+ return bytes_read;
+}
+
+/* Read the contents of a file (specified by an open file descriptor) and copy
+ * the data to the store. Returns the size of the file (number of bytes
+ * dumped), or -1 on error. */
+int64_t dumpfile(int fd, dictionary &file_info)
+{
+ struct stat stat_buf;
+ fstat(fd, &stat_buf);
+ int64_t size = 0;
+ list<string> object_list;
+
+ if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
+ fprintf(stderr, "file is no longer a regular file!\n");
+ return -1;
+ }
+
+ /* The index data consists of a sequence of pointers to the data blocks
+ * that actually comprise the file data. This level of indirection is used
+ * so that the same data block can be used in multiple files, or multiple
+ * versions of the same file. */
+ SHA1Checksum hash;
+ while (true) {
+ size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
+ if (bytes == 0)
+ break;
+
+ hash.process(block_buf, bytes);
+
+ // Either find a copy of this block in an already-existing segment, or
+ // index it so it can be re-used in the future
+ SHA1Checksum block_hash;
+ block_hash.process(block_buf, bytes);
+ string block_csum = block_hash.checksum_str();
+ ObjectReference ref = db->FindObject(block_csum, bytes);
+
+ // Store a copy of the object if one does not yet exist
+ if (ref.get_segment().size() == 0) {
+ LbsObject *o = new LbsObject;
+
+ /* We might still have seen this checksum before, if the object was
+ * stored at some time in the past, but we have decided to clean
+ * the segment the object was originally stored in (FindObject will
+ * not return such objects). When rewriting the object contents,
+ * put it in a separate group, so that old objects get grouped
+ * together. The hope is that these old objects will continue to
+ * be used in the future, and we obtain segments which will
+ * continue to be well-utilized. */
+ if (db->IsOldObject(block_csum, bytes))
+ o->set_group("compacted");
+ else
+ o->set_group("data");
+
+ o->set_data(block_buf, bytes);
+ o->write(tss);
+ ref = o->get_ref();
+ db->StoreObject(ref, block_csum, bytes);
+ delete o;
+ }
+
+ object_list.push_back(ref.to_string());
+ segment_list.insert(ref.get_segment());
+ db->UseObject(ref);
+ size += bytes;
+ }
+
+ file_info["checksum"] = hash.checksum_str();
+
+ /* For files that only need to be broken apart into a few objects, store
+ * the list of objects directly. For larger files, store the data
+ * out-of-line and provide a pointer to the indrect object. */
+ if (object_list.size() < 8) {
+ string blocklist = "";
+ for (list<string>::iterator i = object_list.begin();
+ i != object_list.end(); ++i) {
+ if (i != object_list.begin())
+ blocklist += " ";
+ blocklist += *i;
+ }
+ file_info["data"] = blocklist;
+ } else {
+ string blocklist = "";
+ for (list<string>::iterator i = object_list.begin();
+ i != object_list.end(); ++i) {
+ blocklist += *i + "\n";
+ }
+
+ LbsObject *i = new LbsObject;
+ i->set_group("indirect");
+ i->set_data(blocklist.data(), blocklist.size());
+ i->write(tss);
+ file_info["data"] = "@" + i->get_name();
+ segment_list.insert(i->get_ref().get_segment());
+ delete i;
+ }
+
+ return size;