lustre/obdclass/llog.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/obdclass/llog.c
  32  *
  33  * OST<->MDS recovery logging infrastructure.
  34  * Invariants in implementation:
  35  * - we do not share logs among different OST<->MDS connections, so that
  36  *   if an OST or MDS fails it need only look at log(s) relevant to itself
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  40  * Author: Mikhail Pershin <tappro@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_LOG
  44
  45 #include <linux/fs_struct.h>
  46 #include <linux/pid_namespace.h>
  47 #include <linux/kthread.h>
  48 #include <llog_swab.h>
  49 #include <lustre_log.h>
  50 #include <obd_support.h>
  51 #include <obd_class.h>
  52 #include "llog_internal.h"
  53
  54 /*
  55  * Allocate a new log or catalog handle
  56  * Used inside llog_open().
  57  */
  58 static struct llog_handle *llog_alloc_handle(void)
  59 {
  60         struct llog_handle *loghandle;
  61
  62         OBD_ALLOC_PTR(loghandle);
  63         if (loghandle == NULL)
  64                 return NULL;
  65
  66         init_rwsem(&loghandle->lgh_lock);
  67         mutex_init(&loghandle->lgh_hdr_mutex);
  68         init_rwsem(&loghandle->lgh_last_sem);
  69         INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
  70         refcount_set(&loghandle->lgh_refcount, 1);
  71
  72         return loghandle;
  73 }
  74
  75 /*
  76  * Free llog handle and header data if exists. Used in llog_close() only
  77  */
  78 static void llog_free_handle(struct llog_handle *loghandle)
  79 {
  80         LASSERT(loghandle != NULL);
  81
  82         /* failed llog_init_handle */
  83         if (loghandle->lgh_hdr == NULL)
  84                 goto out;
  85
  86         if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
  87                 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
  88         else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
  89                 LASSERT(list_empty(&loghandle->u.chd.chd_head));
  90         OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size);
  91 out:
  92         OBD_FREE_PTR(loghandle);
  93 }
  94
  95 struct llog_handle *llog_handle_get(struct llog_handle *loghandle)
  96 {
  97         if (refcount_inc_not_zero(&loghandle->lgh_refcount))
  98                 return loghandle;
  99         return NULL;
 100 }
 101
 102 int llog_handle_put(const struct lu_env *env, struct llog_handle *loghandle)
 103 {
 104         int rc = 0;
 105
 106         if (refcount_dec_and_test(&loghandle->lgh_refcount)) {
 107                 const struct llog_operations *lop;
 108
 109                 rc = llog_handle2ops(loghandle, &lop);
 110                 if (!rc) {
 111                         if (lop->lop_close)
 112                                 rc = lop->lop_close(env, loghandle);
 113                         else
 114                                 rc = -EOPNOTSUPP;
 115                 }
 116                 llog_free_handle(loghandle);
 117         }
 118         return rc;
 119 }
 120
 121 static int llog_declare_destroy(const struct lu_env *env,
 122                                 struct llog_handle *handle,
 123                                 struct thandle *th)
 124 {
 125         const struct llog_operations *lop;
 126         int rc;
 127
 128         ENTRY;
 129
 130         rc = llog_handle2ops(handle, &lop);
 131         if (rc)
 132                 RETURN(rc);
 133         if (lop->lop_declare_destroy == NULL)
 134                 RETURN(-EOPNOTSUPP);
 135
 136         rc = lop->lop_declare_destroy(env, handle, th);
 137
 138         RETURN(rc);
 139 }
 140
 141 int llog_trans_destroy(const struct lu_env *env, struct llog_handle *handle,
 142                        struct thandle *th)
 143 {
 144         const struct llog_operations *lop;
 145         int rc;
 146         ENTRY;
 147
 148         rc = llog_handle2ops(handle, &lop);
 149         if (rc < 0)
 150                 RETURN(rc);
 151         if (lop->lop_destroy == NULL)
 152                 RETURN(-EOPNOTSUPP);
 153
 154         LASSERT(handle->lgh_obj != NULL);
 155         if (!llog_exist(handle))
 156                 RETURN(0);
 157
 158         rc = lop->lop_destroy(env, handle, th);
 159
 160         RETURN(rc);
 161 }
 162
 163 int llog_destroy(const struct lu_env *env, struct llog_handle *handle)
 164 {
 165         const struct llog_operations *lop;
 166         struct dt_device *dt;
 167         struct thandle *th;
 168         int rc;
 169
 170         ENTRY;
 171
 172         rc = llog_handle2ops(handle, &lop);
 173         if (rc < 0)
 174                 RETURN(rc);
 175         if (lop->lop_destroy == NULL)
 176                 RETURN(-EOPNOTSUPP);
 177
 178         if (handle->lgh_obj == NULL) {
 179                 /* if lgh_obj == NULL, then it is from client side destroy */
 180                 rc = lop->lop_destroy(env, handle, NULL);
 181                 RETURN(rc);
 182         }
 183
 184         if (!llog_exist(handle))
 185                 RETURN(0);
 186
 187         dt = lu2dt_dev(handle->lgh_obj->do_lu.lo_dev);
 188
 189         if (unlikely(unlikely(dt->dd_rdonly)))
 190                 RETURN(-EROFS);
 191
 192         th = dt_trans_create(env, dt);
 193         if (IS_ERR(th))
 194                 RETURN(PTR_ERR(th));
 195         th->th_wait_submit = 1;
 196
 197         rc = llog_declare_destroy(env, handle, th);
 198         if (rc != 0)
 199                 GOTO(out_trans, rc);
 200
 201         rc = dt_trans_start_local(env, dt, th);
 202         if (rc < 0)
 203                 GOTO(out_trans, rc);
 204
 205         rc = lop->lop_destroy(env, handle, th);
 206
 207 out_trans:
 208         dt_trans_stop(env, dt, th);
 209
 210         RETURN(rc);
 211 }
 212 EXPORT_SYMBOL(llog_destroy);
 213
 214 /* returns negative on error; 0 if success; 1 if success & log destroyed */
 215 int llog_cancel_arr_rec(const struct lu_env *env, struct llog_handle *loghandle,
 216                         int num, int *index)
 217 {
 218         struct llog_thread_info *lgi = llog_info(env);
 219         struct dt_device        *dt;
 220         struct llog_log_hdr     *llh;
 221         struct thandle          *th;
 222         __u32                    tmp_lgc_index;
 223         int                      rc, i = 0;
 224         int rc1;
 225         bool subtract_count = false;
 226
 227         ENTRY;
 228
 229         LASSERT(loghandle != NULL);
 230         LASSERT(loghandle->lgh_ctxt != NULL);
 231         LASSERT(loghandle->lgh_obj != NULL);
 232
 233         llh = loghandle->lgh_hdr;
 234
 235         CDEBUG(D_OTHER, "Canceling %d records, first %d in log "DFID"\n",
 236                num, index[0], PLOGID(&loghandle->lgh_id));
 237
 238         dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
 239
 240         if (unlikely(unlikely(dt->dd_rdonly)))
 241                 RETURN(0);
 242
 243         th = dt_trans_create(env, dt);
 244         if (IS_ERR(th))
 245                 RETURN(PTR_ERR(th));
 246
 247         rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, 0, th);
 248         if (rc < 0)
 249                 GOTO(out_trans, rc);
 250
 251         if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY)) {
 252                 rc = llog_declare_destroy(env, loghandle, th);
 253                 if (rc < 0)
 254                         GOTO(out_trans, rc);
 255         }
 256
 257         th->th_wait_submit = 1;
 258         rc = dt_trans_start_local(env, dt, th);
 259         if (rc < 0)
 260                 GOTO(out_trans, rc);
