/* Create a tar archive.
- Copyright (C) 1985, 1992, 1993 Free Software Foundation
+ Copyright 1985, 92, 93, 94, 96, 97, 1999 Free Software Foundation, Inc.
+ Written by John Gilmore, on 1985-08-25.
-This file is part of GNU Tar.
+ This program is free software; you can redistribute it and/or modify it
+ under the terms of the GNU General Public License as published by the
+ Free Software Foundation; either version 2, or (at your option) any later
+ version.
-GNU Tar is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2, or (at your option)
-any later version.
+ This program is distributed in the hope that it will be useful, but
+ WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
+ Public License for more details.
-GNU Tar is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-GNU General Public License for more details.
+ You should have received a copy of the GNU General Public License along
+ with this program; if not, write to the Free Software Foundation, Inc.,
+ 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
-You should have received a copy of the GNU General Public License
-along with GNU Tar; see the file COPYING. If not, write to
-the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
+#include "system.h"
-/*
- * Create a tar archive.
- *
- * Written 25 Aug 1985 by John Gilmore, ihnp4!hoptoad!gnu.
- */
-
-#ifdef _AIX
- #pragma alloca
-#endif
-#include <sys/types.h>
-#include <stdio.h>
-#include <errno.h>
-#ifndef STDC_HEADERS
-extern int errno;
+#if !MSDOS
+# include <pwd.h>
+# include <grp.h>
#endif
-#ifdef BSD42
-#include <sys/file.h>
-#else
-#ifndef V7
-#include <fcntl.h>
-#endif
-#endif
-
-#include "tar.h"
-#include "port.h"
-
-#ifndef __MSDOS__
-#include <pwd.h>
-#include <grp.h>
-#endif
-
-#if defined (_POSIX_VERSION)
-#include <utime.h>
+#if HAVE_UTIME_H
+# include <utime.h>
#else
struct utimbuf
-{
- long actime;
- long modtime;
-};
-
+ {
+ long actime;
+ long modtime;
+ };
#endif
-extern struct stat hstat; /* Stat struct corresponding */
+#include "common.h"
-#ifndef __MSDOS__
+#ifndef MSDOS
extern dev_t ar_dev;
extern ino_t ar_ino;
#endif
-/* JF */
extern struct name *gnu_list_name;
-/*
- * If there are no symbolic links, there is no lstat(). Use stat().
- */
-#ifndef S_ISLNK
-#define lstat stat
-#endif
-
-extern void print_header ();
-
-union record *start_header ();
-void blank_name_list ();
-int check_exclude ();
-PTR ck_malloc ();
-PTR ck_realloc ();
-void clear_buffer ();
-void close_archive ();
-void collect_and_sort_names ();
-int confirm ();
-int deal_with_sparse ();
-void find_new_file_size ();
-void finish_header ();
-int finish_sparse_file ();
-void finduname ();
-void findgname ();
-int is_dot_or_dotdot ();
-void open_archive ();
-char *name_next ();
-void name_close ();
-void to_oct ();
-void dump_file ();
-void write_dir_file ();
-void write_eot ();
-void write_long ();
-int zero_record ();
-
-/* This code moved from tar.h since create.c is the only file that cares
- about 'struct link's. This means that other files might not have to
- include sys/types.h any more. */
+/* This module is the only one that cares about `struct link's. */
struct link
{
struct link *next;
dev_t dev;
ino_t ino;
- short linkcount;
char name[1];
};
-struct link *linklist; /* Points to first link in list */
-
-static nolinks; /* Gets set if we run out of RAM */
+static struct link *linklist; /* points to first link in list */
+\f
+/* The maximum uintmax_t value that can be represented with DIGITS digits,
+ assuming that each digit is BITS_PER_DIGIT wide. */
+#define MAX_VAL_WITH_DIGITS(digits, bits_per_digit) \
+ ((digits) * (bits_per_digit) < sizeof (uintmax_t) * CHAR_BIT \
+ ? ((uintmax_t) 1 << ((digits) * (bits_per_digit))) - 1 \
+ : (uintmax_t) -1)
+
+/* Convert VALUE to an octal representation suitable for tar headers.
+ Output to buffer WHERE with size SIZE.
+ The result is undefined if SIZE is 0 or if VALUE is too large to fit. */
+
+static void
+to_octal (uintmax_t value, char *where, size_t size)
+{
+ uintmax_t v = value;
+ size_t i = size;
-/*
- * "Scratch" space to store the information about a sparse file before
- * writing the info into the header or extended header
- */
-/* struct sp_array *sparsearray;*/
+ do
+ {
+ where[--i] = '0' + (v & ((1 << LG_8) - 1));
+ v >>= LG_8;
+ }
+ while (i);
+}
-/* number of elts storable in the sparsearray */
-/*int sparse_array_size = 10;*/
+/* Convert NEGATIVE VALUE to a base-256 representation suitable for
+ tar headers. NEGATIVE is 1 if VALUE was negative before being cast
+ to uintmax_t, 0 otherwise. Output to buffer WHERE with size SIZE.
+ The result is undefined if SIZE is 0 or if VALUE is too large to
+ fit. */
-void
-create_archive ()
+static void
+to_base256 (int negative, uintmax_t value, char *where, size_t size)
{
- register char *p;
- char *name_from_list ();
+ uintmax_t v = value;
+ uintmax_t propagated_sign_bits =
+ ((uintmax_t) - negative << (CHAR_BIT * sizeof v - LG_256));
+ size_t i = size;
- open_archive (0); /* Open for writing */
+ do
+ {
+ where[--i] = v & ((1 << LG_256) - 1);
+ v = propagated_sign_bits | (v >> LG_256);
+ }
+ while (i);
+}
+
+/* Convert NEGATIVE VALUE (which was originally of size VALSIZE) to
+ external form, using SUBSTITUTE (...) if VALUE won't fit. Output
+ to buffer WHERE with size SIZE. NEGATIVE is 1 iff VALUE was
+ negative before being cast to uintmax_t; its original bitpattern
+ can be deduced from VALSIZE, its original size before casting.
+ TYPE is the kind of value being output (useful for diagnostics).
+ Prefer the POSIX format of SIZE - 1 octal digits (with leading zero
+ digits), followed by '\0'. If this won't work, and if GNU or
+ OLDGNU format is allowed, use '\200' followed by base-256, or (if
+ NEGATIVE is nonzero) '\377' followed by two's complement base-256.
+ If neither format works, use SUBSTITUTE (...) instead. Pass to
+ SUBSTITUTE the address of an 0-or-1 flag recording whether the
+ substitute value is negative. */
+
+static void
+to_chars (int negative, uintmax_t value, size_t valsize,
+ uintmax_t (*substitute) PARAMS ((int *)),
+ char *where, size_t size, const char *type)
+{
+ int base256_allowed = (archive_format == GNU_FORMAT
+ || archive_format == OLDGNU_FORMAT);
- if (f_gnudump)
+ /* Generate the POSIX octal representation if the number fits. */
+ if (! negative && value <= MAX_VAL_WITH_DIGITS (size - 1, LG_8))
{
- char *buf = ck_malloc (PATH_MAX);
- char *q, *bufp;
+ where[size - 1] = '\0';
+ to_octal (value, where, size - 1);
+ }
- collect_and_sort_names ();
+ /* Otherwise, generate the base-256 representation if we are
+ generating an old or new GNU format and if the number fits. */
+ else if (((negative ? -1 - value : value)
+ <= MAX_VAL_WITH_DIGITS (size - 1, LG_256))
+ && base256_allowed)
+ {
+ where[0] = negative ? -1 : 1 << (LG_256 - 1);
+ to_base256 (negative, value, where + 1, size - 1);
+ }
- while (p = name_from_list ())
- dump_file (p, -1, 1);
- /* if(!f_dironly) { */
- blank_name_list ();
- while (p = name_from_list ())
+ /* Otherwise, if the number is negative, and if it would not cause
+ ambiguity on this host by confusing positive with negative
+ values, then generate the POSIX octal representation of the value
+ modulo 2**(field bits). The resulting tar file is
+ machine-dependent, since it depends on the host word size. Yuck!
+ But this is the traditional behavior. */
+ else if (negative && valsize * CHAR_BIT <= (size - 1) * LG_8)
+ {
+ static int warned_once;
+ if (! warned_once)
{
- strcpy (buf, p);
- if (p[strlen (p) - 1] != '/')
- strcat (buf, "/");
- bufp = buf + strlen (buf);
- for (q = gnu_list_name->dir_contents; q && *q; q += strlen (q) + 1)
- {
- if (*q == 'Y')
- {
- strcpy (bufp, q + 1);
- dump_file (buf, -1, 1);
- }
- }
+ warned_once = 1;
+ WARN ((0, 0, _("Generating negative octal headers")));
}
- /* } */
- free (buf);
+ where[size - 1] = '\0';
+ to_octal (value & MAX_VAL_WITH_DIGITS (valsize * CHAR_BIT, 1),
+ where, size - 1);
}
+
+ /* Otherwise, output a substitute value if possible (with a
+ warning), and an error message if not. */
else
{
- while (p = name_next (1))
- dump_file (p, -1, 1);
+ uintmax_t maxval = (base256_allowed
+ ? MAX_VAL_WITH_DIGITS (size - 1, LG_256)
+ : MAX_VAL_WITH_DIGITS (size - 1, LG_8));
+ char valbuf[UINTMAX_STRSIZE_BOUND + 1];
+ char maxbuf[UINTMAX_STRSIZE_BOUND];
+ char minbuf[UINTMAX_STRSIZE_BOUND + 1];
+ char const *minval_string;
+ char const *maxval_string = STRINGIFY_BIGINT (maxval, maxbuf);
+ char const *value_string;
+
+ if (base256_allowed)
+ {
+ uintmax_t m = maxval + 1 ? maxval + 1 : maxval / 2 + 1;
+ char *p = STRINGIFY_BIGINT (m, minbuf + 1);
+ *--p = '-';
+ minval_string = p;
+ }
+ else
+ minval_string = "0";
+
+ if (negative)
+ {
+ char *p = STRINGIFY_BIGINT (- value, valbuf + 1);
+ *--p = '-';
+ value_string = p;
+ }
+ else
+ value_string = STRINGIFY_BIGINT (value, valbuf);
+
+ if (substitute)
+ {
+ int negsub;
+ uintmax_t sub = substitute (&negsub) & maxval;
+ uintmax_t s = (negsub &= archive_format == GNU_FORMAT) ? - sub : sub;
+ char subbuf[UINTMAX_STRSIZE_BOUND + 1];
+ char *sub_string = STRINGIFY_BIGINT (s, subbuf + 1);
+ if (negsub)
+ *--sub_string = '-';
+ WARN ((0, 0, _("value %s out of %s range %s..%s; substituting %s"),
+ value_string, type, minval_string, maxval_string,
+ sub_string));
+ to_chars (negsub, s, valsize, 0, where, size, type);
+ }
+ else
+ ERROR ((0, 0, _("value %s out of %s range %s..%s"),
+ value_string, type, minval_string, maxval_string));
}
+}
- write_eot ();
- close_archive ();
- if (f_gnudump)
- write_dir_file ();
- name_close ();
+static uintmax_t
+gid_substitute (int *negative)
+{
+ gid_t r;
+#ifdef GID_NOBODY
+ r = GID_NOBODY;
+#else
+ static gid_t gid_nobody;
+ if (!gid_nobody && !gname_to_gid ("nobody", &gid_nobody))
+ gid_nobody = -2;
+ r = gid_nobody;
+#endif
+ *negative = r < 0;
+ return r;
}
-/*
- * Dump a single file. If it's a directory, recurse.
