1 /* Recursively descend the filesystem and visit each file. */
13 #include <sys/sysmacros.h>
14 #include <sys/types.h>
38 /* Version information. This will be filled in by the Makefile. */
40 #define LBS_VERSION Unknown
42 #define LBS_STRINGIFY(s) LBS_STRINGIFY2(s)
43 #define LBS_STRINGIFY2(s) #s
44 static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION);
46 static TarSegmentStore *tss = NULL;
47 static MetadataWriter *metawriter = NULL;
49 /* Buffer for holding a single block of data read from a file. */
50 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
51 static char *block_buf;
53 /* Local database, which tracks objects written in this and previous
54 * invocations to help in creating incremental snapshots. */
57 /* Keep track of all segments which are needed to reconstruct the snapshot. */
58 std::set<string> segment_list;
60 /* Selection of files to include/exclude in the snapshot. */
61 std::list<string> includes; // Paths in which files should be saved
62 std::list<string> excludes; // Paths which will not be saved
63 std::list<string> searches; // Directories we don't want to save, but
64 // do want to descend searching for data
67 bool relative_paths = true;
69 /* Ensure that the given segment is listed as a dependency of the current
71 void add_segment(const string& segment)
73 segment_list.insert(segment);
76 /* Read data from a file descriptor and return the amount of data read. A
77 * short read (less than the requested size) will only occur if end-of-file is
79 ssize_t file_read(int fd, char *buf, size_t maxlen)
81 size_t bytes_read = 0;
84 ssize_t res = read(fd, buf, maxlen);
88 fprintf(stderr, "error reading file: %m\n");
90 } else if (res == 0) {
102 /* Read the contents of a file (specified by an open file descriptor) and copy
103 * the data to the store. Returns the size of the file (number of bytes
104 * dumped), or -1 on error. */
105 int64_t dumpfile(int fd, dictionary &file_info, const string &path,
106 struct stat& stat_buf)
109 list<string> object_list;
110 const char *status = NULL; /* Status indicator printed out */
112 /* Look up this file in the old stat cache, if we can. If the stat
113 * information indicates that the file has not changed, do not bother
114 * re-reading the entire contents. */
117 if (metawriter->find(path) && metawriter->is_unchanged(&stat_buf)) {
119 list<ObjectReference> blocks = metawriter->get_blocks();
121 /* If any of the blocks in the object have been expired, then we should
122 * fall back to fully reading in the file. */
123 for (list<ObjectReference>::const_iterator i = blocks.begin();
124 i != blocks.end(); ++i) {
125 const ObjectReference &ref = *i;
126 if (!db->IsAvailable(ref)) {
133 /* If everything looks okay, use the cached information */
135 file_info["checksum"] = metawriter->get_checksum();
136 for (list<ObjectReference>::const_iterator i = blocks.begin();
137 i != blocks.end(); ++i) {
138 const ObjectReference &ref = *i;
139 object_list.push_back(ref.to_string());
140 segment_list.insert(ref.get_segment());
143 size = stat_buf.st_size;
147 /* If the file is new or changed, we must read in the contents a block at a
152 ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
156 fprintf(stderr, "Backup contents for %s may be incorrect\n",
161 hash.process(block_buf, bytes);
163 // Either find a copy of this block in an already-existing segment,
164 // or index it so it can be re-used in the future
165 double block_age = 0.0;
166 SHA1Checksum block_hash;
167 block_hash.process(block_buf, bytes);
168 string block_csum = block_hash.checksum_str();
169 ObjectReference ref = db->FindObject(block_csum, bytes);
171 // Store a copy of the object if one does not yet exist
172 if (ref.get_segment().size() == 0) {
173 LbsObject *o = new LbsObject;
176 /* We might still have seen this checksum before, if the object
177 * was stored at some time in the past, but we have decided to
178 * clean the segment the object was originally stored in
179 * (FindObject will not return such objects). When rewriting
180 * the object contents, put it in a separate group, so that old
181 * objects get grouped together. The hope is that these old
182 * objects will continue to be used in the future, and we
183 * obtain segments which will continue to be well-utilized.
