freebsd-dev/gnu/usr.bin/cpio/util.c
Jordan K. Hubbard 8843e04abc This is GNU cpio 2.3. It handles tar files and is fully compatible
with SYSV cpio.  It's also supposed to integrate rmt support, though
I haven't tried this.
1993-08-07 22:33:48 +00:00

1103 lines
30 KiB
C
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/* util.c - Several utility routines for cpio.
Copyright (C) 1990, 1991, 1992 Free Software Foundation, Inc.
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.
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.
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., 675 Mass Ave, Cambridge, MA 02139, USA. */
#include <stdio.h>
#include <sys/types.h>
#ifdef HPUX_CDF
#include <sys/stat.h>
#endif
#include "system.h"
#include "cpiohdr.h"
#include "dstring.h"
#include "extern.h"
#include "rmt.h"
#ifndef __MSDOS__
#include <sys/ioctl.h>
#else
#include <io.h>
#endif
#ifdef HAVE_SYS_MTIO_H
#ifdef HAVE_SYS_IO_TRIOCTL_H
#include <sys/io/trioctl.h>
#endif
#include <sys/mtio.h>
#endif
static void empty_output_buffer_swap ();
static void hash_insert ();
/* Write `output_size' bytes of `output_buffer' to file
descriptor OUT_DES and reset `output_size' and `out_buff'. */
void
empty_output_buffer (out_des)
int out_des;
{
int bytes_written;
#ifdef BROKEN_LONG_TAPE_DRIVER
static long output_bytes_before_lseek = 0;
#endif
if (swapping_halfwords || swapping_bytes)
{
empty_output_buffer_swap (out_des);
return;
}
#ifdef BROKEN_LONG_TAPE_DRIVER
/* Some tape drivers seem to have a signed internal seek pointer and
they lose if it overflows and becomes negative (e.g. when writing
tapes > 2Gb). Doing an lseek (des, 0, SEEK_SET) seems to reset the
seek pointer and prevent it from overflowing. */
if (output_is_special
&& (output_bytes_before_lseek += output_size) < 0L)
{
lseek(out_des, 0L, SEEK_SET);
output_bytes_before_lseek = 0;
}
#endif
bytes_written = rmtwrite (out_des, output_buffer, output_size);
if (bytes_written != output_size)
{
int rest_bytes_written;
int rest_output_size;
if (output_is_special
&& (bytes_written >= 0
|| (bytes_written < 0
&& (errno == ENOSPC || errno == EIO || errno == ENXIO))))
{
get_next_reel (out_des);
if (bytes_written > 0)
rest_output_size = output_size - bytes_written;
else
rest_output_size = output_size;
rest_bytes_written = rmtwrite (out_des, output_buffer,
rest_output_size);
if (rest_bytes_written != rest_output_size)
error (1, errno, "write error");
}
else
error (1, errno, "write error");
}
output_bytes += output_size;
out_buff = output_buffer;
output_size = 0;
}
/* Write `output_size' bytes of `output_buffer' to file
descriptor OUT_DES with byte and/or halfword swapping and reset
`output_size' and `out_buff'. This routine should not be called
with `swapping_bytes' set unless the caller knows that the
file being written has an even number of bytes, and it should not be
called with `swapping_halfwords' set unless the caller knows
that the file being written has a length divisible by 4. If either
of those restrictions are not met, bytes may be lost in the output
file. OUT_DES must refer to a file that we are creating during
a process_copy_in, so we don't have to check for end of media
errors or be careful about only writing in blocks of `output_size'
bytes. */
static void
empty_output_buffer_swap (out_des)
int out_des;
{
/* Since `output_size' might not be divisible by 4 or 2, we might
not be able to be able to swap all the bytes and halfwords in
`output_buffer' (e.g., if `output_size' is odd), so we might not be
able to write them all. We will swap and write as many bytes as
we can, and save the rest in `left_overs' for the next time we are
called. */
