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, 2014 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 <obd.h>
  48 #include <obd_class.h>
  49 #include <lustre_fid.h>
  50 #include <dt_object.h>
  51
  52 #include "llog_internal.h"
  53 #include "local_storage.h"
  54
  55 /**
  56  * Implementation of the llog_operations::lop_declare_create
  57  *
  58  * This function is a wrapper over local_storage API function
  59  * local_object_declare_create().
  60  *
  61  * \param[in] env       execution environment
  62  * \param[in] los       local_storage for bottom storage device
  63  * \param[in] o         dt_object to create
  64  * \param[in] th        current transaction handle
  65  *
  66  * \retval              0 on successful declaration of the new object
  67  * \retval              negative error if declaration was failed
  68  */
  69 static int llog_osd_declare_new_object(const struct lu_env *env,
  70                                        struct local_oid_storage *los,
  71                                        struct dt_object *o,
  72                                        struct thandle *th)
  73 {
  74         struct llog_thread_info *lgi = llog_info(env);
  75
  76         lgi->lgi_attr.la_valid = LA_MODE;
  77         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
  78         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
  79
  80         return local_object_declare_create(env, los, o, &lgi->lgi_attr,
  81                                            &lgi->lgi_dof, th);
  82 }
  83
  84 /**
  85  * Implementation of the llog_operations::lop_create
  86  *
  87  * This function is a wrapper over local_storage API function
  88  * local_object_create().
  89  *
  90  * \param[in] env       execution environment
  91  * \param[in] los       local_storage for bottom storage device
  92  * \param[in] o         dt_object to create
  93  * \param[in] th        current transaction handle
  94  *
  95  * \retval              0 on successful creation of the new object
  96  * \retval              negative error if creation was failed
  97  */
  98 static int llog_osd_create_new_object(const struct lu_env *env,
  99                                       struct local_oid_storage *los,
 100                                       struct dt_object *o,
 101                                       struct thandle *th)
 102 {
 103         struct llog_thread_info *lgi = llog_info(env);
 104
 105         lgi->lgi_attr.la_valid = LA_MODE;
 106         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
 107         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
 108
 109         return local_object_create(env, los, o, &lgi->lgi_attr,
 110                                    &lgi->lgi_dof, th);
 111 }
 112
 113 /**
 114  * Write a padding record to the llog
 115  *
 116  * This function writes a padding record to the end of llog. That may
 117  * be needed if llog contains records of variable size, e.g. config logs
 118  * or changelogs.
 119  * The padding record just aligns llog to the llog chunk_size boundary if
 120  * the current record doesn't fit in the remaining space.
 121  *
 122  * It allocates full length to avoid two separate writes for header and tail.
 123  * Such 2-steps scheme needs extra protection and complex error handling.
 124  *
 125  * \param[in]     env   execution environment
 126  * \param[in]     o     dt_object to create
 127  * \param[in,out] off   pointer to the padding start offset
 128  * \param[in]     len   padding length
 129  * \param[in]     index index of the padding record in a llog
 130  * \param[in]     th    current transaction handle
 131  *
 132  * \retval              0 on successful padding write
 133  * \retval              negative error if write failed
 134  */
 135 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
 136                         loff_t *off, int len, int index, struct thandle *th)
 137 {
 138         struct llog_thread_info *lgi = llog_info(env);
 139         struct llog_rec_hdr     *rec;
 140         struct llog_rec_tail    *tail;
 141         int                      rc;
 142
 143         ENTRY;
 144
 145         LASSERT(th);
 146         LASSERT(off);
 147         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
 148
 149         OBD_ALLOC(rec, len);
 150         if (rec == NULL)
 151                 RETURN(-ENOMEM);
 152
 153         rec->lrh_len = len;
 154         rec->lrh_index = index;
 155         rec->lrh_type = LLOG_PAD_MAGIC;
 156
 157         tail = rec_tail(rec);
 158         tail->lrt_len = len;
 159         tail->lrt_index = index;
 160
 161         lgi->lgi_buf.lb_buf = rec;
 162         lgi->lgi_buf.lb_len = len;
 163         rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
 164         if (rc)
 165                 CERROR("%s: error writing padding record: rc = %d\n",
 166                        o->do_lu.lo_dev->ld_obd->obd_name, rc);
 167
 168         OBD_FREE(rec, len);
 169         RETURN(rc);
 170 }
 171
 172 /**
 173  * Implementation of the llog_operations::lop_read_header
 174  *
 175  * This function reads the current llog header from the bottom storage
 176  * device.
 177  *
 178  * \param[in] env       execution environment
 179  * \param[in] handle    llog handle of the current llog
 180  *
 181  * \retval              0 on successful header read
 182  * \retval              negative error if read failed
 183  */
 184 static int llog_osd_read_header(const struct lu_env *env,
 185                                 struct llog_handle *handle)
 186 {
 187         struct llog_rec_hdr     *llh_hdr;
 188         struct dt_object        *o;
 189         struct llog_thread_info *lgi;
 190         enum llog_flag           flags;
 191         int                      rc;
 192
 193         ENTRY;
 194
 195         o = handle->lgh_obj;
 196         LASSERT(o);
 197
 198         lgi = llog_info(env);
 199
 200         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 201         if (rc)
 202                 RETURN(rc);
 203
 204         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 205
 206         if (lgi->lgi_attr.la_size == 0) {
 207                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 208                 RETURN(LLOG_EEMPTY);
 209         }
 210
 211         flags = handle->lgh_hdr->llh_flags;
 212
 213         lgi->lgi_off = 0;
 214         lgi->lgi_buf.lb_buf = handle->lgh_hdr;
 215         lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
 216         rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
 217         llh_hdr = &handle->lgh_hdr->llh_hdr;
 218         if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
 219                 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
 220                        o->do_lu.lo_dev->ld_obd->obd_name,
 221                        PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
 222                        -EFAULT);
 223
 224                 if (rc >= 0)
 225                         rc = -EFAULT;
 226
 227                 RETURN(rc);
 228         }
 229
 230         if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
 231                 lustre_swab_llog_hdr(handle->lgh_hdr);
 232
 233         if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
 234                 CERROR("%s: bad log %s "DFID" header magic: %#x "
 235                        "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
 236                        handle->lgh_name ? handle->lgh_name : "",
 237                        PFID(lu_object_fid(&o->do_lu)),
 238                        llh_hdr->lrh_type, LLOG_HDR_MAGIC);
 239                 RETURN(-EIO);
 240         } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
 241                    llh_hdr->lrh_len > handle->lgh_hdr_size) {
 242                 CERROR("%s: incorrectly sized log %s "DFID" header: "
 243                        "%#x (expected at least %#x)\n"
 244                        "you may need to re-run lconf --write_conf.\n",
 245                        o->do_lu.lo_dev->ld_obd->obd_name,
 246                        handle->lgh_name ? handle->lgh_name : "",
 247                        PFID(lu_object_fid(&o->do_lu)),
 248                        llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
 249                 RETURN(-EIO);
 250         } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
 251                    LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
 252                    LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
 253                         llh_hdr->lrh_len) {
 254                 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
 255                        "%#x : rc = %d\n",
 256                        o->do_lu.lo_dev->ld_obd->obd_name,
 257                        handle->lgh_name ? handle->lgh_name : "",
 258                        PFID(lu_object_fid(&o->do_lu)),
 259                        LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
 260                 RETURN(-EIO);
 261         }
 262
 263         handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
 264         handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
 265
 266         RETURN(0);
 267 }
 268
 269 /**
 270  * Implementation of the llog_operations::lop_declare_write
 271  *
 272  * This function declares the new record write.
