Whamcloud - gitweb
LU-7340 mdd: changelogs garbage collection
[fs/lustre-release.git] / lustre / obdclass / llog.c
index 79a4dea..d3b1e04 100644 (file)
@@ -204,8 +204,8 @@ int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
 
        ENTRY;
 
-       CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
-              POSTID(&loghandle->lgh_id.lgl_oi));
+       CDEBUG(D_RPCTRACE, "Canceling %d in log "DFID"\n", index,
+              PFID(&loghandle->lgh_id.lgl_oi.oi_fid));
 
        if (index == 0) {
                CERROR("Can't cancel index 0 which is header\n");
@@ -270,12 +270,10 @@ int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
                         * the bitmap has been clearly, so the record can not
                         * be accessed anymore, let's return 0 for now, and
                         * the orphan will be handled by LFSCK. */
-                       CERROR("%s: can't destroy empty llog #"DOSTID
-                              "#%08x: rc = %d\n",
+                       CERROR("%s: can't destroy empty llog "DFID": rc = %d\n",
                               loghandle->lgh_ctxt->loc_obd->obd_name,
-                              POSTID(&loghandle->lgh_id.lgl_oi),
-                              loghandle->lgh_id.lgl_ogen, rc);
-                       GOTO(out_unlock, rc);
+                              PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rc);
+                       GOTO(out_unlock, rc = 0);
                }
                rc = LLOG_DEL_PLAIN;
        }
@@ -423,10 +421,11 @@ static int llog_process_thread(void *arg)
        struct llog_process_cat_data    *cd  = lpi->lpi_catdata;
        char                            *buf;
        size_t                           chunk_size;
-       __u64                            cur_offset, tmp_offset;
+       __u64                            cur_offset;
        int                              rc = 0, index = 1, last_index;
        int                              saved_index = 0;
        int                              last_called_index = 0;
+       bool                             repeated = false;
 
        ENTRY;
 
@@ -454,9 +453,10 @@ static int llog_process_thread(void *arg)
 
        while (rc == 0) {
                struct llog_rec_hdr *rec;
-               off_t chunk_offset;
+               off_t chunk_offset = 0;
                unsigned int buf_offset = 0;
                bool partial_chunk;
+               int     lh_last_idx;
 
                /* skip records not set in bitmap */
                while (index <= last_index &&
@@ -473,8 +473,19 @@ static int llog_process_thread(void *arg)
 repeat:
                /* get the buf with our target record; avoid old garbage */
                memset(buf, 0, chunk_size);
+               /* the record index for outdated chunk data */
+               lh_last_idx = loghandle->lgh_last_idx + 1;
                rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
                                     index, &cur_offset, buf, chunk_size);
+               if (repeated && rc)
+                       CDEBUG(D_OTHER, "cur_offset %llu, chunk_offset %llu,"
+                              " buf_offset %u, rc = %d\n", cur_offset,
+                              (__u64)chunk_offset, buf_offset, rc);
+               /* we`ve tried to reread the chunk, but there is no
+                * new records */
+               if (rc == -EIO && repeated && (chunk_offset + buf_offset) ==
+                   cur_offset)
+                       GOTO(out, rc = 0);
                if (rc != 0)
                        GOTO(out, rc);
 
@@ -483,8 +494,7 @@ repeat:
                 * The absolute offset of the current chunk is calculated
                 * from cur_offset value and stored in chunk_offset variable.
                 */
-               tmp_offset = cur_offset;
-               if (do_div(tmp_offset, chunk_size) != 0) {
+               if ((cur_offset & (chunk_size - 1)) != 0) {
                        partial_chunk = true;
                        chunk_offset = cur_offset & ~(chunk_size - 1);
                } else {
@@ -508,15 +518,28 @@ repeat:
                        CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
                               rec->lrh_type, rec->lrh_index);
 