 261
 262         down_write(&loghandle->lgh_lock);
 263         /* clear bitmap */
 264         mutex_lock(&loghandle->lgh_hdr_mutex);
 265         for (i = 0; i < num; ++i) {
 266                 if (index[i] == 0) {
 267                         CERROR("Can't cancel index 0 which is header\n");
 268                         GOTO(out_unlock, rc = -EINVAL);
 269                 }
 270                 if (!__test_and_clear_bit_le(index[i], LLOG_HDR_BITMAP(llh))) {
 271                         CDEBUG(D_OTHER, "Catalog index %u already clear?\n",
 272                                index[i]);
 273                         GOTO(out_unlock, rc = -ENOENT);
 274                 }
 275         }
 276         loghandle->lgh_hdr->llh_count -= num;
 277         subtract_count = true;
 278
 279         /* Since llog_process_thread use lgi_cookie, it`s better to save them
 280          * and restore after using
 281          */
 282         tmp_lgc_index = lgi->lgi_cookie.lgc_index;
 283         /* Pass this index to llog_osd_write_rec(), which will use the index
 284          * to only update the necesary bitmap. */
 285         lgi->lgi_cookie.lgc_index = index[0];
 286         /* update header */
 287         rc = llog_write_rec(env, loghandle, &llh->llh_hdr, (num != 1 ? NULL :
 288                             &lgi->lgi_cookie), LLOG_HEADER_IDX, th);
 289         lgi->lgi_cookie.lgc_index = tmp_lgc_index;
 290
 291         if (rc != 0)
 292                 GOTO(out_unlock, rc);
 293
 294         if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
 295             (llh->llh_count == 1) &&
 296             ((loghandle->lgh_last_idx == llog_max_idx(llh)) ||
 297              (loghandle->u.phd.phd_cat_handle != NULL &&
 298               loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
 299                 loghandle))) {
 300                 /* never try to destroy it again */
 301                 llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
 302                 rc = llog_trans_destroy(env, loghandle, th);
 303                 if (rc < 0) {
 304                         /* Sigh, can not destroy the final plain llog, but
 305                          * the bitmap has been clearly, so the record can not
 306                          * be accessed anymore, let's return 0 for now, and
 307                          * the orphan will be handled by LFSCK. */
 308                         CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
 309                                loghandle2name(loghandle),
 310                                PLOGID(&loghandle->lgh_id), rc);
 311                         GOTO(out_unlock, rc = 0);
 312                 }
 313                 rc = LLOG_DEL_PLAIN;
 314         }
 315
 316 out_unlock:
 317         if (rc < 0) {
 318                 /* restore bitmap while holding a mutex */
 319                 if (subtract_count) {
 320                         loghandle->lgh_hdr->llh_count += num;
 321                         subtract_count = false;
 322                 }
 323                 for (i = i - 1; i >= 0; i--)
 324                         set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
 325         }
 326         mutex_unlock(&loghandle->lgh_hdr_mutex);
 327         up_write(&loghandle->lgh_lock);
 328 out_trans:
 329         rc1 = dt_trans_stop(env, dt, th);
 330         if (rc == 0)
 331                 rc = rc1;
 332         if (rc1 < 0) {
 333                 mutex_lock(&loghandle->lgh_hdr_mutex);
 334                 if (subtract_count)
 335                         loghandle->lgh_hdr->llh_count += num;
 336                 for (i = i - 1; i >= 0; i--)
 337                         set_bit_le(index[i], LLOG_HDR_BITMAP(llh));
 338                 mutex_unlock(&loghandle->lgh_hdr_mutex);
 339         }
 340         RETURN(rc);
 341 }
 342
 343 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
 344                     int index)
 345 {
 346         return llog_cancel_arr_rec(env, loghandle, 1, &index);
 347 }
 348
 349 int llog_read_header(const struct lu_env *env, struct llog_handle *handle,
 350                      const struct obd_uuid *uuid)
 351 {
 352         const struct llog_operations *lop;
 353         int rc;
 354         ENTRY;
 355
 356         rc = llog_handle2ops(handle, &lop);
 357         if (rc)
 358                 RETURN(rc);
 359
 360         if (lop->lop_read_header == NULL)
 361                 RETURN(-EOPNOTSUPP);
 362
 363         rc = lop->lop_read_header(env, handle);
 364         if (rc == LLOG_EEMPTY) {
 365                 struct llog_log_hdr *llh = handle->lgh_hdr;
 366
 367                 /* lrh_len should be initialized in llog_init_handle */
 368                 handle->lgh_last_idx = 0; /* header is record with index 0 */
 369                 llh->llh_count = 1;         /* for the header record */
 370                 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
 371                 LASSERT(handle->lgh_ctxt->loc_chunk_size >=
 372                                                 LLOG_MIN_CHUNK_SIZE);
 373                 llh->llh_hdr.lrh_len = handle->lgh_ctxt->loc_chunk_size;
 374                 llh->llh_hdr.lrh_index = 0;
 375                 llh->llh_timestamp = ktime_get_real_seconds();
 376                 if (uuid)
 377                         memcpy(&llh->llh_tgtuuid, uuid,
 378                                sizeof(llh->llh_tgtuuid));
 379                 llh->llh_bitmap_offset = offsetof(typeof(*llh), llh_bitmap);
 380                 /* Since update llog header might also call this function,
 381                  * let's reset the bitmap to 0 here */
 382                 memset(LLOG_HDR_BITMAP(llh), 0, llh->llh_hdr.lrh_len -
 383                                                 llh->llh_bitmap_offset -
 384                                                 sizeof(llh->llh_tail));
 385                 set_bit_le(0, LLOG_HDR_BITMAP(llh));
 386                 LLOG_HDR_TAIL(llh)->lrt_len = llh->llh_hdr.lrh_len;
 387                 LLOG_HDR_TAIL(llh)->lrt_index = llh->llh_hdr.lrh_index;
 388                 rc = 0;
 389         }
 390         RETURN(rc);
 391 }
 392 EXPORT_SYMBOL(llog_read_header);
 393
 394 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
 395                      int flags, struct obd_uuid *uuid)
 396 {
 397         struct llog_log_hdr     *llh;
 398         enum llog_flag           fmt = flags & LLOG_F_EXT_MASK;
 399         int                      rc;
 400         int                     chunk_size = handle->lgh_ctxt->loc_chunk_size;
 401         ENTRY;
 402
 403         LASSERT(handle->lgh_hdr == NULL);
 404
 405         LASSERT(chunk_size >= LLOG_MIN_CHUNK_SIZE);
 406         OBD_ALLOC_LARGE(llh, chunk_size);
 407         if (llh == NULL)
 408                 RETURN(-ENOMEM);
 409
 410         handle->lgh_hdr = llh;
 411         handle->lgh_hdr_size = chunk_size;
 412         /* first assign flags to use llog_client_ops */
 413         llh->llh_flags = flags;
 414         rc = llog_read_header(env, handle, uuid);
 415         if (rc == 0) {
 416                 if (unlikely((llh->llh_flags & LLOG_F_IS_PLAIN &&
 417                               flags & LLOG_F_IS_CAT) ||
 418                              (llh->llh_flags & LLOG_F_IS_CAT &&
 419                               flags & LLOG_F_IS_PLAIN))) {
 420                         CERROR("%s: llog type is %s but initializing %s\n",
 421                                loghandle2name(handle),
 422                                llh->llh_flags & LLOG_F_IS_CAT ?
