lustre/obdclass/llog_osd.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  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 /*
  33  * lustre/obdclass/llog_osd.c
  34  *
  35  * Low level llog routines on top of OSD API
  36  *
  37  * This file provides set of methods for llog operations on top of
  38  * dt_device. It contains all supported llog_operations interfaces and
  39  * supplimental functions.
  40  *
  41  * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
  42  * Author: Mikhail Pershin <mike.pershin@intel.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_LOG
  46
  47 #include <dt_object.h>
  48 #include <llog_swab.h>
  49 #include <lustre_fid.h>
  50 #include <obd.h>
  51 #include <obd_class.h>
  52
  53 #include "llog_internal.h"
  54 #include "local_storage.h"
  55
  56 /**
  57  * Implementation of the llog_operations::lop_declare_create
  58  *
  59  * This function is a wrapper over local_storage API function
  60  * local_object_declare_create().
  61  *
  62  * \param[in] env       execution environment
  63  * \param[in] los       local_storage for bottom storage device
  64  * \param[in] o         dt_object to create
  65  * \param[in] th        current transaction handle
  66  *
  67  * \retval              0 on successful declaration of the new object
  68  * \retval              negative error if declaration was failed
  69  */
  70 static int llog_osd_declare_new_object(const struct lu_env *env,
  71                                        struct local_oid_storage *los,
  72                                        struct dt_object *o,
  73                                        struct thandle *th)
  74 {
  75         struct llog_thread_info *lgi = llog_info(env);
  76
  77         lgi->lgi_attr.la_valid = LA_MODE;
  78         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
  79         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
  80
  81         return local_object_declare_create(env, los, o, &lgi->lgi_attr,
  82                                            &lgi->lgi_dof, th);
  83 }
  84
  85 /**
  86  * Implementation of the llog_operations::lop_create
  87  *
  88  * This function is a wrapper over local_storage API function
  89  * local_object_create().
  90  *
  91  * \param[in] env       execution environment
  92  * \param[in] los       local_storage for bottom storage device
  93  * \param[in] o         dt_object to create
  94  * \param[in] th        current transaction handle
  95  *
  96  * \retval              0 on successful creation of the new object
  97  * \retval              negative error if creation was failed
  98  */
  99 static int llog_osd_create_new_object(const struct lu_env *env,
 100                                       struct local_oid_storage *los,
 101                                       struct dt_object *o,
 102                                       struct thandle *th)
 103 {
 104         struct llog_thread_info *lgi = llog_info(env);
 105
 106         lgi->lgi_attr.la_valid = LA_MODE;
 107         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
 108         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
 109
 110         return local_object_create(env, los, o, &lgi->lgi_attr,
 111                                    &lgi->lgi_dof, th);
 112 }
 113
 114 /**
 115  * Implementation of the llog_operations::lop_exist
 116  *
 117  * This function checks that llog exists on storage.
 118  *
 119  * \param[in] handle    llog handle of the current llog
 120  *
 121  * \retval              true if llog object exists and is not just destroyed
 122  * \retval              false if llog doesn't exist or just destroyed
 123  */
 124 static int llog_osd_exist(struct llog_handle *handle)
 125 {
 126         LASSERT(handle->lgh_obj);
 127         return dt_object_exists(handle->lgh_obj) &&
 128                 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header);
 129 }
 130
 131 static void *rec_tail(struct llog_rec_hdr *rec)
 132 {
 133         return (void *)((char *)rec + rec->lrh_len -
 134                         sizeof(struct llog_rec_tail));
 135 }
 136
 137 /**
 138  * Write a padding record to the llog
 139  *
 140  * This function writes a padding record to the end of llog. That may
 141  * be needed if llog contains records of variable size, e.g. config logs
 142  * or changelogs.
 143  * The padding record just aligns llog to the llog chunk_size boundary if
 144  * the current record doesn't fit in the remaining space.
 145  *
 146  * It allocates full length to avoid two separate writes for header and tail.
 147  * Such 2-steps scheme needs extra protection and complex error handling.
 148  *
 149  * \param[in]     env   execution environment
 150  * \param[in]     o     dt_object to create
 151  * \param[in,out] off   pointer to the padding start offset
 152  * \param[in]     len   padding length
 153  * \param[in]     index index of the padding record in a llog
 154  * \param[in]     th    current transaction handle
 155  *
 156  * \retval              0 on successful padding write
 157  * \retval              negative error if write failed
 158  */
 159 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
 160                         loff_t *off, int len, int index, struct thandle *th)
 161 {
 162         struct llog_thread_info *lgi = llog_info(env);
 163         struct llog_rec_hdr     *rec;
 164         struct llog_rec_tail    *tail;
 165         int                      rc;
 166
 167         ENTRY;
 168
 169         LASSERT(th);
 170         LASSERT(off);
 171         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
 172
 173         OBD_ALLOC(rec, len);
 174         if (rec == NULL)
 175                 RETURN(-ENOMEM);
 176
 177         rec->lrh_len = len;
 178         rec->lrh_index = index;
 179         rec->lrh_type = LLOG_PAD_MAGIC;
 180
 181         tail = rec_tail(rec);
 182         tail->lrt_len = len;
 183         tail->lrt_index = index;
 184
 185         lgi->lgi_buf.lb_buf = rec;
 186         lgi->lgi_buf.lb_len = len;
 187         rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
 188         if (rc)
 189                 CERROR("%s: error writing padding record: rc = %d\n",
 190                        o->do_lu.lo_dev->ld_obd->obd_name, rc);
 191
 192         OBD_FREE(rec, len);
 193         RETURN(rc);
 194 }
 195
 196 /**
 197  * Implementation of the llog_operations::lop_read_header
 198  *
 199  * This function reads the current llog header from the bottom storage
 200  * device.
 201  *
 202  * \param[in] env       execution environment
 203  * \param[in] handle    llog handle of the current llog
 204  *
 205  * \retval              0 on successful header read
 206  * \retval              negative error if read failed
 207  */
 208 static int llog_osd_read_header(const struct lu_env *env,
 209                                 struct llog_handle *handle)
 210 {
 211         struct llog_rec_hdr     *llh_hdr;
 212         struct dt_object        *o;
 213         struct llog_thread_info *lgi;
 214         enum llog_flag           flags;
 215         int                      rc;
 216
 217         ENTRY;
 218
 219         o = handle->lgh_obj;
 220         LASSERT(o);
 221
 222         lgi = llog_info(env);
 223
 224         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 225         if (rc)
 226                 RETURN(rc);
 227
 228         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 229
 230         if (lgi->lgi_attr.la_size == 0) {
 231                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 232                 RETURN(LLOG_EEMPTY);
 233         }
 234
 235         flags = handle->lgh_hdr->llh_flags;
 236
 237         lgi->lgi_off = 0;
 238         lgi->lgi_buf.lb_buf = handle->lgh_hdr;
 239         lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
 240         rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
 241         llh_hdr = &handle->lgh_hdr->llh_hdr;
 242         if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
 243                 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
 244                        o->do_lu.lo_dev->ld_obd->obd_name,
 245                        PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
 246                        -EFAULT);
 247
 248                 if (rc >= 0)
 249                         rc = -EFAULT;
 250
 251                 RETURN(rc);
 252         }
 253
 254         if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
 255                 lustre_swab_llog_hdr(handle->lgh_hdr);
 256
 257         if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
 258                 CERROR("%s: bad log %s "DFID" header magic: %#x "
 259                        "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
 260                        handle->lgh_name ? handle->lgh_name : "",
 261                        PFID(lu_object_fid(&o->do_lu)),
 262                        llh_hdr->lrh_type, LLOG_HDR_MAGIC);
 263                 RETURN(-EIO);
 264         } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
 265                    llh_hdr->lrh_len > handle->lgh_hdr_size) {
 266                 CERROR("%s: incorrectly sized log %s "DFID" header: "
 267                        "%#x (expected at least %#x)\n"
 268                        "you may need to re-run lconf --write_conf.\n",
 269                        o->do_lu.lo_dev->ld_obd->obd_name,
 270                        handle->lgh_name ? handle->lgh_name : "",
 271                        PFID(lu_object_fid(&o->do_lu)),
 272                        llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
 273                 RETURN(-EIO);
 274         } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
 275                    LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
 276                    LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
 277                         llh_hdr->lrh_len) {
 278                 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
 279                        "%#x : rc = %d\n",
 280                        o->do_lu.lo_dev->ld_obd->obd_name,
 281                        handle->lgh_name ? handle->lgh_name : "",
 282                        PFID(lu_object_fid(&o->do_lu)),
 283                        LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
 284                 RETURN(-EIO);
 285         }
 286
 287         handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
 288         handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
 289
 290         RETURN(0);
 291 }
 292
 293 /**
 294  * Implementation of the llog_operations::lop_declare_write
 295  *
 296  * This function declares the new record write.