- * Result is 1 for success, 0 for failure.
- * Sets global "hstat" to stat() output for this file.
- */
void
-dump_file (p, curdev, toplevel)
- char *p; /* File name to dump */
- int curdev; /* Device our parent dir was on */
- int toplevel; /* Whether we are a toplevel call */
+gid_to_chars (gid_t v, char *p, size_t s)
{
- union record *header;
- char type;
- extern char *save_name; /* JF for multi-volume support */
- extern long save_totsize;
- extern long save_sizeleft;
- union record *exhdr;
- char save_linkflag;
- extern time_t new_time;
- int critical_error = 0;
- struct utimbuf restore_times;
- /* int sparse_ind = 0;*/
+ to_chars (v < 0, (uintmax_t) v, sizeof v, gid_substitute, p, s, "gid_t");
+}
+void
+major_to_chars (major_t v, char *p, size_t s)
+{
+ to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "major_t");
+}
- if (f_confirm && !confirm ("add", p))
- return;
+void
+minor_to_chars (minor_t v, char *p, size_t s)
+{
+ to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "minor_t");
+}
- /*
- * Use stat if following (rather than dumping) 4.2BSD's
- * symbolic links. Otherwise, use lstat (which, on non-4.2
- * systems, is #define'd to stat anyway.
- */
-#ifdef STX_HIDDEN /* AIX */
- if (0 != f_follow_links ?
- statx (p, &hstat, STATSIZE, STX_HIDDEN) :
- statx (p, &hstat, STATSIZE, STX_HIDDEN | STX_LINK))
+void
+mode_to_chars (mode_t v, char *p, size_t s)
+{
+ /* In the common case where the internal and external mode bits are the same,
+ and we are not using POSIX or GNU format,
+ propagate all unknown bits to the external mode.
+ This matches historical practice.
+ Otherwise, just copy the bits we know about. */
+ int negative;
+ uintmax_t u;
+ if (S_ISUID == TSUID && S_ISGID == TSGID && S_ISVTX == TSVTX
+ && S_IRUSR == TUREAD && S_IWUSR == TUWRITE && S_IXUSR == TUEXEC
+ && S_IRGRP == TGREAD && S_IWGRP == TGWRITE && S_IXGRP == TGEXEC
+ && S_IROTH == TOREAD && S_IWOTH == TOWRITE && S_IXOTH == TOEXEC
+ && archive_format != POSIX_FORMAT
+ && archive_format != GNU_FORMAT)
+ {
+ negative = v < 0;
+ u = v;
+ }
+ else
+ {
+ negative = 0;
+ u = ((v & S_ISUID ? TSUID : 0)
+ | (v & S_ISGID ? TSGID : 0)
+ | (v & S_ISVTX ? TSVTX : 0)
+ | (v & S_IRUSR ? TUREAD : 0)
+ | (v & S_IWUSR ? TUWRITE : 0)
+ | (v & S_IXUSR ? TUEXEC : 0)
+ | (v & S_IRGRP ? TGREAD : 0)
+ | (v & S_IWGRP ? TGWRITE : 0)
+ | (v & S_IXGRP ? TGEXEC : 0)
+ | (v & S_IROTH ? TOREAD : 0)
+ | (v & S_IWOTH ? TOWRITE : 0)
+ | (v & S_IXOTH ? TOEXEC : 0));
+ }
+ to_chars (negative, u, sizeof v, 0, p, s, "mode_t");
+}
+
+void
+off_to_chars (off_t v, char *p, size_t s)
+{
+ to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "off_t");
+}
+
+void
+size_to_chars (size_t v, char *p, size_t s)
+{
+ to_chars (0, (uintmax_t) v, sizeof v, 0, p, s, "size_t");
+}
+
+void
+time_to_chars (time_t v, char *p, size_t s)
+{
+ to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "time_t");
+}
+
+static uintmax_t
+uid_substitute (int *negative)
+{
+ uid_t r;
+#ifdef UID_NOBODY
+ r = UID_NOBODY;
#else
- if (0 != f_follow_links ? stat (p, &hstat) : lstat (p, &hstat))
+ static uid_t uid_nobody;
+ if (!uid_nobody && !uname_to_uid ("nobody", &uid_nobody))
+ uid_nobody = -2;
+ r = uid_nobody;
#endif
+ *negative = r < 0;
+ return r;
+}
+
+void
+uid_to_chars (uid_t v, char *p, size_t s)
+{
+ to_chars (v < 0, (uintmax_t) v, sizeof v, uid_substitute, p, s, "uid_t");
+}
+
+void
+uintmax_to_chars (uintmax_t v, char *p, size_t s)
+{
+ to_chars (0, v, sizeof v, 0, p, s, "uintmax_t");
+}
+\f
+/* Writing routines. */
+
+/*-----------------------------------------------------------------------.
+| Just zeroes out the buffer so we don't confuse ourselves with leftover |
+| data. |
+`-----------------------------------------------------------------------*/
+
+static void
+clear_buffer (char *buffer)
+{
+ memset (buffer, 0, BLOCKSIZE);
+}
+
+/*-------------------------------------------------------------------------.
+| Write the EOT block(s). We zero at least two blocks, through |
+| the end of the record. Old tar, as previous versions of GNU tar, writes |
+| garbage after two zeroed blocks. |
+`-------------------------------------------------------------------------*/
+
+void
+write_eot (void)
+{
+ union block *pointer = find_next_block ();
+ memset (pointer->buffer, 0, BLOCKSIZE);
+ set_next_block_after (pointer);
+ pointer = find_next_block ();
+ memset (pointer->buffer, 0, available_space_after (pointer));
+ set_next_block_after (pointer);
+}
+
+/*-----------------------------------------------------.
+| Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. |
+`-----------------------------------------------------*/
+
+/* FIXME: Cross recursion between start_header and write_long! */
+
+static union block *start_header PARAMS ((const char *, struct stat *));
+
+static void
+write_long (const char *p, char type)
+{
+ size_t size = strlen (p) + 1;
+ size_t bufsize;
+ union block *header;
+ struct stat foo;
+
+ memset (&foo, 0, sizeof foo);
+ foo.st_size = size;
+
+ header = start_header ("././@LongLink", &foo);
+ header->header.typeflag = type;
+ finish_header (header);
+
+ header = find_next_block ();
+
+ bufsize = available_space_after (header);
+
+ while (bufsize < size)
{
- badperror:
- msg_perror ("can't add file %s", p);
- badfile:
- if (!f_ignore_failed_read || critical_error)
- errors++;
- return;
+ memcpy (header->buffer, p, bufsize);
+ p += bufsize;
+ size -= bufsize;
+ set_next_block_after (header + (bufsize - 1) / BLOCKSIZE);
+ header = find_next_block ();
+ bufsize = available_space_after (header);
}
+ memcpy (header->buffer, p, size);
+ memset (header->buffer + size, 0, bufsize - size);
+ set_next_block_after (header + (size - 1) / BLOCKSIZE);
+}
+\f
+/* Header handling. */
- restore_times.actime = hstat.st_atime;
- restore_times.modtime = hstat.st_mtime;
+/* Make a header block for the file whose stat info is st,
+ and return its address. */
-#ifdef S_ISHIDDEN
- if (S_ISHIDDEN (hstat.st_mode))
+static union block *
+start_header (const char *name, struct stat *st)
+{
+ union block *header;
+
+ if (!absolute_names_option)
{
- char *new = (char *) alloca (strlen (p) + 2);
- if (new)
+ size_t prefix_len = FILESYSTEM_PREFIX_LEN (name);
+
+ if (prefix_len)
{
- strcpy (new, p);
- strcat (new, "@");
- p = new;
+ static int warned_once;
+ if (!warned_once)
+ {
+ warned_once = 1;
+ WARN ((0, 0, _("Removing `%.*s' prefix from archive names"),
+ (int) prefix_len, name));
+ }
+ name += prefix_len;
}
- }
-#endif
- /* See if we only want new files, and check if this one is too old to
- put in the archive. */
- if (f_new_files
- && !f_gnudump
- && new_time > hstat.st_mtime
- && !S_ISDIR (hstat.st_mode)
- && (f_new_files > 1 || new_time > hstat.st_ctime))
- {
- if (curdev < 0)
+ while (*name == '/')
{
- msg ("%s: is unchanged; not dumped", p);
+ static int warned_once;
+ if (!warned_once)
+ {
+ warned_once = 1;
+ WARN ((0, 0, _("Removing leading `/' from archive names")));
+ }
+ name++;
}
- return;
}
-#ifndef __MSDOS__
- /* See if we are trying to dump the archive */
- if (ar_dev && hstat.st_dev == ar_dev && hstat.st_ino == ar_ino)
+ if (sizeof header->header.name <= strlen (name))
+ write_long (name, GNUTYPE_LONGNAME);
+ header = find_next_block ();
+ memset (header->buffer, 0, sizeof (union block));
+
+ assign_string (¤t_file_name, name);
+
+ strncpy (header->header.name, name, NAME_FIELD_SIZE);
+ header->header.name[NAME_FIELD_SIZE - 1] = '\0';
+
+ /* Override some stat fields, if requested to do so. */
+
+ if (owner_option != (uid_t) -1)
+ st->st_uid = owner_option;
+ if (group_option != (gid_t) -1)
+ st->st_gid = group_option;
+ if (mode_option)
+ st->st_mode = ((st->st_mode & ~MODE_ALL)
+ | mode_adjust (st->st_mode, mode_option));
+
+ /* Paul Eggert tried the trivial test ($WRITER cf a b; $READER tvf a)
+ for a few tars and came up with the following interoperability
+ matrix:
+
+ WRITER
+ 1 2 3 4 5 6 7 8 9 READER
+ . . . . . . . . . 1 = SunOS 4.2 tar
+ # . . # # . . # # 2 = NEC SVR4.0.2 tar
+ . . . # # . . # . 3 = Solaris 2.1 tar
+ . . . . . . . . . 4 = GNU tar 1.11.1
+ . . . . . . . . . 5 = HP-UX 8.07 tar
+ . . . . . . . . . 6 = Ultrix 4.1
+ . . . . . . . . . 7 = AIX 3.2
+ . . . . . . . . . 8 = Hitachi HI-UX 1.03
+ . . . . . . . . . 9 = Omron UNIOS-B 4.3BSD 1.60Beta
+
+ . = works
+ # = ``impossible file type''
+
+ The following mask for old archive removes the `#'s in column 4
+ above, thus making GNU tar both a universal donor and a universal
+ acceptor for Paul's test. */
+
+ if (archive_format == V7_FORMAT)
+ MODE_TO_CHARS (st->st_mode & MODE_ALL, header->header.mode);
+ else
+ MODE_TO_CHARS (st->st_mode, header->header.mode);
+
+ UID_TO_CHARS (st->st_uid, header->header.uid);
+ GID_TO_CHARS (st->st_gid, header->header.gid);
+ OFF_TO_CHARS (st->st_size, header->header.size);
+ TIME_TO_CHARS (st->st_mtime, header->header.mtime);
+
+ if (incremental_option)
+ if (archive_format == OLDGNU_FORMAT)
+ {
+ TIME_TO_CHARS (st->st_atime, header->oldgnu_header.atime);
+ TIME_TO_CHARS (st->st_ctime, header->oldgnu_header.ctime);
+ }
+
+ header->header.typeflag = archive_format == V7_FORMAT ? AREGTYPE : REGTYPE;
+
+ switch (archive_format)
{
- msg ("%s is the archive; not dumped", p);
- return;
+ case V7_FORMAT:
+ break;
+
+ case OLDGNU_FORMAT:
+ /* Overwrite header->header.magic and header.version in one blow. */
+ strcpy (header->header.magic, OLDGNU_MAGIC);
+ break;
+
+ case POSIX_FORMAT:
+ case GNU_FORMAT:
+ strncpy (header->header.magic, TMAGIC, TMAGLEN);
+ strncpy (header->header.version, TVERSION, TVERSLEN);
+ break;
+
+ default:
+ abort ();
}
-#endif
- /*
- * Check for multiple links.