184 * Additionally, keep track of the age of the data by looking
185 * up the age of the block which was expired and using that
186 * instead of the current time. */
187 if (db->IsOldObject(block_csum, bytes,
188 &block_age, &object_group)) {
189 if (object_group == 0) {
190 o->set_group("data");
193 sprintf(group, "compacted-%d", object_group);
199 o->set_group("data");
203 o->set_data(block_buf, bytes);
206 db->StoreObject(ref, block_csum, bytes, block_age);
210 object_list.push_back(ref.to_string());
211 segment_list.insert(ref.get_segment());
219 file_info["checksum"] = hash.checksum_str();
223 printf(" [%s]\n", status);
225 string blocklist = "";
226 for (list<string>::iterator i = object_list.begin();
227 i != object_list.end(); ++i) {
228 if (i != object_list.begin())
232 file_info["data"] = blocklist;
237 /* Dump a specified filesystem object (file, directory, etc.) based on its
238 * inode information. If the object is a regular file, an open filehandle is
240 void dump_inode(const string& path, // Path within snapshot
241 const string& fullpath, // Path to object in filesystem
242 struct stat& stat_buf, // Results of stat() call
243 int fd) // Open filehandle if regular file
246 dictionary file_info;
250 printf("%s\n", path.c_str());
251 metawriter->find(path);
253 file_info["name"] = uri_encode(path);
254 file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8);
255 file_info["ctime"] = encode_int(stat_buf.st_ctime);
256 file_info["mtime"] = encode_int(stat_buf.st_mtime);
257 file_info["user"] = encode_int(stat_buf.st_uid);
258 file_info["group"] = encode_int(stat_buf.st_gid);
260 time_t now = time(NULL);
261 if (now - stat_buf.st_ctime < 30 || now - stat_buf.st_mtime < 30)
262 if ((stat_buf.st_mode & S_IFMT) != S_IFDIR)
263 file_info["volatile"] = "1";
265 struct passwd *pwd = getpwuid(stat_buf.st_uid);
267 file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
270 struct group *grp = getgrgid(stat_buf.st_gid);
272 file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
275 if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) {
276 file_info["links"] = encode_int(stat_buf.st_nlink);
279 file_info["inode"] = encode_int(major(stat_buf.st_dev))
280 + "/" + encode_int(minor(stat_buf.st_dev))
281 + "/" + encode_int(stat_buf.st_ino);
285 switch (stat_buf.st_mode & S_IFMT) {
294 inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c';
295 file_info["device"] = encode_int(major(stat_buf.st_rdev))
296 + "/" + encode_int(minor(stat_buf.st_rdev));
301 /* Use the reported file size to allocate a buffer large enough to read
302 * the symlink. Allocate slightly more space, so that we ask for more
303 * bytes than we expect and so check for truncation. */
304 buf = new char[stat_buf.st_size + 2];
305 len = readlink(fullpath.c_str(), buf, stat_buf.st_size + 1);
307 fprintf(stderr, "error reading symlink: %m\n");
308 } else if (len <= stat_buf.st_size) {
310 file_info["target"] = uri_encode(buf);
311 } else if (len > stat_buf.st_size) {
312 fprintf(stderr, "error reading symlink: name truncated\n");
320 file_size = dumpfile(fd, file_info, path, stat_buf);
321 file_info["size"] = encode_int(file_size);
324 return; // error occurred; do not dump file
326 if (file_size != stat_buf.st_size) {
327 fprintf(stderr, "Warning: Size of %s changed during reading\n",
329 file_info["volatile"] = "1";
338 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
342 file_info["type"] = string(1, inode_type);
344 metawriter->add(file_info);
347 void scanfile(const string& path, bool include)
351 struct stat stat_buf;
358 true_path = "/" + path;
360 // Set to true if we should scan through the contents of this directory,
361 // but not actually back files up
362 bool scan_only = false;
364 // Check this file against the include/exclude list to see if it should be
366 for (list<string>::iterator i = includes.begin();
367 i != includes.end(); ++i) {
369 printf("Including %s\n", path.c_str());
374 for (list<string>::iterator i = excludes.begin();
375 i != excludes.end(); ++i) {
377 printf("Excluding %s\n", path.c_str());
382 for (list<string>::iterator i = searches.begin();
383 i != searches.end(); ++i) {
385 printf("Scanning %s\n", path.c_str());
390 if (!include && !scan_only)
393 if (lstat(true_path.c_str(), &stat_buf) < 0) {
394 fprintf(stderr, "lstat(%s): %m\n", path.c_str());
398 if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
399 /* Be paranoid when opening the file. We have no guarantee that the
400 * file was not replaced between the stat() call above and the open()
401 * call below, so we might not even be opening a regular file. We
402 * supply flags to open to to guard against various conditions before
403 * we can perform an lstat to check that the file is still a regular
405 * - O_NOFOLLOW: in the event the file was replaced by a symlink
406 * - O_NONBLOCK: prevents open() from blocking if the file was
408 * We also add in O_NOATIME, since this may reduce disk writes (for
409 * inode updates). However, O_NOATIME may result in EPERM, so if the
410 * initial open fails, try again without O_NOATIME. */
411 fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
413 fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK);
416 fprintf(stderr, "Unable to open file %s: %m\n", path.c_str());
420 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
422 flags = fcntl(fd, F_GETFL);
423 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
425 /* Perform the stat call again, and check that we still have a regular
427 if (fstat(fd, &stat_buf) < 0) {
428 fprintf(stderr, "fstat: %m\n");
433 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
434 fprintf(stderr, "file is no longer a regular file!\n");
440 dump_inode(path, true_path, stat_buf, fd);
445 // If we hit a directory, now that we've written the directory itself,
446 // recursively scan the directory.
