Whamcloud - gitweb
LU-5859 llog: do not cleanup orphans in remote catalogs
[fs/lustre-release.git] / lustre / obdclass / llog_cat.c
index 1bd130e..e7114cf 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -48,9 +48,6 @@
 
 #define DEBUG_SUBSYSTEM S_LOG
 
-#ifndef __KERNEL__
-#include <liblustre.h>
-#endif
 
 #include <obd_class.h>
 
@@ -66,22 +63,11 @@ static int llog_cat_new_log(const struct lu_env *env,
                            struct llog_handle *loghandle,
                            struct thandle *th)
 {
+       struct llog_thread_info *lgi = llog_info(env);
+       struct llog_logid_rec   *rec = &lgi->lgi_logid;
+       int                      rc;
 
-        struct llog_log_hdr *llh;
-        struct llog_logid_rec rec = { { 0 }, };
-        int rc, index, bitmap_size;
-        ENTRY;
-
-        llh = cathandle->lgh_hdr;
-        bitmap_size = LLOG_BITMAP_SIZE(llh);
-
-        index = (cathandle->lgh_last_idx + 1) % bitmap_size;
-
-        /* maximum number of available slots in catlog is bitmap_size - 2 */
-        if (llh->llh_cat_idx == index) {
-                CERROR("no free catalog slots for log...\n");
-               RETURN(-ENOSPC);
-       }
+       ENTRY;
 
        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
                RETURN(-ENOSPC);
@@ -97,45 +83,29 @@ static int llog_cat_new_log(const struct lu_env *env,
        }
 
        rc = llog_init_handle(env, loghandle,
-                              LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
-                              &cathandle->lgh_hdr->llh_tgtuuid);
-        if (rc)
-                GOTO(out_destroy, rc);
-
-        if (index == 0)
-                index = 1;
-
-       spin_lock(&loghandle->lgh_hdr_lock);
-       llh->llh_count++;
-       if (ext2_set_bit(index, llh->llh_bitmap)) {
-               CERROR("argh, index %u already set in log bitmap?\n",
-                      index);
-               spin_unlock(&loghandle->lgh_hdr_lock);
-               LBUG(); /* should never happen */
-       }
-       spin_unlock(&loghandle->lgh_hdr_lock);
-
-        cathandle->lgh_last_idx = index;
-        llh->llh_tail.lrt_index = index;
-
-        CDEBUG(D_RPCTRACE,"new recovery log "LPX64":%x for index %u of catalog "
-               LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen,
-               index, cathandle->lgh_id.lgl_oid);
-        /* build the record for this log in the catalog */
-        rec.lid_hdr.lrh_len = sizeof(rec);
-        rec.lid_hdr.lrh_index = index;
-        rec.lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
-        rec.lid_id = loghandle->lgh_id;
-        rec.lid_tail.lrt_len = sizeof(rec);
-        rec.lid_tail.lrt_index = index;
-
-        /* update the catalog: header and record */
-       rc = llog_write_rec(env, cathandle, &rec.lid_hdr,
-                           &loghandle->u.phd.phd_cookie, 1, NULL, index, th);
+                             LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
+                             &cathandle->lgh_hdr->llh_tgtuuid);
+       if (rc)
+               GOTO(out_destroy, rc);
+
+       /* build the record for this log in the catalog */
+       rec->lid_hdr.lrh_len = sizeof(*rec);
+       rec->lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
+       rec->lid_id = loghandle->lgh_id;
+
+       /* append the new record into catalog. The new index will be
+        * assigned to the record and updated in rec header */
+       rc = llog_write_rec(env, cathandle, &rec->lid_hdr,
+                           &loghandle->u.phd.phd_cookie, LLOG_NEXT_IDX, th);
        if (rc < 0)
                GOTO(out_destroy, rc);
 
