tar: support for tar --numeric-owner. By Natanael Copa.
[oweals/busybox.git] / archival / tar.c
index 57da0b6b83d68d181fad47a234063c48382a40a4..03d66a6929e9e8b2dc8ecca730c6039c0c162d11 100644 (file)
@@ -3,7 +3,7 @@
  * Mini tar implementation for busybox
  *
  * Modified to use common extraction code used by ar, cpio, dpkg-deb, dpkg
- *  Glenn McGrath <bug1@iinet.net.au>
+ *  by Glenn McGrath
  *
  * Note, that as of BusyBox-0.43, tar has been completely rewritten from the
  * ground up.  It still has remnants of the old code lying about, but it is
  */
 
 #include <fnmatch.h>
-#include <getopt.h>
-#include "busybox.h"
+#include "libbb.h"
 #include "unarchive.h"
 
+/* FIXME: Stop using this non-standard feature */
+#ifndef FNM_LEADING_DIR
+#define FNM_LEADING_DIR 0
+#endif
+
+
+#define block_buf bb_common_bufsiz1
+
+
+#if !ENABLE_FEATURE_SEAMLESS_GZ && !ENABLE_FEATURE_SEAMLESS_BZ2
+/* Do not pass gzip flag to writeTarFile() */
+#define writeTarFile(tar_fd, verboseFlag, dereferenceFlag, include, exclude, gzip) \
+       writeTarFile(tar_fd, verboseFlag, dereferenceFlag, include, exclude)
+#endif
+
+
 #if ENABLE_FEATURE_TAR_CREATE
 
 /* Tar file constants  */
@@ -48,14 +63,18 @@ struct TarHeader {            /* byte offset */
        char chksum[8];           /* 148-155 */
        char typeflag;            /* 156-156 */
        char linkname[NAME_SIZE]; /* 157-256 */
-       char magic[6];            /* 257-262 */
-       char version[2];          /* 263-264 */
+       /* POSIX:   "ustar" NUL "00" */
+       /* GNU tar: "ustar  " NUL */
+       /* Normally it's defined as magic[6] followed by
+        * version[2], but we put them together to save code.
+        */
+       char magic[8];            /* 257-264 */
        char uname[32];           /* 265-296 */
        char gname[32];           /* 297-328 */
        char devmajor[8];         /* 329-336 */
        char devminor[8];         /* 337-344 */
        char prefix[155];         /* 345-499 */
-       char padding[12];         /* 500-512 (pad to exactly the TAR_BLOCK_SIZE) */
+       char padding[12];         /* 500-512 (pad to exactly TAR_BLOCK_SIZE) */
 };
 
 /*
@@ -65,26 +84,26 @@ struct TarHeader {            /* byte offset */
 */
 typedef struct HardLinkInfo HardLinkInfo;
 struct HardLinkInfo {
-       HardLinkInfo *next;     /* Next entry in list */
-       dev_t dev;                      /* Device number */
-       ino_t ino;                      /* Inode number */
-       short linkCount;        /* (Hard) Link Count */
-       char name[1];           /* Start of filename (must be last) */
+       HardLinkInfo *next;     /* Next entry in list */
+       dev_t dev;              /* Device number */
+       ino_t ino;              /* Inode number */
+       short linkCount;        /* (Hard) Link Count */
+       char name[1];           /* Start of filename (must be last) */
 };
 
 /* Some info to be carried along when creating a new tarball */
 typedef struct TarBallInfo TarBallInfo;
 struct TarBallInfo {
-       int tarFd;                              /* Open-for-write file descriptor
-                                                          for the tarball */
-       struct stat statBuf;    /* Stat info for the tarball, letting
-                                                          us know the inode and device that the
-                                                          tarball lives, so we can avoid trying
-                                                          to include the tarball into itself */
-       int verboseFlag;                /* Whether to print extra stuff or not */
-       const llist_t *excludeList;     /* List of files to not include */
-       HardLinkInfo *hlInfoHead;       /* Hard Link Tracking Information */
-       HardLinkInfo *hlInfo;   /* Hard Link Info for the current file */
+       int tarFd;                      /* Open-for-write file descriptor
+                                        * for the tarball */
+       struct stat statBuf;            /* Stat info for the tarball, letting
+                                        * us know the inode and device that the
+                                        * tarball lives, so we can avoid trying
+                                        * to include the tarball into itself */
+       int verboseFlag;                /* Whether to print extra stuff or not */
+       const llist_t *excludeList;     /* List of files to not include */
+       HardLinkInfo *hlInfoHead;       /* Hard Link Tracking Information */
+       HardLinkInfo *hlInfo;           /* Hard Link Info for the current file */
 };
 
 /* A nice enum with all the possible tar file content types */
@@ -104,7 +123,7 @@ enum TarFileType {
 typedef enum TarFileType TarFileType;
 
 /* Might be faster (and bigger) if the dev/ino were stored in numeric order;) */
-static void addHardLinkInfo(HardLinkInfo ** hlInfoHeadPtr,
+static void addHardLinkInfo(HardLinkInfo **hlInfoHeadPtr,
                                        struct stat *statbuf,
                                        const char *fileName)
 {
@@ -120,7 +139,7 @@ static void addHardLinkInfo(HardLinkInfo ** hlInfoHeadPtr,
        strcpy(hlInfo->name, fileName);
 }
 