 423                                "catalog" : "plain",
 424                                flags & LLOG_F_IS_CAT ? "catalog" : "plain");
 425                         GOTO(out, rc = -EINVAL);
 426                 } else if (llh->llh_flags &
 427                            (LLOG_F_IS_PLAIN | LLOG_F_IS_CAT)) {
 428                         /*
 429                          * it is possible to open llog without specifying llog
 430                          * type so it is taken from llh_flags
 431                          */
 432                         flags = llh->llh_flags;
 433                 } else {
 434                         /* for some reason the llh_flags has no type set */
 435                         CERROR("llog type is not specified!\n");
 436                         GOTO(out, rc = -EINVAL);
 437                 }
 438                 if (unlikely(uuid &&
 439                              !obd_uuid_equals(uuid, &llh->llh_tgtuuid))) {
 440                         CERROR("%s: llog uuid mismatch: %s/%s\n",
 441                                loghandle2name(handle),
 442                                (char *)uuid->uuid,
 443                                (char *)llh->llh_tgtuuid.uuid);
 444                         GOTO(out, rc = -EEXIST);
 445                 }
 446         }
 447         if (flags & LLOG_F_IS_CAT) {
 448                 LASSERT(list_empty(&handle->u.chd.chd_head));
 449                 INIT_LIST_HEAD(&handle->u.chd.chd_head);
 450                 llh->llh_size = sizeof(struct llog_logid_rec);
 451                 llh->llh_flags |= LLOG_F_IS_FIXSIZE;
 452         } else if (!(flags & LLOG_F_IS_PLAIN)) {
 453                 CERROR("%s: unknown flags: %#x (expected %#x or %#x)\n",
 454                        loghandle2name(handle), flags, LLOG_F_IS_CAT,
 455                        LLOG_F_IS_PLAIN);
 456                 rc = -EINVAL;
 457         }
 458         llh->llh_flags |= fmt;
 459 out:
 460         if (rc) {
 461                 OBD_FREE_LARGE(llh, chunk_size);
 462                 handle->lgh_hdr = NULL;
 463         }
 464         RETURN(rc);
 465 }
 466 EXPORT_SYMBOL(llog_init_handle);
 467
 468 #define LLOG_ERROR_REC(lgh, rec, format, a...) \
 469         CERROR("%s: "DFID" rec type=%x idx=%u len=%u, " format "\n" , \
 470                loghandle2name(lgh), PLOGID(&lgh->lgh_id), (rec)->lrh_type, \
 471                (rec)->lrh_index, (rec)->lrh_len, ##a)
 472
 473 int llog_verify_record(const struct llog_handle *llh, struct llog_rec_hdr *rec)
 474 {
 475         int chunk_size = llh->lgh_hdr->llh_hdr.lrh_len;
 476
 477         if ((rec->lrh_type & LLOG_OP_MASK) != LLOG_OP_MAGIC)
 478                 LLOG_ERROR_REC(llh, rec, "magic is bad");
 479         else if (rec->lrh_len == 0 || rec->lrh_len > chunk_size)
 480                 LLOG_ERROR_REC(llh, rec, "bad record len, chunk size is %d",
 481                                chunk_size);
 482         else if (rec->lrh_index > llog_max_idx(llh->lgh_hdr))
 483                 LLOG_ERROR_REC(llh, rec, "index is too high");
 484         else
 485                 return 0;
 486
 487         return -EINVAL;
 488 }
 489 EXPORT_SYMBOL(llog_verify_record);
 490
 491 static inline bool llog_is_index_skipable(int idx, struct llog_log_hdr *llh,
 492                                           struct llog_process_cat_data *cd)
 493 {
 494         if (cd && (cd->lpcd_read_mode & LLOG_READ_MODE_RAW))
 495                 return false;
 496
 497         return !test_bit_le(idx, LLOG_HDR_BITMAP(llh));
 498 }
 499
 500 static int llog_process_thread(void *arg)
 501 {
 502         struct llog_process_info *lpi = arg;
 503         struct llog_handle *loghandle = lpi->lpi_loghandle;
 504         struct llog_log_hdr *llh = loghandle->lgh_hdr;
 505         struct llog_process_cat_data *cd  = lpi->lpi_catdata;
 506         struct llog_thread_info *lti;
 507         char *buf;
 508         size_t chunk_size;
 509         __u64 cur_offset;
 510         int rc = 0, index = 1, last_index;
 511         int saved_index = 0;
 512         int last_called_index = 0;
 513         bool repeated = false;
 514
 515         ENTRY;
 516
 517         if (llh == NULL)
 518                 RETURN(-EINVAL);
 519
 520         lti = lpi->lpi_env == NULL ? NULL : llog_info(lpi->lpi_env);
 521
 522         cur_offset = chunk_size = llh->llh_hdr.lrh_len;
 523         /* expect chunk_size to be power of two */
 524         LASSERT(is_power_of_2(chunk_size));
 525
 526         OBD_ALLOC_LARGE(buf, chunk_size);
 527         if (buf == NULL) {
 528                 lpi->lpi_rc = -ENOMEM;
 529                 RETURN(0);
 530         }
 531
 532         last_index = llog_max_idx(llh);
 533         if (cd) {
 534                 if (cd->lpcd_first_idx >= llog_max_idx(llh))
 535                         /* End of the indexes -> Nothing to do */
 536                         GOTO(out, rc = 0);
 537
 538                 index = cd->lpcd_first_idx + 1;
 539                 last_called_index = cd->lpcd_first_idx;
 540                 if (cd->lpcd_last_idx > 0 &&
 541                     cd->lpcd_last_idx <= llog_max_idx(llh))
 542                         last_index = cd->lpcd_last_idx;
 543                 else if (cd->lpcd_read_mode & LLOG_READ_MODE_RAW)
 544                         last_index = loghandle->lgh_last_idx;
 545         }
 546
 547         while (rc == 0) {
 548                 struct llog_rec_hdr *rec;
 549                 off_t chunk_offset = 0;
 550                 unsigned int buf_offset = 0;
 551                 int lh_last_idx;
 552                 int synced_idx = 0;
 553
 554                 /* skip records not set in bitmap */
 555                 while (index <= last_index &&
 556                        llog_is_index_skipable(index, llh, cd))
 557                         ++index;
 558
 559                 /* There are no indices prior the last_index */
 560                 if (index > last_index)
 561                         break;
 562
 563                 CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
 564                        last_index);
 565
 566 repeat:
 567                 /* get the buf with our target record; avoid old garbage */
 568                 memset(buf, 0, chunk_size);
 569                 /* the record index for outdated chunk data */
 570                 /* it is safe to process buffer until saved lgh_last_idx */
 571                 lh_last_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 572                 rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
 573                                      index, &cur_offset, buf, chunk_size);
 574                 if (repeated && rc)
 575                         CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
 576                                " buf_offset %u, rc = %d\n", cur_offset,
 577                                (__u64)chunk_offset, buf_offset, rc);
 578                 if (rc == -ESTALE)
 579                         GOTO(out, rc = 0);
 580                 /* we`ve tried to reread the chunk, but there is no
 581                  * new records */
 582                 if (repeated && (chunk_offset + buf_offset) == cur_offset &&
 583                     (rc == -EBADR || rc == -EIO))
 584                         GOTO(out, rc = 0);
 585                 /* EOF while trying to skip to the next chunk */
 586                 if (!index && rc == -EBADR)
 587                         GOTO(out, rc = 0);
 588                 if (rc != 0)
 589                         GOTO(out, rc);
 590
 591                 /* NB: after llog_next_block() call the cur_offset is the
 592                  * offset of the next block after read one.
