Whamcloud - gitweb
LU-17000 utils: handle_yaml_no_op() has wrong signature
[fs/lustre-release.git] / lustre / obdclass / llog_cat.c
index ec28fa8..93b9ac1 100644 (file)
@@ -1,35 +1,14 @@
-/*
- * GPL HEADER START
- *
- * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 only,
- * as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License version 2 for more details (a copy is included
- * in the LICENSE file that accompanied this code).
- *
- * You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see
- * http://www.gnu.org/licenses/gpl-2.0.html
- *
- * GPL HEADER END
- */
+// SPDX-License-Identifier: GPL-2.0
+
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
  * Copyright (c) 2012, 2017, Intel Corporation.
  */
+
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
- *
- * lustre/obdclass/llog_cat.c
  *
  * OST<->MDS recovery logging infrastructure.
  *
 
 #include "llog_internal.h"
 
+
+/**
+ * lockdep markers for nested struct llog_handle::lgh_lock locking.
+ */
+enum {
+       LLOGH_CAT,
+       LLOGH_LOG,
+};
+
 /* Create a new log handle and add it to the open list.
  * This log handle will be closed when all of the records in it are removed.
  *
@@ -68,9 +56,7 @@ static int llog_cat_new_log(const struct lu_env *env,
 
        ENTRY;
 
-       index = (cathandle->lgh_last_idx + 1) %
-               (OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) ? (cfs_fail_val + 1) :
-                                               LLOG_HDR_BITMAP_SIZE(llh));
+       index = (cathandle->lgh_last_idx + 1) % (llog_max_idx(llh) + 1);
 
        /* check that new llog index will not overlap with the first one.
         * - llh_cat_idx is the index just before the first/oldest still in-use
@@ -86,21 +72,17 @@ static int llog_cat_new_log(const struct lu_env *env,
        if ((index == llh->llh_cat_idx + 1 && llh->llh_count > 1) ||
            (index == 0 && llh->llh_cat_idx == 0)) {
                if (cathandle->lgh_name == NULL) {
-                       CWARN("%s: there are no more free slots in catalog "
-                             DFID":%x\n",
-                             loghandle->lgh_ctxt->loc_obd->obd_name,
-                             PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
-                             cathandle->lgh_id.lgl_ogen);
+                       CWARN("%s: there are no more free slots in catalog "DFID"\n",
+                             loghandle2name(loghandle),
+                             PLOGID(&cathandle->lgh_id));
                } else {
-                       CWARN("%s: there are no more free slots in "
-                             "catalog %s\n",
-                             loghandle->lgh_ctxt->loc_obd->obd_name,
-                             cathandle->lgh_name);
+                       CWARN("%s: there are no more free slots in catalog %s\n",
+                             loghandle2name(loghandle), cathandle->lgh_name);
                }
                RETURN(-ENOSPC);
        }
 
-       if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
+       if (CFS_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
                RETURN(-ENOSPC);
 
        if (loghandle->lgh_hdr != NULL) {
@@ -139,6 +121,7 @@ static int llog_cat_new_log(const struct lu_env *env,
                                            handle);
                if (rc != 0)
                        GOTO(out, rc);
+               dt_declare_attr_set(env, cathandle->lgh_obj, NULL, handle);
 
                rc = dt_trans_start_local(env, dt, handle);
                if (rc != 0)
@@ -153,11 +136,12 @@ static int llog_cat_new_log(const struct lu_env *env,
                GOTO(out, rc = 0);
        } else if (rc != 0) {
                CERROR("%s: can't create new plain llog in catalog: rc = %d\n",
-                      loghandle->lgh_ctxt->loc_obd->obd_name, rc);
+                      loghandle2name(loghandle), rc);
                GOTO(out, rc);
        }
 
-       rc = llog_init_handle(env, loghandle,
+       rc = llog_init_handle(env, loghandle, (cathandle->lgh_hdr->llh_flags &
+                             LLOG_F_EXT_MASK) |
                              LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
                              &cathandle->lgh_hdr->llh_tgtuuid);
        if (rc < 0)
@@ -174,10 +158,14 @@ static int llog_cat_new_log(const struct lu_env *env,
                            &loghandle->u.phd.phd_cookie, LLOG_NEXT_IDX, th);
        if (rc < 0)
                GOTO(out_destroy, rc);
+       /* update for catalog which doesn't happen very often */
+       lgi->lgi_attr.la_valid = LA_MTIME;
+       lgi->lgi_attr.la_mtime = ktime_get_real_seconds();
+       dt_attr_set(env, cathandle->lgh_obj, &lgi->lgi_attr, th);
 
        CDEBUG(D_OTHER, "new plain log "DFID".%u of catalog "DFID"\n",
-              PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rec->lid_hdr.lrh_index,
-              PFID(&cathandle->lgh_id.lgl_oi.oi_fid));
+              PLOGID(&loghandle->lgh_id), rec->lid_hdr.lrh_index,
+              PLOGID(&cathandle->lgh_id));
 
        loghandle->lgh_hdr->llh_cat_idx = rec->lid_hdr.lrh_index;
 
@@ -197,6 +185,12 @@ static int llog_cat_new_log(const struct lu_env *env,
                if (freespace > (128 << 20))
                        loghandle->lgh_max_size = 128 << 20;
        }
+       if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PLAIN_RECORDS) ||
+                    CFS_FAIL_PRECHECK(OBD_FAIL_CATALOG_FULL_CHECK))) {
+               // limit the numer of plain records for test
+               loghandle->lgh_max_size = loghandle->lgh_hdr_size +
+                      cfs_fail_val * 64;
+       }
        rc = 0;
 
 out:
@@ -213,11 +207,145 @@ out_destroy:
        loghandle->lgh_hdr->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
        /* this is to mimic full log, so another llog_cat_current_log()
         * can skip it and ask for another onet */
-       loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(llh) + 1;
+       loghandle->lgh_last_idx = llog_max_idx(loghandle->lgh_hdr) + 1;
        llog_trans_destroy(env, loghandle, th);
+       if (handle != NULL)
+               dt_trans_stop(env, dt, handle);
        RETURN(rc);
 }
 