static char left_overs[4];
static int left_over_bytes = 0;
int bytes_written;
int complete_halfwords;
int complete_words;
int extra_bytes;
output_bytes += output_size;
out_buff = output_buffer;
if (swapping_halfwords)
{
if (left_over_bytes != 0)
{
while (output_size > 0 && left_over_bytes < 4)
{
left_overs[left_over_bytes++] = *out_buff++;
--output_size;
}
if (left_over_bytes < 4)
{
out_buff = output_buffer;
output_size = 0;
return;
}
swahw_array (left_overs, 1);
if (swapping_bytes)
swab_array (left_overs, 2);
bytes_written = rmtwrite (out_des, left_overs, 4);
if (bytes_written != 4)
error (1, errno, "write error");
left_over_bytes = 0;
}
complete_words = output_size / 4;
if (complete_words > 0)
{
swahw_array (out_buff, complete_words);
if (swapping_bytes)
swab_array (out_buff, 2 * complete_words);
bytes_written = rmtwrite (out_des, out_buff, 4 * complete_words);
if (bytes_written != (4 * complete_words))
error (1, errno, "write error");
}
out_buff += (4 * complete_words);
extra_bytes = output_size % 4;
while (extra_bytes > 0)
{
left_overs[left_over_bytes++] = *out_buff++;
--extra_bytes;
}
}
else
{
if (left_over_bytes != 0)
{
while (output_size > 0 && left_over_bytes < 2)
{
left_overs[left_over_bytes++] = *out_buff++;
--output_size;
}
if (left_over_bytes < 2)
{
out_buff = output_buffer;
output_size = 0;
return;
}
swab_array (left_overs, 1);
bytes_written = rmtwrite (out_des, left_overs, 2);
if (bytes_written != 2)
error (1, errno, "write error");
left_over_bytes = 0;
}
complete_halfwords = output_size / 2;
if (complete_halfwords > 0)
{
swab_array (out_buff, complete_halfwords);
bytes_written = rmtwrite (out_des, out_buff, 2 * complete_halfwords);
if (bytes_written != (2 * complete_halfwords))
error (1, errno, "write error");
}
out_buff += (2 * complete_halfwords);
extra_bytes = output_size % 2;
while (extra_bytes > 0)
{
left_overs[left_over_bytes++] = *out_buff++;
--extra_bytes;
}
}
out_buff = output_buffer;
output_size = 0;
}
/* Exchange the halfwords of each element of the array of COUNT longs
starting at PTR. PTR does not have to be aligned at a word
boundary. */
void
swahw_array (ptr, count)
char *ptr;
int count;
{
char tmp;
for (; count > 0; --count)
{
tmp = *ptr;
*ptr = *(ptr + 2);
*(ptr + 2) = tmp;
++ptr;
tmp = *ptr;
*ptr = *(ptr + 2);
*(ptr + 2) = tmp;
ptr += 3;
}
}
/* Read at most NUM_BYTES or `io_block_size' bytes, whichever is smaller,
into the start of `input_buffer' from file descriptor IN_DES.
Set `input_size' to the number of bytes read and reset `in_buff'.
Exit with an error if end of file is reached. */
#ifdef BROKEN_LONG_TAPE_DRIVER
static long input_bytes_before_lseek = 0;
#endif
void
fill_input_buffer (in_des, num_bytes)
int in_des;
int num_bytes;
{
#ifdef BROKEN_LONG_TAPE_DRIVER
/* Some tape drivers seem to have a signed internal seek pointer and
they lose if it overflows and becomes negative (e.g. when writing
tapes > 4Gb). Doing an lseek (des, 0, SEEK_SET) seems to reset the
seek pointer and prevent it from overflowing. */
if (input_is_special
&& (input_bytes_before_lseek += num_bytes) < 0L)
{
lseek(in_des, 0L, SEEK_SET);
input_bytes_before_lseek = 0;
}
#endif
in_buff = input_buffer;
num_bytes = (num_bytes < io_block_size) ? num_bytes : io_block_size;
input_size = rmtread (in_des, input_buffer, num_bytes);
if (input_size == 0 && input_is_special)
{
get_next_reel (in_des);
input_size = rmtread (in_des, input_buffer, num_bytes);
}
if (input_size < 0)
error (1, errno, "read error");
if (input_size == 0)
{
error (0, 0, "premature end of file");
exit (1);
}
input_bytes += input_size;
}
/* Copy NUM_BYTES of buffer IN_BUF to `out_buff', which may be partly full.