 273  *
 274  * \param[in] env       execution environment
 275  * \param[in] loghandle llog handle of the current llog
 276  * \param[in] rec       llog record header. This is a real header of the full
 277  *                      llog record to write. This is the beginning of buffer
 278  *                      to write, the length of buffer is stored in
 279  *                      \a rec::lrh_len
 280  * \param[in] idx       index of the llog record. If \a idx == -1 then this is
 281  *                      append case, otherwise \a idx is the index of record
 282  *                      to modify
 283  * \param[in] th        current transaction handle
 284  *
 285  * \retval              0 on successful declaration
 286  * \retval              negative error if declaration failed
 287  */
 288 static int llog_osd_declare_write_rec(const struct lu_env *env,
 289                                       struct llog_handle *loghandle,
 290                                       struct llog_rec_hdr *rec,
 291                                       int idx, struct thandle *th)
 292 {
 293         struct llog_thread_info *lgi = llog_info(env);
 294         __u32                   chunk_size;
 295         struct dt_object        *o;
 296         int                      rc;
 297
 298         ENTRY;
 299
 300         LASSERT(env);
 301         LASSERT(th);
 302         LASSERT(loghandle);
 303         LASSERT(rec);
 304         LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
 305
 306         o = loghandle->lgh_obj;
 307         LASSERT(o);
 308
 309         chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
 310         lgi->lgi_buf.lb_len = chunk_size;
 311         lgi->lgi_buf.lb_buf = NULL;
 312         /* each time we update header */
 313         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
 314                                      th);
 315         if (rc || idx == 0) /* if error or just header */
 316                 RETURN(rc);
 317
 318         /**
 319          * the pad record can be inserted so take into account double
 320          * record size
 321          */
 322         lgi->lgi_buf.lb_len = chunk_size * 2;
 323         lgi->lgi_buf.lb_buf = NULL;
 324         /* XXX: implement declared window or multi-chunks approach */
 325         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
 326
 327         RETURN(rc);
 328 }
 329
 330 /**
 331  * Implementation of the llog_operations::lop_write
 332  *
 333  * This function writes the new record in the llog or modify the existed one.
 334  *
 335  * \param[in]  env              execution environment
 336  * \param[in]  loghandle        llog handle of the current llog
 337  * \param[in]  rec              llog record header. This is a real header of
 338  *                              the full llog record to write. This is
 339  *                              the beginning of buffer to write, the length
 340  *                              of buffer is stored in \a rec::lrh_len
 341  * \param[out] reccookie        pointer to the cookie to return back if needed.
 342  *                              It is used for further cancel of this llog
 343  *                              record.
 344  * \param[in]  idx              index of the llog record. If \a idx == -1 then
 345  *                              this is append case, otherwise \a idx is
 346  *                              the index of record to modify
 347  * \param[in]  th               current transaction handle
 348  *
 349  * \retval                      0 on successful write && \a reccookie == NULL
 350  *                              1 on successful write && \a reccookie != NULL
 351  * \retval                      negative error if write failed
 352  */
 353 static int llog_osd_write_rec(const struct lu_env *env,
 354                               struct llog_handle *loghandle,
 355                               struct llog_rec_hdr *rec,
 356                               struct llog_cookie *reccookie,
 357                               int idx, struct thandle *th)
 358 {
 359         struct llog_thread_info *lgi = llog_info(env);
 360         struct llog_log_hdr     *llh;
 361         int                      reclen = rec->lrh_len;
 362         int                      index, rc;
 363         struct llog_rec_tail    *lrt;
 364         struct dt_object        *o;
 365         __u32                   chunk_size;
 366         size_t                   left;
 367
 368         ENTRY;
 369
 370         LASSERT(env);
 371         llh = loghandle->lgh_hdr;
 372         LASSERT(llh);
 373         o = loghandle->lgh_obj;
 374         LASSERT(o);
 375         LASSERT(th);
 376
 377         chunk_size = llh->llh_hdr.lrh_len;
 378         CDEBUG(D_OTHER, "new record %x to "DFID"\n",
 379                rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
 380
 381         /* record length should not bigger than  */
 382         if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
 383                 RETURN(-E2BIG);
 384
 385         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 386         if (rc)
 387                 RETURN(rc);
 388
 389         /**
 390          * The modification case.
 391          * If idx set then the record with that index must be modified.
 392          * There are three cases possible:
 393          * 1) the common case is the llog header update (idx == 0)
 394          * 2) the llog record modification during llog process.
 395          *    This is indicated by the \a loghandle::lgh_cur_idx > 0.
 396          *    In that case the \a loghandle::lgh_cur_offset
 397          * 3) otherwise this is assumed that llog consist of records of
 398          *    fixed size, i.e. catalog. The llog header must has llh_size
 399          *    field equal to record size. The record offset is calculated
 400          *    just by /a idx value
 401          *
 402          * During modification we don't need extra header update because
 403          * the bitmap and record count are not changed. The record header
 404          * and tail remains the same too.
 405          */
 406         if (idx != LLOG_NEXT_IDX) {
 407                 /* llog can be empty only when first record is being written */
 408                 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
 409
 410                 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
 411                         CERROR("%s: modify unset record %u\n",
 412                                o->do_lu.lo_dev->ld_obd->obd_name, idx);
 413                         RETURN(-ENOENT);
 414                 }
 415
 416                 if (idx != rec->lrh_index) {
 417                         CERROR("%s: modify index mismatch %d %u\n",
 418                                o->do_lu.lo_dev->ld_obd->obd_name, idx,
 419                                rec->lrh_index);
 420                         RETURN(-EFAULT);
 421                 }
 422
 423                 if (idx == LLOG_HEADER_IDX) {
 424                         /* llog header update */
 425                         LASSERT(reclen >= sizeof(struct llog_log_hdr));
 426                         LASSERT(rec == &llh->llh_hdr);
 427
 428                         lgi->lgi_off = 0;
 429                         lgi->lgi_buf.lb_len = reclen;
 430                         lgi->lgi_buf.lb_buf = rec;
 431                         rc = dt_record_write(env, o, &lgi->lgi_buf,
 432                                              &lgi->lgi_off, th);
 433                         RETURN(rc);
 434                 } else if (loghandle->lgh_cur_idx > 0) {
 435                         /**
 436                          * The lgh_cur_offset can be used only if index is
 437                          * the same.
 438                          */
 439                         if (idx != loghandle->lgh_cur_idx) {
 440                                 CERROR("%s: modify index mismatch %d %d\n",
 441                                        o->do_lu.lo_dev->ld_obd->obd_name, idx,
 442                                        loghandle->lgh_cur_idx);
 443                                 RETURN(-EFAULT);
 444                         }
 445
 446                         lgi->lgi_off = loghandle->lgh_cur_offset;
 447                         CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
 448                                "len:%u offset %llu\n",
 449                                POSTID(&loghandle->lgh_id.lgl_oi), idx,
 450                                rec->lrh_len, (long long)lgi->lgi_off);
 451                 } else if (llh->llh_size > 0) {
 452                         if (llh->llh_size != rec->lrh_len) {
 453                                 CERROR("%s: wrong record size, llh_size is %u"
 454                                        " but record size is %u\n",
 455                                        o->do_lu.lo_dev->ld_obd->obd_name,
 456                                        llh->llh_size, rec->lrh_len);
 457                                 RETURN(-EINVAL);
 458                         }
 459                         lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
 460                 } else {
 461                         /* This can be result of lgh_cur_idx is not set during
 462                          * llog processing or llh_size is not set to proper
 463                          * record size for fixed records llog. Therefore it is
 464                          * impossible to get record offset. */
 465                         CERROR("%s: can't get record offset, idx:%d, "
 466                                "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
 467                                idx, rec->lrh_len);
 468                         RETURN(-EFAULT);
 469                 }
 470
 471                 /* update only data, header and tail remain the same */
 472                 lgi->lgi_off += sizeof(struct llog_rec_hdr);
 473                 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
 474                 lgi->lgi_buf.lb_buf = REC_DATA(rec);
 475                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 476                 if (rc == 0 && reccookie) {
 477                         reccookie->lgc_lgl = loghandle->lgh_id;
 478                         reccookie->lgc_index = idx;
 479                         rc = 1;
 480                 }
 481                 RETURN(rc);
 482         }
 483
 484         /**
 485          * The append case.
 486          * The most common case of using llog. The new index is assigned to
 487          * the new record, new bit is set in llog bitmap and llog count is
 488          * incremented.
 489          *
 490          * Make sure that records don't cross a chunk boundary, so we can
 491          * process them page-at-a-time if needed.  If it will cross a chunk
 492          * boundary, write in a fake (but referenced) entry to pad the chunk.
