1 /* LBS: An LFS-inspired filesystem backup system
2 * Copyright (C) 2006-2008 Michael Vrable
4 * Main entry point for LBS. Contains logic for traversing the filesystem and
5 * constructing a backup.
18 #include <sys/sysmacros.h>
19 #include <sys/types.h>
44 /* Version information. This will be filled in by the Makefile. */
46 #define LBS_VERSION Unknown
48 #define LBS_STRINGIFY(s) LBS_STRINGIFY2(s)
49 #define LBS_STRINGIFY2(s) #s
50 static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION);
52 static RemoteStore *remote = NULL;
53 static TarSegmentStore *tss = NULL;
54 static MetadataWriter *metawriter = NULL;
56 /* Buffer for holding a single block of data read from a file. */
57 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
58 static char *block_buf;
60 /* Local database, which tracks objects written in this and previous
61 * invocations to help in creating incremental snapshots. */
64 /* Keep track of all segments which are needed to reconstruct the snapshot. */
65 std::set<string> segment_list;
67 /* Snapshot intent: 1=daily, 7=weekly, etc. This is not used directly, but is
68 * stored in the local database and can help guide segment cleaning and
69 * snapshot expiration policies. */
70 double snapshot_intent = 1.0;
72 /* Selection of files to include/exclude in the snapshot. */
73 std::list<string> includes; // Paths in which files should be saved
74 std::list<string> excludes; // Paths which will not be saved
75 std::list<string> searches; // Directories we don't want to save, but
76 // do want to descend searching for data
79 bool relative_paths = true;
81 /* Ensure that the given segment is listed as a dependency of the current
83 void add_segment(const string& segment)
85 segment_list.insert(segment);
88 /* Read data from a file descriptor and return the amount of data read. A
89 * short read (less than the requested size) will only occur if end-of-file is
91 ssize_t file_read(int fd, char *buf, size_t maxlen)
93 size_t bytes_read = 0;
96 ssize_t res = read(fd, buf, maxlen);
100 fprintf(stderr, "error reading file: %m\n");
102 } else if (res == 0) {
114 /* Read the contents of a file (specified by an open file descriptor) and copy
115 * the data to the store. Returns the size of the file (number of bytes
116 * dumped), or -1 on error. */
117 int64_t dumpfile(int fd, dictionary &file_info, const string &path,
118 struct stat& stat_buf)
121 list<string> object_list;
122 const char *status = NULL; /* Status indicator printed out */
124 /* Look up this file in the old stat cache, if we can. If the stat
125 * information indicates that the file has not changed, do not bother
126 * re-reading the entire contents. */
129 if (metawriter->find(path) && metawriter->is_unchanged(&stat_buf)) {
131 list<ObjectReference> blocks = metawriter->get_blocks();
133 /* If any of the blocks in the object have been expired, then we should
134 * fall back to fully reading in the file. */
135 for (list<ObjectReference>::const_iterator i = blocks.begin();
136 i != blocks.end(); ++i) {
137 const ObjectReference &ref = *i;
138 if (!db->IsAvailable(ref)) {
145 /* If everything looks okay, use the cached information */
147 file_info["checksum"] = metawriter->get_checksum();
148 for (list<ObjectReference>::const_iterator i = blocks.begin();
149 i != blocks.end(); ++i) {
150 const ObjectReference &ref = *i;
151 object_list.push_back(ref.to_string());
153 add_segment(ref.get_segment());
156 size = stat_buf.st_size;
160 /* If the file is new or changed, we must read in the contents a block at a
165 ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
169 fprintf(stderr, "Backup contents for %s may be incorrect\n",
174 hash.process(block_buf, bytes);
176 // Sparse file processing: if we read a block of all zeroes, encode
178 bool all_zero = true;
179 for (int i = 0; i < bytes; i++) {
180 if (block_buf[i] != 0) {
186 // Either find a copy of this block in an already-existing segment,
187 // or index it so it can be re-used in the future
188 double block_age = 0.0;
191 SHA1Checksum block_hash;
192 block_hash.process(block_buf, bytes);
193 string block_csum = block_hash.checksum_str();
196 ref = ObjectReference(ObjectReference::REF_ZERO);
197 ref.set_range(0, bytes);
199 ref = db->FindObject(block_csum, bytes);
202 // Store a copy of the object if one does not yet exist
204 LbsObject *o = new LbsObject;
207 /* We might still have seen this checksum before, if the object
208 * was stored at some time in the past, but we have decided to
209 * clean the segment the object was originally stored in
210 * (FindObject will not return such objects). When rewriting
211 * the object contents, put it in a separate group, so that old
212 * objects get grouped together. The hope is that these old
213 * objects will continue to be used in the future, and we
214 * obtain segments which will continue to be well-utilized.