+static int llog_cat_refresh(const struct lu_env *env,
+                           struct llog_handle *cathandle)
+{
+       struct llog_handle *loghandle;
+       int rc;
+
+       LASSERT(rwsem_is_locked(&cathandle->lgh_lock));
+
+       list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
+                           u.phd.phd_entry) {
+               if (!llog_exist(loghandle))
+                       continue;
+
+               down_write(&loghandle->lgh_lock);
+               rc = llog_read_header(env, loghandle, NULL);
+               up_write(&loghandle->lgh_lock);
+               if (rc)
+                       goto out;
+       }
+
+       rc = llog_read_header(env, cathandle, NULL);
+out:
+       return rc;
+}
+
+static inline int llog_cat_declare_create(const struct lu_env *env,
+                                         struct llog_handle *cathandle,
+                                         struct llog_handle *loghandle,
+                                         struct thandle *th)
+{
+
+       struct llog_thread_info *lgi = llog_info(env);
+       struct llog_logid_rec *lirec = &lgi->lgi_logid;
+       int rc;
+
+       if (dt_object_remote(cathandle->lgh_obj)) {
+               down_write(&loghandle->lgh_lock);
+               if (!llog_exist(loghandle))
+                       rc = llog_cat_new_log(env, cathandle, loghandle, NULL);
+               else
+                       rc = 0;
+               up_write(&loghandle->lgh_lock);
+       } else {
+
+               rc = llog_declare_create(env, loghandle, th);
+               if (rc)
+                       return rc;
+
+               lirec->lid_hdr.lrh_len = sizeof(*lirec);
+               rc = llog_declare_write_rec(env, cathandle, &lirec->lid_hdr, -1,
+                                           th);
+               if (!rc)
+                       dt_declare_attr_set(env, cathandle->lgh_obj, NULL, th);
+       }
+       return rc;
+}
+/*
+ * prepare current/next log for catalog.
+ *
+ * if \a *ploghandle is NULL, open it, and declare create, NB, if \a
+ * *ploghandle is remote, create it synchronously here, see comments
+ * below.
+ *
+ * \a cathandle->lgh_lock is down_read-ed, it gets down_write-ed if \a
+ * *ploghandle has to be opened.
+ */
+static int llog_cat_prep_log(const struct lu_env *env,
+                            struct llog_handle *cathandle,
+                            struct llog_handle **ploghandle,
+                            struct thandle *th)
+{
+       struct llog_handle *loghandle;
+       int rc;
+
+       rc = 0;
+       loghandle = *ploghandle;
+       if (!IS_ERR_OR_NULL(loghandle)) {
+               loghandle = llog_handle_get(loghandle);
+               if (loghandle) {
+                       if (llog_exist(loghandle) == 0)
+                               rc = llog_cat_declare_create(env, cathandle,
+                                                            loghandle, th);
+                       llog_handle_put(env, loghandle);
+               }
+               return rc;
+       }
+
+       down_write(&cathandle->lgh_lock);
+       if (!IS_ERR_OR_NULL(*ploghandle)) {
+               loghandle = *ploghandle;
+               up_write(&cathandle->lgh_lock);
+               loghandle = llog_handle_get(loghandle);
+               if (loghandle) {
+                       if (llog_exist(loghandle) == 0)
+                               rc = llog_cat_declare_create(env, cathandle,
+                                                            loghandle, th);
+                       llog_handle_put(env, loghandle);
+               }
+               return rc;
+       }
+
+       /* Slow path with open/create declare, only one thread do all stuff
+        * and share loghandle at the end
+        */
+       rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, NULL, NULL,
+                      LLOG_OPEN_NEW);
+       if (rc) {
+               up_write(&cathandle->lgh_lock);
+               CDEBUG(D_OTHER, "%s: failed to open log, catalog "DFID" %d\n",
+                      loghandle2name(cathandle), PLOGID(&cathandle->lgh_id),
+                      rc);
+               return rc;
+       }
+
+       rc = llog_cat_declare_create(env, cathandle, loghandle, th);
+       if (!rc) {
+               list_add(&loghandle->u.phd.phd_entry,
+                        &cathandle->u.chd.chd_head);
+               *ploghandle = loghandle;
+       }
+
+       up_write(&cathandle->lgh_lock);
+       CDEBUG(D_OTHER, "%s: open log "DFID" for catalog "DFID" rc=%d\n",
+              loghandle2name(cathandle), PLOGID(&loghandle->lgh_id),
+              PLOGID(&cathandle->lgh_id), rc);
+
+       if (rc)
+               llog_close(env, loghandle);
+
+       return rc;
+}
+
 /* Open an existent log handle and add it to the open list.
  * This log handle will be closed when all of the records in it are removed.
  *
@@ -240,56 +368,54 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
                RETURN(-EBADF);
 
        fmt = cathandle->lgh_hdr->llh_flags & LLOG_F_EXT_MASK;
-       down_write(&cathandle->lgh_lock);
+       down_read(&cathandle->lgh_lock);
        list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
                            u.phd.phd_entry) {
                struct llog_logid *cgl = &loghandle->lgh_id;
 
                if (ostid_id(&cgl->lgl_oi) == ostid_id(&logid->lgl_oi) &&
                    ostid_seq(&cgl->lgl_oi) == ostid_seq(&logid->lgl_oi)) {
-                       if (cgl->lgl_ogen != logid->lgl_ogen) {
-                               CWARN("%s: log "DFID" generation %x != %x\n",
-                                     loghandle->lgh_ctxt->loc_obd->obd_name,
-                                     PFID(&logid->lgl_oi.oi_fid),
-                                     cgl->lgl_ogen, logid->lgl_ogen);
+                       *res = llog_handle_get(loghandle);
+                       if (!*res) {
+                               CERROR("%s: log "DFID" refcount is zero!\n",
+                                      loghandle2name(loghandle),
+                                      PLOGID(logid));
                                continue;
                        }
                        loghandle->u.phd.phd_cat_handle = cathandle;
-                       up_write(&cathandle->lgh_lock);
-                       GOTO(out, rc = 0);
+                       up_read(&cathandle->lgh_lock);
+                       RETURN(rc);
                }
        }
-       up_write(&cathandle->lgh_lock);
+       up_read(&cathandle->lgh_lock);
 
        rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, logid, NULL,
                       LLOG_OPEN_EXISTS);
        if (rc < 0) {
-               CERROR("%s: error opening log id "DFID":%x: rc = %d\n",
-                      cathandle->lgh_ctxt->loc_obd->obd_name,
-                      PFID(&logid->lgl_oi.oi_fid), logid->lgl_ogen, rc);
+               CERROR("%s: error opening log id "DFID": rc = %d\n",
+                      loghandle2name(cathandle), PLOGID(logid), rc);
                RETURN(rc);
        }
 
-       rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN | fmt, NULL);
+       rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN |
+                             LLOG_F_ZAP_WHEN_EMPTY | fmt, NULL);
        if (rc < 0) {
                llog_close(env, loghandle);
-               loghandle = NULL;
+               *res = NULL;
                RETURN(rc);
        }
 
+       *res = llog_handle_get(loghandle);
+       LASSERT(*res);
        down_write(&cathandle->lgh_lock);
-       list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
+       list_add(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
        up_write(&cathandle->lgh_lock);
 
        loghandle->u.phd.phd_cat_handle = cathandle;
        loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
        loghandle->u.phd.phd_cookie.lgc_index =
                                loghandle->lgh_hdr->llh_cat_idx;
-       EXIT;
-out:
-       llog_handle_get(loghandle);
-       *res = loghandle;
-       return 0;
+       RETURN(0);
 }
 
 int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
@@ -314,8 +440,7 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
                        if (rc)
                                CERROR("%s: failure destroying log during "
                                       "cleanup: rc = %d\n",
-                                      loghandle->lgh_ctxt->loc_obd->obd_name,
-                                      rc);
+                                      loghandle2name(loghandle), rc);
 