- *
- * We maintain a list of all such files that we've written so
- * far. Any time we see another, we check the list and
- * avoid dumping the data again if we've done it once already.
- */
- if (hstat.st_nlink > 1
- && (S_ISREG (hstat.st_mode)
-#ifdef S_ISCTG
- || S_ISCTG (hstat.st_mode)
-#endif
-#ifdef S_ISCHR
- || S_ISCHR (hstat.st_mode)
-#endif
-#ifdef S_ISBLK
- || S_ISBLK (hstat.st_mode)
-#endif
-#ifdef S_ISFIFO
- || S_ISFIFO (hstat.st_mode)
-#endif
- ))
+
+ if (archive_format == V7_FORMAT || numeric_owner_option)
{
- register struct link *lp;
+ /* header->header.[ug]name are left as the empty string. */
+ }
+ else
+ {
+ uid_to_uname (st->st_uid, header->header.uname);
+ gid_to_gname (st->st_gid, header->header.gname);
+ }
- /* First quick and dirty. Hashing, etc later FIXME */
- for (lp = linklist; lp; lp = lp->next)
- {
- if (lp->ino == hstat.st_ino &&
- lp->dev == hstat.st_dev)
- {
- char *link_name = lp->name;
+ return header;
+}
- /* We found a link. */
- while (!f_absolute_paths && *link_name == '/')
- {
- static int link_warn = 0;
+/*-------------------------------------------------------------------------.
+| Finish off a filled-in header block and write it out. We also print the |
+| file name and/or full info if verbose is on. |
+`-------------------------------------------------------------------------*/
- if (!link_warn)
- {
- msg ("Removing leading / from absolute links");
- link_warn++;
- }
- link_name++;
- }
- if (link_name - lp->name >= NAMSIZ)
- write_long (link_name, LF_LONGLINK);
- current_link_name = link_name;
+void
+finish_header (union block *header)
+{
+ size_t i;
+ int sum;
+ char *p;
- hstat.st_size = 0;
- header = start_header (p, &hstat);
- if (header == NULL)
- {
- critical_error = 1;
- goto badfile;
- }
- strncpy (header->header.arch_linkname,
- link_name, NAMSIZ);
+ memcpy (header->header.chksum, CHKBLANKS, sizeof header->header.chksum);
- /* Force null truncated */
- header->header.arch_linkname[NAMSIZ - 1] = 0;
+ sum = 0;
+ p = header->buffer;
+ for (i = sizeof *header; i-- != 0; )
+ /* We can't use unsigned char here because of old compilers, e.g. V7. */
+ sum += 0xFF & *p++;
- header->header.linkflag = LF_LINK;
- finish_header (header);
- /* FIXME: Maybe remove from list after all links found? */
- if (f_remove_files)
- {
- if (unlink (p) == -1)
- msg_perror ("cannot remove %s", p);
- }
- return; /* We dumped it */
- }
- }
+ /* Fill in the checksum field. It's formatted differently from the
+ other fields: it has [6] digits, a null, then a space -- rather than
+ digits, then a null. We use to_chars.
+ The final space is already there, from
+ checksumming, and to_chars doesn't modify it.
- /* Not found. Add it to the list of possible links. */
- lp = (struct link *) ck_malloc ((unsigned) (sizeof (struct link) + strlen (p)));
- if (!lp)
- {
- if (!nolinks)
- {
- msg (
- "no memory for links, they will be dumped as separate files");
- nolinks++;
- }
- }
- lp->ino = hstat.st_ino;
- lp->dev = hstat.st_dev;
- strcpy (lp->name, p);
- lp->next = linklist;
- linklist = lp;
+ This is a fast way to do:
+
+ sprintf(header->header.chksum, "%6o", sum); */
+
+ uintmax_to_chars ((uintmax_t) sum, header->header.chksum, 7);
+
+ if (verbose_option
+ && header->header.typeflag != GNUTYPE_LONGLINK
+ && header->header.typeflag != GNUTYPE_LONGNAME)
+ {
+ /* These globals are parameters to print_header, sigh. */
+
+ current_header = header;
+ /* current_stat is already set up. */
+ current_format = archive_format;
+ print_header ();
}
- /*
- * This is not a link to a previously dumped file, so dump it.
- */
- if (S_ISREG (hstat.st_mode)
-#ifdef S_ISCTG
- || S_ISCTG (hstat.st_mode)
-#endif
- )
+ set_next_block_after (header);
+}
+\f
+/* Sparse file processing. */
+
+/*-------------------------------------------------------------------------.
+| Takes a blockful of data and basically cruises through it to see if it's |
+| made *entirely* of zeros, returning a 0 the instant it finds something |
+| that is a nonzero, i.e., useful data. |
+`-------------------------------------------------------------------------*/
+
+static int
+zero_block_p (char *buffer)
+{
+ int counter;
+
+ for (counter = 0; counter < BLOCKSIZE; counter++)
+ if (buffer[counter] != '\0')
+ return 0;
+ return 1;
+}
+
+/*---.
+| ? |
+`---*/
+
+static void
+init_sparsearray (void)
+{
+ int counter;
+
+ sp_array_size = 10;
+
+ /* Make room for our scratch space -- initially is 10 elts long. */
+
+ sparsearray = xmalloc (sp_array_size * sizeof (struct sp_array));
+ for (counter = 0; counter < sp_array_size; counter++)
{
- int f; /* File descriptor */
- long bufsize, count;
- long sizeleft;
- register union record *start;
- int header_moved;
- char isextended = 0;
- int upperbound;
- /* int end_nulls = 0; */
-
- header_moved = 0;
-
-#ifdef BSD42
- if (f_sparse_files)
- {
- /*
- * JK - This is the test for sparseness: whether the
- * "size" of the file matches the number of blocks
- * allocated for it. If there is a smaller number
- * of blocks that would be necessary to accommodate
- * a file of this size, we have a sparse file, i.e.,
- * at least one of those records in the file is just
- * a useless hole.
- */
-#ifdef hpux /* Nice of HPUX to gratuitiously change it, huh? - mib */
- if (hstat.st_size - (hstat.st_blocks * 1024) > 1024)
-#else
- if (hstat.st_size - (hstat.st_blocks * RECORDSIZE) > RECORDSIZE)
-#endif
- {
- int filesize = hstat.st_size;
- register int i;
+ sparsearray[counter].offset = 0;
+ sparsearray[counter].numbytes = 0;
+ }
+}
- header = start_header (p, &hstat);
- if (header == NULL)
- {
- critical_error = 1;
- goto badfile;
- }
- header->header.linkflag = LF_SPARSE;
- header_moved++;
-
- /*
- * Call the routine that figures out the
- * layout of the sparse file in question.
- * UPPERBOUND is the index of the last
- * element of the "sparsearray," i.e.,
- * the number of elements it needed to
- * describe the file.
- */
-
- upperbound = deal_with_sparse (p, header);
-
- /*
- * See if we'll need an extended header
- * later
- */
- if (upperbound > SPARSE_IN_HDR - 1)
- header->header.isextended++;
- /*
- * We store the "real" file size so
- * we can show that in case someone wants
- * to list the archive, i.e., tar tvf <file>.
- * It might be kind of disconcerting if the
- * shrunken file size was the one that showed
- * up.
- */
- to_oct ((long) hstat.st_size, 1 + 12,
- header->header.realsize);
-
- /*
- * This will be the new "size" of the
- * file, i.e., the size of the file
- * minus the records of holes that we're
- * skipping over.
- */
-
- find_new_file_size (&filesize, upperbound);
- hstat.st_size = filesize;
- to_oct ((long) filesize, 1 + 12,
- header->header.size);
- /* to_oct((long) end_nulls, 1+12,
- header->header.ending_blanks);*/
-
- for (i = 0; i < SPARSE_IN_HDR; i++)
- {
- if (!sparsearray[i].numbytes)
- break;
- to_oct (sparsearray[i].offset, 1 + 12,
- header->header.sp[i].offset);
- to_oct (sparsearray[i].numbytes, 1 + 12,
- header->header.sp[i].numbytes);
- }
+/*---.
+| ? |
+`---*/
- }
- }
-#else
- upperbound = SPARSE_IN_HDR - 1;
-#endif
+static void
+find_new_file_size (off_t *filesize, int highest_index)
+{
+ int counter;
+
+ *filesize = 0;
+ for (counter = 0;
+ sparsearray[counter].numbytes && counter <= highest_index;
+ counter++)
+ *filesize += sparsearray[counter].numbytes;
+}
+
+/*-----------------------------------------------------------------------.