When `out_buff' fills up, flush it to file descriptor OUT_DES. */
void
copy_buf_out (in_buf, out_des, num_bytes)
char *in_buf;
int out_des;
long num_bytes;
{
register long bytes_left = num_bytes; /* Bytes needing to be copied. */
register long space_left; /* Room left in output buffer. */
while (bytes_left > 0)
{
space_left = io_block_size - output_size;
if (space_left == 0)
empty_output_buffer (out_des);
else
{
if (bytes_left < space_left)
space_left = bytes_left;
bcopy (in_buf, out_buff, (unsigned) space_left);
out_buff += space_left;
output_size += space_left;
in_buf += space_left;
bytes_left -= space_left;
}
}
}
/* Copy NUM_BYTES of buffer `in_buff' into IN_BUF.
`in_buff' may be partly full.
When `in_buff' is exhausted, refill it from file descriptor IN_DES. */
void
copy_in_buf (in_buf, in_des, num_bytes)
char *in_buf;
int in_des;
long num_bytes;
{
register long bytes_left = num_bytes; /* Bytes needing to be copied. */
register long space_left; /* Bytes to copy from input buffer. */
while (bytes_left > 0)
{
if (input_size == 0)
fill_input_buffer (in_des, io_block_size);
if (bytes_left < input_size)
space_left = bytes_left;
else
space_left = input_size;
bcopy (in_buff, in_buf, (unsigned) space_left);
in_buff += space_left;
in_buf += space_left;
input_size -= space_left;
bytes_left -= space_left;
}
}
/* Copy the the next NUM_BYTES bytes of `input_buffer' into PEEK_BUF.
If NUM_BYTES bytes are not available, read the next `io_block_size' bytes
into the end of `input_buffer' and update `input_size'.
Return the number of bytes copied into PEEK_BUF.
If the number of bytes returned is less than NUM_BYTES,
then EOF has been reached. */
int
peek_in_buf (peek_buf, in_des, num_bytes)
char *peek_buf;
int in_des;
int num_bytes;
{
long tmp_input_size;
long got_bytes;
char *append_buf;
#ifdef BROKEN_LONG_TAPE_DRIVER
/* Some tape drivers seem to have a signed internal seek pointer and
they lose if it overflows and becomes negative (e.g. when writing
tapes > 4Gb). Doing an lseek (des, 0, SEEK_SET) seems to reset the
seek pointer and prevent it from overflowing. */
if (input_is_special
&& (input_bytes_before_lseek += num_bytes) < 0L)
{
lseek(in_des, 0L, SEEK_SET);
input_bytes_before_lseek = 0;
}
#endif
while (input_size < num_bytes)
{
append_buf = in_buff + input_size;
tmp_input_size = rmtread (in_des, append_buf, io_block_size);
if (tmp_input_size == 0)
{
if (input_is_special)
{
get_next_reel (in_des);
tmp_input_size = rmtread (in_des, append_buf, io_block_size);
}
else
break;
}
if (tmp_input_size < 0)
error (1, errno, "read error");
input_bytes += tmp_input_size;
input_size += tmp_input_size;
}
if (num_bytes <= input_size)
got_bytes = num_bytes;
else
got_bytes = input_size;
bcopy (in_buff, peek_buf, (unsigned) got_bytes);
return got_bytes;
}
/* Skip the next NUM_BYTES bytes of file descriptor IN_DES. */
void
toss_input (in_des, num_bytes)
int in_des;
long num_bytes;
{
register long bytes_left = num_bytes; /* Bytes needing to be copied. */
register long space_left; /* Bytes to copy from input buffer. */
while (bytes_left > 0)
{
if (input_size == 0)
fill_input_buffer (in_des, io_block_size);
if (bytes_left < input_size)
space_left = bytes_left;
else
space_left = input_size;
in_buff += space_left;
input_size -= space_left;
bytes_left -= space_left;
}
}
/* Copy a file using the input and output buffers, which may start out
partly full. After the copy, the files are not closed nor the last
block flushed to output, and the input buffer may still be partly
full. If `crc_i_flag' is set, add each byte to `crc'.
IN_DES is the file descriptor for input;
OUT_DES is the file descriptor for output;
NUM_BYTES is the number of bytes to copy. */
void
copy_files (in_des, out_des, num_bytes)
int in_des;
int out_des;
long num_bytes;
{
long size;
long k;
while (num_bytes > 0)
{
if (input_size == 0)
fill_input_buffer (in_des, io_block_size);
size = (input_size < num_bytes) ? input_size : num_bytes;
if (crc_i_flag)
{
for (k = 0; k < size; ++k)
crc += in_buff[k] & 0xff;
}
copy_buf_out (in_buff, out_des, size);
num_bytes -= size;
input_size -= size;
in_buff += size;
}
}
/* Create all directories up to but not including the last part of NAME.