-       loghandle->lgh_hdr->llh_cat_idx = index;
+       CDEBUG(D_OTHER, "new recovery log "DOSTID":%x for index %u of catalog"
+              DOSTID"\n", POSTID(&loghandle->lgh_id.lgl_oi),
+              loghandle->lgh_id.lgl_ogen, rec->lid_hdr.lrh_index,
+              POSTID(&cathandle->lgh_id.lgl_oi));
+
+       loghandle->lgh_hdr->llh_cat_idx = rec->lid_hdr.lrh_index;
        RETURN(0);
 out_destroy:
        llog_destroy(env, loghandle);
@@ -147,11 +117,15 @@ out_destroy:
  *
  * Assumes caller has already pushed us into the kernel context and is locking.
  * We return a lock on the handle to ensure nobody yanks it from us.
+ *
+ * This takes extra reference on llog_handle via llog_handle_get() and require
+ * this reference to be put by caller using llog_handle_put()
  */
 int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
                       struct llog_handle **res, struct llog_logid *logid)
 {
        struct llog_handle      *loghandle;
+       enum llog_flag           fmt;
        int                      rc = 0;
 
        ENTRY;
@@ -159,16 +133,18 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
        if (cathandle == NULL)
                RETURN(-EBADF);
 
+       fmt = cathandle->lgh_hdr->llh_flags & LLOG_F_EXT_MASK;
        down_write(&cathandle->lgh_lock);
-       cfs_list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
-                               u.phd.phd_entry) {
+       list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
+                           u.phd.phd_entry) {
                struct llog_logid *cgl = &loghandle->lgh_id;
 
-               if (cgl->lgl_oid == logid->lgl_oid) {
+               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) {
-                               CERROR("%s: log "LPX64" generation %x != %x\n",
+                               CERROR("%s: log "DOSTID" generation %x != %x\n",
                                       loghandle->lgh_ctxt->loc_obd->obd_name,
-                                      logid->lgl_oid, cgl->lgl_ogen,
+                                      POSTID(&logid->lgl_oi), cgl->lgl_ogen,
                                       logid->lgl_ogen);
                                continue;
                        }
@@ -182,20 +158,21 @@ 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 "LPX64":%x: rc = %d\n",
+               CERROR("%s: error opening log id "DOSTID":%x: rc = %d\n",
                       cathandle->lgh_ctxt->loc_obd->obd_name,
-                      logid->lgl_oid, logid->lgl_ogen, rc);
-               GOTO(out, rc);
+                      POSTID(&logid->lgl_oi), logid->lgl_ogen, rc);
+               RETURN(rc);
        }
 
-       rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN, NULL);
+       rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN | fmt, NULL);
        if (rc < 0) {
                llog_close(env, loghandle);
-               GOTO(out, rc);
+               loghandle = NULL;
+               RETURN(rc);
        }
 
        down_write(&cathandle->lgh_lock);
-       cfs_list_add(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
+       list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
        up_write(&cathandle->lgh_lock);
 
        loghandle->u.phd.phd_cat_handle = cathandle;
@@ -204,8 +181,9 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle,
                                loghandle->lgh_hdr->llh_cat_idx;
        EXIT;
 out:
+       llog_handle_get(loghandle);
        *res = loghandle;
-       return rc;
+       return 0;
 }
 
 int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
@@ -215,13 +193,13 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
 
        ENTRY;
 
-       cfs_list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
-                                    u.phd.phd_entry) {
+       list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
+                                u.phd.phd_entry) {
                struct llog_log_hdr     *llh = loghandle->lgh_hdr;
                int                      index;
 
                /* unlink open-not-created llogs */
-               cfs_list_del_init(&loghandle->u.phd.phd_entry);
+               list_del_init(&loghandle->u.phd.phd_entry);
                llh = loghandle->lgh_hdr;
                if (loghandle->lgh_obj != NULL && llh != NULL &&
                    (llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
@@ -234,15 +212,7 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle)
                                       rc);
 
