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_mutex protects llog header data from concurrent
 527          * update/cancel, the llh_count and llh_bitmap are protected */
 528         mutex_lock(&loghandle->lgh_hdr_mutex);
 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                 mutex_unlock(&loghandle->lgh_hdr_mutex);
 533                 LBUG(); /* should never happen */
 534         }
 535         llh->llh_count++;
 536
 537         if (lgi->lgi_attr.la_size == 0) {
 538                 lgi->lgi_off = 0;
 539                 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
 540                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 541                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 542                 if (rc != 0)
 543                         GOTO(out_unlock, rc);
 544         } else {
 545                 __u32   *bitmap = LLOG_HDR_BITMAP(llh);
 546
 547                 /* Note: If this is not initialization (size == 0), then do not
 548                  * write the whole header (8k bytes), only update header/tail
 549                  * and bits needs to be updated. Because this update might be
 550                  * part of cross-MDT operation, which needs to write these
 551                  * updates into the update log(32KB limit) and also pack inside
 552                  * the RPC (1MB limit), if we write 8K for each operation, which
 553                  * will cost a lot space, and keep us adding more updates to one
 554                  * update log.*/
 555                 lgi->lgi_off = offsetof(typeof(*llh), llh_count);
 556                 lgi->lgi_buf.lb_len = sizeof(llh->llh_count);
 557                 lgi->lgi_buf.lb_buf = &llh->llh_count;
 558                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 559                 if (rc != 0)
 560                         GOTO(out_unlock, rc);
 561
 562                 lgi->lgi_off = llh->llh_bitmap_offset +
 563                               (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
 564                 lgi->lgi_buf.lb_len = sizeof(*bitmap);
 565                 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
 566                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 567                 if (rc != 0)
 568                         GOTO(out_unlock, rc);
 569
 570                 lgi->lgi_off =  (unsigned long)LLOG_HDR_TAIL(llh) -
 571                                 (unsigned long)llh;
 572                 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
 573                 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
 574                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 575                 if (rc != 0)
 576                         GOTO(out_unlock, rc);
 577         }
 578
 579 out_unlock:
 580         /* unlock here for remote object */
 581         mutex_unlock(&loghandle->lgh_hdr_mutex);
 582         if (rc)
 583                 GOTO(out, rc);
 584
 585         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 586         if (rc)
 587                 GOTO(out, rc);
 588
 589         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 590         lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size, lgi->lgi_off);
 591         lgi->lgi_buf.lb_len = reclen;
 592         lgi->lgi_buf.lb_buf = rec;
 593         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 594         if (rc < 0)
 595                 GOTO(out, rc);
 596
 597         CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u off"LPU64"\n",
 598                POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len,
 599                lgi->lgi_off);
 600         if (reccookie != NULL) {
 601                 reccookie->lgc_lgl = loghandle->lgh_id;
 602                 reccookie->lgc_index = index;
 603                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 604                     (rec->lrh_type == MDS_SETATTR64_REC))
 605                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 606                 else if (rec->lrh_type == OST_SZ_REC)
 607                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 608                 else
 609                         reccookie->lgc_subsys = -1;
 610                 rc = 1;
 611         }
 612         RETURN(rc);
 613 out:
 614         /* cleanup llog for error case */
 615         mutex_lock(&loghandle->lgh_hdr_mutex);
 616         ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
 617         llh->llh_count--;
 618         mutex_unlock(&loghandle->lgh_hdr_mutex);
 619
 620         /* restore llog last_idx */
 621         loghandle->lgh_last_idx--;
 622         LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
 623
 624         RETURN(rc);
 625 }
 626
 627 /**
 628  * We can skip reading at least as many log blocks as the number of
 629  * minimum sized log records we are skipping.  If it turns out
 630  * that we are not far enough along the log (because the
 631  * actual records are larger than minimum size) we just skip
 632  * some more records.
 633  */
 634 static inline void llog_skip_over(struct llog_log_hdr *llh, __u64 *off,
 635                                   int curr, int goal, __u32 chunk_size)
 636 {
 637         if (goal > curr) {
 638                 if (llh->llh_size == 0) {
 639                         /* variable size records */
 640                         *off = *off + (goal - curr - 1) * LLOG_MIN_REC_SIZE;
 641                 } else {
 642                         *off = chunk_size + (goal - 1) * llh->llh_size;
 643                 }
 644         }
 645         /* always align with lower chunk boundary*/
 646         *off &= ~(chunk_size - 1);
 647 }
 648
 649 /**
 650  * Remove optional fields that the client doesn't expect.
 651  * This is typically in order to ensure compatibility with older clients.
 652  * It is assumed that since we exclusively remove fields, the block will be
 653  * big enough to handle the remapped records. It is also assumed that records
 654  * of a block have the same format (i.e.: the same features enabled).
 655  *
 656  * \param[in,out]    hdr        Header of the block of records to remap.
 657  * \param[in,out]    last_hdr   Last header, don't read past this point.
 658  * \param[in]        flags      Flags describing the fields to keep.
 659  */
 660 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
 661                                      struct llog_rec_hdr *last_hdr,
 662                                      enum changelog_rec_flags flags)
 663 {
 664         if (hdr->lrh_type != CHANGELOG_REC)
 665                 return;
 666
 667         do {
 668                 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
 669
 670                 changelog_remap_rec(rec, rec->cr_flags & flags);
 671                 hdr = llog_rec_hdr_next(hdr);
 672         } while ((char *)hdr <= (char *)last_hdr);
 673 }
 674
 675 /**
 676  * Implementation of the llog_operations::lop_next_block
 677  *
 678  * This function finds the the next llog block to return which contains
 679  * record with required index. It is main part of llog processing.
 680  *
 681  * \param[in]     env           execution environment
 682  * \param[in]     loghandle     llog handle of the current llog
 683  * \param[in,out] cur_idx       index preceeding cur_offset
 684  * \param[in]     next_idx      target index to find
 685  * \param[in,out] cur_offset    furtherst point read in the file
 686  * \param[in]     buf           pointer to data buffer to fill
 687  * \param[in]     len           required len to read, it is
 688  *                              usually llog chunk_size.
