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 inline bool llog_is_index_skipable(int idx, struct llog_log_hdr *llh,
 491                                           struct llog_process_cat_data *cd)
 492 {
 493         if (cd && (cd->lpcd_read_mode & LLOG_READ_MODE_RAW))
 494                 return false;
 495
 496         return !test_bit_le(idx, LLOG_HDR_BITMAP(llh));
 497 }
 498
 499 static int llog_process_thread(void *arg)
 500 {
 501         struct llog_process_info        *lpi = arg;
 502         struct llog_handle              *loghandle = lpi->lpi_loghandle;
 503         struct llog_log_hdr             *llh = loghandle->lgh_hdr;
 504         struct llog_process_cat_data    *cd  = lpi->lpi_catdata;
 505         struct llog_thread_info         *lti;
 506         char                            *buf;
 507         size_t                           chunk_size;
 508         __u64                            cur_offset;
 509         int                              rc = 0, index = 1, last_index;
 510         int                              saved_index = 0;
 511         int                              last_called_index = 0;
 512         bool                             repeated = false;
 513         bool                            refresh_idx = false;
 514
 515         ENTRY;
 516
 517         if (llh == NULL)
 518                 RETURN(-EINVAL);
 519
 520         lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
 521
 522         cur_offset = chunk_size = llh->llh_hdr.lrh_len;
 523         /* expect chunk_size to be power of two */
 524         LASSERT(is_power_of_2(chunk_size));
 525
 526         OBD_ALLOC_LARGE(buf, chunk_size);
 527         if (buf == NULL) {
 528                 lpi->lpi_rc = -ENOMEM;
 529                 RETURN(0);
 530         }
 531
 532         if (cd != NULL) {
 533                 last_called_index = cd->lpcd_first_idx;
 534                 index = cd->lpcd_first_idx + 1;
 535         }
 536         if (cd && cd->lpcd_last_idx)
 537                 last_index = cd->lpcd_last_idx;
 538         else if (cd && (cd->lpcd_read_mode & LLOG_READ_MODE_RAW))
 539                 last_index = loghandle->lgh_last_idx;
 540         else
 541                 last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 542
 543         while (rc == 0) {
 544                 struct llog_rec_hdr *rec;
 545                 off_t chunk_offset = 0;
 546                 unsigned int buf_offset = 0;
 547                 int     lh_last_idx;
 548                 int     synced_idx = 0;
 549
 550                 /* skip records not set in bitmap */
 551                 while (index <= last_index &&
 552                        llog_is_index_skipable(index, llh, cd))
 553                         ++index;
 554
 555                 /* There are no indices prior the last_index */
 556                 if (index > last_index)
 557                         break;
 558
 559                 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
 560                        last_index);
 561
 562 repeat:
 563                 /* get the buf with our target record; avoid old garbage */
 564                 memset(buf, 0, chunk_size);
 565                 /* the record index for outdated chunk data */
 566                 /* it is safe to process buffer until saved lgh_last_idx */
 567                 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 568                 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
 569                                      index, &cur_offset, buf, chunk_size);
 570                 if (repeated && rc)
 571                         CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
 572                                " buf_offset %u, rc = %d\n", cur_offset,
 573                                (__u64)chunk_offset, buf_offset, rc);
 574                 if (rc == -ESTALE)
 575                         GOTO(out, rc = 0);
 576                 /* we`ve tried to reread the chunk, but there is no
 577                  * new records */
 578                 if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
 579                     cur_offset)
 580                         GOTO(out, rc = 0);
 581                 if (rc != 0)
 582                         GOTO(out, rc);
 583
 584                 /* NB: after llog_next_block() call the cur_offset is the
 585                  * offset of the next block after read one.
 586                  * The absolute offset of the current chunk is calculated
 587                  * from cur_offset value and stored in chunk_offset variable.
 588                  */
 589                 if ((cur_offset & (chunk_size - 1)) != 0)
 590                         chunk_offset = cur_offset & ~(chunk_size - 1);
 591                 else
 592                         chunk_offset = cur_offset - chunk_size;
 593
 594                 /* NB: when rec->lrh_len is accessed it is already swabbed
 595                  * since it is used at the "end" of the loop and the rec
 596                  * swabbing is done at the beginning of the loop. */
 597                 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
 598                      (char *)rec < buf + chunk_size;
 599                      rec = llog_rec_hdr_next(rec)) {
 600
 601                         CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
 602                                rec, rec->lrh_type);
 603
 604                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 605                                 lustre_swab_llog_rec(rec);
 606
 607                         CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
 608                                rec->lrh_type, rec->lrh_index);
 609
 610                         if (index == (synced_idx + 1) &&
 611                             synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
 612                                 GOTO(out, rc = 0);
 613
 614                         if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
 615                                 cfs_fail_val == (unsigned int)
 616                                         (loghandle->lgh_id.lgl_oi.oi.oi_id &
 617                                          0xFFFFFFFF)) {
 618                                 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
 619                         }
 620
 621                         /* the bitmap could be changed during processing
 622                          * records from the chunk. For wrapped catalog
 623                          * it means we can read deleted record and try to
 624                          * process it. Check this case and reread the chunk.