                        index = loghandle->u.phd.phd_cookie.lgc_index;
                        llog_cat_cleanup(env, cathandle, NULL, index);
@@ -330,14 +455,6 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
 }
 EXPORT_SYMBOL(llog_cat_close);
 
-/**
- * lockdep markers for nested struct llog_handle::lgh_lock locking.
- */
-enum {
-        LLOGH_CAT,
-        LLOGH_LOG
-};
-
 /** Return the currently active log handle.  If the current log handle doesn't
  * have enough space left for the current record, start a new one.
  *
@@ -351,73 +468,70 @@ enum {
 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
                                                struct thandle *th)
 {
-        struct llog_handle *loghandle = NULL;
-        ENTRY;
+       struct llog_handle *loghandle = NULL;
 
+       ENTRY;
 
-       if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
-               down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
+       if (CFS_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
+               loghandle = cathandle->u.chd.chd_current_log;
                GOTO(next, loghandle);
        }
 
-       down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
-        loghandle = cathandle->u.chd.chd_current_log;
-        if (loghandle) {
-               struct llog_log_hdr *llh;
-
-               down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
-               llh = loghandle->lgh_hdr;
-               if (llh == NULL || !llog_is_full(loghandle)) {
-                       up_read(&cathandle->lgh_lock);
-                        RETURN(loghandle);
-                } else {
-                       up_write(&loghandle->lgh_lock);
-                }
-        }
-       up_read(&cathandle->lgh_lock);
-
-       /* time to use next log */
-
-       /* first, we have to make sure the state hasn't changed */
-       down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
+retry:
        loghandle = cathandle->u.chd.chd_current_log;
-       if (loghandle) {
+       if (likely(loghandle)) {
                struct llog_log_hdr *llh;
 
                down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
                llh = loghandle->lgh_hdr;
                if (llh == NULL || !llog_is_full(loghandle))
-                       GOTO(out_unlock, loghandle);
+                       RETURN(loghandle);
                else
                        up_write(&loghandle->lgh_lock);
        }
 
+       /* time to use next log */
 next:
-       /* Sigh, the chd_next_log and chd_current_log is initialized
-        * in declare phase, and we do not serialize the catlog
-        * accessing, so it might be possible the llog creation
-        * thread (see llog_cat_declare_add_rec()) did not create
-        * llog successfully, then the following thread might
-        * meet this situation. */
-       if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) {
-               CERROR("%s: next log does not exist!\n",
-                      cathandle->lgh_ctxt->loc_obd->obd_name);
-               loghandle = ERR_PTR(-EIO);
-               if (cathandle->u.chd.chd_next_log == NULL) {
-                       /* Store the error in chd_next_log, so
-                        * the following process can get correct
-                        * failure value */
-                       cathandle->u.chd.chd_next_log = loghandle;
+       /* first, we have to make sure the state hasn't changed */
+       down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
+       if (unlikely(loghandle == cathandle->u.chd.chd_current_log)) {
+               struct llog_logid lid = {.lgl_oi.oi.oi_id = 0,
+                                        .lgl_oi.oi.oi_seq = 0,
+                                        .lgl_ogen = 0};
+               /* Sigh, the chd_next_log and chd_current_log is initialized
+                * in declare phase, and we do not serialize the catlog
+                * accessing, so it might be possible the llog creation
+                * thread (see llog_cat_declare_add_rec()) did not create
+                * llog successfully, then the following thread might
+                * meet this situation.
+                */
+               if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) {
+                       CERROR("%s: next log does not exist, catalog "DFID" rc=%d\n",
+                              loghandle2name(cathandle),
+                              PLOGID(&cathandle->lgh_id), -EIO);
+                       loghandle = ERR_PTR(-EIO);
+                       if (cathandle->u.chd.chd_next_log == NULL) {
+                               /* Store the error in chd_next_log, so
+                                * the following process can get correct
+                                * failure value
+                                */
+                               cathandle->u.chd.chd_next_log = loghandle;
+                       }
+                       GOTO(out_unlock, loghandle);
                }
-               GOTO(out_unlock, loghandle);
+               if (!IS_ERR_OR_NULL(loghandle))
+                       lid = loghandle->lgh_id;
+
+               CDEBUG(D_OTHER, "%s: use next log "DFID"->"DFID" catalog "DFID"\n",
+                      loghandle2name(cathandle), PLOGID(&lid),
+                      PLOGID(&cathandle->u.chd.chd_next_log->lgh_id),
+                      PLOGID(&cathandle->lgh_id));
+               loghandle = cathandle->u.chd.chd_next_log;
+               cathandle->u.chd.chd_current_log = loghandle;
+               cathandle->u.chd.chd_next_log = NULL;
        }
-
-       CDEBUG(D_INODE, "use next log\n");
-
-       loghandle = cathandle->u.chd.chd_next_log;
-       cathandle->u.chd.chd_current_log = loghandle;
-       cathandle->u.chd.chd_next_log = NULL;
-       down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
+       up_write(&cathandle->lgh_lock);
+       GOTO(retry, loghandle);
 
 out_unlock:
        up_write(&cathandle->lgh_lock);
@@ -425,40 +539,6 @@ out_unlock:
        RETURN(loghandle);
 }
 
-static int llog_cat_update_header(const struct lu_env *env,
-                          struct llog_handle *cathandle)
-{
-       struct llog_handle *loghandle;
-       int rc;
-       ENTRY;
-
-       /* refresh llog */
-       down_write(&cathandle->lgh_lock);
-       if (!cathandle->lgh_stale) {
-               up_write(&cathandle->lgh_lock);
-               RETURN(0);
-       }
-       list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
-                           u.phd.phd_entry) {
-               if (!llog_exist(loghandle))
-                       continue;
-
-               rc = llog_read_header(env, loghandle, NULL);
-               if (rc != 0) {
-                       up_write(&cathandle->lgh_lock);
-                       GOTO(out, rc);
-               }
-       }
-       rc = llog_read_header(env, cathandle, NULL);
-       if (rc == 0)
-               cathandle->lgh_stale = 0;
-       up_write(&cathandle->lgh_lock);
-       if (rc != 0)
-               GOTO(out, rc);
-out:
-       RETURN(rc);
-}
-
 /* Add a single record to the recovery log(s) using a catalog
  * Returns as llog_write_record
  *
@@ -468,7 +548,8 @@ int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
                     struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
                     struct thandle *th)
 {
-        struct llog_handle *loghandle;
+       struct llog_handle *loghandle;
+       struct llog_thread_info *lgi = llog_info(env);
        int rc, retried = 0;
        ENTRY;
 