 689  *
 690  * \retval                      0 on successful buffer read
 691  * \retval                      negative value on error
 692  */
 693 static int llog_osd_next_block(const struct lu_env *env,
 694                                struct llog_handle *loghandle, int *cur_idx,
 695                                int next_idx, __u64 *cur_offset, void *buf,
 696                                int len)
 697 {
 698         struct llog_thread_info *lgi = llog_info(env);
 699         struct dt_object        *o;
 700         struct dt_device        *dt;
 701         int                      rc;
 702         __u32                   chunk_size;
 703
 704         ENTRY;
 705
 706         LASSERT(env);
 707         LASSERT(lgi);
 708
 709         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 710         if (len == 0 || len & (chunk_size - 1))
 711                 RETURN(-EINVAL);
 712
 713         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
 714                next_idx, *cur_idx, *cur_offset);
 715
 716         LASSERT(loghandle);
 717         LASSERT(loghandle->lgh_ctxt);
 718
 719         o = loghandle->lgh_obj;
 720         LASSERT(o);
 721         LASSERT(dt_object_exists(o));
 722         dt = lu2dt_dev(o->do_lu.lo_dev);
 723         LASSERT(dt);
 724
 725         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 726         if (rc)
 727                 GOTO(out, rc);
 728
 729         while (*cur_offset < lgi->lgi_attr.la_size) {
 730                 struct llog_rec_hdr     *rec, *last_rec;
 731                 struct llog_rec_tail    *tail;
 732
 733                 llog_skip_over(loghandle->lgh_hdr, cur_offset, *cur_idx,
 734                                next_idx, chunk_size);
 735
 736                 /* read up to next llog chunk_size block */
 737                 lgi->lgi_buf.lb_len = chunk_size -
 738                                       (*cur_offset & (chunk_size - 1));
 739                 lgi->lgi_buf.lb_buf = buf;
 740
 741                 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
 742                 if (rc < 0) {
 743                         CERROR("%s: can't read llog block from log "DFID
 744                                " offset "LPU64": rc = %d\n",
 745                                o->do_lu.lo_dev->ld_obd->obd_name,
 746                                PFID(lu_object_fid(&o->do_lu)), *cur_offset,
 747                                rc);
 748                         GOTO(out, rc);
 749                 }
 750
 751                 if (rc < len) {
 752                         /* signal the end of the valid buffer to
 753                          * llog_process */
 754                         memset(buf + rc, 0, len - rc);
 755                 }
 756
 757                 if (rc == 0) /* end of file, nothing to do */
 758                         GOTO(out, rc);
 759
 760                 if (rc < sizeof(*tail)) {
 761                         CERROR("%s: invalid llog block at log id "DOSTID"/%u "
 762                                "offset "LPU64"\n",
 763                                o->do_lu.lo_dev->ld_obd->obd_name,
 764                                POSTID(&loghandle->lgh_id.lgl_oi),
 765                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 766                         GOTO(out, rc = -EINVAL);
 767                 }
 768
 769                 rec = buf;
 770                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 771                         lustre_swab_llog_rec(rec);
 772
 773                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 774                                                 sizeof(struct llog_rec_tail));
 775                 /* get the last record in block */
 776                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 777                                                    tail->lrt_len);
 778
 779                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 780                         lustre_swab_llog_rec(last_rec);
 781                 LASSERT(last_rec->lrh_index == tail->lrt_index);
 782
 783                 *cur_idx = tail->lrt_index;
 784
 785                 /* this shouldn't happen */
 786                 if (tail->lrt_index == 0) {
 787                         CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
 788                                "offset "LPU64"\n",
 789                                o->do_lu.lo_dev->ld_obd->obd_name,
 790                                POSTID(&loghandle->lgh_id.lgl_oi),
 791                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 792                         GOTO(out, rc = -EINVAL);
 793                 }
 794                 if (tail->lrt_index < next_idx)
 795                         continue;
 796
 797                 /* sanity check that the start of the new buffer is no farther
 798                  * than the record that we wanted.  This shouldn't happen. */
 799                 if (rec->lrh_index > next_idx) {
 800                         CERROR("%s: missed desired record? %u > %u\n",
 801                                o->do_lu.lo_dev->ld_obd->obd_name,
 802                                rec->lrh_index, next_idx);
 803                         GOTO(out, rc = -ENOENT);
 804                 }
 805
 806                 /* Trim unsupported extensions for compat w/ older clients */
 807                 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 808                         changelog_block_trim_ext(rec, last_rec,
 809                                                  CLF_VERSION | CLF_RENAME);
 810
 811                 GOTO(out, rc = 0);
 812         }
 813         GOTO(out, rc = -EIO);
 814 out:
 815         return rc;
 816 }
 817
 818 /**
 819  * Implementation of the llog_operations::lop_prev_block
 820  *
 821  * This function finds the llog block to return which contains
 822  * record with required index but in reverse order - from end of llog
 823  * to the beginning.
 824  * It is main part of reverse llog processing.
 825  *
 826  * \param[in] env       execution environment
 827  * \param[in] loghandle llog handle of the current llog
 828  * \param[in] prev_idx  target index to find
 829  * \param[in] buf       pointer to data buffer to fill
 830  * \param[in] len       required len to read, it is llog_chunk_size usually.