447 if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) {
448 DIR *dir = opendir(true_path.c_str());
451 fprintf(stderr, "Error: %m\n");
456 vector<string> contents;
457 while ((ent = readdir(dir)) != NULL) {
458 string filename(ent->d_name);
459 if (filename == "." || filename == "..")
461 contents.push_back(filename);
466 sort(contents.begin(), contents.end());
468 for (vector<string>::iterator i = contents.begin();
469 i != contents.end(); ++i) {
470 const string& filename = *i;
472 scanfile(filename, include);
474 scanfile(path + "/" + filename, include);
479 /* Include the specified file path in the backups. Append the path to the
480 * includes list, and to ensure that we actually see the path when scanning the
481 * directory tree, add all the parent directories to the search list, which
482 * means we will scan through the directory listing even if the files
483 * themselves are excluded from being backed up. */
484 void add_include(const char *path)
486 printf("Add: %s\n", path);
487 /* Was an absolute path specified? If so, we'll need to start scanning
488 * from the root directory. Make sure that the user was consistent in
489 * providing either all relative paths or all absolute paths. */
490 if (path[0] == '/') {
491 if (includes.size() > 0 && relative_paths == true) {
493 "Error: Cannot mix relative and absolute paths!\n");
497 relative_paths = false;
499 // Skip over leading '/'
501 } else if (relative_paths == false && path[0] != '/') {
502 fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n");
506 includes.push_back(path);
508 /* Split the specified path into directory components, and ensure that we
509 * descend into all the directories along the path. */
510 const char *slash = path;
515 while ((slash = strchr(slash + 1, '/')) != NULL) {
516 string component(path, slash - path);
517 searches.push_back(component);
521 void usage(const char *program)
526 "Usage: %s [OPTION]... --dest=DEST PATHS...\n"
527 "Produce backup snapshot of files in SOURCE and store to DEST.\n"
530 " --dest=PATH path where backup is to be written [REQUIRED]\n"
531 " --exclude=PATH exclude files in PATH from snapshot\n"
532 " --localdb=PATH local backup metadata is stored in PATH\n"
533 " --filter=COMMAND program through which to filter segment data\n"
534 " (defaults to \"bzip2 -c\")\n"
535 " --filter-extension=EXT\n"
536 " string to append to segment files\n"
537 " (defaults to \".bz2\")\n"
538 " --signature-filter=COMMAND\n"
539 " program though which to filter descriptor\n"
540 " --scheme=NAME optional name for this snapshot\n",
545 int main(int argc, char *argv[])
547 string backup_dest = "";
548 string localdb_dir = "";
549 string backup_scheme = "";
550 string signature_filter = "";
553 static struct option long_options[] = {
554 {"localdb", 1, 0, 0}, // 0
555 {"exclude", 1, 0, 0}, // 1
556 {"filter", 1, 0, 0}, // 2
557 {"filter-extension", 1, 0, 0}, // 3
558 {"dest", 1, 0, 0}, // 4
559 {"scheme", 1, 0, 0}, // 5
560 {"signature-filter", 1, 0, 0}, // 6
565 int c = getopt_long(argc, argv, "", long_options, &long_index);
571 switch (long_index) {
573 localdb_dir = optarg;
576 if (optarg[0] != '/')
577 excludes.push_back(optarg);
579 excludes.push_back(optarg + 1);
582 filter_program = optarg;
584 case 3: // --filter-extension
585 filter_extension = optarg;
588 backup_dest = optarg;
591 backup_scheme = optarg;
593 case 6: // --signature-filter
594 signature_filter = optarg;
597 fprintf(stderr, "Unhandled long option!\n");
606 if (optind == argc) {
611 searches.push_back(".");
612 for (int i = optind; i < argc; i++)
613 add_include(argv[i]);
615 if (backup_dest == "") {
617 "Error: Backup destination must be specified with --dest=\n");
622 // Default for --localdb is the same as --dest
623 if (localdb_dir == "") {
624 localdb_dir = backup_dest;
627 // Dump paths for debugging/informational purposes
629 list<string>::const_iterator i;
631 printf("LBS Version: %s\n", lbs_version);
633 printf("--dest=%s\n--localdb=%s\n\n",
634 backup_dest.c_str(), localdb_dir.c_str());
636 printf("Includes:\n");
637 for (i = includes.begin(); i != includes.end(); ++i)