 297  *
 298  * \param[in] env       execution environment
 299  * \param[in] loghandle llog handle of the current llog
 300  * \param[in] rec       llog record header. This is a real header of the full
 301  *                      llog record to write. This is the beginning of buffer
 302  *                      to write, the length of buffer is stored in
 303  *                      \a rec::lrh_len
 304  * \param[in] idx       index of the llog record. If \a idx == -1 then this is
 305  *                      append case, otherwise \a idx is the index of record
 306  *                      to modify
 307  * \param[in] th        current transaction handle
 308  *
 309  * \retval              0 on successful declaration
 310  * \retval              negative error if declaration failed
 311  */
 312 static int llog_osd_declare_write_rec(const struct lu_env *env,
 313                                       struct llog_handle *loghandle,
 314                                       struct llog_rec_hdr *rec,
 315                                       int idx, struct thandle *th)
 316 {
 317         struct llog_thread_info *lgi = llog_info(env);
 318         __u32                   chunk_size;
 319         struct dt_object        *o;
 320         int                      rc;
 321
 322         ENTRY;
 323
 324         LASSERT(env);
 325         LASSERT(th);
 326         LASSERT(loghandle);
 327         LASSERT(rec);
 328         LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
 329
 330         o = loghandle->lgh_obj;
 331         LASSERT(o);
 332
 333         chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
 334         lgi->lgi_buf.lb_len = chunk_size;
 335         lgi->lgi_buf.lb_buf = NULL;
 336         /* each time we update header */
 337         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
 338                                      th);
 339         if (rc || idx == 0) /* if error or just header */
 340                 RETURN(rc);
 341
 342         /**
 343          * the pad record can be inserted so take into account double
 344          * record size
 345          */
 346         lgi->lgi_buf.lb_len = chunk_size * 2;
 347         lgi->lgi_buf.lb_buf = NULL;
 348         /* XXX: implement declared window or multi-chunks approach */
 349         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
 350
 351         RETURN(rc);
 352 }
 353
 354 /**
 355  * Implementation of the llog_operations::lop_write
 356  *
 357  * This function writes the new record in the llog or modify the existed one.
 358  *
 359  * \param[in]  env              execution environment
 360  * \param[in]  loghandle        llog handle of the current llog
 361  * \param[in]  rec              llog record header. This is a real header of
 362  *                              the full llog record to write. This is
 363  *                              the beginning of buffer to write, the length
 364  *                              of buffer is stored in \a rec::lrh_len
 365  * \param[in,out] reccookie     pointer to the cookie to return back if needed.
 366  *                              It is used for further cancel of this llog
 367  *                              record.
 368  * \param[in]  idx              index of the llog record. If \a idx == -1 then
 369  *                              this is append case, otherwise \a idx is
 370  *                              the index of record to modify
 371  * \param[in]  th               current transaction handle
 372  *
 373  * \retval                      0 on successful write && \a reccookie == NULL
 374  *                              1 on successful write && \a reccookie != NULL
 375  * \retval                      negative error if write failed
 376  */
 377 static int llog_osd_write_rec(const struct lu_env *env,
 378                               struct llog_handle *loghandle,
 379                               struct llog_rec_hdr *rec,
 380                               struct llog_cookie *reccookie,
 381                               int idx, struct thandle *th)
 382 {
 383         struct llog_thread_info *lgi = llog_info(env);
 384         struct llog_log_hdr     *llh;
 385         int                      reclen = rec->lrh_len;
 386         int                      index, rc;
 387         struct llog_rec_tail    *lrt;
 388         struct dt_object        *o;
 389         __u32                   chunk_size;
 390         size_t                   left;
 391         __u32                   orig_last_idx;
 392         ENTRY;
 393
 394         llh = loghandle->lgh_hdr;
 395         o = loghandle->lgh_obj;
 396
 397         chunk_size = llh->llh_hdr.lrh_len;
 398         CDEBUG(D_OTHER, "new record %x to "DFID"\n",
 399                rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
 400
 401         if (!llog_osd_exist(loghandle))
 402                 RETURN(-ENOENT);
 403
 404         /* record length should not bigger than  */
 405         if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
 406                 RETURN(-E2BIG);
 407
 408         /* sanity check for fixed-records llog */
 409         if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
 410                 LASSERT(llh->llh_size != 0);
 411                 LASSERT(llh->llh_size == reclen);
 412         }
 413
 414         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 415         if (rc)
 416                 RETURN(rc);
 417
 418         /**
 419          * The modification case.
 420          * If idx set then the record with that index must be modified.
 421          * There are three cases possible:
 422          * 1) the common case is the llog header update (idx == 0)
 423          * 2) the llog record modification during llog process.
 424          *    This is indicated by the \a loghandle::lgh_cur_idx > 0.
 425          *    In that case the \a loghandle::lgh_cur_offset
 426          * 3) otherwise this is assumed that llog consist of records of
 427          *    fixed size, i.e. catalog. The llog header must has llh_size
 428          *    field equal to record size. The record offset is calculated
 429          *    just by /a idx value
 430          *
 431          * During modification we don't need extra header update because
 432          * the bitmap and record count are not changed. The record header
 433          * and tail remains the same too.
 434          */
 435         if (idx != LLOG_NEXT_IDX) {
 436                 /* llog can be empty only when first record is being written */
 437                 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
 438
 439                 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
 440                         CERROR("%s: modify unset record %u\n",
 441                                o->do_lu.lo_dev->ld_obd->obd_name, idx);
 442                         RETURN(-ENOENT);
 443                 }
 444
 445                 if (idx != rec->lrh_index) {
 446                         CERROR("%s: modify index mismatch %d %u\n",
 447                                o->do_lu.lo_dev->ld_obd->obd_name, idx,
 448                                rec->lrh_index);
 449                         RETURN(-EFAULT);
 450                 }
 451
 452                 if (idx == LLOG_HEADER_IDX) {
 453                         /* llog header update */
 454                         __u32   *bitmap = LLOG_HDR_BITMAP(llh);
 455
 456                         lgi->lgi_off = 0;
 457
 458                         /* If it does not indicate the bitmap index
 459                          * (reccookie == NULL), then it means update
 460                          * the whole update header. Otherwise only
 461                          * update header and bits needs to be updated,
 462                          * and in DNE cases, it will signaficantly
 463                          * shrink the RPC size.
 464                          * see distribute_txn_cancel_records()*/
 465                         if (reccookie == NULL) {
 466                                 lgi->lgi_buf.lb_len = reclen;
 467                                 lgi->lgi_buf.lb_buf = rec;
 468                                 rc = dt_record_write(env, o, &lgi->lgi_buf,
 469                                                      &lgi->lgi_off, th);
 470                                 RETURN(rc);
 471                         }
 472
 473                         /* update the header */
 474                         lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
 475                         lgi->lgi_buf.lb_buf = llh;
 476                         rc = dt_record_write(env, o, &lgi->lgi_buf,
 477                                              &lgi->lgi_off, th);
 478                         if (rc != 0)
 479                                 RETURN(rc);
 480
 481                         /* update the bitmap */
 482                         index = reccookie->lgc_index;
 483                         lgi->lgi_off = llh->llh_bitmap_offset +
 484                                       (index / (sizeof(*bitmap) * 8)) *
 485                                                         sizeof(*bitmap);
 486                         lgi->lgi_buf.lb_len = sizeof(*bitmap);
 487                         lgi->lgi_buf.lb_buf =
 488                                         &bitmap[index/(sizeof(*bitmap)*8)];
 489                         rc = dt_record_write(env, o, &lgi->lgi_buf,
 490                                              &lgi->lgi_off, th);
 491
 492                         RETURN(rc);
 493                 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 494                         lgi->lgi_off = llh->llh_hdr.lrh_len +
 495                                        (idx - 1) * reclen;
 496                 } else if (reccookie != NULL && reccookie->lgc_index > 0) {
 497                         /**
 498                          * The lgc_offset can be used only if index is
 499                          * the same.
 500                          */
 501                         if (idx != reccookie->lgc_index) {
 502                                 CERROR("%s: modify index mismatch %d %d\n",
 503                                        o->do_lu.lo_dev->ld_obd->obd_name, idx,
 504                                        reccookie->lgc_index);
 505                                 RETURN(-EFAULT);
 506                         }
 507
 508                         lgi->lgi_off = reccookie->lgc_offset;
 509                         CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
 510                                "len:%u offset %llu\n",
 511                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid), idx,
 512                                rec->lrh_len, (long long)lgi->lgi_off);
 513                 } else {
 514                         /* This can be result of lgh_cur_idx is not set during
 515                          * llog processing or llh_size is not set to proper
 516                          * record size for fixed records llog. Therefore it is
 517                          * impossible to get record offset. */
 518                         CERROR("%s: can't get record offset, idx:%d, "
 519                                "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
 520                                idx, rec->lrh_len);
 521                         RETURN(-EFAULT);
 522                 }
 523
 524                 /* update only data, header and tail remain the same */
 525                 lgi->lgi_off += sizeof(struct llog_rec_hdr);
 526                 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
 527                 lgi->lgi_buf.lb_buf = REC_DATA(rec);
 528                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 529                 if (rc == 0 && reccookie) {
 530                         reccookie->lgc_lgl = loghandle->lgh_id;
 531                         reccookie->lgc_index = idx;
 532                         rc = 1;
 533                 }
 534                 RETURN(rc);
 535         }
 536
 537         /**
 538          * The append case.
 539          * The most common case of using llog. The new index is assigned to
 540          * the new record, new bit is set in llog bitmap and llog count is
 541          * incremented.
 542          *
 543          * Make sure that records don't cross a chunk boundary, so we can
 544          * process them page-at-a-time if needed.  If it will cross a chunk
 545          * boundary, write in a fake (but referenced) entry to pad the chunk.