 831  *
 832  * \retval              0 on successful buffer read
 833  * \retval              negative value on error
 834  */
 835 static int llog_osd_prev_block(const struct lu_env *env,
 836                                struct llog_handle *loghandle,
 837                                int prev_idx, void *buf, int len)
 838 {
 839         struct llog_thread_info *lgi = llog_info(env);
 840         struct dt_object        *o;
 841         struct dt_device        *dt;
 842         loff_t                   cur_offset;
 843         __u32                   chunk_size;
 844         int                      rc;
 845
 846         ENTRY;
 847
 848         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 849         if (len == 0 || len & (chunk_size - 1))
 850                 RETURN(-EINVAL);
 851
 852         CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
 853
 854         LASSERT(loghandle);
 855         LASSERT(loghandle->lgh_ctxt);
 856
 857         o = loghandle->lgh_obj;
 858         LASSERT(o);
 859         LASSERT(dt_object_exists(o));
 860         dt = lu2dt_dev(o->do_lu.lo_dev);
 861         LASSERT(dt);
 862
 863         cur_offset = chunk_size;
 864         llog_skip_over(loghandle->lgh_hdr, &cur_offset, 0, prev_idx,
 865                        chunk_size);
 866
 867         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 868         if (rc)
 869                 GOTO(out, rc);
 870
 871         while (cur_offset < lgi->lgi_attr.la_size) {
 872                 struct llog_rec_hdr     *rec, *last_rec;
 873                 struct llog_rec_tail    *tail;
 874
 875                 lgi->lgi_buf.lb_len = len;
 876                 lgi->lgi_buf.lb_buf = buf;
 877                 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
 878                 if (rc < 0) {
 879                         CERROR("%s: can't read llog block from log "DFID
 880                                " offset "LPU64": rc = %d\n",
 881                                o->do_lu.lo_dev->ld_obd->obd_name,
 882                                PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
 883                         GOTO(out, rc);
 884                 }
 885
 886                 if (rc == 0) /* end of file, nothing to do */
 887                         GOTO(out, rc);
 888
 889                 if (rc < sizeof(*tail)) {
 890                         CERROR("%s: invalid llog block at log id "DOSTID"/%u "
 891                                "offset "LPU64"\n",
 892                                o->do_lu.lo_dev->ld_obd->obd_name,
 893                                POSTID(&loghandle->lgh_id.lgl_oi),
 894                                loghandle->lgh_id.lgl_ogen, cur_offset);
 895                         GOTO(out, rc = -EINVAL);
 896                 }
 897
 898                 rec = buf;
 899                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 900                         lustre_swab_llog_rec(rec);
 901
 902                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 903                                                 sizeof(struct llog_rec_tail));
 904                 /* get the last record in block */
 905                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 906                                                    le32_to_cpu(tail->lrt_len));
 907
 908                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 909                         lustre_swab_llog_rec(last_rec);
 910                 LASSERT(last_rec->lrh_index == tail->lrt_index);
 911
 912                 /* this shouldn't happen */
 913                 if (tail->lrt_index == 0) {
 914                         CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
 915                                "offset "LPU64"\n",
 916                                o->do_lu.lo_dev->ld_obd->obd_name,
 917                                POSTID(&loghandle->lgh_id.lgl_oi),
 918                                loghandle->lgh_id.lgl_ogen, cur_offset);
 919                         GOTO(out, rc = -EINVAL);
 920                 }
 921                 if (tail->lrt_index < prev_idx)
 922                         continue;
 923
 924                 /* sanity check that the start of the new buffer is no farther
 925                  * than the record that we wanted.  This shouldn't happen. */
 926                 if (rec->lrh_index > prev_idx) {
 927                         CERROR("%s: missed desired record? %u > %u\n",
 928                                o->do_lu.lo_dev->ld_obd->obd_name,
 929                                rec->lrh_index, prev_idx);
 930                         GOTO(out, rc = -ENOENT);
 931                 }
 932
 933                 /* Trim unsupported extensions for compat w/ older clients */
 934                 if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 935                         changelog_block_trim_ext(rec, last_rec,
 936                                                  CLF_VERSION | CLF_RENAME);
 937
 938                 GOTO(out, rc = 0);
 939         }
 940         GOTO(out, rc = -EIO);
 941 out:
 942         return rc;
 943 }
 944
 945 /**
 946  * This is helper function to get llog directory object. It is used by named
 947  * llog operations to find/insert/delete llog entry from llog directory.
 948  *
 949  * \param[in] env       execution environment
 950  * \param[in] ctxt      llog context
 951  *
 952  * \retval              dt_object of llog directory
 953  * \retval              ERR_PTR of negative value on error
 954  */
 955 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
 956                                           struct llog_ctxt *ctxt)
 957 {
 958         struct dt_device        *dt;
 959         struct dt_thread_info   *dti = dt_info(env);
 960         struct dt_object        *dir;
 961         int                      rc;
 962
 963         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
 964         if (ctxt->loc_dir == NULL) {
 965                 rc = dt_root_get(env, dt, &dti->dti_fid);
 966                 if (rc)
 967                         return ERR_PTR(rc);
 968                 dir = dt_locate(env, dt, &dti->dti_fid);
 969
 970                 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
 971                         lu_object_put(env, &dir->do_lu);
 972                         return ERR_PTR(-ENOTDIR);
 973                 }
 974         } else {
 975                 lu_object_get(&ctxt->loc_dir->do_lu);
 976                 dir = ctxt->loc_dir;
 977         }
 978
 979         return dir;
 980 }
 981
 982 /**
 983  * Implementation of the llog_operations::lop_open
 984  *
 985  * This function opens the llog by its logid or by name, it may open also
 986  * non existent llog and assing then new id to it.
 987  * The llog_open/llog_close pair works similar to lu_object_find/put,
 988  * the object may not exist prior open. The result of open is just dt_object
 989  * in the llog header.
 990  *
 991  * \param[in] env               execution environment
 992  * \param[in] handle            llog handle of the current llog
 993  * \param[in] logid             logid of llog to open (nameless llog)
 994  * \param[in] name              name of llog to open (named llog)
 995  * \param[in] open_param
 996  *                              LLOG_OPEN_NEW - new llog, may not exist
 997  *                              LLOG_OPEN_EXIST - old llog, must exist
 998  *
 999  * \retval                      0 on successful open, llog_handle::lgh_obj
1000  *                              contains the dt_object of the llog.