Do not destroy any nondirectories while creating directories. */
void
create_all_directories (name)
char *name;
{
char *dir;
int mode;
#ifdef HPUX_CDF
int cdf;
#endif
dir = dirname (name);
mode = 0700;
#ifdef HPUX_CDF
cdf = islastparentcdf (name);
if (cdf)
{
dir [strlen (dir) - 1] = '\0'; /* remove final + */
mode = 04700;
}
#endif
if (dir == NULL)
error (2, 0, "virtual memory exhausted");
if (dir[0] != '.' || dir[1] != '\0')
make_path (dir, mode, 0700, -1, -1, (char *) NULL);
free (dir);
}
/* Prepare to append to an archive. We have been in
process_copy_in, keeping track of the position where
the last header started in `last_header_start'. Now we
have the starting position of the last header (the TRAILER!!!
header, or blank record for tar archives) and we want to start
writing (appending) over the last header. The last header may
be in the middle of a block, so to keep the buffering in sync
we lseek back to the start of the block, read everything up
to but not including the last header, lseek back to the start
of the block, and then do a copy_buf_out of what we read.
Actually, we probably don't have to worry so much about keeping the
buffering perfect since you can only append to archives that
are disk files. */
void
prepare_append (out_file_des)
int out_file_des;
{
int start_of_header;
int start_of_block;
int useful_bytes_in_block;
char *tmp_buf;
start_of_header = last_header_start;
/* Figure out how many bytes we will rewrite, and where they start. */
useful_bytes_in_block = start_of_header % io_block_size;
start_of_block = start_of_header - useful_bytes_in_block;
if (lseek (out_file_des, start_of_block, SEEK_SET) < 0)
error (1, errno, "cannot seek on output");
if (useful_bytes_in_block > 0)
{
tmp_buf = (char *) xmalloc (useful_bytes_in_block);
read (out_file_des, tmp_buf, useful_bytes_in_block);
if (lseek (out_file_des, start_of_block, SEEK_SET) < 0)
error (1, errno, "cannot seek on output");
copy_buf_out (tmp_buf, out_file_des, useful_bytes_in_block);
free (tmp_buf);
}
/* We are done reading the archive, so clear these since they
will now be used for reading in files that we are appending
to the archive. */
input_size = 0;
input_bytes = 0;
in_buff = input_buffer;
}
/* Support for remembering inodes with multiple links. Used in the
"copy in" and "copy pass" modes for making links instead of copying
the file. */
struct inode_val
{
unsigned long inode;
unsigned long major_num;
unsigned long minor_num;
char *file_name;
};
/* Inode hash table. Allocated by first call to add_inode. */
static struct inode_val **hash_table = NULL;
/* Size of current hash table. Initial size is 47. (47 = 2*22 + 3) */
static int hash_size = 22;
/* Number of elements in current hash table. */
static int hash_num;
/* Find the file name associated with NODE_NUM. If there is no file
associated with NODE_NUM, return NULL. */
char *
find_inode_file (node_num, major_num, minor_num)
unsigned long node_num;
unsigned long major_num;
unsigned long minor_num;
{
#ifndef __MSDOS__
int start; /* Initial hash location. */
int temp; /* Rehash search variable. */
if (hash_table != NULL)
{
/* Hash function is node number modulo the table size. */
start = node_num % hash_size;
/* Initial look into the table. */
if (hash_table[start] == NULL)
return NULL;
if (hash_table[start]->inode == node_num
&& hash_table[start]->major_num == major_num
&& hash_table[start]->minor_num == minor_num)
return hash_table[start]->file_name;
/* The home position is full with a different inode record.