 546          */
 547
 548
 549         /* simulate ENOSPC when new plain llog is being added to the
 550          * catalog */
 551         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
 552             llh->llh_flags & LLOG_F_IS_CAT)
 553                 RETURN(-ENOSPC);
 554
 555         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 556         orig_last_idx = loghandle->lgh_last_idx;
 557         lgi->lgi_off = lgi->lgi_attr.la_size;
 558
 559         if (loghandle->lgh_max_size > 0 &&
 560             lgi->lgi_off >= loghandle->lgh_max_size) {
 561                 CDEBUG(D_OTHER, "llog is getting too large (%u > %u) at %u "
 562                        DFID"\n", (unsigned)lgi->lgi_off,
 563                        loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
 564                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
 565                 /* this is to signal that this llog is full */
 566                 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 567                 RETURN(-ENOSPC);
 568         }
 569
 570         left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
 571         /* NOTE: padding is a record, but no bit is set */
 572         if (left != 0 && left != reclen &&
 573             left < (reclen + LLOG_MIN_REC_SIZE)) {
 574                 index = loghandle->lgh_last_idx + 1;
 575                 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
 576                 if (rc)
 577                         RETURN(rc);
 578
 579                 loghandle->lgh_last_idx++; /* for pad rec */
 580         }
 581         /* if it's the last idx in log file, then return -ENOSPC
 582          * or wrap around if a catalog */
 583         if (llog_is_full(loghandle) ||
 584             unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
 585                      OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
 586                      loghandle->lgh_last_idx >= cfs_fail_val)) {
 587                 if (llh->llh_flags & LLOG_F_IS_CAT)
 588                         loghandle->lgh_last_idx = 0;
 589                 else
 590                         RETURN(-ENOSPC);
 591         }
 592
 593         /* increment the last_idx along with llh_tail index, they should
 594          * be equal for a llog lifetime */
 595         loghandle->lgh_last_idx++;
 596         index = loghandle->lgh_last_idx;
 597         LLOG_HDR_TAIL(llh)->lrt_index = index;
 598         /**
 599          * NB: the caller should make sure only 1 process access
 600          * the lgh_last_idx, e.g. append should be exclusive.
 601          * Otherwise it might hit the assert.
 602          */
 603         LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
 604         rec->lrh_index = index;
 605         lrt = rec_tail(rec);
 606         lrt->lrt_len = rec->lrh_len;
 607         lrt->lrt_index = rec->lrh_index;
 608
 609         /* the lgh_hdr_mutex protects llog header data from concurrent
 610          * update/cancel, the llh_count and llh_bitmap are protected */
 611         mutex_lock(&loghandle->lgh_hdr_mutex);
 612         if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
 613                 CERROR("%s: index %u already set in log bitmap\n",
 614                        o->do_lu.lo_dev->ld_obd->obd_name, index);
 615                 mutex_unlock(&loghandle->lgh_hdr_mutex);
 616                 LBUG(); /* should never happen */
 617         }
 618         llh->llh_count++;
 619
 620         if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
 621                 /* Update the minimum size of the llog record */
 622                 if (llh->llh_size == 0)
 623                         llh->llh_size = reclen;
 624                 else if (reclen < llh->llh_size)
 625                         llh->llh_size = reclen;
 626         }
 627
 628         if (lgi->lgi_attr.la_size == 0) {
 629                 lgi->lgi_off = 0;
 630                 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
 631                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 632                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 633                 if (rc != 0)
 634                         GOTO(out_unlock, rc);
 635         } else {
 636                 __u32   *bitmap = LLOG_HDR_BITMAP(llh);
 637
 638                 /* Note: If this is not initialization (size == 0), then do not
 639                  * write the whole header (8k bytes), only update header/tail
 640                  * and bits needs to be updated. Because this update might be
 641                  * part of cross-MDT operation, which needs to write these
 642                  * updates into the update log(32KB limit) and also pack inside
 643                  * the RPC (1MB limit), if we write 8K for each operation, which
 644                  * will cost a lot space, and keep us adding more updates to one
 645                  * update log.*/
 646                 lgi->lgi_off = 0;
 647                 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
 648                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 649                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 650                 if (rc != 0)
 651                         GOTO(out_unlock, rc);
 652
 653                 lgi->lgi_off = llh->llh_bitmap_offset +
 654                               (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
 655                 lgi->lgi_buf.lb_len = sizeof(*bitmap);
 656                 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
 657                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 658                 if (rc != 0)
 659                         GOTO(out_unlock, rc);
 660
 661                 lgi->lgi_off =  (unsigned long)LLOG_HDR_TAIL(llh) -
 662                                 (unsigned long)llh;
 663                 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
 664                 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
 665                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 666                 if (rc != 0)
 667                         GOTO(out_unlock, rc);
 668         }
 669
 670 out_unlock:
 671         /* unlock here for remote object */
 672         mutex_unlock(&loghandle->lgh_hdr_mutex);
 673         if (rc)
 674                 GOTO(out, rc);
 675
 676         /* computed index can be used to determine offset for fixed-size
 677          * records. This also allows to handle Catalog wrap around case */
 678         if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 679                 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
 680         } else {
 681                 rc = dt_attr_get(env, o, &lgi->lgi_attr);
 682                 if (rc)
 683                         GOTO(out, rc);
 684
 685                 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 686                 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
 687                                      lgi->lgi_off);
 688         }
 689
 690         lgi->lgi_buf.lb_len = reclen;
 691         lgi->lgi_buf.lb_buf = rec;
 692         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 693         if (rc < 0)
 694                 GOTO(out, rc);
 695
 696         CDEBUG(D_HA, "added record "DFID".%u, %u off%llu\n",
 697                PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
 698                lgi->lgi_off);
 699         if (reccookie != NULL) {
 700                 reccookie->lgc_lgl = loghandle->lgh_id;
 701                 reccookie->lgc_index = index;
 702                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 703                     (rec->lrh_type == MDS_SETATTR64_REC))
 704                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 705                 else if (rec->lrh_type == OST_SZ_REC)
 706                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 707                 else
 708                         reccookie->lgc_subsys = -1;
 709                 rc = 1;
 710         }
 711         RETURN(rc);
 712 out:
 713         /* cleanup llog for error case */
 714         mutex_lock(&loghandle->lgh_hdr_mutex);
 715         ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
 716         llh->llh_count--;
 717         mutex_unlock(&loghandle->lgh_hdr_mutex);
 718
 719         /* restore llog last_idx */
 720         if (dt_object_remote(o)) {
 721                 loghandle->lgh_last_idx = orig_last_idx;
 722         } else if (--loghandle->lgh_last_idx == 0 &&
 723             (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
 724                 /* catalog had just wrap-around case */
 725                 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 726         }
 727
 728         LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
 729
 730         RETURN(rc);
 731 }
 732
 733 /**
 734  * We can skip reading at least as many log blocks as the number of
 735  * minimum sized log records we are skipping.  If it turns out
 736  * that we are not far enough along the log (because the
 737  * actual records are larger than minimum size) we just skip
 738  * some more records.
 739  *
 740  * Note: in llog_process_thread, it will use bitmap offset as
 741  * the index to locate the record, which also includs some pad
 742  * records, whose record size is very small, and it also does not
 743  * consider pad record when recording minimum record size (otherwise
 744  * min_record size might be too small), so in some rare cases,
 745  * it might skip too much record for @goal, see llog_osd_next_block().
 746  *
 747  * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
 748  * as the min record size to skip over, usually because in the previous
 749  * try, it skip too much record, see loog_osd_next(prev)_block().
 750  */
 751 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
 752                                   int curr, int goal, __u32 chunk_size,
 753                                   bool force_mini_rec)
 754 {
 755         struct llog_log_hdr *llh = lgh->lgh_hdr;
 756
 757         /* Goal should not bigger than the record count */
 758         if (goal > lgh->lgh_last_idx)
 759                 goal = lgh->lgh_last_idx;
 760
 761         if (goal > curr) {
 762                 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 763                         *off = chunk_size + (goal - 1) * llh->llh_size;
 764                 } else {
 765                         __u64 min_rec_size = LLOG_MIN_REC_SIZE;
 766
 767                         if (llh->llh_size > 0 && !force_mini_rec)
 768                                 min_rec_size = llh->llh_size;
 769
 770                         *off = *off + (goal - curr - 1) * min_rec_size;
 771                 }
 772         }
 773         /* always align with lower chunk boundary*/
 774         *off &= ~(chunk_size - 1);
 775 }
 776
 777 /**
 778  * Remove optional fields that the client doesn't expect.
 779  * This is typically in order to ensure compatibility with older clients.
 780  * It is assumed that since we exclusively remove fields, the block will be
 781  * big enough to handle the remapped records. It is also assumed that records
 782  * of a block have the same format (i.e.: the same features enabled).
 783  *
 784  * \param[in,out]    hdr           Header of the block of records to remap.
 785  * \param[in,out]    last_hdr      Last header, don't read past this point.
 786  * \param[in]        flags         Flags describing the fields to keep.
 787  * \param[in]        extra_flags   Flags describing the extra fields to keep.