 625                          * It is safe to process to lh_last_idx, including
 626                          * lh_last_idx if it was synced. We can not do <=
 627                          * comparison, cause for wrapped catalog lgh_last_idx
 628                          * could be less than index. So we detect last index
 629                          * for processing as index == lh_last_idx+1. But when
 630                          * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
 631                          * the first processing index is llh_cat_idx+1.The
 632                          * exception is !(lgh_last_idx == llh_cat_idx &&
 633                          * index == llh_cat_idx + 1), and after simplification
 634                          * it turns to
 635                          * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
 636                          * This exception is working for catalog only.
 637                          */
 638
 639                         if ((index == lh_last_idx && synced_idx != index) ||
 640                             (index == (lh_last_idx + 1) &&
 641                              lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
 642                             (rec->lrh_index == 0 && !repeated)) {
 643
 644                                 /* save offset inside buffer for the re-read */
 645                                 buf_offset = (char *)rec - (char *)buf;
 646                                 cur_offset = chunk_offset;
 647                                 repeated = true;
 648                                 /* We need to be sure lgh_last_idx
 649                                  * record was saved to disk
 650                                  */
 651                                 down_read(&loghandle->lgh_last_sem);
 652                                 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 653                                 up_read(&loghandle->lgh_last_sem);
 654                                 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
 655                                 goto repeat;
 656
 657                         }
 658
 659                         repeated = false;
 660
 661                         rc = llog_verify_record(loghandle, rec);
 662                         if (rc) {
 663                                 CERROR("%s: invalid record in llog "DFID
 664                                        " record for index %d/%d: rc = %d\n",
 665                                        loghandle2name(loghandle),
 666                                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 667                                        rec->lrh_index, index, rc);
 668                                 /*
 669                                  * the block seem to be corrupted, let's try
 670                                  * with the next one. reset rc to go to the
 671                                  * next chunk.
 672                                  */
 673                                 refresh_idx = true;
 674                                 index = 0;
 675                                 GOTO(repeat, rc = 0);
 676                         }
 677
 678                         if (rec->lrh_index < index) {
 679                                 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
 680                                        rec->lrh_index);
 681                                 continue;
 682                         }
 683
 684                         if (rec->lrh_index > index) {
 685                                 /* the record itself looks good, but we met a
 686                                  * gap which can be result of old bugs, just
 687                                  * keep going */
 688                                 CERROR("%s: "DFID" index %u, expected %u\n",
 689                                        loghandle2name(loghandle),
 690                                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 691                                        rec->lrh_index, index);
 692                                 index = rec->lrh_index;
 693                         }
 694
 695                         CDEBUG(D_OTHER,
 696                                "lrh_index: %d lrh_len: %d (%d remains)\n",
 697                                rec->lrh_index, rec->lrh_len,
 698                                (int)(buf + chunk_size - (char *)rec));
 699
 700                         /* lgh_cur_offset is used only at llog_test_3 */
 701                         loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
 702                                                     chunk_offset;
 703
 704                         /* if needed, process the callback on this record */
 705                         if (!llog_is_index_skipable(index, llh, cd)) {
 706                                 struct llog_cookie *lgc;
 707                                 __u64   tmp_off;
 708                                 int     tmp_idx;
 709
 710                                 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
 711                                         D_HA : D_OTHER),
 712                                        "index: %d, lh_last_idx: %d "
 713                                        "synced_idx: %d lgh_last_idx: %d\n",
 714                                        index, lh_last_idx, synced_idx,
 715                                        loghandle->lgh_last_idx);
 716
 717                                 if (lti != NULL) {
