X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=f5ab2b6011274adc2e621f220107a6dfa996dee0;hb=87c6a5ed4ec515c2e154f98529c1d2f077d2a7e4;hp=5d7c9c570076d564d5a8dbb25ace3c8acd207358;hpb=117fafdd16cf070cda130d6d8ac321947692055a;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index 5d7c9c5..f5ab2b6 100644 --- a/scandir.cc +++ b/scandir.cc @@ -10,6 +10,7 @@ #include #include #include +#include #include #include @@ -22,11 +23,11 @@ #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; @@ -67,8 +68,6 @@ std::ostringstream metadata; /* Keep track of all segments which are needed to reconstruct the snapshot. */ std::set segment_list; -void scandir(const string& path, bool include); - /* Selection of files to include/exclude in the snapshot. */ std::list includes; // Paths in which files should be saved std::list excludes; // Paths which will not be saved @@ -105,7 +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; @@ -114,7 +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 { @@ -130,18 +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; - } - /* 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 * re-reading the entire contents. */ @@ -183,9 +177,14 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path) 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); @@ -266,65 +265,22 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path) return size; } -void scanfile(const string& path, bool include) +/* 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; - - string true_path; - if (relative_paths) - true_path = path; - else - true_path = "/" + path; - - // Set to true if the item is a directory and we should recursively scan - bool recurse = false; - - // 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; - dictionary file_info; - - lstat(true_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); @@ -348,11 +304,11 @@ void scanfile(const string& path, bool include) 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'; @@ -361,7 +317,7 @@ void scanfile(const string& path, bool include) * 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(true_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) { @@ -376,33 +332,7 @@ void scanfile(const string& path, bool include) 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(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); - - 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); @@ -417,7 +347,6 @@ void scanfile(const string& path, bool include) break; case S_IFDIR: inode_type = 'd'; - recurse = true; break; default: @@ -434,49 +363,138 @@ void scanfile(const string& path, bool include) // 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, include); } -void scandir(const string& path, bool include) +void scanfile(const string& path, bool include) { + int fd = -1; + long flags; + struct stat stat_buf; + list refs; + string true_path; if (relative_paths) true_path = path; else true_path = "/" + path; - DIR *dir = opendir(true_path.c_str()); + // Set to true if we should scan through the contents of this directory, + // but not actually back files up + bool scan_only = false; - if (dir == NULL) { - fprintf(stderr, "Error: %m\n"); - return; + // 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; + } } - struct dirent *ent; - vector contents; - while ((ent = readdir(dir)) != NULL) { - string filename(ent->d_name); - if (filename == "." || filename == "..") - continue; - contents.push_back(filename); + 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 (lstat(true_path.c_str(), &stat_buf) < 0) { + fprintf(stderr, "lstat(%s): %m\n", path.c_str()); + return; } - sort(contents.begin(), contents.end()); + 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); - for (vector::iterator i = contents.begin(); - i != contents.end(); ++i) { - const string& filename = *i; - if (path == ".") - scanfile(filename, include); - else - scanfile(path + "/" + filename, include); + /* 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; + } } - closedir(dir); + 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); + + 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 @@ -536,7 +554,8 @@ void usage(const char *program) " (defaults to \"bzip2 -c\")\n" " --filter-extension=EXT\n" " string to append to segment files\n" - " (defaults to \".bz2\")\n", + " (defaults to \".bz2\")\n" + " --scheme=NAME optional name for this snapshot\n", program ); } @@ -546,6 +565,7 @@ int main(int argc, char *argv[]) string backup_source = "."; string backup_dest = ""; string localdb_dir = ""; + string backup_scheme = ""; while (1) { static struct option long_options[] = { @@ -554,6 +574,7 @@ int main(int argc, char *argv[]) {"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}, }; @@ -583,6 +604,9 @@ int main(int argc, char *argv[]) case 4: // --dest backup_dest = optarg; break; + case 5: // --scheme + backup_scheme = optarg; + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; @@ -659,7 +683,8 @@ 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); /* Initialize the stat cache, for skipping over unchanged files. */ statcache = new StatCache; @@ -692,13 +717,18 @@ int main(int argc, char *argv[]) /* 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 << "Producer: " << lbs_version << "\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"; descriptor << "Segments:\n";