lustre/obdclass/llog_osd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 /*
  33  * lustre/obdclass/llog_osd.c
  34  *
  35  * Low level llog routines on top of OSD API
  36  *
  37  * This file provides set of methods for llog operations on top of
  38  * dt_device. It contains all supported llog_operations interfaces and
  39  * supplimental functions.
  40  *
  41  * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
  42  * Author: Mikhail Pershin <mike.pershin@intel.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_LOG
  46
  47 #include <linux/delay.h>
  48
  49 #include <dt_object.h>
  50 #include <llog_swab.h>
  51 #include <lustre_fid.h>
  52 #include <obd.h>
  53 #include <obd_class.h>
  54
  55 #include "llog_internal.h"
  56 #include "local_storage.h"
  57
  58 /**
  59  * Implementation of the llog_operations::lop_declare_create
  60  *
  61  * This function is a wrapper over local_storage API function
  62  * local_object_declare_create().
  63  *
  64  * \param[in] env       execution environment
  65  * \param[in] los       local_storage for bottom storage device
  66  * \param[in] o         dt_object to create
  67  * \param[in] th        current transaction handle
  68  *
  69  * \retval              0 on successful declaration of the new object
  70  * \retval              negative error if declaration was failed
  71  */
  72 static int llog_osd_declare_new_object(const struct lu_env *env,
  73                                        struct local_oid_storage *los,
  74                                        struct dt_object *o,
  75                                        struct thandle *th)
  76 {
  77         struct llog_thread_info *lgi = llog_info(env);
  78
  79         lgi->lgi_attr.la_valid = LA_MODE;
  80         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
  81         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
  82
  83         return local_object_declare_create(env, los, o, &lgi->lgi_attr,
  84                                            &lgi->lgi_dof, th);
  85 }
  86
  87 /**
  88  * Implementation of the llog_operations::lop_create
  89  *
  90  * This function is a wrapper over local_storage API function
  91  * local_object_create().
  92  *
  93  * \param[in] env       execution environment
  94  * \param[in] los       local_storage for bottom storage device
  95  * \param[in] o         dt_object to create
  96  * \param[in] th        current transaction handle
  97  *
  98  * \retval              0 on successful creation of the new object
  99  * \retval              negative error if creation was failed
 100  */
 101 static int llog_osd_create_new_object(const struct lu_env *env,
 102                                       struct local_oid_storage *los,
 103                                       struct dt_object *o,
 104                                       struct thandle *th)
 105 {
 106         struct llog_thread_info *lgi = llog_info(env);
 107
 108         lgi->lgi_attr.la_valid = LA_MODE;
 109         lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
 110         lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
 111
 112         return local_object_create(env, los, o, &lgi->lgi_attr,
 113                                    &lgi->lgi_dof, th);
 114 }
 115
 116 /**
 117  * Implementation of the llog_operations::lop_exist
 118  *
 119  * This function checks that llog exists on storage.
 120  *
 121  * \param[in] handle    llog handle of the current llog
 122  *
 123  * \retval              true if llog object exists and is not just destroyed
 124  * \retval              false if llog doesn't exist or just destroyed
 125  */
 126 static int llog_osd_exist(struct llog_handle *handle)
 127 {
 128         LASSERT(handle->lgh_obj);
 129         return dt_object_exists(handle->lgh_obj) && !handle->lgh_destroyed;
 130 }
 131
 132 static void *rec_tail(struct llog_rec_hdr *rec)
 133 {
 134         return (void *)((char *)rec + rec->lrh_len -
 135                         sizeof(struct llog_rec_tail));
 136 }
 137
 138 /**
 139  * Write a padding record to the llog
 140  *
 141  * This function writes a padding record to the end of llog. That may
 142  * be needed if llog contains records of variable size, e.g. config logs
 143  * or changelogs.
 144  * The padding record just aligns llog to the llog chunk_size boundary if
 145  * the current record doesn't fit in the remaining space.
 146  *
 147  * It allocates full length to avoid two separate writes for header and tail.
 148  * Such 2-steps scheme needs extra protection and complex error handling.
 149  *
 150  * \param[in]     env   execution environment
 151  * \param[in]     o     dt_object to create
 152  * \param[in,out] off   pointer to the padding start offset
 153  * \param[in]     len   padding length
 154  * \param[in]     index index of the padding record in a llog
 155  * \param[in]     th    current transaction handle
 156  *
 157  * \retval              0 on successful padding write
 158  * \retval              negative error if write failed
 159  */
 160 static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
 161                         loff_t *off, int len, int index, struct thandle *th)
 162 {
 163         struct llog_thread_info *lgi = llog_info(env);
 164         struct llog_rec_hdr     *rec;
 165         struct llog_rec_tail    *tail;
 166         int                      rc;
 167
 168         ENTRY;
 169
 170         LASSERT(th);
 171         LASSERT(off);
 172         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
 173
 174         OBD_ALLOC(rec, len);
 175         if (rec == NULL)
 176                 RETURN(-ENOMEM);
 177
 178         rec->lrh_len = len;
 179         rec->lrh_index = index;
 180         rec->lrh_type = LLOG_PAD_MAGIC;
 181
 182         tail = rec_tail(rec);
 183         tail->lrt_len = len;
 184         tail->lrt_index = index;
 185
 186         lgi->lgi_buf.lb_buf = rec;
 187         lgi->lgi_buf.lb_len = len;
 188         rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
 189         if (rc)
 190                 CERROR("%s: error writing padding record: rc = %d\n",
 191                        o->do_lu.lo_dev->ld_obd->obd_name, rc);
 192
 193         OBD_FREE(rec, len);
 194         RETURN(rc);
 195 }
 196
 197 /**
 198  * Implementation of the llog_operations::lop_read_header
 199  *
 200  * This function reads the current llog header from the bottom storage
 201  * device.
 202  *
 203  * \param[in] env       execution environment
 204  * \param[in] handle    llog handle of the current llog
 205  *
 206  * \retval              0 on successful header read
 207  * \retval              negative error if read failed
 208  */
 209 static int llog_osd_read_header(const struct lu_env *env,
 210                                 struct llog_handle *handle)
 211 {
 212         struct llog_rec_hdr     *llh_hdr;
 213         struct dt_object        *o;
 214         struct llog_thread_info *lgi;
 215         enum llog_flag           flags;
 216         int                      rc;
 217
 218         ENTRY;
 219
 220         o = handle->lgh_obj;
 221         LASSERT(o);
 222
 223         lgi = llog_info(env);
 224
 225         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 226         if (rc)
 227                 RETURN(rc);
 228
 229         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 230
 231         if (lgi->lgi_attr.la_size == 0) {
 232                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
 233                 RETURN(LLOG_EEMPTY);
 234         }
 235
 236         flags = handle->lgh_hdr->llh_flags;
 237
 238         lgi->lgi_off = 0;
 239         lgi->lgi_buf.lb_buf = handle->lgh_hdr;
 240         lgi->lgi_buf.lb_len = handle->lgh_hdr_size;
 241         rc = dt_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
 242         llh_hdr = &handle->lgh_hdr->llh_hdr;
 243         if (rc < sizeof(*llh_hdr) || rc < llh_hdr->lrh_len) {
 244                 CERROR("%s: error reading "DFID" log header size %d: rc = %d\n",
 245                        o->do_lu.lo_dev->ld_obd->obd_name,
 246                        PFID(lu_object_fid(&o->do_lu)), rc < 0 ? 0 : rc,
 247                        -EFAULT);
 248
 249                 if (rc >= 0)
 250                         rc = -EFAULT;
 251
 252                 RETURN(rc);
 253         }
 254
 255         if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
 256                 lustre_swab_llog_hdr(handle->lgh_hdr);
 257
 258         if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
 259                 CERROR("%s: bad log %s "DFID" header magic: %#x "
 260                        "(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
 261                        handle->lgh_name ? handle->lgh_name : "",
 262                        PFID(lu_object_fid(&o->do_lu)),
 263                        llh_hdr->lrh_type, LLOG_HDR_MAGIC);
 264                 RETURN(-EIO);
 265         } else if (llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
 266                    llh_hdr->lrh_len > handle->lgh_hdr_size) {
 267                 CERROR("%s: incorrectly sized log %s "DFID" header: "
 268                        "%#x (expected at least %#x)\n"
 269                        "you may need to re-run lconf --write_conf.\n",
 270                        o->do_lu.lo_dev->ld_obd->obd_name,
 271                        handle->lgh_name ? handle->lgh_name : "",
 272                        PFID(lu_object_fid(&o->do_lu)),
 273                        llh_hdr->lrh_len, LLOG_MIN_CHUNK_SIZE);
 274                 RETURN(-EIO);
 275         } else if (LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index >
 276                    LLOG_HDR_BITMAP_SIZE(handle->lgh_hdr) ||
 277                    LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len !=
 278                         llh_hdr->lrh_len) {
 279                 CERROR("%s: incorrectly sized log %s "DFID" tailer: "
 280                        "%#x : rc = %d\n",
 281                        o->do_lu.lo_dev->ld_obd->obd_name,
 282                        handle->lgh_name ? handle->lgh_name : "",
 283                        PFID(lu_object_fid(&o->do_lu)),
 284                        LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len, -EIO);
 285                 RETURN(-EIO);
 286         }
 287
 288         handle->lgh_hdr->llh_flags |= (flags & LLOG_F_EXT_MASK);
 289         handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_index;
 290
 291         RETURN(0);
 292 }
 293
 294 /**
 295  * Implementation of the llog_operations::lop_declare_write
 296  *
 297  * This function declares the new record write.
