/* Create a tar archive.
- Copyright 1985, 92, 93, 94, 96, 97, 99, 2000 Free Software Foundation, Inc.
+
+ Copyright (C) 1985, 1992, 1993, 1994, 1996, 1997, 1999, 2000, 2001,
+ 2003 Free Software Foundation, Inc.
+
Written by John Gilmore, on 1985-08-25.
This program is free software; you can redistribute it and/or modify it
#include <quotearg.h>
#include "common.h"
+#include <hash.h>
#ifndef MSDOS
extern dev_t ar_dev;
extern ino_t ar_ino;
#endif
-extern struct name *gnu_list_name;
-
-/* This module is the only one that cares about `struct link's. */
-
struct link
{
- struct link *next;
dev_t dev;
ino_t ino;
char name[1];
};
-
-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. */
static void
to_chars (int negative, uintmax_t value, size_t valsize,
- uintmax_t (*substitute) PARAMS ((int *)),
+ uintmax_t (*substitute) (int *),
char *where, size_t size, const char *type)
{
int base256_allowed = (archive_format == GNU_FORMAT
\f
/* Writing routines. */
-/*-----------------------------------------------------------------------.
-| Just zeroes out the buffer so we don't confuse ourselves with leftover |
-| data. |
-`-----------------------------------------------------------------------*/
-
+/* Zero 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. |
-`-------------------------------------------------------------------------*/
-
+/* Write the EOT block(s). 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)
{
set_next_block_after (pointer);
}
-/*-----------------------------------------------------.
-| Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. |
-`-----------------------------------------------------*/
+/* 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 union block *start_header (const char *, struct stat *);
static void
write_long (const char *p, char type)
header = start_header ("././@LongLink", &foo);
header->header.typeflag = type;
- finish_header (header);
+ finish_header (header, -1);
header = find_next_block ();
{
union block *header;
- if (!absolute_names_option)
- {
- size_t prefix_len = FILESYSTEM_PREFIX_LEN (name);
-
- if (prefix_len)
- {
- static int warned_once;
- if (!warned_once)
- {
- warned_once = 1;
- WARN ((0, 0, _("Removing `%.*s' prefix from member names"),
- (int) prefix_len, name));
- }
- name += prefix_len;
- }
-
- while (*name == '/')
- {
- static int warned_once;
- if (!warned_once)
- {
- warned_once = 1;
- WARN ((0, 0, _("Removing leading `/' from member names")));
- }
- name++;
- }
-
- {
- static int warned_once;
- if (! warned_once && contains_dot_dot (name))
- {
- warned_once = 1;
- WARN ((0, 0, _("Member names contain `..'")));
- }
- }
- }
+ name = safer_name_suffix (name, 0);
if (sizeof header->header.name <= strlen (name))
write_long (name, GNUTYPE_LONGNAME);
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);
+ MAJOR_TO_CHARS (0, header->header.devmajor);
+ MINOR_TO_CHARS (0, header->header.devminor);
if (incremental_option)
if (archive_format == OLDGNU_FORMAT)
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. |
-`-------------------------------------------------------------------------*/
-
+/* 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 BLOCK_ORDINAL
+ is not negative, is the block ordinal of the first record for this
+ file, which may be a preceding long name or long link record. */
void
-finish_header (union block *header)
+finish_header (union block *header, off_t block_ordinal)
{
size_t i;
int sum;
current_header = header;
/* current_stat is already set up. */
current_format = archive_format;
- print_header ();
+ print_header (block_ordinal);
}
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. |
-`-------------------------------------------------------------------------*/
-
+/* 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)
{
return 1;
}
-/*---.
-| ? |
-`---*/
-
-static void
+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++)
- {
- sparsearray[counter].offset = 0;
- sparsearray[counter].numbytes = 0;
- }
+ if (! sp_array_size)
+ sp_array_size = SPARSES_IN_OLDGNU_HEADER;
+ sparsearray = xmalloc (sp_array_size * sizeof *sparsearray);
}
-/*---.