Do a linear search terminated by a NULL pointer. */
for (temp = (start + 1) % hash_size;
hash_table[temp] != NULL && temp != start;
temp = (temp + 1) % hash_size)
{
if (hash_table[temp]->inode == node_num
&& hash_table[start]->major_num == major_num
&& hash_table[start]->minor_num == minor_num)
return hash_table[temp]->file_name;
}
}
#endif
return NULL;
}
/* Associate FILE_NAME with the inode NODE_NUM. (Insert into hash table.) */
void
add_inode (node_num, file_name, major_num, minor_num)
unsigned long node_num;
char *file_name;
unsigned long major_num;
unsigned long minor_num;
{
#ifndef __MSDOS__
struct inode_val *temp;
/* Create new inode record. */
temp = (struct inode_val *) xmalloc (sizeof (struct inode_val));
temp->inode = node_num;
temp->major_num = major_num;
temp->minor_num = minor_num;
temp->file_name = xstrdup (file_name);
/* Do we have to increase the size of (or initially allocate)
the hash table? */
if (hash_num == hash_size || hash_table == NULL)
{
struct inode_val **old_table; /* Pointer to old table. */
int i; /* Index for re-insert loop. */
/* Save old table. */
old_table = hash_table;
if (old_table == NULL)
hash_num = 0;
/* Calculate new size of table and allocate it.
Sequence of table sizes is 47, 97, 197, 397, 797, 1597, 3197, 6397 ...
where 3197 and most of the sizes after 6397 are not prime. The other
numbers listed are prime. */
hash_size = 2 * hash_size + 3;
hash_table = (struct inode_val **)
xmalloc (hash_size * sizeof (struct inode_val *));
bzero (hash_table, hash_size * sizeof (struct inode_val *));
/* Insert the values from the old table into the new table. */
for (i = 0; i < hash_num; i++)
hash_insert (old_table[i]);
if (old_table != NULL)
free (old_table);
}
/* Insert the new record and increment the count of elements in the
hash table. */
hash_insert (temp);
hash_num++;
#endif /* __MSDOS__ */
}
/* Do the hash insert. Used in normal inserts and resizing the hash
table. It is guaranteed that there is room to insert the item.
NEW_VALUE is the pointer to the previously allocated inode, file
name association record. */
static void
hash_insert (new_value)
struct inode_val *new_value;
{
int start; /* Home position for the value. */
int temp; /* Used for rehashing. */
/* Hash function is node number modulo the table size. */
start = new_value->inode % hash_size;
/* Do the initial look into the table. */
if (hash_table[start] == NULL)
{
hash_table[start] = new_value;
return;
}
/* If we get to here, the home position is full with a different inode
record. Do a linear search for the first NULL pointer and insert
the new item there. */
temp = (start + 1) % hash_size;
while (hash_table[temp] != NULL)
temp = (temp + 1) % hash_size;
/* Insert at the NULL. */
hash_table[temp] = new_value;
}
/* Open FILE in the mode specified by the command line options
and return an open file descriptor for it,
or -1 if it can't be opened. */
int
open_archive (file)
char *file;
{
int fd;
void (*copy_in) (); /* Workaround for pcc bug. */
copy_in = process_copy_in;
if (copy_function == copy_in)
fd = rmtopen (file, O_RDONLY | O_BINARY, 0666);
else
{
if (!append_flag)
fd = rmtopen (file, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0666);
else
fd = rmtopen (file, O_RDWR | O_BINARY, 0666);
}
return fd;
}
/* Attempt to rewind the tape drive on file descriptor TAPE_DES
and take it offline. */
void
tape_offline (tape_des)
int tape_des;
{
#if defined(MTIOCTOP) && defined(MTOFFL)
struct mtop control;
control.mt_op = MTOFFL;
control.mt_count = 1;
rmtioctl (tape_des, MTIOCTOP, &control); /* Don't care if it fails. */
#endif
}
/* The file on file descriptor TAPE_DES is assumed to be magnetic tape
(or floppy disk or other device) and the end of the medium
has been reached. Ask the user for to mount a new "tape" to continue
the processing. If the user specified the device name on the