 593                  * The absolute offset of the current chunk is calculated
 594                  * from cur_offset value and stored in chunk_offset variable.
 595                  */
 596                 if ((cur_offset & (chunk_size - 1)) != 0)
 597                         chunk_offset = cur_offset & ~(chunk_size - 1);
 598                 else
 599                         chunk_offset = cur_offset - chunk_size;
 600
 601                 /* NB: when rec->lrh_len is accessed it is already swabbed
 602                  * since it is used at the "end" of the loop and the rec
 603                  * swabbing is done at the beginning of the loop. */
 604                 for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
 605                      (char *)rec < buf + chunk_size;
 606                      rec = llog_rec_hdr_next(rec)) {
 607
 608                         CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
 609                                rec, rec->lrh_type);
 610
 611                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
 612                                 lustre_swab_llog_rec(rec);
 613
 614                         CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
 615                                rec->lrh_type, rec->lrh_index);
 616
 617                         /* start with first rec if block was skipped */
 618                         if (!index) {
 619                                 CDEBUG(D_OTHER,
 620                                        "%s: skipping to the index %u\n",
 621                                        loghandle2name(loghandle),
 622                                        rec->lrh_index);
 623                                 index = rec->lrh_index;
 624                         }
 625
 626                         if (index == (synced_idx + 1) &&
 627                             synced_idx == LLOG_HDR_TAIL(llh)->lrt_index)
 628                                 GOTO(out, rc = 0);
 629
 630                         if (CFS_FAIL_PRECHECK(OBD_FAIL_LLOG_PROCESS_TIMEOUT) &&
 631                                 cfs_fail_val == (unsigned int)
 632                                         (loghandle->lgh_id.lgl_oi.oi.oi_id &
 633                                          0xFFFFFFFF)) {
 634                                 CFS_RACE(OBD_FAIL_LLOG_PROCESS_TIMEOUT);
 635                         }
 636
 637                         /* the bitmap could be changed during processing
 638                          * records from the chunk. For wrapped catalog
 639                          * it means we can read deleted record and try to
 640                          * process it. Check this case and reread the chunk.
 641                          * It is safe to process to lh_last_idx, including
 642                          * lh_last_idx if it was synced. We can not do <=
 643                          * comparison, cause for wrapped catalog lgh_last_idx
 644                          * could be less than index. So we detect last index
 645                          * for processing as index == lh_last_idx+1. But when
 646                          * catalog is wrapped and full lgh_last_idx=llh_cat_idx,
 647                          * the first processing index is llh_cat_idx+1.The
 648                          * exception is !(lgh_last_idx == llh_cat_idx &&
 649                          * index == llh_cat_idx + 1), and after simplification
 650                          * it turns to
 651                          * lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index
 652                          * This exception is working for catalog only.
 653                          * The last check is for the partial chunk boundary,
 654                          * if it is reached then try to re-read for possible
 655                          * new records once.
 656                          */
 657                         if ((index == lh_last_idx && synced_idx != index) ||
 658                             (index == (lh_last_idx + 1) &&
 659                              lh_last_idx != LLOG_HDR_TAIL(llh)->lrt_index) ||
 660                             (((char *)rec - buf >= cur_offset - chunk_offset) &&
 661                             !repeated)) {
 662                                 /* save offset inside buffer for the re-read */
 663                                 buf_offset = (char *)rec - (char *)buf;
 664                                 cur_offset = chunk_offset;
 665                                 repeated = true;
 666                                 /* We need to be sure lgh_last_idx
 667                                  * record was saved to disk
 668                                  */
 669                                 down_read(&loghandle->lgh_last_sem);
 670                                 synced_idx = LLOG_HDR_TAIL(llh)->lrt_index;
 671                                 up_read(&loghandle->lgh_last_sem);
 672                                 CDEBUG(D_OTHER, "synced_idx: %d\n", synced_idx);
 673                                 goto repeat;
 674                         }
 675                         repeated = false;
 676
 677                         rc = llog_verify_record(loghandle, rec);
 678                         if (rc) {
 679                                 CDEBUG(D_OTHER, "invalid record at index %d\n",
 680                                        index);
 681                                 /*
 682                                  * for fixed-sized llogs we can skip one record
 683                                  * by using llh_size from llog header.
 684                                  * Otherwise skip the next llog chunk.
 685                                  */
 686                                 rc = 0;
 687                                 if (llh->llh_flags & LLOG_F_IS_FIXSIZE) {
 688                                         rec->lrh_len = llh->llh_size;
 689                                         goto next_rec;
 690                                 }
 691                                 /* make sure that is always next block */
 692                                 cur_offset = chunk_offset + chunk_size;
 693                                 /* no goal to find, just next block to read */
 694                                 index = 0;
 695                                 break;
 696                         }
 697
 698                         if (rec->lrh_index < index) {
 699                                 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
 700                                        rec->lrh_index);
 701                                 continue;
 702                         }
 703
 704                         if (rec->lrh_index > index) {
 705                                 /* the record itself looks good, but we met a
 706                                  * gap which can be result of old bugs, just
 707                                  * keep going */
 708                                 LLOG_ERROR_REC(loghandle, rec,
 709                                                "gap in index, expected %u",
 710                                                index);
 711                                 index = rec->lrh_index;
 712                         }
 713
 714                         CDEBUG(D_OTHER,
 715                                "lrh_index: %d lrh_len: %d (%d remains)\n",
 716                                rec->lrh_index, rec->lrh_len,
 717                                (int)(buf + chunk_size - (char *)rec));
 718
 719                         /* lgh_cur_offset is used only at llog_test_3 */
 720                         loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
 721                                                     chunk_offset;
 722
 723                         /* if needed, process the callback on this record */
 724                         if (!llog_is_index_skipable(index, llh, cd)) {
 725                                 struct llog_cookie *lgc;
 726                                 __u64   tmp_off;
 727                                 int     tmp_idx;