 298  *
 299  * \param[in] env       execution environment
 300  * \param[in] loghandle llog handle of the current llog
 301  * \param[in] rec       llog record header. This is a real header of the full
 302  *                      llog record to write. This is the beginning of buffer
 303  *                      to write, the length of buffer is stored in
 304  *                      \a rec::lrh_len
 305  * \param[in] idx       index of the llog record. If \a idx == -1 then this is
 306  *                      append case, otherwise \a idx is the index of record
 307  *                      to modify
 308  * \param[in] th        current transaction handle
 309  *
 310  * \retval              0 on successful declaration
 311  * \retval              negative error if declaration failed
 312  */
 313 static int llog_osd_declare_write_rec(const struct lu_env *env,
 314                                       struct llog_handle *loghandle,
 315                                       struct llog_rec_hdr *rec,
 316                                       int idx, struct thandle *th)
 317 {
 318         struct llog_thread_info *lgi = llog_info(env);
 319         __u32                   chunk_size;
 320         struct dt_object        *o;
 321         int                      rc;
 322
 323         ENTRY;
 324
 325         LASSERT(env);
 326         LASSERT(th);
 327         LASSERT(loghandle);
 328         LASSERT(rec);
 329         LASSERT(rec->lrh_len <= loghandle->lgh_ctxt->loc_chunk_size);
 330
 331         o = loghandle->lgh_obj;
 332         LASSERT(o);
 333
 334         chunk_size = loghandle->lgh_ctxt->loc_chunk_size;
 335         lgi->lgi_buf.lb_len = chunk_size;
 336         lgi->lgi_buf.lb_buf = NULL;
 337         /* each time we update header */
 338         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, 0,
 339                                      th);
 340         if (rc || idx == 0) /* if error or just header */
 341                 RETURN(rc);
 342
 343         /**
 344          * the pad record can be inserted so take into account double
 345          * record size
 346          */
 347         lgi->lgi_buf.lb_len = chunk_size * 2;
 348         lgi->lgi_buf.lb_buf = NULL;
 349         /* XXX: implement declared window or multi-chunks approach */
 350         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, -1, th);
 351
 352         RETURN(rc);
 353 }
 354
 355 /**
 356  * Implementation of the llog_operations::lop_write
 357  *
 358  * This function writes the new record in the llog or modify the existed one.
 359  *
 360  * \param[in]  env              execution environment
 361  * \param[in]  loghandle        llog handle of the current llog
 362  * \param[in]  rec              llog record header. This is a real header of
 363  *                              the full llog record to write. This is
 364  *                              the beginning of buffer to write, the length
 365  *                              of buffer is stored in \a rec::lrh_len
 366  * \param[in,out] reccookie     pointer to the cookie to return back if needed.
 367  *                              It is used for further cancel of this llog
 368  *                              record.
 369  * \param[in]  idx              index of the llog record. If \a idx == -1 then
 370  *                              this is append case, otherwise \a idx is
 371  *                              the index of record to modify
 372  * \param[in]  th               current transaction handle
 373  *
 374  * \retval                      0 on successful write && \a reccookie == NULL
 375  *                              1 on successful write && \a reccookie != NULL
 376  * \retval                      negative error if write failed
 377  */
 378 static int llog_osd_write_rec(const struct lu_env *env,
 379                               struct llog_handle *loghandle,
 380                               struct llog_rec_hdr *rec,
 381                               struct llog_cookie *reccookie,
 382                               int idx, struct thandle *th)
 383 {
 384         struct llog_thread_info *lgi = llog_info(env);
 385         struct llog_log_hdr     *llh;
 386         int                      reclen = rec->lrh_len;
 387         int                      index, rc;
 388         struct llog_rec_tail    *lrt;
 389         struct dt_object        *o;
 390         __u32                   chunk_size;
 391         size_t                   left;
 392         __u32                   orig_last_idx;
 393         ENTRY;
 394
 395         llh = loghandle->lgh_hdr;
 396         o = loghandle->lgh_obj;
 397
 398         chunk_size = llh->llh_hdr.lrh_len;
 399         CDEBUG(D_OTHER, "new record %x to "DFID"\n",
 400                rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
 401
 402         if (!llog_osd_exist(loghandle))
 403                 RETURN(-ENOENT);
 404
 405         /* record length should not bigger than  */
 406         if (reclen > loghandle->lgh_hdr->llh_hdr.lrh_len)
 407                 RETURN(-E2BIG);
 408
 409         /* sanity check for fixed-records llog */
 410         if (idx != LLOG_HEADER_IDX && (llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
 411                 LASSERT(llh->llh_size != 0);
 412                 LASSERT(llh->llh_size == reclen);
 413         }
 414
 415         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 416         if (rc)
 417                 RETURN(rc);
 418
 419         /**
 420          * The modification case.
 421          * If idx set then the record with that index must be modified.
 422          * There are three cases possible:
 423          * 1) the common case is the llog header update (idx == 0)
 424          * 2) the llog record modification during llog process.
 425          *    This is indicated by the \a loghandle::lgh_cur_idx > 0.
 426          *    In that case the \a loghandle::lgh_cur_offset
 427          * 3) otherwise this is assumed that llog consist of records of
 428          *    fixed size, i.e. catalog. The llog header must has llh_size
 429          *    field equal to record size. The record offset is calculated
 430          *    just by /a idx value
 431          *
 432          * During modification we don't need extra header update because
 433          * the bitmap and record count are not changed. The record header
 434          * and tail remains the same too.
 435          */
 436         if (idx != LLOG_NEXT_IDX) {
 437                 /* llog can be empty only when first record is being written */
 438                 LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
 439
 440                 if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
 441                         CERROR("%s: modify unset record %u\n",
 442                                o->do_lu.lo_dev->ld_obd->obd_name, idx);
 443                         RETURN(-ENOENT);
 444                 }
 445
 446                 if (idx != rec->lrh_index) {
 447                         CERROR("%s: modify index mismatch %d %u\n",
 448                                o->do_lu.lo_dev->ld_obd->obd_name, idx,
 449                                rec->lrh_index);
 450                         RETURN(-EFAULT);
 451                 }
 452
 453                 if (idx == LLOG_HEADER_IDX) {
 454                         /* llog header update */
 455                         __u32   *bitmap = LLOG_HDR_BITMAP(llh);
 456
 457                         lgi->lgi_off = 0;
 458
 459                         /* If it does not indicate the bitmap index
 460                          * (reccookie == NULL), then it means update
 461                          * the whole update header. Otherwise only
 462                          * update header and bits needs to be updated,
 463                          * and in DNE cases, it will signaficantly
 464                          * shrink the RPC size.
 465                          * see distribute_txn_cancel_records()*/
 466                         if (reccookie == NULL) {
 467                                 lgi->lgi_buf.lb_len = reclen;
 468                                 lgi->lgi_buf.lb_buf = rec;
 469                                 rc = dt_record_write(env, o, &lgi->lgi_buf,
 470                                                      &lgi->lgi_off, th);
 471                                 RETURN(rc);
 472                         }
 473
 474                         /* update the header */
 475                         lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
 476                         lgi->lgi_buf.lb_buf = llh;
 477                         rc = dt_record_write(env, o, &lgi->lgi_buf,
 478                                              &lgi->lgi_off, th);
 479                         if (rc != 0)
 480                                 RETURN(rc);
 481
 482                         /* update the bitmap */
 483                         index = reccookie->lgc_index;
 484                         lgi->lgi_off = llh->llh_bitmap_offset +
 485                                       (index / (sizeof(*bitmap) * 8)) *
 486                                                         sizeof(*bitmap);
 487                         lgi->lgi_buf.lb_len = sizeof(*bitmap);
 488                         lgi->lgi_buf.lb_buf =
 489                                         &bitmap[index/(sizeof(*bitmap)*8)];
 490                         rc = dt_record_write(env, o, &lgi->lgi_buf,
 491                                              &lgi->lgi_off, th);
 492
 493                         RETURN(rc);
 494                 } else if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 495                         lgi->lgi_off = llh->llh_hdr.lrh_len +
 496                                        (idx - 1) * reclen;
 497                 } else if (reccookie != NULL && reccookie->lgc_index > 0) {
 498                         /**
 499                          * The lgc_offset can be used only if index is
 500                          * the same.
 501                          */
 502                         if (idx != reccookie->lgc_index) {
 503                                 CERROR("%s: modify index mismatch %d %d\n",
 504                                        o->do_lu.lo_dev->ld_obd->obd_name, idx,
 505                                        reccookie->lgc_index);
 506                                 RETURN(-EFAULT);
 507                         }
 508
 509                         lgi->lgi_off = reccookie->lgc_offset;
 510                         CDEBUG(D_OTHER, "modify record "DFID": idx:%u, "
 511                                "len:%u offset %llu\n",
 512                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid), idx,
 513                                rec->lrh_len, (long long)lgi->lgi_off);
 514                 } else {
 515                         /* This can be result of lgh_cur_idx is not set during
 516                          * llog processing or llh_size is not set to proper
 517                          * record size for fixed records llog. Therefore it is
 518                          * impossible to get record offset. */
 519                         CERROR("%s: can't get record offset, idx:%d, "
 520                                "len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
 521                                idx, rec->lrh_len);
 522                         RETURN(-EFAULT);
 523                 }
 524
 525                 /* update only data, header and tail remain the same */
 526                 lgi->lgi_off += sizeof(struct llog_rec_hdr);
 527                 lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
 528                 lgi->lgi_buf.lb_buf = REC_DATA(rec);
 529                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 530                 if (rc == 0 && reccookie) {
 531                         reccookie->lgc_lgl = loghandle->lgh_id;
 532                         reccookie->lgc_index = idx;
 533                         rc = 1;
 534                 }
 535                 RETURN(rc);
 536         }
 537
 538         /**
 539          * The append case.
 540          * The most common case of using llog. The new index is assigned to
 541          * the new record, new bit is set in llog bitmap and llog count is
 542          * incremented.
 543          *
 544          * Make sure that records don't cross a chunk boundary, so we can
 545          * process them page-at-a-time if needed.  If it will cross a chunk
 546          * boundary, write in a fake (but referenced) entry to pad the chunk.