215 * Additionally, keep track of the age of the data by looking
216 * up the age of the block which was expired and using that
217 * instead of the current time. */
218 if (db->IsOldObject(block_csum, bytes,
219 &block_age, &object_group)) {
220 if (object_group == 0) {
221 o->set_group("data");
224 sprintf(group, "compacted-%d", object_group);
230 o->set_group("data");
234 o->set_data(block_buf, bytes);
237 db->StoreObject(ref, block_csum, bytes, block_age);
238 ref.set_range(0, bytes);
242 object_list.push_back(ref.to_string());
244 add_segment(ref.get_segment());
252 file_info["checksum"] = hash.checksum_str();
256 printf(" [%s]\n", status);
258 string blocklist = "";
259 for (list<string>::iterator i = object_list.begin();
260 i != object_list.end(); ++i) {
261 if (i != object_list.begin())
265 file_info["data"] = blocklist;
270 /* Dump a specified filesystem object (file, directory, etc.) based on its
271 * inode information. If the object is a regular file, an open filehandle is
273 void dump_inode(const string& path, // Path within snapshot
274 const string& fullpath, // Path to object in filesystem
275 struct stat& stat_buf, // Results of stat() call
276 int fd) // Open filehandle if regular file
279 dictionary file_info;
283 printf("%s\n", path.c_str());
284 metawriter->find(path);
286 file_info["name"] = uri_encode(path);
287 file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8);
288 file_info["ctime"] = encode_int(stat_buf.st_ctime);
289 file_info["mtime"] = encode_int(stat_buf.st_mtime);
290 file_info["user"] = encode_int(stat_buf.st_uid);
291 file_info["group"] = encode_int(stat_buf.st_gid);
293 time_t now = time(NULL);
294 if (now - stat_buf.st_ctime < 30 || now - stat_buf.st_mtime < 30)
295 if ((stat_buf.st_mode & S_IFMT) != S_IFDIR)
296 file_info["volatile"] = "1";
298 struct passwd *pwd = getpwuid(stat_buf.st_uid);
300 file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
303 struct group *grp = getgrgid(stat_buf.st_gid);
305 file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
308 if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) {
309 file_info["links"] = encode_int(stat_buf.st_nlink);
312 file_info["inode"] = encode_int(major(stat_buf.st_dev))
313 + "/" + encode_int(minor(stat_buf.st_dev))
314 + "/" + encode_int(stat_buf.st_ino);
318 switch (stat_buf.st_mode & S_IFMT) {
327 inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c';
328 file_info["device"] = encode_int(major(stat_buf.st_rdev))
329 + "/" + encode_int(minor(stat_buf.st_rdev));
334 /* Use the reported file size to allocate a buffer large enough to read
335 * the symlink. Allocate slightly more space, so that we ask for more
336 * bytes than we expect and so check for truncation. */
337 buf = new char[stat_buf.st_size + 2];
338 len = readlink(fullpath.c_str(), buf, stat_buf.st_size + 1);
340 fprintf(stderr, "error reading symlink: %m\n");
341 } else if (len <= stat_buf.st_size) {
343 file_info["target"] = uri_encode(buf);
344 } else if (len > stat_buf.st_size) {
345 fprintf(stderr, "error reading symlink: name truncated\n");
353 file_size = dumpfile(fd, file_info, path, stat_buf);
354 file_info["size"] = encode_int(file_size);
357 return; // error occurred; do not dump file
359 if (file_size != stat_buf.st_size) {
360 fprintf(stderr, "Warning: Size of %s changed during reading\n",
362 file_info["volatile"] = "1";
371 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
375 file_info["type"] = string(1, inode_type);
377 metawriter->add(file_info);
380 void scanfile(const string& path, bool include)
384 struct stat stat_buf;
391 true_path = "/" + path;
393 // Set to true if we should scan through the contents of this directory,
394 // but not actually back files up
395 bool scan_only = false;
397 // Check this file against the include/exclude list to see if it should be
399 for (list<string>::iterator i = includes.begin();
400 i != includes.end(); ++i) {
402 printf("Including %s\n", path.c_str());
407 for (list<string>::iterator i = excludes.begin();
408 i != excludes.end(); ++i) {