-| ? |
-`---*/
-
static off_t
find_new_file_size (int sparses)
{
return s;
}
-/*-----------------------------------------------------------------------.
-| 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. |
-`-----------------------------------------------------------------------*/
+/* 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. */
init_sparsearray ();
clear_buffer (buffer);
- while (0 < (count = safe_read (file, buffer, sizeof buffer)))
+ for (;;)
{
/* Realloc the scratch area as necessary. FIXME: should reallocate
only at beginning of a new instance of non-zero data. */
2 * sp_array_size * sizeof (struct sp_array));
sp_array_size *= 2;
}
+
+ count = safe_read (file, buffer, sizeof buffer);
+ if (count <= 0)
+ break;
/* Process one block. */
sparsearray[sparses++].numbytes = 1;
}
- close (file);
- return count < 0 ? 0 : sparses;
+ return close (file) == 0 && 0 <= count ? sparses : 0;
}
-/*---.
-| ? |
-`---*/
-
static int
finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name)
{
\f
/* Main functions of this module. */
-/*---.
-| ? |
-`---*/
-
void
create_archive (void)
{
if (incremental_option)
{
- char *buffer = xmalloc (PATH_MAX);
+ size_t buffer_size = 1000;
+ char *buffer = xmalloc (buffer_size);
const char *q;
- char *bufp;
collect_and_sort_names ();
while (p = name_from_list (), p)
if (!excluded_name (p))
{
- strcpy (buffer, p);
- if (p[strlen (p) - 1] != '/')
- strcat (buffer, "/");
- bufp = buffer + strlen (buffer);
+ size_t plen = strlen (p);
+ if (buffer_size <= plen)
+ {
+ while ((buffer_size *= 2) <= plen)
+ continue;
+ buffer = xrealloc (buffer, buffer_size);
+ }
+ memcpy (buffer, p, plen);
+ if (! ISSLASH (buffer[plen - 1]))
+ buffer[plen++] = '/';
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);
- }
+ while (*q)
+ {
+ size_t qlen = strlen (q);
+ if (*q == 'Y')
+ {
+ if (buffer_size < plen + qlen)
+ {
+ while ((buffer_size *=2 ) < plen + qlen)
+ continue;
+ buffer = xrealloc (buffer, buffer_size);
+ }
+ strcpy (buffer + plen, q + 1);
+ dump_file (buffer, -1, (dev_t) 0);
+ }
+ q += qlen + 1;
+ }
}
free (buffer);
}
write_directory_file ();
}
+
+/* Calculate the hash of a link. */
+static unsigned
+hash_link (void const *entry, unsigned n_buckets)
+{
+ struct link const *link = entry;
+ return (uintmax_t) (link->dev ^ link->ino) % n_buckets;
+}
+
+/* Compare two links for equality. */
+static bool
+compare_links (void const *entry1, void const *entry2)
+{
+ struct link const *link1 = entry1;
+ struct link const *link2 = entry2;
+ return ((link1->dev ^ link2->dev) | (link1->ino ^ link2->ino)) == 0;
+}
+
/* 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
+ means no, positive means yes, and negative means the top level
+ of an incremental dump. PARENT_DEVICE is the device of P's
parent directory; it is examined only if TOP_LEVEL is zero.
