1311         rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1312         if (rc)
1313                 GOTO(out_trans, rc);
1314
1315         th->th_wait_submit = 1;
1316         rc = dt_trans_start_local(env, dt, th);
1317         if (rc)
1318                 GOTO(out_trans, rc);
1319
1320         need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1321
1322         down_write(&loghandle->lgh_lock);
1323         if (need_cookie) {
1324                 struct llog_thread_info *lti = llog_info(env);
1325
1326                 /* cookie comes from llog_process_thread */
1327                 rc = llog_write_rec(env, loghandle, rec, &lti->lgi_cookie,
1328                                     rec->lrh_index, th);
1329                 /* upper layer didn`t pass cookie so change rc */
1330                 rc = (rc == 1 ? 0 : rc);
1331         } else {
1332                 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1333         }
1334
1335         up_write(&loghandle->lgh_lock);
1336 out_trans:
1337         dt_trans_stop(env, dt, th);
1338         RETURN(rc);
1339 }
1340 EXPORT_SYMBOL(llog_write);
1341
1342 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1343               struct llog_handle **lgh, struct llog_logid *logid,
1344               char *name, enum llog_open_param open_param)
1345 {
1346         const struct cred *old_cred;
1347         int      rc;
1348
1349         ENTRY;
1350
1351         LASSERT(ctxt);
1352         LASSERT(ctxt->loc_logops);
1353
1354         if (ctxt->loc_logops->lop_open == NULL) {
1355                 *lgh = NULL;
1356                 RETURN(-EOPNOTSUPP);
1357         }
1358
1359         *lgh = llog_alloc_handle();
1360         if (*lgh == NULL)
1361                 RETURN(-ENOMEM);
1362         (*lgh)->lgh_ctxt = ctxt;
1363         (*lgh)->lgh_logops = ctxt->loc_logops;
1364
1365         old_cred = llog_raise_resource();
1366         rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1367         llog_restore_resource(old_cred);
1368         if (rc) {
1369                 llog_free_handle(*lgh);
1370                 *lgh = NULL;
1371         }
1372         RETURN(rc);
1373 }
1374 EXPORT_SYMBOL(llog_open);
1375
1376 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1377 {
1378         return llog_handle_put(env, loghandle);
1379 }
1380 EXPORT_SYMBOL(llog_close);
1381
1382 /**
1383  * Helper function to get the llog size in records. It is used by MGS
1384  * mostly to check that config llog exists and contains data.
1385  *
1386  * \param[in] env       execution environment
1387  * \param[in] ctxt      llog context
1388  * \param[in] name      llog name
1389  *
1390  * \retval              true if there are records in llog besides a header
1391  * \retval              false on error or llog without records
1392  */
1393 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1394                   char *name)
1395 {
1396         struct llog_handle      *llh;
1397         int                      rc = 0;
1398
1399         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1400         if (rc < 0) {
1401                 if (likely(rc == -ENOENT))
1402                         rc = 0;
1403                 GOTO(out, rc);
1404         }
1405
1406         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1407         if (rc)
1408                 GOTO(out_close, rc);
1409         rc = llog_get_size(llh);
1410
1411 out_close:
1412         llog_close(env, llh);
1413 out:
1414         /* The header is record 1, the llog is still considered as empty
1415          * if there is only header */
1416         return (rc <= 1);
1417 }
1418 EXPORT_SYMBOL(llog_is_empty);
1419
1420 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1421                       struct llog_rec_hdr *rec, void *data)
1422 {
1423         struct llog_handle      *copy_llh = data;
1424
1425         /* Append all records */
1426         return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1427 }
1428
1429 /* backup plain llog */
1430 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1431                 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1432                 char *name, char *backup)
1433 {
1434         struct llog_handle      *llh, *bllh;
1435         int                      rc;
1436
1437         ENTRY;
1438
1439         /* open original log */
1440         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1441         if (rc < 0) {
1442                 /* the -ENOENT case is also reported to the caller
1443                  * but silently so it should handle that if needed.
1444                  */
1445                 if (rc != -ENOENT)
1446                         CERROR("%s: failed to open log %s: rc = %d\n",
1447                                obd->obd_name, name, rc);
1448                 RETURN(rc);
1449         }
1450
1451         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1452         if (rc)
1453                 GOTO(out_close, rc);
1454
1455         /* Make sure there's no old backup log */
1456         rc = llog_erase(env, bctxt, NULL, backup);
1457         if (rc < 0 && rc != -ENOENT)
1458                 GOTO(out_close, rc);
1459
1460         /* open backup log */
1461         rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1462         if (rc) {
1463                 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1464                        obd->obd_name, backup, rc);
1465                 GOTO(out_close, rc);
1466         }
1467
1468         /* check that backup llog is not the same object as original one */
1469         if (llh->lgh_obj == bllh->lgh_obj) {
1470                 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1471                        obd->obd_name, name, backup, llh->lgh_obj,
1472                        bllh->lgh_obj);
1473                 GOTO(out_backup, rc = -EEXIST);
1474         }
1475
1476         rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1477         if (rc)
1478                 GOTO(out_backup, rc);
1479
1480         /* Copy log record by record */
1481         rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1482                                   NULL, false);
1483         if (rc)
1484                 CERROR("%s: failed to backup log %s: rc = %d\n",
1485                        obd->obd_name, name, rc);
1486 out_backup:
1487         llog_close(env, bllh);
1488 out_close:
1489         llog_close(env, llh);
1490         RETURN(rc);
1491 }
1492 EXPORT_SYMBOL(llog_backup);
1493
1494 /* Get size of llog */
1495 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1496 {
1497         int rc;
1498         struct lu_attr la;
1499
1500         rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1501         if (rc) {
1502                 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1503                        loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
1504                        rc);
1505                 return 0;
1506         }
1507
1508         return la.la_size;
1509 }
1510 EXPORT_SYMBOL(llog_size);
1511