-static void freeHardLinkInfo(HardLinkInfo ** hlInfoHeadPtr)
+static void freeHardLinkInfo(HardLinkInfo **hlInfoHeadPtr)
 {
        HardLinkInfo *hlInfo;
        HardLinkInfo *hlInfoNext;
@@ -134,11 +153,10 @@ static void freeHardLinkInfo(HardLinkInfo ** hlInfoHeadPtr)
                }
                *hlInfoHeadPtr = NULL;
        }
-       return;
 }
 
 /* Might be faster (and bigger) if the dev/ino were stored in numeric order;) */
-static HardLinkInfo *findHardLinkInfo(HardLinkInfo * hlInfo, struct stat *statbuf)
+static HardLinkInfo *findHardLinkInfo(HardLinkInfo *hlInfo, struct stat *statbuf)
 {
        while (hlInfo) {
                if ((statbuf->st_ino == hlInfo->ino) && (statbuf->st_dev == hlInfo->dev))
@@ -173,6 +191,9 @@ static void putOctal(char *cp, int len, off_t value)
 
 static void chksum_and_xwrite(int fd, struct TarHeader* hp)
 {
+       /* POSIX says that checksum is done on unsigned bytes
+        * (Sun and HP-UX gets it wrong... more details in
+        * GNU tar source) */
        const unsigned char *cp;
        int chksum, size;
 
@@ -336,7 +357,8 @@ static int writeTarHeader(struct TarBallInfo *tbInfo,
        if (tbInfo->verboseFlag) {
                FILE *vbFd = stdout;
 
-               if (tbInfo->tarFd == STDOUT_FILENO)     /* If the archive goes to stdout, verbose to stderr */
+               /* If archive goes to stdout, verbose goes to stderr */
+               if (tbInfo->tarFd == STDOUT_FILENO)
                        vbFd = stderr;
                /* GNU "tar cvvf" prints "extended" listing a-la "ls -l" */
                /* We don't have such excesses here: for us "v" == "vv" */
@@ -375,13 +397,34 @@ static int exclude_file(const llist_t *excluded_files, const char *file)
 #define exclude_file(excluded_files, file) 0
 #endif
 
-static int writeFileToTarball(const char *fileName, struct stat *statbuf,
-                       void *userData, int depth ATTRIBUTE_UNUSED)
+static int FAST_FUNC writeFileToTarball(const char *fileName, struct stat *statbuf,
+                       void *userData, int depth UNUSED_PARAM)
 {
        struct TarBallInfo *tbInfo = (struct TarBallInfo *) userData;
        const char *header_name;
        int inputFileFd = -1;
 
+       /* Strip leading '/' (must be before memorizing hardlink's name) */
+       header_name = fileName;
+       while (header_name[0] == '/') {
+               static smallint warned;
+
+               if (!warned) {
+                       bb_error_msg("removing leading '/' from member names");
+                       warned = 1;
+               }
+               header_name++;
+       }
+
+       if (header_name[0] == '\0')
+               return TRUE;
+
+       /* It is against the rules to archive a socket */
+       if (S_ISSOCK(statbuf->st_mode)) {
+               bb_error_msg("%s: socket ignored", fileName);
+               return TRUE;
+       }
+
        /*
         * Check to see if we are dealing with a hard link.
         * If so -
@@ -393,54 +436,34 @@ static int writeFileToTarball(const char *fileName, struct stat *statbuf,
        if (statbuf->st_nlink > 1) {
                tbInfo->hlInfo = findHardLinkInfo(tbInfo->hlInfoHead, statbuf);
                if (tbInfo->hlInfo == NULL)
-                       addHardLinkInfo(&tbInfo->hlInfoHead, statbuf, fileName);
-       }
-
-       /* It is against the rules to archive a socket */
-       if (S_ISSOCK(statbuf->st_mode)) {
-               bb_error_msg("%s: socket ignored", fileName);
-               return TRUE;
+                       addHardLinkInfo(&tbInfo->hlInfoHead, statbuf, header_name);
        }
 
        /* It is a bad idea to store the archive we are in the process of creating,
         * so check the device and inode to be sure that this particular file isn't
         * the new tarball */
-       if (tbInfo->statBuf.st_dev == statbuf->st_dev &&
-               tbInfo->statBuf.st_ino == statbuf->st_ino) {
+       if (tbInfo->statBuf.st_dev == statbuf->st_dev
+        && tbInfo->statBuf.st_ino == statbuf->st_ino
+       ) {
                bb_error_msg("%s: file is the archive; skipping", fileName);
                return TRUE;
        }
 
-       header_name = fileName;
-       while (header_name[0] == '/') {
-               static int alreadyWarned = FALSE;
-
-               if (alreadyWarned == FALSE) {
-                       bb_error_msg("removing leading '/' from member names");
-                       alreadyWarned = TRUE;
-               }
-               header_name++;
-       }
+       if (exclude_file(tbInfo->excludeList, header_name))
+               return SKIP;
 