 788  */
 789 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
 790                                      struct llog_rec_hdr *last_hdr,
 791                                      struct llog_handle *loghandle)
 792 {
 793         enum changelog_rec_flags flags = CLF_SUPPORTED;
 794         enum changelog_rec_extra_flags extra_flags = CLFE_SUPPORTED;
 795
 796         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_XATTR))
 797                 extra_flags &= ~CLFE_XATTR;
 798         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_OMODE))
 799                 extra_flags &= ~CLFE_OPEN;
 800         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID))
 801                 extra_flags &= ~CLFE_NID;
 802         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_UIDGID))
 803                 extra_flags &= ~CLFE_UIDGID;
 804         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_EXTRA_FLAGS))
 805                 flags &= ~CLF_EXTRA_FLAGS;
 806         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 807                 flags &= ~CLF_JOBID;
 808
 809         if (flags == CLF_SUPPORTED && extra_flags == CLFE_SUPPORTED)
 810                 return;
 811
 812         if (hdr->lrh_type != CHANGELOG_REC)
 813                 return;
 814
 815         do {
 816                 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
 817                 enum changelog_rec_extra_flags xflag = CLFE_INVALID;
 818
 819                 if (flags & CLF_EXTRA_FLAGS &&
 820                     rec->cr_flags & CLF_EXTRA_FLAGS) {
 821                         xflag = changelog_rec_extra_flags(rec)->cr_extra_flags &
 822                                 extra_flags;
 823                 }
 824
 825                 if (unlikely(hdr->lrh_len == 0)) {
 826                         /* It is corruption case, we cannot know the next rec,
 827                          * jump to the last one directly to avoid dead loop. */
 828                         LCONSOLE(D_WARNING, "Hit invalid llog record: "
 829                                  "idx %u, type %u, id %u\n",
 830                                  hdr->lrh_index, hdr->lrh_type, hdr->lrh_id);
 831                         hdr = llog_rec_hdr_next(last_hdr);
 832                         if (unlikely(hdr == last_hdr))
 833                                 LCONSOLE(D_WARNING, "The last record crashed: "
 834                                          "idx %u, type %u, id %u\n",
 835                                          hdr->lrh_index, hdr->lrh_type,
 836                                          hdr->lrh_id);
 837                         break;
 838                 }
 839
 840                 changelog_remap_rec(rec, rec->cr_flags & flags, xflag);
 841                 hdr = llog_rec_hdr_next(hdr);
 842                 /* Yield CPU to avoid soft-lockup if there are too many records
 843                  * to be handled. */
 844                 cond_resched();
 845         } while ((char *)hdr <= (char *)last_hdr);
 846 }
 847
 848 /**
 849  * Implementation of the llog_operations::lop_next_block
 850  *
 851  * This function finds the the next llog block to return which contains
 852  * record with required index. It is main part of llog processing.
 853  *
 854  * \param[in]     env           execution environment
 855  * \param[in]     loghandle     llog handle of the current llog
 856  * \param[in,out] cur_idx       index preceeding cur_offset
 857  * \param[in]     next_idx      target index to find
 858  * \param[in,out] cur_offset    furtherst point read in the file
 859  * \param[in]     buf           pointer to data buffer to fill
 860  * \param[in]     len           required len to read, it is
 861  *                              usually llog chunk_size.
 862  *
 863  * \retval                      0 on successful buffer read
 864  * \retval                      negative value on error
 865  */
 866 static int llog_osd_next_block(const struct lu_env *env,
 867                                struct llog_handle *loghandle, int *cur_idx,
 868                                int next_idx, __u64 *cur_offset, void *buf,
 869                                int len)
 870 {
 871         struct llog_thread_info *lgi = llog_info(env);
 872         struct dt_object        *o;
 873         struct dt_device        *dt;
 874         int                      rc;
 875         __u32                   chunk_size;
 876         int last_idx = *cur_idx;
 877         __u64 last_offset = *cur_offset;
 878         bool force_mini_rec = false;
 879
 880         ENTRY;
 881
 882         LASSERT(env);
 883         LASSERT(lgi);
 884
 885         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 886         if (len == 0 || len & (chunk_size - 1))
 887                 RETURN(-EINVAL);
 888
 889         LASSERT(loghandle);
 890         LASSERT(loghandle->lgh_ctxt);
 891
 892         o = loghandle->lgh_obj;
 893         LASSERT(o);
 894         LASSERT(dt_object_exists(o));
 895         dt = lu2dt_dev(o->do_lu.lo_dev);
 896         LASSERT(dt);
 897
 898         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 899         if (rc)
 900                 GOTO(out, rc);
 901
 902         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
 903                "%llu), size %llu\n", next_idx, *cur_idx,
 904                *cur_offset, lgi->lgi_attr.la_size);
 905
 906         while (*cur_offset < lgi->lgi_attr.la_size) {
 907                 struct llog_rec_hdr     *rec, *last_rec;
 908                 struct llog_rec_tail    *tail;
 909
 910                 llog_skip_over(loghandle, cur_offset, *cur_idx,
 911                                next_idx, chunk_size, force_mini_rec);
 912
 913                 /* read up to next llog chunk_size block */
 914                 lgi->lgi_buf.lb_len = chunk_size -
 915                                       (*cur_offset & (chunk_size - 1));
 916                 lgi->lgi_buf.lb_buf = buf;
 917
 918                 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
 919                 if (rc < 0) {
 920                         if (rc == -EBADR && !force_mini_rec)
 921                                 goto retry;
 922
 923                         CERROR("%s: can't read llog block from log "DFID
 924                                " offset %llu: rc = %d\n",
 925                                o->do_lu.lo_dev->ld_obd->obd_name,
 926                                PFID(lu_object_fid(&o->do_lu)), *cur_offset,
 927                                rc);
 928                         GOTO(out, rc);
 929                 }
 930
 931                 if (rc < len) {
 932                         /* signal the end of the valid buffer to
 933                          * llog_process */
 934                         memset(buf + rc, 0, len - rc);
 935                 }
 936
 937                 if (rc == 0) { /* end of file, nothing to do */
 938                         if (!force_mini_rec)
 939                                 goto retry;
 940                         GOTO(out, rc);
 941                 }
 942
 943                 if (rc < sizeof(*tail)) {
 944                         if (!force_mini_rec)
 945                                 goto retry;
 946
 947                         CERROR("%s: invalid llog block at log id "DFID":%x "
 948                                "offset %llu\n",
 949                                o->do_lu.lo_dev->ld_obd->obd_name,
 950                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 951                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 952                         GOTO(out, rc = -EINVAL);
 953                 }
 954
 955                 rec = buf;
 956                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 957                         lustre_swab_llog_rec(rec);
 958
 959                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 960                                                 sizeof(struct llog_rec_tail));
 961                 /* get the last record in block */
 962                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 963                                                    tail->lrt_len);
 964
 965                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 966                         lustre_swab_llog_rec(last_rec);
 967
 968                 if (last_rec->lrh_index != tail->lrt_index) {
 969                         CERROR("%s: invalid llog tail at log id "DFID":%x "
 970                                "offset %llu last_rec idx %u tail idx %u"
 971                                "lrt len %u read_size %d\n",
 972                                o->do_lu.lo_dev->ld_obd->obd_name,
 973                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 974                                loghandle->lgh_id.lgl_ogen, *cur_offset,
 975                                last_rec->lrh_index, tail->lrt_index,
 976                                tail->lrt_len, rc);
 977                         GOTO(out, rc = -EINVAL);
 978                 }
 979
 980                 *cur_idx = tail->lrt_index;
 981
 982                 /* this shouldn't happen */
 983                 if (tail->lrt_index == 0) {
 984                         CERROR("%s: invalid llog tail at log id "DFID":%x "
 985                                "offset %llu bytes %d\n",
 986                                o->do_lu.lo_dev->ld_obd->obd_name,
 987                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 988                                loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
 989                         GOTO(out, rc = -EINVAL);
 990                 }
 991                 if (tail->lrt_index < next_idx) {
 992                         last_idx = *cur_idx;
 993                         last_offset = *cur_offset;
 994                         continue;
 995                 }
 996
 997                 /* sanity check that the start of the new buffer is no farther
 998                  * than the record that we wanted.  This shouldn't happen. */
 999                 if (rec->lrh_index > next_idx) {
1000                         if (!force_mini_rec && next_idx > last_idx)
1001                                 goto retry;
1002
1003                         CERROR("%s: missed desired record? %u > %u\n",
1004                                o->do_lu.lo_dev->ld_obd->obd_name,
1005                                rec->lrh_index, next_idx);
1006                         GOTO(out, rc = -ENOENT);
1007                 }
1008
1009                 /* Trim unsupported extensions for compat w/ older clients */
1010                 changelog_block_trim_ext(rec, last_rec, loghandle);
1011
1012                 GOTO(out, rc = 0);
1013
1014 retry:
1015                 /* Note: because there are some pad records in the
1016                  * llog, so llog_skip_over() might skip too much
1017                  * records, let's try skip again with minimum record */
1018                 force_mini_rec = true;
1019                 *cur_offset = last_offset;
1020                 *cur_idx = last_idx;
1021         }
1022         GOTO(out, rc = -EIO);
1023 out:
1024         return rc;
1025 }
1026
1027 /**
1028  * Implementation of the llog_operations::lop_prev_block
1029  *
1030  * This function finds the llog block to return which contains
1031  * record with required index but in reverse order - from end of llog
1032  * to the beginning.
1033  * It is main part of reverse llog processing.
1034  *
1035  * \param[in] env       execution environment
1036  * \param[in] loghandle llog handle of the current llog
1037  * \param[in] prev_idx  target index to find
1038  * \param[in] buf       pointer to data buffer to fill
1039  * \param[in] len       required len to read, it is llog_chunk_size usually.