                        index = loghandle->u.phd.phd_cookie.lgc_index;
-
-                       LASSERT(index);
-                       llog_cat_set_first_idx(cathandle, index);
-                       rc = llog_cancel_rec(env, cathandle, index);
-                       if (rc == 0)
-                               CDEBUG(D_RPCTRACE,
-                                      "cancel plain log at index %u of "
-                                      "catalog "LPX64"\n",
-                                      index, cathandle->lgh_id.lgl_oid);
+                       llog_cat_cleanup(env, cathandle, NULL, index);
                }
                llog_close(env, loghandle);
        }
@@ -332,7 +302,7 @@ static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
  */
 int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
                     struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
-                    void *buf, struct thandle *th)
+                    struct thandle *th)
 {
         struct llog_handle *loghandle;
         int rc;
@@ -351,9 +321,10 @@ int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
                }
        }
        /* now let's try to add the record */
-       rc = llog_write_rec(env, loghandle, rec, reccookie, 1, buf, -1, th);
-        if (rc < 0)
-                CERROR("llog_write_rec %d: lh=%p\n", rc, loghandle);
+       rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th);
+       if (rc < 0)
+               CDEBUG_LIMIT(rc == -ENOSPC ? D_HA : D_ERROR,
+                            "llog_write_rec %d: lh=%p\n", rc, loghandle);
        up_write(&loghandle->lgh_lock);
         if (rc == -ENOSPC) {
                /* try to use next log */
@@ -368,8 +339,8 @@ int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
                        }
                }
                /* now let's try to add the record */
-               rc = llog_write_rec(env, loghandle, rec, reccookie, 1, buf,
-                                   -1, th);
+               rc = llog_write_rec(env, loghandle, rec, reccookie,
+                                   LLOG_NEXT_IDX, th);
                if (rc < 0)
                        CERROR("llog_write_rec %d: lh=%p\n", rc, loghandle);
                up_write(&loghandle->lgh_lock);
@@ -383,6 +354,8 @@ 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;
 
@@ -396,8 +369,8 @@ int llog_cat_declare_add_rec(const struct lu_env *env,
                                       NULL, NULL, LLOG_OPEN_NEW);
                        if (rc == 0) {
                                cathandle->u.chd.chd_current_log = loghandle;
-                               cfs_list_add_tail(&loghandle->u.phd.phd_entry,
-                                                 &cathandle->u.chd.chd_head);
+                               list_add_tail(&loghandle->u.phd.phd_entry,
+                                             &cathandle->u.chd.chd_head);
                        }
                }
                up_write(&cathandle->lgh_lock);
@@ -409,8 +382,8 @@ int llog_cat_declare_add_rec(const struct lu_env *env,
                                       NULL, NULL, LLOG_OPEN_NEW);
                        if (rc == 0) {
                                cathandle->u.chd.chd_next_log = loghandle;
-                               cfs_list_add_tail(&loghandle->u.phd.phd_entry,
-                                                 &cathandle->u.chd.chd_head);
+                               list_add_tail(&loghandle->u.phd.phd_entry,
+                                             &cathandle->u.chd.chd_head);
                        }
                }
                up_write(&cathandle->lgh_lock);
@@ -418,12 +391,14 @@ int llog_cat_declare_add_rec(const struct lu_env *env,
        if (rc)
                GOTO(out, rc);
 
+       lirec->lid_hdr.lrh_len = sizeof(*lirec);
+
        if (!llog_exist(cathandle->u.chd.chd_current_log)) {
                rc = llog_declare_create(env, cathandle->u.chd.chd_current_log,
                                         th);
                if (rc)
                        GOTO(out, rc);
-               llog_declare_write_rec(env, cathandle, NULL, -1, th);
+               llog_declare_write_rec(env, cathandle, &lirec->lid_hdr, -1, th);
        }
        /* declare records in the llogs */
        rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log,
@@ -435,9 +410,14 @@ int llog_cat_declare_add_rec(const struct lu_env *env,
        if (next) {
                if (!llog_exist(next)) {
                        rc = llog_declare_create(env, next, th);
-                       llog_declare_write_rec(env, cathandle, NULL, -1, th);
+                       llog_declare_write_rec(env, cathandle, &lirec->lid_hdr,
+                                              -1, th);
                }
-               llog_declare_write_rec(env, next, rec, -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:
        RETURN(rc);
@@ -445,8 +425,7 @@ out:
 EXPORT_SYMBOL(llog_cat_declare_add_rec);
 
