1 /* Recursively descend the filesystem and visit each file. */
13 #include <sys/sysmacros.h>
14 #include <sys/types.h>
39 /* Version information. This will be filled in by the Makefile. */
41 #define LBS_VERSION Unknown
43 #define LBS_STRINGIFY(s) LBS_STRINGIFY2(s)
44 #define LBS_STRINGIFY2(s) #s
45 static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION);
47 static RemoteStore *remote = NULL;
48 static TarSegmentStore *tss = NULL;
49 static MetadataWriter *metawriter = NULL;
51 /* Buffer for holding a single block of data read from a file. */
52 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
53 static char *block_buf;
55 /* Local database, which tracks objects written in this and previous
56 * invocations to help in creating incremental snapshots. */
59 /* Keep track of all segments which are needed to reconstruct the snapshot. */
60 std::set<string> segment_list;
62 /* Snapshot intent: 1=daily, 7=weekly, etc. This is not used directly, but is
63 * stored in the local database and can help guide segment cleaning and
64 * snapshot expiration policies. */
65 double snapshot_intent = 1.0;
67 /* Selection of files to include/exclude in the snapshot. */
68 std::list<string> includes; // Paths in which files should be saved
69 std::list<string> excludes; // Paths which will not be saved
70 std::list<string> searches; // Directories we don't want to save, but
71 // do want to descend searching for data
74 bool relative_paths = true;
76 /* Ensure that the given segment is listed as a dependency of the current
78 void add_segment(const string& segment)
80 segment_list.insert(segment);
83 /* Read data from a file descriptor and return the amount of data read. A
84 * short read (less than the requested size) will only occur if end-of-file is
86 ssize_t file_read(int fd, char *buf, size_t maxlen)
88 size_t bytes_read = 0;
91 ssize_t res = read(fd, buf, maxlen);
95 fprintf(stderr, "error reading file: %m\n");
97 } else if (res == 0) {
109 /* Read the contents of a file (specified by an open file descriptor) and copy
110 * the data to the store. Returns the size of the file (number of bytes
111 * dumped), or -1 on error. */
112 int64_t dumpfile(int fd, dictionary &file_info, const string &path,
113 struct stat& stat_buf)
116 list<string> object_list;
117 const char *status = NULL; /* Status indicator printed out */
119 /* Look up this file in the old stat cache, if we can. If the stat
120 * information indicates that the file has not changed, do not bother
121 * re-reading the entire contents. */
124 if (metawriter->find(path) && metawriter->is_unchanged(&stat_buf)) {
126 list<ObjectReference> blocks = metawriter->get_blocks();
128 /* If any of the blocks in the object have been expired, then we should
129 * fall back to fully reading in the file. */
130 for (list<ObjectReference>::const_iterator i = blocks.begin();
131 i != blocks.end(); ++i) {
132 const ObjectReference &ref = *i;
133 if (!db->IsAvailable(ref)) {
140 /* If everything looks okay, use the cached information */
142 file_info["checksum"] = metawriter->get_checksum();
143 for (list<ObjectReference>::const_iterator i = blocks.begin();
144 i != blocks.end(); ++i) {
145 const ObjectReference &ref = *i;
146 object_list.push_back(ref.to_string());
148 add_segment(ref.get_segment());
151 size = stat_buf.st_size;
155 /* If the file is new or changed, we must read in the contents a block at a
160 ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
164 fprintf(stderr, "Backup contents for %s may be incorrect\n",
169 hash.process(block_buf, bytes);
171 // Sparse file processing: if we read a block of all zeroes, encode
173 bool all_zero = true;
174 for (int i = 0; i < bytes; i++) {
175 if (block_buf[i] != 0) {
181 // Either find a copy of this block in an already-existing segment,
182 // or index it so it can be re-used in the future
183 double block_age = 0.0;
186 SHA1Checksum block_hash;
187 block_hash.process(block_buf, bytes);
188 string block_csum = block_hash.checksum_str();
191 ref = ObjectReference(ObjectReference::REF_ZERO);
192 ref.set_range(0, bytes);
194 ref = db->FindObject(block_csum, bytes);
197 // Store a copy of the object if one does not yet exist
199 LbsObject *o = new LbsObject;
202 /* We might still have seen this checksum before, if the object
203 * was stored at some time in the past, but we have decided to
204 * clean the segment the object was originally stored in
205 * (FindObject will not return such objects). When rewriting
206 * the object contents, put it in a separate group, so that old
207 * objects get grouped together. The hope is that these old
208 * objects will continue to be used in the future, and we
209 * obtain segments which will continue to be well-utilized.