410 printf("Excluding %s\n", path.c_str());
415 for (list<string>::iterator i = searches.begin();
416 i != searches.end(); ++i) {
418 printf("Scanning %s\n", path.c_str());
423 if (!include && !scan_only)
426 if (lstat(true_path.c_str(), &stat_buf) < 0) {
427 fprintf(stderr, "lstat(%s): %m\n", path.c_str());
431 if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
432 /* Be paranoid when opening the file. We have no guarantee that the
433 * file was not replaced between the stat() call above and the open()
434 * call below, so we might not even be opening a regular file. We
435 * supply flags to open to to guard against various conditions before
436 * we can perform an lstat to check that the file is still a regular
438 * - O_NOFOLLOW: in the event the file was replaced by a symlink
439 * - O_NONBLOCK: prevents open() from blocking if the file was
441 * We also add in O_NOATIME, since this may reduce disk writes (for
442 * inode updates). However, O_NOATIME may result in EPERM, so if the
443 * initial open fails, try again without O_NOATIME. */
444 fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
446 fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK);
449 fprintf(stderr, "Unable to open file %s: %m\n", path.c_str());
453 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
455 flags = fcntl(fd, F_GETFL);
456 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
458 /* Perform the stat call again, and check that we still have a regular
460 if (fstat(fd, &stat_buf) < 0) {
461 fprintf(stderr, "fstat: %m\n");
466 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
467 fprintf(stderr, "file is no longer a regular file!\n");
473 dump_inode(path, true_path, stat_buf, fd);
478 // If we hit a directory, now that we've written the directory itself,
479 // recursively scan the directory.
480 if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) {
481 DIR *dir = opendir(true_path.c_str());
484 fprintf(stderr, "Error: %m\n");
489 vector<string> contents;
490 while ((ent = readdir(dir)) != NULL) {
491 string filename(ent->d_name);
492 if (filename == "." || filename == "..")
494 contents.push_back(filename);
499 sort(contents.begin(), contents.end());
501 for (vector<string>::iterator i = contents.begin();
502 i != contents.end(); ++i) {
503 const string& filename = *i;
505 scanfile(filename, include);
507 scanfile(path + "/" + filename, include);
512 /* Include the specified file path in the backups. Append the path to the
513 * includes list, and to ensure that we actually see the path when scanning the
514 * directory tree, add all the parent directories to the search list, which
515 * means we will scan through the directory listing even if the files
516 * themselves are excluded from being backed up. */
517 void add_include(const char *path)
519 printf("Add: %s\n", path);
520 /* Was an absolute path specified? If so, we'll need to start scanning
521 * from the root directory. Make sure that the user was consistent in
522 * providing either all relative paths or all absolute paths. */
523 if (path[0] == '/') {
524 if (includes.size() > 0 && relative_paths == true) {
526 "Error: Cannot mix relative and absolute paths!\n");
530 relative_paths = false;
532 // Skip over leading '/'
534 } else if (relative_paths == false && path[0] != '/') {
535 fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n");
539 includes.push_back(path);
541 /* Split the specified path into directory components, and ensure that we
542 * descend into all the directories along the path. */
543 const char *slash = path;
548 while ((slash = strchr(slash + 1, '/')) != NULL) {
549 string component(path, slash - path);
550 searches.push_back(component);
554 void usage(const char *program)
559 "Usage: %s [OPTION]... --dest=DEST PATHS...\n"
560 "Produce backup snapshot of files in SOURCE and store to DEST.\n"
563 " --dest=PATH path where backup is to be written\n"
564 " --upload-script=COMMAND\n"
565 " program to invoke for each backup file generated\n"
566 " --exclude=PATH exclude files in PATH from snapshot\n"