1040  *
1041  * \retval              0 on successful buffer read
1042  * \retval              negative value on error
1043  */
1044 static int llog_osd_prev_block(const struct lu_env *env,
1045                                struct llog_handle *loghandle,
1046                                int prev_idx, void *buf, int len)
1047 {
1048         struct llog_thread_info *lgi = llog_info(env);
1049         struct dt_object        *o;
1050         struct dt_device        *dt;
1051         loff_t                   cur_offset;
1052         __u32                   chunk_size;
1053         int                      rc;
1054
1055         ENTRY;
1056
1057         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1058         if (len == 0 || len & (chunk_size - 1))
1059                 RETURN(-EINVAL);
1060
1061         CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1062
1063         LASSERT(loghandle);
1064         LASSERT(loghandle->lgh_ctxt);
1065
1066         o = loghandle->lgh_obj;
1067         LASSERT(o);
1068         LASSERT(dt_object_exists(o));
1069         dt = lu2dt_dev(o->do_lu.lo_dev);
1070         LASSERT(dt);
1071
1072         /* Let's only use mini record size for previous block read
1073          * for now XXX */
1074         cur_offset = chunk_size;
1075         llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1076                        chunk_size, true);
1077
1078         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1079         if (rc)
1080                 GOTO(out, rc);
1081
1082         while (cur_offset < lgi->lgi_attr.la_size) {
1083                 struct llog_rec_hdr     *rec, *last_rec;
1084                 struct llog_rec_tail    *tail;
1085
1086                 lgi->lgi_buf.lb_len = len;
1087                 lgi->lgi_buf.lb_buf = buf;
1088                 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1089                 if (rc < 0) {
1090                         CERROR("%s: can't read llog block from log "DFID
1091                                " offset %llu: rc = %d\n",
1092                                o->do_lu.lo_dev->ld_obd->obd_name,
1093                                PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1094                         GOTO(out, rc);
1095                 }
1096
1097                 if (rc == 0) /* end of file, nothing to do */
1098                         GOTO(out, rc);
1099
1100                 if (rc < sizeof(*tail)) {
1101                         CERROR("%s: invalid llog block at log id "DFID":%x "
1102                                "offset %llu\n",
1103                                o->do_lu.lo_dev->ld_obd->obd_name,
1104                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1105                                loghandle->lgh_id.lgl_ogen, cur_offset);
1106                         GOTO(out, rc = -EINVAL);
1107                 }
1108
1109                 rec = buf;
1110                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1111                         lustre_swab_llog_rec(rec);
1112
1113                 tail = (struct llog_rec_tail *)((char *)buf + rc -
1114                                                 sizeof(struct llog_rec_tail));
1115                 /* get the last record in block */
1116                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1117                                                    le32_to_cpu(tail->lrt_len));
1118
1119                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1120                         lustre_swab_llog_rec(last_rec);
1121                 LASSERT(last_rec->lrh_index == tail->lrt_index);
1122
1123                 /* this shouldn't happen */
1124                 if (tail->lrt_index == 0) {
1125                         CERROR("%s: invalid llog tail at log id "DFID":%x "
1126                                "offset %llu\n",
1127                                o->do_lu.lo_dev->ld_obd->obd_name,
1128                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1129                                loghandle->lgh_id.lgl_ogen, cur_offset);
1130                         GOTO(out, rc = -EINVAL);
1131                 }
1132                 if (tail->lrt_index < prev_idx)
1133                         continue;
1134
1135                 /* sanity check that the start of the new buffer is no farther
1136                  * than the record that we wanted.  This shouldn't happen. */
1137                 if (rec->lrh_index > prev_idx) {
1138                         CERROR("%s: missed desired record? %u > %u\n",
1139                                o->do_lu.lo_dev->ld_obd->obd_name,
1140                                rec->lrh_index, prev_idx);
1141                         GOTO(out, rc = -ENOENT);
1142                 }
1143
1144                 /* Trim unsupported extensions for compat w/ older clients */
1145                 changelog_block_trim_ext(rec, last_rec, loghandle);
1146
1147                 GOTO(out, rc = 0);
1148         }
1149         GOTO(out, rc = -EIO);
1150 out:
1151         return rc;
1152 }
1153
1154 /**
1155  * This is helper function to get llog directory object. It is used by named
1156  * llog operations to find/insert/delete llog entry from llog directory.
1157  *
1158  * \param[in] env       execution environment
1159  * \param[in] ctxt      llog context
1160  *
1161  * \retval              dt_object of llog directory
1162  * \retval              ERR_PTR of negative value on error
1163  */
1164 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1165                                           struct llog_ctxt *ctxt)
1166 {
1167         struct dt_device        *dt;
1168         struct dt_thread_info   *dti = dt_info(env);
1169         struct dt_object        *dir;
1170         int                      rc;
1171
1172         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1173         if (ctxt->loc_dir == NULL) {
1174                 rc = dt_root_get(env, dt, &dti->dti_fid);
1175                 if (rc)
1176                         return ERR_PTR(rc);
1177                 dir = dt_locate(env, dt, &dti->dti_fid);
1178
1179                 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1180                         dt_object_put(env, dir);
1181                         return ERR_PTR(-ENOTDIR);
1182                 }
1183         } else {
1184                 lu_object_get(&ctxt->loc_dir->do_lu);
1185                 dir = ctxt->loc_dir;
1186         }
1187
1188         return dir;
1189 }
1190
1191 /**
1192  * Implementation of the llog_operations::lop_open
1193  *
1194  * This function opens the llog by its logid or by name, it may open also
1195  * non existent llog and assing then new id to it.
1196  * The llog_open/llog_close pair works similar to lu_object_find/put,
1197  * the object may not exist prior open. The result of open is just dt_object
1198  * in the llog header.
1199  *
1200  * \param[in] env               execution environment
1201  * \param[in] handle            llog handle of the current llog
1202  * \param[in] logid             logid of llog to open (nameless llog)
1203  * \param[in] name              name of llog to open (named llog)
1204  * \param[in] open_param
1205  *                              LLOG_OPEN_NEW - new llog, may not exist
1206  *                              LLOG_OPEN_EXIST - old llog, must exist
1207  *
1208  * \retval                      0 on successful open, llog_handle::lgh_obj
1209  *                              contains the dt_object of the llog.
1210  * \retval                      negative value on error
1211  */
1212 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1213                          struct llog_logid *logid, char *name,
1214                          enum llog_open_param open_param)
1215 {
1216         struct llog_thread_info         *lgi = llog_info(env);
1217         struct llog_ctxt                *ctxt = handle->lgh_ctxt;
1218         struct dt_object                *o;
1219         struct dt_device                *dt;
1220         struct ls_device                *ls;
1221         struct local_oid_storage        *los = NULL;
1222         int                              rc = 0;
1223         bool new_id = false;
1224
1225         ENTRY;
1226
1227         LASSERT(env);
1228         LASSERT(ctxt);
1229         LASSERT(ctxt->loc_exp);
1230         LASSERT(ctxt->loc_exp->exp_obd);
1231         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1232         LASSERT(dt);
1233         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1234                 struct lu_object_conf conf = { 0 };
1235                 if (logid != NULL) {
1236                         logid_to_fid(logid, &lgi->lgi_fid);
1237                 } else {
1238                         /* If logid == NULL, then it means the caller needs
1239                          * to allocate new FID (llog_cat_declare_add_rec()). */
1240                         rc = obd_fid_alloc(env, ctxt->loc_exp,
1241                                            &lgi->lgi_fid, NULL);
1242                         if (rc < 0)
1243                                 RETURN(rc);
1244                         rc = 0;
1245                         conf.loc_flags = LOC_F_NEW;
1246                 }
1247
1248                 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1249                                  dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1250                 if (IS_ERR(o))
1251                         RETURN(PTR_ERR(o));
1252
1253                 goto after_open;
1254         }
1255
1256         ls = ls_device_get(dt);
1257         if (IS_ERR(ls))
1258                 RETURN(PTR_ERR(ls));
1259
1260         mutex_lock(&ls->ls_los_mutex);
1261         los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1262         mutex_unlock(&ls->ls_los_mutex);
1263         LASSERT(los);
1264         ls_device_put(env, ls);
1265
1266         LASSERT(handle);
1267
1268         if (logid != NULL) {
1269                 logid_to_fid(logid, &lgi->lgi_fid);
1270         } else if (name) {
1271                 struct dt_object *llog_dir;
1272
1273                 llog_dir = llog_osd_dir_get(env, ctxt);
1274                 if (IS_ERR(llog_dir))
1275                         GOTO(out, rc = PTR_ERR(llog_dir));
1276                 dt_read_lock(env, llog_dir, 0);
1277                 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1278                 dt_read_unlock(env, llog_dir);
1279                 dt_object_put(env, llog_dir);
1280                 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1281                         /* generate fid for new llog */
1282                         rc = local_object_fid_generate(env, los,
1283                                                        &lgi->lgi_fid);
1284                         new_id = true;
1285                 }
1286                 if (rc < 0)
1287                         GOTO(out, rc);
1288                 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1289                 if (handle->lgh_name)
1290                         strcpy(handle->lgh_name, name);
1291                 else
1292                         GOTO(out, rc = -ENOMEM);
1293         } else {
1294                 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1295                 /* generate fid for new llog */
1296 generate:
1297                 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1298                 if (rc < 0)
1299                         GOTO(out, rc);
1300                 new_id = true;
1301         }
1302
1303         o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1304         if (IS_ERR(o))
1305                 GOTO(out_name, rc = PTR_ERR(o));
1306
1307         if (dt_object_exists(o) && new_id) {
1308                 /* llog exists with just generated ID, e.g. some old llog file
1309                  * still is in use or is orphan, drop a warn and skip it. */
1310                 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1311                        ", skipping\n",
1312                        o->do_lu.lo_dev->ld_obd->obd_name,
1313                        PFID(lu_object_fid(&o->do_lu)));
1314                 dt_object_put(env, o);
1315                 /* just skip this llog ID, we shouldn't delete it because we
1316                  * don't know exactly what is its purpose and state. */
1317                 goto generate;
1318         }
1319
1320 after_open:
1321         /* No new llog is expected but doesn't exist */
1322         if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
1323                 CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
1324                        o->do_lu.lo_dev->ld_obd->obd_name,
1325                        PFID(lu_object_fid(&o->do_lu)), o);
1326                 GOTO(out_put, rc = -ENOENT);
1327         }
1328         fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1329         handle->lgh_obj = o;
1330         handle->private_data = los;
1331         LASSERT(handle->lgh_ctxt);
1332
1333         RETURN(rc);
1334
1335 out_put:
1336         dt_object_put(env, o);
1337 out_name:
1338         if (handle->lgh_name != NULL)
1339                 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1340 out:
1341         if (los != NULL)
1342                 dt_los_put(los);
1343         RETURN(rc);
1344 }
1345
1346 /**
1347  * Get dir for regular fid log object
1348  *
1349  * Get directory for regular fid log object, and these regular fid log
1350  * object will be inserted under this directory, to satisfy the FS
1351  * consistency check, e2fsck etc.