 547          */
 548
 549
 550         /* simulate ENOSPC when new plain llog is being added to the
 551          * catalog */
 552         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2) &&
 553             llh->llh_flags & LLOG_F_IS_CAT)
 554                 RETURN(-ENOSPC);
 555
 556         LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 557         orig_last_idx = loghandle->lgh_last_idx;
 558         lgi->lgi_off = lgi->lgi_attr.la_size;
 559
 560         if (loghandle->lgh_max_size > 0 &&
 561             lgi->lgi_off >= loghandle->lgh_max_size) {
 562                 CDEBUG(D_OTHER, "llog is getting too large (%u > %u) at %u "
 563                        DFID"\n", (unsigned)lgi->lgi_off,
 564                        loghandle->lgh_max_size, (int)loghandle->lgh_last_idx,
 565                        PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
 566                 /* this is to signal that this llog is full */
 567                 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 568                 RETURN(-ENOSPC);
 569         }
 570
 571         left = chunk_size - (lgi->lgi_off & (chunk_size - 1));
 572         /* NOTE: padding is a record, but no bit is set */
 573         if (left != 0 && left != reclen &&
 574             left < (reclen + LLOG_MIN_REC_SIZE)) {
 575                 index = loghandle->lgh_last_idx + 1;
 576                 rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
 577                 if (rc)
 578                         RETURN(rc);
 579
 580                 loghandle->lgh_last_idx++; /* for pad rec */
 581         }
 582         /* if it's the last idx in log file, then return -ENOSPC
 583          * or wrap around if a catalog */
 584         if (llog_is_full(loghandle) ||
 585             unlikely(llh->llh_flags & LLOG_F_IS_CAT &&
 586                      OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) &&
 587                      loghandle->lgh_last_idx >= cfs_fail_val)) {
 588                 if (llh->llh_flags & LLOG_F_IS_CAT)
 589                         loghandle->lgh_last_idx = 0;
 590                 else
 591                         RETURN(-ENOSPC);
 592         }
 593
 594         down_write(&loghandle->lgh_last_sem);
 595         /* increment the last_idx along with llh_tail index, they should
 596          * be equal for a llog lifetime */
 597         loghandle->lgh_last_idx++;
 598         index = loghandle->lgh_last_idx;
 599         LLOG_HDR_TAIL(llh)->lrt_index = index;
 600         /**
 601          * NB: the caller should make sure only 1 process access
 602          * the lgh_last_idx, e.g. append should be exclusive.
 603          * Otherwise it might hit the assert.
 604          */
 605         LASSERT(index < LLOG_HDR_BITMAP_SIZE(llh));
 606         rec->lrh_index = index;
 607         lrt = rec_tail(rec);
 608         lrt->lrt_len = rec->lrh_len;
 609         lrt->lrt_index = rec->lrh_index;
 610
 611         /* the lgh_hdr_mutex protects llog header data from concurrent
 612          * update/cancel, the llh_count and llh_bitmap are protected */
 613         mutex_lock(&loghandle->lgh_hdr_mutex);
 614         if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
 615                 CERROR("%s: index %u already set in log bitmap\n",
 616                        o->do_lu.lo_dev->ld_obd->obd_name, index);
 617                 mutex_unlock(&loghandle->lgh_hdr_mutex);
 618                 LBUG(); /* should never happen */
 619         }
 620         llh->llh_count++;
 621
 622         if (!(llh->llh_flags & LLOG_F_IS_FIXSIZE)) {
 623                 /* Update the minimum size of the llog record */
 624                 if (llh->llh_size == 0)
 625                         llh->llh_size = reclen;
 626                 else if (reclen < llh->llh_size)
 627                         llh->llh_size = reclen;
 628         }
 629
 630         if (lgi->lgi_attr.la_size == 0) {
 631                 lgi->lgi_off = 0;
 632                 lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
 633                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 634                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 635                 if (rc != 0)
 636                         GOTO(out_unlock, rc);
 637         } else {
 638                 __u32   *bitmap = LLOG_HDR_BITMAP(llh);
 639
 640                 /* Note: If this is not initialization (size == 0), then do not
 641                  * write the whole header (8k bytes), only update header/tail
 642                  * and bits needs to be updated. Because this update might be
 643                  * part of cross-MDT operation, which needs to write these
 644                  * updates into the update log(32KB limit) and also pack inside
 645                  * the RPC (1MB limit), if we write 8K for each operation, which
 646                  * will cost a lot space, and keep us adding more updates to one
 647                  * update log.*/
 648                 lgi->lgi_off = 0;
 649                 lgi->lgi_buf.lb_len = llh->llh_bitmap_offset;
 650                 lgi->lgi_buf.lb_buf = &llh->llh_hdr;
 651                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 652                 if (rc != 0)
 653                         GOTO(out_unlock, rc);
 654
 655                 lgi->lgi_off = llh->llh_bitmap_offset +
 656                               (index / (sizeof(*bitmap) * 8)) * sizeof(*bitmap);
 657                 lgi->lgi_buf.lb_len = sizeof(*bitmap);
 658                 lgi->lgi_buf.lb_buf = &bitmap[index/(sizeof(*bitmap)*8)];
 659                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 660                 if (rc != 0)
 661                         GOTO(out_unlock, rc);
 662
 663                 lgi->lgi_off =  (unsigned long)LLOG_HDR_TAIL(llh) -
 664                                 (unsigned long)llh;
 665                 lgi->lgi_buf.lb_len = sizeof(llh->llh_tail);
 666                 lgi->lgi_buf.lb_buf = LLOG_HDR_TAIL(llh);
 667                 rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 668                 if (rc != 0)
 669                         GOTO(out_unlock, rc);
 670         }
 671
 672 out_unlock:
 673         /* unlock here for remote object */
 674         mutex_unlock(&loghandle->lgh_hdr_mutex);
 675         if (rc)
 676                 GOTO(out, rc);
 677
 678         if (OBD_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
 679            cfs_fail_val == (unsigned int)(loghandle->lgh_id.lgl_oi.oi.oi_id &
 680                                           0xFFFFFFFF)) {
 681                 OBD_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
 682                 msleep(1 * MSEC_PER_SEC);
 683         }
 684         /* computed index can be used to determine offset for fixed-size
 685          * records. This also allows to handle Catalog wrap around case */
 686         if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 687                 lgi->lgi_off = llh->llh_hdr.lrh_len + (index - 1) * reclen;
 688         } else {
 689                 rc = dt_attr_get(env, o, &lgi->lgi_attr);
 690                 if (rc)
 691                         GOTO(out, rc);
 692
 693                 LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
 694                 lgi->lgi_off = max_t(__u64, lgi->lgi_attr.la_size,
 695                                      lgi->lgi_off);
 696         }
 697
 698         lgi->lgi_buf.lb_len = reclen;
 699         lgi->lgi_buf.lb_buf = rec;
 700         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
 701         if (rc < 0)
 702                 GOTO(out, rc);
 703
 704         up_write(&loghandle->lgh_last_sem);
 705
 706         CDEBUG(D_HA, "added record "DFID".%u, %u off%llu\n",
 707                PFID(lu_object_fid(&o->do_lu)), index, rec->lrh_len,
 708                lgi->lgi_off);
 709         if (reccookie != NULL) {
 710                 reccookie->lgc_lgl = loghandle->lgh_id;
 711                 reccookie->lgc_index = index;
 712                 if ((rec->lrh_type == MDS_UNLINK_REC) ||
 713                     (rec->lrh_type == MDS_SETATTR64_REC))
 714                         reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
 715                 else if (rec->lrh_type == OST_SZ_REC)
 716                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
 717                 else
 718                         reccookie->lgc_subsys = -1;
 719                 rc = 1;
 720         }
 721         RETURN(rc);
 722 out:
 723         /* cleanup llog for error case */
 724         mutex_lock(&loghandle->lgh_hdr_mutex);
 725         ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
 726         llh->llh_count--;
 727         mutex_unlock(&loghandle->lgh_hdr_mutex);
 728
 729         /* restore llog last_idx */
 730         if (dt_object_remote(o)) {
 731                 loghandle->lgh_last_idx = orig_last_idx;
 732         } else if (--loghandle->lgh_last_idx == 0 &&
 733             (llh->llh_flags & LLOG_F_IS_CAT) && llh->llh_cat_idx != 0) {
 734                 /* catalog had just wrap-around case */
 735                 loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) - 1;
 736         }
 737
 738         LLOG_HDR_TAIL(llh)->lrt_index = loghandle->lgh_last_idx;
 739         up_write(&loghandle->lgh_last_sem);
 740
 741         RETURN(rc);
 742 }
 743
 744 /**
 745  * We can skip reading at least as many log blocks as the number of
 746  * minimum sized log records we are skipping.  If it turns out
 747  * that we are not far enough along the log (because the
 748  * actual records are larger than minimum size) we just skip
 749  * some more records.
 750  *
 751  * Note: in llog_process_thread, it will use bitmap offset as
 752  * the index to locate the record, which also includs some pad
 753  * records, whose record size is very small, and it also does not
 754  * consider pad record when recording minimum record size (otherwise
 755  * min_record size might be too small), so in some rare cases,
 756  * it might skip too much record for @goal, see llog_osd_next_block().
 757  *
 758  * When force_mini_rec is true, it means we have to use LLOG_MIN_REC_SIZE
 759  * as the min record size to skip over, usually because in the previous
 760  * try, it skip too much record, see loog_osd_next(prev)_block().
 761  */
 762 static inline void llog_skip_over(struct llog_handle *lgh, __u64 *off,
 763                                   int curr, int goal, __u32 chunk_size,
 764                                   bool force_mini_rec)
 765 {
 766         struct llog_log_hdr *llh = lgh->lgh_hdr;
 767
 768         /* Goal should not bigger than the record count */
 769         if (goal > lgh->lgh_last_idx)
 770                 goal = lgh->lgh_last_idx;
 771
 772         if (goal > curr) {
 773                 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 774                         *off = chunk_size + (goal - 1) * llh->llh_size;
 775                 } else {
 776                         __u64 min_rec_size = LLOG_MIN_REC_SIZE;
 777
 778                         if (llh->llh_size > 0 && !force_mini_rec)
 779                                 min_rec_size = llh->llh_size;
 780
 781                         *off = *off + (goal - curr - 1) * min_rec_size;
 782                 }
 783         }
 784         /* always align with lower chunk boundary*/
 785         *off &= ~(chunk_size - 1);
 786 }
 787
 788 /**
 789  * Remove optional fields that the client doesn't expect.
 790  * This is typically in order to ensure compatibility with older clients.
 791  * It is assumed that since we exclusively remove fields, the block will be
 792  * big enough to handle the remapped records. It is also assumed that records
 793  * of a block have the same format (i.e.: the same features enabled).
 794  *
 795  * \param[in,out]    hdr           Header of the block of records to remap.
 796  * \param[in,out]    last_hdr      Last header, don't read past this point.
 797  * \param[in]        flags         Flags describing the fields to keep.
 798  * \param[in]        extra_flags   Flags describing the extra fields to keep.