+| Make one pass over the file NAME, studying where any non-zero data is, |
+| that is, how far into the file each instance of data is, and how many |
+| bytes are there. Save this information in the sparsearray, which will |
+| later be translated into header information. |
+`-----------------------------------------------------------------------*/
+
+/* There is little point in trimming small amounts of null data at the head
+ and tail of blocks, only avoid dumping full null blocks. */
+
+/* FIXME: this routine might accept bits of algorithmic cleanup, it is
+ too kludgey for my taste... */
+
+static int
+deal_with_sparse (char *name, union block *header)
+{
+ size_t numbytes = 0;
+ off_t offset = 0;
+ int file;
+ int sparse_index = 0;
+ ssize_t count;
+ char buffer[BLOCKSIZE];
+
+ if (archive_format == OLDGNU_FORMAT)
+ header->oldgnu_header.isextended = 0;
+
+ if (file = open (name, O_RDONLY), file < 0)
+ /* This problem will be caught later on, so just return. */
+ return 0;
+
+ init_sparsearray ();
+ clear_buffer (buffer);
+
+ while (count = safe_read (file, buffer, sizeof buffer), count != 0)
+ {
+ /* Realloc the scratch area as necessary. FIXME: should reallocate
+ only at beginning of a new instance of non-zero data. */
- sizeleft = hstat.st_size;
- /* Don't bother opening empty, world readable files. */
- if (sizeleft > 0 || 0444 != (0444 & hstat.st_mode))
+ if (sparse_index > sp_array_size - 1)
{
- f = open (p, O_RDONLY | O_BINARY);
- if (f < 0)
- goto badperror;
+ sparsearray =
+ xrealloc (sparsearray,
+ 2 * sp_array_size * sizeof (struct sp_array));
+ sp_array_size *= 2;
}
+
+ /* Process one block. */
+
+ if (count == sizeof buffer)
+
+ if (zero_block_p (buffer))
+ {
+ if (numbytes)
+ {
+ sparsearray[sparse_index++].numbytes = numbytes;
+ numbytes = 0;
+ }
+ }
+ else
+ {
+ if (!numbytes)
+ sparsearray[sparse_index].offset = offset;
+ numbytes += count;
+ }
+
else
- {
- f = -1;
- }
- /* If the file is sparse, we've already taken care of this */
- if (!header_moved)
+ /* Since count < sizeof buffer, we have the last bit of the file. */
+
+ if (!zero_block_p (buffer))
+ {
+ if (!numbytes)
+ sparsearray[sparse_index].offset = offset;
+ numbytes += count;
+ }
+ else
+ /* The next two lines are suggested by Andreas Degert, who says
+ they are required for trailing full blocks to be written to the
+ archive, when all zeroed. Yet, it seems to me that the case
+ does not apply. Further, at restore time, the file is not as
+ sparse as it should. So, some serious cleanup is *also* needed
+ in this area. Just one more... :-(. FIXME. */
+ if (numbytes)
+ numbytes += count;
+
+ /* Prepare for next block. */
+
+ offset += count;
+ /* FIXME: do not clear unless necessary. */
+ clear_buffer (buffer);
+ }
+
+ if (numbytes)
+ sparsearray[sparse_index++].numbytes = numbytes;
+ else
+ {
+ sparsearray[sparse_index].offset = offset - 1;
+ sparsearray[sparse_index++].numbytes = 1;
+ }
+
+ close (file);
+ return sparse_index - 1;
+}
+
+/*---.
+| ? |
+`---*/
+
+static int
+finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name)
+{
+ union block *start;
+ size_t bufsize;
+ int sparse_index = 0;
+ ssize_t count;
+
+ while (*sizeleft > 0)
+ {
+ start = find_next_block ();
+ memset (start->buffer, 0, BLOCKSIZE);
+ bufsize = sparsearray[sparse_index].numbytes;
+ if (!bufsize)
{
- header = start_header (p, &hstat);
- if (header == NULL)
- {
- if (f >= 0)
- (void) close (f);
- critical_error = 1;
- goto badfile;
- }
+ /* We blew it, maybe. */
+ char buf1[UINTMAX_STRSIZE_BOUND];
+ char buf2[UINTMAX_STRSIZE_BOUND];
+
+ ERROR ((0, 0, _("Wrote %s of %s bytes to file %s"),
+ STRINGIFY_BIGINT (fullsize - *sizeleft, buf1),
+ STRINGIFY_BIGINT (fullsize, buf2),
+ name));
+ break;
}
-#ifdef S_ISCTG
- /* Mark contiguous files, if we support them */
- if (f_standard && S_ISCTG (hstat.st_mode))
+
+ if (lseek (file, sparsearray[sparse_index++].offset, SEEK_SET) < 0)
{
- header->header.linkflag = LF_CONTIG;
+ char buf[UINTMAX_STRSIZE_BOUND];
+ ERROR ((0, errno, _("lseek error at byte %s in file %s"),
+ STRINGIFY_BIGINT (sparsearray[sparse_index - 1].offset, buf),
+ name));
+ break;
}
-#endif
- isextended = header->header.isextended;
- save_linkflag = header->header.linkflag;
- finish_header (header);
- if (isextended)
- {
- /* int sum = 0;*/
- register int i;
- /* register union record *exhdr;*/
- /* int arraybound = SPARSE_EXT_HDR;*/
- /* static */ int index_offset = SPARSE_IN_HDR;
- extend:exhdr = findrec ();
+ /* If the number of bytes to be written here exceeds the size of
+ the temporary buffer, do it in steps. */
- if (exhdr == NULL)
- {
- critical_error = 1;
- goto badfile;
- }
- bzero (exhdr->charptr, RECORDSIZE);
- for (i = 0; i < SPARSE_EXT_HDR; i++)
+ while (bufsize > BLOCKSIZE)
+ {
+#if 0
+ if (amount_read)
{
- if (i + index_offset > upperbound)
- break;
- to_oct ((long) sparsearray[i + index_offset].numbytes,
- 1 + 12,
- exhdr->ext_hdr.sp[i].numbytes);
- to_oct ((long) sparsearray[i + index_offset].offset,
- 1 + 12,
- exhdr->ext_hdr.sp[i].offset);
+ count = safe_read (file, start->buffer + amount_read,
+ BLOCKSIZE - amount_read);
+ bufsize -= BLOCKSIZE - amount_read;
+ amount_read = 0;
+ set_next_block_after (start);
+ start = find_next_block ();
+ memset (start->buffer, 0, BLOCKSIZE);
}
- userec (exhdr);
- /* sum += i;
- if (sum < upperbound)
- goto extend;*/
- if (index_offset + i <= upperbound)
+#endif
+ /* Store the data. */
+
+ count = safe_read (file, start->buffer, BLOCKSIZE);
+ if (count < 0)
{
- index_offset += i;
- exhdr->ext_hdr.isextended++;
- goto extend;
+ char buf[UINTMAX_STRSIZE_BOUND];
+ ERROR ((0, errno,
+ _("Read error at byte %s, reading %lu bytes, in file %s"),
+ STRINGIFY_BIGINT (fullsize - *sizeleft, buf),
+ (unsigned long) bufsize, name));
+ return 1;
}
+ bufsize -= count;
+ *sizeleft -= count;
+ set_next_block_after (start);
+ start = find_next_block ();
+ memset (start->buffer, 0, BLOCKSIZE);
+ }
+
+ {
+ char buffer[BLOCKSIZE];
+ clear_buffer (buffer);
+ count = safe_read (file, buffer, bufsize);
+ memcpy (start->buffer, buffer, BLOCKSIZE);
+ }
+
+ if (count < 0)
+ {
+ char buf[UINTMAX_STRSIZE_BOUND];
+
+ ERROR ((0, errno,
+ _("Read error at byte %s, reading %lu bytes, in file %s"),
+ STRINGIFY_BIGINT (fullsize - *sizeleft, buf),
+ (unsigned long) bufsize, name));
+ return 1;
}
- if (save_linkflag == LF_SPARSE)
+#if 0
+ if (amount_read >= BLOCKSIZE)
{
- if (finish_sparse_file (f, &sizeleft, hstat.st_size, p))
- goto padit;
+ amount_read = 0;
+ set_next_block_after (start + (count - 1) / BLOCKSIZE);
+ if (count != bufsize)
+ {
+ ERROR ((0, 0,
+ _("File %s shrunk, padding with zeros"),
+ name));
+ return 1;
+ }
+ start = find_next_block ();
}
else
- while (sizeleft > 0)
- {
+ amount_read += bufsize;
+#endif
+ *sizeleft -= count;
+ set_next_block_after (start);
- if (f_multivol)
- {
- save_name = p;
- save_sizeleft = sizeleft;
- save_totsize = hstat.st_size;
- }
- start = findrec ();
+ }
+ free (sparsearray);
+#if 0
+ set_next_block_after (start + (count - 1) / BLOCKSIZE);
+#endif
+ return 0;
+}
+\f
+/* Main functions of this module. */
- bufsize = endofrecs ()->charptr - start->charptr;
+/*---.
+| ? |
+`---*/
- if (sizeleft < bufsize)
- {
- /* Last read -- zero out area beyond */
- bufsize = (int) sizeleft;
- count = bufsize % RECORDSIZE;
- if (count)
- bzero (start->charptr + sizeleft,
- (int) (RECORDSIZE - count));
- }
- count = read (f, start->charptr, bufsize);
- if (count < 0)
- {
- msg_perror ("read error at byte %ld, reading\
- %d bytes, in file %s", hstat.st_size - sizeleft, bufsize, p);
- goto padit;
- }
- sizeleft -= count;
+void
+create_archive (void)
+{
+ char *p;
- /* This is nonportable (the type of userec's arg). */
- userec (start + (count - 1) / RECORDSIZE);
+ open_archive (ACCESS_WRITE);
- if (count == bufsize)
- continue;
- msg ("file %s shrunk by %d bytes, padding with zeros.", p, sizeleft);
- goto padit; /* Short read */
+ if (incremental_option)
+ {
+ char *buffer = xmalloc (PATH_MAX);
+ const char *q;
+ char *bufp;
+
+ collect_and_sort_names ();
+
+ while (p = name_from_list (), p)
+ if (!excluded_name (p))
+ dump_file (p, -1, (dev_t) 0);
+
+ blank_name_list ();
+ while (p = name_from_list (), p)
+ if (!excluded_name (p))
+ {
+ strcpy (buffer, p);
+ if (p[strlen (p) - 1] != '/')
+ strcat (buffer, "/");
+ bufp = buffer + strlen (buffer);
+ q = gnu_list_name->dir_contents;
+ if (q)
+ for (; *q; q += strlen (q) + 1)
+ if (*q == 'Y')
+ {
+ strcpy (bufp, q + 1);
+ dump_file (buffer, -1, (dev_t) 0);
+ }
}
+ free (buffer);
+ }
+ else
+ {
+ while (p = name_next (1), p)
+ if (!excluded_name (p))
+ dump_file (p, 1, (dev_t) 0);
+ }
+
+ write_eot ();
+ close_archive ();
- if (f_multivol)
- save_name = 0;
+ if (listed_incremental_option)
+ write_directory_file ();
+}
- if (f >= 0)
- (void) close (f);
+/* Dump a single file, recursing on directories. P is the file name
+ to dump. TOP_LEVEL tells whether this is a top-level call; zero
+ means no, positive means yes, and negative means an incremental
+ dump. PARENT_DEVICE is the device of P's
+ parent directory; it is examined only if TOP_LEVEL is zero.
