Whamcloud - gitweb
LU-17705 ptlrpc: replace synchronize_rcu() with rcu_barrier()
[fs/lustre-release.git] / lustre / obdclass / llog_cat.c
index ec28fa8..3d9cceb 100644 (file)
@@ -27,7 +27,6 @@
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  *
  * lustre/obdclass/llog_cat.c
  *
 
 #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 +76,7 @@ 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 +92,17 @@ 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) {
@@ -153,7 +155,7 @@ 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);
        }
 
@@ -176,8 +178,8 @@ static int llog_cat_new_log(const struct lu_env *env,
                GOTO(out_destroy, rc);
 
        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 +199,13 @@ 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 +222,139 @@ 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;
+
+       down_write(&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 unlock;
+       }
+
+       rc = llog_read_header(env, cathandle, NULL);
+unlock:
+       up_write(&cathandle->lgh_lock);
+
+       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)
+{
+       int rc;
+       int sem_upgraded;
+
+start:
+       rc = 0;
+       sem_upgraded = 0;
+       if (IS_ERR_OR_NULL(*ploghandle)) {
+               up_read(&cathandle->lgh_lock);
+               down_write(&cathandle->lgh_lock);
+               sem_upgraded = 1;
+               if (IS_ERR_OR_NULL(*ploghandle)) {
+                       struct llog_handle *loghandle;
+
+                       rc = llog_open(env, cathandle->lgh_ctxt, &loghandle,
+                                      NULL, NULL, LLOG_OPEN_NEW);
+                       if (!rc) {
+                               *ploghandle = loghandle;
+                               list_add_tail(&loghandle->u.phd.phd_entry,
+                                             &cathandle->u.chd.chd_head);
+                       }
+               }
+               if (rc)
+                       GOTO(out, rc);
+       }
+
+       rc = llog_exist(*ploghandle);
+       if (rc < 0)
+               GOTO(out, rc);
+       if (rc)
+               GOTO(out, rc = 0);
+
+       if (dt_object_remote(cathandle->lgh_obj)) {
+               down_write_nested(&(*ploghandle)->lgh_lock, LLOGH_LOG);
+               if (!llog_exist(*ploghandle)) {
+                       /* 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. */
+                       rc = llog_cat_new_log(env, cathandle, *ploghandle,
+                                             NULL);
+                       if (rc == -ESTALE) {
+                               up_write(&(*ploghandle)->lgh_lock);
+                               if (sem_upgraded)
+                                       up_write(&cathandle->lgh_lock);
+                               else
+                                       up_read(&cathandle->lgh_lock);
+
+                               rc = llog_cat_refresh(env, cathandle);
+                               down_read_nested(&cathandle->lgh_lock,
+                                                LLOGH_CAT);
+                               if (rc)
+                                       return rc;
+                               /* *ploghandle might become NULL, restart */
+                               goto start;
+                       }
+               }
+               up_write(&(*ploghandle)->lgh_lock);
+       } else {
+               struct llog_thread_info *lgi = llog_info(env);
+               struct llog_logid_rec *lirec = &lgi->lgi_logid;
+
+               rc = llog_declare_create(env, *ploghandle, th);
+               if (rc)
+                       GOTO(out, rc);
+
+               lirec->lid_hdr.lrh_len = sizeof(*lirec);
+               rc = llog_declare_write_rec(env, cathandle, &lirec->lid_hdr, -1,
+                                           th);
+       }
+
+out:
+       if (sem_upgraded) {
+               up_write(&cathandle->lgh_lock);
+               down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
+               if (rc == 0)
+                       goto start;
+       }
+       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.
  *
@@ -247,16 +384,16 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
 
                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);
+                       RETURN(rc);
                }
        }
        up_write(&cathandle->lgh_lock);
@@ -264,19 +401,21 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
        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);
        up_write(&cathandle->lgh_lock);
@@ -285,11 +424,7 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_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 +449,7 @@ 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 +464,6 @@ 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.
  *
@@ -355,7 +481,7 @@ static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
         ENTRY;
 