 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
-                struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
-                void *buf)
+                struct llog_rec_hdr *rec, struct llog_cookie *reccookie)
 {
        struct llog_ctxt        *ctxt;
        struct dt_device        *dt;
@@ -457,31 +436,23 @@ int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
        LASSERT(ctxt);
        LASSERT(ctxt->loc_exp);
 
-       if (cathandle->lgh_obj != NULL) {
-               dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
-               LASSERT(dt);
+       LASSERT(cathandle->lgh_obj != NULL);
+       dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev);
 
-               th = dt_trans_create(env, dt);
-               if (IS_ERR(th))
-                       RETURN(PTR_ERR(th));
+       th = dt_trans_create(env, dt);
+       if (IS_ERR(th))
+               RETURN(PTR_ERR(th));
 
-               rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
-               if (rc)
-                       GOTO(out_trans, rc);
+       rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
+       if (rc)
+               GOTO(out_trans, rc);
 
-               rc = dt_trans_start_local(env, dt, th);
-               if (rc)
-                       GOTO(out_trans, rc);
-               rc = llog_cat_add_rec(env, cathandle, rec, reccookie, buf, th);
+       rc = dt_trans_start_local(env, dt, th);
+       if (rc)
+               GOTO(out_trans, rc);
+       rc = llog_cat_add_rec(env, cathandle, rec, reccookie, th);
 out_trans:
-               dt_trans_stop(env, dt, th);
-       } else { /* lvfs compat code */
-               LASSERT(cathandle->lgh_file != NULL);
-               rc = llog_cat_declare_add_rec(env, cathandle, rec, th);
-               if (rc == 0)
-                       rc = llog_cat_add_rec(env, cathandle, rec, reccookie,
-                                             buf, th);
-       }
+       dt_trans_stop(env, dt, th);
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_cat_add);
@@ -510,28 +481,18 @@ int llog_cat_cancel_records(const struct lu_env *env,
 
                rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
                if (rc) {
-                       CERROR("%s: cannot find handle for llog "LPX64": %d\n",
+                       CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
                               cathandle->lgh_ctxt->loc_obd->obd_name,
-                              lgl->lgl_oid, rc);
-                       break;
+                              POSTID(&lgl->lgl_oi), rc);
+                       failed++;
+                       continue;
                }
 
                lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index);
-               if (lrc == 1) {          /* log has been destroyed */
+               if (lrc == LLOG_DEL_PLAIN) { /* log has been destroyed */
                        index = loghandle->u.phd.phd_cookie.lgc_index;
-                       down_write(&cathandle->lgh_lock);
-                       if (cathandle->u.chd.chd_current_log == loghandle)
-                               cathandle->u.chd.chd_current_log = NULL;
-                       up_write(&cathandle->lgh_lock);
-                       llog_close(env, loghandle);
-
-                       LASSERT(index);
-                       llog_cat_set_first_idx(cathandle, index);
-                       lrc = llog_cancel_rec(env, cathandle, index);
-                       if (lrc == 0)
-                               CDEBUG(D_RPCTRACE, "cancel plain log at index"
-                                      " %u of catalog "LPX64"\n",
-                                      index, cathandle->lgh_id.lgl_oid);
+                       rc = llog_cat_cleanup(env, cathandle, loghandle,
+                                             index);
                } else if (lrc == -ENOENT) {
                        if (rc == 0) /* ENOENT shouldn't rewrite any error */
                                rc = lrc;
@@ -539,6 +500,7 @@ int llog_cat_cancel_records(const struct lu_env *env,
                        failed++;
                        rc = lrc;
                }
+               llog_handle_put(loghandle);
        }
        if (rc)
                CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n",
@@ -549,12 +511,14 @@ int llog_cat_cancel_records(const struct lu_env *env,
 }
 EXPORT_SYMBOL(llog_cat_cancel_records);
 