 718                                         lgc = &lti->lgi_cookie;
 719                                         /* store lu_env for recursive calls */
 720                                         tmp_off = lgc->lgc_offset;
 721                                         tmp_idx = lgc->lgc_index;
 722
 723                                         lgc->lgc_offset = (char *)rec -
 724                                                 (char *)buf + chunk_offset;
 725                                         lgc->lgc_index = rec->lrh_index;
 726                                 }
 727                                 /* using lu_env for passing record offset to
 728                                  * llog_write through various callbacks */
 729                                 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
 730                                                  lpi->lpi_cbdata);
 731                                 last_called_index = index;
 732
 733                                 if (lti != NULL) {
 734                                         lgc->lgc_offset = tmp_off;
 735                                         lgc->lgc_index = tmp_idx;
 736                                 }
 737
 738                                 if (rc == LLOG_PROC_BREAK) {
 739                                         GOTO(out, rc);
 740                                 } else if (rc == LLOG_DEL_RECORD) {
 741                                         rc = llog_cancel_rec(lpi->lpi_env,
 742                                                              loghandle,
 743                                                              rec->lrh_index);
 744                                         /* Allow parallel cancelling, ENOENT
 745                                          * means record was canceled at another
 746                                          * processing thread or callback
 747                                          */
 748                                         if (rc == -ENOENT)
 749                                                 rc = 0;
 750                                 }
 751                                 if (rc)
 752                                         GOTO(out, rc);
 753                                 /* some stupid callbacks directly cancel records
 754                                  * and delete llog. Check it and stop
 755                                  * processing. */
 756                                 if (loghandle->lgh_hdr == NULL ||
 757                                     loghandle->lgh_hdr->llh_count == 1)
 758                                         GOTO(out, rc = 0);
 759                         }
 760                         /* exit if the last index is reached */
 761                         if (index >= last_index)
 762                                 GOTO(out, rc = 0);
 763                         ++index;
 764                 }
 765         }
 766
 767 out:
 768         CDEBUG(D_HA, "stop processing %s "DOSTID":%x index %d count %d\n",
 769                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
 770                POSTID(&loghandle->lgh_id.lgl_oi), loghandle->lgh_id.lgl_ogen,
 771                index, llh->llh_count);
 772
 773         if (cd != NULL)
 774                 cd->lpcd_last_idx = last_called_index;
 775
 776         if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
 777                 if (dt_object_remote(loghandle->lgh_obj)) {
 778                         /* If it is remote object, then -EIO might means
 779                          * disconnection or eviction, let's return -EAGAIN,
 780                          * so for update recovery log processing, it will
 781                          * retry until the umount or abort recovery, see
 782                          * lod_sub_recovery_thread() */
 783                         CERROR("%s retry remote llog process\n",
 784                                loghandle2name(loghandle));
 785                         rc = -EAGAIN;
 786                 } else {
 787                         /* something bad happened to the processing of a local
 788                          * llog file, probably I/O error or the log got
 789                          * corrupted to be able to finally release the log we
 790                          * discard any remaining bits in the header */
 791                         CERROR("%s: Local llog found corrupted #"DOSTID":%x"
 792                                " %s index %d count %d\n",
 793                                loghandle2name(loghandle),
 794                                POSTID(&loghandle->lgh_id.lgl_oi),
 795                                loghandle->lgh_id.lgl_ogen,
 796                                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
 797                                 "plain"), index, llh->llh_count);
 798
 799                         while (index <= last_index) {
 800                                 if (test_bit_le(index,
 801                                                   LLOG_HDR_BITMAP(llh)) != 0)
 802                                         llog_cancel_rec(lpi->lpi_env, loghandle,
 803                                                         index);
 804                                 index++;
 805                         }
 806                         rc = 0;
 807                 }
 808         }
 809
 810         OBD_FREE_LARGE(buf, chunk_size);
 811         lpi->lpi_rc = rc;
 812         return 0;
 813 }
 814
 815 static int llog_process_thread_daemonize(void *arg)
 816 {
 817         struct llog_process_info        *lpi = arg;
