X-Git-Url: http://git.vrable.net/?p=cumulus.git;a=blobdiff_plain;f=main.cc;h=1335bea5ac7e23c1f88ad72c4c6b170aa6560dc3;hp=6b9f457eae60f3a29f268071981e134f683c7545;hb=HEAD;hpb=36e15463221ffb8f2e1dc9903705d7be81d1d1f8 diff --git a/main.cc b/main.cc index 6b9f457..1335bea 100644 --- a/main.cc +++ b/main.cc @@ -1,8 +1,6 @@ -/* Cumulus: Smart Filesystem Backup to Dumb Servers - * - * Copyright (C) 2006-2009 The Regents of the University of California - * Copyright (C) 2012 Google Inc. - * Written by Michael Vrable +/* Cumulus: Efficient Filesystem Backup to the Cloud + * Copyright (C) 2006-2009, 2012 The Cumulus Developers + * See the AUTHORS file for a list of contributors. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -43,25 +41,28 @@ #include #include #include +#include #include #include #include #include #include "exclude.h" +#include "hash.h" #include "localdb.h" #include "metadata.h" #include "remote.h" #include "store.h" -#include "sha1.h" #include "subfile.h" #include "util.h" +#include "third_party/sha1.h" using std::list; using std::map; using std::string; using std::vector; using std::ostream; +using std::unique_ptr; /* Version information. This will be filled in by the Makefile. */ #ifndef CUMULUS_VERSION @@ -83,14 +84,6 @@ static char *block_buf; * invocations to help in creating incremental snapshots. */ LocalDb *db; -/* 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. */ PathFilterList filter_rules; @@ -99,13 +92,6 @@ bool flag_rebuild_statcache = false; /* Whether verbose output is enabled. */ bool verbose = false; -/* Ensure that the given segment is listed as a dependency of the current - * snapshot. */ -void add_segment(const string& segment) -{ - segment_list.insert(segment); -} - /* Attempts to open a regular file read-only, but with safety checks for files * that might not be fully trusted. */ int safe_open(const string& path, struct stat *stat_buf) @@ -233,8 +219,6 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, i != old_blocks.end(); ++i) { const ObjectReference &ref = *i; object_list.push_back(ref.to_string()); - if (ref.is_normal()) - add_segment(ref.get_segment()); db->UseObject(ref); } size = stat_buf.st_size; @@ -244,7 +228,7 @@ 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) { - SHA1Checksum hash; + unique_ptr file_hash(Hash::New()); Subfile subfile(db); subfile.load_old_blocks(old_blocks); @@ -258,7 +242,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, break; } - hash.process(block_buf, bytes); + file_hash->update(block_buf, bytes); // Sparse file processing: if we read a block of all zeroes, encode // that explicitly. @@ -275,9 +259,9 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, double block_age = 0.0; ObjectReference ref; - SHA1Checksum block_hash; - block_hash.process(block_buf, bytes); - string block_csum = block_hash.checksum_str(); + unique_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); @@ -309,9 +293,8 @@ 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"; @@ -332,9 +315,13 @@ 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()); - if (ref.is_normal()) - add_segment(ref.get_segment()); db->UseObject(ref); } size += bytes; @@ -343,25 +330,22 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, status = "old"; } - file_info["checksum"] = hash.checksum_str(); + 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) @@ -541,6 +525,7 @@ void try_merge_filter(const string& path, const string& basedir) * one block (1 MB) worth of data. If the file doesn't seems like it might * be larger than that, don't parse the rules in it. */ ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); + close(fd); if (bytes < 0 || bytes >= static_cast(LBS_BLOCK_SIZE - 1)) { /* TODO: Add more strict resource limits on merge files? */ fprintf(stderr, @@ -585,7 +570,8 @@ void scanfile(const string& path) DIR *dir = opendir(path.c_str()); if (dir == NULL) { - fprintf(stderr, "Error: %m\n"); + fprintf(stderr, "Error reading directory %s: %m\n", + path.c_str()); return; } @@ -668,8 +654,7 @@ void usage(const char *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" + " --intent=FLOAT DEPRECATED: ignored, and will be removed soon\n" " --full-metadata do not re-use metadata from previous backups\n" " --rebuild-statcache re-read all file data to verify statcache\n" " -v --verbose list files as they are backed up\n" @@ -681,6 +666,8 @@ void usage(const char *program) int main(int argc, char *argv[]) { + hash_init(); + string backup_dest = "", backup_script = ""; string localdb_dir = ""; string backup_scheme = ""; @@ -698,7 +685,7 @@ int main(int argc, char *argv[]) {"dest", 1, 0, 0}, // 3 {"scheme", 1, 0, 0}, // 4 {"signature-filter", 1, 0, 0}, // 5 - {"intent", 1, 0, 0}, // 6 + {"intent", 1, 0, 0}, // 6, DEPRECATED {"full-metadata", 0, 0, 0}, // 7 {"tmpdir", 1, 0, 0}, // 8 {"upload-script", 1, 0, 0}, // 9 @@ -738,9 +725,9 @@ int main(int argc, char *argv[]) signature_filter = optarg; break; case 6: // --intent - snapshot_intent = atof(optarg); - if (snapshot_intent <= 0) - snapshot_intent = 1; + fprintf(stderr, + "Warning: The --intent= option is deprecated and will " + "be removed in the future.