X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=f653f4335d9ab69e7fff9c66a49669844fd39670;hb=cc2d3611ed50f5965a9138ffaf3262417993c4f8;hp=d91701f09633d8a21cd30bbbb174f154c140bb77;hpb=d3519c88bf18df470a2cc35286448181e1544cbb;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index d91701f..f653f43 100644 --- a/scandir.cc +++ b/scandir.cc @@ -12,6 +12,7 @@ #include #include #include +#include #include #include @@ -24,9 +25,9 @@ #include #include "localdb.h" +#include "metadata.h" #include "store.h" #include "sha1.h" -#include "statcache.h" #include "util.h" using std::list; @@ -43,31 +44,24 @@ 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; +/* Snapshot intent: 1=daily, 7=weekly, etc. This is not used directly, but is + * stored in the local database and can help guide segment cleaning and + * snapshot expiration policies. */ +double snapshot_intent = 1.0; + /* Selection of files to include/exclude in the snapshot. */ std::list includes; // Paths in which files should be saved std::list excludes; // Paths which will not be saved @@ -77,28 +71,11 @@ 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 @@ -142,9 +119,9 @@ 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->find(path) && 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. */ @@ -160,12 +137,13 @@ 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; object_list.push_back(ref.to_string()); - segment_list.insert(ref.get_segment()); + if (ref.is_normal()) + add_segment(ref.get_segment()); db->UseObject(ref); } size = stat_buf.st_size; @@ -188,17 +166,36 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, hash.process(block_buf, bytes); + // Sparse file processing: if we read a block of all zeroes, encode + // that explicitly. + bool all_zero = true; + for (int i = 0; i < bytes; i++) { + if (block_buf[i] != 0) { + all_zero = false; + break; + } + } + // Either find a copy of this block in an already-existing segment, // or index it so it can be re-used in the future double block_age = 0.0; + ObjectReference ref; + SHA1Checksum block_hash; block_hash.process(block_buf, bytes); string block_csum = block_hash.checksum_str(); - ObjectReference ref = db->FindObject(block_csum, bytes); + + if (all_zero) { + ref = ObjectReference(ObjectReference::REF_ZERO); + ref.set_range(0, bytes); + } else { + ObjectReference ref = db->FindObject(block_csum, bytes); + } // Store a copy of the object if one does not yet exist - if (ref.get_segment().size() == 0) { + if (ref.is_null()) { 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 @@ -211,8 +208,15 @@ 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 { @@ -228,7 +232,8 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, } object_list.push_back(ref.to_string()); - segment_list.insert(ref.get_segment()); + if (ref.is_normal()) + add_segment(ref.get_segment()); db->UseObject(ref); size += bytes; @@ -242,35 +247,14 @@ 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; } @@ -289,12 +273,20 @@ void dump_inode(const string& path, // Path within snapshot ssize_t len; printf("%s\n", path.c_str()); + metawriter->find(path); + file_info["name"] = 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); + time_t now = time(NULL); + if (now - stat_buf.st_ctime < 30 || now - stat_buf.st_mtime < 30) + if ((stat_buf.st_mode & S_IFMT) != S_IFDIR) + file_info["volatile"] = "1"; + struct passwd *pwd = getpwuid(stat_buf.st_uid); if (pwd != NULL) { file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")"; @@ -307,11 +299,12 @@ 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) { @@ -339,7 +332,7 @@ 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"); } @@ -347,7 +340,7 @@ 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); @@ -358,6 +351,7 @@ void dump_inode(const string& path, // Path within snapshot if (file_size != stat_buf.st_size) { fprintf(stderr, "Warning: Size of %s changed during reading\n", path.c_str()); + file_info["volatile"] = "1"; } break; @@ -372,13 +366,7 @@ 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) @@ -559,6 +547,7 @@ void usage(const char *program) { fprintf( stderr, + "LBS %s\n\n" "Usage: %s [OPTION]... --dest=DEST PATHS...