Set global CURRENT_STAT to stat output for this file. */
char save_typeflag;
time_t original_ctime;
struct utimbuf restore_times;
+ off_t block_ordinal = -1;
+
+ /* Table of all non-directories that we've written so far. Any time
+ we see another, we check the table and avoid dumping the data
+ again if we've done it once already. */
+ static Hash_table *link_table;
- /* FIXME: `header' and `sparses' might be used uninitialized in this
+ /* FIXME: `header' might be used uninitialized in this
function. Reported by Bruno Haible. */
if (interactive_option && !confirm ("add", p))
if (S_ISDIR (current_stat.st_mode))
{
- DIR *directory;
- struct dirent *entry;
+ char *directory;
+ char const *entry;
+ size_t entrylen;
char *namebuf;
size_t buflen;
size_t len;
errno = 0;
- directory = opendir (p);
+ directory = savedir (p);
if (! directory)
{
if (ignore_failed_read_option)
- opendir_warn (p);
+ savedir_warn (p);
else
- opendir_error (p);
+ savedir_error (p);
return;
}
len = strlen (p);
buflen = len + NAME_FIELD_SIZE;
namebuf = xmalloc (buflen + 1);
- strncpy (namebuf, p, buflen);
- while (len >= 1 && namebuf[len - 1] == '/')
+ memcpy (namebuf, p, len);
+ while (len >= 1 && ISSLASH (namebuf[len - 1]))
len--;
namebuf[len++] = '/';
namebuf[len] = '\0';
reasons either, so until these are reported (anew?), just allow
directory blocks to be written even with old archives. */
+ block_ordinal = current_block_ordinal ();
current_stat.st_size = 0; /* force 0 size on dir */
/* FIXME: If people could really read standard archives, this
/* If we're gnudumping, we aren't done yet so don't close it. */
if (!incremental_option)
- finish_header (header); /* done with directory header */
+ finish_header (header, block_ordinal);
}
if (incremental_option && gnu_list_name->dir_contents)
buffer = gnu_list_name->dir_contents; /* FOO */
totsize = 0;
- for (p_buffer = buffer; p_buffer && *p_buffer;)
- {
- size_t tmp;
-
- tmp = strlen (p_buffer) + 1;
- totsize += tmp;
- p_buffer += tmp;
- }
+ if (buffer)
+ for (p_buffer = buffer; *p_buffer; )
+ {
+ size_t size = strlen (p_buffer) + 1;
+ totsize += size;
+ p_buffer += size;
+ }
totsize++;
OFF_TO_CHARS (totsize, header->header.size);
- finish_header (header);
+ finish_header (header, block_ordinal);
p_buffer = buffer;
sizeleft = totsize;
while (sizeleft > 0)
/* 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)
+ if (! recursion_option)
goto finish_dir;
/* See if we are crossing from one file system to another, and
/* FIXME: Should speed this up by cd-ing into the dir. */
- while (errno = 0, (entry = readdir (directory)))
+ for (entry = directory;
+ (entrylen = strlen (entry)) != 0;
+ entry += entrylen + 1)
{
- /* Skip `.', `..', and excluded file names. */
-
- if (is_dot_or_dotdot (entry->d_name))
- continue;
-
- if ((int) NAMLEN (entry) + len >= buflen)
+ if (buflen < len + entrylen)
{
- buflen = len + NAMLEN (entry);
+ buflen = len + entrylen;
namebuf = xrealloc (namebuf, buflen + 1);
}
- strcpy (namebuf + len, entry->d_name);
+ strcpy (namebuf + len, entry);
if (!excluded_name (namebuf))
dump_file (namebuf, 0, our_device);
}
- if (errno)
- {
- if (ignore_failed_read_option)
- readdir_warn (p);
- else
- readdir_error (p);
- }
-
finish_dir:
- if (closedir (directory) != 0)
- {
- if (ignore_failed_read_option)
- closedir_warn (p);
- else
- closedir_error (p);
- }
+ free (directory);
free (namebuf);
if (atime_preserve_option)
utime (p, &restore_times);
return;
else
{
- /* Check for multiple links.