 728
 729                                 CDEBUG((llh->llh_flags & LLOG_F_IS_CAT ?
 730                                         D_HA : D_OTHER),
 731                                        "index: %d, lh_last_idx: %d "
 732                                        "synced_idx: %d lgh_last_idx: %d\n",
 733                                        index, lh_last_idx, synced_idx,
 734                                        loghandle->lgh_last_idx);
 735
 736                                 if (lti != NULL) {
 737                                         lgc = &lti->lgi_cookie;
 738                                         /* store lu_env for recursive calls */
 739                                         tmp_off = lgc->lgc_offset;
 740                                         tmp_idx = lgc->lgc_index;
 741
 742                                         lgc->lgc_offset = (char *)rec -
 743                                                 (char *)buf + chunk_offset;
 744                                         lgc->lgc_index = rec->lrh_index;
 745                                 }
 746                                 /* using lu_env for passing record offset to
 747                                  * llog_write through various callbacks */
 748                                 rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
 749                                                  lpi->lpi_cbdata);
 750                                 last_called_index = index;
 751
 752                                 if (lti != NULL) {
 753                                         lgc->lgc_offset = tmp_off;
 754                                         lgc->lgc_index = tmp_idx;
 755                                 }
 756
 757                                 if (rc == LLOG_PROC_BREAK ||
 758                                     rc == LLOG_SKIP_PLAIN) {
 759                                         GOTO(out, rc);
 760                                 } else if (rc == LLOG_DEL_RECORD) {
 761                                         rc = llog_cancel_rec(lpi->lpi_env,
 762                                                              loghandle,
 763                                                              rec->lrh_index);
 764                                         /* Allow parallel cancelling, ENOENT
 765                                          * means record was canceled at another
 766                                          * processing thread or callback
 767                                          */
 768                                         if (rc == -ENOENT)
 769                                                 rc = 0;
 770                                 }
 771                                 if (rc)
 772                                         GOTO(out, rc);
 773                                 /* some stupid callbacks directly cancel records
 774                                  * and delete llog. Check it and stop
 775                                  * processing. */
 776                                 if (loghandle->lgh_hdr == NULL ||
 777                                     loghandle->lgh_hdr->llh_count == 1)
 778                                         GOTO(out, rc = 0);
 779                         }
 780 next_rec:
 781                         /* exit if the last index is reached */
 782                         if (index >= last_index)
 783                                 GOTO(out, rc = 0);
 784                         ++index;
 785                 }
 786         }
 787
 788 out:
 789         CDEBUG(D_HA, "stop processing %s "DFID" index %d count %d\n",
 790                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" : "plain"),
 791                PLOGID(&loghandle->lgh_id), index, llh->llh_count);
 792
 793         if (cd != NULL)
 794                 cd->lpcd_last_idx = last_called_index;
 795
 796         if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
 797                 if (dt_object_remote(loghandle->lgh_obj)) {
 798                         /* If it is remote object, then -EIO might means
 799                          * disconnection or eviction, let's return -EAGAIN,
 800                          * so for update recovery log processing, it will
 801                          * retry until the umount or abort recovery, see
 802                          * lod_sub_recovery_thread() */
 803                         CERROR("%s retry remote llog process\n",
 804                                loghandle2name(loghandle));
 805                         rc = -EAGAIN;
 806                 } else {
 807                         /* something bad happened to the processing of a local
 808                          * llog file, probably I/O error or the log got
 809                          * corrupted to be able to finally release the log we
 810                          * discard any remaining bits in the header */
 811                         CERROR("%s: local llog is corrupted "DFID" %s index %d count %d\n",
 812                                loghandle2name(loghandle),
 813                                PLOGID(&loghandle->lgh_id),
 814                                ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
 815                                "plain"), index, llh->llh_count);
 816
 817                         while (index <= last_index) {
 818                                 if (test_bit_le(index,
 819                                                   LLOG_HDR_BITMAP(llh)) != 0)
 820                                         llog_cancel_rec(lpi->lpi_env, loghandle,
 821                                                         index);
 822                                 index++;
 823                         }
 824                         rc = 0;
 825                 }
 826         }
 827
 828         OBD_FREE_LARGE(buf, chunk_size);
 829         lpi->lpi_rc = rc;
 830         return 0;
 831 }
 832
 833 static int llog_process_thread_daemonize(void *arg)
 834 {
 835         struct llog_process_info        *lpi = arg;
 836         struct lu_env                    env;
 837         int                              rc;
 838         struct nsproxy                  *new_ns, *curr_ns = current->nsproxy;
 839
 840         task_lock(lpi->lpi_reftask);
 841         new_ns = lpi->lpi_reftask->nsproxy;
 842         if (curr_ns != new_ns) {
 843                 get_nsproxy(new_ns);
 844
 845                 current->nsproxy = new_ns;
 846                 /* XXX: we should call put_nsproxy() instead of
 847                  * atomic_dec(&ns->count) directly. But put_nsproxy() cannot be
 848                  * used outside of the kernel itself, because it calls
 849                  * free_nsproxy() which is not exported by the kernel
 850                  * (defined in kernel/nsproxy.c) */
 851                 if (curr_ns)
 852                         atomic_dec(&curr_ns->count);
 853         }
 854         task_unlock(lpi->lpi_reftask);
 855
 856         unshare_fs_struct();
 857         /* client env has no keys, tags is just 0 */
 858         rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
 859         if (rc)
 860                 goto out;
 861         lpi->lpi_env = &env;
 862
 863         rc = llog_process_thread(arg);
 864
 865         lu_env_fini(&env);
 866 out:
 867         complete(&lpi->lpi_completion);
 868         return rc;
 869 }
 870
 871 int llog_process_or_fork(const struct lu_env *env,
 872                          struct llog_handle *loghandle,
 873                          llog_cb_t cb, void *data, void *catdata, bool fork)
 874 {
 875         struct llog_process_info *lpi;
 876         struct llog_process_data *d = data;
 877         struct llog_process_cat_data *cd = catdata;
 878         __u32 flags = loghandle->lgh_hdr->llh_flags;
 879         int rc;
 880
 881         ENTRY;
 882
 883         OBD_ALLOC_PTR(lpi);
 884         if (lpi == NULL) {
 885                 CERROR("cannot alloc pointer\n");
 886                 RETURN(-ENOMEM);
 887         }
 888         lpi->lpi_loghandle = loghandle;
 889         lpi->lpi_cb        = cb;
 890         lpi->lpi_cbdata    = data;
 891         lpi->lpi_catdata   = catdata;
 892
 893         CDEBUG(D_OTHER, "Processing "DFID" flags 0x%03x startcat %d startidx %d first_idx %d last_idx %d read_mode %d\n",
 894                PLOGID(&loghandle->lgh_id), flags,
 895                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startcat : -1,
 896                (flags & LLOG_F_IS_CAT) && d ? d->lpd_startidx : -1,
 897                cd ? cd->lpcd_first_idx : -1, cd ? cd->lpcd_last_idx : -1,
 898                cd ? cd->lpcd_read_mode : -1);
 899         if (fork) {
 900                 struct task_struct *task;
 901
 902                 /* The new thread can't use parent env,
 903                  * init the new one in llog_process_thread_daemonize. */
 904                 lpi->lpi_env = NULL;
 905                 init_completion(&lpi->lpi_completion);