command line (with the -I, -O, -F or --file options), then we can
automatically re-open the same device to use the next medium. If the
user did not specify the device name, then we have to ask them which
device to use. */
void
get_next_reel (tape_des)
int tape_des;
{
static int reel_number = 1;
FILE *tty_in; /* File for interacting with user. */
FILE *tty_out; /* File for interacting with user. */
int old_tape_des;
char *next_archive_name;
dynamic_string new_name;
char *str_res;
ds_init (&new_name, 128);
/* Open files for interactive communication. */
tty_in = fopen (CONSOLE, "r");
if (tty_in == NULL)
error (2, errno, CONSOLE);
tty_out = fopen (CONSOLE, "w");
if (tty_out == NULL)
error (2, errno, CONSOLE);
old_tape_des = tape_des;
tape_offline (tape_des);
rmtclose (tape_des);
/* Give message and wait for carrage return. User should hit carrage return
only after loading the next tape. */
++reel_number;
if (new_media_message)
fprintf (tty_out, "%s", new_media_message);
else if (new_media_message_with_number)
fprintf (tty_out, "%s%d%s", new_media_message_with_number, reel_number,
new_media_message_after_number);
else if (archive_name)
fprintf (tty_out, "Found end of tape. Load next tape and press RETURN. ");
else
fprintf (tty_out, "Found end of tape. To continue, type device/file name when ready.\n");
fflush (tty_out);
if (archive_name)
{
int c;
do
c = getc (tty_in);
while (c != EOF && c != '\n');
tape_des = open_archive (archive_name);
if (tape_des == -1)
error (1, errno, "%s", archive_name);
}
else
{
do
{
if (tape_des < 0)
{
fprintf (tty_out,
"To continue, type device/file name when ready.\n");
fflush (tty_out);
}
str_res = ds_fgets (tty_in, &new_name);
if (str_res == NULL || str_res[0] == '\0')
exit (1);
next_archive_name = str_res;
tape_des = open_archive (next_archive_name);
if (tape_des == -1)
error (0, errno, "%s", next_archive_name);
}
while (tape_des < 0);
}
/* We have to make sure that `tape_des' has not changed its value even
though we closed it and reopened it, since there are local
copies of it in other routines. This works fine on Unix (even with
rmtread and rmtwrite) since open will always return the lowest
available file descriptor and we haven't closed any files (e.g.,
stdin, stdout or stderr) that were opened before we originally opened
the archive. */
if (tape_des != old_tape_des)
error (1, 0, "internal error: tape descriptor changed from %d to %d",
old_tape_des, tape_des);
free (new_name.ds_string);
fclose (tty_in);
fclose (tty_out);
}
/* If MESSAGE does not contain the string "%d", make `new_media_message'
a copy of MESSAGE. If MESSAGES does contain the string "%d", make
`new_media_message_with_number' a copy of MESSAGE up to, but
not including, the string "%d", and make `new_media_message_after_number'
a copy of MESSAGE after the string "%d". */
void
set_new_media_message (message)
char *message;
{
char *p;
int prev_was_percent;
p = message;
prev_was_percent = 0;
while (*p != '\0')
{
if (*p == 'd' && prev_was_percent)
break;
prev_was_percent = (*p == '%');
++p;
}
if (*p == '\0')
{
new_media_message = xstrdup (message);
}
else
{
int length = p - message - 1;
new_media_message_with_number = xmalloc (length + 1);
strncpy (new_media_message_with_number, message, length);
new_media_message_with_number[length] = '\0';
length = strlen (p + 1);
new_media_message_after_number = xmalloc (length + 1);
strcpy (new_media_message_after_number, message);
}
}
#ifdef SYMLINK_USES_UMASK
/* Most machines always create symlinks with rwxrwxrwx protection,
but some (HP/UX 8.07; maybe DEC's OSF on MIPS, too?) use the
umask when creating symlinks, so if your umask is 022 you end
up with rwxr-xr-x symlinks (although HP/UX seems to completely
ignore the protection). There doesn't seem to be any way to
manipulate the modes once the symlinks are created (e.g.