+ /* 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 (1 < current_stat.st_nlink)
+ if (1 < current_stat.st_nlink && link_table)
{
- struct link *lp;
-
- /* FIXME: First quick and dirty. Hashing, etc later. */
-
- 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;
-
- /* We found a link. */
+ struct link lp;
+ struct link *dup;
+ lp.ino = current_stat.st_ino;
+ lp.dev = current_stat.st_dev;
- 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);
-
- current_stat.st_size = 0;
- header = start_header (p, ¤t_stat);
- strncpy (header->header.linkname,
- link_name, NAME_FIELD_SIZE);
-
- /* Force null truncated. */
+ if ((dup = hash_lookup (link_table, &lp)))
+ {
+ /* We found a link. */
+ char const *link_name = safer_name_suffix (dup->name, 1);
- header->header.linkname[NAME_FIELD_SIZE - 1] = 0;
+ block_ordinal = current_block_ordinal ();
+ if (NAME_FIELD_SIZE <= strlen (link_name))
+ write_long (link_name, GNUTYPE_LONGLINK);
+ assign_string (¤t_link_name, link_name);
- header->header.typeflag = LNKTYPE;
- finish_header (header);
+ current_stat.st_size = 0;
+ header = start_header (p, ¤t_stat);
+ strncpy (header->header.linkname, link_name, NAME_FIELD_SIZE);
- /* FIXME: Maybe remove from list after all links found? */
+ /* Force null termination. */
+ header->header.linkname[NAME_FIELD_SIZE - 1] = 0;
- if (remove_files_option)
- if (unlink (p) == -1)
- unlink_error (p);
+ header->header.typeflag = LNKTYPE;
+ finish_header (header, block_ordinal);
- /* We dumped it. */
- return;
- }
+ /* FIXME: Maybe remove from table after all links found? */
- /* Not found. Add it to the list of possible links. */
+ if (remove_files_option && unlink (p) != 0)
+ unlink_error (p);
- 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;
+ /* We dumped it, and we don't need to put it in the
+ table again. */
+ return;
+ }
}
/* This is not a link to a previously dumped file, so dump it. */
union block *start;
int header_moved;
char isextended = 0;
- int sparses;
+ int sparses = 0;
header_moved = 0;
{
/* 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
+ If there is a smaller number of blocks than 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
{
int counter;
+ block_ordinal = current_block_ordinal ();
header = start_header (p, ¤t_stat);
header->header.typeflag = GNUTYPE_SPARSE;
header_moved = 1;
}
}
}
- else
- sparses = SPARSES_IN_OLDGNU_HEADER;
sizeleft = current_stat.st_size;
/* If the file is sparse, we've already taken care of this. */
if (!header_moved)
- header = start_header (p, ¤t_stat);
+ {
+ block_ordinal = current_block_ordinal ();
+ header = start_header (p, ¤t_stat);
+ }
/* Mark contiguous files, if we support them. */
isextended = header->oldgnu_header.isextended;
save_typeflag = header->header.typeflag;
- finish_header (header);
+ finish_header (header, block_ordinal);
if (isextended)
{
- int counter;
- /* static */ int index_offset = SPARSES_IN_OLDGNU_HEADER;
+ int sparses_emitted = SPARSES_IN_OLDGNU_HEADER;
- extend:
- exhdr = find_next_block ();
- memset (exhdr->buffer, 0, BLOCKSIZE);
- for (counter = 0; counter < SPARSES_IN_SPARSE_HEADER; counter++)
+ for (;;)
{
- if (sparses <= counter + index_offset)
+ int i;
+ exhdr = find_next_block ();
+ memset (exhdr->buffer, 0, BLOCKSIZE);
+ for (i = 0;
+ (i < SPARSES_IN_SPARSE_HEADER
+ && sparses_emitted + i < sparses);