 #if !ENABLE_FEATURE_TAR_GNU_EXTENSIONS
-       if (strlen(fileName) >= NAME_SIZE) {
+       if (strlen(header_name) >= NAME_SIZE) {
                bb_error_msg("names longer than "NAME_SIZE_STR" chars not supported");
                return TRUE;
        }
 #endif
 
-       if (header_name[0] == '\0')
-               return TRUE;
-
-       if (exclude_file(tbInfo->excludeList, header_name))
-               return SKIP;
-
        /* Is this a regular file? */
        if (tbInfo->hlInfo == NULL && S_ISREG(statbuf->st_mode)) {
                /* open the file we want to archive, and make sure all is well */
-               inputFileFd = open(fileName, O_RDONLY);
+               inputFileFd = open_or_warn(fileName, O_RDONLY);
                if (inputFileFd < 0) {
-                       bb_perror_msg("%s: cannot open", fileName);
                        return FALSE;
                }
        }
@@ -453,7 +476,7 @@ static int writeFileToTarball(const char *fileName, struct stat *statbuf,
        /* If it was a regular file, write out the body */
        if (inputFileFd >= 0) {
                size_t readSize;
-               /* Wwrite the file to the archive. */
+               /* Write the file to the archive. */
                /* We record size into header first, */
                /* and then write out file. If file shrinks in between, */
                /* tar will be corrupted. So we don't allow for that. */
@@ -474,24 +497,101 @@ static int writeFileToTarball(const char *fileName, struct stat *statbuf,
                /* Pad the file up to the tar block size */
                /* (a few tricks here in the name of code size) */
                readSize = (-(int)statbuf->st_size) & (TAR_BLOCK_SIZE-1);
-               memset(bb_common_bufsiz1, 0, readSize);
-               xwrite(tbInfo->tarFd, bb_common_bufsiz1, readSize);
+               memset(block_buf, 0, readSize);
+               xwrite(tbInfo->tarFd, block_buf, readSize);
        }
 
        return TRUE;
 }
 
-static int writeTarFile(const int tar_fd, const int verboseFlag,
-       const unsigned long dereferenceFlag, const llist_t *include,
-       const llist_t *exclude, const int gzip)
+#if ENABLE_FEATURE_SEAMLESS_GZ || ENABLE_FEATURE_SEAMLESS_BZ2
+#if !(ENABLE_FEATURE_SEAMLESS_GZ && ENABLE_FEATURE_SEAMLESS_BZ2)
+#define vfork_compressor(tar_fd, gzip) vfork_compressor(tar_fd)
+#endif
+/* Don't inline: vfork scares gcc and pessimizes code */
+static void NOINLINE vfork_compressor(int tar_fd, int gzip)
+{
+       pid_t gzipPid;
+#if ENABLE_FEATURE_SEAMLESS_GZ && ENABLE_FEATURE_SEAMLESS_BZ2
+       const char *zip_exec = (gzip == 1) ? "gzip" : "bzip2";
+#elif ENABLE_FEATURE_SEAMLESS_GZ
+       const char *zip_exec = "gzip";
+#else /* only ENABLE_FEATURE_SEAMLESS_BZ2 */
+       const char *zip_exec = "bzip2";
+#endif
+       // On Linux, vfork never unpauses parent early, although standard
+       // allows for that. Do we want to waste bytes checking for it?
+#define WAIT_FOR_CHILD 0
+       volatile int vfork_exec_errno = 0;
+       struct fd_pair gzipDataPipe;
+#if WAIT_FOR_CHILD
+       struct fd_pair gzipStatusPipe;
+       xpiped_pair(gzipStatusPipe);
+#endif
+       xpiped_pair(gzipDataPipe);
+
+       signal(SIGPIPE, SIG_IGN); /* we only want EPIPE on errors */
+
+#if defined(__GNUC__) && __GNUC__
+       /* Avoid vfork clobbering */
+       (void) &zip_exec;
+#endif
+
+       gzipPid = vfork();
+       if (gzipPid < 0)
+               bb_perror_msg_and_die("vfork");
+
+       if (gzipPid == 0) {
+               /* child */
+               /* NB: close _first_, then move fds! */
+               close(gzipDataPipe.wr);
+#if WAIT_FOR_CHILD
+               close(gzipStatusPipe.rd);
+               /* gzipStatusPipe.wr will close only on exec -
+                * parent waits for this close to happen */
+               fcntl(gzipStatusPipe.wr, F_SETFD, FD_CLOEXEC);
+#endif
+               xmove_fd(gzipDataPipe.rd, 0);
+               xmove_fd(tar_fd, 1);
+               /* exec gzip/bzip2 program/applet */
+               BB_EXECLP(zip_exec, zip_exec, "-f", NULL);
+               vfork_exec_errno = errno;
+               _exit(EXIT_FAILURE);
+       }
+
+       /* parent */
+       xmove_fd(gzipDataPipe.wr, tar_fd);
+       close(gzipDataPipe.rd);
+#if WAIT_FOR_CHILD
+       close(gzipStatusPipe.wr);
+       while (1) {
+               char buf;
+               int n;
+
+               /* Wait until child execs (or fails to) */
+               n = full_read(gzipStatusPipe.rd, &buf, 1);
+               if (n < 0 /* && errno == EAGAIN */)
+                       continue;       /* try it again */
+       }
+       close(gzipStatusPipe.rd);
+#endif
+       if (vfork_exec_errno) {
+               errno = vfork_exec_errno;
+               bb_perror_msg_and_die("cannot exec %s", zip_exec);
+       }
+}
+#endif /* ENABLE_FEATURE_SEAMLESS_GZ || ENABLE_FEATURE_SEAMLESS_BZ2 */
+
+
+/* gcc 4.2.1 inlines it, making code bigger */
+static NOINLINE int writeTarFile(int tar_fd, int verboseFlag,
+       int dereferenceFlag, const llist_t *include,
+       const llist_t *exclude, int gzip)
 {
-       pid_t gzipPid = 0;
        int errorFlag = FALSE;
        struct TarBallInfo tbInfo;
 