- if (f_remove_files)
- {
- if (unlink (p) == -1)
- msg_perror ("cannot remove %s", p);
- }
- if (f_atime_preserve)
- utime (p, &restore_times);
+ Set global CURRENT_STAT to stat output for this file. */
+
+/* FIXME: One should make sure that for *every* path leading to setting
+ exit_status to failure, a clear diagnostic has been issued. */
+
+void
+dump_file (char *p, int top_level, dev_t parent_device)
+{
+ union block *header;
+ char type;
+ union block *exhdr;
+ char save_typeflag;
+ time_t original_ctime;
+ struct utimbuf restore_times;
+
+ /* FIXME: `header' and `upperbound' might be used uninitialized in this
+ function. Reported by Bruno Haible. */
+
+ if (interactive_option && !confirm ("add", p))
+ return;
+
+ if (deref_stat (dereference_option, p, ¤t_stat) != 0)
+ {
+ WARN ((0, errno, _("Cannot add file %s"), p));
+ if (!ignore_failed_read_option)
+ exit_status = TAREXIT_FAILURE;
return;
+ }
+
+ original_ctime = current_stat.st_ctime;
+ restore_times.actime = current_stat.st_atime;
+ restore_times.modtime = current_stat.st_mtime;
- /*
- * File shrunk or gave error, pad out tape to match
- * the size we specified in the header.
- */
- padit:
- while (sizeleft > 0)
+#ifdef S_ISHIDDEN
+ if (S_ISHIDDEN (current_stat.st_mode))
+ {
+ char *new = (char *) alloca (strlen (p) + 2);
+ if (new)
{
- save_sizeleft = sizeleft;
- start = findrec ();
- bzero (start->charptr, RECORDSIZE);
- userec (start);
- sizeleft -= RECORDSIZE;
+ strcpy (new, p);
+ strcat (new, "@");
+ p = new;
}
- if (f_multivol)
- save_name = 0;
- if (f >= 0)
- (void) close (f);
- if (f_atime_preserve)
- utime (p, &restore_times);
+ }
+#endif
+
+ /* See if we want only new files, and check if this one is too old to
+ put in the archive. */
+
+ if ((0 < top_level || !incremental_option)
+ && !S_ISDIR (current_stat.st_mode)
+ && current_stat.st_mtime < newer_mtime_option
+ && (!after_date_option || current_stat.st_ctime < newer_ctime_option))
+ {
+ if (0 < top_level)
+ WARN ((0, 0, _("%s: is unchanged; not dumped"), p));
+ /* FIXME: recheck this return. */
return;
}
-#ifdef S_ISLNK
- else if (S_ISLNK (hstat.st_mode))
+#if !MSDOS
+ /* See if we are trying to dump the archive. */
+
+ if (ar_dev && current_stat.st_dev == ar_dev && current_stat.st_ino == ar_ino)
{
- int size;
- char *buf = alloca (PATH_MAX + 1);
-
- size = readlink (p, buf, PATH_MAX + 1);
- if (size < 0)
- goto badperror;
- buf[size] = '\0';
- if (size >= NAMSIZ)
- write_long (buf, LF_LONGLINK);
- current_link_name = buf;
-
- hstat.st_size = 0; /* Force 0 size on symlink */
- header = start_header (p, &hstat);
- if (header == NULL)
- {
- critical_error = 1;
- goto badfile;
- }
- strncpy (header->header.arch_linkname, buf, NAMSIZ);
- header->header.arch_linkname[NAMSIZ - 1] = '\0';
- header->header.linkflag = LF_SYMLINK;
- finish_header (header); /* Nothing more to do to it */
- if (f_remove_files)
- {
- if (unlink (p) == -1)
- msg_perror ("cannot remove %s", p);
- }
+ WARN ((0, 0, _("%s is the archive; not dumped"), p));
return;
}
#endif
- else if (S_ISDIR (hstat.st_mode))
+ if (S_ISDIR (current_stat.st_mode))
{
- register DIR *dirp;
- register struct dirent *d;
+ DIR *directory;
+ struct dirent *entry;
char *namebuf;
- int buflen;
- register int len;
- int our_device = hstat.st_dev;
+ size_t buflen;
+ size_t len;
+ dev_t our_device = current_stat.st_dev;
+
+ /* If this tar program is installed suid root, like for Amanda, the
+ access might look like denied, while it is not really.
+
+ FIXME: I have the feeling this test is done too early. Couldn't it
+ just be bundled in later actions? I guess that the proper support
+ of --ignore-failed-read is the key of the current writing. */
+
+ if (access (p, R_OK) == -1 && geteuid () != 0)
+ {
+ WARN ((0, errno, _("Cannot add directory %s"), p));
+ if (!ignore_failed_read_option)
+ exit_status = TAREXIT_FAILURE;
+ return;
+ }
+
+ /* Build new prototype name. Ensure exactly one trailing slash. */
- /* Build new prototype name */
len = strlen (p);
- buflen = len + NAMSIZ;
- namebuf = ck_malloc (buflen + 1);
+ buflen = len + NAME_FIELD_SIZE;
+ namebuf = xmalloc (buflen + 1);
strncpy (namebuf, p, buflen);
- while (len >= 1 && '/' == namebuf[len - 1])
- len--; /* Delete trailing slashes */
- namebuf[len++] = '/'; /* Now add exactly one back */
- namebuf[len] = '\0'; /* Make sure null-terminated */
-
- /*
- * Output directory header record with permissions
- * FIXME, do this AFTER files, to avoid R/O dir problems?
- * If old archive format, don't write record at all.
- */
- if (!f_oldarch)
+ while (len >= 1 && namebuf[len - 1] == '/')
+ len--;
+ namebuf[len++] = '/';
+ namebuf[len] = '\0';
+
+ if (! is_avoided_name (namebuf))
{
- hstat.st_size = 0; /* Force 0 size on dir */
- /*
- * If people could really read standard archives,
- * this should be: (FIXME)
- header = start_header(f_standard? p: namebuf, &hstat);
- * but since they'd interpret LF_DIR records as
- * regular files, we'd better put the / on the name.
- */
- header = start_header (namebuf, &hstat);
- if (header == NULL)
- {
- critical_error = 1;
- goto badfile; /* eg name too long */
- }
+ /* The condition above used to be "archive_format != V7_FORMAT".
+ GNU tar was not writing directory blocks at all. Daniel Trinkle
+ writes: ``All old versions of tar I have ever seen have
+ correctly archived an empty directory. The really old ones I
+ checked included HP-UX 7 and Mt. Xinu More/BSD. There may be
+ some subtle reason for the exclusion that I don't know, but the
+ current behavior is broken.'' I do not know those subtle
+ reasons either, so until these are reported (anew?), just allow
+ directory blocks to be written even with old archives. */
+
+ current_stat.st_size = 0; /* force 0 size on dir */
- if (f_gnudump)
- header->header.linkflag = LF_DUMPDIR;
- else if (f_standard)
- header->header.linkflag = LF_DIR;
+ /* FIXME: If people could really read standard archives, this
+ should be:
- /* If we're gnudumping, we aren't done yet so don't close it. */
- if (!f_gnudump)
- finish_header (header); /* Done with directory header */
+ header
+ = start_header (standard_option ? p : namebuf, ¤t_stat);
+
+ but since they'd interpret DIRTYPE blocks as regular
+ files, we'd better put the / on the name. */
+
+ header = start_header (namebuf, ¤t_stat);
+
+ if (incremental_option)
+ header->header.typeflag = GNUTYPE_DUMPDIR;
+ else /* if (standard_option) */
+ header->header.typeflag = DIRTYPE;
+
+ /* If we're gnudumping, we aren't done yet so don't close it. */
+
+ if (!incremental_option)
+ finish_header (header); /* done with directory header */
}
- if (f_gnudump)
+ if (incremental_option && gnu_list_name->dir_contents)
{
- int sizeleft;
- int totsize;
- int bufsize;
- union record *start;
- int count;
- char *buf, *p_buf;
-
- buf = gnu_list_name->dir_contents; /* FOO */
+ off_t sizeleft;
+ off_t totsize;
+ size_t bufsize;
+ union block *start;
+ ssize_t count;
+ const char *buffer, *p_buffer;
+
+ buffer = gnu_list_name->dir_contents; /* FOO */
totsize = 0;
- for (p_buf = buf; p_buf && *p_buf;)
+ for (p_buffer = buffer; p_buffer && *p_buffer;)
{
- int tmp;
+ size_t tmp;
- tmp = strlen (p_buf) + 1;
+ tmp = strlen (p_buffer) + 1;
totsize += tmp;
- p_buf += tmp;
+ p_buffer += tmp;
}
totsize++;
- to_oct ((long) totsize, 1 + 12, header->header.size);
+ OFF_TO_CHARS (totsize, header->header.size);
finish_header (header);
- p_buf = buf;
+ p_buffer = buffer;
sizeleft = totsize;
while (sizeleft > 0)
{
- if (f_multivol)
+ if (multi_volume_option)
{
- save_name = p;
+ assign_string (&save_name, p);
save_sizeleft = sizeleft;
save_totsize = totsize;
}
- start = findrec ();
- bufsize = endofrecs ()->charptr - start->charptr;
+ start = find_next_block ();
+ bufsize = available_space_after (start);
if (sizeleft < bufsize)
{
bufsize = sizeleft;
- count = bufsize % RECORDSIZE;
+ count = bufsize % BLOCKSIZE;
if (count)
- bzero (start->charptr + sizeleft, RECORDSIZE - count);
+ memset (start->buffer + sizeleft, 0, BLOCKSIZE - count);
}
- bcopy (p_buf, start->charptr, bufsize);
+ memcpy (start->buffer, p_buffer, bufsize);
sizeleft -= bufsize;
- p_buf += bufsize;
- userec (start + (bufsize - 1) / RECORDSIZE);
+ p_buffer += bufsize;
+ set_next_block_after (start + (bufsize - 1) / BLOCKSIZE);
}
- if (f_multivol)
- save_name = 0;
- if (f_atime_preserve)
+ if (multi_volume_option)
+ assign_string (&save_name, 0);
+ if (atime_preserve_option)
utime (p, &restore_times);
return;
}
- /* Now output all the files in the directory */
-#if 0
- if (f_dironly)
- return; /* Unless the cmdline said not to */
-#endif
- /*
- * See if we are crossing from one file system to another,
- * and avoid doing so if the user only wants to dump one file system.
