X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=main.cc;h=daad05806925ca084c4af36a181d9351d2c06ffd;hb=ae5a596823b8ca4860f4487f5e228e93f73bb050;hp=91d1cef68dbc2644be9aa0ef3699033809df1c12;hpb=262f66f165516f2ae3923b708d860a8711fc7c58;p=cumulus.git diff --git a/main.cc b/main.cc index 91d1cef..daad058 100644 --- a/main.cc +++ b/main.cc @@ -46,6 +46,7 @@ #include #include +#include "cumulus.h" #include "exclude.h" #include "hash.h" #include "localdb.h" @@ -226,7 +227,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, /* If the file is new or changed, we must read in the contents a block at a * time. */ if (!cached) { - Hash *hash = Hash::New(); + scoped_ptr file_hash(Hash::New()); Subfile subfile(db); subfile.load_old_blocks(old_blocks); @@ -240,7 +241,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, break; } - hash->update(block_buf, bytes); + file_hash->update(block_buf, bytes); // Sparse file processing: if we read a block of all zeroes, encode // that explicitly. @@ -257,10 +258,9 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, double block_age = 0.0; ObjectReference ref; - Hash *hash = Hash::New(); - hash->update(block_buf, bytes); - string block_csum = hash->digest_str(); - delete hash; + scoped_ptr block_hash(Hash::New()); + block_hash->update(block_buf, bytes); + string block_csum = block_hash->digest_str(); if (all_zero) { ref = ObjectReference(ObjectReference::REF_ZERO); @@ -292,9 +292,8 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, if (object_group == 0) { o->set_group("data"); } else { - char group[32]; - sprintf(group, "compacted-%d", object_group); - o->set_group(group); + o->set_group(string_printf("compacted-%d", + object_group)); } if (status == NULL) status = "partial"; @@ -315,6 +314,12 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, while (!refs.empty()) { ref = refs.front(); refs.pop_front(); + + // The file-level checksum guarantees integrity of the data. + // To reduce the metadata log size, do not include checksums on + // individual objects. + ref.clear_checksum(); + object_list.push_back(ref.to_string()); db->UseObject(ref); } @@ -324,8 +329,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, status = "old"; } - file_info["checksum"] = hash->digest_str(); - delete hash; + file_info["checksum"] = file_hash->digest_str(); } // Sanity check: if we are rebuilding the statcache, but the file looks @@ -844,64 +848,23 @@ int main(int argc, char *argv[]) tss->dump_stats(); delete tss; - /* Write out a checksums file which lists the checksums for all the - * 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 = "snapshot-"; - if (backup_scheme.size() > 0) - checksum_filename += backup_scheme + "-"; - checksum_filename - = checksum_filename + timestamp + "." + csum_type + "sums"; - RemoteFile *checksum_file = remote->alloc_file(checksum_filename, - "meta"); - FILE *checksums = fdopen(checksum_file->get_fd(), "w"); - - std::set segment_list = db->GetUsedSegments(); - for (std::set::iterator i = segment_list.begin(); - i != segment_list.end(); ++i) { - map segment_metadata = db->GetSegmentMetadata(*i); - if (segment_metadata.count("path") - && segment_metadata.count("checksum")) - { - string seg_path = segment_metadata["path"]; - string seg_csum = segment_metadata["checksum"]; - 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); - - SHA1Checksum checksum_csum; - string csum; - checksum_filename = checksum_file->get_local_path(); - if (checksum_csum.process_file(checksum_filename.c_str())) { - csum = checksum_csum.checksum_str(); - } - - checksum_file->send(); - /* Write out a summary file with metadata for all the segments in this - * snapshot (can be used to reconstruct database contents if needed). */ + * snapshot (can be used to reconstruct database contents if needed), and + * contains hash values for the segments for quick integrity checks. */ string dbmeta_filename = "snapshot-"; if (backup_scheme.size() > 0) dbmeta_filename += backup_scheme + "-"; - dbmeta_filename += timestamp + ".meta"; - RemoteFile *dbmeta_file = remote->alloc_file(dbmeta_filename, - "meta"); - FILE *dbmeta = fdopen(dbmeta_file->get_fd(), "w"); + dbmeta_filename += timestamp + ".meta" + filter_extension; + RemoteFile *dbmeta_file = remote->alloc_file(dbmeta_filename, "meta"); + scoped_ptr dbmeta_filter(FileFilter::New(dbmeta_file->get_fd(), + filter_program)); + if (dbmeta_filter == NULL) { + fprintf(stderr, "Unable to open descriptor output file: %m\n"); + return 1; + } + FILE *dbmeta = fdopen(dbmeta_filter->get_wrapped_fd(), "w"); + std::set segment_list = db->GetUsedSegments(); for (std::set::iterator i = segment_list.begin(); i != segment_list.end(); ++i) { map segment_metadata = db->GetSegmentMetadata(*i); @@ -917,6 +880,7 @@ int main(int argc, char *argv[]) } } fclose(dbmeta); + dbmeta_filter->wait(); string dbmeta_csum = Hash::hash_file(dbmeta_file->get_local_path().c_str()); @@ -941,19 +905,13 @@ int main(int argc, char *argv[]) RemoteFile *descriptor_file = remote->alloc_file(desc_filename, "snapshots"); - int descriptor_fd = descriptor_file->get_fd(); - if (descriptor_fd < 0) { + scoped_ptr descriptor_filter( + FileFilter::New(descriptor_file->get_fd(), signature_filter.c_str())); + if (descriptor_filter == NULL) { 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"); + FILE *descriptor = fdopen(descriptor_filter->get_wrapped_fd(), "w"); fprintf(descriptor, "Format: Cumulus Snapshot v0.11\n"); fprintf(descriptor, "Producer: Cumulus %s\n", cumulus_version); @@ -965,11 +923,7 @@ int main(int argc, char *argv[]) fprintf(descriptor, "Root: %s\n", backup_root.c_str()); if (dbmeta_csum.size() > 0) { - fprintf(descriptor, "Database-state: %s\n", dbmeta_csum.c_str()); - } - - if (csum.size() > 0) { - fprintf(descriptor, "Checksums: %s\n", csum.c_str()); + fprintf(descriptor, "Segment-metadata: %s\n", dbmeta_csum.c_str()); } fprintf(descriptor, "Segments:\n"); @@ -979,14 +933,8 @@ int main(int argc, char *argv[]) } fclose(descriptor); - - if (signature_pid) { - int status; - waitpid(signature_pid, &status, 0); - - if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { - fatal("Signature filter process error"); - } + if (descriptor_filter->wait() < 0) { + fatal("Signature filter process error"); } descriptor_file->send();