+ i++)
+ {
+ SIZE_TO_CHARS (sparsearray[sparses_emitted + i].numbytes,
+ exhdr->sparse_header.sp[i].numbytes);
+ OFF_TO_CHARS (sparsearray[sparses_emitted + i].offset,
+ exhdr->sparse_header.sp[i].offset);
+ }
+ set_next_block_after (exhdr);
+ sparses_emitted += i;
+ if (sparses == sparses_emitted)
break;
-
- 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 (counter + index_offset < sparses)
- {
- index_offset += counter;
exhdr->sparse_header.isextended = 1;
- goto extend;
}
-
}
if (save_typeflag == GNUTYPE_SPARSE)
{
/* This is nonportable (the type of set_next_block_after's arg). */
- set_next_block_after (start + (count - 1) / BLOCKSIZE);
+ set_next_block_after (start + (bufsize - 1) / BLOCKSIZE);
- if (count == bufsize)
- continue;
- else
+
+ if (count != bufsize)
{
char buf[UINTMAX_STRSIZE_BOUND];
+ memset (start->buffer + count, 0, bufsize - count);
WARN ((0, 0,
_("%s: File shrank by %s bytes; padding with zeros"),
quotearg_colon (p),
{
char const *qp = quotearg_colon (p);
WARN ((0, 0, _("%s: file changed as we read it"), qp));
- if (! ignore_failed_read_option)
- exit_status = TAREXIT_FAILURE;
}
if (close (f) != 0)
{
if (unlink (p) == -1)
unlink_error (p);
}
- return;
+ goto file_was_dumped;
/* File shrunk or gave error, pad out tape to match the size we
specified in the header. */
if (atime_preserve_option)
utime (p, &restore_times);
}
- return;
+ goto file_was_dumped;
}
#ifdef HAVE_READLINK
else if (S_ISLNK (current_stat.st_mode))
{
+ char *buffer;
int size;
- char *buffer = (char *) alloca (PATH_MAX + 1);
-
- size = readlink (p, buffer, PATH_MAX + 1);
+ size_t linklen = current_stat.st_size;
+ if (linklen != current_stat.st_size || linklen + 1 == 0)
+ xalloc_die ();
+ buffer = (char *) alloca (linklen + 1);
+ size = readlink (p, buffer, linklen + 1);
if (size < 0)
{
if (ignore_failed_read_option)
write_long (buffer, GNUTYPE_LONGLINK);
assign_string (¤t_link_name, buffer);
+ block_ordinal = current_block_ordinal ();
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 */
+ finish_header (header, block_ordinal);
+ /* nothing more to do to it */
+
if (remove_files_option)
{
if (unlink (p) == -1)
unlink_error (p);
}
- return;
+ goto file_was_dumped;
}
#endif
else if (S_ISCHR (current_stat.st_mode))
if (archive_format == V7_FORMAT)
goto unknown;
+ block_ordinal = current_block_ordinal ();
current_stat.st_size = 0; /* force 0 size */
header = start_header (p, ¤t_stat);
header->header.typeflag = type;
MINOR_TO_CHARS (minor (current_stat.st_rdev), header->header.devminor);
}
- finish_header (header);
+ finish_header (header, block_ordinal);
if (remove_files_option)
{
if (unlink (p) == -1)
unlink_error (p);
}
- return;
+ goto file_was_dumped;
unknown:
WARN ((0, 0, _("%s: Unknown file type; file ignored"),
quotearg_colon (p)));
if (! ignore_failed_read_option)
exit_status = TAREXIT_FAILURE;
+ return;
+
+file_was_dumped:
+ if (1 < current_stat.st_nlink)
+ {
+ struct link *dup;
+ struct link *lp = xmalloc (offsetof (struct link, name)
+ + strlen (p) + 1);
+ lp->ino = current_stat.st_ino;
+ lp->dev = current_stat.st_dev;
+ strcpy (lp->name, p);
+
+ if (! ((link_table
+ || (link_table = hash_initialize (0, 0, hash_link,
+ compare_links, 0)))
+ && (dup = hash_insert (link_table, lp))))
+ xalloc_die ();
+
+ if (dup != lp)
+ abort ();
+ }
+
}
projects / chaz / tar / blobdiff