-int llog_cat_process_cb(const struct lu_env *env, struct llog_handle *cat_llh,
-                       struct llog_rec_hdr *rec, void *data)
+static int llog_cat_process_cb(const struct lu_env *env,
+                              struct llog_handle *cat_llh,
+                              struct llog_rec_hdr *rec, void *data)
 {
         struct llog_process_data *d = data;
         struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
         struct llog_handle *llh;
+       struct llog_log_hdr *hdr;
         int rc;
 
         ENTRY;
@@ -562,37 +526,67 @@ int llog_cat_process_cb(const struct lu_env *env, struct llog_handle *cat_llh,
                 CERROR("invalid record in catalog\n");
                 RETURN(-EINVAL);
         }
-        CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
-               LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
-               rec->lrh_index, cat_llh->lgh_id.lgl_oid);
+       CDEBUG(D_HA, "processing log "DOSTID":%x at index %u of catalog "
+              DOSTID"\n", POSTID(&lir->lid_id.lgl_oi), lir->lid_id.lgl_ogen,
+              rec->lrh_index, POSTID(&cat_llh->lgh_id.lgl_oi));
 
        rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id);
        if (rc) {
-               CERROR("%s: cannot find handle for llog "LPX64": %d\n",
+               CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
                       cat_llh->lgh_ctxt->loc_obd->obd_name,
-                      lir->lid_id.lgl_oid, rc);
+                      POSTID(&lir->lid_id.lgl_oi), rc);
+               if (rc == -ENOENT || rc == -ESTALE) {
+                       /* After a server crash, a stub of index
+                        * record in catlog could be kept, because
+                        * plain log destroy + catlog index record
+                        * deletion are not atomic. So we end up with
+                        * an index but no actual record. Destroy the
+                        * index and move on. */
+                       rc = llog_cat_cleanup(env, cat_llh, NULL,
+                                             rec->lrh_index);
+               }
+
                RETURN(rc);
        }
 
-        if (rec->lrh_index < d->lpd_startcat)
-                /* Skip processing of the logs until startcat */
-                RETURN(0);
+       /* clean old empty llogs, do not consider current llog in use */
+       /* ignore remote (lgh_obj=NULL) llogs */
+       hdr = llh->lgh_hdr;
+       if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
+           hdr->llh_count == 1 && cat_llh->lgh_obj != NULL &&
+           llh != cat_llh->u.chd.chd_current_log) {
+               rc = llog_destroy(env, llh);
+               if (rc)
+                       CERROR("%s: fail to destroy empty log: rc = %d\n",
+                              llh->lgh_ctxt->loc_obd->obd_name, rc);
+               GOTO(out, rc = LLOG_DEL_PLAIN);
+       }
 
-        if (d->lpd_startidx > 0) {
+       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;
 
                 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 */
-                d->lpd_startidx = 0;
-        } else {
+               /* Continue processing the next log from idx 0 */
+               d->lpd_startidx = 0;
+       } else {
                rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data,
                                          NULL, false);
-        }
+       }
 
-        RETURN(rc);
+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);
+       llog_handle_put(llh);
+
+       RETURN(rc);
 }
 
 int llog_cat_process_or_fork(const struct lu_env *env,
@@ -614,18 +608,18 @@ int llog_cat_process_or_fork(const struct lu_env *env,
         if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
                 struct llog_process_cat_data cd;
 
-                CWARN("catlog "LPX64" crosses index zero\n",
-                      cat_llh->lgh_id.lgl_oid);
+                CWARN("catlog "DOSTID" crosses index zero\n",
+                      POSTID(&cat_llh->lgh_id.lgl_oi));
 
                 cd.lpcd_first_idx = llh->llh_cat_idx;
                 cd.lpcd_last_idx = 0;
                rc = llog_process_or_fork(env, cat_llh, llog_cat_process_cb,
                                          &d, &cd, fork);
-                if (rc != 0)
-                        RETURN(rc);
+               if (rc != 0)
+                       RETURN(rc);
 