        tbInfo.hlInfoHead = NULL;
-
-       fchmod(tar_fd, 0644);
        tbInfo.tarFd = tar_fd;
        tbInfo.verboseFlag = verboseFlag;
 
@@ -500,76 +600,26 @@ static int writeTarFile(const int tar_fd, const int verboseFlag,
        if (fstat(tbInfo.tarFd, &tbInfo.statBuf) < 0)
                bb_perror_msg_and_die("cannot stat tar file");
 
-       if ((ENABLE_FEATURE_TAR_GZIP || ENABLE_FEATURE_TAR_BZIP2) && gzip) {
-               int gzipDataPipe[2] = { -1, -1 };
-               int gzipStatusPipe[2] = { -1, -1 };
-               volatile int vfork_exec_errno = 0;
-               const char *zip_exec = (gzip == 1) ? "gzip" : "bzip2";
-
-               if (pipe(gzipDataPipe) < 0 || pipe(gzipStatusPipe) < 0)
-                       bb_perror_msg_and_die("pipe");
-
-               signal(SIGPIPE, SIG_IGN); /* we only want EPIPE on errors */
-
-#if defined(__GNUC__) && __GNUC__
-               /* Avoid vfork clobbering */
-               (void) &include;
-               (void) &errorFlag;
-               (void) &zip_exec;
+#if ENABLE_FEATURE_SEAMLESS_GZ || ENABLE_FEATURE_SEAMLESS_BZ2
+       if (gzip)
+               vfork_compressor(tbInfo.tarFd, gzip);
 #endif
 
-               gzipPid = vfork();
-
-               if (gzipPid == 0) {
-                       dup2(gzipDataPipe[0], 0);
-                       close(gzipDataPipe[1]);
-
-                       dup2(tbInfo.tarFd, 1);
-
-                       close(gzipStatusPipe[0]);
-                       fcntl(gzipStatusPipe[1], F_SETFD, FD_CLOEXEC);  /* close on exec shows success */
-
-                       BB_EXECLP(zip_exec, zip_exec, "-f", NULL);
-                       vfork_exec_errno = errno;
-
-                       close(gzipStatusPipe[1]);
-                       exit(-1);
-               } else if (gzipPid > 0) {
-                       close(gzipDataPipe[0]);
-                       close(gzipStatusPipe[1]);
-
-                       while (1) {
-                               char buf;
-
-                               int n = full_read(gzipStatusPipe[0], &buf, 1);
-
-                               if (n == 0 && vfork_exec_errno != 0) {
-                                       errno = vfork_exec_errno;
-                                       bb_perror_msg_and_die("cannot exec %s", zip_exec);
-                               } else if ((n < 0) && (errno == EAGAIN || errno == EINTR))
-                                       continue;       /* try it again */
-                               break;
-                       }
-                       close(gzipStatusPipe[0]);
-
-                       tbInfo.tarFd = gzipDataPipe[1];
-               } else bb_perror_msg_and_die("vfork gzip");
-       }
-
        tbInfo.excludeList = exclude;
 
        /* Read the directory/files and iterate over them one at a time */
        while (include) {
-               if (!recursive_action(include->data, TRUE, dereferenceFlag,
-                               FALSE, writeFileToTarball, writeFileToTarball, &tbInfo, 0))
+               if (!recursive_action(include->data, ACTION_RECURSE |
+                               (dereferenceFlag ? ACTION_FOLLOWLINKS : 0),
+                               writeFileToTarball, writeFileToTarball, &tbInfo, 0))
                {
                        errorFlag = TRUE;
                }
                include = include->link;
        }
        /* Write two empty blocks to the end of the archive */
-       memset(bb_common_bufsiz1, 0, 2*TAR_BLOCK_SIZE);
-       xwrite(tbInfo.tarFd, bb_common_bufsiz1, 2*TAR_BLOCK_SIZE);
+       memset(block_buf, 0, 2*TAR_BLOCK_SIZE);
+       xwrite(tbInfo.tarFd, block_buf, 2*TAR_BLOCK_SIZE);
 
