static SegmentPartitioner *index_segment, *data_segment;
+/* Buffer for holding a single block of data read from a file. */
+static const int LBS_BLOCK_SIZE = 1024 * 1024;
+static char *block_buf;
+
void scandir(const string& path);
/* Converts time to microseconds since the epoch. */
return (int64_t)time * 1000000;
}
+/* Read data from a file descriptor and return the amount of data read. A
+ * short read (less than the requested size) will only occur if end-of-file is
+ * hit. */
+size_t file_read(int fd, char *buf, size_t maxlen)
+{
+ size_t bytes_read = 0;
+
+ while (true) {
+ ssize_t res = read(fd, buf, maxlen);
+ if (res < 0) {
+ if (errno == EINTR)
+ continue;
+ throw IOException("file_read: error reading");
+ } else if (res == 0) {
+ break;
+ } else {
+ bytes_read += res;
+ buf += res;
+ maxlen -= res;
+ }
+ }
+
+ return bytes_read;
+}
+
+/* Read the contents of a file (specified by an open file descriptor) and copy
+ * the data to the store. */
void dumpfile(int fd, dictionary &file_info)
{
struct stat stat_buf;
fstat(fd, &stat_buf);
int64_t size = 0;
- char buf[4096];
-
if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
printf("file is no longer a regular file!\n");
return;
}
+ /* 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. */
+ struct uuid segment_uuid;
+ int object_id;
+ OutputStream *index_data = index_segment->new_object(&segment_uuid,
+ &object_id);
+
SHA1Checksum hash;
while (true) {
- ssize_t res = read(fd, buf, sizeof(buf));
- if (res < 0) {
- if (errno == EINTR)
- continue;
- printf("Error while reading: %m\n");
- return;
- } else if (res == 0) {
+ struct uuid block_segment_uuid;
+ int block_object_id;
+
+ size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
+ if (bytes == 0)
break;
- } else {
- hash.process(buf, res);
- OutputStream *block = data_segment->new_object();
- block->write(buf, res);
- size += res;
- }
+
+ hash.process(block_buf, bytes);
+ OutputStream *block = data_segment->new_object(&block_segment_uuid,
+ &block_object_id);
+ block->write(block_buf, bytes);
+ index_data->write_uuid(block_segment_uuid);
+ index_data->write_u32(block_object_id);
+
+ size += bytes;
}
file_info["sha1"] = string((const char *)hash.checksum(),
hash.checksum_size());
+ file_info["data"] = encode_objref(segment_uuid, object_id);
}
void scanfile(const string& path)
int main(int argc, char *argv[])
{
+ block_buf = new char[LBS_BLOCK_SIZE];
+
segment_store = new SegmentStore(".");
SegmentWriter *sw = segment_store->new_segment();
- info_dump = sw->new_object();
+ info_dump = sw->new_object(NULL);
index_segment = new SegmentPartitioner(segment_store);
data_segment = new SegmentPartitioner(segment_store);
} while (val);
}
+void OutputStream::write_uuid(const struct uuid &u)
+{
+ write(u.bytes, 16);
+}
+
/* Write an arbitrary string by first writing out the length, followed by the
* data itself. */
void OutputStream::write_string(const string &s)
return s.contents();
}
+string encode_objref(const struct uuid &segment, uint32_t object)
+{
+ StringOutputStream s;
+ s.write_uuid(segment);
+ s.write_u32(object);
+ return s.contents();
+}
+
SegmentWriter::SegmentWriter(OutputStream *output, struct uuid u)
: raw_out(output),
id(u),
/* Write out the segment header first. */
static const char signature[] = "LBSSEG0\n";
out->write(signature, strlen(signature));
- out->write(id.bytes, sizeof(struct uuid));
+ out->write_uuid(id);
}
SegmentWriter::~SegmentWriter()
delete raw_out;
}
-OutputStream *SegmentWriter::new_object()
+OutputStream *SegmentWriter::new_object(int *id)
{
if (object_stream)
finish_object();
object_start_offset = out->get_pos();
object_stream = new WrapperOutputStream(*out);
+ if (id != NULL) {
+ *id = objects.size();
+ }
+
return object_stream;
}
delete segment;
}
-OutputStream *SegmentPartitioner::new_object()
+OutputStream *SegmentPartitioner::new_object(struct uuid *uuid, int *id)
{
if (segment != NULL && segment->get_size() > target_size) {
delete segment;
if (segment == NULL)
segment = store->new_segment();
- return segment->new_object();
+ if (uuid != NULL)
+ *uuid = segment->get_uuid();
+
+ return segment->new_object(id);
}
* metadata. Currently implemented as map<string, string>. */
typedef std::map<std::string, std::string> dictionary;
+/* In-memory representation of a UUID (Universally-Unique Identifier), which is
+ * used to name a segment. */
+struct uuid {
+ uint8_t bytes[16];
+};
+
/* IOException will be thrown if an error occurs while reading or writing in
* one of the I/O wrappers. Depending upon the context; this may be fatal or
* not--typically, errors reading/writing the store will be serious, but errors
void write_varint(uint64_t val);
+ void write_uuid(const struct uuid &u);
void write_string(const std::string &s);
void write_dictionary(const dictionary &d);
std::string encode_u16(uint16_t val);
std::string encode_u32(uint32_t val);
std::string encode_u64(uint64_t val);
-
-struct uuid {
- uint8_t bytes[16];
-};
+std::string encode_objref(const struct uuid &segment, uint32_t object);
/* A class which is used to pack multiple objects into a single segment, with a
* lookup table to quickly locate each object. Call new_object() to get an
struct uuid get_uuid() const { return id; }
// Start writing out a new object to this segment.
- OutputStream *new_object();
+ OutputStream *new_object(int *id);
void finish_object();
// Determine size of segment data written out so far.
explicit SegmentPartitioner(SegmentStore *s);
~SegmentPartitioner();
- OutputStream *new_object();
+ OutputStream *new_object(struct uuid *uuid, int *id);
private:
size_t target_size;
projects / cumulus.git / commitdiff