X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=d39029cd57197933eed2e8e57ed8a6bc23ca3899;hb=bae7cf459803abce603c8b3d3009153936378b42;hp=588a907e0337a774f58713a197bb43601e733507;hpb=edcae8181571b57b058cb2249cb84b01df73d52a;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index 588a907..d39029c 100644 --- a/scandir.cc +++ b/scandir.cc @@ -25,9 +25,9 @@ #include #include "localdb.h" +#include "metadata.h" #include "store.h" #include "sha1.h" -#include "statcache.h" #include "util.h" using std::list; @@ -44,28 +44,16 @@ using std::ostream; static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); static TarSegmentStore *tss = NULL; +static MetadataWriter *metawriter = NULL; /* Buffer for holding a single block of data read from a file. */ static const size_t LBS_BLOCK_SIZE = 1024 * 1024; static char *block_buf; -static const size_t LBS_METADATA_BLOCK_SIZE = 65536; - /* Local database, which tracks objects written in this and previous * invocations to help in creating incremental snapshots. */ LocalDb *db; -/* Stat cache, which stored data locally to speed the backup process by quickly - * skipping files which have not changed. */ -StatCache *statcache; - -/* Contents of the root object. This will contain a set of indirect links to - * the metadata objects. */ -std::ostringstream metadata_root; - -/* Buffer for building up metadata. */ -std::ostringstream metadata; - /* Keep track of all segments which are needed to reconstruct the snapshot. */ std::set segment_list; @@ -78,28 +66,11 @@ std::list searches; // Directories we don't want to save, but bool relative_paths = true; -/* Ensure contents of metadata are flushed to an object. */ -void metadata_flush() +/* Ensure that the given segment is listed as a dependency of the current + * snapshot. */ +void add_segment(const string& segment) { - string m = metadata.str(); - if (m.size() == 0) - return; - - /* Write current metadata information to a new object. */ - LbsObject *meta = new LbsObject; - meta->set_group("metadata"); - meta->set_data(m.data(), m.size()); - meta->write(tss); - meta->checksum(); - - /* Write a reference to this block in the root. */ - ObjectReference ref = meta->get_ref(); - metadata_root << "@" << ref.to_string() << "\n"; - segment_list.insert(ref.get_segment()); - - delete meta; - - metadata.str(""); + segment_list.insert(segment); } /* Read data from a file descriptor and return the amount of data read. A @@ -143,9 +114,9 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, * re-reading the entire contents. */ bool cached = false; - if (statcache->Find(path, &stat_buf)) { + if (metawriter->matched() && metawriter->is_unchanged(&stat_buf)) { cached = true; - const list &blocks = statcache->get_blocks(); + list blocks = metawriter->get_blocks(); /* If any of the blocks in the object have been expired, then we should * fall back to fully reading in the file. */ @@ -161,7 +132,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, /* If everything looks okay, use the cached information */ if (cached) { - file_info["checksum"] = statcache->get_checksum(); + file_info["checksum"] = metawriter->get_checksum(); for (list::const_iterator i = blocks.begin(); i != blocks.end(); ++i) { const ObjectReference &ref = *i; @@ -200,6 +171,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, // Store a copy of the object if one does not yet exist if (ref.get_segment().size() == 0) { LbsObject *o = new LbsObject; + int object_group; /* We might still have seen this checksum before, if the object * was stored at some time in the past, but we have decided to @@ -212,8 +184,15 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, * Additionally, keep track of the age of the data by looking * up the age of the block which was expired and using that * instead of the current time. */ - if (db->IsOldObject(block_csum, bytes, &block_age)) { - o->set_group("compacted"); + if (db->IsOldObject(block_csum, bytes, + &block_age, &object_group)) { + if (object_group == 0) { + o->set_group("data"); + } else { + char group[32]; + sprintf(group, "compacted-%d", object_group); + o->set_group(group); + } if (status == NULL) status = "partial"; } else { @@ -243,35 +222,14 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, if (status != NULL) printf(" [%s]\n", status); - statcache->Save(path, &stat_buf, file_info["checksum"], object_list); - - /* 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::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::iterator i = object_list.begin(); - i != object_list.end(); ++i) { - blocklist += *i + "\n"; - } - - LbsObject *i = new LbsObject; - i->set_group("metadata"); - 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; + string blocklist = ""; + for (list::iterator i = object_list.begin(); + i != object_list.end(); ++i) { + if (i != object_list.begin()) + blocklist += "\n "; + blocklist += *i; } + file_info["data"] = blocklist; return size; } @@ -291,7 +249,11 @@ void dump_inode(const string& path, // Path within snapshot printf("%s\n", path.c_str()); + metawriter->find(path); + + file_info["path"] = uri_encode(path); file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8); + file_info["ctime"] = encode_int(stat_buf.st_ctime); file_info["mtime"] = encode_int(stat_buf.st_mtime); file_info["user"] = encode_int(stat_buf.st_uid); file_info["group"] = encode_int(stat_buf.st_gid); @@ -308,11 +270,12 @@ void dump_inode(const string& path, // Path within snapshot if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) { file_info["links"] = encode_int(stat_buf.st_nlink); - file_info["inode"] = encode_int(major(stat_buf.st_dev)) - + "/" + encode_int(minor(stat_buf.st_dev)) - + "/" + encode_int(stat_buf.st_ino); } + file_info["inode"] = encode_int(major(stat_buf.st_dev)) + + "/" + encode_int(minor(stat_buf.st_dev)) + + "/" + encode_int(stat_buf.st_ino); + char inode_type; switch (stat_buf.st_mode & S_IFMT) { @@ -340,7 +303,7 @@ void dump_inode(const string& path, // Path within snapshot fprintf(stderr, "error reading symlink: %m\n"); } else if (len <= stat_buf.st_size) { buf[len] = '\0'; - file_info["contents"] = uri_encode(buf); + file_info["target"] = uri_encode(buf); } else if (len > stat_buf.st_size) { fprintf(stderr, "error reading symlink: name truncated\n"); } @@ -348,7 +311,7 @@ void dump_inode(const string& path, // Path within snapshot delete[] buf; break; case S_IFREG: - inode_type = '-'; + inode_type = 'f'; file_size = dumpfile(fd, file_info, path, stat_buf); file_info["size"] = encode_int(file_size); @@ -373,13 +336,7 @@ void dump_inode(const string& path, // Path within snapshot file_info["type"] = string(1, inode_type); - metadata << "name: " << uri_encode(path) << "\n"; - dict_output(metadata, file_info); - metadata << "\n"; - - // Break apart metadata listing if it becomes too large. - if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE) - metadata_flush(); + metawriter->add(file_info); } void scanfile(const string& path, bool include) @@ -706,27 +663,18 @@ int main(int argc, char *argv[]) tss = new TarSegmentStore(backup_dest, db); /* Initialize the stat cache, for skipping over unchanged files. */ - statcache = new StatCache; - statcache->Open(localdb_dir.c_str(), desc_buf, - backup_scheme.size() ? backup_scheme.c_str() : NULL); + metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf, + backup_scheme.size() + ? backup_scheme.c_str() + : NULL); scanfile(".", false); - metadata_flush(); - const string md = metadata_root.str(); - - LbsObject *root = new LbsObject; - root->set_group("metadata"); - root->set_data(md.data(), md.size()); - root->write(tss); - root->checksum(); - segment_list.insert(root->get_ref().get_segment()); - - string backup_root = root->get_ref().to_string(); - delete root; + ObjectReference root_ref = metawriter->close(); + add_segment(root_ref.get_segment()); + string backup_root = root_ref.to_string(); - statcache->Close(); - delete statcache; + delete metawriter; tss->sync(); tss->dump_stats(); @@ -791,7 +739,7 @@ int main(int argc, char *argv[]) } FILE *descriptor = fdopen(descriptor_fd, "w"); - fprintf(descriptor, "Format: LBS Snapshot v0.2\n"); + fprintf(descriptor, "Format: LBS Snapshot v0.6\n"); fprintf(descriptor, "Producer: LBS %s\n", lbs_version); strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf); fprintf(descriptor, "Date: %s\n", desc_buf);