        /* To be pedantically correct, we would check if the tarball
         * is smaller than 20 tar blocks, and pad it if it was smaller,
@@ -586,37 +636,34 @@ static int writeTarFile(const int tar_fd, const int verboseFlag,
        if (errorFlag)
                bb_error_msg("error exit delayed from previous errors");
 
-       if (gzipPid) {
+#if ENABLE_FEATURE_SEAMLESS_GZ || ENABLE_FEATURE_SEAMLESS_BZ2
+       if (gzip) {
                int status;
-               if (waitpid(gzipPid, &status, 0) == -1)
+               if (safe_waitpid(-1, &status, 0) == -1)
                        bb_perror_msg("waitpid");
                else if (!WIFEXITED(status) || WEXITSTATUS(status))
                        /* gzip was killed or has exited with nonzero! */
                        errorFlag = TRUE;
        }
+#endif
        return errorFlag;
 }
 #else
-int writeTarFile(const int tar_fd, const int verboseFlag,
-       const unsigned long dereferenceFlag, const llist_t *include,
-       const llist_t *exclude, const int gzip);
+int writeTarFile(int tar_fd, int verboseFlag,
+       int dereferenceFlag, const llist_t *include,
+       const llist_t *exclude, int gzip);
 #endif /* FEATURE_TAR_CREATE */
 
 #if ENABLE_FEATURE_TAR_FROM
 static llist_t *append_file_list_to_list(llist_t *list)
 {
        FILE *src_stream;
-       llist_t *cur = list;
-       llist_t *tmp;
        char *line;
        llist_t *newlist = NULL;
 
-       while (cur) {
-               src_stream = xfopen(cur->data, "r");
-               tmp = cur;
-               cur = cur->link;
-               free(tmp);
-               while ((line = xmalloc_getline(src_stream)) != NULL) {
+       while (list) {
+               src_stream = xfopen_for_read(llist_pop(&list));
+               while ((line = xmalloc_fgetline(src_stream)) != NULL) {
                        /* kill trailing '/' unless the string is just "/" */
                        char *cp = last_char_is(line, '/');
                        if (cp > line)
@@ -631,8 +678,8 @@ static llist_t *append_file_list_to_list(llist_t *list)
 #define append_file_list_to_list(x) 0
 #endif
 
-#if ENABLE_FEATURE_TAR_COMPRESS
-static char get_header_tar_Z(archive_handle_t *archive_handle)
+#if ENABLE_FEATURE_SEAMLESS_Z
+static char FAST_FUNC get_header_tar_Z(archive_handle_t *archive_handle)
 {
        /* Can't lseek over pipes */
        archive_handle->seek = seek_by_read;
@@ -644,10 +691,10 @@ static char get_header_tar_Z(archive_handle_t *archive_handle)
                bb_error_msg_and_die("invalid magic");
        }
 
-       archive_handle->src_fd = open_transformer(archive_handle->src_fd, uncompress);
+       open_transformer(archive_handle->src_fd, unpack_Z_stream, "uncompress");
        archive_handle->offset = 0;
        while (get_header_tar(archive_handle) == EXIT_SUCCESS)
-               /* nothing */;
+               continue;
 
        /* Can only do one file at a time */
        return EXIT_FAILURE;
@@ -666,7 +713,7 @@ static void handle_SIGCHLD(int status)
        /* Actually, 'status' is a signo. We reuse it for other needs */
 
        /* Wait for any child without blocking */
-       if (waitpid(-1, &status, WNOHANG) < 0)
+       if (wait_any_nohang(&status) < 0)
                /* wait failed?! I'm confused... */
                return;
 
@@ -674,23 +721,24 @@ static void handle_SIGCHLD(int status)
                /* child exited with 0 */
                return;
        /* Cannot happen?
-       if(!WIFSIGNALED(status) && !WIFEXITED(status)) return; */
+       if (!WIFSIGNALED(status) && !WIFEXITED(status)) return; */
        child_error = 1;
 }
 #endif
 