210 * Additionally, keep track of the age of the data by looking
211 * up the age of the block which was expired and using that
212 * instead of the current time. */
213 if (db->IsOldObject(block_csum, bytes,
214 &block_age, &object_group)) {
215 if (object_group == 0) {
216 o->set_group("data");
219 sprintf(group, "compacted-%d", object_group);
225 o->set_group("data");
229 o->set_data(block_buf, bytes);
232 db->StoreObject(ref, block_csum, bytes, block_age);
233 ref.set_range(0, bytes);
237 object_list.push_back(ref.to_string());
239 add_segment(ref.get_segment());
247 file_info["checksum"] = hash.checksum_str();
251 printf(" [%s]\n", status);
253 string blocklist = "";
254 for (list<string>::iterator i = object_list.begin();
255 i != object_list.end(); ++i) {
256 if (i != object_list.begin())
260 file_info["data"] = blocklist;
265 /* Dump a specified filesystem object (file, directory, etc.) based on its
266 * inode information. If the object is a regular file, an open filehandle is
268 void dump_inode(const string& path, // Path within snapshot
269 const string& fullpath, // Path to object in filesystem
270 struct stat& stat_buf, // Results of stat() call
271 int fd) // Open filehandle if regular file
274 dictionary file_info;
278 printf("%s\n", path.c_str());
279 metawriter->find(path);
281 file_info["name"] = uri_encode(path);
282 file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8);
283 file_info["ctime"] = encode_int(stat_buf.st_ctime);
284 file_info["mtime"] = encode_int(stat_buf.st_mtime);
285 file_info["user"] = encode_int(stat_buf.st_uid);
286 file_info["group"] = encode_int(stat_buf.st_gid);
288 time_t now = time(NULL);
289 if (now - stat_buf.st_ctime < 30 || now - stat_buf.st_mtime < 30)
290 if ((stat_buf.st_mode & S_IFMT) != S_IFDIR)
291 file_info["volatile"] = "1";
293 struct passwd *pwd = getpwuid(stat_buf.st_uid);
295 file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
298 struct group *grp = getgrgid(stat_buf.st_gid);
300 file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
303 if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) {
304 file_info["links"] = encode_int(stat_buf.st_nlink);
307 file_info["inode"] = encode_int(major(stat_buf.st_dev))
308 + "/" + encode_int(minor(stat_buf.st_dev))
309 + "/" + encode_int(stat_buf.st_ino);
313 switch (stat_buf.st_mode & S_IFMT) {
322 inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c';
323 file_info["device"] = encode_int(major(stat_buf.st_rdev))
324 + "/" + encode_int(minor(stat_buf.st_rdev));
329 /* Use the reported file size to allocate a buffer large enough to read
330 * the symlink. Allocate slightly more space, so that we ask for more
331 * bytes than we expect and so check for truncation. */
332 buf = new char[stat_buf.st_size + 2];
333 len = readlink(fullpath.c_str(), buf, stat_buf.st_size + 1);
335 fprintf(stderr, "error reading symlink: %m\n");
336 } else if (len <= stat_buf.st_size) {
338 file_info["target"] = uri_encode(buf);
339 } else if (len > stat_buf.st_size) {
340 fprintf(stderr, "error reading symlink: name truncated\n");
348 file_size = dumpfile(fd, file_info, path, stat_buf);
349 file_info["size"] = encode_int(file_size);
352 return; // error occurred; do not dump file
354 if (file_size != stat_buf.st_size) {
355 fprintf(stderr, "Warning: Size of %s changed during reading\n",
357 file_info["volatile"] = "1";
366 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
370 file_info["type"] = string(1, inode_type);
372 metawriter->add(file_info);
375 void scanfile(const string& path, bool include)
379 struct stat stat_buf;
386 true_path = "/" + path;
388 // Set to true if we should scan through the contents of this directory,
389 // but not actually back files up
390 bool scan_only = false;
392 // Check this file against the include/exclude list to see if it should be
394 for (list<string>::iterator i = includes.begin();
395 i != includes.end(); ++i) {
397 printf("Including %s\n", path.c_str());
402 for (list<string>::iterator i = excludes.begin();
403 i != excludes.end(); ++i) {
