X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=9075e51b6bc2a26f4acf359e624e930a03c38554;hb=1048a6510710acb5bef352d0acb2dd13ecce0e01;hp=b96b1d308e98859ffc7a88398f9bf478151bc722;hpb=4b89157cdec2da9a016a7ea0f511a0f3ff43f6dc;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index b96b1d3..9075e51 100644 --- a/scandir.cc +++ b/scandir.cc @@ -10,6 +10,7 @@ #include #include #include +#include #include #include @@ -22,17 +23,25 @@ #include #include -#include "format.h" #include "localdb.h" #include "store.h" #include "sha1.h" #include "statcache.h" +#include "util.h" using std::list; using std::string; using std::vector; using std::ostream; +/* Version information. This will be filled in by the Makefile. */ +#ifndef LBS_VERSION +#define LBS_VERSION Unknown +#endif +#define LBS_STRINGIFY(s) LBS_STRINGIFY2(s) +#define LBS_STRINGIFY2(s) #s +static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); + static TarSegmentStore *tss = NULL; /* Buffer for holding a single block of data read from a file. */ @@ -59,10 +68,14 @@ std::ostringstream metadata; /* Keep track of all segments which are needed to reconstruct the snapshot. */ std::set segment_list; -void scandir(const string& path); - /* Selection of files to include/exclude in the snapshot. */ -std::list excludes; +std::list includes; // Paths in which files should be saved +std::list excludes; // Paths 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() @@ -91,7 +104,7 @@ void metadata_flush() /* Read data from a file descriptor and return the amount of data read. A * short read (less than the requested size) will only occur if end-of-file is * hit. */ -size_t file_read(int fd, char *buf, size_t maxlen) +ssize_t file_read(int fd, char *buf, size_t maxlen) { size_t bytes_read = 0; @@ -100,7 +113,8 @@ size_t file_read(int fd, char *buf, size_t maxlen) if (res < 0) { if (errno == EINTR) continue; - throw IOException("file_read: error reading"); + fprintf(stderr, "error reading file: %m\n"); + return -1; } else if (res == 0) { break; } else { @@ -116,17 +130,12 @@ size_t file_read(int fd, char *buf, size_t maxlen) /* Read the contents of a file (specified by an open file descriptor) and copy * the data to the store. Returns the size of the file (number of bytes * dumped), or -1 on error. */ -int64_t dumpfile(int fd, dictionary &file_info, const string &path) +int64_t dumpfile(int fd, dictionary &file_info, const string &path, + struct stat& stat_buf) { - struct stat stat_buf; - fstat(fd, &stat_buf); int64_t size = 0; list object_list; - - if ((stat_buf.st_mode & S_IFMT) != S_IFREG) { - fprintf(stderr, "file is no longer a regular file!\n"); - return -1; - } + 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 @@ -144,6 +153,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path) const ObjectReference &ref = *i; if (!db->IsAvailable(ref)) { cached = false; + status = "repack"; break; } } @@ -165,13 +175,16 @@ 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; while (true) { - size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); + ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); if (bytes == 0) break; + if (bytes < 0) { + fprintf(stderr, "Backup contents for %s may be incorrect\n", + path.c_str()); + break; + } hash.process(block_buf, bytes); @@ -198,10 +211,14 @@ 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)) + if (db->IsOldObject(block_csum, bytes, &block_age)) { o->set_group("compacted"); - else + if (status == NULL) + status = "partial"; + } else { o->set_group("data"); + status = "new"; + } o->set_data(block_buf, bytes); o->write(tss); @@ -214,11 +231,17 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path) segment_list.insert(ref.get_segment()); db->UseObject(ref); size += bytes; + + if (status == NULL) + status = "old"; } file_info["checksum"] = hash.checksum_str(); } + if (status != NULL) + printf(" [%s]\n", status); + statcache->Save(path, &stat_buf, file_info["checksum"], object_list); /* For files that only need to be broken apart into a few objects, store @@ -252,36 +275,22 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path) return size; } -void scanfile(const string& path) +/* Dump a specified filesystem object (file, directory, etc.) based on its + * inode information. If the object is a regular file, an open filehandle is + * provided. */ +void dump_inode(const string& path, // Path within snapshot + const string& fullpath, // Path to object in filesystem + struct stat& stat_buf, // Results of stat() call + int fd) // Open filehandle if regular file { - int fd; - long flags; - struct stat stat_buf; char *buf; - ssize_t len; - int64_t file_size; - list refs; - - // Set to true if the item is a directory and we should recursively scan - bool recurse = false; - - // Check this file against the include/exclude list to see if it should be - // considered - for (list::iterator i = excludes.begin(); - i != excludes.end(); ++i) { - if (path == *i) { - printf("Excluding %s\n", path.c_str()); - return; - } - } - dictionary file_info; - - lstat(path.c_str(), &stat_buf); + int64_t file_size; + ssize_t len; printf("%s\n", path.c_str()); - file_info["mode"] = encode_int(stat_buf.st_mode & 07777); + file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8); 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); @@ -296,6 +305,13 @@ void scanfile(const string& path) 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) { @@ -305,11 +321,11 @@ void scanfile(const string& path) case S_IFSOCK: inode_type = 's'; break; - case S_IFCHR: - inode_type = 'c'; - break; case S_IFBLK: - inode_type = 'b'; + case S_IFCHR: + inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c'; + file_info["device"] = encode_int(major(stat_buf.st_rdev)) + + "/" + encode_int(minor(stat_buf.st_rdev)); break; case S_IFLNK: inode_type = 'l'; @@ -318,7 +334,7 @@ void scanfile(const string& path) * the symlink. Allocate slightly more space, so that we ask for more * bytes than we expect and so check for truncation. */ buf = new char[stat_buf.st_size + 2]; - len = readlink(path.c_str(), buf, stat_buf.st_size + 1); + len = readlink(fullpath.c_str(), buf, stat_buf.st_size + 1); if (len < 0) { fprintf(stderr, "error reading symlink: %m\n"); } else if (len <= stat_buf.st_size) { @@ -333,35 +349,8 @@ void scanfile(const string& path) case S_IFREG: inode_type = '-'; - /* Be paranoid when opening the file. We have no guarantee that the - * file was not replaced between the stat() call above and the open() - * call below, so we might not even be opening a regular file. That - * the file descriptor refers to a regular file is checked in - * dumpfile(). But we also supply flags to open to to guard against - * various conditions before we can perform that verification: - * - O_NOFOLLOW: in the event the file was replaced by a symlink - * - O_NONBLOCK: prevents open() from blocking if the file was - * replaced by a fifo - * We also add in O_NOATIME, since this may reduce disk writes (for - * inode updates). However, O_NOATIME may result in EPERM, so if the - * initial open fails, try again without O_NOATIME. */ - fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK); - if (fd < 0) { - fd = open(path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK); - } - if (fd < 0) { - fprintf(stderr, "Unable to open file %s: %m\n", path.c_str()); - return; - } - - /* Drop the use of the O_NONBLOCK flag; we only wanted that for file - * open. */ - flags = fcntl(fd, F_GETFL); - fcntl(fd, F_SETFL, flags & ~O_NONBLOCK); - - file_size = dumpfile(fd, file_info, path); + 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 @@ -374,7 +363,6 @@ void scanfile(const string& path) break; case S_IFDIR: inode_type = 'd'; - recurse = true; break; default: @@ -391,62 +379,208 @@ void scanfile(const string& path) // Break apart metadata listing if it becomes too large. if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE) metadata_flush(); - - // If we hit a directory, now that we've written the directory itself, - // recursively scan the directory. - if (recurse) - scandir(path); } -void scandir(const string& path) +void scanfile(const string& path, bool include) { - DIR *dir = opendir(path.c_str()); + int fd = -1; + long flags; + struct stat stat_buf; + list refs; + + string true_path; + if (relative_paths) + true_path = path; + else + true_path = "/" + path; + + // Set to true if we should scan through the contents of this directory, + // but not actually back files up + bool scan_only = false; + + // Check this file against the include/exclude list to see if it should be + // considered + for (list::iterator i = includes.begin(); + i != includes.end(); ++i) { + if (path == *i) { + printf("Including %s\n", path.c_str()); + include = true; + } + } + + for (list::iterator i = excludes.begin(); + i != excludes.end(); ++i) { + if (path == *i) { + printf("Excluding %s\n", path.c_str()); + 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; + } + } + + if (!include && !scan_only) + return; - if (dir == NULL) { - fprintf(stderr, "Error: %m\n"); + if (lstat(true_path.c_str(), &stat_buf) < 0) { + fprintf(stderr, "lstat(%s): %m\n", path.c_str()); return; } - struct dirent *ent; - vector contents; - while ((ent = readdir(dir)) != NULL) { - string filename(ent->d_name); - if (filename == "." || filename == "..") - continue; - contents.push_back(filename); + if ((stat_buf.st_mode & S_IFMT) == S_IFREG) { + /* Be paranoid when opening the file. We have no guarantee that the + * file was not replaced between the stat() call above and the open() + * call below, so we might not even be opening a regular file. We + * supply flags to open to to guard against various conditions before + * we can perform an lstat to check that the file is still a regular + * file: + * - O_NOFOLLOW: in the event the file was replaced by a symlink + * - O_NONBLOCK: prevents open() from blocking if the file was + * replaced by a fifo + * We also add in O_NOATIME, since this may reduce disk writes (for + * inode updates). However, O_NOATIME may result in EPERM, so if the + * initial open fails, try again without O_NOATIME. */ + fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK); + if (fd < 0) { + fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK); + } + if (fd < 0) { + fprintf(stderr, "Unable to open file %s: %m\n", path.c_str()); + return; + } + + /* Drop the use of the O_NONBLOCK flag; we only wanted that for file + * open. */ + flags = fcntl(fd, F_GETFL); + fcntl(fd, F_SETFL, flags & ~O_NONBLOCK); + + /* Perform the stat call again, and check that we still have a regular + * file. */ + if (fstat(fd, &stat_buf) < 0) { + fprintf(stderr, "fstat: %m\n"); + close(fd); + return; + } + + if ((stat_buf.st_mode & S_IFMT) != S_IFREG) { + fprintf(stderr, "file is no longer a regular file!\n"); + close(fd); + return; + } } - sort(contents.begin(), contents.end()); + dump_inode(path, true_path, stat_buf, fd); + + if (fd >= 0) + close(fd); + + // If we hit a directory, now that we've written the directory itself, + // recursively scan the directory. + if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) { + DIR *dir = opendir(true_path.c_str()); + + if (dir == NULL) { + fprintf(stderr, "Error: %m\n"); + return; + } + + struct dirent *ent; + vector contents; + while ((ent = readdir(dir)) != NULL) { + string filename(ent->d_name); + if (filename == "." || filename == "..") + continue; + contents.push_back(filename); + } + + closedir(dir); - for (vector::iterator i = contents.begin(); - i != contents.end(); ++i) { - const string& filename = *i; - if (path == ".") - scanfile(filename); - else - scanfile(path + "/" + filename); + sort(contents.begin(), contents.end()); + + for (vector::iterator i = contents.begin(); + i != contents.end(); ++i) { + const string& filename = *i; + if (path == ".") + scanfile(filename, include); + else + scanfile(path + "/" + filename, include); + } + } +} + +/* Include the specified file path in the backups. Append the path to the + * includes list, and to ensure that we actually see the path when scanning the + * directory tree, add all the parent directories to the search list, which + * means we will scan through the directory listing even if the files + * 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. */ + if (path[0] == '/') { + if (includes.size() > 0 && relative_paths == true) { + fprintf(stderr, + "Error: Cannot mix relative and absolute paths!\n"); + exit(1); + } + + relative_paths = false; + + // Skip over leading '/' + path++; + } else if (relative_paths == false && path[0] != '/') { + fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n"); + exit(1); } - closedir(dir); + includes.push_back(path); + + /* Split the specified path into directory components, and ensure that we + * descend into all the directories along the path. */ + const char *slash = path; + + if (path[0] == '\0') + return; + + while ((slash = strchr(slash + 1, '/')) != NULL) { + string component(path, slash - path); + searches.push_back(component); + } } void usage(const char *program) { - fprintf(stderr, - "Usage: %s [OPTION]... SOURCE DEST\n" - "Produce backup snapshot of files in SOURCE and store to DEST.