-                cd.lpcd_first_idx = 0;
-                cd.lpcd_last_idx = cat_llh->lgh_last_idx;
+               cd.lpcd_first_idx = 0;
+               cd.lpcd_last_idx = cat_llh->lgh_last_idx;
                rc = llog_process_or_fork(env, cat_llh, llog_cat_process_cb,
                                          &d, &cd, fork);
         } else {
@@ -645,99 +639,65 @@ int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
 }
 EXPORT_SYMBOL(llog_cat_process);
 
-#ifdef __KERNEL__
-int llog_cat_process_thread(void *data)
-{
-        struct llog_process_cat_args *args = data;
-        struct llog_ctxt *ctxt = args->lpca_ctxt;
-        struct llog_handle *llh = NULL;
-        llog_cb_t cb = args->lpca_cb;
-       struct llog_thread_info *lgi;
-       struct lu_env            env;
-        int rc;
-        ENTRY;
-
-        cfs_daemonize_ctxt("ll_log_process");
-
-       rc = lu_env_init(&env, LCT_LOCAL);
-       if (rc)
-               GOTO(out, rc);
-       lgi = llog_info(&env);
-       LASSERT(lgi);
-
-       lgi->lgi_logid = *(struct llog_logid *)(args->lpca_arg);
-       rc = llog_open(&env, ctxt, &llh, &lgi->lgi_logid, NULL,
-                      LLOG_OPEN_EXISTS);
-       if (rc) {
-               CERROR("%s: cannot open llog "LPX64":%x: rc = %d\n",
-                      ctxt->loc_obd->obd_name, lgi->lgi_logid.lgl_oid,
-                      lgi->lgi_logid.lgl_ogen, rc);
-               GOTO(out_env, rc);
-       }
-       rc = llog_init_handle(&env, llh, LLOG_F_IS_CAT, NULL);
-       if (rc) {
-               CERROR("%s: llog_init_handle failed: rc = %d\n",
-                      llh->lgh_ctxt->loc_obd->obd_name, rc);
-               GOTO(release_llh, rc);
-       }
-
-       if (cb) {
-               rc = llog_cat_process(&env, llh, cb, NULL, 0, 0);
-               if (rc != LLOG_PROC_BREAK && rc != 0)
-                       CERROR("%s: llog_cat_process() failed: rc = %d\n",
-                              llh->lgh_ctxt->loc_obd->obd_name, rc);
-               cb(&env, llh, NULL, NULL);
-       } else {
-               CWARN("No callback function for recovery\n");
-       }
-
-       /*
-        * Make sure that all cached data is sent.
-        */
-       llog_sync(ctxt, NULL, 0);
-       GOTO(release_llh, rc);
-release_llh:
-       rc = llog_cat_close(&env, llh);
-       if (rc)
-               CERROR("%s: llog_cat_close() failed: rc = %d\n",
-                      llh->lgh_ctxt->loc_obd->obd_name, rc);
-out_env:
-       lu_env_fini(&env);
-out:
-       llog_ctxt_put(ctxt);
-       OBD_FREE_PTR(args);
-       return rc;
-}
-EXPORT_SYMBOL(llog_cat_process_thread);
-#endif
-
 static int llog_cat_reverse_process_cb(const struct lu_env *env,
                                       struct llog_handle *cat_llh,
                                       struct llog_rec_hdr *rec, void *data)
 {
-        struct llog_process_data *d = data;
-        struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
-        struct llog_handle *llh;
-        int rc;
+       struct llog_process_data *d = data;
+       struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
+       struct llog_handle *llh;
+       struct llog_log_hdr *hdr;
+       int rc;
 