@@ -484,19 +565,26 @@ retry:
                rc = llog_cat_new_log(env, cathandle, loghandle, th);
                if (rc < 0) {
                        up_write(&loghandle->lgh_lock);
-                       /* nobody should be trying to use this llog */
-                       down_write(&cathandle->lgh_lock);
-                       /* only reset current log if still room in catalog, to
-                        * avoid unnecessarily and racy creation of new and
-                        * partially initialized llog_handle
+                       /* When ENOSPC happened no need to drop loghandle
+                        * a new one would be allocated anyway for next llog_add
+                        * so better to stay with the old.
                         */
-                       if ((cathandle->u.chd.chd_current_log == loghandle) &&
-                           rc != -ENOSPC)
-                               cathandle->u.chd.chd_current_log = NULL;
-                       up_write(&cathandle->lgh_lock);
+                       if (rc != -ENOSPC) {
+                               /* nobody should be trying to use this llog */
+                               down_write(&cathandle->lgh_lock);
+                               if (cathandle->u.chd.chd_current_log ==
+                                   loghandle)
+                                       cathandle->u.chd.chd_current_log = NULL;
+                               list_del_init(&loghandle->u.phd.phd_entry);
+                               up_write(&cathandle->lgh_lock);
+                               llog_close(env, loghandle);
+                       }
+                       CERROR("%s: initialization error: rc = %d\n",
+                              loghandle2name(cathandle), rc);
                        RETURN(rc);
                }
        }
+
        /* now let's try to add the record */
        rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th);
        if (rc < 0) {
@@ -510,14 +598,24 @@ retry:
                 * actual cause here */
                if (rc == -ENOSPC && llog_is_full(loghandle))
                        rc = -ENOBUFS;
+       } else {
+               unsigned long timestamp = ktime_get_real_seconds();
+               if (timestamp != loghandle->lgh_timestamp) {
+                       loghandle->lgh_timestamp = timestamp;
+                       lgi->lgi_attr.la_valid = LA_MTIME;
+                       lgi->lgi_attr.la_mtime = timestamp;
+                       dt_attr_set(env, loghandle->lgh_obj, &lgi->lgi_attr, th);
+               }
        }
-       up_write(&loghandle->lgh_lock);
+       /* llog_write_rec could unlock a semaphore */
+       if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_UNLCK_SEM))
+               up_write(&loghandle->lgh_lock);
 
        if (rc == -ENOBUFS) {
                if (retried++ == 0)
                        GOTO(retry, rc);
                CERROR("%s: error on 2nd llog: rc = %d\n",
-                      cathandle->lgh_ctxt->loc_obd->obd_name, rc);
+                      loghandle2name(cathandle), rc);
        }
 
        RETURN(rc);
@@ -528,167 +626,70 @@ int llog_cat_declare_add_rec(const struct lu_env *env,
                             struct llog_handle *cathandle,
                             struct llog_rec_hdr *rec, struct thandle *th)
 {
-       struct llog_thread_info *lgi = llog_info(env);
-       struct llog_logid_rec   *lirec = &lgi->lgi_logid;
-       struct llog_handle      *loghandle, *next;
-       int                      rc = 0;
+       struct llog_handle *loghandle = NULL;
+       int retries = 5;
+       int rc;
 
        ENTRY;
 
-       if (cathandle->u.chd.chd_current_log == NULL) {
-               /* declare new plain llog */
-               down_write(&cathandle->lgh_lock);
-               if (cathandle->u.chd.chd_current_log == NULL) {
-                       rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
-                                      NULL, NULL, LLOG_OPEN_NEW);
-                       if (rc == 0) {
-                               cathandle->u.chd.chd_current_log = loghandle;
-                               list_add_tail(&loghandle->u.phd.phd_entry,
-                                             &cathandle->u.chd.chd_head);
-                       }
-               }
-               up_write(&cathandle->lgh_lock);
-       } else if (cathandle->u.chd.chd_next_log == NULL ||
-                  IS_ERR(cathandle->u.chd.chd_next_log)) {
-               /* declare next plain llog */
-               down_write(&cathandle->lgh_lock);
-               if (cathandle->u.chd.chd_next_log == NULL ||
-                   IS_ERR(cathandle->u.chd.chd_next_log)) {
-                       rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
-                                      NULL, NULL, LLOG_OPEN_NEW);
-                       if (rc == 0) {
-                               cathandle->u.chd.chd_next_log = loghandle;
-                               list_add_tail(&loghandle->u.phd.phd_entry,
-                                             &cathandle->u.chd.chd_head);
-                       }
-               }
-               up_write(&cathandle->lgh_lock);
-       }
+start:
+       CDEBUG(D_INFO, "Declare adding to "DOSTID" flags %x count %d\n",
+              POSTID(&cathandle->lgh_id.lgl_oi),
+              cathandle->lgh_hdr->llh_flags, cathandle->lgh_hdr->llh_count);
+
+
+       rc = llog_cat_prep_log(env, cathandle,
+                              &cathandle->u.chd.chd_current_log, th);
        if (rc)
-               GOTO(out, rc);
+               GOTO(estale, rc);
 
-       lirec->lid_hdr.lrh_len = sizeof(*lirec);
+       loghandle = cathandle->u.chd.chd_current_log;
+       if (IS_ERR_OR_NULL(loghandle)) { /* low chance race, repeat */
+               GOTO(estale, rc = -ESTALE);
+       } else {
+               loghandle = llog_handle_get(loghandle);
+               if (!loghandle)
+                       GOTO(estale, rc = -ESTALE);
+       }
 
-       if (!llog_exist(cathandle->u.chd.chd_current_log)) {
-               if (dt_object_remote(cathandle->lgh_obj)) {
-                       /* For remote operation, if we put the llog object
-                        * creation in the current transaction, then the
-                        * llog object will not be created on the remote
-                        * target until the transaction stop, if other
-                        * operations start before the transaction stop,
-                        * and use the same llog object, will be dependent
-                        * on the success of this transaction. So let's
-                        * create the llog object synchronously here to
-                        * remove the dependency. */
-create_again:
-                       down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
-                       loghandle = cathandle->u.chd.chd_current_log;
-                       down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
-                       if (cathandle->lgh_stale) {
-                               up_write(&loghandle->lgh_lock);
-                               up_read(&cathandle->lgh_lock);
-                               GOTO(out, rc = -EIO);
-                       }
-                       if (!llog_exist(loghandle)) {
-                               rc = llog_cat_new_log(env, cathandle, loghandle,
-                                                     NULL);
-                               if (rc == -ESTALE)
-                                       cathandle->lgh_stale = 1;
-                       }
-                       up_write(&loghandle->lgh_lock);
-                       up_read(&cathandle->lgh_lock);
-                       if (rc == -ESTALE) {
-                               rc = llog_cat_update_header(env, cathandle);
-                               if (rc != 0)
-                                       GOTO(out, rc);
-                               goto create_again;
-                       } else if (rc < 0) {
-                               GOTO(out, rc);
-                       }
-               } else {
-                       rc = llog_declare_create(env,
-                                       cathandle->u.chd.chd_current_log, th);
-                       if (rc)
-                               GOTO(out, rc);
-                       llog_declare_write_rec(env, cathandle,
-                                              &lirec->lid_hdr, -1, th);
-               }
+       /* For local llog this would always reserves credits for creation */
+       rc = llog_cat_prep_log(env, cathandle, &cathandle->u.chd.chd_next_log,
+                              th);
+       if (!rc) {
+               rc = llog_declare_write_rec(env, loghandle, rec, -1, th);
+               if (!rc)
+                       dt_declare_attr_set(env, loghandle->lgh_obj, NULL, th);
        }
 
