X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=eb05243842668f5479c2dfffe005fadad88d94a6;hb=f1f817df56ce99ea544569635698df55d72bd1ee;hp=e4fe21fe121343c8135fe23cfedf9d858a79ef89;hpb=a855c59ca566ee7a55fc3dda56ab70943a85ef32;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index e4fe21f..eb05243 100644 --- a/scandir.cc +++ b/scandir.cc @@ -25,9 +25,10 @@ #include #include "localdb.h" +#include "metadata.h" +#include "remote.h" #include "store.h" #include "sha1.h" -#include "statcache.h" #include "util.h" using std::list; @@ -43,32 +44,26 @@ using std::ostream; #define LBS_STRINGIFY2(s) #s static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); +static RemoteStore *remote = NULL; 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 @@ -78,28 +73,11 @@ 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 +121,9 @@ 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. */ @@ -161,12 +139,13 @@ 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; @@ -189,16 +168,34 @@ 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 { + 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; @@ -233,11 +230,13 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, o->write(tss); ref = o->get_ref(); db->StoreObject(ref, block_csum, bytes, block_age); + ref.set_range(0, bytes); delete o; } 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; @@ -251,35 +250,14 @@ 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; } @@ -298,12 +276,20 @@ 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) + ")"; @@ -316,11 +302,12 @@ 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) { @@ -348,7 +335,7 @@ 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"); } @@ -356,7 +343,7 @@ 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); @@ -367,6 +354,7 @@ 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; @@ -381,13 +369,7 @@ 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) @@ -573,9 +555,13 @@ void usage(const char *program) "Produce backup snapshot of files in SOURCE and store to DEST.\n" "\n" "Options:\n" - " --dest=PATH path where backup is to be written [REQUIRED]\n" + " --dest=PATH path where backup is to be written\n" + " --upload-script=COMMAND\n" + " program to invoke for each backup file generated\n" " --exclude=PATH exclude files in PATH from snapshot\n" " --localdb=PATH local backup metadata is stored in PATH\n" + " --tmpdir=PATH path for temporarily storing backup files\n" + " (defaults to TMPDIR environment variable or /tmp)\n" " --filter=COMMAND program through which to filter segment data\n" " (defaults to \"bzip2 -c\")\n" " --filter-extension=EXT\n" @@ -583,18 +569,27 @@ void usage(const char *program) " (defaults to \".bz2\")\n" " --signature-filter=COMMAND\n" " program though which to filter descriptor\n" - " --scheme=NAME optional name for this snapshot\n", + " --scheme=NAME optional name for this snapshot\n" + " --intent=FLOAT intended backup type: 1=daily, 7=weekly, ...\n" + " (defaults to \"1\")\n" + " --full-metadata do not re-use metadata from previous backups\n" + "\n" + "Exactly one of --dest or --upload-script must be specified.\n", lbs_version, program ); } int main(int argc, char *argv[]) { - string backup_dest = ""; + string backup_dest = "", backup_script = ""; string localdb_dir = ""; string backup_scheme = ""; string signature_filter = ""; + string tmp_dir = "/tmp"; + if (getenv("TMPDIR") != NULL) + tmp_dir = getenv("TMPDIR"); + while (1) { static struct option long_options[] = { {"localdb", 1, 0, 0}, // 0 @@ -604,6 +599,10 @@ int main(int argc, char *argv[]) {"dest", 1, 0, 0}, // 4 {"scheme", 1, 0, 0}, // 5 {"signature-filter", 1, 0, 0}, // 6 + {"intent", 1, 0, 0}, // 7 + {"full-metadata", 0, 0, 0}, // 8 + {"tmpdir", 1, 0, 0}, // 9 + {"upload-script", 1, 0, 0}, // 10 {NULL, 0, 0, 0}, }; @@ -639,6 +638,20 @@ int main(int argc, char *argv[]) case 6: // --signature-filter signature_filter = optarg; break; + case 7: // --intent + snapshot_intent = atof(optarg); + if (snapshot_intent <= 0) + snapshot_intent = 1; + break; + case 8: // --full-metadata + flag_full_metadata = true; + break; + case 9: // --tmpdir + tmp_dir = optarg; + break; + case 10: // --upload-script + backup_script = optarg; + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; @@ -658,9 +671,16 @@ int main(int argc, char *argv[]) for (int i = optind; i < argc; i++) add_include(argv[i]); - if (backup_dest == "") { + if (backup_dest == "" && backup_script == "") { fprintf(stderr, - "Error: Backup destination must be specified with --dest=\n"); + "Error: Backup destination must be specified using --dest= or --upload-script=\n"); + usage(argv[0]); + return 1; + } + + if (backup_dest != "" && backup_script != "") { + fprintf(stderr, + "Error: Cannot specify both --dest= and --upload-script=\n"); usage(argv[0]); return 1; } @@ -669,6 +689,12 @@ int main(int argc, char *argv[]) if (localdb_dir == "") { localdb_dir = backup_dest; } + if (localdb_dir == "") { + fprintf(stderr, + "Error: Must specify local database path with --localdb=\n"); + usage(argv[0]); + return 1; + } // Dump paths for debugging/informational purposes { @@ -676,8 +702,8 @@ int main(int argc, char *argv[]) printf("LBS Version: %s\n", lbs_version); - printf("--dest=%s\n--localdb=%s\n\n", - backup_dest.c_str(), localdb_dir.c_str()); + printf("--dest=%s\n--localdb=%s\n--upload-script=%s\n", + backup_dest.c_str(), localdb_dir.c_str(), backup_script.c_str()); printf("Includes:\n"); for (i = includes.begin(); i != includes.end(); ++i) @@ -694,6 +720,22 @@ int main(int argc, char *argv[]) block_buf = new char[LBS_BLOCK_SIZE]; + /* Initialize the remote storage layer. If using an upload script, create + * a temporary directory for staging files. Otherwise, write backups + * directly to the destination directory. */ + if (backup_script != "") { + tmp_dir = tmp_dir + "/lbs." + generate_uuid(); + if (mkdir(tmp_dir.c_str(), 0700) < 0) { + fprintf(stderr, "Cannot create temporary directory %s: %m\n", + tmp_dir.c_str()); + return 1; + } + remote = new RemoteStore(tmp_dir); + remote->set_script(backup_script); + } else { + remote = new RemoteStore(backup_dest); + } + /* Store the time when the backup started, so it can be included in the * snapshot name. */ time_t now; @@ -709,32 +751,24 @@ 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); + tss = new TarSegmentStore(remote, 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()); + ObjectReference root_ref = metawriter->close(); + add_segment(root_ref.get_segment()); + string backup_root = root_ref.to_string(); - string backup_root = root->get_ref().to_string(); - delete root; - - statcache->Close(); - delete statcache; + delete metawriter; tss->sync(); tss->dump_stats(); @@ -744,48 +778,56 @@ int main(int argc, char *argv[]) * segments included in this snapshot. The format is designed so that it * may be easily verified using the sha1sums command. */ const char csum_type[] = "sha1"; - string checksum_filename = backup_dest + "/snapshot-"; + string checksum_filename = "snapshot-"; if (backup_scheme.size() > 0) checksum_filename += backup_scheme + "-"; checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums"; - FILE *checksums = fopen(checksum_filename.c_str(), "w"); - if (checksums != NULL) { - for (std::set::iterator i = segment_list.begin(); - i != segment_list.end(); ++i) { - string seg_path, seg_csum; - if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) { - const char *raw_checksum = NULL; - if (strncmp(seg_csum.c_str(), csum_type, - strlen(csum_type)) == 0) { - raw_checksum = seg_csum.c_str() + strlen(csum_type); - if (*raw_checksum == '=') - raw_checksum++; - else - raw_checksum = NULL; - } + RemoteFile *checksum_file = remote->alloc_file(checksum_filename, + "checksums"); + FILE *checksums = fdopen(checksum_file->get_fd(), "w"); - if (raw_checksum != NULL) - fprintf(checksums, "%s *%s\n", - raw_checksum, seg_path.c_str()); + for (std::set::iterator i = segment_list.begin(); + i != segment_list.end(); ++i) { + string seg_path, seg_csum; + if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) { + const char *raw_checksum = NULL; + if (strncmp(seg_csum.c_str(), csum_type, + strlen(csum_type)) == 0) { + raw_checksum = seg_csum.c_str() + strlen(csum_type); + if (*raw_checksum == '=') + raw_checksum++; + else + raw_checksum = NULL; } + + if (raw_checksum != NULL) + fprintf(checksums, "%s *%s\n", + raw_checksum, seg_path.c_str()); } - fclose(checksums); - } else { - fprintf(stderr, "ERROR: Unable to write checksums file: %m\n"); } + fclose(checksums); + checksum_file->send(); db->Close(); + /* All other files should be flushed to remote storage before writing the + * backup descriptor below, so that it is not possible to have a backup + * descriptor written out depending on non-existent (not yet written) + * files. */ + remote->sync(); + /* 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. 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-"; + string desc_filename = "snapshot-"; if (backup_scheme.size() > 0) desc_filename += backup_scheme + "-"; desc_filename = desc_filename + desc_buf + ".lbs"; - int descriptor_fd = open(desc_filename.c_str(), O_WRONLY | O_CREAT, 0666); + RemoteFile *descriptor_file = remote->alloc_file(desc_filename, + "snapshots"); + int descriptor_fd = descriptor_file->get_fd(); if (descriptor_fd < 0) { fprintf(stderr, "Unable to open descriptor output file: %m\n"); return 1; @@ -799,12 +841,13 @@ 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); if (backup_scheme.size() > 0) fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str()); + fprintf(descriptor, "Backup-Intent: %g\n", snapshot_intent); fprintf(descriptor, "Root: %s\n", backup_root.c_str()); SHA1Checksum checksum_csum; @@ -830,5 +873,18 @@ int main(int argc, char *argv[]) } } + descriptor_file->send(); + + remote->sync(); + delete remote; + + if (backup_script != "") { + if (rmdir(tmp_dir.c_str()) < 0) { + fprintf(stderr, + "Warning: Cannot delete temporary directory %s: %m\n", + tmp_dir.c_str()); + } + } + return 0; }