-        if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
-                CERROR("invalid record in catalog\n");
-                RETURN(-EINVAL);
-        }
-        CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
-               LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
-               le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
+       if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
+               CERROR("invalid record in catalog\n");
+               RETURN(-EINVAL);
+       }
+       CDEBUG(D_HA, "processing log "DOSTID":%x at index %u of catalog "
+              DOSTID"\n", POSTID(&lir->lid_id.lgl_oi), lir->lid_id.lgl_ogen,
+              le32_to_cpu(rec->lrh_index), POSTID(&cat_llh->lgh_id.lgl_oi));
 
        rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id);
        if (rc) {
-               CERROR("%s: cannot find handle for llog "LPX64": %d\n",
+               CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
                       cat_llh->lgh_ctxt->loc_obd->obd_name,
-                      lir->lid_id.lgl_oid, rc);
+                      POSTID(&lir->lid_id.lgl_oi), rc);
+               if (rc == -ENOENT || rc == -ESTALE) {
+                       /* After a server crash, a stub of index
+                        * record in catlog could be kept, because
+                        * plain log destroy + catlog index record
+                        * deletion are not atomic. So we end up with
+                        * an index but no actual record. Destroy the
+                        * index and move on. */
+                       rc = llog_cat_cleanup(env, cat_llh, NULL,
+                                             rec->lrh_index);
+               }
+
                RETURN(rc);
        }
 
+       /* clean old empty llogs, do not consider current llog in use */
+       hdr = llh->lgh_hdr;
+       if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
+           hdr->llh_count == 1 &&
+           llh != cat_llh->u.chd.chd_current_log) {
+               rc = llog_destroy(env, llh);
+               if (rc)
+                       CERROR("%s: fail to destroy empty log: rc = %d\n",
+                              llh->lgh_ctxt->loc_obd->obd_name, rc);
+               GOTO(out, rc = LLOG_DEL_PLAIN);
+       }
+
        rc = llog_reverse_process(env, llh, d->lpd_cb, d->lpd_data, NULL);
-        RETURN(rc);
+
+out:
+       /* The empty plain was destroyed while processing */
+       if (rc == LLOG_DEL_PLAIN)
+               rc = llog_cat_cleanup(env, cat_llh, llh,
+                                     llh->u.phd.phd_cookie.lgc_index);
+
+       llog_handle_put(llh);
+       RETURN(rc);
 }
 
 int llog_cat_reverse_process(const struct lu_env *env,
@@ -754,20 +714,20 @@ int llog_cat_reverse_process(const struct lu_env *env,
         d.lpd_data = data;
         d.lpd_cb = cb;
 
-        if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
-                CWARN("catalog "LPX64" crosses index zero\n",
-                      cat_llh->lgh_id.lgl_oid);
+       if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
+               CWARN("catalog "DOSTID" crosses index zero\n",
+                     POSTID(&cat_llh->lgh_id.lgl_oi));
 
                 cd.lpcd_first_idx = 0;
                 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
                rc = llog_reverse_process(env, cat_llh,
                                          llog_cat_reverse_process_cb,
                                          &d, &cd);
-                if (rc != 0)
-                        RETURN(rc);
+               if (rc != 0)
+                       RETURN(rc);
 
-                cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
-                cd.lpcd_last_idx = 0;
+               cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
+               cd.lpcd_last_idx = 0;
                rc = llog_reverse_process(env, cat_llh,
                                          llog_cat_reverse_process_cb,
                                          &d, &cd);
@@ -781,7 +741,7 @@ int llog_cat_reverse_process(const struct lu_env *env,
 }
 EXPORT_SYMBOL(llog_cat_reverse_process);
 
-int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
+static int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
 {
         struct llog_log_hdr *llh = cathandle->lgh_hdr;
         int i, bitmap_size, idx;
@@ -806,40 +766,70 @@ int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
                         }
                 }
 out:
-                CDEBUG(D_RPCTRACE, "set catlog "LPX64" first idx %u\n",
-                       cathandle->lgh_id.lgl_oid, llh->llh_cat_idx);
-        }
+               CDEBUG(D_RPCTRACE, "set catlog "DOSTID" first idx %u\n",
+                      POSTID(&cathandle->lgh_id.lgl_oi), llh->llh_cat_idx);
+       }
 