 818         struct lu_env                    env;
 819         int                              rc;
 820         struct nsproxy                  *new_ns, *curr_ns = current->nsproxy;
 821
 822         task_lock(lpi->lpi_reftask);
 823         new_ns = lpi->lpi_reftask->nsproxy;
 824         if (curr_ns != new_ns) {
 825                 get_nsproxy(new_ns);
 826
 827                 current->nsproxy = new_ns;
 828                 /* XXX: we should call put_nsproxy() instead of
 829                  * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
 830                  * used outside of the kernel itself, because it calls
 831                  * free_nsproxy() which is not exported by the kernel
 832                  * (defined in kernel/nsproxy.c) */
 833                 if (curr_ns)
 834                         atomic_dec(&curr_ns->count);
 835         }
 836         task_unlock(lpi->lpi_reftask);
 837
 838         /* client env has no keys, tags is just 0 */
 839         rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
 840         if (rc)
 841                 goto out;
 842         lpi->lpi_env = &env;
 843
 844         rc = llog_process_thread(arg);
 845
 846         lu_env_fini(&env);
 847 out:
 848         complete(&lpi->lpi_completion);
 849         return rc;
 850 }
 851
 852 int llog_process_or_fork(const struct lu_env *env,
 853                          struct llog_handle *loghandle,
 854                          llog_cb_t cb, void *data, void *catdata, bool fork)
 855 {
 856         struct llog_process_info *lpi;
 857         struct llog_process_data *d = data;
 858         struct llog_process_cat_data *cd = catdata;
 859         __u32 flags = loghandle->lgh_hdr->llh_flags;
 860         int rc;
 861
 862         ENTRY;
 863
 864         OBD_ALLOC_PTR(lpi);
 865         if (lpi == NULL) {
 866                 CERROR("cannot alloc pointer\n");
 867                 RETURN(-ENOMEM);
 868         }
 869         lpi->lpi_loghandle = loghandle;
 870         lpi->lpi_cb        = cb;
 871         lpi->lpi_cbdata    = data;
 872         lpi->lpi_catdata   = catdata;
 873
 874         CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d read_mode %d\n",
 875                PFID(&loghandle->lgh_id.lgl_oi.oi_fid), flags,
 876                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
 877                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
 878                cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1,
 879                cd ? cd->lpcd_read_mode : -1);
 880         if (fork) {
 881                 struct task_struct *task;
 882
 883                 /* The new thread can't use parent env,
 884                  * init the new one in llog_process_thread_daemonize. */
 885                 lpi->lpi_env = NULL;
 886                 init_completion(&lpi->lpi_completion);
 887                 /* take reference to current, so that
 888                  * llog_process_thread_daemonize() can use it to switch to
 889                  * namespace associated with current  */
 890                 lpi->lpi_reftask = current;
 891                 task = kthread_run(llog_process_thread_daemonize, lpi,
 892                                    "llog_process_thread");
 893                 if (IS_ERR(task)) {
 894                         rc = PTR_ERR(task);
 895                         CERROR("%s: cannot start thread: rc = %d\n",
 896                                loghandle2name(loghandle), rc);
 897                         GOTO(out_lpi, rc);
 898                 }
 899                 wait_for_completion(&lpi->lpi_completion);
 900         } else {
 901                 lpi->lpi_env = env;
 902                 llog_process_thread(lpi);
 903         }
 904         rc = lpi->lpi_rc;
 905
 906 out_lpi:
 907         OBD_FREE_PTR(lpi);
 908         RETURN(rc);
 909 }
 910 EXPORT_SYMBOL(llog_process_or_fork);
 911
 912 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
 913                  llog_cb_t cb, void *data, void *catdata)
 914 {
 915         int rc;
 916         rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
 917         return rc == LLOG_DEL_PLAIN ? 0 : rc;
 918 }
 919 EXPORT_SYMBOL(llog_process);
 920
 921 static inline const struct cred *llog_raise_resource(void)
 922 {
 923         struct cred *cred = NULL;
 924
 925         if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
 926                 return cred;
 927
 928         cred = prepare_creds();
 929         if (!cred)
 930                 return cred;
 931
 932         cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
 933         return override_creds(cred);
 934 }
 935
 936 static inline void llog_restore_resource(const struct cred *old_cred)
 937 {
 938         if (old_cred)
 939                 revert_creds(old_cred);
 940 }
 941
 942 int llog_reverse_process(const struct lu_env *env,
 943                          struct llog_handle *loghandle, llog_cb_t cb,
 944                          void *data, void *catdata)
 945 {
 946         struct llog_log_hdr *llh = loghandle->lgh_hdr;
 947         struct llog_process_cat_data *cd = catdata;
 948         void *buf;
 949         int rc = 0, first_index = 1, index, idx;
 950         __u32   chunk_size = llh->llh_hdr.lrh_len;
 951         ENTRY;
 952
 953         OBD_ALLOC_LARGE(buf, chunk_size);
 954         if (buf == NULL)
 955                 RETURN(-ENOMEM);
 956
 957         if (cd != NULL)
 958                 first_index = cd->lpcd_first_idx + 1;
 959         if (cd != NULL && cd->lpcd_last_idx)