 906                 /* take reference to current, so that
 907                  * llog_process_thread_daemonize() can use it to switch to
 908                  * namespace associated with current  */
 909                 lpi->lpi_reftask = current;
 910                 task = kthread_run(llog_process_thread_daemonize, lpi,
 911                                    "llog_process_thread");
 912                 if (IS_ERR(task)) {
 913                         rc = PTR_ERR(task);
 914                         CERROR("%s: cannot start thread: rc = %d\n",
 915                                loghandle2name(loghandle), rc);
 916                         GOTO(out_lpi, rc);
 917                 }
 918                 wait_for_completion(&lpi->lpi_completion);
 919         } else {
 920                 lpi->lpi_env = env;
 921                 llog_process_thread(lpi);
 922         }
 923         rc = lpi->lpi_rc;
 924
 925 out_lpi:
 926         OBD_FREE_PTR(lpi);
 927         RETURN(rc);
 928 }
 929 EXPORT_SYMBOL(llog_process_or_fork);
 930
 931 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
 932                  llog_cb_t cb, void *data, void *catdata)
 933 {
 934         int rc;
 935         rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
 936         return rc == LLOG_DEL_PLAIN ? 0 : rc;
 937 }
 938 EXPORT_SYMBOL(llog_process);
 939
 940 static inline const struct cred *llog_raise_resource(void)
 941 {
 942         struct cred *cred = NULL;
 943
 944         if (cap_raised(current_cap(), CAP_SYS_RESOURCE))
 945                 return cred;
 946
 947         cred = prepare_creds();
 948         if (!cred)
 949                 return cred;
 950
 951         cap_raise(cred->cap_effective, CAP_SYS_RESOURCE);
 952         return override_creds(cred);
 953 }
 954
 955 static inline void llog_restore_resource(const struct cred *old_cred)
 956 {
 957         if (old_cred)
 958                 revert_creds(old_cred);
 959 }
 960
 961 int llog_reverse_process(const struct lu_env *env,
 962                          struct llog_handle *loghandle, llog_cb_t cb,
 963                          void *data, void *catdata)
 964 {
 965         struct llog_log_hdr *llh = loghandle->lgh_hdr;
 966         struct llog_process_cat_data *cd = catdata;
 967         void *buf;
 968         int rc = 0, first_index = 1, index, idx;
 969         __u32   chunk_size = llh->llh_hdr.lrh_len;
 970         ENTRY;
 971
 972         OBD_ALLOC_LARGE(buf, chunk_size);
 973         if (buf == NULL)
 974                 RETURN(-ENOMEM);
 975
 976         if (cd != NULL)
 977                 first_index = cd->lpcd_first_idx + 1;
 978         if (cd != NULL && cd->lpcd_last_idx)
 979                 index = cd->lpcd_last_idx;
 980         else
 981                 index = llog_max_idx(llh);
 982
 983         while (rc == 0) {
 984                 struct llog_rec_hdr *rec;
 985                 struct llog_rec_tail *tail;
 986
 987                 /* skip records not set in bitmap */
 988                 while (index >= first_index &&
 989                        llog_is_index_skipable(index, llh, cd))
 990                         --index;
 991
 992                 LASSERT(index >= first_index - 1);
 993                 if (index == first_index - 1)
 994                         break;
 995
 996                 /* get the buf with our target record; avoid old garbage */
 997                 memset(buf, 0, chunk_size);
 998                 rc = llog_prev_block(env, loghandle, index, buf, chunk_size);
 999                 if (rc)
1000                         GOTO(out, rc);
1001
1002                 rec = buf;
1003                 idx = rec->lrh_index;
1004                 CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
1005                 while (idx < index) {
1006                         rec = (void *)rec + rec->lrh_len;
1007                         if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
1008                                 lustre_swab_llog_rec(rec);
1009                         idx ++;
1010                 }
1011                 LASSERT(idx == index);
1012                 tail = (void *)rec + rec->lrh_len - sizeof(*tail);
1013
1014                 /* process records in buffer, starting where we found one */
1015                 while ((void *)tail > buf) {
1016                         if (tail->lrt_index == 0)
1017                                 GOTO(out, rc = 0); /* no more records */
1018
1019                         /* if needed, process the callback on this record */
1020                         if (!llog_is_index_skipable(index, llh, cd)) {
1021                                 rec = (void *)tail - tail->lrt_len +
1022                                       sizeof(*tail);
1023
1024                                 rc = cb(env, loghandle, rec, data);
1025                                 if (rc == LLOG_PROC_BREAK ||
1026                                     rc == LLOG_SKIP_PLAIN) {
1027                                         GOTO(out, rc);
1028                                 } else if (rc == LLOG_DEL_RECORD) {
1029                                         rc = llog_cancel_rec(env, loghandle,
1030                                                              tail->lrt_index);
1031                                 }
1032                                 if (rc)
1033                                         GOTO(out, rc);
1034                         }
1035
1036                         /* previous record, still in buffer? */
1037                         --index;
1038                         if (index < first_index)
1039                                 GOTO(out, rc = 0);
1040                         tail = (void *)tail - tail->lrt_len;
1041                 }
1042         }
1043
1044 out:
1045         if (buf != NULL)
1046                 OBD_FREE_LARGE(buf, chunk_size);
1047         RETURN(rc);
1048 }
1049 EXPORT_SYMBOL(llog_reverse_process);
1050
1051 /**
1052  * new llog API
1053  *
1054  * API functions:
1055  *      llog_open - open llog, may not exist
1056  *      llog_exist - check if llog exists
1057  *      llog_close - close opened llog, pair for open, frees llog_handle
1058  *      llog_declare_create - declare llog creation
1059  *      llog_create - create new llog on disk, need transaction handle
1060  *      llog_declare_write_rec - declaration of llog write
1061  *      llog_write_rec - write llog record on disk, need transaction handle
1062  *      llog_declare_add - declare llog catalog record addition
1063  *      llog_add - add llog record in catalog, need transaction handle
1064  */
1065 int llog_exist(struct llog_handle *loghandle)
1066 {
1067         const struct llog_operations *lop;
1068         int rc;
1069
1070         ENTRY;
1071
1072         rc = llog_handle2ops(loghandle, &lop);
1073         if (rc)
1074                 RETURN(rc);
1075         if (lop->lop_exist == NULL)
1076                 RETURN(-EOPNOTSUPP);
1077
1078         rc = lop->lop_exist(loghandle);
1079         RETURN(rc);
1080 }
1081 EXPORT_SYMBOL(llog_exist);
1082
1083 int llog_declare_create(const struct lu_env *env,
1084                         struct llog_handle *loghandle, struct thandle *th)
1085 {
1086         const struct cred *old_cred;
1087         const struct llog_operations *lop;
1088         int rc;
1089
1090         ENTRY;
1091
1092         rc = llog_handle2ops(loghandle, &lop);
1093         if (rc)
1094                 RETURN(rc);
1095         if (lop->lop_declare_create == NULL)
1096                 RETURN(-EOPNOTSUPP);
1097
1098         old_cred = llog_raise_resource();
1099         rc = lop->lop_declare_create(env, loghandle, th);
1100         llog_restore_resource(old_cred);
1101         RETURN(rc);
1102 }
1103
1104 int llog_create(const struct lu_env *env, struct llog_handle *handle,
1105                 struct thandle *th)
1106 {
1107         const struct cred *old_cred;
1108         const struct llog_operations *lop;
1109         int rc;
1110
1111         ENTRY;
1112
1113         rc = llog_handle2ops(handle, &lop);
1114         if (rc)
1115                 RETURN(rc);
1116         if (lop->lop_create == NULL)
1117                 RETURN(-EOPNOTSUPP);
1118
1119         old_cred = llog_raise_resource();
1120         rc = lop->lop_create(env, handle, th);
1121         llog_restore_resource(old_cred);
1122         RETURN(rc);
1123 }
1124
1125 int llog_declare_write_rec(const struct lu_env *env,
1126                            struct llog_handle *handle,
1127                            struct llog_rec_hdr *rec, int idx,
1128                            struct thandle *th)
1129 {
1130         const struct cred *old_cred;
1131         const struct llog_operations *lop;
1132         int rc;
1133
1134         ENTRY;
1135
1136         rc = llog_handle2ops(handle, &lop);
1137         if (rc)
1138                 RETURN(rc);
1139         LASSERT(lop);
1140         if (lop->lop_declare_write_rec == NULL)
1141                 RETURN(-EOPNOTSUPP);
1142
1143         old_cred = llog_raise_resource();