567 " --localdb=PATH local backup metadata is stored in PATH\n"
568 " --tmpdir=PATH path for temporarily storing backup files\n"
569 " (defaults to TMPDIR environment variable or /tmp)\n"
570 " --filter=COMMAND program through which to filter segment data\n"
571 " (defaults to \"bzip2 -c\")\n"
572 " --filter-extension=EXT\n"
573 " string to append to segment files\n"
574 " (defaults to \".bz2\")\n"
575 " --signature-filter=COMMAND\n"
576 " program though which to filter descriptor\n"
577 " --scheme=NAME optional name for this snapshot\n"
578 " --intent=FLOAT intended backup type: 1=daily, 7=weekly, ...\n"
579 " (defaults to \"1\")\n"
580 " --full-metadata do not re-use metadata from previous backups\n"
582 "Exactly one of --dest or --upload-script must be specified.\n",
587 int main(int argc, char *argv[])
589 string backup_dest = "", backup_script = "";
590 string localdb_dir = "";
591 string backup_scheme = "";
592 string signature_filter = "";
594 string tmp_dir = "/tmp";
595 if (getenv("TMPDIR") != NULL)
596 tmp_dir = getenv("TMPDIR");
599 static struct option long_options[] = {
600 {"localdb", 1, 0, 0}, // 0
601 {"exclude", 1, 0, 0}, // 1
602 {"filter", 1, 0, 0}, // 2
603 {"filter-extension", 1, 0, 0}, // 3
604 {"dest", 1, 0, 0}, // 4
605 {"scheme", 1, 0, 0}, // 5
606 {"signature-filter", 1, 0, 0}, // 6
607 {"intent", 1, 0, 0}, // 7
608 {"full-metadata", 0, 0, 0}, // 8
609 {"tmpdir", 1, 0, 0}, // 9
610 {"upload-script", 1, 0, 0}, // 10
615 int c = getopt_long(argc, argv, "", long_options, &long_index);
621 switch (long_index) {
623 localdb_dir = optarg;
626 if (optarg[0] != '/')
627 excludes.push_back(optarg);
629 excludes.push_back(optarg + 1);
632 filter_program = optarg;
634 case 3: // --filter-extension
635 filter_extension = optarg;
638 backup_dest = optarg;
641 backup_scheme = optarg;
643 case 6: // --signature-filter
644 signature_filter = optarg;
647 snapshot_intent = atof(optarg);
648 if (snapshot_intent <= 0)
651 case 8: // --full-metadata
652 flag_full_metadata = true;
657 case 10: // --upload-script
658 backup_script = optarg;
661 fprintf(stderr, "Unhandled long option!\n");
670 if (optind == argc) {
675 searches.push_back(".");
676 for (int i = optind; i < argc; i++)
677 add_include(argv[i]);
679 if (backup_dest == "" && backup_script == "") {
681 "Error: Backup destination must be specified using --dest= or --upload-script=\n");
686 if (backup_dest != "" && backup_script != "") {
688 "Error: Cannot specify both --dest= and --upload-script=\n");
693 // Default for --localdb is the same as --dest
694 if (localdb_dir == "") {
695 localdb_dir = backup_dest;
697 if (localdb_dir == "") {
699 "Error: Must specify local database path with --localdb=\n");
704 // Dump paths for debugging/informational purposes
706 list<string>::const_iterator i;
708 printf("LBS Version: %s\n", lbs_version);
710 printf("--dest=%s\n--localdb=%s\n--upload-script=%s\n",
711 backup_dest.c_str(), localdb_dir.c_str(), backup_script.c_str());
713 printf("Includes:\n");
714 for (i = includes.begin(); i != includes.end(); ++i)
715 printf(" %s\n", i->c_str());
717 printf("Excludes:\n");
718 for (i = excludes.begin(); i != excludes.end(); ++i)
719 printf(" %s\n", i->c_str());
721 printf("Searching:\n");
722 for (i = searches.begin(); i != searches.end(); ++i)
723 printf(" %s\n", i->c_str());
726 block_buf = new char[LBS_BLOCK_SIZE];
728 /* Initialize the remote storage layer. If using an upload script, create
729 * a temporary directory for staging files. Otherwise, write backups
730 * directly to the destination directory. */
731 if (backup_script != "") {
732 tmp_dir = tmp_dir + "/lbs." + generate_uuid();
733 if (mkdir(tmp_dir.c_str(), 0700) < 0) {
734 fprintf(stderr, "Cannot create temporary directory %s: %m\n",
738 remote = new RemoteStore(tmp_dir);
739 remote->set_script(backup_script);
741 remote = new RemoteStore(backup_dest);
744 /* Store the time when the backup started, so it can be included in the
750 localtime_r(&now, &time_buf);
751 strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