 799  */
 800 static void changelog_block_trim_ext(struct llog_rec_hdr *hdr,
 801                                      struct llog_rec_hdr *last_hdr,
 802                                      struct llog_handle *loghandle)
 803 {
 804         enum changelog_rec_flags flags = CLF_SUPPORTED;
 805         enum changelog_rec_extra_flags extra_flags = CLFE_SUPPORTED;
 806
 807         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_XATTR))
 808                 extra_flags &= ~CLFE_XATTR;
 809         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_OMODE))
 810                 extra_flags &= ~CLFE_OPEN;
 811         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_NID))
 812                 extra_flags &= ~CLFE_NID;
 813         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_X_UIDGID))
 814                 extra_flags &= ~CLFE_UIDGID;
 815         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_EXTRA_FLAGS))
 816                 flags &= ~CLF_EXTRA_FLAGS;
 817         if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
 818                 flags &= ~CLF_JOBID;
 819
 820         if (flags == CLF_SUPPORTED && extra_flags == CLFE_SUPPORTED)
 821                 return;
 822
 823         if (hdr->lrh_type != CHANGELOG_REC)
 824                 return;
 825
 826         do {
 827                 struct changelog_rec *rec = (struct changelog_rec *)(hdr + 1);
 828                 enum changelog_rec_extra_flags xflag = CLFE_INVALID;
 829
 830                 if (flags & CLF_EXTRA_FLAGS &&
 831                     rec->cr_flags & CLF_EXTRA_FLAGS) {
 832                         xflag = changelog_rec_extra_flags(rec)->cr_extra_flags &
 833                                 extra_flags;
 834                 }
 835
 836                 if (unlikely(hdr->lrh_len == 0)) {
 837                         /* It is corruption case, we cannot know the next rec,
 838                          * jump to the last one directly to avoid dead loop. */
 839                         LCONSOLE(D_WARNING, "Hit invalid llog record: "
 840                                  "idx %u, type %u, id %u\n",
 841                                  hdr->lrh_index, hdr->lrh_type, hdr->lrh_id);
 842                         hdr = llog_rec_hdr_next(last_hdr);
 843                         if (unlikely(hdr == last_hdr))
 844                                 LCONSOLE(D_WARNING, "The last record crashed: "
 845                                          "idx %u, type %u, id %u\n",
 846                                          hdr->lrh_index, hdr->lrh_type,
 847                                          hdr->lrh_id);
 848                         break;
 849                 }
 850
 851                 changelog_remap_rec(rec, rec->cr_flags & flags, xflag);
 852                 hdr = llog_rec_hdr_next(hdr);
 853                 /* Yield CPU to avoid soft-lockup if there are too many records
 854                  * to be handled. */
 855                 cond_resched();
 856         } while ((char *)hdr <= (char *)last_hdr);
 857 }
 858
 859 /**
 860  * Implementation of the llog_operations::lop_next_block
 861  *
 862  * This function finds the the next llog block to return which contains
 863  * record with required index. It is main part of llog processing.
 864  *
 865  * \param[in]     env           execution environment
 866  * \param[in]     loghandle     llog handle of the current llog
 867  * \param[in,out] cur_idx       index preceeding cur_offset
 868  * \param[in]     next_idx      target index to find
 869  * \param[in,out] cur_offset    furtherst point read in the file
 870  * \param[in]     buf           pointer to data buffer to fill
 871  * \param[in]     len           required len to read, it is
 872  *                              usually llog chunk_size.
 873  *
 874  * \retval                      0 on successful buffer read
 875  * \retval                      negative value on error
 876  */
 877 static int llog_osd_next_block(const struct lu_env *env,
 878                                struct llog_handle *loghandle, int *cur_idx,
 879                                int next_idx, __u64 *cur_offset, void *buf,
 880                                int len)
 881 {
 882         struct llog_thread_info *lgi = llog_info(env);
 883         struct dt_object        *o;
 884         struct dt_device        *dt;
 885         int                      rc;
 886         __u32                   chunk_size;
 887         int last_idx = *cur_idx;
 888         __u64 last_offset = *cur_offset;
 889         bool force_mini_rec = false;
 890
 891         ENTRY;
 892
 893         LASSERT(env);
 894         LASSERT(lgi);
 895
 896         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
 897         if (len == 0 || len & (chunk_size - 1))
 898                 RETURN(-EINVAL);
 899
 900         LASSERT(loghandle);
 901         LASSERT(loghandle->lgh_ctxt);
 902
 903         o = loghandle->lgh_obj;
 904         LASSERT(o);
 905         LASSERT(llog_osd_exist(loghandle));
 906         dt = lu2dt_dev(o->do_lu.lo_dev);
 907         LASSERT(dt);
 908
 909         rc = dt_attr_get(env, o, &lgi->lgi_attr);
 910         if (rc)
 911                 GOTO(out, rc);
 912
 913         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
 914                "%llu), size %llu\n", next_idx, *cur_idx,
 915                *cur_offset, lgi->lgi_attr.la_size);
 916
 917         while (*cur_offset < lgi->lgi_attr.la_size) {
 918                 struct llog_rec_hdr     *rec, *last_rec;
 919                 struct llog_rec_tail    *tail;
 920
 921                 llog_skip_over(loghandle, cur_offset, *cur_idx,
 922                                next_idx, chunk_size, force_mini_rec);
 923
 924                 /* read up to next llog chunk_size block */
 925                 lgi->lgi_buf.lb_len = chunk_size -
 926                                       (*cur_offset & (chunk_size - 1));
 927                 lgi->lgi_buf.lb_buf = buf;
 928
 929                 rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
 930                 if (rc < 0) {
 931                         if (rc == -EBADR && !force_mini_rec)
 932                                 goto retry;
 933
 934                         CERROR("%s: can't read llog block from log "DFID
 935                                " offset %llu: rc = %d\n",
 936                                o->do_lu.lo_dev->ld_obd->obd_name,
 937                                PFID(lu_object_fid(&o->do_lu)), *cur_offset,
 938                                rc);
 939                         GOTO(out, rc);
 940                 }
 941
 942                 if (rc < len) {
 943                         /* signal the end of the valid buffer to
 944                          * llog_process */
 945                         memset(buf + rc, 0, len - rc);
 946                 }
 947
 948                 if (rc == 0) { /* end of file, nothing to do */
 949                         if (!force_mini_rec)
 950                                 goto retry;
 951                         GOTO(out, rc);
 952                 }
 953
 954                 if (rc < sizeof(*tail)) {
 955                         if (!force_mini_rec)
 956                                 goto retry;
 957
 958                         CERROR("%s: invalid llog block at log id "DFID":%x "
 959                                "offset %llu\n",
 960                                o->do_lu.lo_dev->ld_obd->obd_name,
 961                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
 962                                loghandle->lgh_id.lgl_ogen, *cur_offset);
 963                         GOTO(out, rc = -EINVAL);
 964                 }
 965
 966                 rec = buf;
 967                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 968                         lustre_swab_llog_rec(rec);
 969                 tail = (struct llog_rec_tail *)((char *)buf + rc -
 970                                                 sizeof(struct llog_rec_tail));
 971
 972                 if (llog_verify_record(loghandle, rec)) {
 973                         /*
 974                          * the block seems corrupted. make a pad record so the
 975                          * caller can skip the block and try with the next one
 976                          */
 977                         rec->lrh_len = rc;
 978                         rec->lrh_index = next_idx;
 979                         rec->lrh_type = LLOG_PAD_MAGIC;
 980
 981                         tail = rec_tail(rec);
 982                         tail->lrt_len = rc;
 983                         tail->lrt_index = next_idx;
 984
 985                         GOTO(out, rc = 0);
 986                 }
 987
 988                 /* get the last record in block */
 989                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
 990                                                    tail->lrt_len);
 991
 992                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
 993                         lustre_swab_llog_rec(last_rec);
 994
 995                 if (last_rec->lrh_index != tail->lrt_index) {
 996                         CERROR("%s: invalid llog tail at log id "DFID":%x "
 997                                "offset %llu last_rec idx %u tail idx %u"
 998                                "lrt len %u read_size %d\n",
 999                                o->do_lu.lo_dev->ld_obd->obd_name,
1000                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1001                                loghandle->lgh_id.lgl_ogen, *cur_offset,
1002                                last_rec->lrh_index, tail->lrt_index,
1003                                tail->lrt_len, rc);
1004                         GOTO(out, rc = -EINVAL);
1005                 }
1006
1007                 *cur_idx = tail->lrt_index;
1008
1009                 /* this shouldn't happen */
1010                 if (tail->lrt_index == 0) {
1011                         CERROR("%s: invalid llog tail at log id "DFID":%x "
1012                                "offset %llu bytes %d\n",
1013                                o->do_lu.lo_dev->ld_obd->obd_name,
1014                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1015                                loghandle->lgh_id.lgl_ogen, *cur_offset, rc);
1016                         GOTO(out, rc = -EINVAL);
1017                 }
1018                 if (tail->lrt_index < next_idx) {
1019                         last_idx = *cur_idx;
1020                         last_offset = *cur_offset;
1021                         continue;
1022                 }
1023
1024                 /* sanity check that the start of the new buffer is no farther
1025                  * than the record that we wanted.  This shouldn't happen. */
1026                 if (next_idx && rec->lrh_index > next_idx) {
1027                         if (!force_mini_rec && next_idx > last_idx)
1028                                 goto retry;
1029
1030                         CERROR("%s: missed desired record? %u > %u\n",
1031                                o->do_lu.lo_dev->ld_obd->obd_name,
1032                                rec->lrh_index, next_idx);
1033                         GOTO(out, rc = -ENOENT);
1034                 }
1035
1036                 /* Trim unsupported extensions for compat w/ older clients */
1037                 changelog_block_trim_ext(rec, last_rec, loghandle);
1038
1039                 GOTO(out, rc = 0);
1040
1041 retry:
1042                 /* Note: because there are some pad records in the
1043                  * llog, so llog_skip_over() might skip too much
1044                  * records, let's try skip again with minimum record */
1045                 force_mini_rec = true;
1046                 *cur_offset = last_offset;
1047                 *cur_idx = last_idx;
1048         }
1049         GOTO(out, rc = -EIO);
1050 out:
1051         return rc;
1052 }
1053
1054 /**
1055  * Implementation of the llog_operations::lop_prev_block
1056  *
1057  * This function finds the llog block to return which contains
1058  * record with required index but in reverse order - from end of llog
1059  * to the beginning.
1060  * It is main part of reverse llog processing.
1061  *
1062  * \param[in] env       execution environment
1063  * \param[in] loghandle llog handle of the current llog
1064  * \param[in] prev_idx  target index to find
1065  * \param[in] buf       pointer to data buffer to fill
1066  * \param[in] len       required len to read, it is llog_chunk_size usually.