1001  * \retval                      negative value on error
1002  */
1003 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1004                          struct llog_logid *logid, char *name,
1005                          enum llog_open_param open_param)
1006 {
1007         struct llog_thread_info         *lgi = llog_info(env);
1008         struct llog_ctxt                *ctxt = handle->lgh_ctxt;
1009         struct dt_object                *o;
1010         struct dt_device                *dt;
1011         struct ls_device                *ls;
1012         struct local_oid_storage        *los = NULL;
1013         int                              rc = 0;
1014
1015         ENTRY;
1016
1017         LASSERT(env);
1018         LASSERT(ctxt);
1019         LASSERT(ctxt->loc_exp);
1020         LASSERT(ctxt->loc_exp->exp_obd);
1021         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1022         LASSERT(dt);
1023         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1024                 if (logid != NULL) {
1025                         logid_to_fid(logid, &lgi->lgi_fid);
1026                 } else {
1027                         /* If logid == NULL, then it means the caller needs
1028                          * to allocate new FID (llog_cat_declare_add_rec()). */
1029                         rc = obd_fid_alloc(env, ctxt->loc_exp,
1030                                            &lgi->lgi_fid, NULL);
1031                         if (rc < 0)
1032                                 RETURN(rc);
1033                         rc = 0;
1034                 }
1035
1036                 o = dt_locate(env, dt, &lgi->lgi_fid);
1037                 if (IS_ERR(o))
1038                         RETURN(PTR_ERR(o));
1039
1040                 goto after_open;
1041         }
1042
1043         ls = ls_device_get(dt);
1044         if (IS_ERR(ls))
1045                 RETURN(PTR_ERR(ls));
1046
1047         mutex_lock(&ls->ls_los_mutex);
1048         los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1049         mutex_unlock(&ls->ls_los_mutex);
1050         LASSERT(los);
1051         ls_device_put(env, ls);
1052
1053         LASSERT(handle);
1054
1055         if (logid != NULL) {
1056                 logid_to_fid(logid, &lgi->lgi_fid);
1057         } else if (name) {
1058                 struct dt_object *llog_dir;
1059
1060                 llog_dir = llog_osd_dir_get(env, ctxt);
1061                 if (IS_ERR(llog_dir))
1062                         GOTO(out, rc = PTR_ERR(llog_dir));
1063                 dt_read_lock(env, llog_dir, 0);
1064                 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1065                 dt_read_unlock(env, llog_dir);
1066                 lu_object_put(env, &llog_dir->do_lu);
1067                 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1068                         /* generate fid for new llog */
1069                         rc = local_object_fid_generate(env, los,
1070                                                        &lgi->lgi_fid);
1071                 }
1072                 if (rc < 0)
1073                         GOTO(out, rc);
1074                 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1075                 if (handle->lgh_name)
1076                         strcpy(handle->lgh_name, name);
1077                 else
1078                         GOTO(out, rc = -ENOMEM);
1079         } else {
1080                 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1081                 /* generate fid for new llog */
1082                 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1083                 if (rc < 0)
1084                         GOTO(out, rc);
1085         }
1086
1087         o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1088         if (IS_ERR(o))
1089                 GOTO(out_name, rc = PTR_ERR(o));
1090
1091 after_open:
1092         /* No new llog is expected but doesn't exist */
1093         if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
1094                 GOTO(out_put, rc = -ENOENT);
1095
1096         fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1097         handle->lgh_obj = o;
1098         handle->private_data = los;
1099         LASSERT(handle->lgh_ctxt);
1100
1101         RETURN(rc);
1102
1103 out_put:
1104         lu_object_put(env, &o->do_lu);
1105 out_name:
1106         if (handle->lgh_name != NULL)
1107                 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1108 out:
1109         if (los != NULL)
1110                 dt_los_put(los);
1111         RETURN(rc);
1112 }
1113
1114 /**
1115  * Implementation of the llog_operations::lop_exist
1116  *
1117  * This function checks that llog exists on storage.
1118  *
1119  * \param[in] handle    llog handle of the current llog
1120  *
1121  * \retval              true if llog object exists and is not just destroyed
1122  * \retval              false if llog doesn't exist or just destroyed
1123  */
1124 static int llog_osd_exist(struct llog_handle *handle)
1125 {
1126         LASSERT(handle->lgh_obj);
1127         return (dt_object_exists(handle->lgh_obj) &&
1128                 !lu_object_is_dying(handle->lgh_obj->do_lu.lo_header));
1129 }
1130
1131 /**
1132  * Get dir for regular fid log object
1133  *
1134  * Get directory for regular fid log object, and these regular fid log
1135  * object will be inserted under this directory, to satisfy the FS
1136  * consistency check, e2fsck etc.
1137  *
1138  * \param [in] env      execution environment
1139  * \param [in] dto      llog object
1140  *
1141  * \retval              pointer to the directory if it is found.
1142  * \retval              ERR_PTR(negative errno) if it fails.
1143  */
1144 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1145                                                struct dt_object *dto)
1146 {
1147         struct llog_thread_info *lgi = llog_info(env);
1148         struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1149         struct lu_seq_range     *range = &lgi->lgi_range;
1150         struct lu_fid           *dir_fid = &lgi->lgi_fid;
1151         struct dt_object        *dir;
1152         int                     rc;
1153         ENTRY;
1154
1155         fld_range_set_any(range);
1156         LASSERT(ss != NULL);
1157         rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1158                                    fid_seq(lu_object_fid(&dto->do_lu)), range);
1159         if (rc < 0)
1160                 RETURN(ERR_PTR(rc));
1161
1162         lu_update_log_dir_fid(dir_fid, range->lsr_index);
1163         dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1164         if (IS_ERR(dir))
1165                 RETURN(dir);
1166
1167         if (!dt_try_as_dir(env, dir)) {
1168                 lu_object_put(env, &dir->do_lu);
1169                 RETURN(ERR_PTR(-ENOTDIR));
1170         }
1171
1172         RETURN(dir);
1173 }
1174
1175 /**
1176  * Add llog object with regular FID to name entry
1177  *
1178  * Add llog object with regular FID to name space, and each llog
1179  * object on each MDT will be /update_log_dir/[seq:oid:ver],
1180  * so to satisfy the namespace consistency check, e2fsck etc.
1181  *
1182  * \param [in] env      execution environment
1183  * \param [in] dto      llog object
1184  * \param [in] th       thandle
1185  * \param [in] declare  if it is declare or execution
1186  *
1187  * \retval              0 if insertion succeeds.
1188  * \retval              negative errno if insertion fails.