1352  *
1353  * \param [in] env      execution environment
1354  * \param [in] dto      llog object
1355  *
1356  * \retval              pointer to the directory if it is found.
1357  * \retval              ERR_PTR(negative errno) if it fails.
1358  */
1359 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1360                                                struct dt_object *dto)
1361 {
1362         struct llog_thread_info *lgi = llog_info(env);
1363         struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1364         struct lu_seq_range     *range = &lgi->lgi_range;
1365         struct lu_fid           *dir_fid = &lgi->lgi_fid;
1366         struct dt_object        *dir;
1367         int                     rc;
1368         ENTRY;
1369
1370         fld_range_set_any(range);
1371         LASSERT(ss != NULL);
1372         rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1373                                    fid_seq(lu_object_fid(&dto->do_lu)), range);
1374         if (rc < 0)
1375                 RETURN(ERR_PTR(rc));
1376
1377         lu_update_log_dir_fid(dir_fid, range->lsr_index);
1378         dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1379         if (IS_ERR(dir))
1380                 RETURN(dir);
1381
1382         if (!dt_try_as_dir(env, dir)) {
1383                 dt_object_put(env, dir);
1384                 RETURN(ERR_PTR(-ENOTDIR));
1385         }
1386
1387         RETURN(dir);
1388 }
1389
1390 /**
1391  * Add llog object with regular FID to name entry
1392  *
1393  * Add llog object with regular FID to name space, and each llog
1394  * object on each MDT will be /update_log_dir/[seq:oid:ver],
1395  * so to satisfy the namespace consistency check, e2fsck etc.
1396  *
1397  * \param [in] env      execution environment
1398  * \param [in] dto      llog object
1399  * \param [in] th       thandle
1400  * \param [in] declare  if it is declare or execution
1401  *
1402  * \retval              0 if insertion succeeds.
1403  * \retval              negative errno if insertion fails.
1404  */
1405 static int
1406 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1407                                     struct dt_object *dto,
1408                                     struct thandle *th, bool declare)
1409 {
1410         struct llog_thread_info *lgi = llog_info(env);
1411         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1412         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1413         struct dt_object        *dir;
1414         char                    *name = lgi->lgi_name;
1415         int                     rc;
1416         ENTRY;
1417
1418         if (!fid_is_norm(fid))
1419                 RETURN(0);
1420
1421         dir = llog_osd_get_regular_fid_dir(env, dto);
1422         if (IS_ERR(dir))
1423                 RETURN(PTR_ERR(dir));
1424
1425         rec->rec_fid = fid;
1426         rec->rec_type = S_IFREG;
1427         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1428         dt_write_lock(env, dir, 0);
1429         if (declare) {
1430                 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1431                                (struct dt_key *)name, th);
1432         } else {
1433                 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1434                                (struct dt_key *)name, th);
1435         }
1436         dt_write_unlock(env, dir);
1437
1438         dt_object_put(env, dir);
1439         RETURN(rc);
1440 }
1441
1442
1443 /**
1444  * Implementation of the llog_operations::lop_declare_create
1445  *
1446  * This function declares the llog create. It declares also name insert
1447  * into llog directory in case of named llog.
1448  *
1449  * \param[in] env       execution environment
1450  * \param[in] res       llog handle of the current llog
1451  * \param[in] th        current transaction handle
1452  *
1453  * \retval              0 on successful create declaration
1454  * \retval              negative value on error
1455  */
1456 static int llog_osd_declare_create(const struct lu_env *env,
1457                                    struct llog_handle *res, struct thandle *th)
1458 {
1459         struct llog_thread_info         *lgi = llog_info(env);
1460         struct dt_insert_rec            *rec = &lgi->lgi_dt_rec;
1461         struct local_oid_storage        *los;
1462         struct dt_object                *o;
1463         int                              rc;
1464
1465         ENTRY;
1466
1467         LASSERT(res->lgh_obj);
1468         LASSERT(th);
1469
1470         /* object can be created by another thread */
1471         o = res->lgh_obj;
1472         if (dt_object_exists(o))
1473                 RETURN(0);
1474
1475         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1476                 struct llog_thread_info *lgi = llog_info(env);
1477
1478                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1479                 lgi->lgi_attr.la_size = 0;
1480                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1481                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1482
1483                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1484                                        &lgi->lgi_dof, th);
1485                 if (rc < 0)
1486                         RETURN(rc);
1487
1488
1489                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1490
1491                 RETURN(rc);
1492         }
1493         los = res->private_data;
1494         LASSERT(los);
1495
1496         rc = llog_osd_declare_new_object(env, los, o, th);
1497         if (rc)
1498                 RETURN(rc);
1499
1500         /* do not declare header initialization here as it's declared
1501          * in llog_osd_declare_write_rec() which is always called */
1502
1503         if (res->lgh_name) {
1504                 struct dt_object *llog_dir;
1505
1506                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1507                 if (IS_ERR(llog_dir))
1508                         RETURN(PTR_ERR(llog_dir));
1509                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1510                 rec->rec_fid = &lgi->lgi_fid;
1511                 rec->rec_type = S_IFREG;
1512                 rc = dt_declare_insert(env, llog_dir,
1513                                        (struct dt_rec *)rec,
1514                                        (struct dt_key *)res->lgh_name, th);
1515                 dt_object_put(env, llog_dir);
1516                 if (rc)
1517                         CERROR("%s: can't declare named llog %s: rc = %d\n",
1518                                o->do_lu.lo_dev->ld_obd->obd_name,
1519                                res->lgh_name, rc);
1520         }
1521         RETURN(rc);
1522 }
1523
1524 /**
1525  * Implementation of the llog_operations::lop_create
1526  *
1527  * This function creates the llog according with llog_handle::lgh_obj
1528  * and llog_handle::lgh_name.
1529  *
1530  * \param[in] env       execution environment
1531  * \param[in] res       llog handle of the current llog
1532  * \param[in] th        current transaction handle
1533  *
1534  * \retval              0 on successful create
1535  * \retval              negative value on error
1536  */
1537 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1538                            struct thandle *th)
1539 {
1540         struct llog_thread_info *lgi = llog_info(env);
1541         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1542         struct local_oid_storage *los;
1543         struct dt_object        *o;
1544         int                      rc = 0;
1545
1546         ENTRY;
1547
1548         LASSERT(env);
1549         o = res->lgh_obj;
1550         LASSERT(o);
1551
1552         /* llog can be already created */
1553         if (dt_object_exists(o))
1554                 RETURN(-EEXIST);
1555
1556         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1557                 struct llog_thread_info *lgi = llog_info(env);
1558
1559                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1560                 lgi->lgi_attr.la_size = 0;
1561                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1562                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1563
1564                 dt_write_lock(env, o, 0);
1565                 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1566                                &lgi->lgi_dof, th);
1567                 dt_write_unlock(env, o);
1568                 if (rc < 0)
1569                         RETURN(rc);
1570
1571                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1572
1573                 RETURN(rc);
1574         }
1575
1576         los = res->private_data;
1577         LASSERT(los);
1578
1579         dt_write_lock(env, o, 0);
1580         if (!dt_object_exists(o))
1581                 rc = llog_osd_create_new_object(env, los, o, th);
1582         else
1583                 rc = -EEXIST;
1584
1585         dt_write_unlock(env, o);
1586         if (rc)
1587                 RETURN(rc);
1588
1589         if (res->lgh_name) {
1590                 struct dt_object *llog_dir;
1591
1592                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1593                 if (IS_ERR(llog_dir))
1594                         RETURN(PTR_ERR(llog_dir));
1595
1596                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1597                 rec->rec_fid = &lgi->lgi_fid;
1598                 rec->rec_type = S_IFREG;
1599                 dt_read_lock(env, llog_dir, 0);
1600                 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1601                                (struct dt_key *)res->lgh_name, th);
1602                 dt_read_unlock(env, llog_dir);
1603                 dt_object_put(env, llog_dir);
1604                 if (rc)
1605                         CERROR("%s: can't create named llog %s: rc = %d\n",
1606                                o->do_lu.lo_dev->ld_obd->obd_name,
1607                                res->lgh_name, rc);
1608         }
1609         RETURN(rc);
1610 }
1611
1612 /**
1613  * Implementation of the llog_operations::lop_close
1614  *
1615  * This function closes the llog. It just put llog object and referenced
1616  * local storage.