\n" - "\n" - "Options:\n" - " --exclude=PATH exclude files in PATH from snapshot\n" - " --localdb=PATH local backup metadata is stored in PATH\n", - program); + fprintf( + stderr, + "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" + " --exclude=PATH exclude files in PATH from snapshot\n" + " --localdb=PATH local backup metadata is stored in PATH\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 + ); } int main(int argc, char *argv[]) { - string backup_source = "."; - string backup_dest = "."; + string backup_dest = ""; string localdb_dir = ""; + string backup_scheme = ""; while (1) { static struct option long_options[] = { @@ -454,6 +588,8 @@ int main(int argc, char *argv[]) {"exclude", 1, 0, 0}, // 1 {"filter", 1, 0, 0}, // 2 {"filter-extension", 1, 0, 0}, // 3 + {"dest", 1, 0, 0}, // 4 + {"scheme", 1, 0, 0}, // 5 {NULL, 0, 0, 0}, }; @@ -469,7 +605,10 @@ int main(int argc, char *argv[]) localdb_dir = optarg; break; case 1: // --exclude - excludes.push_back(optarg); + if (optarg[0] != '/') + excludes.push_back(optarg); + else + excludes.push_back(optarg + 1); break; case 2: // --filter filter_program = optarg; @@ -477,6 +616,12 @@ int main(int argc, char *argv[]) case 3: // --filter-extension filter_extension = optarg; break; + case 4: // --dest + backup_dest = optarg; + break; + case 5: // --scheme + backup_scheme = optarg; + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; @@ -487,22 +632,49 @@ int main(int argc, char *argv[]) } } - if (argc < optind + 2) { + if (optind == argc) { usage(argv[0]); return 1; } - backup_source = argv[optind]; - backup_dest = argv[argc - 1]; + searches.push_back("."); + for (int i = optind; i < argc; i++) + add_include(argv[i]); + + if (backup_dest == "") { + fprintf(stderr, + "Error: Backup destination must be specified with --dest=\n"); + usage(argv[0]); + return 1; + } + // Default for --localdb is the same as --dest if (localdb_dir == "") { localdb_dir = backup_dest; } - printf("Source: %s\nDest: %s\nDatabase: %s\n\n", - backup_source.c_str(), backup_dest.c_str(), localdb_dir.c_str()); + // 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()); + } - tss = new TarSegmentStore(backup_dest); block_buf = new char[LBS_BLOCK_SIZE]; /* Store the time when the backup started, so it can be included in the @@ -519,17 +691,17 @@ 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); + db->Open(database_path.c_str(), desc_buf, + backup_scheme.size() ? backup_scheme.c_str() : NULL); + + tss = new TarSegmentStore(backup_dest, db); /* Initialize the stat cache, for skipping over unchanged files. */ statcache = new StatCache; - statcache->Open(localdb_dir.c_str(), desc_buf); + statcache->Open(localdb_dir.c_str(), desc_buf, + backup_scheme.size() ? backup_scheme.c_str() : NULL); - try { - scanfile("."); - } catch (IOException e) { - fprintf(stderr, "IOException: %s\n", e.getError().c_str()); - } + scanfile(".", false); metadata_flush(); const string md = metadata_root.str(); @@ -544,25 +716,71 @@ int main(int argc, char *argv[]) string backup_root = root->get_ref().to_string(); delete root; - db->Close(); - statcache->Close(); delete statcache; 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 = backup_dest + "/snapshot-"; + if (backup_scheme.size() > 0) + checksum_filename += backup_scheme + "-"; + checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums"; + FILE *checksums = fopen(checksum_filename.c_str(), "w"); + if (checksums != NULL) { + 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); + } else { + fprintf(stderr, "ERROR: Unable to write checksums file: %m\n"); + } + + db->Close(); + /* 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-" + desc_buf + ".lbs"; + string desc_filename = backup_dest + "/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 << "Format: LBS Snapshot v0.2\n"; + descriptor << "Producer: LBS " << lbs_version << "\n"; strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf); descriptor << "Date: " << desc_buf << "\n"; + if (backup_scheme.size() > 0) + descriptor << "Scheme: " << backup_scheme << "\n"; descriptor << "Root: " << backup_root << "\n"; + SHA1Checksum checksum_csum; + if (checksum_csum.process_file(checksum_filename.c_str())) { + descriptor << "Checksum-File: " << checksum_csum.checksum_str() << "\n"; + } + descriptor << "Segments:\n"; for (std::set::iterator i = segment_list.begin(); i != segment_list.end(); ++i) {