1067  *
1068  * \retval              0 on successful buffer read
1069  * \retval              negative value on error
1070  */
1071 static int llog_osd_prev_block(const struct lu_env *env,
1072                                struct llog_handle *loghandle,
1073                                int prev_idx, void *buf, int len)
1074 {
1075         struct llog_thread_info *lgi = llog_info(env);
1076         struct dt_object        *o;
1077         struct dt_device        *dt;
1078         loff_t                   cur_offset;
1079         __u32                   chunk_size;
1080         int                      rc;
1081
1082         ENTRY;
1083
1084         chunk_size = loghandle->lgh_hdr->llh_hdr.lrh_len;
1085         if (len == 0 || len & (chunk_size - 1))
1086                 RETURN(-EINVAL);
1087
1088         CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
1089
1090         LASSERT(loghandle);
1091         LASSERT(loghandle->lgh_ctxt);
1092
1093         o = loghandle->lgh_obj;
1094         LASSERT(o);
1095         LASSERT(llog_osd_exist(loghandle));
1096         dt = lu2dt_dev(o->do_lu.lo_dev);
1097         LASSERT(dt);
1098
1099         /* Let's only use mini record size for previous block read
1100          * for now XXX */
1101         cur_offset = chunk_size;
1102         llog_skip_over(loghandle, &cur_offset, 0, prev_idx,
1103                        chunk_size, true);
1104
1105         rc = dt_attr_get(env, o, &lgi->lgi_attr);
1106         if (rc)
1107                 GOTO(out, rc);
1108
1109         while (cur_offset < lgi->lgi_attr.la_size) {
1110                 struct llog_rec_hdr     *rec, *last_rec;
1111                 struct llog_rec_tail    *tail;
1112
1113                 lgi->lgi_buf.lb_len = len;
1114                 lgi->lgi_buf.lb_buf = buf;
1115                 rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
1116                 if (rc < 0) {
1117                         CERROR("%s: can't read llog block from log "DFID
1118                                " offset %llu: rc = %d\n",
1119                                o->do_lu.lo_dev->ld_obd->obd_name,
1120                                PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
1121                         GOTO(out, rc);
1122                 }
1123
1124                 if (rc == 0) /* end of file, nothing to do */
1125                         GOTO(out, rc);
1126
1127                 if (rc < sizeof(*tail)) {
1128                         CERROR("%s: invalid llog block at log id "DFID":%x "
1129                                "offset %llu\n",
1130                                o->do_lu.lo_dev->ld_obd->obd_name,
1131                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1132                                loghandle->lgh_id.lgl_ogen, cur_offset);
1133                         GOTO(out, rc = -EINVAL);
1134                 }
1135
1136                 rec = buf;
1137                 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1138                         lustre_swab_llog_rec(rec);
1139
1140                 tail = (struct llog_rec_tail *)((char *)buf + rc -
1141                                                 sizeof(struct llog_rec_tail));
1142                 /* get the last record in block */
1143                 last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
1144                                                    le32_to_cpu(tail->lrt_len));
1145
1146                 if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
1147                         lustre_swab_llog_rec(last_rec);
1148                 LASSERT(last_rec->lrh_index == tail->lrt_index);
1149
1150                 /* this shouldn't happen */
1151                 if (tail->lrt_index == 0) {
1152                         CERROR("%s: invalid llog tail at log id "DFID":%x "
1153                                "offset %llu\n",
1154                                o->do_lu.lo_dev->ld_obd->obd_name,
1155                                PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
1156                                loghandle->lgh_id.lgl_ogen, cur_offset);
1157                         GOTO(out, rc = -EINVAL);
1158                 }
1159                 if (tail->lrt_index < prev_idx)
1160                         continue;
1161
1162                 /* sanity check that the start of the new buffer is no farther
1163                  * than the record that we wanted.  This shouldn't happen. */
1164                 if (rec->lrh_index > prev_idx) {
1165                         CERROR("%s: missed desired record? %u > %u\n",
1166                                o->do_lu.lo_dev->ld_obd->obd_name,
1167                                rec->lrh_index, prev_idx);
1168                         GOTO(out, rc = -ENOENT);
1169                 }
1170
1171                 /* Trim unsupported extensions for compat w/ older clients */
1172                 changelog_block_trim_ext(rec, last_rec, loghandle);
1173
1174                 GOTO(out, rc = 0);
1175         }
1176         GOTO(out, rc = -EIO);
1177 out:
1178         return rc;
1179 }
1180
1181 /**
1182  * This is helper function to get llog directory object. It is used by named
1183  * llog operations to find/insert/delete llog entry from llog directory.
1184  *
1185  * \param[in] env       execution environment
1186  * \param[in] ctxt      llog context
1187  *
1188  * \retval              dt_object of llog directory
1189  * \retval              ERR_PTR of negative value on error
1190  */
1191 static struct dt_object *llog_osd_dir_get(const struct lu_env *env,
1192                                           struct llog_ctxt *ctxt)
1193 {
1194         struct dt_device        *dt;
1195         struct dt_thread_info   *dti = dt_info(env);
1196         struct dt_object        *dir;
1197         int                      rc;
1198
1199         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1200         if (ctxt->loc_dir == NULL) {
1201                 rc = dt_root_get(env, dt, &dti->dti_fid);
1202                 if (rc)
1203                         return ERR_PTR(rc);
1204                 dir = dt_locate(env, dt, &dti->dti_fid);
1205
1206                 if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) {
1207                         dt_object_put(env, dir);
1208                         return ERR_PTR(-ENOTDIR);
1209                 }
1210         } else {
1211                 lu_object_get(&ctxt->loc_dir->do_lu);
1212                 dir = ctxt->loc_dir;
1213         }
1214
1215         return dir;
1216 }
1217
1218 /**
1219  * Implementation of the llog_operations::lop_open
1220  *
1221  * This function opens the llog by its logid or by name, it may open also
1222  * non existent llog and assing then new id to it.
1223  * The llog_open/llog_close pair works similar to lu_object_find/put,
1224  * the object may not exist prior open. The result of open is just dt_object
1225  * in the llog header.
1226  *
1227  * \param[in] env               execution environment
1228  * \param[in] handle            llog handle of the current llog
1229  * \param[in] logid             logid of llog to open (nameless llog)
1230  * \param[in] name              name of llog to open (named llog)
1231  * \param[in] open_param
1232  *                              LLOG_OPEN_NEW - new llog, may not exist
1233  *                              LLOG_OPEN_EXIST - old llog, must exist
1234  *
1235  * \retval                      0 on successful open, llog_handle::lgh_obj
1236  *                              contains the dt_object of the llog.
1237  * \retval                      negative value on error
1238  */
1239 static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
1240                          struct llog_logid *logid, char *name,
1241                          enum llog_open_param open_param)
1242 {
1243         struct llog_thread_info         *lgi = llog_info(env);
1244         struct llog_ctxt                *ctxt = handle->lgh_ctxt;
1245         struct dt_object                *o;
1246         struct dt_device                *dt;
1247         struct ls_device                *ls;
1248         struct local_oid_storage        *los = NULL;
1249         int                              rc = 0;
1250         bool new_id = false;
1251
1252         ENTRY;
1253
1254         LASSERT(env);
1255         LASSERT(ctxt);
1256         LASSERT(ctxt->loc_exp);
1257         LASSERT(ctxt->loc_exp->exp_obd);
1258         dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
1259         LASSERT(dt);
1260         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1261                 struct lu_object_conf conf = { 0 };
1262                 if (logid != NULL) {
1263                         logid_to_fid(logid, &lgi->lgi_fid);
1264                 } else {
1265                         /* If logid == NULL, then it means the caller needs
1266                          * to allocate new FID (llog_cat_declare_add_rec()). */
1267                         rc = obd_fid_alloc(env, ctxt->loc_exp,
1268                                            &lgi->lgi_fid, NULL);
1269                         if (rc < 0)
1270                                 RETURN(rc);
1271                         rc = 0;
1272                         conf.loc_flags = LOC_F_NEW;
1273                 }
1274
1275                 o = dt_locate_at(env, dt, &lgi->lgi_fid,
1276                                  dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1277                 if (IS_ERR(o))
1278                         RETURN(PTR_ERR(o));
1279
1280                 goto after_open;
1281         }
1282
1283         ls = ls_device_get(dt);
1284         if (IS_ERR(ls))
1285                 RETURN(PTR_ERR(ls));
1286
1287         mutex_lock(&ls->ls_los_mutex);
1288         los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
1289         mutex_unlock(&ls->ls_los_mutex);
1290         LASSERT(los);
1291         ls_device_put(env, ls);
1292
1293         LASSERT(handle);
1294
1295         if (logid != NULL) {
1296                 logid_to_fid(logid, &lgi->lgi_fid);
1297         } else if (name) {
1298                 struct dt_object *llog_dir;
1299
1300                 llog_dir = llog_osd_dir_get(env, ctxt);
1301                 if (IS_ERR(llog_dir))
1302                         GOTO(out, rc = PTR_ERR(llog_dir));
1303                 dt_read_lock(env, llog_dir, 0);
1304                 rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
1305                 dt_read_unlock(env, llog_dir);
1306                 dt_object_put(env, llog_dir);
1307                 if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
1308                         /* generate fid for new llog */
1309                         rc = local_object_fid_generate(env, los,
1310                                                        &lgi->lgi_fid);
1311                         new_id = true;
1312                 }
1313                 if (rc < 0)
1314                         GOTO(out, rc);
1315                 OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
1316                 if (handle->lgh_name)
1317                         strcpy(handle->lgh_name, name);
1318                 else
1319                         GOTO(out, rc = -ENOMEM);
1320         } else {
1321                 LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
1322                 /* generate fid for new llog */
1323 generate:
1324                 rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
1325                 if (rc < 0)
1326                         GOTO(out, rc);
1327                 new_id = true;
1328         }
1329
1330         o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
1331         if (IS_ERR(o))
1332                 GOTO(out_name, rc = PTR_ERR(o));
1333
1334         if (dt_object_exists(o) && new_id) {
1335                 /* llog exists with just generated ID, e.g. some old llog file
1336                  * still is in use or is orphan, drop a warn and skip it. */
1337                 CDEBUG(D_INFO, "%s: llog exists with the same FID: "DFID
1338                        ", skipping\n",
1339                        o->do_lu.lo_dev->ld_obd->obd_name,
1340                        PFID(lu_object_fid(&o->do_lu)));
1341                 dt_object_put(env, o);
1342                 /* just skip this llog ID, we shouldn't delete it because we
1343                  * don't know exactly what is its purpose and state. */
1344                 goto generate;
1345         }
1346
1347 after_open:
1348         /* No new llog is expected but doesn't exist */
1349         if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o)) {
1350                 CDEBUG(D_INFO, "%s: llog FID: "DFID" obj %p doesn`t exist\n",
1351                        o->do_lu.lo_dev->ld_obd->obd_name,
1352                        PFID(lu_object_fid(&o->do_lu)), o);
1353                 GOTO(out_put, rc = -ENOENT);
1354         }
1355         fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
1356         handle->lgh_obj = o;
1357         handle->private_data = los;
1358         LASSERT(handle->lgh_ctxt);
1359
1360         RETURN(rc);
1361
1362 out_put:
1363         dt_object_put(env, o);
1364 out_name:
1365         if (handle->lgh_name != NULL)
1366                 OBD_FREE(handle->lgh_name, strlen(name) + 1);
1367 out:
1368         if (los != NULL)
1369                 dt_los_put(los);
1370         RETURN(rc);
1371 }
1372
1373 /**
1374  * Get dir for regular fid log object
1375  *
1376  * Get directory for regular fid log object, and these regular fid log
1377  * object will be inserted under this directory, to satisfy the FS
1378  * consistency check, e2fsck etc.