\n" "Produce backup snapshot of files in SOURCE and store to DEST.\n" "\n" @@ -571,8 +560,12 @@ void usage(const char *program) " --filter-extension=EXT\n" " string to append to segment files\n" " (defaults to \".bz2\")\n" - " --scheme=NAME optional name for this snapshot\n", - program + " --signature-filter=COMMAND\n" + " program though which to filter descriptor\n" + " --scheme=NAME optional name for this snapshot\n" + " --intent=FLOAT intended backup type: 1=daily, 7=weekly, ...\n" + " (defaults to \"1\")\n", + lbs_version, program ); } @@ -581,6 +574,7 @@ int main(int argc, char *argv[]) string backup_dest = ""; string localdb_dir = ""; string backup_scheme = ""; + string signature_filter = ""; while (1) { static struct option long_options[] = { @@ -590,6 +584,8 @@ int main(int argc, char *argv[]) {"filter-extension", 1, 0, 0}, // 3 {"dest", 1, 0, 0}, // 4 {"scheme", 1, 0, 0}, // 5 + {"signature-filter", 1, 0, 0}, // 6 + {"intent", 1, 0, 0}, // 7 {NULL, 0, 0, 0}, }; @@ -622,6 +618,14 @@ int main(int argc, char *argv[]) case 5: // --scheme backup_scheme = optarg; break; + case 6: // --signature-filter + signature_filter = optarg; + break; + case 7: // --intent + snapshot_intent = atof(optarg); + if (snapshot_intent <= 0) + snapshot_intent = 1; + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; @@ -692,32 +696,24 @@ int main(int argc, char *argv[]) string database_path = localdb_dir + "/localdb.sqlite"; db = new LocalDb; db->Open(database_path.c_str(), desc_buf, - backup_scheme.size() ? backup_scheme.c_str() : NULL); + backup_scheme.size() ? backup_scheme.c_str() : NULL, + snapshot_intent); 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(); @@ -761,30 +757,56 @@ int main(int argc, char *argv[]) /* Write a backup descriptor file, which says which segments are needed and * where to start to restore this snapshot. The filename is based on the - * current time. */ + * current time. If a signature filter program was specified, filter the + * data through that to give a chance to sign the descriptor contents. */ string desc_filename = backup_dest + "/snapshot-"; if (backup_scheme.size() > 0) desc_filename += backup_scheme + "-"; desc_filename = desc_filename + desc_buf + ".lbs"; - std::ofstream descriptor(desc_filename.c_str()); - descriptor << "Format: LBS Snapshot v0.2\n"; - descriptor << "Producer: LBS " << lbs_version << "\n"; + int descriptor_fd = open(desc_filename.c_str(), O_WRONLY | O_CREAT, 0666); + if (descriptor_fd < 0) { + fprintf(stderr, "Unable to open descriptor output file: %m\n"); + return 1; + } + pid_t signature_pid = 0; + if (signature_filter.size() > 0) { + int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(), + &signature_pid); + close(descriptor_fd); + descriptor_fd = new_fd; + } + FILE *descriptor = fdopen(descriptor_fd, "w"); + + 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); - descriptor << "Date: " << desc_buf << "\n"; + fprintf(descriptor, "Date: %s\n", desc_buf); if (backup_scheme.size() > 0) - descriptor << "Scheme: " << backup_scheme << "\n"; - descriptor << "Root: " << backup_root << "\n"; + fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str()); + fprintf(descriptor, "Root: %s\n", backup_root.c_str()); SHA1Checksum checksum_csum; if (checksum_csum.process_file(checksum_filename.c_str())) { - descriptor << "Checksums: " << checksum_csum.checksum_str() << "\n"; + string csum = checksum_csum.checksum_str(); + fprintf(descriptor, "Checksums: %s\n", csum.c_str()); } - descriptor << "Segments:\n"; + fprintf(descriptor, "Segments:\n"); for (std::set::iterator i = segment_list.begin(); i != segment_list.end(); ++i) { - descriptor << " " << *i << "\n"; + fprintf(descriptor, " %s\n", i->c_str()); + } + + fclose(descriptor); + + if (signature_pid) { + int status; + waitpid(signature_pid, &status, 0); + + if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { + throw IOException("Signature filter process error"); + } } return 0;