 960                 index = cd->lpcd_last_idx;
 961         else
 962                 index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 963
 964         while (rc == 0) {
 965                 struct llog_rec_hdr *rec;
 966                 struct llog_rec_tail *tail;
 967
 968                 /* skip records not set in bitmap */
 969                 while (index >= first_index &&
 970                        llog_is_index_skipable(index, llh, cd))
 971                         --index;
 972
 973                 LASSERT(index >= first_index - 1);
 974                 if (index == first_index - 1)
 975                         break;
 976
 977                 /* get the buf with our target record; avoid old garbage */
 978                 memset(buf, 0, chunk_size);
 979                 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
 980                 if (rc)
 981                         GOTO(out, rc);
 982
 983                 rec = buf;
 984                 idx = rec->lrh_index;
 985                 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
 986                 while (idx < index) {
 987                         rec = (void *)rec + rec->lrh_len;
 988                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 989                                 lustre_swab_llog_rec(rec);
 990                         idx ++;
 991                 }
 992                 LASSERT(idx == index);
 993                 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
 994
 995                 /* process records in buffer, starting where we found one */
 996                 while ((void *)tail > buf) {
 997                         if (tail->lrt_index == 0)
 998                                 GOTO(out, rc = 0); /* no more records */
 999
1000                         /* if needed, process the callback on this record */
1001                         if (!llog_is_index_skipable(index, llh, cd)) {
1002                                 rec = (void *)tail - tail->lrt_len +
1003                                       sizeof(*tail);
1004
1005                                 rc = cb(env, loghandle, rec, data);
1006                                 if (rc == LLOG_PROC_BREAK) {
1007                                         GOTO(out, rc);
1008                                 } else if (rc == LLOG_DEL_RECORD) {
1009                                         rc = llog_cancel_rec(env, loghandle,
1010                                                              tail->lrt_index);
1011                                 }
1012                                 if (rc)
1013                                         GOTO(out, rc);
1014                         }
1015
1016                         /* previous record, still in buffer? */
1017                         --index;
1018                         if (index < first_index)
1019                                 GOTO(out, rc = 0);
1020                         tail = (void *)tail - tail->lrt_len;
1021                 }
1022         }
1023
1024 out:
1025         if (buf != NULL)
1026                 OBD_FREE_LARGE(buf, chunk_size);
1027         RETURN(rc);
1028 }
1029 EXPORT_SYMBOL(llog_reverse_process);
1030
1031 /**
1032  * new llog API
1033  *
1034  * API functions:
1035  *      llog_open - open llog, may not exist
1036  *      llog_exist - check if llog exists
1037  *      llog_close - close opened llog, pair for open, frees llog_handle
1038  *      llog_declare_create - declare llog creation
1039  *      llog_create - create new llog on disk, need transaction handle
1040  *      llog_declare_write_rec - declaration of llog write
1041  *      llog_write_rec - write llog record on disk, need transaction handle
1042  *      llog_declare_add - declare llog catalog record addition
1043  *      llog_add - add llog record in catalog, need transaction handle
1044  */
1045 int llog_exist(struct llog_handle *loghandle)
1046 {
1047         const struct llog_operations *lop;
1048         int rc;
1049
1050         ENTRY;
1051
1052         rc = llog_handle2ops(loghandle, &lop);
1053         if (rc)
1054                 RETURN(rc);
1055         if (lop->lop_exist == NULL)
1056                 RETURN(-EOPNOTSUPP);
1057
1058         rc = lop->lop_exist(loghandle);
1059         RETURN(rc);
1060 }
1061 EXPORT_SYMBOL(llog_exist);
1062
1063 int llog_declare_create(const struct lu_env *env,
1064                         struct llog_handle *loghandle, struct thandle *th)
1065 {
1066         const struct cred *old_cred;
1067         const struct llog_operations *lop;
1068         int rc;
1069
1070         ENTRY;
1071
1072         rc = llog_handle2ops(loghandle, &lop);
1073         if (rc)
1074                 RETURN(rc);
1075         if (lop->lop_declare_create == NULL)
1076                 RETURN(-EOPNOTSUPP);
1077
1078         old_cred = llog_raise_resource();
1079         rc = lop->lop_declare_create(env, loghandle, th);
1080         llog_restore_resource(old_cred);
1081         RETURN(rc);
1082 }
1083
1084 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1085                 struct thandle *th)
1086 {
1087         const struct cred *old_cred;
1088         const struct llog_operations *lop;
1089         int rc;
1090
1091         ENTRY;
1092
1093         rc = llog_handle2ops(handle, &lop);
1094         if (rc)
1095                 RETURN(rc);
1096         if (lop->lop_create == NULL)
1097                 RETURN(-EOPNOTSUPP);
1098
1099         old_cred = llog_raise_resource();
1100         rc = lop->lop_create(env, handle, th);
1101         llog_restore_resource(old_cred);
1102         RETURN(rc);