 enum {
        OPTBIT_KEEP_OLD = 7,
-       USE_FEATURE_TAR_CREATE(  OPTBIT_CREATE      ,)
-       USE_FEATURE_TAR_CREATE(  OPTBIT_DEREFERENCE ,)
-       USE_FEATURE_TAR_BZIP2(   OPTBIT_BZIP2       ,)
-       USE_FEATURE_TAR_LZMA(    OPTBIT_LZMA        ,)
-       USE_FEATURE_TAR_FROM(    OPTBIT_INCLUDE_FROM,)
-       USE_FEATURE_TAR_FROM(    OPTBIT_EXCLUDE_FROM,)
-       USE_FEATURE_TAR_GZIP(    OPTBIT_GZIP        ,)
-       USE_FEATURE_TAR_COMPRESS(OPTBIT_COMPRESS    ,)
+       USE_FEATURE_TAR_CREATE(   OPTBIT_CREATE      ,)
+       USE_FEATURE_TAR_CREATE(   OPTBIT_DEREFERENCE ,)
+       USE_FEATURE_SEAMLESS_BZ2( OPTBIT_BZIP2       ,)
+       USE_FEATURE_SEAMLESS_LZMA(OPTBIT_LZMA        ,)
+       USE_FEATURE_TAR_FROM(     OPTBIT_INCLUDE_FROM,)
+       USE_FEATURE_TAR_FROM(     OPTBIT_EXCLUDE_FROM,)
+       USE_FEATURE_SEAMLESS_GZ(  OPTBIT_GZIP        ,)
+       USE_FEATURE_SEAMLESS_Z(   OPTBIT_COMPRESS    ,)
        OPTBIT_NOPRESERVE_OWN,
        OPTBIT_NOPRESERVE_PERM,
+       OPTBIT_NUMERIC_OWNER,
        OPT_TEST         = 1 << 0, // t
        OPT_EXTRACT      = 1 << 1, // x
        OPT_BASEDIR      = 1 << 2, // C
@@ -699,63 +747,64 @@ enum {
        OPT_P            = 1 << 5, // p
        OPT_VERBOSE      = 1 << 6, // v
        OPT_KEEP_OLD     = 1 << 7, // k
-       OPT_CREATE       = USE_FEATURE_TAR_CREATE(  (1<<OPTBIT_CREATE      )) + 0, // c
-       OPT_DEREFERENCE  = USE_FEATURE_TAR_CREATE(  (1<<OPTBIT_DEREFERENCE )) + 0, // h
-       OPT_BZIP2        = USE_FEATURE_TAR_BZIP2(   (1<<OPTBIT_BZIP2       )) + 0, // j
-       OPT_LZMA         = USE_FEATURE_TAR_LZMA(    (1<<OPTBIT_LZMA        )) + 0, // a
-       OPT_INCLUDE_FROM = USE_FEATURE_TAR_FROM(    (1<<OPTBIT_INCLUDE_FROM)) + 0, // T
-       OPT_EXCLUDE_FROM = USE_FEATURE_TAR_FROM(    (1<<OPTBIT_EXCLUDE_FROM)) + 0, // X
-       OPT_GZIP         = USE_FEATURE_TAR_GZIP(    (1<<OPTBIT_GZIP        )) + 0, // z
-       OPT_COMPRESS     = USE_FEATURE_TAR_COMPRESS((1<<OPTBIT_COMPRESS    )) + 0, // Z
+       OPT_CREATE       = USE_FEATURE_TAR_CREATE(   (1 << OPTBIT_CREATE      )) + 0, // c
+       OPT_DEREFERENCE  = USE_FEATURE_TAR_CREATE(   (1 << OPTBIT_DEREFERENCE )) + 0, // h
+       OPT_BZIP2        = USE_FEATURE_SEAMLESS_BZ2( (1 << OPTBIT_BZIP2       )) + 0, // j
+       OPT_LZMA         = USE_FEATURE_SEAMLESS_LZMA((1 << OPTBIT_LZMA        )) + 0, // a
+       OPT_INCLUDE_FROM = USE_FEATURE_TAR_FROM(     (1 << OPTBIT_INCLUDE_FROM)) + 0, // T
+       OPT_EXCLUDE_FROM = USE_FEATURE_TAR_FROM(     (1 << OPTBIT_EXCLUDE_FROM)) + 0, // X
+       OPT_GZIP         = USE_FEATURE_SEAMLESS_GZ(  (1 << OPTBIT_GZIP        )) + 0, // z
+       OPT_COMPRESS     = USE_FEATURE_SEAMLESS_Z(   (1 << OPTBIT_COMPRESS    )) + 0, // Z
        OPT_NOPRESERVE_OWN  = 1 << OPTBIT_NOPRESERVE_OWN , // no-same-owner
        OPT_NOPRESERVE_PERM = 1 << OPTBIT_NOPRESERVE_PERM, // no-same-permissions
+       OPT_NUMERIC_OWNER = 1 << OPTBIT_NUMERIC_OWNER,
 };
 #if ENABLE_FEATURE_TAR_LONG_OPTIONS
-static const struct option tar_long_options[] = {
-       { "list",               0,  NULL,   't' },
-       { "extract",            0,  NULL,   'x' },
-       { "directory",          1,  NULL,   'C' },
-       { "file",               1,  NULL,   'f' },
-       { "to-stdout",          0,  NULL,   'O' },
-       { "same-permissions",   0,  NULL,   'p' },
-       { "verbose",            0,  NULL,   'v' },
-       { "keep-old",           0,  NULL,   'k' },
+static const char tar_longopts[] ALIGN1 =
+       "list\0"                No_argument       "t"
+       "extract\0"             No_argument       "x"
+       "directory\0"           Required_argument "C"
+       "file\0"                Required_argument "f"
+       "to-stdout\0"           No_argument       "O"
+       "same-permissions\0"    No_argument       "p"
+       "verbose\0"             No_argument       "v"
+       "keep-old\0"            No_argument       "k"
 # if ENABLE_FEATURE_TAR_CREATE
-       { "create",             0,  NULL,   'c' },
-       { "dereference",        0,  NULL,   'h' },
+       "create\0"              No_argument       "c"
+       "dereference\0"         No_argument       "h"
 # endif
-# if ENABLE_FEATURE_TAR_BZIP2
-       { "bzip2",              0,  NULL,   'j' },
+# if ENABLE_FEATURE_SEAMLESS_BZ2
+       "bzip2\0"               No_argument       "j"
 # endif
-# if ENABLE_FEATURE_TAR_LZMA
-       { "lzma",               0,  NULL,   'a' },
+# if ENABLE_FEATURE_SEAMLESS_LZMA
+       "lzma\0"                No_argument       "a"
 # endif
 # if ENABLE_FEATURE_TAR_FROM
-       { "files-from",         1,  NULL,   'T' },
-       { "exclude-from",       1,  NULL,   'X' },
+       "files-from\0"          Required_argument "T"
+       "exclude-from\0"        Required_argument "X"
 # endif
-# if ENABLE_FEATURE_TAR_GZIP
-       { "gzip",               0,  NULL,   'z' },
+# if ENABLE_FEATURE_SEAMLESS_GZ
+       "gzip\0"                No_argument       "z"
 # endif
-# if ENABLE_FEATURE_TAR_COMPRESS
-       { "compress",           0,  NULL,   'Z' },
+# if ENABLE_FEATURE_SEAMLESS_Z
+       "compress\0"            No_argument       "Z"
 # endif
-       { "no-same-owner",      0,  NULL,   0xfd },
-       { "no-same-permissions",0,  NULL,   0xfe },
+       "numeric-owner\0"       No_argument       "\xfc"
+       "no-same-owner\0"       No_argument       "\xfd"
+       "no-same-permissions\0" No_argument       "\xfe"
        /* --exclude takes next bit position in option mask, */
        /* therefore we have to either put it _after_ --no-same-perm */
        /* or add OPT[BIT]_EXCLUDE before OPT[BIT]_NOPRESERVE_OWN */
 # if ENABLE_FEATURE_TAR_FROM
-       { "exclude",            1,  NULL,   0xff },
+       "exclude\0"             Required_argument "\xff"
 # endif
-       { 0,                    0, 0, 0 }
-};
+       ;
 #endif
 
