Whamcloud - gitweb
LU-11116 llog: error handling cleanup
[fs/lustre-release.git] / lustre / obdclass / llog_cat.c
index 0d57ce3..59a0950 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -213,8 +213,10 @@ 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_HDR_BITMAP_SIZE(loghandle->lgh_hdr) + 1;
        llog_trans_destroy(env, loghandle, th);
+       if (handle != NULL)
+               dt_trans_stop(env, dt, handle);
        RETURN(rc);
 }
 
@@ -386,8 +388,7 @@ static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
 
                down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
                llh = loghandle->lgh_hdr;
-               LASSERT(llh);
-               if (!llog_is_full(loghandle))
+               if (llh == NULL || !llog_is_full(loghandle))
                        GOTO(out_unlock, loghandle);
                else
                        up_write(&loghandle->lgh_lock);
@@ -487,7 +488,12 @@ retry:
                        up_write(&loghandle->lgh_lock);
                        /* nobody should be trying to use this llog */
                        down_write(&cathandle->lgh_lock);
-                       if (cathandle->u.chd.chd_current_log == loghandle)
+                       /* 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)
                                cathandle->u.chd.chd_current_log = NULL;
                        up_write(&cathandle->lgh_lock);
                        RETURN(rc);
@@ -754,16 +760,38 @@ int llog_cat_cancel_records(const struct lu_env *env,
                        continue;
                }
 
+               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;
-                       rc = llog_cat_cleanup(env, cathandle, loghandle, 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++;
-                       rc = lrc;
+                       if (rc == 0)
+                               rc = lrc;
                }
                llog_handle_put(loghandle);
        }
@@ -806,9 +834,8 @@ static int llog_cat_process_common(const struct lu_env *env,
                 * deletion are not atomic. So we end up with an index but no
                 * actual record. Destroy the index and move on. */
                if (rc == -ENOENT || rc == -ESTALE)
-                       rc = llog_cat_cleanup(env, cat_llh, NULL,
-                                             rec->lrh_index);
-               if (rc)
+                       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),
@@ -866,9 +893,14 @@ static int llog_cat_process_cb(const struct lu_env *env,
 
 out:
        /* The empty plain log was destroyed while processing */
-       if (rc == LLOG_DEL_PLAIN)
+       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) {
+               /* clear wrong catalog entry */
+               rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
+       }
+
        if (llh)
                llog_handle_put(llh);
 
@@ -938,13 +970,14 @@ static int llog_cat_size_cb(const struct lu_env *env,
 
        ENTRY;
        rc = llog_cat_process_common(env, cat_llh, rec, &llh);
-       if (llh == NULL)
-               RETURN(0);
 
        if (rc == LLOG_DEL_PLAIN) {
                /* empty log was deleted, don't count it */
                rc = llog_cat_cleanup(env, cat_llh, llh,
                                      llh->u.phd.phd_cookie.lgc_index);
+       } else if (rc == LLOG_DEL_RECORD) {
+               /* clear wrong catalog entry */
+               rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
        } else {
                size = llog_size(env, llh);
                *cum_size += size;
@@ -953,7 +986,8 @@ static int llog_cat_size_cb(const struct lu_env *env,
                       PFID(&llh->lgh_id.lgl_oi.oi_fid), size, *cum_size);
        }
 
-       llog_handle_put(llh);
+       if (llh != NULL)
+               llog_handle_put(llh);
 
        RETURN(0);
 }
@@ -969,6 +1003,28 @@ __u64 llog_cat_size(const struct lu_env *env, struct llog_handle *cat_llh)
 }
 EXPORT_SYMBOL(llog_cat_size);
 
+/* currently returns the number of "free" entries in catalog,
+ * ie the available entries for a new plain LLOG file creation,
+ * even if catalog has wrapped
+ */
+__u32 llog_cat_free_space(struct llog_handle *cat_llh)
+{
+       /* simulate almost full Catalog */
+       if (OBD_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;
+
+       if (cat_llh->lgh_last_idx > cat_llh->lgh_hdr->llh_cat_idx)
+               return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1 +
+                      cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
+
+       /* catalog is presently wrapped */
+       return cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx;
+}
+EXPORT_SYMBOL(llog_cat_free_space);
+
 static int llog_cat_reverse_process_cb(const struct lu_env *env,
                                       struct llog_handle *cat_llh,
                                       struct llog_rec_hdr *rec, void *data)
@@ -979,6 +1035,15 @@ static int llog_cat_reverse_process_cb(const struct lu_env *env,
 
        ENTRY;
        rc = llog_cat_process_common(env, cat_llh, rec, &llh);
+
+       /* 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) {
+               /* clear wrong catalog entry */
+               rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index);
+       }
        if (rc)
                RETURN(rc);