 493          */
 494         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 495         lgi->lgi_off = lgi->lgi_attr.la_size;
 496         left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
 497         /* NOTE: padding is a record, but no bit is set */
 498         if (left != 0 && left != reclen &&
 499             left < (reclen + LLOG_MIN_REC_SIZE)) {
 500                 index = loghandle->lgh_last_idx + 1;
 501                 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
 502                 if (rc)
 503                         RETURN(rc);
 504                 loghandle->lgh_last_idx++; /* for pad rec */
 505         }
 506         /* if it's the last idx in log file, then return -ENOSPC */
 507         if (loghandle->lgh_last_idx >= LLOG_HDR_BITMAP_SIZE(llh) - 1)
 508                 RETURN(-ENOSPC);
 509
 510         /* increment the last_idx along with llh_tail index, they should
 511          * be equal for a llog lifetime */
 512         loghandle->lgh_last_idx++;
 513         index = loghandle->lgh_last_idx;
 514         LLOG_HDR_TAIL(llh)->lrt_index = index;
 515         /**
 516          * NB: the caller should make sure only 1 process access
 517          * the lgh_last_idx, e.g. append should be exclusive.
 518          * Otherwise it might hit the assert.
 519          */
 520         LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
 521         rec->lrh_index = index;
 522         lrt = rec_tail(rec);
 523         lrt->lrt_len = rec->lrh_len;
 524         lrt->lrt_index = rec->lrh_index;
 525
 526         /* the lgh_hdr_lock protects llog header data from concurrent
 527          * update/cancel, the llh_count and llh_bitmap are protected */
 528         down_write(&loghandle->lgh_hdr_lock);
 529         if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
 530                 CERROR("%s: index %u already set in log bitmap\n",
 531                        o->do_lu.lo_dev->ld_obd->obd_name, index);
 532                 up_write(&loghandle->lgh_hdr_lock);
 533                 LBUG(); /* should never happen */
 534         }
 535         llh->llh_count++;
 536
 537         /* XXX It is a bit tricky here, if the log object is local,
 538          * we do not need lock during write here, because if there is
 539          * race, the transaction(jbd2, what about ZFS?) will make sure the
 540          * conflicts will all committed in the same transaction group.
 541          * But for remote object, we need lock the whole process, so to
 542          * set the version of the remote transaction to make sure they
 543          * are being sent in order. (see osp_md_write()) */
 544         if (!dt_object_remote(o))
 545                 up_write(&loghandle->lgh_hdr_lock);
 546
 547         if (lgi->lgi_attr.la_size == 0) {
 548                 lgi->lgi_off = 0;
 549                 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
 550                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 551                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 552                 if (rc != 0)
 553                         GOTO(out_remote_unlock, rc);
 554         } else {
 555                 /* Note: If this is not initialization (size == 0), then do not
 556                  * write the whole header (8k bytes), only update header/tail
 557                  * and bits needs to be updated. Because this update might be
 558                  * part of cross-MDT operation, which needs to write these
 559                  * updates into the update log(32KB limit) and also pack inside
 560                  * the RPC (1MB limit), if we write 8K for each operation, which
 561                  * will cost a lot space, and keep us adding more updates to one
 562                  * update log.*/
 563                 lgi->lgi_off = offsetof(typeof(*llh), llh_count);
 564                 lgi->lgi_buf.lb_len = sizeof(llh->llh_count);
 565                 lgi->lgi_buf.lb_buf = &llh->llh_count;
 566                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 567                 if (rc != 0)
 568                         GOTO(out_remote_unlock, rc);
 569
 570                 lgi->lgi_off = offsetof(typeof(*llh),
 571                         llh_bitmap[index / (sizeof(*llh->llh_bitmap) * 8)]);
 572                 lgi->lgi_buf.lb_len = sizeof(*llh->llh_bitmap);
 573                 lgi->lgi_buf.lb_buf =
 574                         &llh->llh_bitmap[index/(sizeof(*llh->llh_bitmap)*8)];
 575                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 576                 if (rc != 0)
 577                         GOTO(out_remote_unlock, rc);
 578
 579                 lgi->lgi_off =  (unsigned long)LLOG_HDR_TAIL(llh) -
 580                                 (unsigned long)llh;
 581                 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
 582                 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
 583                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 584                 if (rc != 0)
 585                         GOTO(out_remote_unlock, rc);
 586         }
 587
 588 out_remote_unlock:
 589         /* unlock here for remote object */
 590         if (dt_object_remote(o))
 591                 up_write(&loghandle->lgh_hdr_lock);
 592         if (rc)
 593                 GOTO(out, rc);
 594
 595         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 596         if (rc)
 597                 GOTO(out, rc);
 598
 599         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 600         lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size, lgi->lgi_off);
 601         lgi->lgi_buf.lb_len = reclen;
 602         lgi->lgi_buf.lb_buf = rec;
 603         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 604         if (rc < 0)
 605                 GOTO(out, rc);
 606
 607         CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u\n",
 608                POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len);
 609         if (reccookie != NULL) {
 610                 reccookie->lgc_lgl = loghandle->lgh_id;
 611                 reccookie->lgc_index = index;
 612                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 613                     (rec->lrh_type == MDS_SETATTR64_REC))
 614                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 615                 else if (rec->lrh_type == OST_SZ_REC)
 616                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 617                 else
 618                         reccookie->lgc_subsys = -1;
 619                 rc = 1;
 620         }
 621         RETURN(rc);
 622 out:
 623         /* cleanup llog for error case */
 624         down_write(&loghandle->lgh_hdr_lock);
 625         ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
 626         llh->llh_count--;
 627         up_write(&loghandle->lgh_hdr_lock);
 628
 629         /* restore llog last_idx */
 630         loghandle->lgh_last_idx--;
 631         LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
 632
 633         RETURN(rc);
 634 }
 635
 636 /**
 637  * We can skip reading at least as many log blocks as the number of
 638  * minimum sized log records we are skipping.  If it turns out
 639  * that we are not far enough along the log (because the
 640  * actual records are larger than minimum size) we just skip
 641  * some more records.
 642  */
 643 static inline void llog_skip_over(__u64 *off, int curr, int goal,
 644                                   __u32 chunk_size)
 645 {
 646         if (goal <= curr)
 647                 return;
 648         *off = (*off + (goal - curr - 1) * LLOG_MIN_REC_SIZE) &
 649                 ~(chunk_size - 1);
 650 }
 651
 652 /**
 653  * Remove optional fields that the client doesn't expect.
 654  * This is typically in order to ensure compatibility with older clients.
 655  * It is assumed that since we exclusively remove fields, the block will be
 656  * big enough to handle the remapped records. It is also assumed that records
 657  * of a block have the same format (i.e.: the same features enabled).
 658  *
 659  * \param[in,out]    hdr        Header of the block of records to remap.
 660  * \param[in,out]    last_hdr   Last header, don't read past this point.
 661  * \param[in]        flags      Flags describing the fields to keep.
 662  */
 663 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
 664                                      struct llog_rec_hdr *last_hdr,
 665                                      enum changelog_rec_flags flags)
 666 {
 667         if (hdr->lrh_type != CHANGELOG_REC)
 668                 return;
 669
 670         do {
 671                 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
 672
 673                 changelog_remap_rec(rec, rec->cr_flags & flags);
 674                 hdr = llog_rec_hdr_next(hdr);
 675         } while ((char *)hdr <= (char *)last_hdr);
 676 }
 677
 678 /**
 679  * Implementation of the llog_operations::lop_next_block
 680  *
 681  * This function finds the the next llog block to return which contains
 682  * record with required index. It is main part of llog processing.
 683  *
 684  * \param[in]     env           execution environment
 685  * \param[in]     loghandle     llog handle of the current llog
 686  * \param[in,out] cur_idx       index preceeding cur_offset
 687  * \param[in]     next_idx      target index to find
 688  * \param[in,out] cur_offset    furtherst point read in the file
 689  * \param[in]     buf           pointer to data buffer to fill
 690  * \param[in]     len           required len to read, it is
 691  *                              usually llog chunk_size.