1103 }
1104
1105 int llog_declare_write_rec(const struct lu_env *env,
1106                            struct llog_handle *handle,
1107                            struct llog_rec_hdr *rec, int idx,
1108                            struct thandle *th)
1109 {
1110         const struct cred *old_cred;
1111         const struct llog_operations *lop;
1112         int rc;
1113
1114         ENTRY;
1115
1116         rc = llog_handle2ops(handle, &lop);
1117         if (rc)
1118                 RETURN(rc);
1119         LASSERT(lop);
1120         if (lop->lop_declare_write_rec == NULL)
1121                 RETURN(-EOPNOTSUPP);
1122
1123         old_cred = llog_raise_resource();
1124         rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1125         llog_restore_resource(old_cred);
1126         RETURN(rc);
1127 }
1128
1129 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1130                    struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1131                    int idx, struct thandle *th)
1132 {
1133         const struct cred *old_cred;
1134         const struct llog_operations *lop;
1135         int rc, buflen;
1136
1137         ENTRY;
1138
1139         /* API sanity checks */
1140         if (handle == NULL) {
1141                 CERROR("loghandle is missed\n");
1142                 RETURN(-EPROTO);
1143         } else if (handle->lgh_obj == NULL) {
1144                 CERROR("loghandle %p with NULL object\n",
1145                         handle);
1146                 RETURN(-EPROTO);
1147         } else if (th == NULL) {
1148                 CERROR("%s: missed transaction handle\n",
1149                        loghandle2name(handle));
1150                 RETURN(-EPROTO);
1151         } else if (handle->lgh_hdr == NULL) {
1152                 CERROR("%s: loghandle %p with no header\n",
1153                        loghandle2name(handle), handle);
1154                 RETURN(-EPROTO);
1155         }
1156
1157         rc = llog_handle2ops(handle, &lop);
1158         if (rc)
1159                 RETURN(rc);
1160
1161         if (lop->lop_write_rec == NULL)
1162                 RETURN(-EOPNOTSUPP);
1163
1164         buflen = rec->lrh_len;
1165         LASSERT(cfs_size_round(buflen) == buflen);
1166
1167         old_cred = llog_raise_resource();
1168         rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1169         llog_restore_resource(old_cred);
1170         RETURN(rc);
1171 }
1172
1173 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1174              struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1175              struct thandle *th)
1176 {
1177         const struct cred *old_cred;
1178         int rc;
1179
1180         ENTRY;
1181
1182         if (lgh->lgh_logops->lop_add == NULL)
1183                 RETURN(-EOPNOTSUPP);
1184
1185         old_cred = llog_raise_resource();
1186         rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1187         llog_restore_resource(old_cred);
1188         RETURN(rc);
1189 }
1190 EXPORT_SYMBOL(llog_add);
1191
1192 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1193                      struct llog_rec_hdr *rec, struct thandle *th)
1194 {
1195         const struct cred *old_cred;
1196         int rc;
1197
1198         ENTRY;
1199
1200         if (lgh->lgh_logops->lop_declare_add == NULL)
1201                 RETURN(-EOPNOTSUPP);
1202
1203         old_cred = llog_raise_resource();
1204         rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1205         llog_restore_resource(old_cred);
1206         RETURN(rc);
1207 }
1208 EXPORT_SYMBOL(llog_declare_add);
1209
1210 /**
1211  * Helper function to open llog or create it if doesn't exist.
1212  * It hides all transaction handling from caller.
1213  */
1214 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1215                      struct llog_handle **res, struct llog_logid *logid,
1216                      char *name)
1217 {
1218         struct dt_device        *d;
1219         struct thandle          *th;
1220         int                      rc;
1221
1222         ENTRY;
1223
1224         rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1225         if (rc)
1226                 RETURN(rc);
1227
1228         if (llog_exist(*res))
1229                 RETURN(0);
1230
1231         LASSERT((*res)->lgh_obj != NULL);
1232
1233         d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1234
1235         if (unlikely(unlikely(d->dd_rdonly)))
1236                 RETURN(-EROFS);
1237
1238         th = dt_trans_create(env, d);
1239         if (IS_ERR(th))
1240                 GOTO(out, rc = PTR_ERR(th));
1241
1242         /* Create update llog object synchronously, which
1243          * happens during inialization process see
1244          * lod_sub_prep_llog(), to make sure the update
1245          * llog object is created before corss-MDT writing
1246          * updates into the llog object */
1247         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1248                 th->th_sync = 1;
1249
1250         th->th_wait_submit = 1;
1251         rc = llog_declare_create(env, *res, th);
1252         if (rc == 0) {
1253                 rc = dt_trans_start_local(env, d, th);
1254                 if (rc == 0)
1255                         rc = llog_create(env, *res, th);
1256         }
1257         dt_trans_stop(env, d, th);
1258 out:
1259         if (rc)
1260                 llog_close(env, *res);
1261         RETURN(rc);
1262 }
1263 EXPORT_SYMBOL(llog_open_create);
1264
1265 /**
1266  * Helper function to delete existent llog.