-write_again:
-       /* declare records in the llogs */
-       rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
-                                   rec, -1, th);
+       llog_handle_put(env, loghandle);
+estale:
        if (rc == -ESTALE) {
-               down_write(&cathandle->lgh_lock);
-               if (cathandle->lgh_stale) {
+               if (dt_object_remote(cathandle->lgh_obj)) {
+                       down_write(&cathandle->lgh_lock);
+                       rc = llog_cat_refresh(env, cathandle);
                        up_write(&cathandle->lgh_lock);
-                       GOTO(out, rc = -EIO);
+                       if (rc)
+                               RETURN(rc);
                }
-
-               cathandle->lgh_stale = 1;
-               up_write(&cathandle->lgh_lock);
-               rc = llog_cat_update_header(env, cathandle);
-               if (rc != 0)
-                       GOTO(out, rc);
-               goto write_again;
-       } else if (rc < 0) {
-               GOTO(out, rc);
+               retries--;
+               if (retries > 0)
+                       goto start;
        }
 
-       next = cathandle->u.chd.chd_next_log;
-       if (!IS_ERR_OR_NULL(next)) {
-               if (!llog_exist(next)) {
-                       if (dt_object_remote(cathandle->lgh_obj)) {
-                               /* For remote operation, if we put the llog
-                                * object creation in the current transaction,
-                                * then the llog object will not be created on
-                                * the remote target until the transaction stop,
-                                * if other operations start before the
-                                * transaction stop, and use the same llog
-                                * object, will be dependent on the success of
-                                * this transaction. So let's create the llog
-                                * object synchronously here to remove the
-                                * dependency. */
-                               down_write_nested(&cathandle->lgh_lock,
-                                                LLOGH_CAT);
-                               next = cathandle->u.chd.chd_next_log;
-                               if (IS_ERR_OR_NULL(next)) {
-                                       /* Sigh, another thread just tried,
-                                        * let's fail as well */
-                                       up_write(&cathandle->lgh_lock);
-                                       if (next == NULL)
-                                               rc = -EIO;
-                                       else
-                                               rc = PTR_ERR(next);
-                                       GOTO(out, rc);
-                               }
-
-                               down_write_nested(&next->lgh_lock, LLOGH_LOG);
-                               if (!llog_exist(next)) {
-                                       rc = llog_cat_new_log(env, cathandle,
-                                                             next, NULL);
-                                       if (rc < 0)
-                                               cathandle->u.chd.chd_next_log =
-                                                               ERR_PTR(rc);
-                               }
-                               up_write(&next->lgh_lock);
-                               up_write(&cathandle->lgh_lock);
-                               if (rc < 0)
-                                       GOTO(out, rc);
-                       } else {
-                               rc = llog_declare_create(env, next, th);
-                               llog_declare_write_rec(env, cathandle,
-                                               &lirec->lid_hdr, -1, th);
-                       }
-               }
-               /* XXX: we hope for declarations made for existing llog
-                *      this might be not correct with some backends
-                *      where declarations are expected against specific
-                *      object like ZFS with full debugging enabled */
-               /*llog_declare_write_rec(env, next, rec, -1, th);*/
-       }
-out:
+#if 0
+       /*
+        * XXX: we hope for declarations made for existing llog this might be
+        * not correct with some backends where declarations are expected
+        * against specific object like ZFS with full debugging enabled.
+        */
+       rc = llog_declare_write_rec(env, cathandle->u.chd.chd_next_log, rec, -1,
+                                   th);
+#endif
+       if (rc)
+               CWARN("%s: declaration failed, catalog "DFID": rc = %d\n",
+                     loghandle2name(cathandle),
+                     PLOGID(&cathandle->lgh_id), rc);
+
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_cat_declare_add_rec);
@@ -726,6 +727,56 @@ out_trans:
 }
 EXPORT_SYMBOL(llog_cat_add);
 
+int llog_cat_cancel_arr_rec(const struct lu_env *env,
+                           struct llog_handle *cathandle,
+                           struct llog_logid *lgl, int count, int *index)
+{
+       struct llog_handle *loghandle;
+       int  rc;
+
+       ENTRY;
+       rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
+       if (rc) {
+               CDEBUG(D_HA, "%s: can't find llog handle for "DFID": rc = %d\n",
+                      loghandle2name(cathandle), PLOGID(lgl), rc);
+               RETURN(rc);
+       }
+
+       if ((cathandle->lgh_ctxt->loc_flags &
+            LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
+               /* For update log, some of loghandles of cathandle
+                * might not exist because remote llog creation might
+                * be failed, so let's skip the record cancellation
+                * for these non-exist llogs.
+                */
+               rc = -ENOENT;
+               CDEBUG(D_HA, "%s: llog "DFID" does not exist: rc = %d\n",
+                      loghandle2name(cathandle), PLOGID(lgl), rc);
+               llog_handle_put(env, loghandle);
+               RETURN(rc);
+       }
+
+       rc = llog_cancel_arr_rec(env, loghandle, count, index);
+       if (rc == LLOG_DEL_PLAIN) { /* log has been destroyed */
+               int cat_index;
+
+               cat_index = loghandle->u.phd.phd_cookie.lgc_index;
+               rc = llog_cat_cleanup(env, cathandle, loghandle, cat_index);
+               if (rc)
+                       CDEBUG(D_HA,
+                              "%s: fail to cancel catalog record: rc = %d\n",
+                              loghandle2name(cathandle), rc);
+               rc = 0;
+
+       }
+       llog_handle_put(env, loghandle);
+       if (rc && rc != -ENOENT && rc != -ESTALE && rc != -EIO)
+               CWARN("%s: fail to cancel %d records in "DFID": rc = %d\n",
+                     loghandle2name(cathandle), count, PLOGID(lgl), rc);
+       RETURN(rc);
+}
+EXPORT_SYMBOL(llog_cat_cancel_arr_rec);
+
 /* For each cookie in the cookie array, we clear the log in-use bit and either:
  * - the log is empty, so mark it free in the catalog header and delete it
  * - the log is not empty, just write out the log header
@@ -739,64 +790,18 @@ int llog_cat_cancel_records(const struct lu_env *env,
                            struct llog_handle *cathandle, int count,
                            struct llog_cookie *cookies)
 {
-       int i, index, rc = 0, failed = 0;
+       int i, rc = 0;
 
