* Copyright (C) 2006 Michael Vrable
*
* Backup data is stored in a collection of objects, which are grouped together
- * into segments for storage purposes. This file provides interfaces for
- * reading and writing objects and segments. */
+ * into segments for storage purposes. This implementation of the object store
+ * is built on top of libtar, and represents segments as TAR files and objects
+ * as files within them. */
#ifndef _LBS_STORE_H
#define _LBS_STORE_H
#include <stdint.h>
+#include <libtar.h>
-#include <exception>
+#include <list>
#include <map>
+#include <set>
#include <string>
+#include <iostream>
#include <sstream>
-#include <vector>
#include "sha1.h"
+#include "ref.h"
+
+class LbsObject;
/* In memory datatype to represent key/value pairs of information, such as file
* 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
std::string getError() const { return error; }
};
-/* OutputStream is an abstract interface for writing data without seeking.
- * Output could be to a file, to an object within a segment, or even to a
- * memory buffer to help serialize data. */
-class OutputStream {
-public:
- OutputStream();
- virtual ~OutputStream() { }
-
- // Write the given data buffer
- void write(const void *data, size_t len);
-
- // Return the total number of bytes written so far
- int64_t get_pos() const { return bytes_written; }
-
- // Convenience functions for writing other data types. Values are always
- // written out in little-endian order.
- void write_u8(uint8_t val);
- void write_u16(uint16_t val);
- void write_u32(uint32_t val);
- void write_u64(uint64_t val);
-
- void write_s32(int32_t val) { write_u32((uint32_t)val); }
- void write_s64(int64_t val) { write_u64((uint64_t)val); }
-
- 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);
-
-protected:
- // Function which actually causes a write: must be overridden by
- // implementation.
- virtual void write_internal(const void *data, size_t len) = 0;
-
-private:
- int64_t bytes_written;
-};
-
-/* An OutputStream implementation which writes data to memory and returns the
- * result as a string. */
-class StringOutputStream : public OutputStream {
-public:
- StringOutputStream();
- std::string contents() const { return buf.str(); }
-
-protected:
- virtual void write_internal(const void *data, size_t len);
-
-private:
- std::stringstream buf;
-};
-
-/* An OutputStream implementation which writes data via the C stdio layer. */
-class FileOutputStream : public OutputStream {
-public:
- explicit FileOutputStream(FILE *file);
- virtual ~FileOutputStream();
-
-protected:
- virtual void write_internal(const void *data, size_t len);
-
-private:
- FILE *f;
-};
-
-/* An OutputStream which is simply sends writes to another OutputStream, but
- * does provide separate tracking of bytes written. */
-class WrapperOutputStream : public OutputStream {
+/* A simple wrapper around a single TAR file to represent a segment. Objects
+ * may only be written out all at once, since the tar header must be written
+ * first; incremental writing is not supported. */
+class Tarfile {
public:
- explicit WrapperOutputStream(OutputStream &o);
- virtual ~WrapperOutputStream() { }
+ Tarfile(const std::string &path, const std::string &segment);
+ ~Tarfile();
-protected:
- virtual void write_internal(const void *data, size_t len);
+ void write_object(int id, const char *data, size_t len);
-private:
- OutputStream ℜ
-};
-
-/* Like WrapperOutputStream, but additionally computes a checksum of data as it
- * is written. */
-class ChecksumOutputStream : public OutputStream {
-public:
- explicit ChecksumOutputStream(OutputStream &o);
- virtual ~ChecksumOutputStream() { }
+ // Return an estimate of the size of the file.
+ size_t size_estimate() { return size; }
- /* Once a checksum is computed, no further data should be written to the
- * stream. */
- const uint8_t *finish_and_checksum();
- size_t checksum_size() const { return csum.checksum_size(); }
-
-protected:
- virtual void write_internal(const void *data, size_t len);
+ void internal_write_object(const std::string &path,
+ const char *data, size_t len);
private:
- OutputStream ℜ
- SHA1Checksum csum;
+ size_t size;
+ std::string segment_name;
+ TAR *t;
};
-/* Simple wrappers that encode integers using a StringOutputStream and return
- * the encoded result. */
-std::string encode_u16(uint16_t val);
-std::string encode_u32(uint32_t val);
-std::string encode_u64(uint64_t val);
-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
- * OutputStream to which a new object may be written, and optionally
- * finish_object() when finished writing the current object. Only one object
- * may be written to a segment at a time; if multiple objects must be written
- * concurrently, they must be to different segments. */
-class SegmentWriter {
+class TarSegmentStore {
public:
- SegmentWriter(OutputStream *output, struct uuid u);
- ~SegmentWriter();
-
- struct uuid get_uuid() const { return id; }
-
- // Start writing out a new object to this segment.