1267  */
1268 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1269                struct llog_logid *logid, char *name)
1270 {
1271         struct llog_handle      *handle;
1272         int                      rc = 0, rc2;
1273
1274         ENTRY;
1275
1276         /* nothing to erase */
1277         if (name == NULL && logid == NULL)
1278                 RETURN(0);
1279
1280         rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1281         if (rc < 0)
1282                 RETURN(rc);
1283
1284         rc = llog_init_handle(env, handle, LLOG_F_IS_PLAIN, NULL);
1285         if (rc == 0)
1286                 rc = llog_destroy(env, handle);
1287
1288         rc2 = llog_close(env, handle);
1289         if (rc == 0)
1290                 rc = rc2;
1291         RETURN(rc);
1292 }
1293 EXPORT_SYMBOL(llog_erase);
1294
1295 /*
1296  * Helper function for write record in llog.
1297  * It hides all transaction handling from caller.
1298  * Valid only with local llog.
1299  */
1300 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1301                struct llog_rec_hdr *rec, int idx)
1302 {
1303         struct dt_device        *dt;
1304         struct thandle          *th;
1305         bool                    need_cookie;
1306         int                     rc;
1307
1308         ENTRY;
1309
1310         LASSERT(loghandle);
1311         LASSERT(loghandle->lgh_ctxt);
1312         LASSERT(loghandle->lgh_obj != NULL);
1313
1314         dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1315
1316         if (unlikely(unlikely(dt->dd_rdonly)))
1317                 RETURN(-EROFS);
1318
1319         th = dt_trans_create(env, dt);
1320         if (IS_ERR(th))
1321                 RETURN(PTR_ERR(th));
1322
1323         rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1324         if (rc)
1325                 GOTO(out_trans, rc);
1326
1327         th->th_wait_submit = 1;
1328         rc = dt_trans_start_local(env, dt, th);
1329         if (rc)
1330                 GOTO(out_trans, rc);
1331
1332         need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1333
1334         down_write(&loghandle->lgh_lock);
1335         if (need_cookie) {
1336                 struct llog_thread_info *lti = llog_info(env);
1337
1338                 /* cookie comes from llog_process_thread */
1339                 rc = llog_write_rec(env, loghandle, rec, &lti->lgi_cookie,
1340                                     rec->lrh_index, th);
1341                 /* upper layer didn`t pass cookie so change rc */
1342                 rc = (rc == 1 ? 0 : rc);
1343         } else {
1344                 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1345         }
1346
1347         up_write(&loghandle->lgh_lock);
1348 out_trans:
1349         dt_trans_stop(env, dt, th);
1350         RETURN(rc);
1351 }
1352 EXPORT_SYMBOL(llog_write);
1353
1354 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1355               struct llog_handle **lgh, struct llog_logid *logid,
1356               char *name, enum llog_open_param open_param)
1357 {
1358         const struct cred *old_cred;
1359         int      rc;
1360
1361         ENTRY;
1362
1363         LASSERT(ctxt);
1364         LASSERT(ctxt->loc_logops);
1365
1366         if (ctxt->loc_logops->lop_open == NULL) {
1367                 *lgh = NULL;
1368                 RETURN(-EOPNOTSUPP);
1369         }
1370
1371         *lgh = llog_alloc_handle();
1372         if (*lgh == NULL)
1373                 RETURN(-ENOMEM);
1374         (*lgh)->lgh_ctxt = ctxt;
1375         (*lgh)->lgh_logops = ctxt->loc_logops;
1376
1377         old_cred = llog_raise_resource();
1378         rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1379         llog_restore_resource(old_cred);
1380         if (rc) {
1381                 llog_free_handle(*lgh);
1382                 *lgh = NULL;
1383         }
1384         RETURN(rc);
1385 }
1386 EXPORT_SYMBOL(llog_open);
1387
1388 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1389 {
1390         return llog_handle_put(env, loghandle);
1391 }
1392 EXPORT_SYMBOL(llog_close);
1393
1394 /**
1395  * Helper function to get the llog size in records. It is used by MGS
1396  * mostly to check that config llog exists and contains data.
