X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=main.cc;h=0d02727bb757406eebc9994926ac5c7500f23ad8;hb=5a6aa5c252a70554a6987bfbb721ef7d722e1e20;hp=4c8f2bbb40093ab28acc698c7c0ee7629dd51e4d;hpb=7efae40a865fce46b74538745b17901785062e5f;p=cumulus.git diff --git a/main.cc b/main.cc index 4c8f2bb..0d02727 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); @@ -314,6 +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); } @@ -323,26 +329,22 @@ 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 - // like it hasn't changed, then the newly-computed checksum should match - // the checksum in the statcache. If not, we have possible disk corruption - // and report a warning. - if (flag_rebuild_statcache) { - if (found - && metawriter->is_unchanged(&stat_buf) - && file_info["checksum"] != metawriter->get_checksum()) { - fprintf(stderr, - "Warning: Checksum for %s does not match expected value\n" - " expected: %s\n" - " actual: %s\n", - path.c_str(), - metawriter->get_checksum().c_str(), - file_info["checksum"].c_str()); - } + // Sanity check: if the file looks like it hasn't changed, then the + // newly-computed checksum should match the checksum in the statcache. If + // not, we have possible disk corruption and report a warning. + if (found + && metawriter->is_unchanged(&stat_buf) + && file_info["checksum"] != metawriter->get_checksum()) { + fprintf(stderr, + "Warning: Checksum for %s does not match expected value\n" + " expected: %s\n" + " actual: %s\n", + path.c_str(), + metawriter->get_checksum().c_str(), + file_info["checksum"].c_str()); } if (verbose && status != NULL) @@ -843,69 +845,23 @@ 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" + filter_extension; RemoteFile *dbmeta_file = remote->alloc_file(dbmeta_filename, "meta"); - FileFilter *dbmeta_filter = FileFilter::New(dbmeta_file->get_fd(), - filter_program); + 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); @@ -921,6 +877,7 @@ int main(int argc, char *argv[]) } } fclose(dbmeta); + dbmeta_filter->wait(); string dbmeta_csum = Hash::hash_file(dbmeta_file->get_local_path().c_str()); @@ -945,8 +902,8 @@ int main(int argc, char *argv[]) RemoteFile *descriptor_file = remote->alloc_file(desc_filename, "snapshots"); - FileFilter *descriptor_filter = FileFilter::New(descriptor_file->get_fd(), - signature_filter.c_str()); + 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; @@ -963,11 +920,7 @@ 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");