 
-       if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
+       if (CFS_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) {
                down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
                GOTO(next, loghandle);
        }
@@ -401,7 +527,7 @@ next:
         * 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);
+                      loghandle2name(cathandle));
                loghandle = ERR_PTR(-EIO);
                if (cathandle->u.chd.chd_next_log == NULL) {
                        /* Store the error in chd_next_log, so
@@ -425,40 +551,6 @@ 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
  *
@@ -486,12 +578,7 @@ retry:
                        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
-                        */
-                       if ((cathandle->u.chd.chd_current_log == loghandle) &&
-                           rc != -ENOSPC)
+                       if (cathandle->u.chd.chd_current_log == loghandle)
                                cathandle->u.chd.chd_current_log = NULL;
                        up_write(&cathandle->lgh_lock);
                        RETURN(rc);
@@ -517,7 +604,7 @@ retry:
                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 +615,43 @@ 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;
+       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:
+       down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
+       rc = llog_cat_prep_log(env, cathandle,
+                              &cathandle->u.chd.chd_current_log, th);
        if (rc)
-               GOTO(out, rc);
+               GOTO(unlock, rc);
 
-       lirec->lid_hdr.lrh_len = sizeof(*lirec);
-
-       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);
-               }
-       }
+       rc = llog_cat_prep_log(env, cathandle, &cathandle->u.chd.chd_next_log,
+                              th);
+       if (rc)
+               GOTO(unlock, rc);
 
-write_again:
-       /* declare records in the llogs */
        rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
                                    rec, -1, th);
-       if (rc == -ESTALE) {
-               down_write(&cathandle->lgh_lock);
-               if (cathandle->lgh_stale) {
-                       up_write(&cathandle->lgh_lock);
-                       GOTO(out, rc = -EIO);
-               }
-
-               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);
+       if (rc == -ESTALE && dt_object_remote(cathandle->lgh_obj)) {
+               up_read(&cathandle->lgh_lock);
+               rc = llog_cat_refresh(env, cathandle);
+               if (rc)
+                       RETURN(rc);
+               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
+unlock:
+       up_read(&cathandle->lgh_lock);
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_cat_declare_add_rec);
@@ -726,6 +689,57 @@ 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)
+                       CERROR("%s: fail to cancel catalog record: rc = %d\n",
+                              loghandle2name(cathandle), rc);
+               rc = 0;
+
+       }
+       llog_handle_put(env, loghandle);
+
+       if (rc)
+               CERROR("%s: fail to cancel %d llog-records: rc = %d\n",
+                      loghandle2name(cathandle), count, 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
@@ -739,65 +753,24 @@ 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, failed = 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) {
+               lrc = llog_cat_cancel_arr_rec(env, cathandle, &cookies->lgc_lgl,
+                                            1, &cookies->lgc_index);
+               if (lrc) {
                        failed++;
-                       if (rc == 0)
+                       if (!rc)
                                rc = lrc;
                }
-               llog_handle_put(loghandle);
        }
-       if (rc)
+       if (failed)
                CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
-                      cathandle->lgh_ctxt->loc_obd->obd_name, failed, count,
-                      rc);
-
+                      loghandle2name(cathandle), failed, count, rc);
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_cat_cancel_records);
@@ -814,16 +787,13 @@ 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 +804,9 @@ 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 +816,13 @@ 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 +838,26 @@ 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 +866,27 @@ 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 +896,96 @@ 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 +1019,11 @@ 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 +1046,14 @@ 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 +1079,9 @@ 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 +1093,7 @@ 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 +1108,15 @@ 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 +1144,11 @@ 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 +1158,8 @@ 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 +1173,9 @@ 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 +1196,8 @@ 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 +1209,35 @@ 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);
+
+