- */
- if (f_local_filesys && !toplevel && curdev != hstat.st_dev)
+ /* See if we are about to recurse into a directory, and avoid doing
+ so if the user wants that we do not descend into directories. */
+
+ if (no_recurse_option)
+ return;
+
+ /* See if we are crossing from one file system to another, and
+ avoid doing so if the user only wants to dump one file system. */
+
+ if (one_file_system_option && !top_level
+ && parent_device != current_stat.st_dev)
{
- if (f_verbose)
- msg ("%s: is on a different filesystem; not dumped", p);
+ if (verbose_option)
+ WARN ((0, 0, _("%s: On a different filesystem; not dumped"), p));
return;
}
+ /* Now output all the files in the directory. */
errno = 0;
- dirp = opendir (p);
- if (!dirp)
+
+ directory = opendir (p);
+ if (!directory)
{
- if (errno)
- {
- msg_perror ("can't open directory %s", p);
- }
- else
- {
- msg ("error opening directory %s",
- p);
- }
+ ERROR ((0, errno, _("Cannot open directory %s"), p));
return;
}
- /* Hack to remove "./" from the front of all the file names */
- if (len == 2 && namebuf[0] == '.' && namebuf[1] == '/')
- len = 0;
+ /* FIXME: Should speed this up by cd-ing into the dir. */
- /* Should speed this up by cd-ing into the dir, FIXME */
- while (NULL != (d = readdir (dirp)))
+ while (entry = readdir (directory), entry)
{
- /* Skip . and .. */
- if (is_dot_or_dotdot (d->d_name))
+ /* Skip `.', `..', and excluded file names. */
+
+ if (is_dot_or_dotdot (entry->d_name))
continue;
- if (NLENGTH (d) + len >= buflen)
+ if ((int) NAMLEN (entry) + len >= buflen)
{
- buflen = len + NLENGTH (d);
- namebuf = ck_realloc (namebuf, buflen + 1);
- /* namebuf[len]='\0';
- msg("file name %s%s too long",
- namebuf, d->d_name);
- continue; */
+ buflen = len + NAMLEN (entry);
+ namebuf = xrealloc (namebuf, buflen + 1);
}
- strcpy (namebuf + len, d->d_name);
- if (f_exclude && check_exclude (namebuf))
- continue;
- dump_file (namebuf, our_device, 0);
+ strcpy (namebuf + len, entry->d_name);
+ if (!excluded_name (namebuf))
+ dump_file (namebuf, 0, our_device);
}
- closedir (dirp);
+ closedir (directory);
free (namebuf);
- if (f_atime_preserve)
+ if (atime_preserve_option)
utime (p, &restore_times);
return;
}
-
-#ifdef S_ISCHR
- else if (S_ISCHR (hstat.st_mode))
+ else if (is_avoided_name (p))
+ return;
+ else
{
- type = LF_CHR;
- }
-#endif
+ /* Check for multiple links.
-#ifdef S_ISBLK
- else if (S_ISBLK (hstat.st_mode))
- {
- type = LF_BLK;
- }
-#endif
+ We maintain a list of all such files that we've written so far. Any
+ time we see another, we check the list and avoid dumping the data
+ again if we've done it once already. */
- /* Avoid screwy apollo lossage where S_IFIFO == S_IFSOCK */
-#if (_ISP__M68K == 0) && (_ISP__A88K == 0) && defined(S_ISFIFO)
- else if (S_ISFIFO (hstat.st_mode))
- {
- type = LF_FIFO;
- }
-#endif
+ if (1 < current_stat.st_nlink)
+ {
+ struct link *lp;
-#ifdef S_ISSOCK
- else if (S_ISSOCK (hstat.st_mode))
- {
- type = LF_FIFO;
- }
-#endif
- else
- goto unknown;
+ /* FIXME: First quick and dirty. Hashing, etc later. */
- if (!f_standard)
- goto unknown;
+ for (lp = linklist; lp; lp = lp->next)
+ if (lp->ino == current_stat.st_ino
+ && lp->dev == current_stat.st_dev)
+ {
+ char *link_name = lp->name;
- hstat.st_size = 0; /* Force 0 size */
- header = start_header (p, &hstat);
- if (header == NULL)
- {
- critical_error = 1;
- goto badfile; /* eg name too long */
- }
+ /* We found a link. */
- header->header.linkflag = type;
-#if defined(S_IFBLK) || defined(S_IFCHR)
- if (type != LF_FIFO)
- {
- to_oct ((long) major (hstat.st_rdev), 8,
- header->header.devmajor);
- to_oct ((long) minor (hstat.st_rdev), 8,
- header->header.devminor);
- }
-#endif
+ while (!absolute_names_option && *link_name == '/')
+ {
+ static int warned_once;
+ if (!warned_once)
+ {
+ warned_once = 1;
+ WARN ((0, 0,
+ _("Removing leading `/' from link names")));
+ }
+ link_name++;
+ }
+ if (strlen (link_name) >= NAME_FIELD_SIZE)
+ write_long (link_name, GNUTYPE_LONGLINK);
+ assign_string (¤t_link_name, link_name);
- finish_header (header);
- if (f_remove_files)
- {
- if (unlink (p) == -1)
- msg_perror ("cannot remove %s", p);
- }
- return;
+ current_stat.st_size = 0;
+ header = start_header (p, ¤t_stat);
+ strncpy (header->header.linkname,
+ link_name, NAME_FIELD_SIZE);
-unknown:
- msg ("%s: Unknown file type; file ignored.", p);
-}
+ /* Force null truncated. */
-int
-finish_sparse_file (fd, sizeleft, fullsize, name)
- int fd;
- long *sizeleft, fullsize;
- char *name;
-{
- union record *start;
- char tempbuf[RECORDSIZE];
- int bufsize, sparse_ind = 0, count;
- long pos;
- long nwritten = 0;
+ header->header.linkname[NAME_FIELD_SIZE - 1] = 0;
+ header->header.typeflag = LNKTYPE;
+ finish_header (header);
- while (*sizeleft > 0)
- {
- start = findrec ();
- bzero (start->charptr, RECORDSIZE);
- bufsize = sparsearray[sparse_ind].numbytes;
- if (!bufsize)
- { /* we blew it, maybe */
- msg ("Wrote %ld of %ld bytes to file %s",
- fullsize - *sizeleft, fullsize, name);
- break;
- }
- pos = lseek (fd, sparsearray[sparse_ind++].offset, 0);
- /*
- * If the number of bytes to be written here exceeds
- * the size of the temporary buffer, do it in steps.
- */
- while (bufsize > RECORDSIZE)
- {
- /* if (amt_read) {
- count = read(fd, start->charptr+amt_read, RECORDSIZE-amt_read);
- bufsize -= RECORDSIZE - amt_read;
- amt_read = 0;
- userec(start);
- start = findrec();
- bzero(start->charptr, RECORDSIZE);
- }*/
- /* store the data */
- count = read (fd, start->charptr, RECORDSIZE);
- if (count < 0)
- {
- msg_perror ("read error at byte %ld, reading %d bytes, in file %s",
- fullsize - *sizeleft, bufsize, name);
- return 1;
- }
- bufsize -= count;
- *sizeleft -= count;
- userec (start);
- nwritten += RECORDSIZE; /* XXX */
- start = findrec ();
- bzero (start->charptr, RECORDSIZE);
+ /* FIXME: Maybe remove from list after all links found? */
+
+ if (remove_files_option)
+ if (unlink (p) == -1)
+ ERROR ((0, errno, _("Cannot remove %s"), p));
+
+ /* We dumped it. */
+ return;
+ }
+
+ /* Not found. Add it to the list of possible links. */
+
+ lp = xmalloc (sizeof (struct link) + strlen (p));
+ lp->ino = current_stat.st_ino;
+ lp->dev = current_stat.st_dev;
+ strcpy (lp->name, p);
+ lp->next = linklist;
+ linklist = lp;
}
+ /* This is not a link to a previously dumped file, so dump it. */
- clear_buffer (tempbuf);
- count = read (fd, tempbuf, bufsize);
- bcopy (tempbuf, start->charptr, RECORDSIZE);
- if (count < 0)
+ if (S_ISREG (current_stat.st_mode)
+ || S_ISCTG (current_stat.st_mode))
{
- msg_perror ("read error at byte %ld, reading %d bytes, in file %s",
- fullsize - *sizeleft, bufsize, name);
- return 1;
- }
- /* if (amt_read >= RECORDSIZE) {
- amt_read = 0;
- userec(start+(count-1)/RECORDSIZE);
- if (count != bufsize) {
- msg("file %s shrunk by %d bytes, padding with zeros.", name, sizeleft);
- return 1;
- }
- start = findrec();
- } else
- amt_read += bufsize;*/
- nwritten += count; /* XXX */
- *sizeleft -= count;
- userec (start);
+ int f; /* file descriptor */
+ size_t bufsize;
+ ssize_t count;
+ off_t sizeleft;
+ union block *start;
+ int header_moved;
+ char isextended = 0;
+ int upperbound;
+
+ header_moved = 0;
+
+ if (sparse_option)
+ {
+ /* Check the size of the file against the number of blocks
+ allocated for it, counting both data and indirect blocks.
+ If there is a smaller number of blocks that would be
+ necessary to accommodate a file of this size, this is safe
+ to say that we have a sparse file: at least one of those
+ blocks in the file is just a useless hole. For sparse
+ files not having more hole blocks than indirect blocks, the
+ sparseness will go undetected. */
+
+ /* Bruno Haible sent me these statistics for Linux. It seems
+ that some filesystems count indirect blocks in st_blocks,
+ while others do not seem to:
+
+ minix-fs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18
+ extfs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18
+ ext2fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16
+ msdos-fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16
+
+ Dick Streefland reports the previous numbers as misleading,
+ because ext2fs use 12 direct blocks, while minix-fs uses only
+ 6 direct blocks. Dick gets:
+
+ ext2 size=20480 ls listed blocks=21
+ minix size=20480 ls listed blocks=21
+ msdos size=20480 ls listed blocks=20
+
+ It seems that indirect blocks *are* included in st_blocks.