- OutputStream *new_object(int *id, const char *type);
- void finish_object();
+ // New segments will be stored in the given directory.
+ TarSegmentStore(const std::string &path) { this->path = path; }
+ ~TarSegmentStore() { sync(); }
- // Determine size of segment data written out so far.
- size_t get_size() const { return raw_out->get_pos(); }
+ // Writes an object to segment in the store, and returns the name
+ // (segment/object) to refer to it. The optional parameter group can be
+ // used to control object placement; objects with different group
+ // parameters are kept in separate segments.
+ ObjectReference write_object(const char *data, size_t len,
+ const std::string &group = "");
- // Utility functions for generating and formatting UUIDs for display.
- static struct uuid generate_uuid();
- static std::string format_uuid(const struct uuid u);
+ // Ensure all segments have been fully written.
+ void sync();
private:
- struct index_info {
- int64_t offset; // File offset at which object starts
- int64_t size; // Size of object in bytes
- char type[4]; // Object type code
+ struct segment_info {
+ Tarfile *file;
+ std::string name; // UUID
+ int count; // Objects written to this segment
};
- typedef std::vector<struct index_info> object_table;
-
- ChecksumOutputStream *out; // Output stream with checksumming enabled
- OutputStream *raw_out; // Raw output stream, without checksumming
- struct uuid id;
+ std::string path;
+ std::map<std::string, struct segment_info *> segments;
- OutputStream *object_stream;
+ // Ensure that all segments in the given group have been fully written.
+ void close_segment(const std::string &group);
- object_table objects;
+ // Parse an object reference string and return just the segment name
+ // portion.
+ std::string object_reference_to_segment(const std::string &object);
};
-/* A SegmentStore, as the name suggests, is used to store the contents of many
- * segments. The SegmentStore internally tracks where data should be placed
- * (such as a local directory or remote storage), and allows new segments to be
- * easily created as needed. */
-class SegmentStore {
+/* An in-memory representation of an object, which can be incrementally built
+ * before it is written out to a segment. */
+class LbsObject {
public:
- // New segments will be stored in the given directory.
- SegmentStore(const std::string &path);
-
- SegmentWriter *new_segment();
-
-private:
- std::string directory;
-};
-
-/* A SegmentPartitioner helps to divide objects up among a collection of
- * segments to meet a rough size limit per segment. Like a SegmentWriter, only
- * one object should be written at a time; however, multiple
- * SegmentPartitioners can be created using the same base SegmentStore. */
-class SegmentPartitioner {
-public:
- explicit SegmentPartitioner(SegmentStore *s);
- ~SegmentPartitioner();
-
- OutputStream *new_object(struct uuid *uuid, int *id, const char *type);
+ LbsObject();
+ ~LbsObject();
+
+ // If an object is placed in a group, it will be written out to segments
+ // only containing other objects in the same group. A group name is simply
+ // a string.
+ //std::string get_group() const { return group; }
+ void set_group(const std::string &g) { group = g; }
+
+ // Data in an object must be written all at once, and cannot be generated
+ // incrementally. Data can be an arbitrary block of binary data of any
+ // size. The pointer to the data need only remain valid until write() is
+ // called.
+ void set_data(const char *d, size_t len) { data = d; data_len = len; }
+
+ // Write an object to a segment, thus making it permanent. This function
+ // can be called at most once.
+ void write(TarSegmentStore *store);
+
+ // Compute the checksum of an object, and include it in the object
+ // reference. This should be called after write(), and the data specified
+ // by set_data() must remain valid through the call to checksum().
+ void checksum();
+
+ // An object is assigned a permanent name once it has been written to a
+ // segment. Until that time, its name cannot be determined.
+ std::string get_name() const { return ref.to_string(); }
+ ObjectReference get_ref() { return ref; }
private:
- size_t target_size;
+ std::string group;
+ const char *data;
+ size_t data_len;
- SegmentStore *store;
- SegmentWriter *segment;
- OutputStream *object;
+ bool written;
+ ObjectReference ref;
};
#endif // _LBS_STORE_H