 692  *
 693  * \retval                      0 on successful buffer read
 694  * \retval                      negative value on error
 695  */
 696 static int llog_osd_next_block(const struct lu_env *env,
 697                                struct llog_handle *loghandle, int *cur_idx,
 698                                int next_idx, __u64 *cur_offset, void *buf,
 699                                int len)
 700 {
 701         struct llog_thread_info *lgi = llog_info(env);
 702         struct dt_object        *o;
 703         struct dt_device        *dt;
 704         int                      rc;
 705         __u32                   chunk_size;
 706
 707         ENTRY;
 708
 709         LASSERT(env);
 710         LASSERT(lgi);
 711
 712         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 713         if (len == 0 || len & (chunk_size - 1))
 714                 RETURN(-EINVAL);
 715
 716         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
 717                next_idx, *cur_idx, *cur_offset);
 718
 719         LASSERT(loghandle);
 720         LASSERT(loghandle->lgh_ctxt);
 721
 722         o = loghandle->lgh_obj;
 723         LASSERT(o);
 724         LASSERT(dt_object_exists(o));
 725         dt = lu2dt_dev(o->do_lu.lo_dev);
 726         LASSERT(dt);
 727
 728         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 729         if (rc)
 730                 GOTO(out, rc);
 731
 732         while (*cur_offset < lgi->lgi_attr.la_size) {
 733                 struct llog_rec_hdr     *rec, *last_rec;
 734                 struct llog_rec_tail    *tail;
 735
 736                 llog_skip_over(cur_offset, *cur_idx, next_idx, chunk_size);
 737
 738                 /* read up to next llog chunk_size block */
 739                 lgi->lgi_buf.lb_len = chunk_size -
 740                                       (*cur_offset & (chunk_size - 1));
 741                 lgi->lgi_buf.lb_buf = buf;
 742
 743                 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
 744                 if (rc < 0) {
 745                         CERROR("%s: can't read llog block from log "DFID
 746                                " offset "LPU64": rc = %d\n",
 747                                o->do_lu.lo_dev->ld_obd->obd_name,
 748                                PFID(lu_object_fid(&o->do_lu)), *cur_offset,
 749                                rc);
 750                         GOTO(out, rc);
 751                 }
 752
 753                 if (rc < len) {
 754                         /* signal the end of the valid buffer to
 755                          * llog_process */
 756                         memset(buf + rc, 0, len - rc);
 757                 }
 758
 759                 if (rc == 0) /* end of file, nothing to do */
 760                         GOTO(out, rc);
 761
 762                 if (rc < sizeof(*tail)) {
 763                         CERROR("%s: invalid llog block at log id "DOSTID"/%u "
 764                                "offset "LPU64"\n",
 765                                o->do_lu.lo_dev->ld_obd->obd_name,
 766                                POSTID(&loghandle->lgh_id.lgl_oi),
 767                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 768                         GOTO(out, rc = -EINVAL);
 769                 }
 770
 771                 rec = buf;
 772                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 773                         lustre_swab_llog_rec(rec);
 774
 775                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 776                                                 sizeof(struct llog_rec_tail));
 777                 /* get the last record in block */
 778                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 779                                                    tail->lrt_len);
 780
 781                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 782                         lustre_swab_llog_rec(last_rec);
 783                 LASSERT(last_rec->lrh_index == tail->lrt_index);
 784
 785                 *cur_idx = tail->lrt_index;
 786
 787                 /* this shouldn't happen */
 788                 if (tail->lrt_index == 0) {
 789                         CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
 790                                "offset "LPU64"\n",
 791                                o->do_lu.lo_dev->ld_obd->obd_name,
 792                                POSTID(&loghandle->lgh_id.lgl_oi),
 793                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 794                         GOTO(out, rc = -EINVAL);
 795                 }
 796                 if (tail->lrt_index < next_idx)
 797                         continue;
 798
 799                 /* sanity check that the start of the new buffer is no farther
 800                  * than the record that we wanted.  This shouldn't happen. */
 801                 if (rec->lrh_index > next_idx) {
 802                         CERROR("%s: missed desired record? %u > %u\n",
 803                                o->do_lu.lo_dev->ld_obd->obd_name,
 804                                rec->lrh_index, next_idx);
 805                         GOTO(out, rc = -ENOENT);
 806                 }
 807
 808                 /* Trim unsupported extensions for compat w/ older clients */
 809                 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 810                         changelog_block_trim_ext(rec, last_rec,
 811                                                  CLF_VERSION | CLF_RENAME);
 812
 813                 GOTO(out, rc = 0);
 814         }
 815         GOTO(out, rc = -EIO);
 816 out:
 817         return rc;
 818 }
 819
 820 /**
 821  * Implementation of the llog_operations::lop_prev_block
 822  *
 823  * This function finds the llog block to return which contains
 824  * record with required index but in reverse order - from end of llog
 825  * to the beginning.
 826  * It is main part of reverse llog processing.
 827  *
 828  * \param[in] env       execution environment
 829  * \param[in] loghandle llog handle of the current llog
 830  * \param[in] prev_idx  target index to find
 831  * \param[in] buf       pointer to data buffer to fill
 832  * \param[in] len       required len to read, it is llog_chunk_size usually.
 833  *
 834  * \retval              0 on successful buffer read
 835  * \retval              negative value on error
 836  */
 837 static int llog_osd_prev_block(const struct lu_env *env,
 838                                struct llog_handle *loghandle,
 839                                int prev_idx, void *buf, int len)
 840 {
 841         struct llog_thread_info *lgi = llog_info(env);
 842         struct dt_object        *o;
 843         struct dt_device        *dt;
 844         loff_t                   cur_offset;
 845         __u32                   chunk_size;
 846         int                      rc;
 847
 848         ENTRY;
 849
 850         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 851         if (len == 0 || len & (chunk_size - 1))
 852                 RETURN(-EINVAL);
 853
 854         CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
 855
 856         LASSERT(loghandle);
 857         LASSERT(loghandle->lgh_ctxt);
 858
 859         o = loghandle->lgh_obj;
 860         LASSERT(o);
 861         LASSERT(dt_object_exists(o));
 862         dt = lu2dt_dev(o->do_lu.lo_dev);
 863         LASSERT(dt);
 864
 865         cur_offset = chunk_size;
 866         llog_skip_over(&cur_offset, 0, prev_idx, chunk_size);
 867
 868         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 869         if (rc)
 870                 GOTO(out, rc);
 871
 872         while (cur_offset < lgi->lgi_attr.la_size) {
 873                 struct llog_rec_hdr     *rec, *last_rec;
 874                 struct llog_rec_tail    *tail;
 875
 876                 lgi->lgi_buf.lb_len = len;
 877                 lgi->lgi_buf.lb_buf = buf;
 878                 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
 879                 if (rc < 0) {
 880                         CERROR("%s: can't read llog block from log "DFID
 881                                " offset "LPU64": rc = %d\n",
 882                                o->do_lu.lo_dev->ld_obd->obd_name,
 883                                PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
 884                         GOTO(out, rc);
 885                 }
 886
 887                 if (rc == 0) /* end of file, nothing to do */
 888                         GOTO(out, rc);
 889
 890                 if (rc < sizeof(*tail)) {
 891                         CERROR("%s: invalid llog block at log id "DOSTID"/%u "
 892                                "offset "LPU64"\n",
 893                                o->do_lu.lo_dev->ld_obd->obd_name,
 894                                POSTID(&loghandle->lgh_id.lgl_oi),
 895                                loghandle->lgh_id.lgl_ogen, cur_offset);
 896                         GOTO(out, rc = -EINVAL);
 897                 }
 898
 899                 rec = buf;
 900                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 901                         lustre_swab_llog_rec(rec);
 902
 903                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 904                                                 sizeof(struct llog_rec_tail));
 905                 /* get the last record in block */
 906                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 907                                                    le32_to_cpu(tail->lrt_len));
 908
 909                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 910                         lustre_swab_llog_rec(last_rec);
 911                 LASSERT(last_rec->lrh_index == tail->lrt_index);
 912
 913                 /* this shouldn't happen */
 914                 if (tail->lrt_index == 0) {
 915                         CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
 916                                "offset "LPU64"\n",
 917                                o->do_lu.lo_dev->ld_obd->obd_name,
 918                                POSTID(&loghandle->lgh_id.lgl_oi),
 919                                loghandle->lgh_id.lgl_ogen, cur_offset);
 920                         GOTO(out, rc = -EINVAL);
 921                 }
 922                 if (tail->lrt_index < prev_idx)
 923                         continue;
 924
 925                 /* sanity check that the start of the new buffer is no farther
 926                  * than the record that we wanted.  This shouldn't happen. */
 927                 if (rec->lrh_index > prev_idx) {
 928                         CERROR("%s: missed desired record? %u > %u\n",
 929                                o->do_lu.lo_dev->ld_obd->obd_name,
 930                                rec->lrh_index, prev_idx);
 931                         GOTO(out, rc = -ENOENT);
 932                 }
 933
 934                 /* Trim unsupported extensions for compat w/ older clients */
 935                 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 936                         changelog_block_trim_ext(rec, last_rec,
 937                                                  CLF_VERSION | CLF_RENAME);
 938
 939                 GOTO(out, rc = 0);
 940         }
 941         GOTO(out, rc = -EIO);
 942 out:
 943         return rc;
 944 }
 945
 946 /**
 947  * This is helper function to get llog directory object. It is used by named
 948  * llog operations to find/insert/delete llog entry from llog directory.