1144         rc = lop->lop_declare_write_rec(env, handle, rec, idx, th);
1145         llog_restore_resource(old_cred);
1146         RETURN(rc);
1147 }
1148
1149 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
1150                    struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1151                    int idx, struct thandle *th)
1152 {
1153         const struct cred *old_cred;
1154         const struct llog_operations *lop;
1155         int rc, buflen;
1156
1157         ENTRY;
1158
1159         /* API sanity checks */
1160         if (handle == NULL) {
1161                 CERROR("loghandle is missed\n");
1162                 RETURN(-EPROTO);
1163         } else if (handle->lgh_obj == NULL) {
1164                 CERROR("loghandle %p with NULL object\n",
1165                         handle);
1166                 RETURN(-EPROTO);
1167         } else if (th == NULL) {
1168                 CERROR("%s: missed transaction handle\n",
1169                        loghandle2name(handle));
1170                 RETURN(-EPROTO);
1171         } else if (handle->lgh_hdr == NULL) {
1172                 CERROR("%s: loghandle %p with no header\n",
1173                        loghandle2name(handle), handle);
1174                 RETURN(-EPROTO);
1175         }
1176
1177         rc = llog_handle2ops(handle, &lop);
1178         if (rc)
1179                 RETURN(rc);
1180
1181         if (lop->lop_write_rec == NULL)
1182                 RETURN(-EOPNOTSUPP);
1183
1184         buflen = rec->lrh_len;
1185         LASSERT(round_up(buflen, 8) == buflen);
1186
1187         old_cred = llog_raise_resource();
1188         rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
1189         llog_restore_resource(old_cred);
1190         RETURN(rc);
1191 }
1192
1193 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
1194              struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
1195              struct thandle *th)
1196 {
1197         const struct cred *old_cred;
1198         int rc;
1199
1200         ENTRY;
1201
1202         if (lgh->lgh_logops->lop_add == NULL)
1203                 RETURN(-EOPNOTSUPP);
1204
1205         old_cred = llog_raise_resource();
1206         rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
1207         llog_restore_resource(old_cred);
1208         RETURN(rc);
1209 }
1210 EXPORT_SYMBOL(llog_add);
1211
1212 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
1213                      struct llog_rec_hdr *rec, struct thandle *th)
1214 {
1215         const struct cred *old_cred;
1216         int rc;
1217
1218         ENTRY;
1219
1220         if (lgh->lgh_logops->lop_declare_add == NULL)
1221                 RETURN(-EOPNOTSUPP);
1222
1223         old_cred = llog_raise_resource();
1224         rc = lgh->lgh_logops->lop_declare_add(env, lgh, rec, th);
1225         llog_restore_resource(old_cred);
1226         RETURN(rc);
1227 }
1228 EXPORT_SYMBOL(llog_declare_add);
1229
1230 /**
1231  * Helper function to open llog or create it if doesn't exist.
1232  * It hides all transaction handling from caller.
1233  */
1234 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
1235                      struct llog_handle **res, struct llog_logid *logid,
1236                      char *name)
1237 {
1238         struct dt_device        *d;
1239         struct thandle          *th;
1240         int                      rc;
1241
1242         ENTRY;
1243
1244         rc = llog_open(env, ctxt, res, logid, name, LLOG_OPEN_NEW);
1245         if (rc)
1246                 RETURN(rc);
1247
1248         if (llog_exist(*res))
1249                 RETURN(0);
1250
1251         LASSERT((*res)->lgh_obj != NULL);
1252
1253         d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
1254
1255         if (unlikely(unlikely(d->dd_rdonly)))
1256                 RETURN(-EROFS);
1257
1258         th = dt_trans_create(env, d);
1259         if (IS_ERR(th))
1260                 GOTO(out, rc = PTR_ERR(th));
1261
1262         /* Create update llog object synchronously, which
1263          * happens during inialization process see
1264          * lod_sub_prep_llog(), to make sure the update
1265          * llog object is created before corss-MDT writing
1266          * updates into the llog object */
1267         if (ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID)
1268                 th->th_sync = 1;
1269
1270         th->th_wait_submit = 1;
1271         rc = llog_declare_create(env, *res, th);
1272         if (rc == 0) {
1273                 rc = dt_trans_start_local(env, d, th);
1274                 if (rc == 0)
1275                         rc = llog_create(env, *res, th);
1276         }
1277         dt_trans_stop(env, d, th);
1278 out:
1279         if (rc)
1280                 llog_close(env, *res);
1281         RETURN(rc);
1282 }
1283 EXPORT_SYMBOL(llog_open_create);
1284
1285 /**
1286  * Helper function to delete existent llog.
1287  */
1288 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
1289                struct llog_logid *logid, char *name)
1290 {
1291         struct llog_handle *handle;
1292         int rc;
1293
1294         ENTRY;
1295
1296         /* nothing to erase */
1297         if (name == NULL && logid == NULL)
1298                 RETURN(0);
1299
1300         rc = llog_open(env, ctxt, &handle, logid, name, LLOG_OPEN_EXISTS);
1301         if (rc < 0)
1302                 RETURN(rc);
1303
1304         rc = llog_destroy(env, handle);
1305         llog_close(env, handle);
1306
1307         RETURN(rc);
1308 }
1309 EXPORT_SYMBOL(llog_erase);
1310
1311 /*
1312  * Helper function for write record in llog.
1313  * It hides all transaction handling from caller.
1314  * Valid only with local llog.
1315  */
1316 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
1317                struct llog_rec_hdr *rec, int idx)
1318 {
1319         struct dt_device        *dt;
1320         struct thandle          *th;
1321         bool                    need_cookie;
1322         int                     rc;
1323
1324         ENTRY;
1325
1326         LASSERT(loghandle);
1327         LASSERT(loghandle->lgh_ctxt);
1328         LASSERT(loghandle->lgh_obj != NULL);
1329
1330         dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
1331
1332         if (unlikely(unlikely(dt->dd_rdonly)))
1333                 RETURN(-EROFS);
1334
1335         th = dt_trans_create(env, dt);
1336         if (IS_ERR(th))
1337                 RETURN(PTR_ERR(th));
1338
1339         rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
1340         if (rc)
1341                 GOTO(out_trans, rc);
1342
1343         th->th_wait_submit = 1;
1344         rc = dt_trans_start_local(env, dt, th);
1345         if (rc)
1346                 GOTO(out_trans, rc);
1347
1348         need_cookie = !(idx == LLOG_HEADER_IDX || idx == LLOG_NEXT_IDX);
1349
1350         down_write(&loghandle->lgh_lock);
1351         if (need_cookie) {
1352                 struct llog_thread_info *lti = llog_info(env);
1353
1354                 /* cookie comes from llog_process_thread */
1355                 rc = llog_write_rec(env, loghandle, rec, &lti->lgi_cookie,
1356                                     rec->lrh_index, th);
1357                 /* upper layer didn`t pass cookie so change rc */
1358                 rc = (rc == 1 ? 0 : rc);
1359         } else {
1360                 rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
1361         }
1362
1363         up_write(&loghandle->lgh_lock);
1364 out_trans:
1365         dt_trans_stop(env, dt, th);
1366         RETURN(rc);
1367 }
1368 EXPORT_SYMBOL(llog_write);
1369
1370 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
1371               struct llog_handle **lgh, struct llog_logid *logid,
1372               char *name, enum llog_open_param open_param)
1373 {
1374         const struct cred *old_cred;
1375         int      rc;
1376
1377         ENTRY;
1378
1379         LASSERT(ctxt);
1380         LASSERT(ctxt->loc_logops);
1381
1382         if (ctxt->loc_logops->lop_open == NULL) {
1383                 *lgh = NULL;
1384                 RETURN(-EOPNOTSUPP);
1385         }
1386
1387         *lgh = llog_alloc_handle();
1388         if (*lgh == NULL)
1389                 RETURN(-ENOMEM);
1390         (*lgh)->lgh_ctxt = ctxt;
1391         (*lgh)->lgh_logops = ctxt->loc_logops;
1392
1393         old_cred = llog_raise_resource();
1394         rc = ctxt->loc_logops->lop_open(env, *lgh, logid, name, open_param);
1395         llog_restore_resource(old_cred);
1396         if (rc) {
1397                 llog_free_handle(*lgh);
1398                 *lgh = NULL;
1399         }
1400         RETURN(rc);
1401 }
1402 EXPORT_SYMBOL(llog_open);
1403
1404 int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
1405 {
1406         return llog_handle_put(env, loghandle);
1407 }
1408 EXPORT_SYMBOL(llog_close);
1409
1410 /**
1411  * Helper function to get the llog size in records. It is used by MGS
1412  * mostly to check that config llog exists and contains data.