753 /* Open the local database which tracks all objects that are stored
754 * remotely, for efficient incrementals. Provide it with the name of this
756 string database_path = localdb_dir + "/localdb.sqlite";
758 db->Open(database_path.c_str(), desc_buf,
759 backup_scheme.size() ? backup_scheme.c_str() : NULL,
762 tss = new TarSegmentStore(remote, db);
764 /* Initialize the stat cache, for skipping over unchanged files. */
765 metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf,
767 ? backup_scheme.c_str()
770 scanfile(".", false);
772 ObjectReference root_ref = metawriter->close();
773 add_segment(root_ref.get_segment());
774 string backup_root = root_ref.to_string();
782 /* Write out a checksums file which lists the checksums for all the
783 * segments included in this snapshot. The format is designed so that it
784 * may be easily verified using the sha1sums command. */
785 const char csum_type[] = "sha1";
786 string checksum_filename = "snapshot-";
787 if (backup_scheme.size() > 0)
788 checksum_filename += backup_scheme + "-";
789 checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums";
790 RemoteFile *checksum_file = remote->alloc_file(checksum_filename,
792 FILE *checksums = fdopen(checksum_file->get_fd(), "w");
794 for (std::set<string>::iterator i = segment_list.begin();
795 i != segment_list.end(); ++i) {
796 string seg_path, seg_csum;
797 if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) {
798 const char *raw_checksum = NULL;
799 if (strncmp(seg_csum.c_str(), csum_type,
800 strlen(csum_type)) == 0) {
801 raw_checksum = seg_csum.c_str() + strlen(csum_type);
802 if (*raw_checksum == '=')
808 if (raw_checksum != NULL)
809 fprintf(checksums, "%s *%s\n",
810 raw_checksum, seg_path.c_str());
814 checksum_file->send();
818 /* All other files should be flushed to remote storage before writing the
819 * backup descriptor below, so that it is not possible to have a backup
820 * descriptor written out depending on non-existent (not yet written)
824 /* Write a backup descriptor file, which says which segments are needed and
825 * where to start to restore this snapshot. The filename is based on the
826 * current time. If a signature filter program was specified, filter the
827 * data through that to give a chance to sign the descriptor contents. */
828 string desc_filename = "snapshot-";
829 if (backup_scheme.size() > 0)
830 desc_filename += backup_scheme + "-";
831 desc_filename = desc_filename + desc_buf + ".lbs";
833 RemoteFile *descriptor_file = remote->alloc_file(desc_filename,
835 int descriptor_fd = descriptor_file->get_fd();
836 if (descriptor_fd < 0) {
837 fprintf(stderr, "Unable to open descriptor output file: %m\n");
840 pid_t signature_pid = 0;
841 if (signature_filter.size() > 0) {
842 int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(),
844 close(descriptor_fd);
845 descriptor_fd = new_fd;
847 FILE *descriptor = fdopen(descriptor_fd, "w");
849 fprintf(descriptor, "Format: LBS Snapshot v0.6\n");
850 fprintf(descriptor, "Producer: LBS %s\n", lbs_version);
851 strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
852 fprintf(descriptor, "Date: %s\n", desc_buf);
853 if (backup_scheme.size() > 0)
854 fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str());
855 fprintf(descriptor, "Backup-Intent: %g\n", snapshot_intent);
856 fprintf(descriptor, "Root: %s\n", backup_root.c_str());
858 SHA1Checksum checksum_csum;
859 if (checksum_csum.process_file(checksum_filename.c_str())) {
860 string csum = checksum_csum.checksum_str();
861 fprintf(descriptor, "Checksums: %s\n", csum.c_str());
864 fprintf(descriptor, "Segments:\n");
865 for (std::set<string>::iterator i = segment_list.begin();
866 i != segment_list.end(); ++i) {
867 fprintf(descriptor, " %s\n", i->c_str());
874 waitpid(signature_pid, &status, 0);
876 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
877 throw IOException("Signature filter process error");
881 descriptor_file->send();
886 if (backup_script != "") {
887 if (rmdir(tmp_dir.c_str()) < 0) {
889 "Warning: Cannot delete temporary directory %s: %m\n",