1397  *
1398  * \param[in] env       execution environment
1399  * \param[in] ctxt      llog context
1400  * \param[in] name      llog name
1401  *
1402  * \retval              true if there are records in llog besides a header
1403  * \retval              false on error or llog without records
1404  */
1405 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1406                   char *name)
1407 {
1408         struct llog_handle      *llh;
1409         int                      rc = 0;
1410
1411         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1412         if (rc < 0) {
1413                 if (likely(rc == -ENOENT))
1414                         rc = 0;
1415                 GOTO(out, rc);
1416         }
1417
1418         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1419         if (rc)
1420                 GOTO(out_close, rc);
1421         rc = llog_get_size(llh);
1422
1423 out_close:
1424         llog_close(env, llh);
1425 out:
1426         /* The header is record 1, the llog is still considered as empty
1427          * if there is only header */
1428         return (rc <= 1);
1429 }
1430 EXPORT_SYMBOL(llog_is_empty);
1431
1432 /* this callback run in raw read mode (canceled record are processed) */
1433 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1434                       struct llog_rec_hdr *rec, void *data)
1435 {
1436         struct llog_handle *copy_llh = data;
1437         int idx = rec->lrh_index;
1438         int rc;
1439
1440         ENTRY;
1441
1442         /* Append all records */
1443         rc = llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1444
1445         /* Cancel the record if it is canceled on the source */
1446         if (!rc && !test_bit_le(idx, LLOG_HDR_BITMAP(llh->lgh_hdr)))
1447                 rc = llog_cancel_rec(env, copy_llh, copy_llh->lgh_last_idx);
1448
1449         RETURN(rc);
1450 }
1451
1452 /* backup plain llog */
1453 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1454                 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1455                 char *name, char *backup)
1456 {
1457         struct llog_handle *llh, *bllh;
1458         struct llog_process_cat_data cd = {0};
1459         int rc;
1460
1461         ENTRY;
1462
1463         /* open original log */
1464         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1465         if (rc < 0) {
1466                 /* the -ENOENT case is also reported to the caller
1467                  * but silently so it should handle that if needed.
1468                  */
1469                 if (rc != -ENOENT)
1470                         CERROR("%s: failed to open log %s: rc = %d\n",
1471                                obd->obd_name, name, rc);
1472                 RETURN(rc);
1473         }
1474
1475         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1476         if (rc)
1477                 GOTO(out_close, rc);
1478
1479         /* Make sure there's no old backup log */
1480         rc = llog_erase(env, bctxt, NULL, backup);
1481         if (rc < 0 && rc != -ENOENT)
1482                 GOTO(out_close, rc);
1483
1484         /* open backup log */
1485         rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1486         if (rc) {
1487                 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1488                        obd->obd_name, backup, rc);
1489                 GOTO(out_close, rc);
1490         }
1491
1492         /* check that backup llog is not the same object as original one */
1493         if (llh->lgh_obj == bllh->lgh_obj) {
1494                 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1495                        obd->obd_name, name, backup, llh->lgh_obj,
1496                        bllh->lgh_obj);
1497                 GOTO(out_backup, rc = -EEXIST);
1498         }
1499
1500         rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1501         if (rc)
1502                 GOTO(out_backup, rc);
1503
1504         /* Read canceled records to have an exact copy */
1505         cd.lpcd_read_mode = LLOG_READ_MODE_RAW;
1506         /* Copy log record by record */
1507         rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1508                                   &cd, false);
1509         if (rc)
1510                 CERROR("%s: failed to backup log %s: rc = %d\n",
1511                        obd->obd_name, name, rc);
1512 out_backup:
1513         llog_close(env, bllh);
1514 out_close:
1515         llog_close(env, llh);
1516         RETURN(rc);
1517 }
1518 EXPORT_SYMBOL(llog_backup);
1519
1520 /* Get size of llog */
1521 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1522 {
1523         int rc;
1524         struct lu_attr la;
1525
1526         rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1527         if (rc) {
1528                 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1529                        loghandle2name(llh), PFID(&llh->lgh_id.lgl_oi.oi_fid),
1530                        rc);
1531                 return 0;
1532         }
1533
1534         return la.la_size;
1535 }
1536 EXPORT_SYMBOL(llog_size);
1537