X-Git-Url: http://git.vrable.net/?p=cumulus.git;a=blobdiff_plain;f=scandir.cc;h=8af3175b64c3b6007a5497f4c4ac8d81e5d9ff7c;hp=f5ab2b6011274adc2e621f220107a6dfa996dee0;hb=f38dd9bcb0caffd3fc9126b05788c936690e8288;hpb=87c6a5ed4ec515c2e154f98529c1d2f077d2a7e4 diff --git a/scandir.cc b/scandir.cc index f5ab2b6..8af3175 100644 --- a/scandir.cc +++ b/scandir.cc @@ -1,4 +1,25 @@ -/* Recursively descend the filesystem and visit each file. */ +/* Cumulus: Smart Filesystem Backup to Dumb Servers + * + * Copyright (C) 2006-2008 The Regents of the University of California + * Written by Michael Vrable + * + * 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 + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +/* Main entry point for LBS. Contains logic for traversing the filesystem and + * constructing a backup. */ #include #include @@ -9,9 +30,11 @@ #include #include #include +#include #include #include #include +#include #include #include @@ -24,9 +47,11 @@ #include #include "localdb.h" +#include "metadata.h" +#include "remote.h" #include "store.h" #include "sha1.h" -#include "statcache.h" +#include "subfile.h" #include "util.h" using std::list; @@ -35,70 +60,53 @@ using std::vector; using std::ostream; /* Version information. This will be filled in by the Makefile. */ -#ifndef LBS_VERSION -#define LBS_VERSION Unknown +#ifndef CUMULUS_VERSION +#define CUMULUS_VERSION Unknown #endif -#define LBS_STRINGIFY(s) LBS_STRINGIFY2(s) -#define LBS_STRINGIFY2(s) #s -static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); +#define CUMULUS_STRINGIFY(s) CUMULUS_STRINGIFY2(s) +#define CUMULUS_STRINGIFY2(s) #s +static const char cumulus_version[] = CUMULUS_STRINGIFY(CUMULUS_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 +std::list excluded_names; // Directories which will not be saved std::list searches; // Directories we don't want to save, but // do want to descend searching for data // in included paths bool relative_paths = true; -/* Ensure contents of metadata are flushed to an object. */ -void metadata_flush() -{ - 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(); +bool flag_rebuild_statcache = false; - /* 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()); +/* Whether verbose output is enabled. */ +bool verbose = false; - delete meta; - - metadata.str(""); +/* Ensure that the given segment is listed as a dependency of the current + * snapshot. */ +void add_segment(const string& segment) +{ + segment_list.insert(segment); } /* Read data from a file descriptor and return the amount of data read. A @@ -135,35 +143,46 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, { int64_t size = 0; list object_list; + const char *status = NULL; /* Status indicator printed out */ /* Look up this file in the old stat cache, if we can. If the stat * information indicates that the file has not changed, do not bother - * re-reading the entire contents. */ + * re-reading the entire contents. Even if the information has been + * changed, we can use the list of old blocks in the search for a sub-block + * incremental representation. */ bool cached = false; + list old_blocks; - if (statcache->Find(path, &stat_buf)) { + bool found = metawriter->find(path); + if (found) + old_blocks = metawriter->get_blocks(); + + if (found + && !flag_rebuild_statcache + && metawriter->is_unchanged(&stat_buf)) { cached = true; - const list &blocks = statcache->get_blocks(); /* If any of the blocks in the object have been expired, then we should * fall back to fully reading in the file. */ - for (list::const_iterator i = blocks.begin(); - i != blocks.end(); ++i) { + for (list::const_iterator i = old_blocks.begin(); + i != old_blocks.end(); ++i) { const ObjectReference &ref = *i; if (!db->IsAvailable(ref)) { cached = false; + status = "repack"; break; } } /* If everything looks okay, use the cached information */ if (cached) { - file_info["checksum"] = statcache->get_checksum(); - for (list::const_iterator i = blocks.begin(); - i != blocks.end(); ++i) { + file_info["checksum"] = metawriter->get_checksum(); + for (list::const_iterator i = old_blocks.begin(); + i != old_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; @@ -173,9 +192,10 @@ 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) { - printf(" [new]\n"); - SHA1Checksum hash; + Subfile subfile(db); + subfile.load_old_blocks(old_blocks); + while (true) { ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); if (bytes == 0) @@ -188,17 +208,38 @@ 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); + } + + list refs; // 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 @@ -211,56 +252,77 @@ 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"); - else + 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 { o->set_group("data"); - - o->set_data(block_buf, bytes); - o->write(tss); - ref = o->get_ref(); - db->StoreObject(ref, block_csum, bytes, block_age); - delete o; + status = "new"; + } + + subfile.analyze_new_block(block_buf, bytes); + refs = subfile.create_incremental(tss, o, block_age); + } else { + if (flag_rebuild_statcache && ref.is_normal()) { + subfile.analyze_new_block(block_buf, bytes); + subfile.store_analyzed_signatures(ref); + } + refs.push_back(ref); } - object_list.push_back(ref.to_string()); - segment_list.insert(ref.get_segment()); - db->UseObject(ref); + while (!refs.empty()) { + ref = refs.front(); refs.pop_front(); + object_list.push_back(ref.to_string()); + if (ref.is_normal()) + add_segment(ref.get_segment()); + db->UseObject(ref); + } size += bytes; + + if (status == NULL) + status = "old"; } file_info["checksum"] = hash.checksum_str(); } - 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"; + // 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()); } + } - 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; + if (verbose && status != NULL) + printf(" [%s]\n", status); + + 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; } @@ -278,23 +340,40 @@ void dump_inode(const string& path, // Path within snapshot int64_t file_size; ssize_t len; - printf("%s\n", path.c_str()); + if (verbose) + 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) { + if (pwd != NULL && pwd->pw_name != NULL) { file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")"; } struct group *grp = getgrgid(stat_buf.st_gid); - if (pwd != NULL) { + if (grp != NULL && grp->gr_name != NULL) { file_info["group"] += " (" + uri_encode(grp->gr_name) + ")"; } + 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); + char inode_type; switch (stat_buf.st_mode & S_IFMT) { @@ -322,7 +401,7 @@ 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"); } @@ -330,11 +409,10 @@ 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); - close(fd); if (file_size < 0) return; // error occurred; do not dump file @@ -342,6 +420,7 @@ 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; @@ -356,13 +435,7 @@ 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) @@ -387,7 +460,6 @@ void scanfile(const string& path, bool include) for (list::iterator i = includes.begin(); i != includes.end(); ++i) { if (path == *i) { - printf("Including %s\n", path.c_str()); include = true; } } @@ -395,15 +467,28 @@ void scanfile(const string& path, bool include) for (list::iterator i = excludes.begin(); i != excludes.end(); ++i) { if (path == *i) { - printf("Excluding %s\n", path.c_str()); include = false; } } + if (excluded_names.size() > 0) { + std::string name = path; + std::string::size_type last_slash = name.rfind('/'); + if (last_slash != std::string::npos) { + name.replace(0, last_slash + 1, ""); + } + + for (list::iterator i = excluded_names.begin(); + i != excluded_names.end(); ++i) { + if (name == *i) { + include = false; + } + } + } + for (list::iterator i = searches.begin(); i != searches.end(); ++i) { if (path == *i) { - printf("Scanning %s\n", path.c_str()); scan_only = true; } } @@ -504,7 +589,6 @@ void scanfile(const string& path, bool include) * themselves are excluded from being backed up. */ void add_include(const char *path) { - printf("Add: %s\n", path); /* Was an absolute path specified? If so, we'll need to start scanning * from the root directory. Make sure that the user was consistent in * providing either all relative paths or all absolute paths. */ @@ -543,29 +627,48 @@ void usage(const char *program) { fprintf( stderr, + "Cumulus %s\n\n" "Usage: %s [OPTION]... --dest=DEST PATHS...\n" "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" + " --exclude-name=NAME exclude files called NAME 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" " 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" + " --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" + "\n" + "Exactly one of --dest or --upload-script must be specified.