        ENTRY;
 
        for (i = 0; i < count; i++, cookies++) {
-               struct llog_handle *loghandle;
-               struct llog_logid *lgl = &cookies->lgc_lgl;
-               int  lrc;
-
-               rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
-               if (rc) {
-                       CDEBUG(D_HA, "%s: cannot find llog for handle "DFID":%x"
-                              ": rc = %d\n",
-                              cathandle->lgh_ctxt->loc_obd->obd_name,
-                              PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc);
-                       failed++;
-                       continue;
-               }
+               int lrc;
 
-               if ((cathandle->lgh_ctxt->loc_flags &
-                    LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
-                       /* For update log, some of loghandles of cathandle
-                        * might not exist because remote llog creation might
-                        * be failed, so let's skip the record cancellation
-                        * for these non-exist llogs.
-                        */
-                       lrc = -ENOENT;
-                       CDEBUG(D_HA, "%s: llog "DFID":%x does not exist"
-                              ": rc = %d\n",
-                              cathandle->lgh_ctxt->loc_obd->obd_name,
-                              PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, lrc);
-                       failed++;
-                       if (rc == 0)
-                               rc = lrc;
-                       continue;
-               }
-
-               lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index);
-               if (lrc == LLOG_DEL_PLAIN) { /* log has been destroyed */
-                       index = loghandle->u.phd.phd_cookie.lgc_index;
-                       lrc = llog_cat_cleanup(env, cathandle, loghandle,
-                                              index);
-                       if (rc == 0)
-                               rc = lrc;
-               } else if (lrc == -ENOENT) {
-                       if (rc == 0) /* ENOENT shouldn't rewrite any error */
-                               rc = lrc;
-               } else if (lrc < 0) {
-                       failed++;
-                       if (rc == 0)
-                               rc = lrc;
-               }
-               llog_handle_put(loghandle);
+               lrc = llog_cat_cancel_arr_rec(env, cathandle, &cookies->lgc_lgl,
+                                             1, &cookies->lgc_index);
+               if (lrc && !rc)
+                       rc = lrc;
        }
-       if (rc)
-               CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
-                      cathandle->lgh_ctxt->loc_obd->obd_name, failed, count,
-                      rc);
 
        RETURN(rc);
 }
@@ -814,16 +819,13 @@ static int llog_cat_process_common(const struct lu_env *env,
        ENTRY;
        if (rec->lrh_type != le32_to_cpu(LLOG_LOGID_MAGIC)) {
                rc = -EINVAL;
-               CWARN("%s: invalid record in catalog "DFID":%x: rc = %d\n",
-                     cat_llh->lgh_ctxt->loc_obd->obd_name,
-                     PFID(&cat_llh->lgh_id.lgl_oi.oi_fid),
-                     cat_llh->lgh_id.lgl_ogen, rc);
+               CWARN("%s: invalid record in catalog "DFID": rc = %d\n",
+                     loghandle2name(cat_llh), PLOGID(&cat_llh->lgh_id), rc);
                RETURN(rc);
        }
-       CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "DFID"\n",
-              PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen,
-              le32_to_cpu(rec->lrh_index),
-              PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
+       CDEBUG(D_HA, "processing log "DFID" at index %u of catalog "DFID"\n",
+              PLOGID(&lir->lid_id), le32_to_cpu(rec->lrh_index),
+              PLOGID(&cat_llh->lgh_id));
 
        rc = llog_cat_id2handle(env, cat_llh, llhp, &lir->lid_id);
        if (rc) {
@@ -834,10 +836,9 @@ static int llog_cat_process_common(const struct lu_env *env,
                if (rc == -ENOENT || rc == -ESTALE)
                        rc = LLOG_DEL_RECORD;
                else if (rc)
-                       CWARN("%s: can't find llog handle "DFID":%x: rc = %d\n",
-                             cat_llh->lgh_ctxt->loc_obd->obd_name,
-                             PFID(&lir->lid_id.lgl_oi.oi_fid),
-                             lir->lid_id.lgl_ogen, rc);
+                       CWARN("%s: can't find llog handle "DFID": rc = %d\n",
+                             loghandle2name(cat_llh), PLOGID(&lir->lid_id),
+                             rc);
 
                RETURN(rc);
        }
@@ -847,12 +848,13 @@ static int llog_cat_process_common(const struct lu_env *env,
        hdr = (*llhp)->lgh_hdr;
        if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
            hdr->llh_count == 1 && cat_llh->lgh_obj != NULL &&
-           *llhp != cat_llh->u.chd.chd_current_log) {
+           *llhp != cat_llh->u.chd.chd_current_log &&
+           *llhp != cat_llh->u.chd.chd_next_log) {
                rc = llog_destroy(env, *llhp);
                if (rc)
                        CWARN("%s: can't destroy empty log "DFID": rc = %d\n",
-                             (*llhp)->lgh_ctxt->loc_obd->obd_name,
-                             PFID(&lir->lid_id.lgl_oi.oi_fid), rc);
+                             loghandle2name((*llhp)), PLOGID(&lir->lid_id),
+                             rc);
                rc = LLOG_DEL_PLAIN;
        }
 
@@ -868,18 +870,26 @@ static int llog_cat_process_cb(const struct lu_env *env,
        int rc;
 
        ENTRY;
+
+       /* Skip processing of the logs until startcat */
+       if (rec->lrh_index < d->lpd_startcat)
+               RETURN(0);
+
        rc = llog_cat_process_common(env, cat_llh, rec, &llh);
        if (rc)
                GOTO(out, rc);
 
-       if (rec->lrh_index < d->lpd_startcat) {
-               /* Skip processing of the logs until startcat */
-               rc = 0;
-       } else if (d->lpd_startidx > 0) {
-                struct llog_process_cat_data cd;
+       if (d->lpd_startidx > 0) {
+               struct llog_process_cat_data cd = {
+                       .lpcd_first_idx = 0,
+                       .lpcd_last_idx = 0,
+                       .lpcd_read_mode = LLOG_READ_MODE_NORMAL,
+               };
+
+               /* startidx is always associated with a catalog index */
+               if (d->lpd_startcat == rec->lrh_index)
+                       cd.lpcd_first_idx = d->lpd_startidx;
 