-int tar_main(int argc, char **argv);
-int tar_main(int argc, char **argv)
+int tar_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
+int tar_main(int argc UNUSED_PARAM, char **argv)
 {
-       char (*get_header_ptr)(archive_handle_t *) = get_header_tar;
+       char FAST_FUNC (*get_header_ptr)(archive_handle_t *) = get_header_tar;
        archive_handle_t *tar_handle;
        char *base_dir = NULL;
        const char *tar_filename = "-";
@@ -767,9 +816,13 @@ int tar_main(int argc, char **argv)
 
        /* Initialise default values */
        tar_handle = init_handle();
-       tar_handle->flags = ARCHIVE_CREATE_LEADING_DIRS
-                         | ARCHIVE_PRESERVE_DATE
-                         | ARCHIVE_EXTRACT_UNCONDITIONAL;
+       tar_handle->ah_flags = ARCHIVE_CREATE_LEADING_DIRS
+                            | ARCHIVE_PRESERVE_DATE
+                            | ARCHIVE_EXTRACT_UNCONDITIONAL;
+
+       /* Apparently only root's tar preserves perms (see bug 3844) */
+       if (getuid() != 0)
+               tar_handle->ah_flags |= ARCHIVE_NOPRESERVE_PERM;
 
        /* Prepend '-' to the first argument if required */
        opt_complementary = "--:" // first arg is options
@@ -783,16 +836,16 @@ int tar_main(int argc, char **argv)
                USE_FEATURE_TAR_CREATE("c--tx:t--cx:x--ct") // mutually exclusive
                SKIP_FEATURE_TAR_CREATE("t--x:x--t"); // mutually exclusive
 #if ENABLE_FEATURE_TAR_LONG_OPTIONS
-       applet_long_options = tar_long_options;
+       applet_long_options = tar_longopts;
 #endif
-       opt = getopt32(argc, argv,
+       opt = getopt32(argv,
                "txC:f:Opvk"
-               USE_FEATURE_TAR_CREATE(  "ch"  )
-               USE_FEATURE_TAR_BZIP2(   "j"   )
-               USE_FEATURE_TAR_LZMA(    "a"   )
-               USE_FEATURE_TAR_FROM(    "T:X:")
-               USE_FEATURE_TAR_GZIP(    "z"   )
-               USE_FEATURE_TAR_COMPRESS("Z"   )
+               USE_FEATURE_TAR_CREATE(   "ch"  )
+               USE_FEATURE_SEAMLESS_BZ2( "j"   )
+               USE_FEATURE_SEAMLESS_LZMA("a"   )
+               USE_FEATURE_TAR_FROM(     "T:X:")
+               USE_FEATURE_SEAMLESS_GZ(  "z"   )
+               USE_FEATURE_SEAMLESS_Z(   "Z"   )
                , &base_dir // -C dir
                , &tar_filename // -f filename
                USE_FEATURE_TAR_FROM(, &(tar_handle->accept)) // T
@@ -803,6 +856,7 @@ int tar_main(int argc, char **argv)
                , &verboseFlag // combined count for -t and -v
                , &verboseFlag // combined count for -t and -v
                );
+       argv += optind;
 
