Add ZIL statistics.
[zfs.git] / module / zfs / zil.c
index 5296b38..9ab02d7 100644 (file)
  */
 
 /*
+ * See zil.h for more information about these fields.
+ */
+zil_stats_t zil_stats = {
+       { "zil_commit_count",              KSTAT_DATA_UINT64 },
+       { "zil_commit_writer_count",       KSTAT_DATA_UINT64 },
+       { "zil_itx_count",                 KSTAT_DATA_UINT64 },
+       { "zil_itx_indirect_count",        KSTAT_DATA_UINT64 },
+       { "zil_itx_indirect_bytes",        KSTAT_DATA_UINT64 },
+       { "zil_itx_copied_count",          KSTAT_DATA_UINT64 },
+       { "zil_itx_copied_bytes",          KSTAT_DATA_UINT64 },
+       { "zil_itx_needcopy_count",        KSTAT_DATA_UINT64 },
+       { "zil_itx_needcopy_bytes",        KSTAT_DATA_UINT64 },
+       { "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 },
+       { "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 },
+       { "zil_itx_metaslab_slog_count",   KSTAT_DATA_UINT64 },
+       { "zil_itx_metaslab_slog_bytes",   KSTAT_DATA_UINT64 },
+};
+
+static kstat_t *zil_ksp;
+
+/*
  * This global ZIL switch affects all pools
  */
 int zil_replay_disable = 0;    /* disable intent logging replay */
@@ -859,7 +880,7 @@ uint64_t zil_block_buckets[] = {
  * is less than the limit or the total list size is less than 2X the limit.
  * Limit checking is disabled by setting zil_slog_limit to UINT64_MAX.
  */
-uint64_t zil_slog_limit = 1024 * 1024;
+unsigned long zil_slog_limit = 1024 * 1024;
 #define        USE_SLOG(zilog) (((zilog)->zl_logbias == ZFS_LOGBIAS_LATENCY) && \
        (((zilog)->zl_cur_used < zil_slog_limit) || \
        ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))))
@@ -879,6 +900,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
        uint64_t txg;
        uint64_t zil_blksz, wsz;
        int i, error;
+       boolean_t use_slog;
 
        if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
                zilc = (zil_chain_t *)lwb->lwb_buf;
@@ -935,8 +957,19 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
 
        BP_ZERO(bp);
        /* pass the old blkptr in order to spread log blocks across devs */
+       use_slog = USE_SLOG(zilog);
        error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz,
-           USE_SLOG(zilog));
+           use_slog);
+       if (use_slog)
+       {
+               ZIL_STAT_BUMP(zil_itx_metaslab_slog_count);
+               ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused);
+       }
+       else
+       {
+               ZIL_STAT_BUMP(zil_itx_metaslab_normal_count);
+               ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused);
+       }
        if (!error) {
                ASSERT3U(bp->blk_birth, ==, txg);
                bp->blk_cksum = lwb->lwb_blk.blk_cksum;
@@ -1022,13 +1055,18 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
        lrc = (lr_t *)lr_buf;
        lrw = (lr_write_t *)lrc;
 
+       ZIL_STAT_BUMP(zil_itx_count);
+
        /*
         * If it's a write, fetch the data or get its blkptr as appropriate.
         */
        if (lrc->lrc_txtype == TX_WRITE) {
                if (txg > spa_freeze_txg(zilog->zl_spa))
                        txg_wait_synced(zilog->zl_dmu_pool, txg);
-               if (itx->itx_wr_state != WR_COPIED) {
+               if (itx->itx_wr_state == WR_COPIED) {
+                       ZIL_STAT_BUMP(zil_itx_copied_count);
+                       ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length);
+               } else {
                        char *dbuf;
                        int error;
 
@@ -1036,9 +1074,13 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
                                ASSERT(itx->itx_wr_state == WR_NEED_COPY);
                                dbuf = lr_buf + reclen;
                                lrw->lr_common.lrc_reclen += dlen;
+                               ZIL_STAT_BUMP(zil_itx_needcopy_count);
+                               ZIL_STAT_INCR(zil_itx_needcopy_bytes, lrw->lr_length);
                        } else {
                                ASSERT(itx->itx_wr_state == WR_INDIRECT);
                                dbuf = NULL;
+                               ZIL_STAT_BUMP(zil_itx_indirect_count);
+                               ZIL_STAT_INCR(zil_itx_indirect_bytes, lrw->lr_length);
                        }
                        error = zilog->zl_get_data(
                            itx->itx_private, lrw, dbuf, lwb->lwb_zio);
@@ -1497,6 +1539,8 @@ zil_commit(zilog_t *zilog, uint64_t foid)
        if (zilog->zl_sync == ZFS_SYNC_DISABLED)
                return;
 
+       ZIL_STAT_BUMP(zil_commit_count);
+
        /* move the async itxs for the foid to the sync queues */
        zil_async_to_sync(zilog, foid);
 
@@ -1512,6 +1556,7 @@ zil_commit(zilog_t *zilog, uint64_t foid)
 
        zilog->zl_next_batch++;
        zilog->zl_writer = B_TRUE;
+       ZIL_STAT_BUMP(zil_commit_writer_count);
        zil_commit_writer(zilog);
        zilog->zl_com_batch = mybatch;
        zilog->zl_writer = B_FALSE;
@@ -1600,12 +1645,26 @@ zil_init(void)
 {
        zil_lwb_cache = kmem_cache_create("zil_lwb_cache",
            sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0);
+
+       zil_ksp = kstat_create("zfs", 0, "zil", "misc",
+           KSTAT_TYPE_NAMED, sizeof(zil_stats) / sizeof(kstat_named_t),
+           KSTAT_FLAG_VIRTUAL);
+
+       if (zil_ksp != NULL) {
+               zil_ksp->ks_data = &zil_stats;
+               kstat_install(zil_ksp);
+       }
 }
 
 void
 zil_fini(void)
 {
        kmem_cache_destroy(zil_lwb_cache);
+
+       if (zil_ksp != NULL) {
+               kstat_delete(zil_ksp);
+               zil_ksp = NULL;
+       }
 }
 
 void
@@ -2010,4 +2069,7 @@ MODULE_PARM_DESC(zil_replay_disable, "Disable intent logging replay");
 
 module_param(zfs_nocacheflush, int, 0644);
 MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes");
+
+module_param(zil_slog_limit, ulong, 0644);
+MODULE_PARM_DESC(zil_slog_limit, "Max commit bytes to separate log device");
 #endif