-        RETURN(0);
+       RETURN(0);
 }
 
-int cat_cancel_cb(const struct lu_env *env, struct llog_handle *cathandle,
+/* Cleanup deleted plain llog traces from catalog */
+int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle,
+                    struct llog_handle *loghandle, int index)
+{
+       int rc;
+
+       LASSERT(index);
+       if (loghandle != NULL) {
+               /* remove destroyed llog from catalog list and
+                * chd_current_log variable */
+               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);
+               LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index);
+               /* llog was opened and keep in a list, close it now */
+               llog_close(env, loghandle);
+       }
+       /* 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 "DOSTID"\n",
+                      index, POSTID(&cathandle->lgh_id.lgl_oi));
+       return rc;
+}
+
+int cat_cancel_cb(const struct lu_env *env,
+                 struct llog_handle *cathandle,
                  struct llog_rec_hdr *rec, void *data)
 {
        struct llog_logid_rec   *lir = (struct llog_logid_rec *)rec;
        struct llog_handle      *loghandle;
        struct llog_log_hdr     *llh;
-       int                      rc, index;
+       int                      rc;
 
        ENTRY;
 
        if (rec->lrh_type != LLOG_LOGID_MAGIC) {
-               CERROR("%s: invalid record in catalog\n",
-                      loghandle->lgh_ctxt->loc_obd->obd_name);
+               CERROR("invalid record in catalog\n");
                RETURN(-EINVAL);
        }
-       CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
-              LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
-              rec->lrh_index, cathandle->lgh_id.lgl_oid);
+
+       CDEBUG(D_HA, "processing log "DOSTID":%x at index %u of catalog "
+              DOSTID"\n", POSTID(&lir->lid_id.lgl_oi), lir->lid_id.lgl_ogen,
+              rec->lrh_index, POSTID(&cathandle->lgh_id.lgl_oi));
 
        rc = llog_cat_id2handle(env, cathandle, &loghandle, &lir->lid_id);
        if (rc) {
-               CERROR("%s: cannot find handle for llog "LPX64": %d\n",
+               CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
                       cathandle->lgh_ctxt->loc_obd->obd_name,
-                      lir->lid_id.lgl_oid, rc);
+                      POSTID(&lir->lid_id.lgl_oi), rc);
                if (rc == -ENOENT || rc == -ESTALE) {
-                       index = rec->lrh_index;
-                       goto cat_cleanup;
+                       /* remove index from catalog */
+                       llog_cat_cleanup(env, cathandle, NULL, rec->lrh_index);
                }
                RETURN(rc);
        }
@@ -852,20 +842,10 @@ int cat_cancel_cb(const struct lu_env *env, struct llog_handle *cathandle,
                        CERROR("%s: fail to destroy empty log: rc = %d\n",
                               loghandle->lgh_ctxt->loc_obd->obd_name, rc);
 
-               index = loghandle->u.phd.phd_cookie.lgc_index;
-               llog_close(env, loghandle);
-
-cat_cleanup:
-               LASSERT(index);
-               llog_cat_set_first_idx(cathandle, index);
-               rc = llog_cancel_rec(env, cathandle, index);
-               if (rc == 0)
-                       CDEBUG(D_HA,
-                              "cancel log "LPX64":%x at index %u of catalog "
-                              LPX64"\n", lir->lid_id.lgl_oid,
-                              lir->lid_id.lgl_ogen, rec->lrh_index,
-                              cathandle->lgh_id.lgl_oid);
+               llog_cat_cleanup(env, cathandle, loghandle,
+                                loghandle->u.phd.phd_cookie.lgc_index);
        }
+       llog_handle_put(loghandle);
 
        RETURN(rc);
 }
@@ -881,11 +861,6 @@ int llog_cat_init_and_process(const struct lu_env *env,
        if (rc)
                RETURN(rc);
 
-       rc = llog_process(env, llh, cat_cancel_cb, NULL, NULL);
-       if (rc)
-               CERROR("%s: llog_process() with cat_cancel_cb failed: rc = "
-                      "%d\n", llh->lgh_ctxt->loc_obd->obd_name, rc);
        RETURN(0);
 }
 EXPORT_SYMBOL(llog_cat_init_and_process);
-