        if (verboseFlag) tar_handle->action_header = header_verbose_list;
        if (verboseFlag == 1) tar_handle->action_header = header_list;
@@ -814,13 +868,16 @@ int tar_main(int argc, char **argv)
                tar_handle->action_data = data_extract_to_stdout;
 
        if (opt & OPT_KEEP_OLD)
-               tar_handle->flags &= ~ARCHIVE_EXTRACT_UNCONDITIONAL;
+               tar_handle->ah_flags &= ~ARCHIVE_EXTRACT_UNCONDITIONAL;
 
        if (opt & OPT_NOPRESERVE_OWN)
-               tar_handle->flags |= ARCHIVE_NOPRESERVE_OWN;
+               tar_handle->ah_flags |= ARCHIVE_NOPRESERVE_OWN;
 
        if (opt & OPT_NOPRESERVE_PERM)
-               tar_handle->flags |= ARCHIVE_NOPRESERVE_PERM;
+               tar_handle->ah_flags |= ARCHIVE_NOPRESERVE_PERM;
+
+       if (opt & OPT_NUMERIC_OWNER)
+               tar_handle->ah_flags |= ARCHIVE_NUMERIC_OWNER;
 
        if (opt & OPT_GZIP)
                get_header_ptr = get_header_tar_gz;
@@ -848,23 +905,16 @@ int tar_main(int argc, char **argv)
        tar_handle->accept = append_file_list_to_list(tar_handle->accept);
 #endif
 
-       /* Check if we are reading from stdin */
-       if (argv[optind] && *argv[optind] == '-') {
-               /* Default is to read from stdin, so just skip to next arg */
-               optind++;
-       }
-
        /* Setup an array of filenames to work with */
        /* TODO: This is the same as in ar, separate function ? */
-       while (optind < argc) {
+       while (*argv) {
                /* kill trailing '/' unless the string is just "/" */
-               char *cp = last_char_is(argv[optind], '/');
-               if (cp > argv[optind])
+               char *cp = last_char_is(*argv, '/');
+               if (cp > *argv)
                        *cp = '\0';
-               llist_add_to(&tar_handle->accept, argv[optind]);
-               optind++;
+               llist_add_to_end(&tar_handle->accept, *argv);
+               argv++;
        }
-       tar_handle->accept = rev_llist(tar_handle->accept);
 
        if (tar_handle->accept || tar_handle->reject)
                tar_handle->filter = filter_accept_reject_list;
@@ -881,8 +931,7 @@ int tar_main(int argc, char **argv)
 
                        tar_stream = stdout;
                        /* Mimicking GNU tar 1.15.1: */
-                       flags = O_WRONLY|O_CREAT|O_TRUNC;
-               /* was doing unlink; open(O_WRONLY|O_CREAT|O_EXCL); why? */
+                       flags = O_WRONLY | O_CREAT | O_TRUNC;
                } else {
                        tar_stream = stdin;
                        flags = O_RDONLY;
@@ -892,7 +941,14 @@ int tar_main(int argc, char **argv)
                        tar_handle->src_fd = fileno(tar_stream);
                        tar_handle->seek = seek_by_read;
                } else {
-                       tar_handle->src_fd = xopen(tar_filename, flags);
+                       if (ENABLE_FEATURE_TAR_AUTODETECT && flags == O_RDONLY) {
+                               get_header_ptr = get_header_tar;
+                               tar_handle->src_fd = open_zipped(tar_filename);
+                               if (tar_handle->src_fd < 0)
+                                       bb_perror_msg_and_die("can't open '%s'", tar_filename);
+                       } else {
+                               tar_handle->src_fd = xopen(tar_filename, flags);
+                       }
                }
        }
 
@@ -906,11 +962,13 @@ int tar_main(int argc, char **argv)
 
        /* create an archive */
        if (opt & OPT_CREATE) {
+#if ENABLE_FEATURE_SEAMLESS_GZ || ENABLE_FEATURE_SEAMLESS_BZ2
                int zipMode = 0;
-               if (ENABLE_FEATURE_TAR_GZIP && get_header_ptr == get_header_tar_gz)
+               if (ENABLE_FEATURE_SEAMLESS_GZ && (opt & OPT_GZIP))
                        zipMode = 1;
-               if (ENABLE_FEATURE_TAR_BZIP2 && get_header_ptr == get_header_tar_bz2)
+               if (ENABLE_FEATURE_SEAMLESS_BZ2 && (opt & OPT_BZIP2))
                        zipMode = 2;
+#endif
                /* NB: writeTarFile() closes tar_handle->src_fd */
                return writeTarFile(tar_handle->src_fd, verboseFlag, opt & OPT_DEREFERENCE,
                                tar_handle->accept,
@@ -918,7 +976,7 @@ int tar_main(int argc, char **argv)
        }
 
        while (get_header_ptr(tar_handle) == EXIT_SUCCESS)
-               /* nothing */;
+               continue;
 
        /* Check that every file that should have been extracted was */
        while (tar_handle->accept) {