-                cd.lpcd_first_idx = d->lpd_startidx;
-                cd.lpcd_last_idx = 0;
                rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
                                          &cd, false);
                /* Continue processing the next log from idx 0 */
@@ -888,19 +898,27 @@ static int llog_cat_process_cb(const struct lu_env *env,
                rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
                                          NULL, false);
        }
+       if (rc == -ENOENT && (cat_llh->lgh_hdr->llh_flags & LLOG_F_RM_ON_ERR)) {
+               /*
+                * plain llog is reported corrupted, so better to just remove
+                * it if the caller is fine with that.
+                */
+               CERROR("%s: remove corrupted/missing llog "DFID"\n",
+                      loghandle2name(cat_llh), PLOGID(&llh->lgh_id));
+               rc = LLOG_DEL_PLAIN;
+       }
 
 out:
        /* The empty plain log was destroyed while processing */
-       if (rc == LLOG_DEL_PLAIN) {
-               rc = llog_cat_cleanup(env, cat_llh, llh,
-                                     llh->u.phd.phd_cookie.lgc_index);
-       } else if (rc == LLOG_DEL_RECORD) {
+       if (rc == LLOG_DEL_PLAIN || rc == LLOG_DEL_RECORD)
                /* clear wrong catalog entry */
-               rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
-       }
+               rc = llog_cat_cleanup(env, cat_llh, llh, rec->lrh_index);
+       else if (rc == LLOG_SKIP_PLAIN)
+               /* processing callback ask to skip the llog -> continue */
+               rc = 0;
 
        if (llh)
-               llog_handle_put(llh);
+               llog_handle_put(env, llh);
 
        RETURN(rc);
 }
@@ -910,44 +928,96 @@ int llog_cat_process_or_fork(const struct lu_env *env,
                             llog_cb_t cb, void *data, int startcat,
                             int startidx, bool fork)
 {
-        struct llog_process_data d;
-        struct llog_log_hdr *llh = cat_llh->lgh_hdr;
-        int rc;
-        ENTRY;
+       struct llog_log_hdr *llh = cat_llh->lgh_hdr;
+       struct llog_process_data d;
+       struct llog_process_cat_data cd;
+       int rc;
 
-        LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
-        d.lpd_data = data;
-        d.lpd_cb = cb;
-        d.lpd_startcat = startcat;
-        d.lpd_startidx = startidx;
+       ENTRY;
 
-       if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
-           llh->llh_count > 1) {
-               struct llog_process_cat_data cd;
+       LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
+       d.lpd_data = data;
+       d.lpd_cb = cb;
+
+       /* default: start from the oldest record */
+       d.lpd_startidx = 0;
+       d.lpd_startcat = llh->llh_cat_idx + 1;
+       cd.lpcd_first_idx = llh->llh_cat_idx;
+       cd.lpcd_last_idx = 0;
+       cd.lpcd_read_mode = LLOG_READ_MODE_NORMAL;
+
+       if (startcat > 0 && startcat <= llog_max_idx(llh)) {
+               /* start from a custom catalog/llog plain indexes*/
+               d.lpd_startidx = startidx;
+               d.lpd_startcat = startcat;
+               cd.lpcd_first_idx = startcat - 1;
+       } else if (startcat != 0) {
+               CWARN("%s: startcat %d out of range for catlog "DFID"\n",
+                     loghandle2name(cat_llh), startcat,
+                     PLOGID(&cat_llh->lgh_id));
+               RETURN(-EINVAL);
+       }
+
+       startcat = d.lpd_startcat;
+
+       /* if startcat <= lgh_last_idx, we only need to process the first part
+        * of the catalog (from startcat).
+        */
+       if (llog_cat_is_wrapped(cat_llh) && startcat > cat_llh->lgh_last_idx) {
+               int cat_idx_origin = llh->llh_cat_idx;
 
                CWARN("%s: catlog "DFID" crosses index zero\n",
-                     cat_llh->lgh_ctxt->loc_obd->obd_name,
-                     PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
+                     loghandle2name(cat_llh),
+                     PLOGID(&cat_llh->lgh_id));
 
-               cd.lpcd_first_idx = llh->llh_cat_idx;
-               cd.lpcd_last_idx = 0;
-               rc = llog_process_or_fork(env, cat_llh, cat_cb,
-                                         &d, &cd, fork);
-               if (rc != 0)
+               /* processing the catalog part at the end */
+               rc = llog_process_or_fork(env, cat_llh, cat_cb, &d, &cd, fork);
+               if (rc)
                        RETURN(rc);
 
+               /* Reset the startcat because it has already reached catalog
+                * bottom.
+                * lgh_last_idx value could be increased during processing. So
+                * we process the remaining of catalog entries to be sure.
+                */
+               d.lpd_startcat = 1;
+               d.lpd_startidx = 0;
                cd.lpcd_first_idx = 0;
-               cd.lpcd_last_idx = cat_llh->lgh_last_idx;
-               rc = llog_process_or_fork(env, cat_llh, cat_cb,
-                                         &d, &cd, fork);
-        } else {
-               rc = llog_process_or_fork(env, cat_llh, cat_cb,
-                                         &d, NULL, fork);
-        }
+               cd.lpcd_last_idx = max(cat_idx_origin, cat_llh->lgh_last_idx);
+       } else if (llog_cat_is_wrapped(cat_llh)) {
+               /* only process 1st part -> stop before reaching 2sd part */
+               cd.lpcd_last_idx = llh->llh_cat_idx;
+       }
 
-        RETURN(rc);
+       /* processing the catalog part at the begining */
+       rc = llog_process_or_fork(env, cat_llh, cat_cb, &d, &cd, fork);
+
+       RETURN(rc);
 }
+EXPORT_SYMBOL(llog_cat_process_or_fork);
 
+/**
+ * Process catalog records with a callback
+ *
+ * \note
+ * If "starcat = 0", this is the default processing. "startidx" argument is
+ * ignored and processing begin from the oldest record.
+ * If "startcat > 0", this is a custom starting point. Processing begin with
+ * the llog plain defined in the catalog record at index "startcat". The first
+ * llog plain record to process is at index "startidx + 1".
+ *
+ * \param env          Lustre environnement
+ * \param cat_llh      Catalog llog handler
+ * \param cb           Callback executed for each records (in llog plain files)
+ * \param data         Callback data argument
+ * \param startcat     Catalog index of the llog plain to start with.
+ * \param startidx     Index of the llog plain to start processing. The first
+ *                     record to process is at startidx + 1.
+ *
+ * \retval 0 processing successfully completed
+ * \retval LLOG_PROC_BREAK processing was stopped by the callback.
+ * \retval -errno on error.
+ */
 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
                     llog_cb_t cb, void *data, int startcat, int startidx)
 {
@@ -981,11 +1051,11 @@ static int llog_cat_size_cb(const struct lu_env *env,
                *cum_size += size;
 
                CDEBUG(D_INFO, "Add llog entry "DFID" size=%llu, tot=%llu\n",
-                      PFID(&llh->lgh_id.lgl_oi.oi_fid), size, *cum_size);
+                      PLOGID(&llh->lgh_id), size, *cum_size);
        }
 
        if (llh != NULL)
-               llog_handle_put(llh);
+               llog_handle_put(env, llh);
 
        RETURN(0);
 }
@@ -1008,14 +1078,14 @@ EXPORT_SYMBOL(llog_cat_size);
 __u32 llog_cat_free_space(struct llog_handle *cat_llh)
 {
        /* simulate almost full Catalog */
-       if (OBD_FAIL_CHECK(OBD_FAIL_CAT_FREE_RECORDS))
+       if (CFS_FAIL_CHECK(OBD_FAIL_CAT_FREE_RECORDS))
                return cfs_fail_val;
 
        if (cat_llh->lgh_hdr->llh_count == 1)
-               return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1;
+               return llog_max_idx(cat_llh->lgh_hdr);
 
        if (cat_llh->lgh_last_idx > cat_llh->lgh_hdr->llh_cat_idx)
-               return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1 +
+               return llog_max_idx(cat_llh->lgh_hdr) +
                       cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
 