+ The minix filesystem does not account for phantom blocks in
+ st_blocks, so `du' and `ls -s' give wrong results. So, the
+ --sparse option would not work on a minix filesystem. */
+
+ if (ST_NBLOCKS (current_stat)
+ < (current_stat.st_size / ST_NBLOCKSIZE
+ + (current_stat.st_size % ST_NBLOCKSIZE != 0)))
+ {
+ off_t filesize = current_stat.st_size;
+ int counter;
- }
- free (sparsearray);
- /* printf ("Amount actually written is (I hope) %d.\n", nwritten); */
- /* userec(start+(count-1)/RECORDSIZE);*/
- return 0;
+ header = start_header (p, ¤t_stat);
+ header->header.typeflag = GNUTYPE_SPARSE;
+ header_moved = 1;
-}
+ /* Call the routine that figures out the layout of the
+ sparse file in question. UPPERBOUND is the index of the
+ last element of the "sparsearray," i.e., the number of
+ elements it needed to describe the file. */
-void
-init_sparsearray ()
-{
- register int i;
+ upperbound = deal_with_sparse (p, header);
- sp_array_size = 10;
- /*
- * Make room for our scratch space -- initially is 10 elts long
- */
- sparsearray = (struct sp_array *) ck_malloc (sp_array_size * sizeof (struct sp_array));
- for (i = 0; i < sp_array_size; i++)
- {
- sparsearray[i].offset = 0;
- sparsearray[i].numbytes = 0;
- }
-}
+ /* See if we'll need an extended header later. */
+ if (upperbound > SPARSES_IN_OLDGNU_HEADER - 1)
+ header->oldgnu_header.isextended = 1;
+ /* We store the "real" file size so we can show that in
+ case someone wants to list the archive, i.e., tar tvf
+ <file>. It might be kind of disconcerting if the
+ shrunken file size was the one that showed up. */
-/*
- * Okay, we've got a sparse file on our hands -- now, what we need to do is
- * make a pass through the file and carefully note where any data is, i.e.,
- * we want to find how far into the file each instance of data is, and how
- * many bytes are there. We store this information in the sparsearray,
- * which will later be translated into header information. For now, we use
- * the sparsearray as convenient storage.
- *
- * As a side note, this routine is a mess. If I could have found a cleaner
- * way to do it, I would have. If anyone wants to find a nicer way to do
- * this, feel free.
- */
-
-/* There is little point in trimming small amounts of null data at the */
-/* head and tail of blocks -- it's ok if we only avoid dumping blocks */
-/* of complete null data */
-int
-deal_with_sparse (name, header, nulls_at_end)
- char *name;
- union record *header;
- int nulls_at_end;
-{
- long numbytes = 0;
- long offset = 0;
- /* long save_offset;*/
- int fd;
- /* int current_size = hstat.st_size;*/
- int sparse_ind = 0, cc;
- char buf[RECORDSIZE];
-#if 0
- int read_last_data = 0; /* did we just read the last record? */
-#endif
- int amidst_data = 0;
-
- header->header.isextended = 0;
- /*
- * Can't open the file -- this problem will be caught later on,
- * so just return.
- */
- if ((fd = open (name, O_RDONLY)) < 0)
- return 0;
+ OFF_TO_CHARS (current_stat.st_size,
+ header->oldgnu_header.realsize);
- init_sparsearray ();
- clear_buffer (buf);
+ /* This will be the new "size" of the file, i.e., the size
+ of the file minus the blocks of holes that we're
+ skipping over. */
- while ((cc = read (fd, buf, sizeof buf)) != 0)
- {
+ find_new_file_size (&filesize, upperbound);
+ current_stat.st_size = filesize;
+ OFF_TO_CHARS (filesize, header->header.size);
- if (sparse_ind > sp_array_size - 1)
- {
+ for (counter = 0; counter < SPARSES_IN_OLDGNU_HEADER; counter++)
+ {
+ if (!sparsearray[counter].numbytes)
+ break;
+
+ OFF_TO_CHARS (sparsearray[counter].offset,
+ header->oldgnu_header.sp[counter].offset);
+ SIZE_TO_CHARS (sparsearray[counter].numbytes,
+ header->oldgnu_header.sp[counter].numbytes);
+ }
- /*
- * realloc the scratch area, since we've run out of room --
- */
- sparsearray = (struct sp_array *)
- ck_realloc (sparsearray,
- 2 * sp_array_size * (sizeof (struct sp_array)));
- sp_array_size *= 2;
- }
- if (cc == sizeof buf)
- {
- if (zero_record (buf))
- {
- if (amidst_data)
- {
- sparsearray[sparse_ind++].numbytes
- = numbytes;
- amidst_data = 0;
}
}
else
- { /* !zero_record(buf) */
- if (amidst_data)
- numbytes += cc;
- else
- {
- amidst_data = 1;
- numbytes = cc;
- sparsearray[sparse_ind].offset
- = offset;
- }
- }
- }
- else if (cc < sizeof buf)
- {
- /* This has to be the last bit of the file, so this */
- /* is somewhat shorter than the above. */
- if (!zero_record (buf))
- {
- if (!amidst_data)
- {
- amidst_data = 1;
- numbytes = cc;
- sparsearray[sparse_ind].offset
- = offset;
- }
- else
- numbytes += cc;
- }
- }
- offset += cc;
- clear_buffer (buf);
- }
- if (amidst_data)
- sparsearray[sparse_ind++].numbytes = numbytes;
- else
- {
- sparsearray[sparse_ind].offset = offset-1;
- sparsearray[sparse_ind++].numbytes = 1;
- }
- close (fd);
-
- return sparse_ind - 1;
-}
-
-/*
- * Just zeroes out the buffer so we don't confuse ourselves with leftover
- * data.
- */
-void
-clear_buffer (buf)
- char *buf;
-{
- register int i;
-
- for (i = 0; i < RECORDSIZE; i++)
- buf[i] = '\0';
-}
-
-#if 0 /* I'm leaving this as a monument to Joy Kendall, who wrote it -mib */
-/*
- * JK -
- * This routine takes a character array, and tells where within that array
- * the data can be found. It skips over any zeros, and sets the first
- * non-zero point in the array to be the "start", and continues until it
- * finds non-data again, which is marked as the "end." This routine is
- * mainly for 1) seeing how far into a file we must lseek to data, given
- * that we have a sparse file, and 2) determining the "real size" of the
- * file, i.e., the number of bytes in the sparse file that are data, as
- * opposed to the zeros we are trying to skip.
- */
-where_is_data (from, to, buffer)
- int *from, *to;
- char *buffer;
-{
- register int i = 0;
- register int save_to = *to;
- int amidst_data = 0;
+ upperbound = SPARSES_IN_OLDGNU_HEADER - 1;
+ sizeleft = current_stat.st_size;
- while (!buffer[i])
- i++;
- *from = i;
+ /* Don't bother opening empty, world readable files. Also do not open
+ files when archive is meant for /dev/null. */
- if (*from < 16) /* don't bother */
- *from = 0;
- /* keep going to make sure there isn't more real
- data in this record */
- while (i < RECORDSIZE)
- {
- if (!buffer[i])
- {
- if (amidst_data)
+ if (dev_null_output
+ || (sizeleft == 0
+ && MODE_R == (MODE_R & current_stat.st_mode)))
+ f = -1;
+ else
{
- save_to = i;
- amidst_data = 0;
+ f = open (p, O_RDONLY | O_BINARY);
+ if (f < 0)
+ {
+ if (! top_level && errno == ENOENT)
+ WARN ((0, 0, _("%s: file removed before we read it"), p));
+ else
+ WARN ((0, errno, _("Cannot add file %s"), p));
+ if (!ignore_failed_read_option)
+ exit_status = TAREXIT_FAILURE;
+ return;
+ }
}
- i++;
- }
- else if (buffer[i])
- {
- if (!amidst_data)
- amidst_data = 1;
- i++;
- }
- }
- if (i == RECORDSIZE)
- *to = i;
- else
- *to = save_to;
-
-}
-
-#endif
-/* Note that this routine is only called if zero_record returned true */
-#if 0 /* But we actually don't need it at all. */
-where_is_data (from, to, buffer)
- int *from, *to;
- char *buffer;
-{
- char *fp, *tp;
-
- for (fp = buffer; !*fp; fp++)
- ;
- for (tp = buffer + RECORDSIZE - 1; !*tp; tp--)
- ;
- *from = fp - buffer;
- *to = tp - buffer + 1;
-}
+ /* If the file is sparse, we've already taken care of this. */
-#endif
+ if (!header_moved)
+ header = start_header (p, ¤t_stat);
+ /* Mark contiguous files, if we support them. */
+ if (archive_format != V7_FORMAT && S_ISCTG (current_stat.st_mode))
+ header->header.typeflag = CONTTYPE;
-/*
- * Takes a recordful of data and basically cruises through it to see if
- * it's made *entirely* of zeros, returning a 0 the instant it finds
- * something that is a non-zero, i.e., useful data.
- */
-int
-zero_record (buffer)
- char *buffer;
-{
- register int i;
+ isextended = header->oldgnu_header.isextended;
+ save_typeflag = header->header.typeflag;
+ finish_header (header);
+ if (isextended)
+ {
+ int counter;
+ /* static */ int index_offset = SPARSES_IN_OLDGNU_HEADER;
- for (i = 0; i < RECORDSIZE; i++)
- if (buffer[i] != '\000')
- return 0;
- return 1;
-}
+ extend:
+ exhdr = find_next_block ();
+ memset (exhdr->buffer, 0, BLOCKSIZE);
+ for (counter = 0; counter < SPARSES_IN_SPARSE_HEADER; counter++)
+ {
+ if (counter + index_offset > upperbound)
+ break;
-void
-find_new_file_size (filesize, highest_index)
- int *filesize;
- int highest_index;
-{
- register int i;
+ SIZE_TO_CHARS (sparsearray[counter + index_offset].numbytes,
+ exhdr->sparse_header.sp[counter].numbytes);
+ OFF_TO_CHARS (sparsearray[counter + index_offset].offset,
+ exhdr->sparse_header.sp[counter].offset);
+ }
+ set_next_block_after (exhdr);
+ if (index_offset + counter <= upperbound)
+ {
+ index_offset += counter;
+ exhdr->sparse_header.isextended = 1;
+ goto extend;
+ }
- *filesize = 0;
- for (i = 0; sparsearray[i].numbytes && i <= highest_index; i++)
- *filesize += sparsearray[i].numbytes;
-}
+ }
+ if (save_typeflag == GNUTYPE_SPARSE)
+ {
+ if (f < 0
+ || finish_sparse_file (f, &sizeleft, current_stat.st_size, p))
+ goto padit;
+ }
+ else
+ while (sizeleft > 0)
+ {
+ if (multi_volume_option)
+ {
+ assign_string (&save_name, p);
+ save_sizeleft = sizeleft;
+ save_totsize = current_stat.st_size;
+ }
+ start = find_next_block ();
+
+ bufsize = available_space_after (start);
+
+ if (sizeleft < bufsize)
+ {
+ /* Last read -- zero out area beyond. */
+
+ bufsize = sizeleft;
+ count = bufsize % BLOCKSIZE;
+ if (count)
+ memset (start->buffer + sizeleft, 0, BLOCKSIZE - count);
+ }
+ if (f < 0)
+ count = bufsize;
+ else
+ count = safe_read (f, start->buffer, bufsize);
+ if (count < 0)
+ {
+ char buf[UINTMAX_STRSIZE_BOUND];
+ ERROR ((0, errno,
+ _("Read error at byte %s, reading %lu bytes, in file %s"),
+ STRINGIFY_BIGINT (current_stat.st_size - sizeleft,
+ buf),
+ (unsigned long) bufsize, p));
+ goto padit;
+ }
+ sizeleft -= count;
+
+ /* This is nonportable (the type of set_next_block_after's arg). */
+
+ set_next_block_after (start + (count - 1) / BLOCKSIZE);
+
+ if (count == bufsize)
+ continue;
+ else
+ {
+ char buf[UINTMAX_STRSIZE_BOUND];
+ ERROR ((0, 0,
+ _("File %s shrunk by %s bytes, padding with zeros"),
+ p, STRINGIFY_BIGINT (sizeleft, buf)));
+ goto padit; /* short read */
+ }
+ }
-/*
- * Make a header block for the file name whose stat info is st .