1189  */
1190 static int
1191 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1192                                     struct dt_object *dto,
1193                                     struct thandle *th, bool declare)
1194 {
1195         struct llog_thread_info *lgi = llog_info(env);
1196         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1197         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1198         struct dt_object        *dir;
1199         char                    *name = lgi->lgi_name;
1200         int                     rc;
1201         ENTRY;
1202
1203         if (!fid_is_norm(fid))
1204                 RETURN(0);
1205
1206         dir = llog_osd_get_regular_fid_dir(env, dto);
1207         if (IS_ERR(dir))
1208                 RETURN(PTR_ERR(dir));
1209
1210         rec->rec_fid = fid;
1211         rec->rec_type = S_IFREG;
1212         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1213         dt_write_lock(env, dir, 0);
1214         if (declare) {
1215                 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1216                                (struct dt_key *)name, th);
1217         } else {
1218                 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1219                                (struct dt_key *)name, th, 1);
1220         }
1221         dt_write_unlock(env, dir);
1222
1223         lu_object_put(env, &dir->do_lu);
1224         RETURN(rc);
1225 }
1226
1227
1228 /**
1229  * Implementation of the llog_operations::lop_declare_create
1230  *
1231  * This function declares the llog create. It declares also name insert
1232  * into llog directory in case of named llog.
1233  *
1234  * \param[in] env       execution environment
1235  * \param[in] res       llog handle of the current llog
1236  * \param[in] th        current transaction handle
1237  *
1238  * \retval              0 on successful create declaration
1239  * \retval              negative value on error
1240  */
1241 static int llog_osd_declare_create(const struct lu_env *env,
1242                                    struct llog_handle *res, struct thandle *th)
1243 {
1244         struct llog_thread_info         *lgi = llog_info(env);
1245         struct dt_insert_rec            *rec = &lgi->lgi_dt_rec;
1246         struct local_oid_storage        *los;
1247         struct dt_object                *o;
1248         int                              rc;
1249
1250         ENTRY;
1251
1252         LASSERT(res->lgh_obj);
1253         LASSERT(th);
1254
1255         /* object can be created by another thread */
1256         o = res->lgh_obj;
1257         if (dt_object_exists(o))
1258                 RETURN(0);
1259
1260         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1261                 struct llog_thread_info *lgi = llog_info(env);
1262
1263                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1264                 lgi->lgi_attr.la_size = 0;
1265                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1266                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1267
1268                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1269                                        &lgi->lgi_dof, th);
1270                 if (rc < 0)
1271                         RETURN(rc);
1272
1273
1274                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1275
1276                 RETURN(rc);
1277         }
1278         los = res->private_data;
1279         LASSERT(los);
1280
1281         rc = llog_osd_declare_new_object(env, los, o, th);
1282         if (rc)
1283                 RETURN(rc);
1284
1285         /* do not declare header initialization here as it's declared
1286          * in llog_osd_declare_write_rec() which is always called */
1287
1288         if (res->lgh_name) {
1289                 struct dt_object *llog_dir;
1290
1291                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1292                 if (IS_ERR(llog_dir))
1293                         RETURN(PTR_ERR(llog_dir));
1294                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1295                 rec->rec_fid = &lgi->lgi_fid;
1296                 rec->rec_type = S_IFREG;
1297                 rc = dt_declare_insert(env, llog_dir,
1298                                        (struct dt_rec *)rec,
1299                                        (struct dt_key *)res->lgh_name, th);
1300                 lu_object_put(env, &llog_dir->do_lu);
1301                 if (rc)
1302                         CERROR("%s: can't declare named llog %s: rc = %d\n",
1303                                o->do_lu.lo_dev->ld_obd->obd_name,
1304                                res->lgh_name, rc);
1305         }
1306         RETURN(rc);
1307 }
1308
1309 /**
1310  * Implementation of the llog_operations::lop_create
1311  *
1312  * This function creates the llog according with llog_handle::lgh_obj
1313  * and llog_handle::lgh_name.
1314  *
1315  * \param[in] env       execution environment
1316  * \param[in] res       llog handle of the current llog
1317  * \param[in] th        current transaction handle
1318  *
1319  * \retval              0 on successful create
1320  * \retval              negative value on error
1321  */
1322 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1323                            struct thandle *th)
1324 {
1325         struct llog_thread_info *lgi = llog_info(env);
1326         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1327         struct local_oid_storage *los;
1328         struct dt_object        *o;
1329         int                      rc = 0;
1330
1331         ENTRY;
1332
1333         LASSERT(env);
1334         o = res->lgh_obj;
1335         LASSERT(o);
1336
1337         /* llog can be already created */
1338         if (dt_object_exists(o))
1339                 RETURN(-EEXIST);
1340
1341         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1342                 struct llog_thread_info *lgi = llog_info(env);
1343
1344                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1345                 lgi->lgi_attr.la_size = 0;
1346                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1347                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1348
1349                 dt_write_lock(env, o, 0);
1350                 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1351                                &lgi->lgi_dof, th);
1352                 dt_write_unlock(env, o);
1353                 if (rc < 0)
1354                         RETURN(rc);
1355
1356                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1357
1358                 RETURN(rc);
1359         }
1360
1361         los = res->private_data;
1362         LASSERT(los);
1363
1364         dt_write_lock(env, o, 0);
1365         if (!dt_object_exists(o))
1366                 rc = llog_osd_create_new_object(env, los, o, th);
1367         else
1368                 rc = -EEXIST;
1369
1370         dt_write_unlock(env, o);
1371         if (rc)
1372                 RETURN(rc);
1373
1374         if (res->lgh_name) {
1375                 struct dt_object *llog_dir;
1376
1377                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1378                 if (IS_ERR(llog_dir))
1379                         RETURN(PTR_ERR(llog_dir));
1380
1381                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1382                 rec->rec_fid = &lgi->lgi_fid;
1383                 rec->rec_type = S_IFREG;
1384                 dt_read_lock(env, llog_dir, 0);
1385                 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1386                                (struct dt_key *)res->lgh_name,
1387                                th, 1);
1388                 dt_read_unlock(env, llog_dir);
1389                 lu_object_put(env, &llog_dir->do_lu);
1390                 if (rc)
1391                         CERROR("%s: can't create named llog %s: rc = %d\n",
1392                                o->do_lu.lo_dev->ld_obd->obd_name,
1393                                res->lgh_name, rc);
1394         }
1395         RETURN(rc);
1396 }
1397
1398 /**
1399  * Implementation of the llog_operations::lop_close
1400  *
1401  * This function closes the llog. It just put llog object and referenced
1402  * local storage.