\n", + cumulus_version, program ); } int main(int argc, char *argv[]) { - string backup_source = "."; - 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[] = { @@ -575,11 +678,20 @@ 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 + {"full-metadata", 0, 0, 0}, // 8 + {"tmpdir", 1, 0, 0}, // 9 + {"upload-script", 1, 0, 0}, // 10 + {"rebuild-statcache", 0, 0, 0}, // 11 + {"exclude-name", 1, 0, 0}, // 12 + // Aliases for short options + {"verbose", 0, 0, 'v'}, {NULL, 0, 0, 0}, }; int long_index; - int c = getopt_long(argc, argv, "", long_options, &long_index); + int c = getopt_long(argc, argv, "v", long_options, &long_index); if (c == -1) break; @@ -607,34 +719,64 @@ 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; + case 8: // --full-metadata + flag_full_metadata = true; + break; + case 9: // --tmpdir + tmp_dir = optarg; + break; + case 10: // --upload-script + backup_script = optarg; + break; + case 11: // --rebuild-statcache + flag_rebuild_statcache = true; + break; + case 12: // --exclude-name + excluded_names.push_back(optarg); + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; } } else { - usage(argv[0]); - return 1; + switch (c) { + case 'v': + verbose = true; + break; + default: + usage(argv[0]); + return 1; + } } } - if (argc < optind + 2) { + if (optind == argc) { usage(argv[0]); return 1; } searches.push_back("."); - if (optind == argc) { - add_include("."); - } else { - for (int i = optind; i < argc; i++) - add_include(argv[i]); - } + for (int i = optind; i < argc; i++) + add_include(argv[i]); - backup_source = argv[optind]; + if (backup_dest == "" && backup_script == "") { + fprintf(stderr, + "Error: Backup destination must be specified using --dest= or --upload-script=\n"); + usage(argv[0]); + return 1; + } - if (backup_dest == "") { + if (backup_dest != "" && backup_script != "") { fprintf(stderr, - "Error: Backup destination must be specified with --dest=\n"); + "Error: Cannot specify both --dest= and --upload-script=\n"); usage(argv[0]); return 1; } @@ -643,32 +785,30 @@ int main(int argc, char *argv[]) if (localdb_dir == "") { localdb_dir = backup_dest; } - - // Dump paths for debugging/informational purposes - { - list::const_iterator i; - - printf("LBS Version: %s\n", lbs_version); - - printf("--dest=%s\n--localdb=%s\n\n", - backup_dest.c_str(), localdb_dir.c_str()); - - printf("Includes:\n"); - for (i = includes.begin(); i != includes.end(); ++i) - printf(" %s\n", i->c_str()); - - printf("Excludes:\n"); - for (i = excludes.begin(); i != excludes.end(); ++i) - printf(" %s\n", i->c_str()); - - printf("Searching:\n"); - for (i = searches.begin(); i != searches.end(); ++i) - printf(" %s\n", i->c_str()); + if (localdb_dir == "") { + fprintf(stderr, + "Error: Must specify local database path with --localdb=\n"); + usage(argv[0]); + return 1; } - tss = new TarSegmentStore(backup_dest); 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, backup_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; @@ -683,58 +823,143 @@ int main(int argc, char *argv[]) * snapshot. */ 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); + db->Open(database_path.c_str(), desc_buf, backup_scheme.c_str(), + snapshot_intent); + + 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); + metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf, + backup_scheme.c_str()); 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; - - db->Close(); - - statcache->Close(); - delete statcache; + delete metawriter; tss->sync(); 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 + desc_buf + "." + csum_type + "sums"; + RemoteFile *checksum_file = remote->alloc_file(checksum_filename, + "checksums"); + FILE *checksums = fdopen(checksum_file->get_fd(), "w"); + + 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); + + 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(); + + 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. */ - string desc_filename = backup_dest + "/snapshot-"; + * 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 = "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 v0.1\n"; - descriptor << "Producer: LBS " << lbs_version << "\n"; + 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; + } + 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 v0.8\n"); + fprintf(descriptor, "Producer: Cumulus %s\n", cumulus_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, "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()); + } - 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) { + fatal("Signature filter process error"); + } + } + + 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;