405 printf("Excluding %s\n", path.c_str());
410 for (list<string>::iterator i = searches.begin();
411 i != searches.end(); ++i) {
413 printf("Scanning %s\n", path.c_str());
418 if (!include && !scan_only)
421 if (lstat(true_path.c_str(), &stat_buf) < 0) {
422 fprintf(stderr, "lstat(%s): %m\n", path.c_str());
426 if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
427 /* Be paranoid when opening the file. We have no guarantee that the
428 * file was not replaced between the stat() call above and the open()
429 * call below, so we might not even be opening a regular file. We
430 * supply flags to open to to guard against various conditions before
431 * we can perform an lstat to check that the file is still a regular
433 * - O_NOFOLLOW: in the event the file was replaced by a symlink
434 * - O_NONBLOCK: prevents open() from blocking if the file was
436 * We also add in O_NOATIME, since this may reduce disk writes (for
437 * inode updates). However, O_NOATIME may result in EPERM, so if the
438 * initial open fails, try again without O_NOATIME. */
439 fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
441 fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK);
444 fprintf(stderr, "Unable to open file %s: %m\n", path.c_str());
448 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
450 flags = fcntl(fd, F_GETFL);
451 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
453 /* Perform the stat call again, and check that we still have a regular
455 if (fstat(fd, &stat_buf) < 0) {
456 fprintf(stderr, "fstat: %m\n");
461 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
462 fprintf(stderr, "file is no longer a regular file!\n");
468 dump_inode(path, true_path, stat_buf, fd);
473 // If we hit a directory, now that we've written the directory itself,
474 // recursively scan the directory.
475 if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) {
476 DIR *dir = opendir(true_path.c_str());
479 fprintf(stderr, "Error: %m\n");
484 vector<string> contents;
485 while ((ent = readdir(dir)) != NULL) {
486 string filename(ent->d_name);
487 if (filename == "." || filename == "..")
489 contents.push_back(filename);
494 sort(contents.begin(), contents.end());
496 for (vector<string>::iterator i = contents.begin();
497 i != contents.end(); ++i) {
498 const string& filename = *i;
500 scanfile(filename, include);
502 scanfile(path + "/" + filename, include);
507 /* Include the specified file path in the backups. Append the path to the
508 * includes list, and to ensure that we actually see the path when scanning the
509 * directory tree, add all the parent directories to the search list, which
510 * means we will scan through the directory listing even if the files
511 * themselves are excluded from being backed up. */
512 void add_include(const char *path)
514 printf("Add: %s\n", path);
515 /* Was an absolute path specified? If so, we'll need to start scanning
516 * from the root directory. Make sure that the user was consistent in
517 * providing either all relative paths or all absolute paths. */
518 if (path[0] == '/') {
519 if (includes.size() > 0 && relative_paths == true) {
521 "Error: Cannot mix relative and absolute paths!\n");
525 relative_paths = false;
527 // Skip over leading '/'
529 } else if (relative_paths == false && path[0] != '/') {
530 fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n");
534 includes.push_back(path);
536 /* Split the specified path into directory components, and ensure that we
537 * descend into all the directories along the path. */
538 const char *slash = path;
543 while ((slash = strchr(slash + 1, '/')) != NULL) {
544 string component(path, slash - path);
545 searches.push_back(component);
549 void usage(const char *program)
554 "Usage: %s [OPTION]... --dest=DEST PATHS...\n"
555 "Produce backup snapshot of files in SOURCE and store to DEST.\n"
558 " --dest=PATH path where backup is to be written\n"
559 " --upload-script=COMMAND\n"
560 " program to invoke for each backup file generated\n"
561 " --exclude=PATH exclude files in PATH from snapshot\n"