1403  *
1404  * \param[in] env       execution environment
1405  * \param[in] handle    llog handle of the current llog
1406  *
1407  * \retval              0 on successful llog close
1408  * \retval              negative value on error
1409  */
1410 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1411 {
1412         struct local_oid_storage        *los;
1413         int                              rc = 0;
1414
1415         ENTRY;
1416
1417         LASSERT(handle->lgh_obj);
1418
1419         if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1420                 /* Remove the object from the cache, otherwise it may
1421                  * hold LOD being released during cleanup process */
1422                 lu_object_put_nocache(env, &handle->lgh_obj->do_lu);
1423                 LASSERT(handle->private_data == NULL);
1424                 RETURN(rc);
1425         } else {
1426                 lu_object_put(env, &handle->lgh_obj->do_lu);
1427         }
1428         los = handle->private_data;
1429         LASSERT(los);
1430         dt_los_put(los);
1431
1432         if (handle->lgh_name)
1433                 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1434
1435         RETURN(rc);
1436 }
1437
1438 /**
1439  * delete llog object name entry
1440  *
1441  * Delete llog object (with regular FID) from name space (under
1442  * update_log_dir).
1443  *
1444  * \param [in] env      execution environment
1445  * \param [in] dto      llog object
1446  * \param [in] th       thandle
1447  * \param [in] declare  if it is declare or execution
1448  *
1449  * \retval              0 if deletion succeeds.
1450  * \retval              negative errno if deletion fails.
1451  */
1452 static int
1453 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1454                                     struct dt_object *dto,
1455                                     struct thandle *th, bool declare)
1456 {
1457         struct llog_thread_info *lgi = llog_info(env);
1458         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1459         struct dt_object        *dir;
1460         char                    *name = lgi->lgi_name;
1461         int                     rc;
1462         ENTRY;
1463
1464         if (!fid_is_norm(fid))
1465                 RETURN(0);
1466
1467         dir = llog_osd_get_regular_fid_dir(env, dto);
1468         if (IS_ERR(dir))
1469                 RETURN(PTR_ERR(dir));
1470
1471         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1472         dt_write_lock(env, dir, 0);
1473         if (declare) {
1474                 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1475                                        th);
1476         } else {
1477                 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1478         }
1479         dt_write_unlock(env, dir);
1480
1481         lu_object_put(env, &dir->do_lu);
1482         RETURN(rc);
1483 }
1484
1485
1486 /**
1487  * Implementation of the llog_operations::lop_destroy
1488  *
1489  * This function destroys the llog and deletes also entry in the
1490  * llog directory in case of named llog. Llog should be opened prior that.
1491  * Destroy method is not part of external transaction and does everything
1492  * inside.
1493  *
1494  * \param[in] env               execution environment
1495  * \param[in] loghandle llog handle of the current llog
1496  *
1497  * \retval              0 on successful destroy
1498  * \retval              negative value on error
1499  */
1500 static int llog_osd_destroy(const struct lu_env *env,
1501                             struct llog_handle *loghandle)
1502 {
1503         struct llog_ctxt        *ctxt;
1504         struct dt_object        *o, *llog_dir = NULL;
1505         struct dt_device        *d;
1506         struct thandle          *th;
1507         char                    *name = NULL;
1508         int                      rc;
1509
1510         ENTRY;
1511
1512         ctxt = loghandle->lgh_ctxt;
1513         LASSERT(ctxt);
1514
1515         o = loghandle->lgh_obj;
1516         LASSERT(o);
1517
1518         d = lu2dt_dev(o->do_lu.lo_dev);
1519         LASSERT(d);
1520         LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
1521
1522         th = dt_trans_create(env, d);
1523         if (IS_ERR(th))
1524                 RETURN(PTR_ERR(th));
1525
1526         if (loghandle->lgh_name) {
1527                 llog_dir = llog_osd_dir_get(env, ctxt);
1528                 if (IS_ERR(llog_dir))
1529                         GOTO(out_trans, rc = PTR_ERR(llog_dir));
1530
1531                 name = loghandle->lgh_name;
1532                 rc = dt_declare_delete(env, llog_dir,
1533                                        (struct dt_key *)name, th);
1534                 if (rc)
1535                         GOTO(out_trans, rc);
1536         }
1537
1538         rc = dt_declare_ref_del(env, o, th);
1539         if (rc < 0)
1540                 GOTO(out_trans, rc);
1541
1542         rc = dt_declare_destroy(env, o, th);
1543         if (rc)
1544                 GOTO(out_trans, rc);
1545
1546         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1547                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1548                 if (rc < 0)
1549                         GOTO(out_trans, rc);
1550         }
1551
1552         rc = dt_trans_start_local(env, d, th);
1553         if (rc)
1554                 GOTO(out_trans, rc);
1555
1556         th->th_wait_submit = 1;
1557
1558         dt_write_lock(env, o, 0);
1559         if (dt_object_exists(o)) {
1560                 if (name) {
1561                         dt_read_lock(env, llog_dir, 0);
1562                         rc = dt_delete(env, llog_dir,
1563                                        (struct dt_key *) name,
1564                                        th);
1565                         dt_read_unlock(env, llog_dir);
1566                         if (rc) {
1567                                 CERROR("%s: can't remove llog %s: rc = %d\n",
1568                                        o->do_lu.lo_dev->ld_obd->obd_name,
1569                                        name, rc);
1570                                 GOTO(out_unlock, rc);
1571                         }
1572                 }
1573                 dt_ref_del(env, o, th);
1574                 rc = dt_destroy(env, o, th);
1575                 if (rc)
1576                         GOTO(out_unlock, rc);
1577
1578                 if (loghandle->lgh_ctxt->loc_flags &
1579                                                 LLOG_CTXT_FLAG_NORMAL_FID) {
1580                         rc = llog_osd_regular_fid_del_name_entry(env, o, th,
1581                                                                  false);
1582                         if (rc < 0)
1583                                 GOTO(out_unlock, rc);
1584                 }
1585         }
1586 out_unlock:
1587         dt_write_unlock(env, o);
1588 out_trans:
1589         dt_trans_stop(env, d, th);
1590         if (!(IS_ERR_OR_NULL(llog_dir)))
1591                 lu_object_put(env, &llog_dir->do_lu);
1592         RETURN(rc);
1593 }
1594
1595 /**
1596  * Implementation of the llog_operations::lop_setup
1597  *
1598  * This function setup the llog on local storage.
