Improve object deserialization: properly parse object headers.

[bluesky.git] / bluesky / cloudlog.c

diff --git a/bluesky/cloudlog.c b/bluesky/cloudlog.c
index beaf1b4..cc186a5 100644 (file)
--- a/bluesky/cloudlog.c
+++ b/bluesky/cloudlog.c
@@ -180,14 +180,29 @@ void bluesky_cloudlog_insert(BlueSkyCloudLog *log)
     g_mutex_unlock(log->fs->lock);
 }
 
-struct cloudlog_header {
-    char magic[4];
-    uint8_t type;
-    BlueSkyCloudID id;
-    uint32_t size1, size2, size3;
-} __attribute__((packed));
+/* Look up the cloud log entry for the given ID.  If create is TRUE and the
+ * item does not exist, create a special pending entry that can later be filled
+ * in when the real item is loaded.  The returned item has a reference held.
+ * As a special case, if a null ID is provided then NULL is returned. */
+BlueSkyCloudLog *bluesky_cloudlog_get(BlueSkyFS *fs, BlueSkyCloudID id)
+{
+    static BlueSkyCloudID id0 = {{0}};
 
-#define CLOUDLOG_MAGIC "AgI-"
+    if (memcmp(&id, &id0, sizeof(BlueSkyCloudID)) == 0)
+        return NULL;
+
+    g_mutex_lock(fs->lock);
+    BlueSkyCloudLog *item;
+    item = g_hash_table_lookup(fs->locations, &id);
+    if (item == NULL) {
+        item = bluesky_cloudlog_new(fs, &id);
+        g_hash_table_insert(fs->locations, &id, item);
+    } else {
+        bluesky_cloudlog_ref(item);
+    }
+    g_mutex_unlock(fs->lock);
+    return item;
+}
 
 /* Ensure that a cloud log item is loaded in memory, and if not read it in.
  * TODO: Make asynchronous, and make this also fetch from the cloud.  Right now
@@ -197,20 +212,23 @@ void bluesky_cloudlog_fetch(BlueSkyCloudLog *log)
     if (log->data != NULL)
         return;
 
+    int offset;
+
     if ((log->location_flags | log->pending_write) & CLOUDLOG_JOURNAL) {
         bluesky_cloudlog_stats_update(log, -1);
+        offset = log->log_offset + sizeof(struct log_header);
         log->data = bluesky_log_map_object(log->fs, -1, log->log_seq,
-                                           log->log_offset, log->log_size);
+                                           offset, log->data_size);
         bluesky_cloudlog_stats_update(log, 1);
     }
 
     if (log->data == NULL && (log->location_flags & CLOUDLOG_CLOUD)) {
         log->location_flags &= ~CLOUDLOG_JOURNAL;
         bluesky_cloudlog_stats_update(log, -1);
+        offset = log->location.offset + sizeof(struct cloudlog_header);
         log->data = bluesky_log_map_object(log->fs, log->location.directory,
                                            log->location.sequence,
-                                           log->location.offset,
-                                           log->location.size);
+                                           offset, log->data_size);
         bluesky_cloudlog_stats_update(log, 1);
     }
 
@@ -251,8 +269,8 @@ BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
     /* TODO: Right now offset/size are set to the raw data, but we should add
      * header parsing to the code which loads objects back in. */
     log->location = state->location;
-    log->location.offset = state->data->len + sizeof(struct cloudlog_header);
-    log->location.size = data1->len;
+    log->location.offset = state->data->len;
+    log->data_size = data1->len;
 
     struct cloudlog_header header;
     memcpy(header.magic, CLOUDLOG_MAGIC, 4);
@@ -267,6 +285,8 @@ BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
     g_string_append_len(state->data, data2->str, data2->len);
     g_string_append_len(state->data, data3->str, data3->len);
 
+    log->location.size = state->data->len - log->location.offset;
+
     /* If the object we flushed was an inode, update the inode map. */
     if (log->type == LOGTYPE_INODE) {
         g_mutex_lock(fs->lock);