X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=store.h;h=a959cce9107f6666a5953e5a3259f69b182d54df;hb=a4cf5f4d8df46fa00992a210d587cd824cedcb08;hp=6acee7fb4b0fce8707eb3aa02a095a0cfabe9ead;hpb=d5afe2c08724b4b263a76f187e39861c1e6e0cf5;p=cumulus.git diff --git a/store.h b/store.h index 6acee7f..a959cce 100644 --- a/store.h +++ b/store.h @@ -3,14 +3,12 @@ * * Backup data is stored in a collection of objects, which are grouped together * 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. */ + * represents segments as TAR files and objects as files within them. */ -#ifndef _LBS_TARSTORE_H -#define _LBS_TARSTORE_H +#ifndef _LBS_STORE_H +#define _LBS_STORE_H #include -#include #include #include @@ -19,7 +17,11 @@ #include #include +#include "localdb.h" #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. */ @@ -38,6 +40,30 @@ public: std::string getError() const { return error; } }; +/* Simplified TAR header--we only need to store regular files, don't need to + * handle long filenames, etc. */ +static const int TAR_BLOCK_SIZE = 512; + +struct tar_header +{ + char name[100]; + char mode[8]; + char uid[8]; + char gid[8]; + char size[12]; + char mtime[12]; + char chksum[8]; + char typeflag; + char linkname[100]; + char magic[8]; + char uname[32]; + char gname[32]; + char devmajor[8]; + char devminor[8]; + char prefix[155]; + char padding[12]; +}; + /* 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. */ @@ -46,10 +72,11 @@ public: Tarfile(const std::string &path, const std::string &segment); ~Tarfile(); + int spawn_filter(int fd_out); void write_object(int id, const char *data, size_t len); // Return an estimate of the size of the file. - size_t size_estimate() { return size; } + size_t size_estimate(); void internal_write_object(const std::string &path, const char *data, size_t len); @@ -57,41 +84,48 @@ public: private: size_t size; std::string segment_name; - std::ostringstream checksums; - TAR *t; + + /* Filter support. */ + int real_fd, filter_fd; + pid_t filter_pid; + + // Write data to the tar file + void tar_write(const char *data, size_t size); }; class TarSegmentStore { public: // New segments will be stored in the given directory. - TarSegmentStore(const std::string &path) { this->path = path; } + TarSegmentStore(const std::string &path, + LocalDb *db = NULL) + { this->path = path; this->db = db; } ~TarSegmentStore() { sync(); } // 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. - std::string write_object(const char *data, size_t len, - const std::string &group = "", - const std::list &refs = norefs); + ObjectReference write_object(const char *data, size_t len, + const std::string &group = ""); // Ensure all segments have been fully written. void sync(); + // Dump statistics to stdout about how much data has been written + void dump_stats(); + private: struct segment_info { Tarfile *file; std::string name; // UUID - std::set refs; // Other segments this one refers to int count; // Objects written to this segment + std::string basename; // Name of segment without directory + std::string fullname; // Full path to stored segment }; std::string path; std::map segments; - - // An empty list which can be used as an argument to write_object to - // indicate that this object depends on no others. - static std::list norefs; + LocalDb *db; // Ensure that all segments in the given group have been fully written. void close_segment(const std::string &group); @@ -101,4 +135,53 @@ private: std::string object_reference_to_segment(const std::string &object); }; -#endif // _LBS_TARSTORE_H +/* An in-memory representation of an object, which can be incrementally built + * before it is written out to a segment. */ +class LbsObject { +public: + 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: + std::string group; + const char *data; + size_t data_len; + + bool written; + ObjectReference ref; +}; + +/* Program through which segment data is piped before being written to file. */ +extern const char *filter_program; + +/* Extension which should be appended to segments written out (.tar is already + * included; this adds to it) */ +extern const char *filter_extension; + +#endif // _LBS_STORE_H