a hypothetical "lchmod"), so to create them with the right
modes we have to set the umask first. */
int
umasked_symlink (name1, name2, mode)
char *name1;
char *name2;
int mode;
{
int old_umask;
int rc;
mode = ~(mode & 0777) & 0777;
old_umask = umask (mode);
rc = symlink (name1, name2);
umask (old_umask);
return rc;
}
#endif /* SYMLINK_USES_UMASK */
#ifdef __MSDOS__
int
chown (path, owner, group)
char *path;
int owner, group;
{
return 0;
}
#endif
#ifdef __TURBOC__
#include <time.h>
#include <fcntl.h>
#include <io.h>
int
utime (char *filename, struct utimbuf *utb)
{
extern int errno;
struct tm *tm;
struct ftime filetime;
time_t when;
int fd;
int status;
if (utb == 0)
when = time (0);
else
when = utb->modtime;
fd = _open (filename, O_RDWR);
if (fd == -1)
return -1;
tm = localtime (&when);
if (tm->tm_year < 80)
filetime.ft_year = 0;
else
filetime.ft_year = tm->tm_year - 80;
filetime.ft_month = tm->tm_mon + 1;
filetime.ft_day = tm->tm_mday;
if (tm->tm_hour < 0)
filetime.ft_hour = 0;
else
filetime.ft_hour = tm->tm_hour;
filetime.ft_min = tm->tm_min;
filetime.ft_tsec = tm->tm_sec / 2;
status = setftime (fd, &filetime);
_close (fd);
return status;
}
#endif
#ifdef HPUX_CDF
/* When we create a cpio archive we mark CDF's by putting an extra `/'
after their component name so we can distinguish the CDF's when we
extract the archive (in case the "hidden" directory's files appear
in the archive before the directory itself). E.g., in the path
"a/b+/c", if b+ is a CDF, we will write this path as "a/b+//c" in
the archive so when we extract the archive we will know that b+
is actually a CDF, and not an ordinary directory whose name happens
to end in `+'. We also do the same thing internally in copypass.c. */
/* Take an input pathname and check it for CDF's. Insert an extra
`/' in the pathname after each "hidden" directory. If we add
any `/'s, return a malloced string (which it will reuse for
later calls so our caller doesn't have to worry about freeing
the string) instead of the original input string. */
char *
add_cdf_double_slashes (input_name)
char *input_name;
{
static char *ret_name = NULL; /* re-usuable return buffer (malloc'ed) */
static int ret_size = -1; /* size of return buffer. */
char *p;
char *q;
int n;
struct stat dir_stat;
/* Search for a `/' preceeded by a `+'. */
for (p = input_name; *p != '\0'; ++p)
{
if ( (*p == '+') && (*(p + 1) == '/') )
break;
}
/* If we didn't find a `/' preceeded by a `+' then there are
no CDF's in this pathname. Return the original pathname. */
if (*p == '\0')
return input_name;
/* There was a `/' preceeded by a `+' in the pathname. If it is a CDF
then we will need to copy the input pathname to our return
buffer so we can insert the extra `/'s. Since we can't tell
yet whether or not it is a CDF we will just always copy the
string to the return buffer. First we have to make sure the
buffer is large enough to hold the string and any number of
extra `/'s we might add. */
n = 2 * (strlen (input_name) + 1);
if (n >= ret_size)
{
if (ret_size < 0)
ret_name = (char *) malloc (n);
else
ret_name = (char *)realloc (ret_name, n);
ret_size = n;
}
/* Clear the `/' after this component, so we can stat the pathname
up to and including this component. */
++p;
*p = '\0';
if ((*xstat) (input_name, &dir_stat) < 0)
{
error (0, errno, "%s", input_name);
return input_name;
}
/* Now put back the `/' after this component and copy the pathname up to
and including this component and its trailing `/' to the return
buffer. */
*p++ = '/';
strncpy (ret_name, input_name, p - input_name);
q = ret_name + (p - input_name);
/* If it was a CDF, add another `/'. */
if (S_ISDIR (dir_stat.st_mode) && (dir_stat.st_mode & 04000) )
*q++ = '/';
/* Go through the rest of the input pathname, copying it to the
return buffer, and adding an extra `/' after each CDF. */
while (*p != '\0')
{
if ( (*p == '+') && (*(p + 1) == '/') )
{
*q++ = *p++;
*p = '\0';
if ((*xstat) (input_name, &dir_stat) < 0)
{
error (0, errno, "%s", input_name);
return input_name;
}
*p = '/';
if (S_ISDIR (dir_stat.st_mode) && (dir_stat.st_mode & 04000) )
*q++ = '/';
}
*q++ = *p++;
}
*q = '\0';
return ret_name;
}
/* Is the last parent directory (e.g., c in a/b/c/d) a CDF? If the
directory name ends in `+' and is followed by 2 `/'s instead of 1
then it is. This is only the case for cpio archives, but we don't
have to worry about tar because tar always has the directory before
its files (or else we lose). */
islastparentcdf(path)
char *path;
{
char *newpath;
char *slash;
int slash_count;
int length; /* Length of result, not including NUL. */
slash = rindex (path, '/');
if (slash == 0)
return 0;
else
{
slash_count = 0;
while (slash > path && *slash == '/')
{
++slash_count;
--slash;
}
if ( (*slash == '+') && (slash_count >= 2) )
return 1;
}
return 0;
}
#endif