 949  *
 950  * \param[in] env       execution environment
 951  * \param[in] ctxt      llog context
 952  *
 953  * \retval              dt_object of llog directory
 954  * \retval              ERR_PTR of negative value on error
 955  */
 956 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
 957                                           struct llog_ctxt *ctxt)
 958 {
 959         struct dt_device        *dt;
 960         struct dt_thread_info   *dti = dt_info(env);
 961         struct dt_object        *dir;
 962         int                      rc;
 963
 964         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
 965         if (ctxt->loc_dir == NULL) {
 966                 rc = dt_root_get(env, dt, &dti->dti_fid);
 967                 if (rc)
 968                         return ERR_PTR(rc);
 969                 dir = dt_locate(env, dt, &dti->dti_fid);
 970
 971                 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
 972                         lu_object_put(env, &dir->do_lu);
 973                         return ERR_PTR(-ENOTDIR);
 974                 }
 975         } else {
 976                 lu_object_get(&ctxt->loc_dir->do_lu);
 977                 dir = ctxt->loc_dir;
 978         }
 979
 980         return dir;
 981 }
 982
 983 /**
 984  * Implementation of the llog_operations::lop_open
 985  *
 986  * This function opens the llog by its logid or by name, it may open also
 987  * non existent llog and assing then new id to it.
 988  * The llog_open/llog_close pair works similar to lu_object_find/put,
 989  * the object may not exist prior open. The result of open is just dt_object
 990  * in the llog header.
 991  *
 992  * \param[in] env               execution environment
 993  * \param[in] handle            llog handle of the current llog
 994  * \param[in] logid             logid of llog to open (nameless llog)
 995  * \param[in] name              name of llog to open (named llog)
 996  * \param[in] open_param
 997  *                              LLOG_OPEN_NEW - new llog, may not exist
 998  *                              LLOG_OPEN_EXIST - old llog, must exist
 999  *
1000  * \retval                      0 on successful open, llog_handle::lgh_obj
1001  *                              contains the dt_object of the llog.
1002  * \retval                      negative value on error
1003  */
1004 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1005                          struct llog_logid *logid, char *name,
1006                          enum llog_open_param open_param)
1007 {
1008         struct llog_thread_info         *lgi = llog_info(env);
1009         struct llog_ctxt                *ctxt = handle->lgh_ctxt;
1010         struct dt_object                *o;
1011         struct dt_device                *dt;
1012         struct ls_device                *ls;
1013         struct local_oid_storage        *los = NULL;
1014         int                              rc = 0;
1015
1016         ENTRY;
1017
1018         LASSERT(env);
1019         LASSERT(ctxt);
1020         LASSERT(ctxt->loc_exp);
1021         LASSERT(ctxt->loc_exp->exp_obd);
1022         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1023         LASSERT(dt);
1024         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1025                 if (logid != NULL) {
1026                         logid_to_fid(logid, &lgi->lgi_fid);
1027                 } else {
1028                         /* If logid == NULL, then it means the caller needs
1029                          * to allocate new FID (llog_cat_declare_add_rec()). */
1030                         rc = obd_fid_alloc(env, ctxt->loc_exp,
1031                                            &lgi->lgi_fid, NULL);
1032                         if (rc < 0)
1033                                 RETURN(rc);
1034                         rc = 0;
1035                 }
1036
1037                 o = dt_locate(env, dt, &lgi->lgi_fid);
1038                 if (IS_ERR(o))
1039                         RETURN(PTR_ERR(o));
1040
1041                 goto after_open;
1042         }
1043
1044         ls = ls_device_get(dt);
1045         if (IS_ERR(ls))
1046                 RETURN(PTR_ERR(ls));
1047
1048         mutex_lock(&ls->ls_los_mutex);
1049         los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1050         mutex_unlock(&ls->ls_los_mutex);
1051         LASSERT(los);
1052         ls_device_put(env, ls);
1053
1054         LASSERT(handle);
1055
1056         if (logid != NULL) {
1057                 logid_to_fid(logid, &lgi->lgi_fid);
1058         } else if (name) {
1059                 struct dt_object *llog_dir;
1060
1061                 llog_dir = llog_osd_dir_get(env, ctxt);
1062                 if (IS_ERR(llog_dir))
1063                         GOTO(out, rc = PTR_ERR(llog_dir));
1064                 dt_read_lock(env, llog_dir, 0);
1065                 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1066                 dt_read_unlock(env, llog_dir);
1067                 lu_object_put(env, &llog_dir->do_lu);
1068                 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1069                         /* generate fid for new llog */
1070                         rc = local_object_fid_generate(env, los,
1071                                                        &lgi->lgi_fid);
1072                 }
1073                 if (rc < 0)
1074                         GOTO(out, rc);
1075                 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1076                 if (handle->lgh_name)
1077                         strcpy(handle->lgh_name, name);
1078                 else
1079                         GOTO(out, rc = -ENOMEM);
1080         } else {
1081                 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1082                 /* generate fid for new llog */
1083                 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1084                 if (rc < 0)
1085                         GOTO(out, rc);
1086         }
1087
1088         o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1089         if (IS_ERR(o))
1090                 GOTO(out_name, rc = PTR_ERR(o));
1091
1092 after_open:
1093         /* No new llog is expected but doesn't exist */
1094         if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1095                 GOTO(out_put, rc = -ENOENT);
1096
1097         fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1098         handle->lgh_obj = o;
1099         handle->private_data = los;
1100         LASSERT(handle->lgh_ctxt);
1101
1102         RETURN(rc);
1103
1104 out_put:
1105         lu_object_put(env, &o->do_lu);
1106 out_name:
1107         if (handle->lgh_name != NULL)
1108                 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1109 out:
1110         if (los != NULL)
1111                 dt_los_put(los);
1112         RETURN(rc);
1113 }
1114
1115 /**
1116  * Implementation of the llog_operations::lop_exist
1117  *
1118  * This function checks that llog exists on storage.
1119  *
1120  * \param[in] handle    llog handle of the current llog
1121  *
1122  * \retval              true if llog object exists and is not just destroyed
1123  * \retval              false if llog doesn't exist or just destroyed
1124  */
1125 static int llog_osd_exist(struct llog_handle *handle)
1126 {
1127         LASSERT(handle->lgh_obj);
1128         return (dt_object_exists(handle->lgh_obj) &&
1129                 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header));
1130 }
1131
1132 /**
1133  * Get dir for regular fid log object
1134  *
1135  * Get directory for regular fid log object, and these regular fid log
1136  * object will be inserted under this directory, to satisfy the FS
1137  * consistency check, e2fsck etc.
1138  *
1139  * \param [in] env      execution environment
1140  * \param [in] dto      llog object
1141  *
1142  * \retval              pointer to the directory if it is found.
1143  * \retval              ERR_PTR(negative errno) if it fails.
1144  */
1145 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1146                                                struct dt_object *dto)
1147 {
1148         struct llog_thread_info *lgi = llog_info(env);
1149         struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1150         struct lu_seq_range     *range = &lgi->lgi_range;
1151         struct lu_fid           *dir_fid = &lgi->lgi_fid;
1152         struct dt_object        *dir;
1153         int                     rc;
1154         ENTRY;
1155
1156         fld_range_set_any(range);
1157         LASSERT(ss != NULL);
1158         rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1159                                    fid_seq(lu_object_fid(&dto->do_lu)), range);
1160         if (rc < 0)
1161                 RETURN(ERR_PTR(rc));
1162
1163         lu_update_log_dir_fid(dir_fid, range->lsr_index);
1164         dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1165         if (IS_ERR(dir))
1166                 RETURN(dir);
1167
1168         if (!dt_try_as_dir(env, dir)) {
1169                 lu_object_put(env, &dir->do_lu);
1170                 RETURN(ERR_PTR(-ENOTDIR));
1171         }
1172
1173         RETURN(dir);
1174 }
1175
1176 /**
1177  * Add llog object with regular FID to name entry
1178  *
1179  * Add llog object with regular FID to name space, and each llog
1180  * object on each MDT will be /update_log_dir/[seq:oid:ver],
1181  * so to satisfy the namespace consistency check, e2fsck etc.