1617  *
1618  * \param[in] env       execution environment
1619  * \param[in] handle    llog handle of the current llog
1620  *
1621  * \retval              0 on successful llog close
1622  * \retval              negative value on error
1623  */
1624 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1625 {
1626         struct local_oid_storage        *los;
1627         int                              rc = 0;
1628
1629         ENTRY;
1630
1631         LASSERT(handle->lgh_obj);
1632
1633         if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1634                 /* Remove the object from the cache, otherwise it may
1635                  * hold LOD being released during cleanup process */
1636                 dt_object_put_nocache(env, handle->lgh_obj);
1637                 LASSERT(handle->private_data == NULL);
1638                 RETURN(rc);
1639         } else {
1640                 dt_object_put(env, handle->lgh_obj);
1641         }
1642         los = handle->private_data;
1643         LASSERT(los);
1644         dt_los_put(los);
1645
1646         if (handle->lgh_name)
1647                 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1648
1649         RETURN(rc);
1650 }
1651
1652 /**
1653  * delete llog object name entry
1654  *
1655  * Delete llog object (with regular FID) from name space (under
1656  * update_log_dir).
1657  *
1658  * \param [in] env      execution environment
1659  * \param [in] dto      llog object
1660  * \param [in] th       thandle
1661  * \param [in] declare  if it is declare or execution
1662  *
1663  * \retval              0 if deletion succeeds.
1664  * \retval              negative errno if deletion fails.
1665  */
1666 static int
1667 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1668                                     struct dt_object *dto,
1669                                     struct thandle *th, bool declare)
1670 {
1671         struct llog_thread_info *lgi = llog_info(env);
1672         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1673         struct dt_object        *dir;
1674         char                    *name = lgi->lgi_name;
1675         int                     rc;
1676         ENTRY;
1677
1678         if (!fid_is_norm(fid))
1679                 RETURN(0);
1680
1681         dir = llog_osd_get_regular_fid_dir(env, dto);
1682         if (IS_ERR(dir))
1683                 RETURN(PTR_ERR(dir));
1684
1685         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1686         dt_write_lock(env, dir, 0);
1687         if (declare) {
1688                 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1689                                        th);
1690         } else {
1691                 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1692         }
1693         dt_write_unlock(env, dir);
1694
1695         dt_object_put(env, dir);
1696         RETURN(rc);
1697 }
1698
1699 /**
1700  * Implementation of the llog_operations::lop_declare_destroy
1701  *
1702  * This function declare destroys the llog and deletes also entry in the
1703  * llog directory in case of named llog. Llog should be opened prior that.
1704  *
1705  * \param[in] env               execution environment
1706  * \param[in] loghandle llog handle of the current llog
1707  *
1708  * \retval              0 on successful destroy
1709  * \retval              negative value on error
1710  */
1711 static int llog_osd_declare_destroy(const struct lu_env *env,
1712                                     struct llog_handle *loghandle,
1713                                     struct thandle *th)
1714 {
1715         struct llog_ctxt        *ctxt;
1716         struct dt_object        *o, *llog_dir = NULL;
1717         int                      rc;
1718
1719         ENTRY;
1720
1721         ctxt = loghandle->lgh_ctxt;
1722         LASSERT(ctxt);
1723
1724         o = loghandle->lgh_obj;
1725         LASSERT(o);
1726
1727         if (loghandle->lgh_name) {
1728                 llog_dir = llog_osd_dir_get(env, ctxt);
1729                 if (IS_ERR(llog_dir))
1730                         RETURN(PTR_ERR(llog_dir));
1731
1732                 rc = dt_declare_delete(env, llog_dir,
1733                                        (struct dt_key *)loghandle->lgh_name,
1734                                        th);
1735                 if (rc < 0)
1736                         GOTO(out_put, rc);
1737         }
1738
1739         rc = dt_declare_ref_del(env, o, th);
1740         if (rc < 0)
1741                 GOTO(out_put, rc);
1742
1743         rc = dt_declare_destroy(env, o, th);
1744         if (rc < 0)
1745                 GOTO(out_put, rc);
1746
1747         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1748                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1749                 if (rc < 0)
1750                         GOTO(out_put, rc);
1751         }
1752
1753 out_put:
1754         if (!(IS_ERR_OR_NULL(llog_dir)))
1755                 dt_object_put(env, llog_dir);
1756
1757         RETURN(rc);
1758 }
1759
1760
1761 /**
1762  * Implementation of the llog_operations::lop_destroy
1763  *
1764  * This function destroys the llog and deletes also entry in the
1765  * llog directory in case of named llog. Llog should be opened prior that.
1766  * Destroy method is not part of external transaction and does everything
1767  * inside.
1768  *
1769  * \param[in] env               execution environment
1770  * \param[in] loghandle llog handle of the current llog
1771  *
1772  * \retval              0 on successful destroy
1773  * \retval              negative value on error
1774  */
1775 static int llog_osd_destroy(const struct lu_env *env,
1776                             struct llog_handle *loghandle, struct thandle *th)
1777 {
1778         struct llog_ctxt        *ctxt;
1779         struct dt_object        *o, *llog_dir = NULL;
1780         int                      rc;
1781
1782         ENTRY;
1783
1784         ctxt = loghandle->lgh_ctxt;
1785         LASSERT(ctxt != NULL);
1786
1787         o = loghandle->lgh_obj;
1788         LASSERT(o != NULL);
1789
1790         dt_write_lock(env, o, 0);
1791         if (!dt_object_exists(o))
1792                 GOTO(out_unlock, rc = 0);
1793
1794         if (loghandle->lgh_name) {
1795                 llog_dir = llog_osd_dir_get(env, ctxt);
1796                 if (IS_ERR(llog_dir))
1797                         GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1798
1799                 dt_read_lock(env, llog_dir, 0);
1800                 rc = dt_delete(env, llog_dir,
1801                                (struct dt_key *)loghandle->lgh_name,
1802                                th);
1803                 dt_read_unlock(env, llog_dir);
1804                 if (rc) {
1805                         CERROR("%s: can't remove llog %s: rc = %d\n",
1806                                o->do_lu.lo_dev->ld_obd->obd_name,
1807                                loghandle->lgh_name, rc);
1808                         GOTO(out_unlock, rc);
1809                 }
1810         }
1811
1812         dt_ref_del(env, o, th);
1813         rc = dt_destroy(env, o, th);
1814         if (rc < 0)
1815                 GOTO(out_unlock, rc);
1816
1817         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1818                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1819                 if (rc < 0)
1820                         GOTO(out_unlock, rc);
1821         }
1822
1823 out_unlock:
1824         dt_write_unlock(env, o);
1825         if (!(IS_ERR_OR_NULL(llog_dir)))
1826                 dt_object_put(env, llog_dir);
1827         RETURN(rc);
1828 }
1829
1830 /**
1831  * Implementation of the llog_operations::lop_setup
1832  *
1833  * This function setup the llog on local storage.
1834  *
1835  * \param[in] env       execution environment
1836  * \param[in] obd       obd device the llog belongs to
1837  * \param[in] olg       the llog group, it is always zero group now.
1838  * \param[in] ctxt_idx  the llog index, it defines the purpose of this llog.
1839  *                      Every new llog type have to use own index.
1840  * \param[in] disk_obd  the storage obd, where llog is stored.
1841  *
1842  * \retval              0 on successful llog setup
1843  * \retval              negative value on error
1844  */
1845 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1846                           struct obd_llog_group *olg, int ctxt_idx,
1847                           struct obd_device *disk_obd)
1848 {
1849         struct llog_thread_info         *lgi = llog_info(env);
1850         struct llog_ctxt                *ctxt;
1851         int                              rc = 0;
1852         ENTRY;
1853
1854         LASSERT(obd);
1855         LASSERT(olg->olg_ctxts[ctxt_idx]);
1856
1857         ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1858         LASSERT(ctxt);
1859
1860         if (disk_obd == NULL)
1861                 GOTO(out, rc = 0);
1862
1863         /* initialize data allowing to generate new fids,
1864          * literally we need a sequece */
1865         lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1866         lgi->lgi_fid.f_oid = 1;
1867         lgi->lgi_fid.f_ver = 0;
1868         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1869                                     &lgi->lgi_fid,
1870                                     &ctxt->loc_los_nameless);
1871         if (rc != 0)
1872                 GOTO(out, rc);
1873
1874         lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1875         lgi->lgi_fid.f_oid = 1;
1876         lgi->lgi_fid.f_ver = 0;
1877         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1878                                     &lgi->lgi_fid,
1879                                     &ctxt->loc_los_named);
1880         if (rc != 0) {
1881                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1882                 ctxt->loc_los_nameless = NULL;
1883         }
1884
1885         GOTO(out, rc);
1886
1887 out:
1888         llog_ctxt_put(ctxt);
1889         return rc;
1890 }
1891
1892 /**
1893  * Implementation of the llog_operations::lop_cleanup
1894  *
1895  * This function cleanups the llog on local storage.