+                       /* the bitmap could be changed during processing
+                        * records from the chunk. For wrapped catalog
+                        * it means we can read deleted record and try to
+                        * process it. Check this case and reread the chunk. */
+
                        /* for partial chunk the end of it is zeroed, check
                         * for index 0 to distinguish it. */
-                       if (partial_chunk && rec->lrh_index == 0) {
+                       if ((partial_chunk && rec->lrh_index == 0) ||
+                            (index == lh_last_idx &&
+                             lh_last_idx != (loghandle->lgh_last_idx + 1))) {
                                /* concurrent llog_add() might add new records
                                 * while llog_processing, check this is not
                                 * the case and re-read the current chunk
                                 * otherwise. */
                                int records;
-                               if (index > loghandle->lgh_last_idx)
+                               /* lgh_last_idx could be less then index
+                                * for catalog, if catalog is wrapped */
+                               if ((index > loghandle->lgh_last_idx &&
+                                   !(loghandle->lgh_hdr->llh_flags &
+                                     LLOG_F_IS_CAT)) || repeated ||
+                                   (loghandle->lgh_obj != NULL &&
+                                    dt_object_remote(loghandle->lgh_obj)))
                                        GOTO(out, rc = 0);
                                /* <2 records means no more records
                                 * if the last record we processed was
@@ -534,15 +557,18 @@ repeat:
                                /* save offset inside buffer for the re-read */
                                buf_offset = (char *)rec - (char *)buf;
                                cur_offset = chunk_offset;
+                               repeated = true;
                                goto repeat;
                        }
 
+                       repeated = false;
+
                        if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
-                               CWARN("%s: invalid length %d in llog "DOSTID
+                               CWARN("%s: invalid length %d in llog "DFID
                                      "record for index %d/%d\n",
                                       loghandle->lgh_ctxt->loc_obd->obd_name,
                                       rec->lrh_len,
-                                      POSTID(&loghandle->lgh_id.lgl_oi),
+                                      PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
                                       rec->lrh_index, index);
 
                                GOTO(out, rc = -EINVAL);
@@ -555,10 +581,10 @@ repeat:
                        }
 
                        if (rec->lrh_index != index) {
-                               CERROR("%s: "DOSTID" Invalid record: index %u"
+                               CERROR("%s: "DFID" Invalid record: index %u"
                                       " but expected %u\n",
                                       loghandle->lgh_ctxt->loc_obd->obd_name,
-                                      POSTID(&loghandle->lgh_id.lgl_oi),
+                                      PFID(&loghandle->lgh_id.lgl_oi.oi_fid),
                                       rec->lrh_index, index);
                                GOTO(out, rc = -ERANGE);
                        }
@@ -586,6 +612,12 @@ repeat:
                                }
                                if (rc)
                                        GOTO(out, rc);
+                               /* some stupid callbacks directly cancel records
+                                * and delete llog. Check it and stop
+                                * processing. */
+                               if (loghandle->lgh_hdr == NULL ||
+                                   loghandle->lgh_hdr->llh_count == 1)
+                                       GOTO(out, rc = 0);
                        }
                        /* exit if the last index is reached */
                        if (index >= last_index)
@@ -613,7 +645,14 @@ out:
                         * llog file, probably I/O error or the log got
                         * corrupted to be able to finally release the log we
                         * discard any remaining bits in the header */
-                       CERROR("Local llog found corrupted\n");
+                       CERROR("%s: Local llog found corrupted #"DOSTID":%x"
+                              " %s index %d count %d\n",
+                              loghandle->lgh_ctxt->loc_obd->obd_name,
+                              POSTID(&loghandle->lgh_id.lgl_oi),
+                              loghandle->lgh_id.lgl_ogen,
+                              ((llh->llh_flags & LLOG_F_IS_CAT) ? "catalog" :
+                               "plain"), index, llh->llh_count);
+
                        while (index <= last_index) {
                                if (ext2_test_bit(index,
                                                  LLOG_HDR_BITMAP(llh)) != 0)
@@ -626,8 +665,8 @@ out:
        }
 
        OBD_FREE_LARGE(buf, chunk_size);
-        lpi->lpi_rc = rc;
-        return 0;
+       lpi->lpi_rc = rc;
+       return 0;
 }
 
 static int llog_process_thread_daemonize(void *arg)
@@ -1309,8 +1348,9 @@ __u64 llog_size(const struct lu_env *env, struct llog_handle *llh)
 
        rc = llh->lgh_obj->do_ops->do_attr_get(env, llh->lgh_obj, &la);
        if (rc) {
-               CERROR("%s: attr_get failed, rc = %d\n",
-                      llh->lgh_ctxt->loc_obd->obd_name, rc);
+               CERROR("%s: attr_get failed for "DFID": rc = %d\n",
+                      llh->lgh_ctxt->loc_obd->obd_name,
+                      PFID(&llh->lgh_id.lgl_oi.oi_fid), rc);
                return 0;
        }