562 " --localdb=PATH local backup metadata is stored in PATH\n"
563 " --tmpdir=PATH path for temporarily storing backup files\n"
564 " (defaults to TMPDIR environment variable or /tmp)\n"
565 " --filter=COMMAND program through which to filter segment data\n"
566 " (defaults to \"bzip2 -c\")\n"
567 " --filter-extension=EXT\n"
568 " string to append to segment files\n"
569 " (defaults to \".bz2\")\n"
570 " --signature-filter=COMMAND\n"
571 " program though which to filter descriptor\n"
572 " --scheme=NAME optional name for this snapshot\n"
573 " --intent=FLOAT intended backup type: 1=daily, 7=weekly, ...\n"
574 " (defaults to \"1\")\n"
575 " --full-metadata do not re-use metadata from previous backups\n"
577 "Exactly one of --dest or --upload-script must be specified.\n",
582 int main(int argc, char *argv[])
584 string backup_dest = "", backup_script = "";
585 string localdb_dir = "";
586 string backup_scheme = "";
587 string signature_filter = "";
589 string tmp_dir = "/tmp";
590 if (getenv("TMPDIR") != NULL)
591 tmp_dir = getenv("TMPDIR");
594 static struct option long_options[] = {
595 {"localdb", 1, 0, 0}, // 0
596 {"exclude", 1, 0, 0}, // 1
597 {"filter", 1, 0, 0}, // 2
598 {"filter-extension", 1, 0, 0}, // 3
599 {"dest", 1, 0, 0}, // 4
600 {"scheme", 1, 0, 0}, // 5
601 {"signature-filter", 1, 0, 0}, // 6
602 {"intent", 1, 0, 0}, // 7
603 {"full-metadata", 0, 0, 0}, // 8
604 {"tmpdir", 1, 0, 0}, // 9
605 {"upload-script", 1, 0, 0}, // 10
610 int c = getopt_long(argc, argv, "", long_options, &long_index);
616 switch (long_index) {
618 localdb_dir = optarg;
621 if (optarg[0] != '/')
622 excludes.push_back(optarg);
624 excludes.push_back(optarg + 1);
627 filter_program = optarg;
629 case 3: // --filter-extension
630 filter_extension = optarg;
633 backup_dest = optarg;
636 backup_scheme = optarg;
638 case 6: // --signature-filter
639 signature_filter = optarg;
642 snapshot_intent = atof(optarg);
643 if (snapshot_intent <= 0)
646 case 8: // --full-metadata
647 flag_full_metadata = true;
652 case 10: // --upload-script
653 backup_script = optarg;
656 fprintf(stderr, "Unhandled long option!\n");
665 if (optind == argc) {
670 searches.push_back(".");
671 for (int i = optind; i < argc; i++)
672 add_include(argv[i]);
674 if (backup_dest == "" && backup_script == "") {
676 "Error: Backup destination must be specified using --dest= or --upload-script=\n");
681 if (backup_dest != "" && backup_script != "") {
683 "Error: Cannot specify both --dest= and --upload-script=\n");
688 // Default for --localdb is the same as --dest
689 if (localdb_dir == "") {
690 localdb_dir = backup_dest;
692 if (localdb_dir == "") {
694 "Error: Must specify local database path with --localdb=\n");
699 // Dump paths for debugging/informational purposes
701 list<string>::const_iterator i;
703 printf("LBS Version: %s\n", lbs_version);
705 printf("--dest=%s\n--localdb=%s\n--upload-script=%s\n",
706 backup_dest.c_str(), localdb_dir.c_str(), backup_script.c_str());
708 printf("Includes:\n");
709 for (i = includes.begin(); i != includes.end(); ++i)
710 printf(" %s\n", i->c_str());
712 printf("Excludes:\n");
713 for (i = excludes.begin(); i != excludes.end(); ++i)
714 printf(" %s\n", i->c_str());
716 printf("Searching:\n");
717 for (i = searches.begin(); i != searches.end(); ++i)
718 printf(" %s\n", i->c_str());
721 block_buf = new char[LBS_BLOCK_SIZE];
723 /* Initialize the remote storage layer. If using an upload script, create
724 * a temporary directory for staging files. Otherwise, write backups
725 * directly to the destination directory. */
726 if (backup_script != "") {
727 tmp_dir = tmp_dir + "/lbs." + generate_uuid();
728 if (mkdir(tmp_dir.c_str(), 0700) < 0) {
729 fprintf(stderr, "Cannot create temporary directory %s: %m\n",
733 remote = new RemoteStore(tmp_dir);
734 remote->set_script(backup_script);
736 remote = new RemoteStore(backup_dest);
739 /* Store the time when the backup started, so it can be included in the
745 localtime_r(&now, &time_buf);
746 strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
