1 /* LBS: An LFS-inspired filesystem backup system
2 * Copyright (C) 2006 Michael Vrable
4 * Backup data is stored in a collection of objects, which are grouped together
5 * into segments for storage purposes. This implementation of the object store
6 * represents segments as TAR files and objects as files within them. */
25 /* In memory datatype to represent key/value pairs of information, such as file
26 * metadata. Currently implemented as map<string, string>. */
27 typedef std::map<std::string, std::string> dictionary;
29 /* IOException will be thrown if an error occurs while reading or writing in
30 * one of the I/O wrappers. Depending upon the context; this may be fatal or
31 * not--typically, errors reading/writing the store will be serious, but errors
32 * reading an individual file are less so. */
33 class IOException : public std::exception {
37 explicit IOException(const std::string &err) { error = err; }
38 virtual ~IOException() throw () { }
39 std::string getError() const { return error; }
42 /* Simplified TAR header--we only need to store regular files, don't need to
43 * handle long filenames, etc. */
44 static const int TAR_BLOCK_SIZE = 512;
66 /* A simple wrapper around a single TAR file to represent a segment. Objects
67 * may only be written out all at once, since the tar header must be written
68 * first; incremental writing is not supported. */
71 Tarfile(const std::string &path, const std::string &segment);
74 int spawn_filter(int fd_out);
75 void write_object(int id, const char *data, size_t len);
77 // Return an estimate of the size of the file.
78 size_t size_estimate();
80 void internal_write_object(const std::string &path,
81 const char *data, size_t len);
85 std::string segment_name;
88 int real_fd, filter_fd;
91 // Write data to the tar file
92 void tar_write(const char *data, size_t size);
95 class TarSegmentStore {
97 // New segments will be stored in the given directory.
98 TarSegmentStore(const std::string &path) { this->path = path; }
99 ~TarSegmentStore() { sync(); }
101 // Writes an object to segment in the store, and returns the name
102 // (segment/object) to refer to it. The optional parameter group can be
103 // used to control object placement; objects with different group
104 // parameters are kept in separate segments.
105 ObjectReference write_object(const char *data, size_t len,
106 const std::string &group = "");
108 // Ensure all segments have been fully written.
111 // Dump statistics to stdout about how much data has been written
115 struct segment_info {
117 std::string name; // UUID
118 int count; // Objects written to this segment
122 std::map<std::string, struct segment_info *> segments;
124 // Ensure that all segments in the given group have been fully written.
125 void close_segment(const std::string &group);
127 // Parse an object reference string and return just the segment name
129 std::string object_reference_to_segment(const std::string &object);
132 /* An in-memory representation of an object, which can be incrementally built
133 * before it is written out to a segment. */
139 // If an object is placed in a group, it will be written out to segments
140 // only containing other objects in the same group. A group name is simply
142 //std::string get_group() const { return group; }
143 void set_group(const std::string &g) { group = g; }
145 // Data in an object must be written all at once, and cannot be generated
146 // incrementally. Data can be an arbitrary block of binary data of any
147 // size. The pointer to the data need only remain valid until write() is
149 void set_data(const char *d, size_t len) { data = d; data_len = len; }
151 // Write an object to a segment, thus making it permanent. This function
152 // can be called at most once.
153 void write(TarSegmentStore *store);
155 // Compute the checksum of an object, and include it in the object
156 // reference. This should be called after write(), and the data specified
157 // by set_data() must remain valid through the call to checksum().
160 // An object is assigned a permanent name once it has been written to a
161 // segment. Until that time, its name cannot be determined.
162 std::string get_name() const { return ref.to_string(); }
163 ObjectReference get_ref() { return ref; }
174 /* Program through which segment data is piped before being written to file. */
175 extern const char *filter_program;
177 /* Extension which should be appended to segments written out (.tar is already
178 * included; this adds to it) */
179 extern const char *filter_extension;
181 #endif // _LBS_STORE_H