638 printf(" %s\n", i->c_str());
640 printf("Excludes:\n");
641 for (i = excludes.begin(); i != excludes.end(); ++i)
642 printf(" %s\n", i->c_str());
644 printf("Searching:\n");
645 for (i = searches.begin(); i != searches.end(); ++i)
646 printf(" %s\n", i->c_str());
649 block_buf = new char[LBS_BLOCK_SIZE];
651 /* Store the time when the backup started, so it can be included in the
657 localtime_r(&now, &time_buf);
658 strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
660 /* Open the local database which tracks all objects that are stored
661 * remotely, for efficient incrementals. Provide it with the name of this
663 string database_path = localdb_dir + "/localdb.sqlite";
665 db->Open(database_path.c_str(), desc_buf,
666 backup_scheme.size() ? backup_scheme.c_str() : NULL);
668 tss = new TarSegmentStore(backup_dest, db);
670 /* Initialize the stat cache, for skipping over unchanged files. */
671 metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf,
673 ? backup_scheme.c_str()
676 scanfile(".", false);
678 ObjectReference root_ref = metawriter->close();
679 add_segment(root_ref.get_segment());
680 string backup_root = root_ref.to_string();
688 /* Write out a checksums file which lists the checksums for all the
689 * segments included in this snapshot. The format is designed so that it
690 * may be easily verified using the sha1sums command. */
691 const char csum_type[] = "sha1";
692 string checksum_filename = backup_dest + "/snapshot-";
693 if (backup_scheme.size() > 0)
694 checksum_filename += backup_scheme + "-";
695 checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums";
696 FILE *checksums = fopen(checksum_filename.c_str(), "w");
697 if (checksums != NULL) {
698 for (std::set<string>::iterator i = segment_list.begin();
699 i != segment_list.end(); ++i) {
700 string seg_path, seg_csum;
701 if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) {
702 const char *raw_checksum = NULL;
703 if (strncmp(seg_csum.c_str(), csum_type,
704 strlen(csum_type)) == 0) {
705 raw_checksum = seg_csum.c_str() + strlen(csum_type);
706 if (*raw_checksum == '=')
712 if (raw_checksum != NULL)
713 fprintf(checksums, "%s *%s\n",
714 raw_checksum, seg_path.c_str());
719 fprintf(stderr, "ERROR: Unable to write checksums file: %m\n");
724 /* Write a backup descriptor file, which says which segments are needed and
725 * where to start to restore this snapshot. The filename is based on the
726 * current time. If a signature filter program was specified, filter the
727 * data through that to give a chance to sign the descriptor contents. */
728 string desc_filename = backup_dest + "/snapshot-";
729 if (backup_scheme.size() > 0)
730 desc_filename += backup_scheme + "-";
731 desc_filename = desc_filename + desc_buf + ".lbs";
733 int descriptor_fd = open(desc_filename.c_str(), O_WRONLY | O_CREAT, 0666);
734 if (descriptor_fd < 0) {
735 fprintf(stderr, "Unable to open descriptor output file: %m\n");
738 pid_t signature_pid = 0;
739 if (signature_filter.size() > 0) {
740 int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(),
742 close(descriptor_fd);
743 descriptor_fd = new_fd;
745 FILE *descriptor = fdopen(descriptor_fd, "w");
747 fprintf(descriptor, "Format: LBS Snapshot v0.6\n");
748 fprintf(descriptor, "Producer: LBS %s\n", lbs_version);
749 strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
750 fprintf(descriptor, "Date: %s\n", desc_buf);
751 if (backup_scheme.size() > 0)
752 fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str());
753 fprintf(descriptor, "Root: %s\n", backup_root.c_str());
755 SHA1Checksum checksum_csum;
756 if (checksum_csum.process_file(checksum_filename.c_str())) {
757 string csum = checksum_csum.checksum_str();
758 fprintf(descriptor, "Checksums: %s\n", csum.c_str());
761 fprintf(descriptor, "Segments:\n");
762 for (std::set<string>::iterator i = segment_list.begin();
763 i != segment_list.end(); ++i) {
764 fprintf(descriptor, " %s\n", i->c_str());
771 waitpid(signature_pid, &status, 0);
773 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
774 throw IOException("Signature filter process error");