1379  *
1380  * \param [in] env      execution environment
1381  * \param [in] dto      llog object
1382  *
1383  * \retval              pointer to the directory if it is found.
1384  * \retval              ERR_PTR(negative errno) if it fails.
1385  */
1386 struct dt_object *llog_osd_get_regular_fid_dir(const struct lu_env *env,
1387                                                struct dt_object *dto)
1388 {
1389         struct llog_thread_info *lgi = llog_info(env);
1390         struct seq_server_site *ss = dto->do_lu.lo_dev->ld_site->ld_seq_site;
1391         struct lu_seq_range     *range = &lgi->lgi_range;
1392         struct lu_fid           *dir_fid = &lgi->lgi_fid;
1393         struct dt_object        *dir;
1394         int                     rc;
1395         ENTRY;
1396
1397         fld_range_set_any(range);
1398         LASSERT(ss != NULL);
1399         rc = ss->ss_server_fld->lsf_seq_lookup(env, ss->ss_server_fld,
1400                                    fid_seq(lu_object_fid(&dto->do_lu)), range);
1401         if (rc < 0)
1402                 RETURN(ERR_PTR(rc));
1403
1404         lu_update_log_dir_fid(dir_fid, range->lsr_index);
1405         dir = dt_locate(env, lu2dt_dev(dto->do_lu.lo_dev), dir_fid);
1406         if (IS_ERR(dir))
1407                 RETURN(dir);
1408
1409         if (!dt_try_as_dir(env, dir)) {
1410                 dt_object_put(env, dir);
1411                 RETURN(ERR_PTR(-ENOTDIR));
1412         }
1413
1414         RETURN(dir);
1415 }
1416
1417 /**
1418  * Add llog object with regular FID to name entry
1419  *
1420  * Add llog object with regular FID to name space, and each llog
1421  * object on each MDT will be /update_log_dir/[seq:oid:ver],
1422  * so to satisfy the namespace consistency check, e2fsck etc.
1423  *
1424  * \param [in] env      execution environment
1425  * \param [in] dto      llog object
1426  * \param [in] th       thandle
1427  * \param [in] declare  if it is declare or execution
1428  *
1429  * \retval              0 if insertion succeeds.
1430  * \retval              negative errno if insertion fails.
1431  */
1432 static int
1433 llog_osd_regular_fid_add_name_entry(const struct lu_env *env,
1434                                     struct dt_object *dto,
1435                                     struct thandle *th, bool declare)
1436 {
1437         struct llog_thread_info *lgi = llog_info(env);
1438         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1439         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1440         struct dt_object        *dir;
1441         char                    *name = lgi->lgi_name;
1442         int                     rc;
1443         ENTRY;
1444
1445         if (!fid_is_norm(fid))
1446                 RETURN(0);
1447
1448         dir = llog_osd_get_regular_fid_dir(env, dto);
1449         if (IS_ERR(dir))
1450                 RETURN(PTR_ERR(dir));
1451
1452         rec->rec_fid = fid;
1453         rec->rec_type = S_IFREG;
1454         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1455         dt_write_lock(env, dir, 0);
1456         if (declare) {
1457                 rc = dt_declare_insert(env, dir, (struct dt_rec *)rec,
1458                                (struct dt_key *)name, th);
1459         } else {
1460                 rc = dt_insert(env, dir, (struct dt_rec *)rec,
1461                                (struct dt_key *)name, th);
1462         }
1463         dt_write_unlock(env, dir);
1464
1465         dt_object_put(env, dir);
1466         RETURN(rc);
1467 }
1468
1469
1470 /**
1471  * Implementation of the llog_operations::lop_declare_create
1472  *
1473  * This function declares the llog create. It declares also name insert
1474  * into llog directory in case of named llog.
1475  *
1476  * \param[in] env       execution environment
1477  * \param[in] res       llog handle of the current llog
1478  * \param[in] th        current transaction handle
1479  *
1480  * \retval              0 on successful create declaration
1481  * \retval              negative value on error
1482  */
1483 static int llog_osd_declare_create(const struct lu_env *env,
1484                                    struct llog_handle *res, struct thandle *th)
1485 {
1486         struct llog_thread_info         *lgi = llog_info(env);
1487         struct dt_insert_rec            *rec = &lgi->lgi_dt_rec;
1488         struct local_oid_storage        *los;
1489         struct dt_object                *o;
1490         int                              rc;
1491
1492         ENTRY;
1493
1494         LASSERT(res->lgh_obj);
1495         LASSERT(th);
1496
1497         /* object can be created by another thread */
1498         o = res->lgh_obj;
1499         if (dt_object_exists(o))
1500                 RETURN(0);
1501
1502         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1503                 struct llog_thread_info *lgi = llog_info(env);
1504
1505                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE;
1506                 lgi->lgi_attr.la_size = 0;
1507                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1508                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1509
1510                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
1511                                        &lgi->lgi_dof, th);
1512                 if (rc < 0)
1513                         RETURN(rc);
1514
1515
1516                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, true);
1517
1518                 RETURN(rc);
1519         }
1520         los = res->private_data;
1521         LASSERT(los);
1522
1523         rc = llog_osd_declare_new_object(env, los, o, th);
1524         if (rc)
1525                 RETURN(rc);
1526
1527         /* do not declare header initialization here as it's declared
1528          * in llog_osd_declare_write_rec() which is always called */
1529
1530         if (res->lgh_name) {
1531                 struct dt_object *llog_dir;
1532
1533                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1534                 if (IS_ERR(llog_dir))
1535                         RETURN(PTR_ERR(llog_dir));
1536                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1537                 rec->rec_fid = &lgi->lgi_fid;
1538                 rec->rec_type = S_IFREG;
1539                 rc = dt_declare_insert(env, llog_dir,
1540                                        (struct dt_rec *)rec,
1541                                        (struct dt_key *)res->lgh_name, th);
1542                 dt_object_put(env, llog_dir);
1543                 if (rc)
1544                         CERROR("%s: can't declare named llog %s: rc = %d\n",
1545                                o->do_lu.lo_dev->ld_obd->obd_name,
1546                                res->lgh_name, rc);
1547         }
1548         RETURN(rc);
1549 }
1550
1551 /**
1552  * Implementation of the llog_operations::lop_create
1553  *
1554  * This function creates the llog according with llog_handle::lgh_obj
1555  * and llog_handle::lgh_name.
1556  *
1557  * \param[in] env       execution environment
1558  * \param[in] res       llog handle of the current llog
1559  * \param[in] th        current transaction handle
1560  *
1561  * \retval              0 on successful create
1562  * \retval              negative value on error
1563  */
1564 static int llog_osd_create(const struct lu_env *env, struct llog_handle *res,
1565                            struct thandle *th)
1566 {
1567         struct llog_thread_info *lgi = llog_info(env);
1568         struct dt_insert_rec    *rec = &lgi->lgi_dt_rec;
1569         struct local_oid_storage *los;
1570         struct dt_object        *o;
1571         int                      rc = 0;
1572
1573         ENTRY;
1574
1575         LASSERT(env);
1576         o = res->lgh_obj;
1577         LASSERT(o);
1578
1579         /* llog can be already created */
1580         if (dt_object_exists(o))
1581                 RETURN(-EEXIST);
1582
1583         if (res->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1584                 struct llog_thread_info *lgi = llog_info(env);
1585
1586                 lgi->lgi_attr.la_valid = LA_MODE | LA_SIZE | LA_TYPE;
1587                 lgi->lgi_attr.la_size = 0;
1588                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1589                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
1590
1591                 dt_write_lock(env, o, 0);
1592                 rc = dt_create(env, o, &lgi->lgi_attr, NULL,
1593                                &lgi->lgi_dof, th);
1594                 dt_write_unlock(env, o);
1595                 if (rc < 0)
1596                         RETURN(rc);
1597
1598                 rc = llog_osd_regular_fid_add_name_entry(env, o, th, false);
1599
1600                 RETURN(rc);
1601         }
1602
1603         los = res->private_data;
1604         LASSERT(los);
1605
1606         dt_write_lock(env, o, 0);
1607         if (!dt_object_exists(o))
1608                 rc = llog_osd_create_new_object(env, los, o, th);
1609         else
1610                 rc = -EEXIST;
1611
1612         dt_write_unlock(env, o);
1613         if (rc)
1614                 RETURN(rc);
1615
1616         if (res->lgh_name) {
1617                 struct dt_object *llog_dir;
1618
1619                 llog_dir = llog_osd_dir_get(env, res->lgh_ctxt);
1620                 if (IS_ERR(llog_dir))
1621                         RETURN(PTR_ERR(llog_dir));
1622
1623                 logid_to_fid(&res->lgh_id, &lgi->lgi_fid);
1624                 rec->rec_fid = &lgi->lgi_fid;
1625                 rec->rec_type = S_IFREG;
1626                 dt_read_lock(env, llog_dir, 0);
1627                 rc = dt_insert(env, llog_dir, (struct dt_rec *)rec,
1628                                (struct dt_key *)res->lgh_name, th);
1629                 dt_read_unlock(env, llog_dir);
1630                 dt_object_put(env, llog_dir);
1631                 if (rc)
1632                         CERROR("%s: can't create named llog %s: rc = %d\n",
1633                                o->do_lu.lo_dev->ld_obd->obd_name,
1634                                res->lgh_name, rc);
1635         }
1636         RETURN(rc);
1637 }
1638
1639 /**
1640  * Implementation of the llog_operations::lop_close
1641  *
1642  * This function closes the llog. It just put llog object and referenced
1643  * local storage.
