1 /* Recursively descend the filesystem and visit each file. */
13 #include <sys/sysmacros.h>
14 #include <sys/types.h>
31 #include "statcache.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 TarSegmentStore *tss = NULL;
48 static MetadataWriter *metawriter = NULL;
50 /* Buffer for holding a single block of data read from a file. */
51 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
52 static char *block_buf;
54 /* Local database, which tracks objects written in this and previous
55 * invocations to help in creating incremental snapshots. */
58 /* Stat cache, which stored data locally to speed the backup process by quickly
59 * skipping files which have not changed. */
62 /* Keep track of all segments which are needed to reconstruct the snapshot. */
63 std::set<string> segment_list;
65 /* Selection of files to include/exclude in the snapshot. */
66 std::list<string> includes; // Paths in which files should be saved
67 std::list<string> excludes; // Paths which will not be saved
68 std::list<string> searches; // Directories we don't want to save, but
69 // do want to descend searching for data
72 bool relative_paths = true;
74 /* Ensure that the given segment is listed as a dependency of the current
76 void add_segment(const string& segment)
78 segment_list.insert(segment);
81 /* Read data from a file descriptor and return the amount of data read. A
82 * short read (less than the requested size) will only occur if end-of-file is
84 ssize_t file_read(int fd, char *buf, size_t maxlen)
86 size_t bytes_read = 0;
89 ssize_t res = read(fd, buf, maxlen);
93 fprintf(stderr, "error reading file: %m\n");
95 } else if (res == 0) {
107 /* Read the contents of a file (specified by an open file descriptor) and copy
108 * the data to the store. Returns the size of the file (number of bytes
109 * dumped), or -1 on error. */
110 int64_t dumpfile(int fd, dictionary &file_info, const string &path,
111 struct stat& stat_buf)
114 list<string> object_list;
115 const char *status = NULL; /* Status indicator printed out */
117 /* Look up this file in the old stat cache, if we can. If the stat
118 * information indicates that the file has not changed, do not bother
119 * re-reading the entire contents. */
122 if (statcache->Find(path, &stat_buf)) {
124 const list<ObjectReference> &blocks = statcache->get_blocks();
126 /* If any of the blocks in the object have been expired, then we should
127 * fall back to fully reading in the file. */
128 for (list<ObjectReference>::const_iterator i = blocks.begin();
129 i != blocks.end(); ++i) {
130 const ObjectReference &ref = *i;
131 if (!db->IsAvailable(ref)) {
138 /* If everything looks okay, use the cached information */
140 file_info["checksum"] = statcache->get_checksum();
141 for (list<ObjectReference>::const_iterator i = blocks.begin();
142 i != blocks.end(); ++i) {
143 const ObjectReference &ref = *i;
144 object_list.push_back(ref.to_string());
145 segment_list.insert(ref.get_segment());
148 size = stat_buf.st_size;
152 /* If the file is new or changed, we must read in the contents a block at a
157 ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
161 fprintf(stderr, "Backup contents for %s may be incorrect\n",
166 hash.process(block_buf, bytes);
168 // Either find a copy of this block in an already-existing segment,
169 // or index it so it can be re-used in the future
170 double block_age = 0.0;
171 SHA1Checksum block_hash;
172 block_hash.process(block_buf, bytes);
173 string block_csum = block_hash.checksum_str();
174 ObjectReference ref = db->FindObject(block_csum, bytes);
176 // Store a copy of the object if one does not yet exist
177 if (ref.get_segment().size() == 0) {
178 LbsObject *o = new LbsObject;
181 /* We might still have seen this checksum before, if the object
182 * was stored at some time in the past, but we have decided to
183 * clean the segment the object was originally stored in
184 * (FindObject will not return such objects). When rewriting
185 * the object contents, put it in a separate group, so that old
186 * objects get grouped together. The hope is that these old
187 * objects will continue to be used in the future, and we
188 * obtain segments which will continue to be well-utilized.