1182  *
1183  * \param [in] env      execution environment
1184  * \param [in] dto      llog object
1185  * \param [in] th       thandle
1186  * \param [in] declare  if it is declare or execution
1187  *
1188  * \retval              0 if insertion succeeds.
1189  * \retval              negative errno if insertion fails.
1190  */
1191 static int
1192 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1193                                     struct dt_object *dto,
1194                                     struct thandle *th, bool declare)
1195 {
1196         struct llog_thread_info *lgi = llog_info(env);
1197         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1198         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1199         struct dt_object        *dir;
1200         char                    *name = lgi->lgi_name;
1201         int                     rc;
1202         ENTRY;
1203
1204         if (!fid_is_norm(fid))
1205                 RETURN(0);
1206
1207         dir = llog_osd_get_regular_fid_dir(env, dto);
1208         if (IS_ERR(dir))
1209                 RETURN(PTR_ERR(dir));
1210
1211         rec->rec_fid = fid;
1212         rec->rec_type = S_IFREG;
1213         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1214         dt_write_lock(env, dir, 0);
1215         if (declare) {
1216                 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1217                                (struct dt_key *)name, th);
1218         } else {
1219                 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1220                                (struct dt_key *)name, th, 1);
1221         }
1222         dt_write_unlock(env, dir);
1223
1224         lu_object_put(env, &dir->do_lu);
1225         RETURN(rc);
1226 }
1227
1228
1229 /**
1230  * Implementation of the llog_operations::lop_declare_create
1231  *
1232  * This function declares the llog create. It declares also name insert
1233  * into llog directory in case of named llog.
1234  *
1235  * \param[in] env       execution environment
1236  * \param[in] res       llog handle of the current llog
1237  * \param[in] th        current transaction handle
1238  *
1239  * \retval              0 on successful create declaration
1240  * \retval              negative value on error
1241  */
1242 static int llog_osd_declare_create(const struct lu_env *env,
1243                                    struct llog_handle *res, struct thandle *th)
1244 {
1245         struct llog_thread_info         *lgi = llog_info(env);
1246         struct dt_insert_rec            *rec = &lgi->lgi_dt_rec;
1247         struct local_oid_storage        *los;
1248         struct dt_object                *o;
1249         int                              rc;
1250
1251         ENTRY;
1252
1253         LASSERT(res->lgh_obj);
1254         LASSERT(th);
1255
1256         /* object can be created by another thread */
1257         o = res->lgh_obj;
1258         if (dt_object_exists(o))
1259                 RETURN(0);
1260
1261         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1262                 struct llog_thread_info *lgi = llog_info(env);
1263
1264                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1265                 lgi->lgi_attr.la_size = 0;
1266                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1267                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1268
1269                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1270                                        &lgi->lgi_dof, th);
1271                 if (rc < 0)
1272                         RETURN(rc);
1273
1274
1275                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1276
1277                 RETURN(rc);
1278         }
1279         los = res->private_data;
1280         LASSERT(los);
1281
1282         rc = llog_osd_declare_new_object(env, los, o, th);
1283         if (rc)
1284                 RETURN(rc);
1285
1286         /* do not declare header initialization here as it's declared
1287          * in llog_osd_declare_write_rec() which is always called */
1288
1289         if (res->lgh_name) {
1290                 struct dt_object *llog_dir;
1291
1292                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1293                 if (IS_ERR(llog_dir))
1294                         RETURN(PTR_ERR(llog_dir));
1295                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1296                 rec->rec_fid = &lgi->lgi_fid;
1297                 rec->rec_type = S_IFREG;
1298                 rc = dt_declare_insert(env, llog_dir,
1299                                        (struct dt_rec *)rec,
1300                                        (struct dt_key *)res->lgh_name, th);
1301                 lu_object_put(env, &llog_dir->do_lu);
1302                 if (rc)
1303                         CERROR("%s: can't declare named llog %s: rc = %d\n",
1304                                o->do_lu.lo_dev->ld_obd->obd_name,
1305                                res->lgh_name, rc);
1306         }
1307         RETURN(rc);
1308 }
1309
1310 /**
1311  * Implementation of the llog_operations::lop_create
1312  *
1313  * This function creates the llog according with llog_handle::lgh_obj
1314  * and llog_handle::lgh_name.
1315  *
1316  * \param[in] env       execution environment
1317  * \param[in] res       llog handle of the current llog
1318  * \param[in] th        current transaction handle
1319  *
1320  * \retval              0 on successful create
1321  * \retval              negative value on error
1322  */
1323 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1324                            struct thandle *th)
1325 {
1326         struct llog_thread_info *lgi = llog_info(env);
1327         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1328         struct local_oid_storage *los;
1329         struct dt_object        *o;
1330         int                      rc = 0;
1331
1332         ENTRY;
1333
1334         LASSERT(env);
1335         o = res->lgh_obj;
1336         LASSERT(o);
1337
1338         /* llog can be already created */
1339         if (dt_object_exists(o))
1340                 RETURN(-EEXIST);
1341
1342         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1343                 struct llog_thread_info *lgi = llog_info(env);
1344
1345                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1346                 lgi->lgi_attr.la_size = 0;
1347                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1348                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1349
1350                 dt_write_lock(env, o, 0);
1351                 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1352                                &lgi->lgi_dof, th);
1353                 dt_write_unlock(env, o);
1354                 if (rc < 0)
1355                         RETURN(rc);
1356
1357                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1358
1359                 RETURN(rc);
1360         }
1361
1362         los = res->private_data;
1363         LASSERT(los);
1364
1365         dt_write_lock(env, o, 0);
1366         if (!dt_object_exists(o))
1367                 rc = llog_osd_create_new_object(env, los, o, th);
1368         else
1369                 rc = -EEXIST;
1370
1371         dt_write_unlock(env, o);
1372         if (rc)
1373                 RETURN(rc);
1374
1375         if (res->lgh_name) {
1376                 struct dt_object *llog_dir;
1377
1378                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1379                 if (IS_ERR(llog_dir))
1380                         RETURN(PTR_ERR(llog_dir));
1381
1382                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1383                 rec->rec_fid = &lgi->lgi_fid;
1384                 rec->rec_type = S_IFREG;
1385                 dt_read_lock(env, llog_dir, 0);
1386                 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1387                                (struct dt_key *)res->lgh_name,
1388                                th, 1);
1389                 dt_read_unlock(env, llog_dir);
1390                 lu_object_put(env, &llog_dir->do_lu);
1391                 if (rc)
1392                         CERROR("%s: can't create named llog %s: rc = %d\n",
1393                                o->do_lu.lo_dev->ld_obd->obd_name,
1394                                res->lgh_name, rc);
1395         }
1396         RETURN(rc);
1397 }
1398
1399 /**
1400  * Implementation of the llog_operations::lop_close
1401  *
1402  * This function closes the llog. It just put llog object and referenced
1403  * local storage.
1404  *
1405  * \param[in] env       execution environment
1406  * \param[in] handle    llog handle of the current llog
1407  *
1408  * \retval              0 on successful llog close
1409  * \retval              negative value on error
1410  */
1411 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1412 {
1413         struct local_oid_storage        *los;
1414         int                              rc = 0;
1415
1416         ENTRY;
1417
1418         LASSERT(handle->lgh_obj);
1419
1420         if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1421                 /* Remove the object from the cache, otherwise it may
1422                  * hold LOD being released during cleanup process */
1423                 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1424                 LASSERT(handle->private_data == NULL);
1425                 RETURN(rc);
1426         } else {
1427                 lu_object_put(env, &handle->lgh_obj->do_lu);
1428         }
1429         los = handle->private_data;
1430         LASSERT(los);
1431         dt_los_put(los);
1432
1433         if (handle->lgh_name)
1434                 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1435
1436         RETURN(rc);
1437 }
1438
1439 /**
1440  * delete llog object name entry
1441  *
1442  * Delete llog object (with regular FID) from name space (under
1443  * update_log_dir).