1644  *
1645  * \param[in] env       execution environment
1646  * \param[in] handle    llog handle of the current llog
1647  *
1648  * \retval              0 on successful llog close
1649  * \retval              negative value on error
1650  */
1651 static int llog_osd_close(const struct lu_env *env, struct llog_handle *handle)
1652 {
1653         struct local_oid_storage        *los;
1654         int                              rc = 0;
1655
1656         ENTRY;
1657
1658         LASSERT(handle->lgh_obj);
1659
1660         if (handle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1661                 /* Remove the object from the cache, otherwise it may
1662                  * hold LOD being released during cleanup process */
1663                 dt_object_put_nocache(env, handle->lgh_obj);
1664                 LASSERT(handle->private_data == NULL);
1665                 RETURN(rc);
1666         } else {
1667                 dt_object_put(env, handle->lgh_obj);
1668         }
1669         los = handle->private_data;
1670         LASSERT(los);
1671         dt_los_put(los);
1672
1673         if (handle->lgh_name)
1674                 OBD_FREE(handle->lgh_name, strlen(handle->lgh_name) + 1);
1675
1676         RETURN(rc);
1677 }
1678
1679 /**
1680  * delete llog object name entry
1681  *
1682  * Delete llog object (with regular FID) from name space (under
1683  * update_log_dir).
1684  *
1685  * \param [in] env      execution environment
1686  * \param [in] dto      llog object
1687  * \param [in] th       thandle
1688  * \param [in] declare  if it is declare or execution
1689  *
1690  * \retval              0 if deletion succeeds.
1691  * \retval              negative errno if deletion fails.
1692  */
1693 static int
1694 llog_osd_regular_fid_del_name_entry(const struct lu_env *env,
1695                                     struct dt_object *dto,
1696                                     struct thandle *th, bool declare)
1697 {
1698         struct llog_thread_info *lgi = llog_info(env);
1699         const struct lu_fid     *fid = lu_object_fid(&dto->do_lu);
1700         struct dt_object        *dir;
1701         char                    *name = lgi->lgi_name;
1702         int                     rc;
1703         ENTRY;
1704
1705         if (!fid_is_norm(fid))
1706                 RETURN(0);
1707
1708         dir = llog_osd_get_regular_fid_dir(env, dto);
1709         if (IS_ERR(dir))
1710                 RETURN(PTR_ERR(dir));
1711
1712         snprintf(name, sizeof(lgi->lgi_name), DFID, PFID(fid));
1713         dt_write_lock(env, dir, 0);
1714         if (declare) {
1715                 rc = dt_declare_delete(env, dir, (struct dt_key *)name,
1716                                        th);
1717         } else {
1718                 rc = dt_delete(env, dir, (struct dt_key *)name, th);
1719         }
1720         dt_write_unlock(env, dir);
1721
1722         dt_object_put(env, dir);
1723         RETURN(rc);
1724 }
1725
1726 /**
1727  * Implementation of the llog_operations::lop_declare_destroy
1728  *
1729  * This function declare destroys the llog and deletes also entry in the
1730  * llog directory in case of named llog. Llog should be opened prior that.
1731  *
1732  * \param[in] env               execution environment
1733  * \param[in] loghandle llog handle of the current llog
1734  *
1735  * \retval              0 on successful destroy
1736  * \retval              negative value on error
1737  */
1738 static int llog_osd_declare_destroy(const struct lu_env *env,
1739                                     struct llog_handle *loghandle,
1740                                     struct thandle *th)
1741 {
1742         struct llog_ctxt        *ctxt;
1743         struct dt_object        *o, *llog_dir = NULL;
1744         int                      rc;
1745
1746         ENTRY;
1747
1748         ctxt = loghandle->lgh_ctxt;
1749         LASSERT(ctxt);
1750
1751         o = loghandle->lgh_obj;
1752         LASSERT(o);
1753
1754         if (loghandle->lgh_name) {
1755                 llog_dir = llog_osd_dir_get(env, ctxt);
1756                 if (IS_ERR(llog_dir))
1757                         RETURN(PTR_ERR(llog_dir));
1758
1759                 rc = dt_declare_delete(env, llog_dir,
1760                                        (struct dt_key *)loghandle->lgh_name,
1761                                        th);
1762                 if (rc < 0)
1763                         GOTO(out_put, rc);
1764         }
1765
1766         rc = dt_declare_ref_del(env, o, th);
1767         if (rc < 0)
1768                 GOTO(out_put, rc);
1769
1770         rc = dt_declare_destroy(env, o, th);
1771         if (rc < 0)
1772                 GOTO(out_put, rc);
1773
1774         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1775                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, true);
1776                 if (rc < 0)
1777                         GOTO(out_put, rc);
1778         }
1779
1780 out_put:
1781         if (!(IS_ERR_OR_NULL(llog_dir)))
1782                 dt_object_put(env, llog_dir);
1783
1784         RETURN(rc);
1785 }
1786
1787
1788 /**
1789  * Implementation of the llog_operations::lop_destroy
1790  *
1791  * This function destroys the llog and deletes also entry in the
1792  * llog directory in case of named llog. Llog should be opened prior that.
1793  * Destroy method is not part of external transaction and does everything
1794  * inside.
1795  *
1796  * \param[in] env               execution environment
1797  * \param[in] loghandle llog handle of the current llog
1798  *
1799  * \retval              0 on successful destroy
1800  * \retval              negative value on error
1801  */
1802 static int llog_osd_destroy(const struct lu_env *env,
1803                             struct llog_handle *loghandle, struct thandle *th)
1804 {
1805         struct llog_ctxt        *ctxt;
1806         struct dt_object        *o, *llog_dir = NULL;
1807         int                      rc;
1808
1809         ENTRY;
1810
1811         ctxt = loghandle->lgh_ctxt;
1812         LASSERT(ctxt != NULL);
1813
1814         o = loghandle->lgh_obj;
1815         LASSERT(o != NULL);
1816
1817         dt_write_lock(env, o, 0);
1818         if (!llog_osd_exist(loghandle))
1819                 GOTO(out_unlock, rc = 0);
1820
1821         if (loghandle->lgh_name) {
1822                 llog_dir = llog_osd_dir_get(env, ctxt);
1823                 if (IS_ERR(llog_dir))
1824                         GOTO(out_unlock, rc = PTR_ERR(llog_dir));
1825
1826                 dt_read_lock(env, llog_dir, 0);
1827                 rc = dt_delete(env, llog_dir,
1828                                (struct dt_key *)loghandle->lgh_name,
1829                                th);
1830                 dt_read_unlock(env, llog_dir);
1831                 if (rc) {
1832                         CERROR("%s: can't remove llog %s: rc = %d\n",
1833                                o->do_lu.lo_dev->ld_obd->obd_name,
1834                                loghandle->lgh_name, rc);
1835                         GOTO(out_unlock, rc);
1836                 }
1837         }
1838
1839         dt_ref_del(env, o, th);
1840         rc = dt_destroy(env, o, th);
1841         if (rc < 0)
1842                 GOTO(out_unlock, rc);
1843
1844         loghandle->lgh_destroyed = true;
1845         if (loghandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) {
1846                 rc = llog_osd_regular_fid_del_name_entry(env, o, th, false);
1847                 if (rc < 0)
1848                         GOTO(out_unlock, rc);
1849         }
1850
1851 out_unlock:
1852         dt_write_unlock(env, o);
1853         if (!(IS_ERR_OR_NULL(llog_dir)))
1854                 dt_object_put(env, llog_dir);
1855         RETURN(rc);
1856 }
1857
1858 /**
1859  * Implementation of the llog_operations::lop_setup
1860  *
1861  * This function setup the llog on local storage.
1862  *
1863  * \param[in] env       execution environment
1864  * \param[in] obd       obd device the llog belongs to
1865  * \param[in] olg       the llog group, it is always zero group now.
1866  * \param[in] ctxt_idx  the llog index, it defines the purpose of this llog.
1867  *                      Every new llog type have to use own index.
1868  * \param[in] disk_obd  the storage obd, where llog is stored.
1869  *
1870  * \retval              0 on successful llog setup
1871  * \retval              negative value on error
1872  */
1873 static int llog_osd_setup(const struct lu_env *env, struct obd_device *obd,
1874                           struct obd_llog_group *olg, int ctxt_idx,
1875                           struct obd_device *disk_obd)
1876 {
1877         struct llog_thread_info         *lgi = llog_info(env);
1878         struct llog_ctxt                *ctxt;
1879         int                              rc = 0;
1880         ENTRY;
1881
1882         LASSERT(obd);
1883         LASSERT(olg->olg_ctxts[ctxt_idx]);
1884
1885         ctxt = llog_ctxt_get(olg->olg_ctxts[ctxt_idx]);
1886         LASSERT(ctxt);
1887
1888         if (disk_obd == NULL)
1889                 GOTO(out, rc = 0);
1890
1891         /* initialize data allowing to generate new fids,
1892          * literally we need a sequece */
1893         lgi->lgi_fid.f_seq = FID_SEQ_LLOG;
1894         lgi->lgi_fid.f_oid = 1;
1895         lgi->lgi_fid.f_ver = 0;
1896         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1897                                     &lgi->lgi_fid,
1898                                     &ctxt->loc_los_nameless);
1899         if (rc != 0)
1900                 GOTO(out, rc);
1901
1902         lgi->lgi_fid.f_seq = FID_SEQ_LLOG_NAME;
1903         lgi->lgi_fid.f_oid = 1;
1904         lgi->lgi_fid.f_ver = 0;
1905         rc = local_oid_storage_init(env, disk_obd->obd_lvfs_ctxt.dt,
1906                                     &lgi->lgi_fid,
1907                                     &ctxt->loc_los_named);
1908         if (rc != 0) {
1909                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1910                 ctxt->loc_los_nameless = NULL;
1911         }
1912
1913         GOTO(out, rc);
1914
1915 out:
1916         llog_ctxt_put(ctxt);
1917         return rc;
1918 }
1919
1920 /**
1921  * Implementation of the llog_operations::lop_cleanup
1922  *
1923  * This function cleanups the llog on local storage.
