[Qemu-devel] [RFC V2 01/16] qcow2: Add qcow2_co_dedup_resume to restart deduplication.

[Benoît Canet]

---
 block/qcow2-dedup.c |  180 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 180 insertions(+)

diff --git a/block/qcow2-dedup.c b/block/qcow2-dedup.c
index 0679a68..cd08594 100644
--- a/block/qcow2-dedup.c
+++ b/block/qcow2-dedup.c
@@ -34,6 +34,7 @@
 #include <skeinApi.h>
 #endif
 
+static void qcow2_dedup_reset(BlockDriverState *bs);
 static int qcow2_dedup_read_write_hash(BlockDriverState *bs,
                                        QCowHash *hash,
                                        uint64_t *first_logical_sect,
@@ -1020,6 +1021,175 @@ bool qcow2_dedup_is_running(BlockDriverState *bs)
     return s->has_dedup && s->dedup_status == DEDUP_STATUS_STARTED;
 }
 
+static bool hash_is_null(QCowHash *hash)
+{
+    QCowHash null_hash;
+    memset(&null_hash.data, 0, HASH_LENGTH);
+    return !memcmp(hash->data, null_hash.data, HASH_LENGTH);
+}
+
+static void qcow2_dedup_insert_hash_node(BlockDriverState *bs,
+                                         QCowHashNode *hash_node)
+{
+    BDRVQcowState *s = bs->opaque;
+
+    g_tree_insert(s->dedup_tree_by_hash, &hash_node->hash, hash_node);
+    g_tree_insert(s->dedup_tree_by_sect, &hash_node->physical_sect, hash_node);
+}
+
+/* This load the QCowHashNode corresponding to a given cluster index into ram
+ *
+ * @index: index of the given physical sector
+ * @ret:   0 on succes, negative on error
+ */
+static int qcow2_load_cluster_hash(BlockDriverState *bs,
+                                   uint64_t index)
+{
+    BDRVQcowState *s = bs->opaque;
+    int ret = 0;
+    QCowHash hash;
+    uint64_t first_logical_sect;
+    QCowHashNode *hash_node;
+
+    /* get the hash */
+    ret = qcow2_dedup_read_write_hash(bs, &hash,
+                                      &first_logical_sect,
+                                      index * s->cluster_sectors,
+                                      false);
+
+    if (ret < 0) {
+        error_report("Failed to load deduplication hash.");
+        return ret;
+    }
+
+    /* if the hash is null don't load it */
+    if (hash_is_null(&hash)) {
+        return ret;
+    }
+
+    hash_node = qcow2_dedup_build_qcow_hash_node(&hash,
+                                                 index * s->cluster_sectors,
+                                                 first_logical_sect);
+    qcow2_dedup_insert_hash_node(bs, hash_node);
+
+    return 0;
+}
+
+/* Load all the actives hashes into RAM
+ *
+ * @ret: 0 on success, negative on error
+ */
+static int qcow2_load_valid_hashes(BlockDriverState *bs)
+{
+    BDRVQcowState *s = bs->opaque;
+    uint64_t max_clusters, i;
+    int nb_hash_in_hash_block = s->hash_block_size / (HASH_LENGTH + 8);
+    int ret = 0;
+
+    max_clusters = s->dedup_table_size * nb_hash_in_hash_block;
+
+    /* load all the hash stored to disk in memory */
+    for (i = 0; i < max_clusters; i++) {
+        if (!(i % nb_hash_in_hash_block)) {
+            co_sleep_ns(rt_clock, s->dedup_co_delay);
+        }
+        qemu_co_mutex_lock(&s->lock);
+        ret = qcow2_load_cluster_hash(bs, i);
+        qemu_co_mutex_unlock(&s->lock);
+        if (ret < 0) {
+            return ret;
+        }
+    }
+
+    return 0;
+}
+
+static int qcow2_drop_to_dedup_stale_hash(BlockDriverState *bs,
+                                          uint64_t index)
+{
+    int ret = 0;
+    bool to_dedup;
+    uint64_t physical_sect;
+
+    to_dedup = qcow2_is_cluster_to_dedup(bs, index, &physical_sect, &ret);
+
+    if (ret < 0) {
+        return ret;
+    }
+
+    if (!to_dedup) {
+        return 0;
+    }
+
+    qcow2_remove_hash_node_by_sector(bs, physical_sect);
+    return 0;
+}
+
+/* For each l2 entry marked as QCOW_OFLAG_TO_DEDUP drop the obsolete hash
+ * from the trees
+ *
+ * @ret: 0 on success, negative on error
+ */
+static int qcow2_drop_to_dedup_hashes(BlockDriverState *bs)
+{
+    BDRVQcowState *s = bs->opaque;
+    uint64_t i;
+    int ret = 0;
+
+    /* for each l2 entry */
+    for (i = 0; i < s->l2_size * s->l1_size; i++) {
+        if (!(i % s->l2_size)) {
+            co_sleep_ns(rt_clock, s->dedup_co_delay);
+        }
+        qemu_co_mutex_lock(&s->lock);
+        ret = qcow2_drop_to_dedup_stale_hash(bs, i);
+        qemu_co_mutex_unlock(&s->lock);
+
+        if (ret < 0) {
+            return ret;
+        }
+    }
+
+    return 0;
+}
+
+/*
+ * This coroutine resume deduplication
+ *
+ * @data: the given BlockDriverState
+ * @ret:  NULL
+ */
+static void coroutine_fn qcow2_co_dedup_resume(void *opaque)
+{
+    BlockDriverState *bs = opaque;
+    BDRVQcowState *s = bs->opaque;
+    int ret = 0;
+
+    ret = qcow2_load_valid_hashes(bs);
+
+    if (ret < 0) {
+        goto fail;
+    }
+
+    ret = qcow2_drop_to_dedup_hashes(bs);
+
+    if (ret < 0) {
+        goto fail;
+    }
+
+    qemu_co_mutex_lock(&s->lock);
+    s->dedup_status = DEDUP_STATUS_STARTED;
+    qemu_co_mutex_unlock(&s->lock);
+
+    return;
+
+fail:
+    qemu_co_mutex_lock(&s->lock);
+    s->dedup_status = DEDUP_STATUS_STOPPED;
+    qcow2_dedup_reset(bs);
+    qemu_co_mutex_unlock(&s->lock);
+}
+
 static gint qcow2_dedup_compare_by_hash(gconstpointer a,
                                         gconstpointer b,
                                         gpointer data)
@@ -1089,6 +1259,12 @@ static void qcow2_dedup_free(BlockDriverState *bs)
     g_tree_destroy(s->dedup_tree_by_hash);
 }
 
+static void qcow2_dedup_reset(BlockDriverState *bs)
+{
+    qcow2_dedup_free(bs);
+    qcow2_dedup_alloc(bs);
+}
+
 int qcow2_dedup_init(BlockDriverState *bs)
 {
     BDRVQcowState *s = bs->opaque;
@@ -1109,6 +1285,10 @@ int qcow2_dedup_init(BlockDriverState *bs)
 
     s->dedup_status = DEDUP_STATUS_STARTING;
 
+    /* resume deduplication */
+    s->dedup_resume_co = qemu_coroutine_create(qcow2_co_dedup_resume);
+    qemu_coroutine_enter(s->dedup_resume_co, bs);
+
     return 0;
 }
 
-- 
1.7.10.4