189 * Additionally, keep track of the age of the data by looking
190 * up the age of the block which was expired and using that
191 * instead of the current time. */
192 if (db->IsOldObject(block_csum, bytes,
193 &block_age, &object_group)) {
194 if (object_group == 0) {
195 o->set_group("data");
198 sprintf(group, "compacted-%d", object_group);
204 o->set_group("data");
208 o->set_data(block_buf, bytes);
211 db->StoreObject(ref, block_csum, bytes, block_age);
215 object_list.push_back(ref.to_string());
216 segment_list.insert(ref.get_segment());
224 file_info["checksum"] = hash.checksum_str();
228 printf(" [%s]\n", status);
230 statcache->Save(path, &stat_buf, file_info["checksum"], object_list);
232 string blocklist = "";
233 for (list<string>::iterator i = object_list.begin();
234 i != object_list.end(); ++i) {
235 if (i != object_list.begin())
239 file_info["data"] = blocklist;
244 /* Dump a specified filesystem object (file, directory, etc.) based on its
245 * inode information. If the object is a regular file, an open filehandle is
247 void dump_inode(const string& path, // Path within snapshot
248 const string& fullpath, // Path to object in filesystem
249 struct stat& stat_buf, // Results of stat() call
250 int fd) // Open filehandle if regular file
253 dictionary file_info;
257 printf("%s\n", path.c_str());
259 if (metawriter->find(path)) {
260 ObjectReference *r = metawriter->old_ref();
262 string s = r->to_string();
263 printf(" cached at %s\n", s.c_str());
268 file_info["path"] = uri_encode(path);
269 file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8);
270 file_info["ctime"] = encode_int(stat_buf.st_ctime);
271 file_info["mtime"] = encode_int(stat_buf.st_mtime);
272 file_info["user"] = encode_int(stat_buf.st_uid);
273 file_info["group"] = encode_int(stat_buf.st_gid);
275 struct passwd *pwd = getpwuid(stat_buf.st_uid);
277 file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
280 struct group *grp = getgrgid(stat_buf.st_gid);
282 file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
285 if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) {
286 file_info["links"] = encode_int(stat_buf.st_nlink);
289 file_info["inode"] = encode_int(major(stat_buf.st_dev))
290 + "/" + encode_int(minor(stat_buf.st_dev))
291 + "/" + encode_int(stat_buf.st_ino);
295 switch (stat_buf.st_mode & S_IFMT) {
304 inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c';
305 file_info["device"] = encode_int(major(stat_buf.st_rdev))
306 + "/" + encode_int(minor(stat_buf.st_rdev));
311 /* Use the reported file size to allocate a buffer large enough to read
312 * the symlink. Allocate slightly more space, so that we ask for more
313 * bytes than we expect and so check for truncation. */
314 buf = new char[stat_buf.st_size + 2];
315 len = readlink(fullpath.c_str(), buf, stat_buf.st_size + 1);
317 fprintf(stderr, "error reading symlink: %m\n");
318 } else if (len <= stat_buf.st_size) {
320 file_info["target"] = uri_encode(buf);
321 } else if (len > stat_buf.st_size) {
322 fprintf(stderr, "error reading symlink: name truncated\n");
330 file_size = dumpfile(fd, file_info, path, stat_buf);
331 file_info["size"] = encode_int(file_size);
334 return; // error occurred; do not dump file
336 if (file_size != stat_buf.st_size) {
337 fprintf(stderr, "Warning: Size of %s changed during reading\n",
347 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
351 file_info["type"] = string(1, inode_type);
353 metawriter->add(file_info);
356 void scanfile(const string& path, bool include)
360 struct stat stat_buf;
367 true_path = "/" + path;
369 // Set to true if we should scan through the contents of this directory,
370 // but not actually back files up
371 bool scan_only = false;
373 // Check this file against the include/exclude list to see if it should be
375 for (list<string>::iterator i = includes.begin();
376 i != includes.end(); ++i) {
378 printf("Including %s\n", path.c_str());
383 for (list<string>::iterator i = excludes.begin();
384 i != excludes.end(); ++i) {
386 printf("Excluding %s\n", path.c_str());
391 for (list<string>::iterator i = searches.begin();
392 i != searches.end(); ++i) {
394 printf("Scanning %s\n", path.c_str());
399 if (!include && !scan_only)
402 if (lstat(true_path.c_str(), &stat_buf) < 0) {
403 fprintf(stderr, "lstat(%s): %m\n", path.c_str());
407 if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
408 /* Be paranoid when opening the file. We have no guarantee that the
409 * file was not replaced between the stat() call above and the open()
410 * call below, so we might not even be opening a regular file. We
411 * supply flags to open to to guard against various conditions before
412 * we can perform an lstat to check that the file is still a regular
414 * - O_NOFOLLOW: in the event the file was replaced by a symlink
415 * - O_NONBLOCK: prevents open() from blocking if the file was
417 * We also add in O_NOATIME, since this may reduce disk writes (for
418 * inode updates). However, O_NOATIME may result in EPERM, so if the
419 * initial open fails, try again without O_NOATIME. */
420 fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
422 fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK);
425 fprintf(stderr, "Unable to open file %s: %m\n", path.c_str());
429 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
431 flags = fcntl(fd, F_GETFL);
432 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
434 /* Perform the stat call again, and check that we still have a regular
436 if (fstat(fd, &stat_buf) < 0) {
437 fprintf(stderr, "fstat: %m\n");
442 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
443 fprintf(stderr, "file is no longer a regular file!\n");
449 dump_inode(path, true_path, stat_buf, fd);
454 // If we hit a directory, now that we've written the directory itself,
455 // recursively scan the directory.