- * Return header pointer for success, NULL if the name is too long.
- */
-union record *
-start_header (name, st)
- char *name;
- register struct stat *st;
-{
- register union record *header;
+ if (multi_volume_option)
+ assign_string (&save_name, 0);
- if (strlen (name) >= NAMSIZ)
- write_long (name, LF_LONGNAME);
+ if (f >= 0)
+ {
+ struct stat final_stat;
+ if (fstat (f, &final_stat) != 0)
+ ERROR ((0, errno, "%s: fstat", p));
+ else if (final_stat.st_ctime != original_ctime)
+ ERROR ((0, 0, _("%s: file changed as we read it"), p));
+ if (close (f) != 0)
+ ERROR ((0, errno, _("%s: close"), p));
+ if (atime_preserve_option)
+ utime (p, &restore_times);
+ }
+ if (remove_files_option)
+ {
+ if (unlink (p) == -1)
+ ERROR ((0, errno, _("Cannot remove %s"), p));
+ }
+ return;
- header = (union record *) findrec ();
- bzero (header->charptr, sizeof (*header)); /* XXX speed up */
+ /* File shrunk or gave error, pad out tape to match the size we
+ specified in the header. */
- /*
- * Check the file name and put it in the record.
- */
- if (!f_absolute_paths)
- {
- static int warned_once = 0;
-#ifdef __MSDOS__
- if (name[1] == ':')
+ padit:
+ while (sizeleft > 0)
+ {
+ save_sizeleft = sizeleft;
+ start = find_next_block ();
+ memset (start->buffer, 0, BLOCKSIZE);
+ set_next_block_after (start);
+ sizeleft -= BLOCKSIZE;
+ }
+ if (multi_volume_option)
+ assign_string (&save_name, 0);
+ if (f >= 0)
+ {
+ close (f);
+ if (atime_preserve_option)
+ utime (p, &restore_times);
+ }
+ return;
+ }
+#ifdef HAVE_READLINK
+ else if (S_ISLNK (current_stat.st_mode))
{
- name += 2;
- if (!warned_once++)
- msg ("Removing drive spec from names in the archive");
+ int size;
+ char *buffer = (char *) alloca (PATH_MAX + 1);
+
+ size = readlink (p, buffer, PATH_MAX + 1);
+ if (size < 0)
+ {
+ WARN ((0, errno, _("Cannot add file %s"), p));
+ if (!ignore_failed_read_option)
+ exit_status = TAREXIT_FAILURE;
+ return;
+ }
+ buffer[size] = '\0';
+ if (size >= NAME_FIELD_SIZE)
+ write_long (buffer, GNUTYPE_LONGLINK);
+ assign_string (¤t_link_name, buffer);
+
+ current_stat.st_size = 0; /* force 0 size on symlink */
+ header = start_header (p, ¤t_stat);
+ strncpy (header->header.linkname, buffer, NAME_FIELD_SIZE);
+ header->header.linkname[NAME_FIELD_SIZE - 1] = '\0';
+ header->header.typeflag = SYMTYPE;
+ finish_header (header); /* nothing more to do to it */
+ if (remove_files_option)
+ {
+ if (unlink (p) == -1)
+ ERROR ((0, errno, _("Cannot remove %s"), p));
+ }
+ return;
}
#endif
- while ('/' == *name)
+ else if (S_ISCHR (current_stat.st_mode))
+ type = CHRTYPE;
+ else if (S_ISBLK (current_stat.st_mode))
+ type = BLKTYPE;
+ else if (S_ISFIFO (current_stat.st_mode)
+ || S_ISSOCK (current_stat.st_mode))
+ type = FIFOTYPE;
+#ifdef S_ISDOOR
+ else if (S_ISDOOR (current_stat.st_mode))
{
- name++; /* Force relative path */
- if (!warned_once++)
- msg ("Removing leading / from absolute path names in the archive.");
+ WARN ((0, 0, _("%s: door ignored"), p));
+ return;
}
- }
- current_file_name = name;
- strncpy (header->header.arch_name, name, NAMSIZ);
- header->header.arch_name[NAMSIZ - 1] = '\0';
-
- to_oct ((long) (f_oldarch ? (st->st_mode & 07777) : st->st_mode),
- 8, header->header.mode);
- to_oct ((long) st->st_uid, 8, header->header.uid);
- to_oct ((long) st->st_gid, 8, header->header.gid);
- to_oct ((long) st->st_size, 1 + 12, header->header.size);
- to_oct ((long) st->st_mtime, 1 + 12, header->header.mtime);
- /* header->header.linkflag is left as null */
- if (f_gnudump)
- {
- to_oct ((long) st->st_atime, 1 + 12, header->header.atime);
- to_oct ((long) st->st_ctime, 1 + 12, header->header.ctime);
- }
-
-#ifndef NONAMES
- /* Fill in new Unix Standard fields if desired. */
- if (f_standard)
- {
- header->header.linkflag = LF_NORMAL; /* New default */
- strcpy (header->header.magic, TMAGIC); /* Mark as Unix Std */
- finduname (header->header.uname, st->st_uid);
- findgname (header->header.gname, st->st_gid);
- }
#endif
- return header;
-}
-
-/*
- * Finish off a filled-in header block and write it out.
- * We also print the file name and/or full info if verbose is on.
- */
-void
-finish_header (header)
- register union record *header;
-{
- register int i, sum;
- register char *p;
-
- bcopy (CHKBLANKS, header->header.chksum, sizeof (header->header.chksum));
-
- sum = 0;
- p = header->charptr;
- for (i = sizeof (*header); --i >= 0;)
- {
- /*
- * We can't use unsigned char here because of old compilers,
- * e.g. V7.
- */
- sum += 0xFF & *p++;
- }
-
- /*
- * Fill in the checksum field. It's formatted differently
- * from the other fields: it has [6] digits, a null, then a
- * space -- rather than digits, a space, then a null.
- * We use to_oct then write the null in over to_oct's space.
- * The final space is already there, from checksumming, and
- * to_oct doesn't modify it.
- *
- * This is a fast way to do:
- * (void) sprintf(header->header.chksum, "%6o", sum);
- */
- to_oct ((long) sum, 8, header->header.chksum);
- header->header.chksum[6] = '\0'; /* Zap the space */
-
- userec (header);
-
- if (f_verbose)
- {
- extern union record *head;/* Points to current tape header */
- extern int head_standard; /* Tape header is in ANSI format */
-
- /* These globals are parameters to print_header, sigh */
- head = header;
- /* hstat is already set up */
- head_standard = f_standard;
- print_header ();
- }
-
- return;
-}
-
-
-/*
- * Quick and dirty octal conversion.
- * Converts long "value" into a "digs"-digit field at "where",
- * including a trailing space and room for a null. "digs"==3 means
- * 1 digit, a space, and room for a null.
- *
- * We assume the trailing null is already there and don't fill it in.
- * This fact is used by start_header and finish_header, so don't change it!
- *
- * This should be equivalent to:
- * (void) sprintf(where, "%*lo ", digs-2, value);
- * except that sprintf fills in the trailing null and we don't.
- */
-void
-to_oct (value, digs, where)
- register long value;
- register int digs;
- register char *where;
-{
-
- --digs; /* Trailing null slot is left alone */
- where[--digs] = ' '; /* Put in the space, though */
-
- /* Produce the digits -- at least one */
- do
- {
- where[--digs] = '0' + (char) (value & 7); /* one octal digit */
- value >>= 3;
+ else
+ goto unknown;
}
- while (digs > 0 && value != 0);
-
- /* Leading spaces, if necessary */
- while (digs > 0)
- where[--digs] = ' ';
-
-}
+ if (archive_format == V7_FORMAT)
+ goto unknown;
-/*
- * Write the EOT record(s).
- * We actually zero at least one record, through the end of the block.
- * Old tar writes garbage after two zeroed records -- and PDtar used to.
- */
-void
-write_eot ()
-{
- union record *p;
- int bufsize;
+ current_stat.st_size = 0; /* force 0 size */
+ header = start_header (p, ¤t_stat);
+ header->header.typeflag = type;
- p = findrec ();
- if (p)
+ if (type != FIFOTYPE)
{
- bufsize = endofrecs ()->charptr - p->charptr;
- bzero (p->charptr, bufsize);
- userec (p);
+ MAJOR_TO_CHARS (major (current_stat.st_rdev), header->header.devmajor);
+ MINOR_TO_CHARS (minor (current_stat.st_rdev), header->header.devminor);
}
-}
-
-/* Write a LF_LONGLINK or LF_LONGNAME record. */
-void
-write_long (p, type)
- char *p;
- char type;
-{
- int size = strlen (p) + 1;
- int bufsize;
- union record *header;
- struct stat foo;
-
-
- bzero (&foo, sizeof foo);
- foo.st_size = size;
- header = start_header ("././@LongLink", &foo);
- header->header.linkflag = type;
finish_header (header);
-
- header = findrec ();
-
- bufsize = endofrecs ()->charptr - header->charptr;
-
- while (bufsize < size)
+ if (remove_files_option)
{
- bcopy (p, header->charptr, bufsize);
- p += bufsize;
- size -= bufsize;
- userec (header + (bufsize - 1) / RECORDSIZE);
- header = findrec ();
- bufsize = endofrecs ()->charptr - header->charptr;
+ if (unlink (p) == -1)
+ ERROR ((0, errno, _("Cannot remove %s"), p));
}
- bcopy (p, header->charptr, size);
- bzero (header->charptr + size, bufsize - size);
- userec (header + (size - 1) / RECORDSIZE);
+ return;
+
+unknown:
+ ERROR ((0, 0, _("%s: Unknown file type; file ignored"), p));
}