1413  *
1414  * \param[in] env       execution environment
1415  * \param[in] ctxt      llog context
1416  * \param[in] name      llog name
1417  *
1418  * \retval              true if there are records in llog besides a header
1419  * \retval              false on error or llog without records
1420  */
1421 int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
1422                   char *name)
1423 {
1424         struct llog_handle      *llh;
1425         int                      rc = 0;
1426
1427         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1428         if (rc < 0) {
1429                 if (likely(rc == -ENOENT))
1430                         rc = 0;
1431                 GOTO(out, rc);
1432         }
1433
1434         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1435         if (rc)
1436                 GOTO(out_close, rc);
1437         rc = llog_get_size(llh);
1438
1439 out_close:
1440         llog_close(env, llh);
1441 out:
1442         /* The header is record 1, the llog is still considered as empty
1443          * if there is only header */
1444         return (rc <= 1);
1445 }
1446 EXPORT_SYMBOL(llog_is_empty);
1447
1448 /* this callback run in raw read mode (canceled record are processed) */
1449 int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
1450                       struct llog_rec_hdr *rec, void *data)
1451 {
1452         struct llog_handle *copy_llh = data;
1453         int idx = rec->lrh_index;
1454         int rc;
1455
1456         ENTRY;
1457
1458         /* Append all records */
1459         rc = llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
1460
1461         /* Cancel the record if it is canceled on the source */
1462         if (!rc && !test_bit_le(idx, LLOG_HDR_BITMAP(llh->lgh_hdr)))
1463                 rc = llog_cancel_rec(env, copy_llh, copy_llh->lgh_last_idx);
1464
1465         RETURN(rc);
1466 }
1467
1468 /* backup plain llog */
1469 int llog_backup(const struct lu_env *env, struct obd_device *obd,
1470                 struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
1471                 char *name, char *backup)
1472 {
1473         struct llog_handle *llh, *bllh;
1474         struct llog_process_cat_data cd = {0};
1475         int rc;
1476
1477         ENTRY;
1478
1479         /* open original log */
1480         rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
1481         if (rc < 0) {
1482                 /* the -ENOENT case is also reported to the caller
1483                  * but silently so it should handle that if needed.
1484                  */
1485                 if (rc != -ENOENT)
1486                         CERROR("%s: failed to open log %s: rc = %d\n",
1487                                obd->obd_name, name, rc);
1488                 RETURN(rc);
1489         }
1490
1491         rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
1492         if (rc)
1493                 GOTO(out_close, rc);
1494
1495         /* Make sure there's no old backup log */
1496         rc = llog_erase(env, bctxt, NULL, backup);
1497         if (rc < 0 && rc != -ENOENT)
1498                 GOTO(out_close, rc);
1499
1500         /* open backup log */
1501         rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
1502         if (rc) {
1503                 CERROR("%s: failed to open backup logfile %s: rc = %d\n",
1504                        obd->obd_name, backup, rc);
1505                 GOTO(out_close, rc);
1506         }
1507
1508         /* check that backup llog is not the same object as original one */
1509         if (llh->lgh_obj == bllh->lgh_obj) {
1510                 CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
1511                        obd->obd_name, name, backup, llh->lgh_obj,
1512                        bllh->lgh_obj);
1513                 GOTO(out_backup, rc = -EEXIST);
1514         }
1515
1516         rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
1517         if (rc)
1518                 GOTO(out_backup, rc);
1519
1520         /* Read canceled records to have an exact copy */
1521         cd.lpcd_read_mode = LLOG_READ_MODE_RAW;
1522         /* Copy log record by record */
1523         rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
1524                                   &cd, false);
1525         if (rc)
1526                 CERROR("%s: failed to backup log %s: rc = %d\n",
1527                        obd->obd_name, name, rc);
1528 out_backup:
1529         llog_close(env, bllh);
1530 out_close:
1531         llog_close(env, llh);
1532         RETURN(rc);
1533 }
1534 EXPORT_SYMBOL(llog_backup);
1535
1536 /* Get size of llog */
1537 __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
1538 {
1539         int rc;
1540         struct lu_attr la;
1541
1542         rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
1543         if (rc) {
1544                 CERROR("%s: attr_get failed for "DFID": rc = %d\n",
1545                        loghandle2name(llh), PLOGID(&llh->lgh_id), rc);
1546                 return 0;
1547         }
1548
1549         return la.la_size;
1550 }
1551 EXPORT_SYMBOL(llog_size);
1552
1553 /* set llog ctime to current, and set LLOG_F_RM_ON_ERR|LLOG_F_MAX_AGE flag in
1554  * log header. It will be reclaimed when expired (UPDATE_LOG_MAX_AGE old).
1555  */
1556 int llog_retain(const struct lu_env *env, struct llog_handle *log)
1557 {
1558         struct dt_object *dto = log->lgh_obj;
1559         struct dt_device *dt = lu2dt_dev(log->lgh_obj->do_lu.lo_dev);
1560         struct lu_attr la = { 0 };
1561         struct thandle *th;
1562         int rc;
1563
1564         la.la_ctime = ktime_get_real_seconds();
1565         la.la_valid = LA_CTIME;
1566
1567         th = dt_trans_create(env, dt);
1568         if (IS_ERR(th))
1569                 return PTR_ERR(th);
1570
1571         th->th_wait_submit = 1;
1572         log->lgh_hdr->llh_flags |= LLOG_F_MAX_AGE | LLOG_F_RM_ON_ERR;
1573         rc = llog_declare_write_rec(env, log, &log->lgh_hdr->llh_hdr, -1, th);
1574         if (rc)
1575                 goto out_trans;
1576
1577         rc = dt_declare_attr_set(env, dto, &la, th);
1578         if (rc)
1579                 goto out_trans;
1580
1581         rc = dt_trans_start_local(env, dt, th);
1582         if (rc)
1583                 goto out_trans;
1584
1585         rc = llog_write_rec(env, log, &log->lgh_hdr->llh_hdr, NULL,
1586                             LLOG_HEADER_IDX, th);
1587         if (rc)
1588                 goto out_trans;
1589
1590         rc = dt_attr_set(env, dto, &la, th);
1591 out_trans:
1592         dt_trans_stop(env, dt, th);
1593
1594         CDEBUG(D_OTHER, "retain log "DFID" rc = %d\n",
1595                PLOGID(&log->lgh_id), rc);
1596         return rc;
1597 }
1598 EXPORT_SYMBOL(llog_retain);