1444  *
1445  * \param [in] env      execution environment
1446  * \param [in] dto      llog object
1447  * \param [in] th       thandle
1448  * \param [in] declare  if it is declare or execution
1449  *
1450  * \retval              0 if deletion succeeds.
1451  * \retval              negative errno if deletion fails.
1452  */
1453 static int
1454 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1455                                     struct dt_object *dto,
1456                                     struct thandle *th, bool declare)
1457 {
1458         struct llog_thread_info *lgi = llog_info(env);
1459         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1460         struct dt_object        *dir;
1461         char                    *name = lgi->lgi_name;
1462         int                     rc;
1463         ENTRY;
1464
1465         if (!fid_is_norm(fid))
1466                 RETURN(0);
1467
1468         dir = llog_osd_get_regular_fid_dir(env, dto);
1469         if (IS_ERR(dir))
1470                 RETURN(PTR_ERR(dir));
1471
1472         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1473         dt_write_lock(env, dir, 0);
1474         if (declare) {
1475                 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1476                                        th);
1477         } else {
1478                 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1479         }
1480         dt_write_unlock(env, dir);
1481
1482         lu_object_put(env, &dir->do_lu);
1483         RETURN(rc);
1484 }
1485
1486
1487 /**
1488  * Implementation of the llog_operations::lop_destroy
1489  *
1490  * This function destroys the llog and deletes also entry in the
1491  * llog directory in case of named llog. Llog should be opened prior that.
1492  * Destroy method is not part of external transaction and does everything
1493  * inside.
1494  *
1495  * \param[in] env               execution environment
1496  * \param[in] loghandle llog handle of the current llog
1497  *
1498  * \retval              0 on successful destroy
1499  * \retval              negative value on error
1500  */
1501 static int llog_osd_destroy(const struct lu_env *env,
1502                             struct llog_handle *loghandle)
1503 {
1504         struct llog_ctxt        *ctxt;
1505         struct dt_object        *o, *llog_dir = NULL;
1506         struct dt_device        *d;
1507         struct thandle          *th;
1508         char                    *name = NULL;
1509         int                      rc;
1510
1511         ENTRY;
1512
1513         ctxt = loghandle->lgh_ctxt;
1514         LASSERT(ctxt);
1515
1516         o = loghandle->lgh_obj;
1517         LASSERT(o);
1518
1519         d = lu2dt_dev(o->do_lu.lo_dev);
1520         LASSERT(d);
1521         LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
1522
1523         th = dt_trans_create(env, d);
1524         if (IS_ERR(th))
1525                 RETURN(PTR_ERR(th));
1526
1527         if (loghandle->lgh_name) {
1528                 llog_dir = llog_osd_dir_get(env, ctxt);
1529                 if (IS_ERR(llog_dir))
1530                         GOTO(out_trans, rc = PTR_ERR(llog_dir));
1531
1532                 name = loghandle->lgh_name;
1533                 rc = dt_declare_delete(env, llog_dir,
1534                                        (struct dt_key *)name, th);
1535                 if (rc)
1536                         GOTO(out_trans, rc);
1537         }
1538
1539         rc = dt_declare_ref_del(env, o, th);
1540         if (rc < 0)
1541                 GOTO(out_trans, rc);
1542
1543         rc = dt_declare_destroy(env, o, th);
1544         if (rc)
1545                 GOTO(out_trans, rc);
1546
1547         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1548                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1549                 if (rc < 0)
1550                         GOTO(out_trans, rc);
1551         }
1552
1553         rc = dt_trans_start_local(env, d, th);
1554         if (rc)
1555                 GOTO(out_trans, rc);
1556
1557         th->th_wait_submit = 1;
1558
1559         dt_write_lock(env, o, 0);
1560         if (dt_object_exists(o)) {
1561                 if (name) {
1562                         dt_read_lock(env, llog_dir, 0);
1563                         rc = dt_delete(env, llog_dir,
1564                                        (struct dt_key *) name,
1565                                        th);
1566                         dt_read_unlock(env, llog_dir);
1567                         if (rc) {
1568                                 CERROR("%s: can't remove llog %s: rc = %d\n",
1569                                        o->do_lu.lo_dev->ld_obd->obd_name,
1570                                        name, rc);
1571                                 GOTO(out_unlock, rc);
1572                         }
1573                 }
1574                 dt_ref_del(env, o, th);
1575                 rc = dt_destroy(env, o, th);
1576                 if (rc)
1577                         GOTO(out_unlock, rc);
1578
1579                 if (loghandle->lgh_ctxt->loc_flags &
1580                                                 LLOG_CTXT_FLAG_NORMAL_FID) {
1581                         rc = llog_osd_regular_fid_del_name_entry(env, o, th,
1582                                                                  false);
1583                         if (rc < 0)
1584                                 GOTO(out_unlock, rc);
1585                 }
1586         }
1587 out_unlock:
1588         dt_write_unlock(env, o);
1589 out_trans:
1590         dt_trans_stop(env, d, th);
1591         if (!(IS_ERR_OR_NULL(llog_dir)))
1592                 lu_object_put(env, &llog_dir->do_lu);
1593         RETURN(rc);
1594 }
1595
1596 /**
1597  * Implementation of the llog_operations::lop_setup
1598  *
1599  * This function setup the llog on local storage.
1600  *
1601  * \param[in] env       execution environment
1602  * \param[in] obd       obd device the llog belongs to
1603  * \param[in] olg       the llog group, it is always zero group now.
1604  * \param[in] ctxt_idx  the llog index, it defines the purpose of this llog.
1605  *                      Every new llog type have to use own index.
1606  * \param[in] disk_obd  the storage obd, where llog is stored.
1607  *
1608  * \retval              0 on successful llog setup
1609  * \retval              negative value on error
1610  */
1611 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1612                           struct obd_llog_group *olg, int ctxt_idx,
1613                           struct obd_device *disk_obd)
1614 {
1615         struct llog_thread_info         *lgi = llog_info(env);
1616         struct llog_ctxt                *ctxt;
1617         int                              rc = 0;
1618         ENTRY;
1619
1620         LASSERT(obd);
1621         LASSERT(olg->olg_ctxts[ctxt_idx]);
1622
1623         ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1624         LASSERT(ctxt);
1625
1626         if (disk_obd == NULL)
1627                 GOTO(out, rc = 0);
1628
1629         /* initialize data allowing to generate new fids,
1630          * literally we need a sequece */
1631         lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1632         lgi->lgi_fid.f_oid = 1;
1633         lgi->lgi_fid.f_ver = 0;
1634         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1635                                     &lgi->lgi_fid,
1636                                     &ctxt->loc_los_nameless);
1637         if (rc != 0)
1638                 GOTO(out, rc);
1639
1640         lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1641         lgi->lgi_fid.f_oid = 1;
1642         lgi->lgi_fid.f_ver = 0;
1643         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1644                                     &lgi->lgi_fid,
1645                                     &ctxt->loc_los_named);
1646         if (rc != 0) {
1647                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1648                 ctxt->loc_los_nameless = NULL;
1649         }
1650
1651         GOTO(out, rc);
1652
1653 out:
1654         llog_ctxt_put(ctxt);
1655         return rc;
1656 }
1657
1658 /**
1659  * Implementation of the llog_operations::lop_cleanup
1660  *
1661  * This function cleanups the llog on local storage.