456 if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) {
457 DIR *dir = opendir(true_path.c_str());
460 fprintf(stderr, "Error: %m\n");
465 vector<string> contents;
466 while ((ent = readdir(dir)) != NULL) {
467 string filename(ent->d_name);
468 if (filename == "." || filename == "..")
470 contents.push_back(filename);
475 sort(contents.begin(), contents.end());
477 for (vector<string>::iterator i = contents.begin();
478 i != contents.end(); ++i) {
479 const string& filename = *i;
481 scanfile(filename, include);
483 scanfile(path + "/" + filename, include);
488 /* Include the specified file path in the backups. Append the path to the
489 * includes list, and to ensure that we actually see the path when scanning the
490 * directory tree, add all the parent directories to the search list, which
491 * means we will scan through the directory listing even if the files
492 * themselves are excluded from being backed up. */
493 void add_include(const char *path)
495 printf("Add: %s\n", path);
496 /* Was an absolute path specified? If so, we'll need to start scanning
497 * from the root directory. Make sure that the user was consistent in
498 * providing either all relative paths or all absolute paths. */
499 if (path[0] == '/') {
500 if (includes.size() > 0 && relative_paths == true) {
502 "Error: Cannot mix relative and absolute paths!\n");
506 relative_paths = false;
508 // Skip over leading '/'
510 } else if (relative_paths == false && path[0] != '/') {
511 fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n");
515 includes.push_back(path);
517 /* Split the specified path into directory components, and ensure that we
518 * descend into all the directories along the path. */
519 const char *slash = path;
524 while ((slash = strchr(slash + 1, '/')) != NULL) {
525 string component(path, slash - path);
526 searches.push_back(component);
530 void usage(const char *program)
535 "Usage: %s [OPTION]... --dest=DEST PATHS...\n"
536 "Produce backup snapshot of files in SOURCE and store to DEST.\n"
539 " --dest=PATH path where backup is to be written [REQUIRED]\n"
540 " --exclude=PATH exclude files in PATH from snapshot\n"
541 " --localdb=PATH local backup metadata is stored in PATH\n"
542 " --filter=COMMAND program through which to filter segment data\n"
543 " (defaults to \"bzip2 -c\")\n"
544 " --filter-extension=EXT\n"
545 " string to append to segment files\n"
546 " (defaults to \".bz2\")\n"
547 " --signature-filter=COMMAND\n"
548 " program though which to filter descriptor\n"
549 " --scheme=NAME optional name for this snapshot\n",
554 int main(int argc, char *argv[])
556 string backup_dest = "";
557 string localdb_dir = "";
558 string backup_scheme = "";
559 string signature_filter = "";
562 static struct option long_options[] = {
563 {"localdb", 1, 0, 0}, // 0
564 {"exclude", 1, 0, 0}, // 1
565 {"filter", 1, 0, 0}, // 2
566 {"filter-extension", 1, 0, 0}, // 3
567 {"dest", 1, 0, 0}, // 4
568 {"scheme", 1, 0, 0}, // 5
569 {"signature-filter", 1, 0, 0}, // 6
574 int c = getopt_long(argc, argv, "", long_options, &long_index);
580 switch (long_index) {
582 localdb_dir = optarg;
585 if (optarg[0] != '/')
586 excludes.push_back(optarg);
588 excludes.push_back(optarg + 1);
591 filter_program = optarg;
593 case 3: // --filter-extension
594 filter_extension = optarg;
597 backup_dest = optarg;
600 backup_scheme = optarg;
602 case 6: // --signature-filter
603 signature_filter = optarg;
606 fprintf(stderr, "Unhandled long option!\n");
615 if (optind == argc) {
620 searches.push_back(".");
621 for (int i = optind; i < argc; i++)
622 add_include(argv[i]);
624 if (backup_dest == "") {
626 "Error: Backup destination must be specified with --dest=\n");
631 // Default for --localdb is the same as --dest
632 if (localdb_dir == "") {
633 localdb_dir = backup_dest;
636 // Dump paths for debugging/informational purposes
638 list<string>::const_iterator i;
640 printf("LBS Version: %s\n", lbs_version);
642 printf("--dest=%s\n--localdb=%s\n\n",
643 backup_dest.c_str(), localdb_dir.c_str());
645 printf("Includes:\n");
646 for (i = includes.begin(); i != includes.end(); ++i)
647 printf(" %s\n", i->c_str());
649 printf("Excludes:\n");