        /* catalog is presently wrapped */
@@ -1041,6 +1111,9 @@ static int llog_cat_reverse_process_cb(const struct lu_env *env,
        } else if (rc == LLOG_DEL_RECORD) {
                /* clear wrong catalog entry */
                rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
+       } else if (rc == LLOG_SKIP_PLAIN) {
+               /* processing callback ask to skip the llog -> continue */
+               rc = 0;
        }
        if (rc)
                RETURN(rc);
@@ -1052,7 +1125,7 @@ static int llog_cat_reverse_process_cb(const struct lu_env *env,
                rc = llog_cat_cleanup(env, cat_llh, llh,
                                      llh->u.phd.phd_cookie.lgc_index);
 
-       llog_handle_put(llh);
+       llog_handle_put(env, llh);
        RETURN(rc);
 }
 
@@ -1067,14 +1140,15 @@ int llog_cat_reverse_process(const struct lu_env *env,
         ENTRY;
 
         LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
+       cd.lpcd_read_mode = LLOG_READ_MODE_NORMAL;
         d.lpd_data = data;
         d.lpd_cb = cb;
 
        if (llh->llh_cat_idx >= cat_llh->lgh_last_idx &&
            llh->llh_count > 1) {
                CWARN("%s: catalog "DFID" crosses index zero\n",
-                     cat_llh->lgh_ctxt->loc_obd->obd_name,
-                     PFID(&cat_llh->lgh_id.lgl_oi.oi_fid));
+                     loghandle2name(cat_llh),
+                     PLOGID(&cat_llh->lgh_id));
 
                cd.lpcd_first_idx = 0;
                cd.lpcd_last_idx = cat_llh->lgh_last_idx;
@@ -1102,11 +1176,11 @@ EXPORT_SYMBOL(llog_cat_reverse_process);
 static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
 {
        struct llog_log_hdr *llh = cathandle->lgh_hdr;
-       int bitmap_size;
+       int idx_nbr;
 
        ENTRY;
 
-       bitmap_size = LLOG_HDR_BITMAP_SIZE(llh);
+       idx_nbr = llog_max_idx(llh) + 1;
        /*
         * The llh_cat_idx equals to the first used index minus 1
         * so if we canceled the first index then llh_cat_idx
@@ -1116,8 +1190,8 @@ static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
                llh->llh_cat_idx = idx;
 
                while (idx != cathandle->lgh_last_idx) {
-                       idx = (idx + 1) % bitmap_size;
-                       if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
+                       idx = (idx + 1) % idx_nbr;
+                       if (!test_bit_le(idx, LLOG_HDR_BITMAP(llh))) {
                                /* update llh_cat_idx for each unset bit,
                                 * expecting the next one is set */
                                llh->llh_cat_idx = idx;
@@ -1131,9 +1205,9 @@ static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx)
                        }
                }
 
-               CDEBUG(D_RPCTRACE, "catlog "DFID" first idx %u, last_idx %u\n",
-                      PFID(&cathandle->lgh_id.lgl_oi.oi_fid),
-                      llh->llh_cat_idx, cathandle->lgh_last_idx);
+               CDEBUG(D_HA, "catlog "DFID" first idx %u, last_idx %u\n",
+                      PLOGID(&cathandle->lgh_id), llh->llh_cat_idx,
+                      cathandle->lgh_last_idx);
        }
 
        RETURN(0);
@@ -1154,7 +1228,8 @@ int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
                        cathandle->u.chd.chd_current_log = NULL;
                list_del_init(&loghandle->u.phd.phd_entry);
                up_write(&cathandle->lgh_lock);
-               LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index);
+               LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index ||
+                       loghandle->u.phd.phd_cookie.lgc_index == 0);
                /* llog was opened and keep in a list, close it now */
                llog_close(env, loghandle);
        }
@@ -1166,8 +1241,67 @@ int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
        /* remove plain llog entry from catalog by index */
        llog_cat_set_first_idx(cathandle, index);
        rc = llog_cancel_rec(env, cathandle, index);
-       if (rc == 0)
-               CDEBUG(D_HA, "cancel plain log at index %u of catalog "DFID"\n",
-                      index, PFID(&cathandle->lgh_id.lgl_oi.oi_fid));
+       if (!rc && loghandle)
+               CDEBUG(D_HA,
+                      "cancel plain log "DFID" at index %u of catalog "DFID"\n",
+                      PLOGID(&loghandle->lgh_id), index,
+                      PLOGID(&cathandle->lgh_id));
+       return rc;
+}
+
+/* retain log in catalog, and zap it if log is empty */
+int llog_cat_retain_cb(const struct lu_env *env, struct llog_handle *cat,
+                      struct llog_rec_hdr *rec, void *data)
+{
+       struct llog_handle *log = NULL;
+       int rc;
+
+       rc = llog_cat_process_common(env, cat, rec, &log);
+
+       /* The empty plain log was destroyed while processing */
+       if (rc == LLOG_DEL_PLAIN || rc == LLOG_DEL_RECORD)
+               /* clear wrong catalog entry */
+               rc = llog_cat_cleanup(env, cat, log, rec->lrh_index);
+       else if (!rc)
+               llog_retain(env, log);
+
+       if (log)
+               llog_handle_put(env, log);
+
        return rc;
 }
+EXPORT_SYMBOL(llog_cat_retain_cb);
+
+/* Modify a llog record base on llog_logid and record cookie,
+ * with valid offset.
+ */
+int llog_cat_modify_rec(const struct lu_env *env, struct llog_handle *cathandle,
+                       struct llog_logid *lid, struct llog_rec_hdr *rec,
+                       struct llog_cookie *cookie)
+{
+       struct llog_handle *llh;
+       int rc;
+
+       ENTRY;
+
+       rc = llog_cat_id2handle(env, cathandle, &llh, lid);
+       if (rc) {
+               CDEBUG(D_OTHER, "%s: failed to find log file "DFID": rc = %d\n",
+                      loghandle2name(llh), PLOGID(lid), rc);
+
+               RETURN(rc);
+       }
+
+       rc = llog_write_cookie(env, llh, rec, cookie, rec->lrh_index);
+       if (rc < 0) {
+               CDEBUG(D_OTHER,
+                      "%s: failed to modify record "DFID".%d: rc = %d\n",
+                      loghandle2name(llh), PLOGID(lid), rec->lrh_index, rc);
+       } else {
+               rc = 0;
+       }
+       llog_handle_put(env, llh);
+
+       RETURN(rc);
+}
+EXPORT_SYMBOL(llog_cat_modify_rec);