748 /* Open the local database which tracks all objects that are stored
749 * remotely, for efficient incrementals. Provide it with the name of this
751 string database_path = localdb_dir + "/localdb.sqlite";
753 db->Open(database_path.c_str(), desc_buf,
754 backup_scheme.size() ? backup_scheme.c_str() : NULL,
757 tss = new TarSegmentStore(remote, db);
759 /* Initialize the stat cache, for skipping over unchanged files. */
760 metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf,
762 ? backup_scheme.c_str()
765 scanfile(".", false);
767 ObjectReference root_ref = metawriter->close();
768 add_segment(root_ref.get_segment());
769 string backup_root = root_ref.to_string();
777 /* Write out a checksums file which lists the checksums for all the
778 * segments included in this snapshot. The format is designed so that it
779 * may be easily verified using the sha1sums command. */
780 const char csum_type[] = "sha1";
781 string checksum_filename = "snapshot-";
782 if (backup_scheme.size() > 0)
783 checksum_filename += backup_scheme + "-";
784 checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums";
785 RemoteFile *checksum_file = remote->alloc_file(checksum_filename);
786 FILE *checksums = fdopen(checksum_file->get_fd(), "w");
788 for (std::set<string>::iterator i = segment_list.begin();
789 i != segment_list.end(); ++i) {
790 string seg_path, seg_csum;
791 if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) {
792 const char *raw_checksum = NULL;
793 if (strncmp(seg_csum.c_str(), csum_type,
794 strlen(csum_type)) == 0) {
795 raw_checksum = seg_csum.c_str() + strlen(csum_type);
796 if (*raw_checksum == '=')
802 if (raw_checksum != NULL)
803 fprintf(checksums, "%s *%s\n",
804 raw_checksum, seg_path.c_str());
808 checksum_file->send();
812 /* All other files should be flushed to remote storage before writing the
813 * backup descriptor below, so that it is not possible to have a backup
814 * descriptor written out depending on non-existent (not yet written)
818 /* Write a backup descriptor file, which says which segments are needed and
819 * where to start to restore this snapshot. The filename is based on the
820 * current time. If a signature filter program was specified, filter the
821 * data through that to give a chance to sign the descriptor contents. */
822 string desc_filename = "snapshot-";
823 if (backup_scheme.size() > 0)
824 desc_filename += backup_scheme + "-";
825 desc_filename = desc_filename + desc_buf + ".lbs";
827 RemoteFile *descriptor_file = remote->alloc_file(desc_filename);
828 int descriptor_fd = descriptor_file->get_fd();
829 if (descriptor_fd < 0) {
830 fprintf(stderr, "Unable to open descriptor output file: %m\n");
833 pid_t signature_pid = 0;
834 if (signature_filter.size() > 0) {
835 int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(),
837 close(descriptor_fd);
838 descriptor_fd = new_fd;
840 FILE *descriptor = fdopen(descriptor_fd, "w");
842 fprintf(descriptor, "Format: LBS Snapshot v0.6\n");
843 fprintf(descriptor, "Producer: LBS %s\n", lbs_version);
844 strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
845 fprintf(descriptor, "Date: %s\n", desc_buf);
846 if (backup_scheme.size() > 0)
847 fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str());
848 fprintf(descriptor, "Backup-Intent: %g\n", snapshot_intent);
849 fprintf(descriptor, "Root: %s\n", backup_root.c_str());
851 SHA1Checksum checksum_csum;
852 if (checksum_csum.process_file(checksum_filename.c_str())) {
853 string csum = checksum_csum.checksum_str();
854 fprintf(descriptor, "Checksums: %s\n", csum.c_str());
857 fprintf(descriptor, "Segments:\n");
858 for (std::set<string>::iterator i = segment_list.begin();
859 i != segment_list.end(); ++i) {
860 fprintf(descriptor, " %s\n", i->c_str());
867 waitpid(signature_pid, &status, 0);
869 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
870 throw IOException("Signature filter process error");
874 descriptor_file->send();
879 if (backup_script != "") {
880 if (rmdir(tmp_dir.c_str()) < 0) {
882 "Warning: Cannot delete temporary directory %s: %m\n",