\n"); break; case 7: // --full-metadata flag_full_metadata = true; @@ -815,7 +802,7 @@ int main(int argc, char *argv[]) * a temporary directory for staging files. Otherwise, write backups * directly to the destination directory. */ if (backup_script != "") { - tmp_dir = tmp_dir + "/lbs." + generate_uuid(); + tmp_dir = tmp_dir + "/cumulus." + generate_uuid(); if (mkdir(tmp_dir.c_str(), 0700) < 0) { fprintf(stderr, "Cannot create temporary directory %s: %m\n", tmp_dir.c_str()); @@ -829,25 +816,21 @@ int main(int argc, char *argv[]) /* Store the time when the backup started, so it can be included in the * snapshot name. */ time_t now; - struct tm time_buf_local, time_buf_utc; - char desc_buf[256]; time(&now); - localtime_r(&now, &time_buf_local); - gmtime_r(&now, &time_buf_utc); - strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf_utc); + string timestamp + = TimeFormat::format(now, TimeFormat::FORMAT_FILENAME, true); /* Open the local database which tracks all objects that are stored * remotely, for efficient incrementals. Provide it with the name of this * snapshot. */ string database_path = localdb_dir + "/localdb.sqlite"; db = new LocalDb; - db->Open(database_path.c_str(), desc_buf, backup_scheme.c_str(), - snapshot_intent); + db->Open(database_path.c_str(), timestamp.c_str(), backup_scheme.c_str()); tss = new TarSegmentStore(remote, db); /* Initialize the stat cache, for skipping over unchanged files. */ - metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf, + metawriter = new MetadataWriter(tss, localdb_dir.c_str(), timestamp.c_str(), backup_scheme.c_str()); for (int i = optind; i < argc; i++) { @@ -855,7 +838,6 @@ int main(int argc, char *argv[]) } ObjectReference root_ref = metawriter->close(); - add_segment(root_ref.get_segment()); string backup_root = root_ref.to_string(); delete metawriter; @@ -864,47 +846,43 @@ 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-"; + /* Write out a summary file with metadata for all the segments in this + * 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) - checksum_filename += backup_scheme + "-"; - checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums"; - RemoteFile *checksum_file = remote->alloc_file(checksum_filename, - "checksums"); - FILE *checksums = fdopen(checksum_file->get_fd(), "w"); + dbmeta_filename += backup_scheme + "-"; + dbmeta_filename += timestamp + ".meta" + filter_extension; + RemoteFile *dbmeta_file = remote->alloc_file(dbmeta_filename, "meta"); + unique_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) { - 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; + map segment_metadata = db->GetSegmentMetadata(*i); + if (segment_metadata.size() > 0) { + map::const_iterator j; + for (j = segment_metadata.begin(); + j != segment_metadata.end(); ++j) + { + fprintf(dbmeta, "%s: %s\n", + j->first.c_str(), j->second.c_str()); } - - if (raw_checksum != NULL) - fprintf(checksums, "%s *%s\n", - raw_checksum, seg_path.c_str()); + fprintf(dbmeta, "\n"); } } - fclose(checksums); + fclose(dbmeta); + dbmeta_filter->wait(); - 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(); + string dbmeta_csum + = Hash::hash_file(dbmeta_file->get_local_path().c_str()); + dbmeta_file->send(); db->Close(); @@ -921,36 +899,29 @@ int main(int argc, char *argv[]) string desc_filename = "snapshot-"; if (backup_scheme.size() > 0) desc_filename += backup_scheme + "-"; - desc_filename = desc_filename + desc_buf + ".lbs"; + desc_filename = desc_filename + timestamp + ".cumulus"; RemoteFile *descriptor_file = remote->alloc_file(desc_filename, "snapshots"); - int descriptor_fd = descriptor_file->get_fd(); - if (descriptor_fd < 0) { + unique_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); - strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", - &time_buf_local); - fprintf(descriptor, "Date: %s\n", desc_buf); + string timestamp_local + = TimeFormat::format(now, TimeFormat::FORMAT_LOCALTIME, false); + fprintf(descriptor, "Date: %s\n", timestamp_local.c_str()); 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()); - if (csum.size() > 0) { - fprintf(descriptor, "Checksums: %s\n", csum.c_str()); + if (dbmeta_csum.size() > 0) { + fprintf(descriptor, "Segment-metadata: %s\n", dbmeta_csum.c_str()); } fprintf(descriptor, "Segments:\n"); @@ -960,14 +931,8 @@ 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();