1924  *
1925  * \param[in] env       execution environment
1926  * \param[in] ctxt      the llog context
1927  *
1928  * \retval              0
1929  */
1930 static int llog_osd_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
1931 {
1932         if (ctxt->loc_los_nameless != NULL) {
1933                 local_oid_storage_fini(env, ctxt->loc_los_nameless);
1934                 ctxt->loc_los_nameless = NULL;
1935         }
1936
1937         if (ctxt->loc_los_named != NULL) {
1938                 local_oid_storage_fini(env, ctxt->loc_los_named);
1939                 ctxt->loc_los_named = NULL;
1940         }
1941
1942         return 0;
1943 }
1944
1945 struct llog_operations llog_osd_ops = {
1946         .lop_next_block         = llog_osd_next_block,
1947         .lop_prev_block         = llog_osd_prev_block,
1948         .lop_read_header        = llog_osd_read_header,
1949         .lop_declare_destroy    = llog_osd_declare_destroy,
1950         .lop_destroy            = llog_osd_destroy,
1951         .lop_setup              = llog_osd_setup,
1952         .lop_cleanup            = llog_osd_cleanup,
1953         .lop_open               = llog_osd_open,
1954         .lop_exist              = llog_osd_exist,
1955         .lop_declare_create     = llog_osd_declare_create,
1956         .lop_create             = llog_osd_create,
1957         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1958         .lop_write_rec          = llog_osd_write_rec,
1959         .lop_close              = llog_osd_close,
1960 };
1961 EXPORT_SYMBOL(llog_osd_ops);
1962
1963 struct llog_operations llog_common_cat_ops = {
1964         .lop_next_block         = llog_osd_next_block,
1965         .lop_prev_block         = llog_osd_prev_block,
1966         .lop_read_header        = llog_osd_read_header,
1967         .lop_declare_destroy    = llog_osd_declare_destroy,
1968         .lop_destroy            = llog_osd_destroy,
1969         .lop_setup              = llog_osd_setup,
1970         .lop_cleanup            = llog_osd_cleanup,
1971         .lop_open               = llog_osd_open,
1972         .lop_exist              = llog_osd_exist,
1973         .lop_declare_create     = llog_osd_declare_create,
1974         .lop_create             = llog_osd_create,
1975         .lop_declare_write_rec  = llog_osd_declare_write_rec,
1976         .lop_write_rec          = llog_osd_write_rec,
1977         .lop_close              = llog_osd_close,
1978         .lop_add                = llog_cat_add_rec,
1979         .lop_declare_add        = llog_cat_declare_add_rec,
1980 };
1981 EXPORT_SYMBOL(llog_common_cat_ops);
1982
1983 /**
1984  * Read the special file which contains the list of llog catalogs IDs
1985  *
1986  * This function reads the CATALOGS file which contains the array of llog
1987  * catalogs IDs. The main purpose of this file is to store OSP llogs indexed
1988  * by OST/MDT number.
1989  *
1990  * \param[in]  env              execution environment
1991  * \param[in]  d                corresponding storage device
1992  * \param[in]  idx              position to start from, usually OST/MDT index
1993  * \param[in]  count            how many catalog IDs to read
1994  * \param[out] idarray          the buffer for the data. If it is NULL then
1995  *                              function returns just number of catalog IDs
1996  *                              in the file.
1997  * \param[in]  fid              LLOG_CATALOGS_OID for CATALOG object
1998  *
1999  * \retval                      0 on successful read of catalog IDs
2000  * \retval                      negative value on error
2001  * \retval                      positive value which is number of records in
2002  *                              the file if \a idarray is NULL
2003  */
2004 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
2005                           int idx, int count, struct llog_catid *idarray,
2006                           const struct lu_fid *fid)
2007 {
2008         struct llog_thread_info *lgi = llog_info(env);
2009         struct dt_object        *o = NULL;
2010         struct thandle          *th;
2011         int                      rc, size;
2012
2013         ENTRY;
2014
2015         LASSERT(d);
2016
2017         size = sizeof(*idarray) * count;
2018         lgi->lgi_off = idx *  sizeof(*idarray);
2019
2020         lgi->lgi_fid = *fid;
2021         o = dt_locate(env, d, &lgi->lgi_fid);
2022         if (IS_ERR(o))
2023                 RETURN(PTR_ERR(o));
2024
2025         if (!dt_object_exists(o)) {
2026                 th = dt_trans_create(env, d);
2027                 if (IS_ERR(th))
2028                         GOTO(out, rc = PTR_ERR(th));
2029
2030                 lgi->lgi_attr.la_valid = LA_MODE;
2031                 lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
2032                 lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
2033
2034                 th->th_wait_submit = 1;
2035                 /* Make the llog object creation synchronization, so
2036                  * it will be reliable to the reference, especially
2037                  * for remote reference */
2038                 th->th_sync = 1;
2039
2040                 rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
2041                                        &lgi->lgi_dof, th);
2042                 if (rc)
2043                         GOTO(out_trans, rc);
2044
2045                 rc = dt_trans_start_local(env, d, th);
2046                 if (rc)
2047                         GOTO(out_trans, rc);
2048
2049                 dt_write_lock(env, o, 0);
2050                 if (!dt_object_exists(o))
2051                         rc = dt_create(env, o, &lgi->lgi_attr, NULL,
2052                                        &lgi->lgi_dof, th);
2053                 dt_write_unlock(env, o);
2054 out_trans:
2055                 dt_trans_stop(env, d, th);
2056                 if (rc)
2057                         GOTO(out, rc);
2058         }
2059
2060         rc = dt_attr_get(env, o, &lgi->lgi_attr);
2061         if (rc)
2062                 GOTO(out, rc);
2063
2064         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2065                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2066                        o->do_lu.lo_dev->ld_obd->obd_name,
2067                        lgi->lgi_attr.la_mode);
2068                 GOTO(out, rc = -ENOENT);
2069         }
2070
2071         CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
2072                (int)lgi->lgi_attr.la_size, size);
2073
2074         /* return just number of llogs */
2075         if (idarray == NULL) {
2076                 rc = lgi->lgi_attr.la_size / sizeof(*idarray);
2077                 GOTO(out, rc);
2078         }
2079
2080         /* read for new ost index or for empty file */
2081         memset(idarray, 0, size);
2082         if (lgi->lgi_attr.la_size <= lgi->lgi_off)
2083                 GOTO(out, rc = 0);
2084         if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
2085                 size = lgi->lgi_attr.la_size - lgi->lgi_off;
2086
2087         lgi->lgi_buf.lb_buf = idarray;
2088         lgi->lgi_buf.lb_len = size;
2089         rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
2090         /* -EFAULT means the llog is a sparse file. This is not an error
2091          * after arbitrary OST index is supported. */
2092         if (rc < 0 && rc != -EFAULT) {
2093                 CERROR("%s: error reading CATALOGS: rc = %d\n",
2094                        o->do_lu.lo_dev->ld_obd->obd_name,  rc);
2095                 GOTO(out, rc);
2096         }
2097
2098         EXIT;
2099 out:
2100         dt_object_put(env, o);
2101         RETURN(rc);
2102 }
2103 EXPORT_SYMBOL(llog_osd_get_cat_list);
2104
2105 /**
2106  * Write the special file which contains the list of llog catalogs IDs
2107  *
2108  * This function writes the CATALOG file which contains the array of llog
2109  * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
2110  * number.
2111  *
2112  * \param[in]  env      execution environment
2113  * \param[in]  d        corresponding storage device
2114  * \param[in]  idx      position to start from, usually OST/MDT index
2115  * \param[in]  count    how many catalog IDs to write
2116  * \param[out] idarray  the buffer with the data to write.
2117  * \param[in]  fid      LLOG_CATALOGS_OID for CATALOG object
2118  *
2119  * \retval              0 on successful write of catalog IDs
2120  * \retval              negative value on error
2121  */
2122 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
2123                           int idx, int count, struct llog_catid *idarray,
2124                           const struct lu_fid *fid)
2125 {
2126         struct llog_thread_info *lgi = llog_info(env);
2127         struct dt_object        *o = NULL;
2128         struct thandle          *th;
2129         int                      rc, size;
2130
2131         if (count == 0)
2132                 RETURN(0);
2133
2134         LASSERT(d);
2135
2136         size = sizeof(*idarray) * count;
2137         lgi->lgi_off = idx * sizeof(*idarray);
2138         lgi->lgi_fid = *fid;
2139
2140         o = dt_locate(env, d, &lgi->lgi_fid);
2141         if (IS_ERR(o))
2142                 RETURN(PTR_ERR(o));
2143
2144         if (!dt_object_exists(o))
2145                 GOTO(out, rc = -ENOENT);
2146
2147         rc = dt_attr_get(env, o, &lgi->lgi_attr);
2148         if (rc)
2149                 GOTO(out, rc);
2150
2151         if (!S_ISREG(lgi->lgi_attr.la_mode)) {
2152                 CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
2153                        o->do_lu.lo_dev->ld_obd->obd_name,
2154                        lgi->lgi_attr.la_mode);
2155                 GOTO(out, rc = -ENOENT);
2156         }
2157
2158         th = dt_trans_create(env, d);
2159         if (IS_ERR(th))
2160                 GOTO(out, rc = PTR_ERR(th));
2161
2162         lgi->lgi_buf.lb_len = size;
2163         lgi->lgi_buf.lb_buf = idarray;
2164         rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
2165         if (rc)
2166                 GOTO(out_trans, rc);
2167
2168         /* For update log, this happens during initialization,
2169          * see lod_sub_prep_llog(), and we need make sure catlog
2170          * file ID is written to catlist file(committed) before
2171          * cross-MDT operation write update records to catlog FILE,
2172          * otherwise, during failover these update records might
2173          * missing */
2174         if (fid_is_update_log(fid))
2175                 th->th_sync = 1;
2176
2177         rc = dt_trans_start_local(env, d, th);
2178         if (rc)
2179                 GOTO(out_trans, rc);
2180
2181         th->th_wait_submit = 1;
2182
2183         rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
2184         if (rc)
2185                 CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
2186                        idx, rc);
2187 out_trans:
2188         dt_trans_stop(env, d, th);
2189 out:
2190         dt_object_put(env, o);
2191         RETURN(rc);
2192 }
2193 EXPORT_SYMBOL(llog_osd_put_cat_list);