1599  *
1600  * \param[in] env       execution environment
1601  * \param[in] obd       obd device the llog belongs to
1602  * \param[in] olg       the llog group, it is always zero group now.
1603  * \param[in] ctxt_idx  the llog index, it defines the purpose of this llog.
1604  *                      Every new llog type have to use own index.
1605  * \param[in] disk_obd  the storage obd, where llog is stored.
1606  *
1607  * \retval              0 on successful llog setup
1608  * \retval              negative value on error
1609  */
1610 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1611                           struct obd_llog_group *olg, int ctxt_idx,
1612                           struct obd_device *disk_obd)
1613 {
1614         struct llog_thread_info         *lgi = llog_info(env);
1615         struct llog_ctxt                *ctxt;
1616         int                              rc = 0;
1617         ENTRY;
1618
1619         LASSERT(obd);
1620         LASSERT(olg->olg_ctxts[ctxt_idx]);
1621
1622         ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1623         LASSERT(ctxt);
1624
1625         if (disk_obd == NULL)
1626                 GOTO(out, rc = 0);
1627
1628         /* initialize data allowing to generate new fids,
1629          * literally we need a sequece */
1630         lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1631         lgi->lgi_fid.f_oid = 1;
1632         lgi->lgi_fid.f_ver = 0;
1633         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1634                                     &lgi->lgi_fid,
1635                                     &ctxt->loc_los_nameless);
1636         if (rc != 0)
1637                 GOTO(out, rc);
1638
1639         lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1640         lgi->lgi_fid.f_oid = 1;
1641         lgi->lgi_fid.f_ver = 0;
1642         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1643                                     &lgi->lgi_fid,
1644                                     &ctxt->loc_los_named);
1645         if (rc != 0) {
1646                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1647                 ctxt->loc_los_nameless = NULL;
1648         }
1649
1650         GOTO(out, rc);
1651
1652 out:
1653         llog_ctxt_put(ctxt);
1654         return rc;
1655 }
1656
1657 /**
1658  * Implementation of the llog_operations::lop_cleanup
1659  *
1660  * This function cleanups the llog on local storage.
1661  *
1662  * \param[in] env       execution environment
1663  * \param[in] ctxt      the llog context
1664  *
1665  * \retval              0
1666  */
1667 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1668 {
1669         if (ctxt->loc_los_nameless != NULL) {
1670                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1671                 ctxt->loc_los_nameless = NULL;
1672         }
1673
1674         if (ctxt->loc_los_named != NULL) {
1675                 local_oid_storage_fini(env, ctxt->loc_los_named);
1676                 ctxt->loc_los_named = NULL;
1677         }
1678
1679         return 0;
1680 }
1681
1682 struct llog_operations llog_osd_ops = {
1683         .lop_next_block         = llog_osd_next_block,
1684         .lop_prev_block         = llog_osd_prev_block,
1685         .lop_read_header        = llog_osd_read_header,
1686         .lop_destroy            = llog_osd_destroy,
1687         .lop_setup              = llog_osd_setup,
1688         .lop_cleanup            = llog_osd_cleanup,
1689         .lop_open               = llog_osd_open,
1690         .lop_exist              = llog_osd_exist,
1691         .lop_declare_create     = llog_osd_declare_create,
1692         .lop_create             = llog_osd_create,
1693         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1694         .lop_write_rec          = llog_osd_write_rec,
1695         .lop_close              = llog_osd_close,
1696 };
1697 EXPORT_SYMBOL(llog_osd_ops);
1698
1699 struct llog_operations llog_common_cat_ops = {
1700         .lop_next_block         = llog_osd_next_block,
1701         .lop_prev_block         = llog_osd_prev_block,
1702         .lop_read_header        = llog_osd_read_header,
1703         .lop_destroy            = llog_osd_destroy,
1704         .lop_setup              = llog_osd_setup,
1705         .lop_cleanup            = llog_osd_cleanup,
1706         .lop_open               = llog_osd_open,
1707         .lop_exist              = llog_osd_exist,
1708         .lop_declare_create     = llog_osd_declare_create,
1709         .lop_create             = llog_osd_create,
1710         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1711         .lop_write_rec          = llog_osd_write_rec,
1712         .lop_close              = llog_osd_close,
1713         .lop_add                = llog_cat_add_rec,
1714         .lop_declare_add        = llog_cat_declare_add_rec,
1715 };
1716 EXPORT_SYMBOL(llog_common_cat_ops);
1717
1718 /**
1719  * Read the special file which contains the list of llog catalogs IDs
1720  *
1721  * This function reads the CATALOGS file which contains the array of llog
1722  * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1723  * by OST/MDT number.
1724  *
1725  * \param[in]  env              execution environment
1726  * \param[in]  d                corresponding storage device
1727  * \param[in]  idx              position to start from, usually OST/MDT index
1728  * \param[in]  count            how many catalog IDs to read
1729  * \param[out] idarray          the buffer for the data. If it is NULL then
1730  *                              function returns just number of catalog IDs
1731  *                              in the file.