1662  *
1663  * \param[in] env       execution environment
1664  * \param[in] ctxt      the llog context
1665  *
1666  * \retval              0
1667  */
1668 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1669 {
1670         if (ctxt->loc_los_nameless != NULL) {
1671                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1672                 ctxt->loc_los_nameless = NULL;
1673         }
1674
1675         if (ctxt->loc_los_named != NULL) {
1676                 local_oid_storage_fini(env, ctxt->loc_los_named);
1677                 ctxt->loc_los_named = NULL;
1678         }
1679
1680         return 0;
1681 }
1682
1683 struct llog_operations llog_osd_ops = {
1684         .lop_next_block         = llog_osd_next_block,
1685         .lop_prev_block         = llog_osd_prev_block,
1686         .lop_read_header        = llog_osd_read_header,
1687         .lop_destroy            = llog_osd_destroy,
1688         .lop_setup              = llog_osd_setup,
1689         .lop_cleanup            = llog_osd_cleanup,
1690         .lop_open               = llog_osd_open,
1691         .lop_exist              = llog_osd_exist,
1692         .lop_declare_create     = llog_osd_declare_create,
1693         .lop_create             = llog_osd_create,
1694         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1695         .lop_write_rec          = llog_osd_write_rec,
1696         .lop_close              = llog_osd_close,
1697 };
1698 EXPORT_SYMBOL(llog_osd_ops);
1699
1700 struct llog_operations llog_common_cat_ops = {
1701         .lop_next_block         = llog_osd_next_block,
1702         .lop_prev_block         = llog_osd_prev_block,
1703         .lop_read_header        = llog_osd_read_header,
1704         .lop_destroy            = llog_osd_destroy,
1705         .lop_setup              = llog_osd_setup,
1706         .lop_cleanup            = llog_osd_cleanup,
1707         .lop_open               = llog_osd_open,
1708         .lop_exist              = llog_osd_exist,
1709         .lop_declare_create     = llog_osd_declare_create,
1710         .lop_create             = llog_osd_create,
1711         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1712         .lop_write_rec          = llog_osd_write_rec,
1713         .lop_close              = llog_osd_close,
1714         .lop_add                = llog_cat_add_rec,
1715         .lop_declare_add        = llog_cat_declare_add_rec,
1716 };
1717 EXPORT_SYMBOL(llog_common_cat_ops);
1718
1719 /**
1720  * Read the special file which contains the list of llog catalogs IDs
1721  *
1722  * This function reads the CATALOGS file which contains the array of llog
1723  * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1724  * by OST/MDT number.
1725  *
1726  * \param[in]  env              execution environment
1727  * \param[in]  d                corresponding storage device
1728  * \param[in]  idx              position to start from, usually OST/MDT index
1729  * \param[in]  count            how many catalog IDs to read
1730  * \param[out] idarray          the buffer for the data. If it is NULL then
1731  *                              function returns just number of catalog IDs
1732  *                              in the file.
1733  * \param[in]  fid              LLOG_CATALOGS_OID for CATALOG object
1734  *
1735  * \retval                      0 on successful read of catalog IDs
1736  * \retval                      negative value on error
1737  * \retval                      positive value which is number of records in
1738  *                              the file if \a idarray is NULL
1739  */
1740 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1741                           int idx, int count, struct llog_catid *idarray,
1742                           const struct lu_fid *fid)
1743 {
1744         struct llog_thread_info *lgi = llog_info(env);
1745         struct dt_object        *o = NULL;
1746         struct thandle          *th;
1747         int                      rc, size;
1748
1749         ENTRY;
1750
1751         LASSERT(d);
1752
1753         size = sizeof(*idarray) * count;
1754         lgi->lgi_off = idx *  sizeof(*idarray);
1755
1756         lgi->lgi_fid = *fid;
1757         o = dt_locate(env, d, &lgi->lgi_fid);
1758         if (IS_ERR(o))
1759                 RETURN(PTR_ERR(o));
1760
1761         if (!dt_object_exists(o)) {
1762                 th = dt_trans_create(env, d);
1763                 if (IS_ERR(th))
1764                         GOTO(out, rc = PTR_ERR(th));
1765
1766                 lgi->lgi_attr.la_valid = LA_MODE;
1767                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1768                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1769
1770                 th->th_wait_submit = 1;
1771                 /* Make the llog object creation synchronization, so
1772                  * it will be reliable to the reference, especially
1773                  * for remote reference */
1774                 th->th_sync = 1;
1775
1776                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1777                                        &lgi->lgi_dof, th);
1778                 if (rc)
1779                         GOTO(out_trans, rc);
1780
1781                 rc = dt_trans_start_local(env, d, th);
1782                 if (rc)
1783                         GOTO(out_trans, rc);
1784
1785                 dt_write_lock(env, o, 0);
1786                 if (!dt_object_exists(o))
1787                         rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1788                                        &lgi->lgi_dof, th);
1789                 dt_write_unlock(env, o);
1790 out_trans:
1791                 dt_trans_stop(env, d, th);
1792                 if (rc)
1793                         GOTO(out, rc);
1794         }
1795
1796         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1797         if (rc)
1798                 GOTO(out, rc);
1799
1800         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1801                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1802                        o->do_lu.lo_dev->ld_obd->obd_name,
1803                        lgi->lgi_attr.la_mode);
1804                 GOTO(out, rc = -ENOENT);
1805         }
1806
1807         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1808                (int)lgi->lgi_attr.la_size, size);
1809
1810         /* return just number of llogs */
1811         if (idarray == NULL) {
1812                 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1813                 GOTO(out, rc);
1814         }
1815
1816         /* read for new ost index or for empty file */
1817         memset(idarray, 0, size);
1818         if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1819                 GOTO(out, rc = 0);
1820         if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1821                 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1822
1823         lgi->lgi_buf.lb_buf = idarray;
1824         lgi->lgi_buf.lb_len = size;
1825         rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1826         /* -EFAULT means the llog is a sparse file. This is not an error
1827          * after arbitrary OST index is supported. */
1828         if (rc < 0 && rc != -EFAULT) {
1829                 CERROR("%s: error reading CATALOGS: rc = %d\n",
1830                        o->do_lu.lo_dev->ld_obd->obd_name,  rc);
1831                 GOTO(out, rc);
1832         }
1833
1834         EXIT;
1835 out:
1836         lu_object_put(env, &o->do_lu);
1837         RETURN(rc);
1838 }
1839 EXPORT_SYMBOL(llog_osd_get_cat_list);
1840
1841 /**
1842  * Write the special file which contains the list of llog catalogs IDs
1843  *
1844  * This function writes the CATALOG file which contains the array of llog
1845  * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
1846  * number.
1847  *
1848  * \param[in]  env      execution environment
1849  * \param[in]  d        corresponding storage device
1850  * \param[in]  idx      position to start from, usually OST/MDT index
1851  * \param[in]  count    how many catalog IDs to write
1852  * \param[out] idarray  the buffer with the data to write.
1853  * \param[in]  fid      LLOG_CATALOGS_OID for CATALOG object
1854  *
1855  * \retval              0 on successful write of catalog IDs
1856  * \retval              negative value on error
1857  */
1858 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
1859                           int idx, int count, struct llog_catid *idarray,
1860                           const struct lu_fid *fid)
1861 {
1862         struct llog_thread_info *lgi = llog_info(env);
1863         struct dt_object        *o = NULL;
1864         struct thandle          *th;
1865         int                      rc, size;
1866
1867         if (count == 0)
1868                 RETURN(0);
1869
1870         LASSERT(d);
1871
1872         size = sizeof(*idarray) * count;
1873         lgi->lgi_off = idx * sizeof(*idarray);
1874         lgi->lgi_fid = *fid;
1875
1876         o = dt_locate(env, d, &lgi->lgi_fid);
1877         if (IS_ERR(o))
1878                 RETURN(PTR_ERR(o));
1879
1880         if (!dt_object_exists(o))
1881                 GOTO(out, rc = -ENOENT);
1882
1883         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1884         if (rc)
1885                 GOTO(out, rc);
1886
1887         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1888                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1889                        o->do_lu.lo_dev->ld_obd->obd_name,
1890                        lgi->lgi_attr.la_mode);
1891                 GOTO(out, rc = -ENOENT);
1892         }
1893
1894         th = dt_trans_create(env, d);
1895         if (IS_ERR(th))
1896                 GOTO(out, rc = PTR_ERR(th));
1897
1898         lgi->lgi_buf.lb_len = size;
1899         lgi->lgi_buf.lb_buf = idarray;
1900         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
1901         if (rc)
1902                 GOTO(out, rc);
1903
1904         rc = dt_trans_start_local(env, d, th);
1905         if (rc)
1906                 GOTO(out_trans, rc);
1907
1908         th->th_wait_submit = 1;
1909
1910         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
1911         if (rc)
1912                 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
1913                        idx, rc);
1914 out_trans:
1915         dt_trans_stop(env, d, th);
1916 out:
1917         lu_object_put(env, &o->do_lu);
1918         RETURN(rc);
1919 }
1920 EXPORT_SYMBOL(llog_osd_put_cat_list);