1896  *
1897  * \param[in] env       execution environment
1898  * \param[in] ctxt      the llog context
1899  *
1900  * \retval              0
1901  */
1902 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1903 {
1904         if (ctxt->loc_los_nameless != NULL) {
1905                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1906                 ctxt->loc_los_nameless = NULL;
1907         }
1908
1909         if (ctxt->loc_los_named != NULL) {
1910                 local_oid_storage_fini(env, ctxt->loc_los_named);
1911                 ctxt->loc_los_named = NULL;
1912         }
1913
1914         return 0;
1915 }
1916
1917 struct llog_operations llog_osd_ops = {
1918         .lop_next_block         = llog_osd_next_block,
1919         .lop_prev_block         = llog_osd_prev_block,
1920         .lop_read_header        = llog_osd_read_header,
1921         .lop_declare_destroy    = llog_osd_declare_destroy,
1922         .lop_destroy            = llog_osd_destroy,
1923         .lop_setup              = llog_osd_setup,
1924         .lop_cleanup            = llog_osd_cleanup,
1925         .lop_open               = llog_osd_open,
1926         .lop_exist              = llog_osd_exist,
1927         .lop_declare_create     = llog_osd_declare_create,
1928         .lop_create             = llog_osd_create,
1929         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1930         .lop_write_rec          = llog_osd_write_rec,
1931         .lop_close              = llog_osd_close,
1932 };
1933 EXPORT_SYMBOL(llog_osd_ops);
1934
1935 struct llog_operations llog_common_cat_ops = {
1936         .lop_next_block         = llog_osd_next_block,
1937         .lop_prev_block         = llog_osd_prev_block,
1938         .lop_read_header        = llog_osd_read_header,
1939         .lop_declare_destroy    = llog_osd_declare_destroy,
1940         .lop_destroy            = llog_osd_destroy,
1941         .lop_setup              = llog_osd_setup,
1942         .lop_cleanup            = llog_osd_cleanup,
1943         .lop_open               = llog_osd_open,
1944         .lop_exist              = llog_osd_exist,
1945         .lop_declare_create     = llog_osd_declare_create,
1946         .lop_create             = llog_osd_create,
1947         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1948         .lop_write_rec          = llog_osd_write_rec,
1949         .lop_close              = llog_osd_close,
1950         .lop_add                = llog_cat_add_rec,
1951         .lop_declare_add        = llog_cat_declare_add_rec,
1952 };
1953 EXPORT_SYMBOL(llog_common_cat_ops);
1954
1955 /**
1956  * Read the special file which contains the list of llog catalogs IDs
1957  *
1958  * This function reads the CATALOGS file which contains the array of llog
1959  * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1960  * by OST/MDT number.
1961  *
1962  * \param[in]  env              execution environment
1963  * \param[in]  d                corresponding storage device
1964  * \param[in]  idx              position to start from, usually OST/MDT index
1965  * \param[in]  count            how many catalog IDs to read
1966  * \param[out] idarray          the buffer for the data. If it is NULL then
1967  *                              function returns just number of catalog IDs
1968  *                              in the file.
1969  * \param[in]  fid              LLOG_CATALOGS_OID for CATALOG object
1970  *
1971  * \retval                      0 on successful read of catalog IDs
1972  * \retval                      negative value on error
1973  * \retval                      positive value which is number of records in
1974  *                              the file if \a idarray is NULL
1975  */
1976 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1977                           int idx, int count, struct llog_catid *idarray,
1978                           const struct lu_fid *fid)
1979 {
1980         struct llog_thread_info *lgi = llog_info(env);
1981         struct dt_object        *o = NULL;
1982         struct thandle          *th;
1983         int                      rc, size;
1984
1985         ENTRY;
1986
1987         LASSERT(d);
1988
1989         size = sizeof(*idarray) * count;
1990         lgi->lgi_off = idx *  sizeof(*idarray);
1991
1992         lgi->lgi_fid = *fid;
1993         o = dt_locate(env, d, &lgi->lgi_fid);
1994         if (IS_ERR(o))
1995                 RETURN(PTR_ERR(o));
1996
1997         if (!dt_object_exists(o)) {
1998                 th = dt_trans_create(env, d);
1999                 if (IS_ERR(th))
2000                         GOTO(out, rc = PTR_ERR(th));
2001
2002                 lgi->lgi_attr.la_valid = LA_MODE;
2003                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
2004                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
2005
2006                 th->th_wait_submit = 1;
2007                 /* Make the llog object creation synchronization, so
2008                  * it will be reliable to the reference, especially
2009                  * for remote reference */
2010                 th->th_sync = 1;
2011
2012                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
2013                                        &lgi->lgi_dof, th);
2014                 if (rc)
2015                         GOTO(out_trans, rc);
2016
2017                 rc = dt_trans_start_local(env, d, th);
2018                 if (rc)
2019                         GOTO(out_trans, rc);
2020
2021                 dt_write_lock(env, o, 0);
2022                 if (!dt_object_exists(o))
2023                         rc = dt_create(env, o, &lgi->lgi_attr, NULL,
2024                                        &lgi->lgi_dof, th);
2025                 dt_write_unlock(env, o);
2026 out_trans:
2027                 dt_trans_stop(env, d, th);
2028                 if (rc)
2029                         GOTO(out, rc);
2030         }
2031
2032         rc = dt_attr_get(env, o, &lgi->lgi_attr);
2033         if (rc)
2034                 GOTO(out, rc);
2035
2036         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2037                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2038                        o->do_lu.lo_dev->ld_obd->obd_name,
2039                        lgi->lgi_attr.la_mode);
2040                 GOTO(out, rc = -ENOENT);
2041         }
2042
2043         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2044                (int)lgi->lgi_attr.la_size, size);
2045
2046         /* return just number of llogs */
2047         if (idarray == NULL) {
2048                 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2049                 GOTO(out, rc);
2050         }
2051
2052         /* read for new ost index or for empty file */
2053         memset(idarray, 0, size);
2054         if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2055                 GOTO(out, rc = 0);
2056         if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2057                 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2058
2059         lgi->lgi_buf.lb_buf = idarray;
2060         lgi->lgi_buf.lb_len = size;
2061         rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2062         /* -EFAULT means the llog is a sparse file. This is not an error
2063          * after arbitrary OST index is supported. */
2064         if (rc < 0 && rc != -EFAULT) {
2065                 CERROR("%s: error reading CATALOGS: rc = %d\n",
2066                        o->do_lu.lo_dev->ld_obd->obd_name,  rc);
2067                 GOTO(out, rc);
2068         }
2069
2070         EXIT;
2071 out:
2072         dt_object_put(env, o);
2073         RETURN(rc);
2074 }
2075 EXPORT_SYMBOL(llog_osd_get_cat_list);
2076
2077 /**
2078  * Write the special file which contains the list of llog catalogs IDs
2079  *
2080  * This function writes the CATALOG file which contains the array of llog
2081  * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2082  * number.
2083  *
2084  * \param[in]  env      execution environment
2085  * \param[in]  d        corresponding storage device
2086  * \param[in]  idx      position to start from, usually OST/MDT index
2087  * \param[in]  count    how many catalog IDs to write
2088  * \param[out] idarray  the buffer with the data to write.
2089  * \param[in]  fid      LLOG_CATALOGS_OID for CATALOG object
2090  *
2091  * \retval              0 on successful write of catalog IDs
2092  * \retval              negative value on error
2093  */
2094 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2095                           int idx, int count, struct llog_catid *idarray,
2096                           const struct lu_fid *fid)
2097 {
2098         struct llog_thread_info *lgi = llog_info(env);
2099         struct dt_object        *o = NULL;
2100         struct thandle          *th;
2101         int                      rc, size;
2102
2103         if (count == 0)
2104                 RETURN(0);
2105
2106         LASSERT(d);
2107
2108         size = sizeof(*idarray) * count;
2109         lgi->lgi_off = idx * sizeof(*idarray);
2110         lgi->lgi_fid = *fid;
2111
2112         o = dt_locate(env, d, &lgi->lgi_fid);
2113         if (IS_ERR(o))
2114                 RETURN(PTR_ERR(o));
2115
2116         if (!dt_object_exists(o))
2117                 GOTO(out, rc = -ENOENT);
2118
2119         rc = dt_attr_get(env, o, &lgi->lgi_attr);
2120         if (rc)
2121                 GOTO(out, rc);
2122
2123         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2124                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2125                        o->do_lu.lo_dev->ld_obd->obd_name,
2126                        lgi->lgi_attr.la_mode);
2127                 GOTO(out, rc = -ENOENT);
2128         }
2129
2130         th = dt_trans_create(env, d);
2131         if (IS_ERR(th))
2132                 GOTO(out, rc = PTR_ERR(th));
2133
2134         lgi->lgi_buf.lb_len = size;
2135         lgi->lgi_buf.lb_buf = idarray;
2136         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2137         if (rc)
2138                 GOTO(out_trans, rc);
2139
2140         /* For update log, this happens during initialization,
2141          * see lod_sub_prep_llog(), and we need make sure catlog
2142          * file ID is written to catlist file(committed) before
2143          * cross-MDT operation write update records to catlog FILE,
2144          * otherwise, during failover these update records might
2145          * missing */
2146         if (fid_is_update_log(fid))
2147                 th->th_sync = 1;
2148
2149         rc = dt_trans_start_local(env, d, th);
2150         if (rc)
2151                 GOTO(out_trans, rc);
2152
2153         th->th_wait_submit = 1;
2154
2155         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2156         if (rc)
2157                 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2158                        idx, rc);
2159 out_trans:
2160         dt_trans_stop(env, d, th);
2161 out:
2162         dt_object_put(env, o);
2163         RETURN(rc);
2164 }
2165 EXPORT_SYMBOL(llog_osd_put_cat_list);