1732  * \param[in]  fid              LLOG_CATALOGS_OID for CATALOG object
1733  *
1734  * \retval                      0 on successful read of catalog IDs
1735  * \retval                      negative value on error
1736  * \retval                      positive value which is number of records in
1737  *                              the file if \a idarray is NULL
1738  */
1739 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
1740                           int idx, int count, struct llog_catid *idarray,
1741                           const struct lu_fid *fid)
1742 {
1743         struct llog_thread_info *lgi = llog_info(env);
1744         struct dt_object        *o = NULL;
1745         struct thandle          *th;
1746         int                      rc, size;
1747
1748         ENTRY;
1749
1750         LASSERT(d);
1751
1752         size = sizeof(*idarray) * count;
1753         lgi->lgi_off = idx *  sizeof(*idarray);
1754
1755         lgi->lgi_fid = *fid;
1756         o = dt_locate(env, d, &lgi->lgi_fid);
1757         if (IS_ERR(o))
1758                 RETURN(PTR_ERR(o));
1759
1760         if (!dt_object_exists(o)) {
1761                 th = dt_trans_create(env, d);
1762                 if (IS_ERR(th))
1763                         GOTO(out, rc = PTR_ERR(th));
1764
1765                 lgi->lgi_attr.la_valid = LA_MODE;
1766                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1767                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1768
1769                 th->th_wait_submit = 1;
1770                 /* Make the llog object creation synchronization, so
1771                  * it will be reliable to the reference, especially
1772                  * for remote reference */
1773                 th->th_sync = 1;
1774
1775                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1776                                        &lgi->lgi_dof, th);
1777                 if (rc)
1778                         GOTO(out_trans, rc);
1779
1780                 rc = dt_trans_start_local(env, d, th);
1781                 if (rc)
1782                         GOTO(out_trans, rc);
1783
1784                 dt_write_lock(env, o, 0);
1785                 if (!dt_object_exists(o))
1786                         rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1787                                        &lgi->lgi_dof, th);
1788                 dt_write_unlock(env, o);
1789 out_trans:
1790                 dt_trans_stop(env, d, th);
1791                 if (rc)
1792                         GOTO(out, rc);
1793         }
1794
1795         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1796         if (rc)
1797                 GOTO(out, rc);
1798
1799         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1800                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1801                        o->do_lu.lo_dev->ld_obd->obd_name,
1802                        lgi->lgi_attr.la_mode);
1803                 GOTO(out, rc = -ENOENT);
1804         }
1805
1806         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
1807                (int)lgi->lgi_attr.la_size, size);
1808
1809         /* return just number of llogs */
1810         if (idarray == NULL) {
1811                 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
1812                 GOTO(out, rc);
1813         }
1814
1815         /* read for new ost index or for empty file */
1816         memset(idarray, 0, size);
1817         if (lgi->lgi_attr.la_size <= lgi->lgi_off)
1818                 GOTO(out, rc = 0);
1819         if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
1820                 size = lgi->lgi_attr.la_size - lgi->lgi_off;
1821
1822         lgi->lgi_buf.lb_buf = idarray;
1823         lgi->lgi_buf.lb_len = size;
1824         rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
1825         /* -EFAULT means the llog is a sparse file. This is not an error
1826          * after arbitrary OST index is supported. */
1827         if (rc < 0 && rc != -EFAULT) {
1828                 CERROR("%s: error reading CATALOGS: rc = %d\n",
1829                        o->do_lu.lo_dev->ld_obd->obd_name,  rc);
1830                 GOTO(out, rc);
1831         }
1832
1833         EXIT;
1834 out:
1835         lu_object_put(env, &o->do_lu);
1836         RETURN(rc);
1837 }
1838 EXPORT_SYMBOL(llog_osd_get_cat_list);
1839
1840 /**
1841  * Write the special file which contains the list of llog catalogs IDs
1842  *
1843  * This function writes the CATALOG file which contains the array of llog
1844  * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
1845  * number.
1846  *
1847  * \param[in]  env      execution environment
1848  * \param[in]  d        corresponding storage device
1849  * \param[in]  idx      position to start from, usually OST/MDT index
1850  * \param[in]  count    how many catalog IDs to write
1851  * \param[out] idarray  the buffer with the data to write.
1852  * \param[in]  fid      LLOG_CATALOGS_OID for CATALOG object
1853  *
1854  * \retval              0 on successful write of catalog IDs
1855  * \retval              negative value on error
1856  */
1857 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
1858                           int idx, int count, struct llog_catid *idarray,
1859                           const struct lu_fid *fid)
1860 {
1861         struct llog_thread_info *lgi = llog_info(env);
1862         struct dt_object        *o = NULL;
1863         struct thandle          *th;
1864         int                      rc, size;
1865
1866         if (count == 0)
1867                 RETURN(0);
1868
1869         LASSERT(d);
1870
1871         size = sizeof(*idarray) * count;
1872         lgi->lgi_off = idx * sizeof(*idarray);
1873         lgi->lgi_fid = *fid;
1874
1875         o = dt_locate(env, d, &lgi->lgi_fid);
1876         if (IS_ERR(o))
1877                 RETURN(PTR_ERR(o));
1878
1879         if (!dt_object_exists(o))
1880                 GOTO(out, rc = -ENOENT);
1881
1882         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1883         if (rc)
1884                 GOTO(out, rc);
1885
1886         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
1887                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
1888                        o->do_lu.lo_dev->ld_obd->obd_name,
1889                        lgi->lgi_attr.la_mode);
1890                 GOTO(out, rc = -ENOENT);
1891         }
1892
1893         th = dt_trans_create(env, d);
1894         if (IS_ERR(th))
1895                 GOTO(out, rc = PTR_ERR(th));
1896
1897         lgi->lgi_buf.lb_len = size;
1898         lgi->lgi_buf.lb_buf = idarray;
1899         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
1900         if (rc)
1901                 GOTO(out, rc);
1902
1903         /* For update log, this happens during initialization,
1904          * see lod_sub_prep_llog(), and we need make sure catlog
1905          * file ID is written to catlist file(committed) before
1906          * cross-MDT operation write update records to catlog FILE,
1907          * otherwise, during failover these update records might
1908          * missing */
1909         if (fid_is_update_log(fid))
1910                 th->th_sync = 1;
1911
1912         rc = dt_trans_start_local(env, d, th);
1913         if (rc)
1914                 GOTO(out_trans, rc);
1915
1916         th->th_wait_submit = 1;
1917
1918         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
1919         if (rc)
1920                 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
1921                        idx, rc);
1922 out_trans:
1923         dt_trans_stop(env, d, th);
1924 out:
1925         lu_object_put(env, &o->do_lu);
1926         RETURN(rc);
1927 }
1928 EXPORT_SYMBOL(llog_osd_put_cat_list);