650 for (i = excludes.begin(); i != excludes.end(); ++i)
651 printf(" %s\n", i->c_str());
653 printf("Searching:\n");
654 for (i = searches.begin(); i != searches.end(); ++i)
655 printf(" %s\n", i->c_str());
658 block_buf = new char[LBS_BLOCK_SIZE];
660 /* Store the time when the backup started, so it can be included in the
666 localtime_r(&now, &time_buf);
667 strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
669 /* Open the local database which tracks all objects that are stored
670 * remotely, for efficient incrementals. Provide it with the name of this
672 string database_path = localdb_dir + "/localdb.sqlite";
674 db->Open(database_path.c_str(), desc_buf,
675 backup_scheme.size() ? backup_scheme.c_str() : NULL);
677 tss = new TarSegmentStore(backup_dest, db);
679 /* Initialize the stat cache, for skipping over unchanged files. */
680 statcache = new StatCache;
681 statcache->Open(localdb_dir.c_str(), desc_buf,
682 backup_scheme.size() ? backup_scheme.c_str() : NULL);
684 metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf,
686 ? backup_scheme.c_str()
689 scanfile(".", false);
691 ObjectReference root_ref = metawriter->close();
692 add_segment(root_ref.get_segment());
693 string backup_root = root_ref.to_string();
704 /* Write out a checksums file which lists the checksums for all the
705 * segments included in this snapshot. The format is designed so that it
706 * may be easily verified using the sha1sums command. */
707 const char csum_type[] = "sha1";
708 string checksum_filename = backup_dest + "/snapshot-";
709 if (backup_scheme.size() > 0)
710 checksum_filename += backup_scheme + "-";
711 checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums";
712 FILE *checksums = fopen(checksum_filename.c_str(), "w");
713 if (checksums != NULL) {
714 for (std::set<string>::iterator i = segment_list.begin();
715 i != segment_list.end(); ++i) {
716 string seg_path, seg_csum;
717 if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) {
718 const char *raw_checksum = NULL;
719 if (strncmp(seg_csum.c_str(), csum_type,
720 strlen(csum_type)) == 0) {
721 raw_checksum = seg_csum.c_str() + strlen(csum_type);
722 if (*raw_checksum == '=')
728 if (raw_checksum != NULL)
729 fprintf(checksums, "%s *%s\n",
730 raw_checksum, seg_path.c_str());
735 fprintf(stderr, "ERROR: Unable to write checksums file: %m\n");
740 /* Write a backup descriptor file, which says which segments are needed and
741 * where to start to restore this snapshot. The filename is based on the
742 * current time. If a signature filter program was specified, filter the
743 * data through that to give a chance to sign the descriptor contents. */
744 string desc_filename = backup_dest + "/snapshot-";
745 if (backup_scheme.size() > 0)
746 desc_filename += backup_scheme + "-";
747 desc_filename = desc_filename + desc_buf + ".lbs";
749 int descriptor_fd = open(desc_filename.c_str(), O_WRONLY | O_CREAT, 0666);
750 if (descriptor_fd < 0) {
751 fprintf(stderr, "Unable to open descriptor output file: %m\n");
754 pid_t signature_pid = 0;
755 if (signature_filter.size() > 0) {
756 int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(),
758 close(descriptor_fd);
759 descriptor_fd = new_fd;
761 FILE *descriptor = fdopen(descriptor_fd, "w");
763 fprintf(descriptor, "Format: LBS Snapshot v0.6\n");
764 fprintf(descriptor, "Producer: LBS %s\n", lbs_version);
765 strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
766 fprintf(descriptor, "Date: %s\n", desc_buf);
767 if (backup_scheme.size() > 0)
768 fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str());
769 fprintf(descriptor, "Root: %s\n", backup_root.c_str());
771 SHA1Checksum checksum_csum;
772 if (checksum_csum.process_file(checksum_filename.c_str())) {
773 string csum = checksum_csum.checksum_str();
774 fprintf(descriptor, "Checksums: %s\n", csum.c_str());
777 fprintf(descriptor, "Segments:\n");
778 for (std::set<string>::iterator i = segment_list.begin();
779 i != segment_list.end(); ++i) {
780 fprintf(descriptor, " %s\n", i->c_str());
787 waitpid(signature_pid, &status, 0);
789 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
790 throw IOException("Signature filter process error");