+/* Writing of data to the cloud in log segments and tracking the location of
+ * various pieces of data (both where in the cloud and where cached locally).
+ * */
+
+/* Eventually we'll want to support multiple writers. But for now, hard-code
+ * separate namespaces in the cloud for the proxy and the cleaner to write to.
+ * */
+#define BLUESKY_CLOUD_DIR_PRIMARY 0
+#define BLUESKY_CLOUD_DIR_CLEANER 1
+
+typedef struct {
+ char bytes[16];
+} BlueSkyCloudID;
+
+typedef struct {
+ uint32_t directory;
+ uint32_t sequence;
+ uint32_t offset;
+ uint32_t size;
+} BlueSkyCloudPointer;
+
+typedef enum {
+ LOGTYPE_INVALID = -1,
+ LOGTYPE_UNKNOWN = 0,
+ LOGTYPE_DATA = 1,
+ LOGTYPE_INODE = 2,
+ LOGTYPE_INODE_MAP = 3,
+ LOGTYPE_CHECKPOINT = 4,
+
+ /* Used only as metadata in the local journal, not loaded as a
+ * BlueSkyCloudLogState nor stored in the cloud */
+ LOGTYPE_JOURNAL_MARKER = 16,
+ LOGTYPE_JOURNAL_CHECKPOINT = 17,
+} BlueSkyCloudLogType;
+
+/* Headers that go on items in local log segments and cloud log segments. */
+struct log_header {
+ uint32_t magic; // HEADER_MAGIC
+ uint8_t type; // Object type + '0'
+ uint32_t offset; // Starting byte offset of the log header
+ uint32_t size1; // Size of the data item (bytes)
+ uint32_t size2; //
+ uint32_t size3; //
+ uint64_t inum; // Inode which owns this data, if any
+ BlueSkyCloudID id; // Object identifier
+} __attribute__((packed));
+
+struct log_footer {
+ uint32_t magic; // FOOTER_MAGIC
+ uint32_t crc; // Computed from log_header to log_footer.magic
+} __attribute__((packed));
+
+struct cloudlog_header {
+ char magic[4];
+ uint8_t crypt_auth[CRYPTO_HASH_SIZE];
+ uint8_t crypt_iv[CRYPTO_BLOCK_SIZE];
+ uint8_t type;
+ BlueSkyCloudID id;
+ uint64_t inum;
+ uint32_t size1, size2, size3;
+} __attribute__((packed));
+
+// Rough size limit for a log segment. This is not a firm limit and there are
+// no absolute guarantees on the size of a log segment.
+#define LOG_SEGMENT_SIZE (1 << 22)
+
+#define JOURNAL_MAGIC "\nLog"
+#define CLOUDLOG_MAGIC "AgI-"
+#define CLOUDLOG_MAGIC_ENCRYPTED "AgI=" // CLOUDLOG_MAGIC[3] ^= 0x10
+
+/* A record which tracks an object which has been written to a local log,
+ * cached, locally, and/or written to the cloud. */
+#define CLOUDLOG_JOURNAL 0x01
+#define CLOUDLOG_CLOUD 0x02
+#define CLOUDLOG_CACHE 0x04
+#define CLOUDLOG_UNCOMMITTED 0x10
+struct BlueSkyCloudLog {
+ gint refcount;
+ GMutex *lock;
+ GCond *cond;
+
+ BlueSkyFS *fs;
+
+ BlueSkyCloudLogType type;
+
+ // Bitmask of CLOUDLOG_* flags indicating where the object exists.
+ int location_flags;
+ int pending_read, pending_write;
+
+ // A stable identifier for the object (only changes when authenticated data
+ // is written out, but stays the same when the in-cloud cleaner relocates
+ // the object).
+ BlueSkyCloudID id;
+
+ // The inode which owns this data, if any, and an offset.
+ uint64_t inum;
+ int32_t inum_offset;
+
+ // The size of encrypted object data, not including any headers
+ int data_size;
+
+ // The location of the object in the cloud, if available.
+ BlueSkyCloudPointer location;
+
+ // TODO: Location in journal/cache
+ int log_seq, log_offset, log_size;
+
+ // Pointers to other objects. Each link counts towards the reference count
+ // of the pointed-to object. To avoid memory leaks there should be no
+ // cycles in the reference graph.
+ GArray *links;
+
+ // Serialized data, if available in memory (otherwise NULL), and a lock
+ // count which tracks if there are users that require the data to be kept
+ // around.
+ BlueSkyRCStr *data;
+ int data_lock_count;
+};
+
+/* Serialize objects into a log segment to be written to the cloud. */
+struct BlueSkyCloudLogState {
+ GString *data;
+ BlueSkyCloudPointer location;
+ GList *inode_list;
+ GSList *writeback_list; // Items which are being serialized right now
+ GList *pending_segments; // Segments which are being uploaded now
+
+ int uploads_pending; // Count of uploads in progress, not completed
+ GMutex *uploads_pending_lock;
+ GCond *uploads_pending_cond;
+
+ /* What is the most recent sequence number written by the cleaner which we
+ * have processed and incorporated into our own log? This gets
+ * incorporated into the version vector written out with our checkpoint
+ * records. */
+ int latest_cleaner_seq_seen;
+};
+
+gboolean bluesky_cloudlog_equal(gconstpointer a, gconstpointer b);
+guint bluesky_cloudlog_hash(gconstpointer a);
+BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs, const BlueSkyCloudID *id);
+gchar *bluesky_cloudlog_id_to_string(BlueSkyCloudID id);
+BlueSkyCloudID bluesky_cloudlog_id_from_string(const gchar *idstr);
+void bluesky_cloudlog_threads_init(BlueSkyFS *fs);
+void bluesky_cloudlog_ref(BlueSkyCloudLog *log);
+void bluesky_cloudlog_unref(BlueSkyCloudLog *log);
+void bluesky_cloudlog_unref_delayed(BlueSkyCloudLog *log);
+void bluesky_cloudlog_erase(BlueSkyCloudLog *log);
+void bluesky_cloudlog_stats_update(BlueSkyCloudLog *log, int type);
+void bluesky_cloudlog_sync(BlueSkyCloudLog *log);
+void bluesky_cloudlog_insert(BlueSkyCloudLog *log);
+void bluesky_cloudlog_insert_locked(BlueSkyCloudLog *log);
+BlueSkyCloudLog *bluesky_cloudlog_get(BlueSkyFS *fs, BlueSkyCloudID id);
+void bluesky_cloudlog_prefetch(BlueSkyCloudLog *log);
+void bluesky_cloudlog_fetch(BlueSkyCloudLog *log);
+void bluesky_cloudlog_background_fetch(BlueSkyCloudLog *item);
+BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
+ BlueSkyFS *fs);
+void bluesky_cloudlog_flush(BlueSkyFS *fs);
+